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);
126 #define MAX_STRING_SIZE 128
127 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
128 struct obd_connect_data *data)
130 struct lov_obd *lov = &obd->u.lov;
131 struct obd_uuid *tgt_uuid;
132 struct obd_device *tgt_obd;
133 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
134 struct obd_import *imp;
137 if (!lov->lov_tgts[index])
140 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
141 tgt_obd = lov->lov_tgts[index]->ltd_obd;
143 if (!tgt_obd->obd_set_up) {
144 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
148 /* override the sp_me from lov */
149 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
151 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
152 data->ocd_index = index;
155 * Divine LOV knows that OBDs under it are OSCs.
157 imp = tgt_obd->u.cli.cl_import;
160 tgt_obd->obd_no_recov = 0;
161 /* FIXME this is probably supposed to be
162 ptlrpc_set_import_active. Horrible naming. */
163 ptlrpc_activate_import(imp);
166 rc = obd_register_observer(tgt_obd, obd);
168 CERROR("Target %s register_observer error %d\n",
169 obd_uuid2str(tgt_uuid), rc);
174 if (imp->imp_invalid) {
175 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
176 obd_uuid2str(tgt_uuid));
180 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
181 &lov_osc_uuid, data, NULL);
182 if (rc || !lov->lov_tgts[index]->ltd_exp) {
183 CERROR("Target %s connect error %d\n",
184 obd_uuid2str(tgt_uuid), rc);
188 lov->lov_tgts[index]->ltd_reap = 0;
190 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
191 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
193 if (lov->lov_tgts_kobj)
194 /* Even if we failed, that's ok */
195 rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj,
201 static int lov_connect(const struct lu_env *env,
202 struct obd_export **exp, struct obd_device *obd,
203 struct obd_uuid *cluuid, struct obd_connect_data *data,
206 struct lov_obd *lov = &obd->u.lov;
207 struct lov_tgt_desc *tgt;
208 struct lustre_handle conn;
211 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
213 rc = class_connect(&conn, obd, cluuid);
217 *exp = class_conn2export(&conn);
219 /* Why should there ever be more than 1 connect? */
221 LASSERT(lov->lov_connects == 1);
223 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
225 lov->lov_ocd = *data;
229 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
232 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
233 tgt = lov->lov_tgts[i];
234 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
236 /* Flags will be lowest common denominator */
237 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
239 CERROR("%s: lov connect tgt %d failed: %d\n",
240 obd->obd_name, i, rc);
243 /* connect to administrative disabled ost */
244 if (!lov->lov_tgts[i]->ltd_exp)
247 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
248 OBD_NOTIFY_CONNECT, (void *)&i);
250 CERROR("%s error sending notify %d\n",
259 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
261 struct lov_obd *lov = &obd->u.lov;
262 struct obd_device *osc_obd;
265 osc_obd = class_exp2obd(tgt->ltd_exp);
266 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
267 obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
269 if (tgt->ltd_active) {
271 lov->desc.ld_active_tgt_count--;
272 tgt->ltd_exp->exp_obd->obd_inactive = 1;
276 if (lov->lov_tgts_kobj)
277 sysfs_remove_link(lov->lov_tgts_kobj,
280 /* Pass it on to our clients.
281 * XXX This should be an argument to disconnect,
282 * XXX not a back-door flag on the OBD. Ah well.
284 osc_obd->obd_force = obd->obd_force;
285 osc_obd->obd_fail = obd->obd_fail;
286 osc_obd->obd_no_recov = obd->obd_no_recov;
289 obd_register_observer(osc_obd, NULL);
291 rc = obd_disconnect(tgt->ltd_exp);
293 CERROR("Target %s disconnect error %d\n",
294 tgt->ltd_uuid.uuid, rc);
302 static int lov_disconnect(struct obd_export *exp)
304 struct obd_device *obd = class_exp2obd(exp);
305 struct lov_obd *lov = &obd->u.lov;
311 /* Only disconnect the underlying layers on the final disconnect. */
313 if (lov->lov_connects != 0) {
314 /* why should there be more than 1 connect? */
315 CERROR("disconnect #%d\n", lov->lov_connects);
319 /* Let's hold another reference so lov_del_obd doesn't spin through
323 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
324 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
325 /* Disconnection is the last we know about an obd */
326 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
333 rc = class_disconnect(exp); /* bz 9811 */
339 * -EINVAL : UUID can't be found in the LOV's target list
340 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
341 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
342 * any >= 0 : is log target index
344 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
345 enum obd_notify_event ev)
347 struct lov_obd *lov = &obd->u.lov;
348 struct lov_tgt_desc *tgt;
349 int index, activate, active;
351 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
352 lov, uuid->uuid, ev);
355 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
356 tgt = lov->lov_tgts[index];
360 * LU-642, initially inactive OSC could miss the obd_connect,
361 * we make up for it here.
363 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
364 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
365 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
367 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
368 &lov_osc_uuid, &lov->lov_ocd, NULL);
373 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
374 index, obd_uuid2str(&tgt->ltd_uuid),
375 tgt->ltd_exp->exp_handle.h_cookie);
376 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
380 if (index == lov->desc.ld_tgt_count) {
385 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
386 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
388 if (lov->lov_tgts[index]->ltd_activate == activate) {
389 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
390 uuid->uuid, activate ? "" : "de");
392 lov->lov_tgts[index]->ltd_activate = activate;
393 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
394 activate ? "" : "de", obd_uuid2str(uuid));
397 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
398 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
400 if (lov->lov_tgts[index]->ltd_active == active) {
401 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
402 uuid->uuid, active ? "" : "in");
405 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
406 obd_uuid2str(uuid), active ? "" : "in");
409 lov->lov_tgts[index]->ltd_active = active;
411 lov->desc.ld_active_tgt_count++;
412 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
414 lov->desc.ld_active_tgt_count--;
415 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
418 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
426 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
427 enum obd_notify_event ev, void *data)
430 struct lov_obd *lov = &obd->u.lov;
432 down_read(&lov->lov_notify_lock);
433 if (!lov->lov_connects) {
434 up_read(&lov->lov_notify_lock);
438 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
439 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
440 struct obd_uuid *uuid;
444 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
445 up_read(&lov->lov_notify_lock);
446 CERROR("unexpected notification of %s %s!\n",
447 watched->obd_type->typ_name,
451 uuid = &watched->u.cli.cl_target_uuid;
453 /* Set OSC as active before notifying the observer, so the
454 * observer can use the OSC normally.
456 rc = lov_set_osc_active(obd, uuid, ev);
458 up_read(&lov->lov_notify_lock);
459 CERROR("event(%d) of %s failed: %d\n", ev,
460 obd_uuid2str(uuid), rc);
463 /* active event should be pass lov target index as data */
467 /* Pass the notification up the chain. */
469 rc = obd_notify_observer(obd, watched, ev, data);
471 /* NULL watched means all osc's in the lov (only for syncs) */
472 /* sync event should be send lov idx as data */
473 struct lov_obd *lov = &obd->u.lov;
477 is_sync = (ev == OBD_NOTIFY_SYNC) ||
478 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
481 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
482 if (!lov->lov_tgts[i])
485 /* don't send sync event if target not
486 * connected/activated */
487 if (is_sync && !lov->lov_tgts[i]->ltd_active)
490 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
493 CERROR("%s: notify %s of %s failed %d\n",
495 obd->obd_observer->obd_name,
496 lov->lov_tgts[i]->ltd_obd->obd_name,
503 up_read(&lov->lov_notify_lock);
507 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
508 __u32 index, int gen, int active)
510 struct lov_obd *lov = &obd->u.lov;
511 struct lov_tgt_desc *tgt;
512 struct obd_device *tgt_obd;
515 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
516 uuidp->uuid, index, gen, active);
519 CERROR("request to add OBD %s with invalid generation: %d\n",
524 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
529 mutex_lock(&lov->lov_lock);
531 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
532 tgt = lov->lov_tgts[index];
533 CERROR("UUID %s already assigned at LOV target index %d\n",
534 obd_uuid2str(&tgt->ltd_uuid), index);
535 mutex_unlock(&lov->lov_lock);
539 if (index >= lov->lov_tgt_size) {
540 /* We need to reallocate the lov target array. */
541 struct lov_tgt_desc **newtgts, **old = NULL;
542 __u32 newsize, oldsize = 0;
544 newsize = max_t(__u32, lov->lov_tgt_size, 2);
545 while (newsize < index + 1)
547 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
548 if (newtgts == NULL) {
549 mutex_unlock(&lov->lov_lock);
553 if (lov->lov_tgt_size) {
554 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
557 oldsize = lov->lov_tgt_size;
560 lov->lov_tgts = newtgts;
561 lov->lov_tgt_size = newsize;
565 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
566 lov->lov_tgts, lov->lov_tgt_size);
569 tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
571 mutex_unlock(&lov->lov_lock);
575 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
577 mutex_unlock(&lov->lov_lock);
582 tgt->ltd_uuid = *uuidp;
583 tgt->ltd_obd = tgt_obd;
584 /* XXX - add a sanity check on the generation number. */
586 tgt->ltd_index = index;
587 tgt->ltd_activate = active;
588 lov->lov_tgts[index] = tgt;
589 if (index >= lov->desc.ld_tgt_count)
590 lov->desc.ld_tgt_count = index + 1;
592 mutex_unlock(&lov->lov_lock);
594 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
595 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
597 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
599 if (lov->lov_connects == 0) {
600 /* lov_connect hasn't been called yet. We'll do the
601 lov_connect_obd on this target when that fn first runs,
602 because we don't know the connect flags yet. */
608 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
612 /* connect to administrative disabled ost */
618 if (lov->lov_cache != NULL) {
619 rc = obd_set_info_async(NULL, tgt->ltd_exp,
620 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
621 sizeof(struct cl_client_cache), lov->lov_cache,
627 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
628 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
633 CERROR("add failed (%d), deleting %s\n", rc,
634 obd_uuid2str(&tgt->ltd_uuid));
635 lov_del_target(obd, index, NULL, 0);
641 /* Schedule a target for deletion */
642 int lov_del_target(struct obd_device *obd, __u32 index,
643 struct obd_uuid *uuidp, int gen)
645 struct lov_obd *lov = &obd->u.lov;
646 int count = lov->desc.ld_tgt_count;
649 if (index >= count) {
650 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
655 /* to make sure there's no ongoing lov_notify() now */
656 down_write(&lov->lov_notify_lock);
659 if (!lov->lov_tgts[index]) {
660 CERROR("LOV target at index %d is not setup.\n", index);
665 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
666 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
667 lov_uuid2str(lov, index), index,
668 obd_uuid2str(uuidp));
673 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
674 lov_uuid2str(lov, index), index,
675 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
676 lov->lov_tgts[index]->ltd_active);
678 lov->lov_tgts[index]->ltd_reap = 1;
679 lov->lov_death_row++;
680 /* we really delete it from obd_putref */
683 up_write(&lov->lov_notify_lock);
688 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
690 struct obd_device *osc_obd;
693 LASSERT(tgt->ltd_reap);
695 osc_obd = class_exp2obd(tgt->ltd_exp);
697 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
699 osc_obd ? osc_obd->obd_name : "<no obd>");
702 lov_disconnect_obd(obd, tgt);
706 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
707 do it ourselves. And we can't do it from lov_cleanup,
708 because we just lost our only reference to it. */
710 class_manual_cleanup(osc_obd);
713 void lov_fix_desc_stripe_size(__u64 *val)
715 if (*val < LOV_MIN_STRIPE_SIZE) {
717 LCONSOLE_INFO("Increasing default stripe size to minimum %u\n",
718 LOV_DESC_STRIPE_SIZE_DEFAULT);
719 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
720 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
721 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
722 LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n",
723 *val, LOV_MIN_STRIPE_SIZE);
727 void lov_fix_desc_stripe_count(__u32 *val)
733 void lov_fix_desc_pattern(__u32 *val)
735 /* from lov_setstripe */
736 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
737 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
742 void lov_fix_desc_qos_maxage(__u32 *val)
745 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
748 void lov_fix_desc(struct lov_desc *desc)
750 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
751 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
752 lov_fix_desc_pattern(&desc->ld_pattern);
753 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
756 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
758 struct lprocfs_static_vars lvars = { NULL };
759 struct lov_desc *desc;
760 struct lov_obd *lov = &obd->u.lov;
763 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
764 CERROR("LOV setup requires a descriptor\n");
768 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
770 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
771 CERROR("descriptor size wrong: %d > %d\n",
772 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
776 if (desc->ld_magic != LOV_DESC_MAGIC) {
777 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
778 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
779 obd->obd_name, desc);
780 lustre_swab_lov_desc(desc);
782 CERROR("%s: Bad lov desc magic: %#x\n",
783 obd->obd_name, desc->ld_magic);
790 desc->ld_active_tgt_count = 0;
792 lov->lov_tgt_size = 0;
794 mutex_init(&lov->lov_lock);
795 atomic_set(&lov->lov_refcount, 0);
796 lov->lov_sp_me = LUSTRE_SP_CLI;
798 init_rwsem(&lov->lov_notify_lock);
800 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
802 HASH_POOLS_BKT_BITS, 0,
805 &pool_hash_operations,
807 INIT_LIST_HEAD(&lov->lov_pool_list);
808 lov->lov_pool_count = 0;
809 rc = lov_ost_pool_init(&lov->lov_packed, 0);
813 lprocfs_lov_init_vars(&lvars);
814 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
816 rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd",
817 0444, &lov_proc_target_fops, obd);
819 CWARN("Error adding the target_obd file\n");
821 lov->lov_pool_debugfs_entry = ldebugfs_register("pools",
822 obd->obd_debugfs_entry,
830 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
832 struct lov_obd *lov = &obd->u.lov;
835 case OBD_CLEANUP_EARLY: {
838 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
839 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
841 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
853 static int lov_cleanup(struct obd_device *obd)
855 struct lov_obd *lov = &obd->u.lov;
856 struct list_head *pos, *tmp;
857 struct pool_desc *pool;
859 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
860 pool = list_entry(pos, struct pool_desc, pool_list);
861 /* free pool structs */
862 CDEBUG(D_INFO, "delete pool %p\n", pool);
863 /* In the function below, .hs_keycmp resolves to
864 * pool_hashkey_keycmp() */
865 /* coverity[overrun-buffer-val] */
866 lov_pool_del(obd, pool->pool_name);
868 cfs_hash_putref(lov->lov_pools_hash_body);
869 lov_ost_pool_free(&lov->lov_packed);
871 lprocfs_obd_cleanup(obd);
876 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
877 if (!lov->lov_tgts[i])
880 /* Inactive targets may never have connected */
881 if (lov->lov_tgts[i]->ltd_active ||
882 atomic_read(&lov->lov_refcount))
883 /* We should never get here - these
884 should have been removed in the
886 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
887 i, lov->lov_death_row,
888 atomic_read(&lov->lov_refcount));
889 lov_del_target(obd, i, NULL, 0);
892 kfree(lov->lov_tgts);
893 lov->lov_tgt_size = 0;
898 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
899 __u32 *indexp, int *genp)
901 struct obd_uuid obd_uuid;
905 switch (cmd = lcfg->lcfg_command) {
906 case LCFG_LOV_ADD_OBD:
907 case LCFG_LOV_ADD_INA:
908 case LCFG_LOV_DEL_OBD: {
911 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
912 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
917 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
919 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1) {
923 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1) {
929 if (cmd == LCFG_LOV_ADD_OBD)
930 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
931 else if (cmd == LCFG_LOV_ADD_INA)
932 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
934 rc = lov_del_target(obd, index, &obd_uuid, gen);
938 struct lprocfs_static_vars lvars = { NULL };
939 struct lov_desc *desc = &(obd->u.lov.desc);
946 lprocfs_lov_init_vars(&lvars);
948 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
961 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
971 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
972 struct lov_stripe_md **ea, struct obd_trans_info *oti)
974 struct lov_stripe_md *obj_mdp, *lsm;
975 struct lov_obd *lov = &exp->exp_obd->u.lov;
979 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
980 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
982 obj_mdp = kzalloc(sizeof(*obj_mdp), GFP_NOFS);
986 ost_idx = src_oa->o_nlink;
992 if (ost_idx >= lov->desc.ld_tgt_count ||
993 !lov->lov_tgts[ost_idx]) {
998 for (i = 0; i < lsm->lsm_stripe_count; i++) {
999 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1001 if (lov_oinfo_is_dummy(loi))
1004 if (loi->loi_ost_idx == ost_idx) {
1005 if (ostid_id(&loi->loi_oi) != ostid_id(&src_oa->o_oi)) {
1012 if (i == lsm->lsm_stripe_count) {
1017 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1018 src_oa, &obj_mdp, oti);
1024 /* the LOV expects oa->o_id to be set to the LOV object id */
1025 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1026 struct obdo *src_oa, struct lov_stripe_md **ea,
1027 struct obd_trans_info *oti)
1029 struct lov_obd *lov;
1032 LASSERT(ea != NULL);
1036 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1037 src_oa->o_flags == OBD_FL_DELORPHAN) {
1038 /* should be used with LOV anymore */
1042 lov = &exp->exp_obd->u.lov;
1043 if (!lov->desc.ld_active_tgt_count)
1046 obd_getref(exp->exp_obd);
1047 /* Recreate a specific object id at the given OST index */
1048 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1049 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1050 rc = lov_recreate(exp, src_oa, ea, oti);
1053 obd_putref(exp->exp_obd);
1057 #define ASSERT_LSM_MAGIC(lsmp) \
1059 LASSERT((lsmp) != NULL); \
1060 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1061 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1062 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1065 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1066 struct obdo *oa, struct lov_stripe_md *lsm,
1067 struct obd_trans_info *oti, struct obd_export *md_exp)
1069 struct lov_request_set *set;
1070 struct obd_info oinfo;
1071 struct lov_request *req;
1072 struct list_head *pos;
1073 struct lov_obd *lov;
1074 int rc = 0, err = 0;
1076 ASSERT_LSM_MAGIC(lsm);
1078 if (!exp || !exp->exp_obd)
1081 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1083 LASSERT(oti->oti_logcookies);
1086 lov = &exp->exp_obd->u.lov;
1087 obd_getref(exp->exp_obd);
1088 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1092 list_for_each(pos, &set->set_list) {
1093 req = list_entry(pos, struct lov_request, rq_link);
1095 if (oa->o_valid & OBD_MD_FLCOOKIE)
1096 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1098 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1099 req->rq_oi.oi_oa, NULL, oti, NULL);
1100 err = lov_update_common_set(set, req, err);
1102 CERROR("%s: destroying objid "DOSTID" subobj "
1103 DOSTID" on OST idx %d: rc = %d\n",
1104 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1105 POSTID(&req->rq_oi.oi_oa->o_oi),
1113 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1114 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1116 err = lov_fini_destroy_set(set);
1118 obd_putref(exp->exp_obd);
1119 return rc ? rc : err;
1122 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1125 struct lov_request_set *lovset = (struct lov_request_set *)data;
1128 /* don't do attribute merge if this async op failed */
1130 atomic_set(&lovset->set_completes, 0);
1131 err = lov_fini_getattr_set(lovset);
1132 return rc ? rc : err;
1135 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1136 struct ptlrpc_request_set *rqset)
1138 struct lov_request_set *lovset;
1139 struct lov_obd *lov;
1140 struct list_head *pos;
1141 struct lov_request *req;
1145 ASSERT_LSM_MAGIC(oinfo->oi_md);
1147 if (!exp || !exp->exp_obd)
1150 lov = &exp->exp_obd->u.lov;
1152 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1156 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1157 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1158 oinfo->oi_md->lsm_stripe_size);
1160 list_for_each(pos, &lovset->set_list) {
1161 req = list_entry(pos, struct lov_request, rq_link);
1163 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1164 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1165 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1166 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1167 &req->rq_oi, rqset);
1169 CERROR("%s: getattr objid "DOSTID" subobj"
1170 DOSTID" on OST idx %d: rc = %d\n",
1171 exp->exp_obd->obd_name,
1172 POSTID(&oinfo->oi_oa->o_oi),
1173 POSTID(&req->rq_oi.oi_oa->o_oi),
1179 if (!list_empty(&rqset->set_requests)) {
1181 LASSERT(rqset->set_interpret == NULL);
1182 rqset->set_interpret = lov_getattr_interpret;
1183 rqset->set_arg = (void *)lovset;
1188 atomic_set(&lovset->set_completes, 0);
1189 err = lov_fini_getattr_set(lovset);
1190 return rc ? rc : err;
1193 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1196 struct lov_request_set *lovset = (struct lov_request_set *)data;
1200 atomic_set(&lovset->set_completes, 0);
1201 err = lov_fini_setattr_set(lovset);
1202 return rc ? rc : err;
1205 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1206 needed. Otherwise, a client is waiting for responses. */
1207 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1208 struct obd_trans_info *oti,
1209 struct ptlrpc_request_set *rqset)
1211 struct lov_request_set *set;
1212 struct lov_request *req;
1213 struct list_head *pos;
1214 struct lov_obd *lov;
1218 ASSERT_LSM_MAGIC(oinfo->oi_md);
1219 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1221 LASSERT(oti->oti_logcookies);
1224 if (!exp || !exp->exp_obd)
1227 lov = &exp->exp_obd->u.lov;
1228 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1232 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1233 POSTID(&oinfo->oi_md->lsm_oi),
1234 oinfo->oi_md->lsm_stripe_count,
1235 oinfo->oi_md->lsm_stripe_size);
1237 list_for_each(pos, &set->set_list) {
1238 req = list_entry(pos, struct lov_request, rq_link);
1240 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1241 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1243 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1244 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1245 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1247 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1248 &req->rq_oi, oti, rqset);
1250 CERROR("error: setattr objid "DOSTID" subobj"
1251 DOSTID" on OST idx %d: rc = %d\n",
1252 POSTID(&set->set_oi->oi_oa->o_oi),
1253 POSTID(&req->rq_oi.oi_oa->o_oi),
1259 /* If we are not waiting for responses on async requests, return. */
1260 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1264 atomic_set(&set->set_completes, 0);
1265 err = lov_fini_setattr_set(set);
1266 return rc ? rc : err;
1269 LASSERT(rqset->set_interpret == NULL);
1270 rqset->set_interpret = lov_setattr_interpret;
1271 rqset->set_arg = (void *)set;
1276 /* find any ldlm lock of the inode in lov
1280 static int lov_find_cbdata(struct obd_export *exp,
1281 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1284 struct lov_obd *lov;
1287 ASSERT_LSM_MAGIC(lsm);
1289 if (!exp || !exp->exp_obd)
1292 lov = &exp->exp_obd->u.lov;
1293 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1294 struct lov_stripe_md submd;
1295 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1297 if (lov_oinfo_is_dummy(loi))
1300 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1301 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1305 submd.lsm_oi = loi->loi_oi;
1306 submd.lsm_stripe_count = 0;
1307 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1315 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1317 struct lov_request_set *lovset = (struct lov_request_set *)data;
1321 atomic_set(&lovset->set_completes, 0);
1323 err = lov_fini_statfs_set(lovset);
1324 return rc ? rc : err;
1327 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1328 __u64 max_age, struct ptlrpc_request_set *rqset)
1330 struct obd_device *obd = class_exp2obd(exp);
1331 struct lov_request_set *set;
1332 struct lov_request *req;
1333 struct list_head *pos;
1334 struct lov_obd *lov;
1337 LASSERT(oinfo != NULL);
1338 LASSERT(oinfo->oi_osfs != NULL);
1341 rc = lov_prep_statfs_set(obd, oinfo, &set);
1345 list_for_each(pos, &set->set_list) {
1346 req = list_entry(pos, struct lov_request, rq_link);
1347 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1348 &req->rq_oi, max_age, rqset);
1353 if (rc || list_empty(&rqset->set_requests)) {
1357 atomic_set(&set->set_completes, 0);
1358 err = lov_fini_statfs_set(set);
1359 return rc ? rc : err;
1362 LASSERT(rqset->set_interpret == NULL);
1363 rqset->set_interpret = lov_statfs_interpret;
1364 rqset->set_arg = (void *)set;
1368 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1369 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1371 struct ptlrpc_request_set *set = NULL;
1372 struct obd_info oinfo = { };
1375 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1376 * statfs requests */
1377 set = ptlrpc_prep_set();
1381 oinfo.oi_osfs = osfs;
1382 oinfo.oi_flags = flags;
1383 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1385 rc = ptlrpc_set_wait(set);
1386 ptlrpc_set_destroy(set);
1391 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1392 void *karg, void *uarg)
1394 struct obd_device *obddev = class_exp2obd(exp);
1395 struct lov_obd *lov = &obddev->u.lov;
1396 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1397 struct obd_uuid *uuidp;
1400 case IOC_OBD_STATFS: {
1401 struct obd_ioctl_data *data = karg;
1402 struct obd_device *osc_obd;
1403 struct obd_statfs stat_buf = {0};
1407 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1411 if (!lov->lov_tgts[index])
1412 /* Try again with the next index */
1414 if (!lov->lov_tgts[index]->ltd_active)
1417 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1422 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1423 min((int) data->ioc_plen2,
1424 (int) sizeof(struct obd_uuid))))
1427 flags = uarg ? *(__u32 *)uarg : 0;
1428 /* got statfs data */
1429 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1430 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1434 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1435 min((int) data->ioc_plen1,
1436 (int) sizeof(stat_buf))))
1440 case OBD_IOC_LOV_GET_CONFIG: {
1441 struct obd_ioctl_data *data;
1442 struct lov_desc *desc;
1447 if (obd_ioctl_getdata(&buf, &len, uarg))
1450 data = (struct obd_ioctl_data *)buf;
1452 if (sizeof(*desc) > data->ioc_inllen1) {
1453 obd_ioctl_freedata(buf, len);
1457 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1458 obd_ioctl_freedata(buf, len);
1462 if (sizeof(__u32) * count > data->ioc_inllen3) {
1463 obd_ioctl_freedata(buf, len);
1467 desc = (struct lov_desc *)data->ioc_inlbuf1;
1468 memcpy(desc, &(lov->desc), sizeof(*desc));
1470 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1471 genp = (__u32 *)data->ioc_inlbuf3;
1472 /* the uuid will be empty for deleted OSTs */
1473 for (i = 0; i < count; i++, uuidp++, genp++) {
1474 if (!lov->lov_tgts[i])
1476 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1477 *genp = lov->lov_tgts[i]->ltd_gen;
1480 if (copy_to_user(uarg, buf, len))
1482 obd_ioctl_freedata(buf, len);
1485 case LL_IOC_LOV_GETSTRIPE:
1486 rc = lov_getstripe(exp, karg, uarg);
1488 case OBD_IOC_QUOTACTL: {
1489 struct if_quotactl *qctl = karg;
1490 struct lov_tgt_desc *tgt = NULL;
1491 struct obd_quotactl *oqctl;
1493 if (qctl->qc_valid == QC_OSTIDX) {
1494 if (count <= qctl->qc_idx)
1497 tgt = lov->lov_tgts[qctl->qc_idx];
1498 if (!tgt || !tgt->ltd_exp)
1500 } else if (qctl->qc_valid == QC_UUID) {
1501 for (i = 0; i < count; i++) {
1502 tgt = lov->lov_tgts[i];
1504 !obd_uuid_equals(&tgt->ltd_uuid,
1508 if (tgt->ltd_exp == NULL)
1520 LASSERT(tgt && tgt->ltd_exp);
1521 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1525 QCTL_COPY(oqctl, qctl);
1526 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1528 QCTL_COPY(qctl, oqctl);
1529 qctl->qc_valid = QC_OSTIDX;
1530 qctl->obd_uuid = tgt->ltd_uuid;
1541 for (i = 0; i < count; i++) {
1543 struct obd_device *osc_obd;
1545 /* OST was disconnected */
1546 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1549 /* ll_umount_begin() sets force flag but for lov, not
1550 * osc. Let's pass it through */
1551 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1552 osc_obd->obd_force = obddev->obd_force;
1553 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1555 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1558 if (lov->lov_tgts[i]->ltd_active) {
1559 CDEBUG(err == -ENOTTY ?
1560 D_IOCTL : D_WARNING,
1561 "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1562 lov_uuid2str(lov, i),
1579 #define FIEMAP_BUFFER_SIZE 4096
1582 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1583 * call. The local end offset and the device are sent in the first
1584 * fm_extent. This function calculates the stripe number from the index.
1585 * This function returns a stripe_no on which mapping is to be restarted.
1587 * This function returns fm_end_offset which is the in-OST offset at which
1588 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1589 * will re-calculate proper offset in next stripe.
1590 * Note that the first extent is passed to lov_get_info via the value field.
1592 * \param fiemap fiemap request header
1593 * \param lsm striping information for the file
1594 * \param fm_start logical start of mapping
1595 * \param fm_end logical end of mapping
1596 * \param start_stripe starting stripe will be returned in this
1598 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1599 struct lov_stripe_md *lsm, u64 fm_start,
1600 u64 fm_end, int *start_stripe)
1602 u64 local_end = fiemap->fm_extents[0].fe_logical;
1603 u64 lun_start, lun_end;
1605 int stripe_no = -1, i;
1607 if (fiemap->fm_extent_count == 0 ||
1608 fiemap->fm_extents[0].fe_logical == 0)
1611 /* Find out stripe_no from ost_index saved in the fe_device */
1612 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1613 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1615 if (lov_oinfo_is_dummy(oinfo))
1618 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1623 if (stripe_no == -1)
1626 /* If we have finished mapping on previous device, shift logical
1627 * offset to start of next device */
1628 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1629 &lun_start, &lun_end)) != 0 &&
1630 local_end < lun_end) {
1631 fm_end_offset = local_end;
1632 *start_stripe = stripe_no;
1634 /* This is a special value to indicate that caller should
1635 * calculate offset in next stripe. */
1637 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1640 return fm_end_offset;
1644 * We calculate on which OST the mapping will end. If the length of mapping
1645 * is greater than (stripe_size * stripe_count) then the last_stripe will
1646 * will be one just before start_stripe. Else we check if the mapping
1647 * intersects each OST and find last_stripe.
1648 * This function returns the last_stripe and also sets the stripe_count
1649 * over which the mapping is spread
1651 * \param lsm striping information for the file
1652 * \param fm_start logical start of mapping
1653 * \param fm_end logical end of mapping
1654 * \param start_stripe starting stripe of the mapping
1655 * \param stripe_count the number of stripes across which to map is returned
1657 * \retval last_stripe return the last stripe of the mapping
1659 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1660 u64 fm_end, int start_stripe,
1664 u64 obd_start, obd_end;
1667 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1668 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1670 *stripe_count = lsm->lsm_stripe_count;
1672 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1673 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1674 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1675 &obd_start, &obd_end)) == 0)
1679 last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count;
1686 * Set fe_device and copy extents from local buffer into main return buffer.
1688 * \param fiemap fiemap request header
1689 * \param lcl_fm_ext array of local fiemap extents to be copied
1690 * \param ost_index OST index to be written into the fm_device field for each
1692 * \param ext_count number of extents to be copied
1693 * \param current_extent where to start copying in main extent array
1695 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1696 struct ll_fiemap_extent *lcl_fm_ext,
1697 int ost_index, unsigned int ext_count,
1703 for (ext = 0; ext < ext_count; ext++) {
1704 lcl_fm_ext[ext].fe_device = ost_index;
1705 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1708 /* Copy fm_extent's from fm_local to return buffer */
1709 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1710 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1714 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1715 * This also handles the restarting of FIEMAP calls in case mapping overflows
1716 * the available number of extents in single call.
1718 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1719 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1721 struct ll_fiemap_info_key *fm_key = key;
1722 struct ll_user_fiemap *fiemap = val;
1723 struct ll_user_fiemap *fm_local = NULL;
1724 struct ll_fiemap_extent *lcl_fm_ext;
1726 unsigned int get_num_extents = 0;
1727 int ost_index = 0, actual_start_stripe, start_stripe;
1728 u64 fm_start, fm_end, fm_length, fm_end_offset;
1730 int current_extent = 0, rc = 0, i;
1731 int ost_eof = 0; /* EOF for object */
1732 int ost_done = 0; /* done with required mapping for this OST? */
1734 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1735 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1737 if (!lsm_has_objects(lsm)) {
1742 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1743 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1745 fm_local = libcfs_kvzalloc(buffer_size, GFP_NOFS);
1746 if (fm_local == NULL) {
1750 lcl_fm_ext = &fm_local->fm_extents[0];
1752 count_local = fiemap_size_to_count(buffer_size);
1754 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1755 fm_start = fiemap->fm_start;
1756 fm_length = fiemap->fm_length;
1757 /* Calculate start stripe, last stripe and length of mapping */
1758 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1759 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1760 fm_start + fm_length - 1);
1761 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1762 if (fm_end > fm_key->oa.o_size)
1763 fm_end = fm_key->oa.o_size;
1765 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1766 actual_start_stripe, &stripe_count);
1768 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1769 fm_end, &start_stripe);
1770 if (fm_end_offset == -EINVAL) {
1775 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1776 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1777 if (fiemap->fm_extent_count == 0) {
1778 get_num_extents = 1;
1781 /* Check each stripe */
1782 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1783 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1784 u64 req_fm_len; /* Stores length of required mapping */
1785 u64 len_mapped_single_call;
1786 u64 lun_start, lun_end, obd_object_end;
1787 unsigned int ext_count;
1789 cur_stripe_wrap = cur_stripe;
1791 /* Find out range of mapping on this stripe */
1792 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1793 &lun_start, &obd_object_end)) == 0)
1796 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1801 /* If this is a continuation FIEMAP call and we are on
1802 * starting stripe then lun_start needs to be set to
1804 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1805 lun_start = fm_end_offset;
1807 if (fm_length != ~0ULL) {
1808 /* Handle fm_start + fm_length overflow */
1809 if (fm_start + fm_length < fm_start)
1810 fm_length = ~0ULL - fm_start;
1811 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1817 if (lun_start == lun_end)
1820 req_fm_len = obd_object_end - lun_start;
1821 fm_local->fm_length = 0;
1822 len_mapped_single_call = 0;
1824 /* If the output buffer is very large and the objects have many
1825 * extents we may need to loop on a single OST repeatedly */
1829 if (get_num_extents == 0) {
1830 /* Don't get too many extents. */
1831 if (current_extent + count_local >
1832 fiemap->fm_extent_count)
1833 count_local = fiemap->fm_extent_count -
1837 lun_start += len_mapped_single_call;
1838 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1839 req_fm_len = fm_local->fm_length;
1840 fm_local->fm_extent_count = count_local;
1841 fm_local->fm_mapped_extents = 0;
1842 fm_local->fm_flags = fiemap->fm_flags;
1844 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1845 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1847 if (ost_index < 0 ||
1848 ost_index >= lov->desc.ld_tgt_count) {
1853 /* If OST is inactive, return extent with UNKNOWN flag */
1854 if (!lov->lov_tgts[ost_index]->ltd_active) {
1855 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1856 fm_local->fm_mapped_extents = 1;
1858 lcl_fm_ext[0].fe_logical = lun_start;
1859 lcl_fm_ext[0].fe_length = obd_object_end -
1861 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1866 fm_local->fm_start = lun_start;
1867 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1868 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1869 *vallen = fiemap_count_to_size(fm_local->fm_extent_count);
1870 rc = obd_get_info(NULL,
1871 lov->lov_tgts[ost_index]->ltd_exp,
1872 keylen, key, vallen, fm_local, lsm);
1877 ext_count = fm_local->fm_mapped_extents;
1878 if (ext_count == 0) {
1880 /* If last stripe has hole at the end,
1881 * then we need to return */
1882 if (cur_stripe_wrap == last_stripe) {
1883 fiemap->fm_mapped_extents = 0;
1889 /* If we just need num of extents then go to next device */
1890 if (get_num_extents) {
1891 current_extent += ext_count;
1895 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1896 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1898 /* Have we finished mapping on this device? */
1899 if (req_fm_len <= len_mapped_single_call)
1902 /* Clear the EXTENT_LAST flag which can be present on
1904 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1905 lcl_fm_ext[ext_count - 1].fe_flags &=
1906 ~FIEMAP_EXTENT_LAST;
1908 curr_loc = lov_stripe_size(lsm,
1909 lcl_fm_ext[ext_count - 1].fe_logical+
1910 lcl_fm_ext[ext_count - 1].fe_length,
1912 if (curr_loc >= fm_key->oa.o_size)
1915 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1916 ost_index, ext_count,
1919 current_extent += ext_count;
1921 /* Ran out of available extents? */
1922 if (current_extent >= fiemap->fm_extent_count)
1924 } while (ost_done == 0 && ost_eof == 0);
1926 if (cur_stripe_wrap == last_stripe)
1931 /* Indicate that we are returning device offsets unless file just has
1933 if (lsm->lsm_stripe_count > 1)
1934 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1936 if (get_num_extents)
1937 goto skip_last_device_calc;
1939 /* Check if we have reached the last stripe and whether mapping for that
1940 * stripe is done. */
1941 if (cur_stripe_wrap == last_stripe) {
1942 if (ost_done || ost_eof)
1943 fiemap->fm_extents[current_extent - 1].fe_flags |=
1947 skip_last_device_calc:
1948 fiemap->fm_mapped_extents = current_extent;
1955 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1956 __u32 keylen, void *key, __u32 *vallen, void *val,
1957 struct lov_stripe_md *lsm)
1959 struct obd_device *obddev = class_exp2obd(exp);
1960 struct lov_obd *lov = &obddev->u.lov;
1963 if (!vallen || !val)
1968 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
1971 struct ldlm_lock *lock;
1973 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
1974 struct lov_oinfo *loi;
1975 __u32 *stripe = val;
1977 if (*vallen < sizeof(*stripe)) {
1981 *vallen = sizeof(*stripe);
1983 /* XXX This is another one of those bits that will need to
1984 * change if we ever actually support nested LOVs. It uses
1985 * the lock's export to find out which stripe it is. */
1986 /* XXX - it's assumed all the locks for deleted OSTs have
1987 * been cancelled. Also, the export for deleted OSTs will
1988 * be NULL and won't match the lock's export. */
1989 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1990 loi = lsm->lsm_oinfo[i];
1991 if (lov_oinfo_is_dummy(loi))
1994 if (!lov->lov_tgts[loi->loi_ost_idx])
1996 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
1997 data->lock->l_conn_export &&
1998 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2004 LDLM_ERROR(data->lock, "lock on inode without such object");
2005 dump_lsm(D_ERROR, lsm);
2008 } else if (KEY_IS(KEY_LAST_ID)) {
2009 struct obd_id_info *info = val;
2010 __u32 size = sizeof(u64);
2011 struct lov_tgt_desc *tgt;
2013 LASSERT(*vallen == sizeof(struct obd_id_info));
2014 tgt = lov->lov_tgts[info->idx];
2016 if (!tgt || !tgt->ltd_active) {
2021 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2022 &size, info->data, NULL);
2025 } else if (KEY_IS(KEY_LOVDESC)) {
2026 struct lov_desc *desc_ret = val;
2027 *desc_ret = lov->desc;
2031 } else if (KEY_IS(KEY_FIEMAP)) {
2032 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2034 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2035 struct lov_tgt_desc *tgt;
2036 __u64 ost_idx = *((__u64 *)val);
2038 LASSERT(*vallen == sizeof(__u64));
2039 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2040 tgt = lov->lov_tgts[ost_idx];
2042 if (!tgt || !tgt->ltd_exp) {
2047 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2050 } else if (KEY_IS(KEY_TGT_COUNT)) {
2051 *((int *)val) = lov->desc.ld_tgt_count;
2063 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2064 u32 keylen, void *key, u32 vallen,
2065 void *val, struct ptlrpc_request_set *set)
2067 struct obd_device *obddev = class_exp2obd(exp);
2068 struct lov_obd *lov = &obddev->u.lov;
2071 struct lov_tgt_desc *tgt;
2072 unsigned incr, check_uuid,
2073 do_inactive, no_set;
2074 unsigned next_id = 0, mds_con = 0;
2076 incr = check_uuid = do_inactive = no_set = 0;
2079 set = ptlrpc_prep_set();
2085 count = lov->desc.ld_tgt_count;
2087 if (KEY_IS(KEY_NEXT_ID)) {
2088 count = vallen / sizeof(struct obd_id_info);
2089 vallen = sizeof(u64);
2090 incr = sizeof(struct obd_id_info);
2093 } else if (KEY_IS(KEY_CHECKSUM)) {
2095 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2096 /* use defaults: do_inactive = incr = 0; */
2097 } else if (KEY_IS(KEY_MDS_CONN)) {
2099 } else if (KEY_IS(KEY_CACHE_SET)) {
2100 LASSERT(lov->lov_cache == NULL);
2101 lov->lov_cache = val;
2105 for (i = 0; i < count; i++, val = (char *)val + incr) {
2107 tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx];
2109 tgt = lov->lov_tgts[i];
2111 /* OST was disconnected */
2112 if (!tgt || !tgt->ltd_exp)
2115 /* OST is inactive and we don't want inactive OSCs */
2116 if (!tgt->ltd_active && !do_inactive)
2120 struct mds_group_info *mgi;
2122 LASSERT(vallen == sizeof(*mgi));
2123 mgi = (struct mds_group_info *)val;
2125 /* Only want a specific OSC */
2126 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2130 err = obd_set_info_async(env, tgt->ltd_exp,
2131 keylen, key, sizeof(int),
2133 } else if (next_id) {
2134 err = obd_set_info_async(env, tgt->ltd_exp,
2135 keylen, key, vallen,
2136 ((struct obd_id_info *)val)->data, set);
2138 /* Only want a specific OSC */
2140 !obd_uuid_equals(val, &tgt->ltd_uuid))
2143 err = obd_set_info_async(env, tgt->ltd_exp,
2144 keylen, key, vallen, val, set);
2153 err = ptlrpc_set_wait(set);
2156 ptlrpc_set_destroy(set);
2161 void lov_stripe_lock(struct lov_stripe_md *md)
2162 __acquires(&md->lsm_lock)
2164 LASSERT(md->lsm_lock_owner != current_pid());
2165 spin_lock(&md->lsm_lock);
2166 LASSERT(md->lsm_lock_owner == 0);
2167 md->lsm_lock_owner = current_pid();
2169 EXPORT_SYMBOL(lov_stripe_lock);
2171 void lov_stripe_unlock(struct lov_stripe_md *md)
2172 __releases(&md->lsm_lock)
2174 LASSERT(md->lsm_lock_owner == current_pid());
2175 md->lsm_lock_owner = 0;
2176 spin_unlock(&md->lsm_lock);
2178 EXPORT_SYMBOL(lov_stripe_unlock);
2180 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2181 struct obd_quotactl *oqctl)
2183 struct lov_obd *lov = &obd->u.lov;
2184 struct lov_tgt_desc *tgt;
2186 __u64 bhardlimit = 0;
2189 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2190 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2191 oqctl->qc_cmd != Q_GETOQUOTA &&
2192 oqctl->qc_cmd != Q_INITQUOTA &&
2193 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2194 oqctl->qc_cmd != Q_FINVALIDATE) {
2195 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2201 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2204 tgt = lov->lov_tgts[i];
2209 if (!tgt->ltd_active || tgt->ltd_reap) {
2210 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2211 lov->lov_tgts[i]->ltd_activate) {
2213 CERROR("ost %d is inactive\n", i);
2215 CDEBUG(D_HA, "ost %d is inactive\n", i);
2220 err = obd_quotactl(tgt->ltd_exp, oqctl);
2222 if (tgt->ltd_active && !rc)
2227 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2228 curspace += oqctl->qc_dqblk.dqb_curspace;
2229 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2234 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2235 oqctl->qc_dqblk.dqb_curspace = curspace;
2236 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2241 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2242 struct obd_quotactl *oqctl)
2244 struct lov_obd *lov = &obd->u.lov;
2249 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2250 if (!lov->lov_tgts[i])
2253 /* Skip quota check on the administratively disabled OSTs. */
2254 if (!lov->lov_tgts[i]->ltd_activate) {
2255 CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
2260 if (!lov->lov_tgts[i]->ltd_active) {
2261 CERROR("lov idx %d inactive\n", i);
2267 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2270 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2273 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2284 static struct obd_ops lov_obd_ops = {
2285 .o_owner = THIS_MODULE,
2286 .o_setup = lov_setup,
2287 .o_precleanup = lov_precleanup,
2288 .o_cleanup = lov_cleanup,
2289 /*.o_process_config = lov_process_config,*/
2290 .o_connect = lov_connect,
2291 .o_disconnect = lov_disconnect,
2292 .o_statfs = lov_statfs,
2293 .o_statfs_async = lov_statfs_async,
2294 .o_packmd = lov_packmd,
2295 .o_unpackmd = lov_unpackmd,
2296 .o_create = lov_create,
2297 .o_destroy = lov_destroy,
2298 .o_getattr_async = lov_getattr_async,
2299 .o_setattr_async = lov_setattr_async,
2300 .o_adjust_kms = lov_adjust_kms,
2301 .o_find_cbdata = lov_find_cbdata,
2302 .o_iocontrol = lov_iocontrol,
2303 .o_get_info = lov_get_info,
2304 .o_set_info_async = lov_set_info_async,
2305 .o_notify = lov_notify,
2306 .o_pool_new = lov_pool_new,
2307 .o_pool_rem = lov_pool_remove,
2308 .o_pool_add = lov_pool_add,
2309 .o_pool_del = lov_pool_del,
2310 .o_getref = lov_getref,
2311 .o_putref = lov_putref,
2312 .o_quotactl = lov_quotactl,
2313 .o_quotacheck = lov_quotacheck,
2316 struct kmem_cache *lov_oinfo_slab;
2318 static int __init lov_init(void)
2320 struct lprocfs_static_vars lvars = { NULL };
2323 /* print an address of _any_ initialized kernel symbol from this
2324 * module, to allow debugging with gdb that doesn't support data
2325 * symbols from modules.*/
2326 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2328 rc = lu_kmem_init(lov_caches);
2332 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2333 sizeof(struct lov_oinfo),
2334 0, SLAB_HWCACHE_ALIGN, NULL);
2335 if (lov_oinfo_slab == NULL) {
2336 lu_kmem_fini(lov_caches);
2339 lprocfs_lov_init_vars(&lvars);
2341 rc = class_register_type(&lov_obd_ops, NULL,
2342 LUSTRE_LOV_NAME, &lov_device_type);
2345 kmem_cache_destroy(lov_oinfo_slab);
2346 lu_kmem_fini(lov_caches);
2352 static void /*__exit*/ lov_exit(void)
2354 class_unregister_type(LUSTRE_LOV_NAME);
2355 kmem_cache_destroy(lov_oinfo_slab);
2357 lu_kmem_fini(lov_caches);
2360 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2361 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2362 MODULE_LICENSE("GPL");
2363 MODULE_VERSION(LUSTRE_VERSION_STRING);
2365 module_init(lov_init);
2366 module_exit(lov_exit);