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
46 #include <libcfs/libcfs.h>
48 #include <liblustre.h>
51 #include <obd_support.h>
52 #include <lustre_lib.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_dlm.h>
56 #include <lustre_mds.h>
57 #include <lustre_debug.h>
58 #include <obd_class.h>
61 #include <lprocfs_status.h>
62 #include <lustre_param.h>
63 #include <cl_object.h>
64 #include <lclient.h> /* for cl_client_lru */
65 #include <lustre/ll_fiemap.h>
66 #include <lustre_log.h>
67 #include <lustre_fid.h>
69 #include "lov_internal.h"
71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
72 Any function that expects lov_tgts to remain stationary must take a ref. */
73 static void lov_getref(struct obd_device *obd)
75 struct lov_obd *lov = &obd->u.lov;
77 /* nobody gets through here until lov_putref is done */
78 mutex_lock(&lov->lov_lock);
79 cfs_atomic_inc(&lov->lov_refcount);
80 mutex_unlock(&lov->lov_lock);
84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
86 static void lov_putref(struct obd_device *obd)
88 struct lov_obd *lov = &obd->u.lov;
90 mutex_lock(&lov->lov_lock);
91 /* ok to dec to 0 more than once -- ltd_exp's will be null */
92 if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
95 struct lov_tgt_desc *tgt, *n;
96 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
98 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
99 tgt = lov->lov_tgts[i];
101 if (!tgt || !tgt->ltd_reap)
103 cfs_list_add(&tgt->ltd_kill, &kill);
104 /* XXX - right now there is a dependency on ld_tgt_count
105 * being the maximum tgt index for computing the
106 * mds_max_easize. So we can't shrink it. */
107 lov_ost_pool_remove(&lov->lov_packed, i);
108 lov->lov_tgts[i] = NULL;
109 lov->lov_death_row--;
111 mutex_unlock(&lov->lov_lock);
113 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
114 cfs_list_del(&tgt->ltd_kill);
116 __lov_del_obd(obd, tgt);
119 mutex_unlock(&lov->lov_lock);
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
124 enum obd_notify_event ev);
125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
126 enum obd_notify_event ev, void *data);
129 #define MAX_STRING_SIZE 128
130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
131 struct obd_connect_data *data)
133 struct lov_obd *lov = &obd->u.lov;
134 struct obd_uuid *tgt_uuid;
135 struct obd_device *tgt_obd;
136 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
137 struct obd_import *imp;
139 cfs_proc_dir_entry_t *lov_proc_dir;
144 if (!lov->lov_tgts[index])
147 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
148 tgt_obd = lov->lov_tgts[index]->ltd_obd;
150 if (!tgt_obd->obd_set_up) {
151 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
155 /* override the sp_me from lov */
156 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
158 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
159 data->ocd_index = index;
162 * Divine LOV knows that OBDs under it are OSCs.
164 imp = tgt_obd->u.cli.cl_import;
167 tgt_obd->obd_no_recov = 0;
168 /* FIXME this is probably supposed to be
169 ptlrpc_set_import_active. Horrible naming. */
170 ptlrpc_activate_import(imp);
173 rc = obd_register_observer(tgt_obd, obd);
175 CERROR("Target %s register_observer error %d\n",
176 obd_uuid2str(tgt_uuid), rc);
181 if (imp->imp_invalid) {
182 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
183 "disabled\n", obd_uuid2str(tgt_uuid));
187 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
188 &lov_osc_uuid, data, NULL);
189 if (rc || !lov->lov_tgts[index]->ltd_exp) {
190 CERROR("Target %s connect error %d\n",
191 obd_uuid2str(tgt_uuid), rc);
195 lov->lov_tgts[index]->ltd_reap = 0;
197 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
198 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
201 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
203 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
204 cfs_proc_dir_entry_t *osc_symlink;
206 LASSERT(osc_obd != NULL);
207 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
208 LASSERT(osc_obd->obd_type->typ_name != NULL);
210 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
213 osc_obd->obd_type->typ_name,
215 if (osc_symlink == NULL) {
216 CERROR("could not register LOV target "
217 "/proc/fs/lustre/%s/%s/target_obds/%s.",
218 obd->obd_type->typ_name, obd->obd_name,
220 lprocfs_remove(&lov_proc_dir);
228 static int lov_connect(const struct lu_env *env,
229 struct obd_export **exp, struct obd_device *obd,
230 struct obd_uuid *cluuid, struct obd_connect_data *data,
233 struct lov_obd *lov = &obd->u.lov;
234 struct lov_tgt_desc *tgt;
235 struct lustre_handle conn;
239 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
241 rc = class_connect(&conn, obd, cluuid);
245 *exp = class_conn2export(&conn);
247 /* Why should there ever be more than 1 connect? */
249 LASSERT(lov->lov_connects == 1);
251 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
253 lov->lov_ocd = *data;
256 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
257 tgt = lov->lov_tgts[i];
258 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
260 /* Flags will be lowest common denominator */
261 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
263 CERROR("%s: lov connect tgt %d failed: %d\n",
264 obd->obd_name, i, rc);
267 /* connect to administrative disabled ost */
268 if (!lov->lov_tgts[i]->ltd_exp)
271 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
272 OBD_NOTIFY_CONNECT, (void *)&i);
274 CERROR("%s error sending notify %d\n",
283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
285 cfs_proc_dir_entry_t *lov_proc_dir;
286 struct lov_obd *lov = &obd->u.lov;
287 struct obd_device *osc_obd;
291 osc_obd = class_exp2obd(tgt->ltd_exp);
292 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
293 obd->obd_name, osc_obd->obd_name);
295 if (tgt->ltd_active) {
297 lov->desc.ld_active_tgt_count--;
298 tgt->ltd_exp->exp_obd->obd_inactive = 1;
301 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
303 cfs_proc_dir_entry_t *osc_symlink;
305 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
307 lprocfs_remove(&osc_symlink);
309 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
310 obd->obd_type->typ_name, obd->obd_name,
316 /* Pass it on to our clients.
317 * XXX This should be an argument to disconnect,
318 * XXX not a back-door flag on the OBD. Ah well.
320 osc_obd->obd_force = obd->obd_force;
321 osc_obd->obd_fail = obd->obd_fail;
322 osc_obd->obd_no_recov = obd->obd_no_recov;
325 obd_register_observer(osc_obd, NULL);
327 rc = obd_disconnect(tgt->ltd_exp);
329 CERROR("Target %s disconnect error %d\n",
330 tgt->ltd_uuid.uuid, rc);
338 static int lov_disconnect(struct obd_export *exp)
340 struct obd_device *obd = class_exp2obd(exp);
341 struct lov_obd *lov = &obd->u.lov;
348 /* Only disconnect the underlying layers on the final disconnect. */
350 if (lov->lov_connects != 0) {
351 /* why should there be more than 1 connect? */
352 CERROR("disconnect #%d\n", lov->lov_connects);
356 /* Let's hold another reference so lov_del_obd doesn't spin through
360 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
361 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
362 /* Disconnection is the last we know about an obd */
363 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
369 rc = class_disconnect(exp); /* bz 9811 */
375 * -EINVAL : UUID can't be found in the LOV's target list
376 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
377 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
378 * any >= 0 : is log target index
380 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
381 enum obd_notify_event ev)
383 struct lov_obd *lov = &obd->u.lov;
384 struct lov_tgt_desc *tgt;
385 int index, activate, active;
388 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
389 lov, uuid->uuid, ev);
392 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
393 tgt = lov->lov_tgts[index];
397 * LU-642, initially inactive OSC could miss the obd_connect,
398 * we make up for it here.
400 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
401 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
402 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
404 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
405 &lov_osc_uuid, &lov->lov_ocd, NULL);
410 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
411 index, obd_uuid2str(&tgt->ltd_uuid),
412 tgt->ltd_exp->exp_handle.h_cookie);
413 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
417 if (index == lov->desc.ld_tgt_count)
418 GOTO(out, index = -EINVAL);
420 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
421 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
423 if (lov->lov_tgts[index]->ltd_activate == activate) {
424 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
425 uuid->uuid, activate ? "" : "de");
427 lov->lov_tgts[index]->ltd_activate = activate;
428 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
429 activate ? "" : "de", obd_uuid2str(uuid));
432 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
433 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
435 if (lov->lov_tgts[index]->ltd_active == active) {
436 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
437 uuid->uuid, active ? "" : "in");
440 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
441 obd_uuid2str(uuid), active ? "" : "in");
444 lov->lov_tgts[index]->ltd_active = active;
446 lov->desc.ld_active_tgt_count++;
447 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
449 lov->desc.ld_active_tgt_count--;
450 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
453 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
461 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
462 enum obd_notify_event ev, void *data)
465 struct lov_obd *lov = &obd->u.lov;
468 down_read(&lov->lov_notify_lock);
469 if (!lov->lov_connects) {
470 up_read(&lov->lov_notify_lock);
474 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
475 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
476 struct obd_uuid *uuid;
480 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
481 up_read(&lov->lov_notify_lock);
482 CERROR("unexpected notification of %s %s!\n",
483 watched->obd_type->typ_name,
487 uuid = &watched->u.cli.cl_target_uuid;
489 /* Set OSC as active before notifying the observer, so the
490 * observer can use the OSC normally.
492 rc = lov_set_osc_active(obd, uuid, ev);
494 up_read(&lov->lov_notify_lock);
495 CERROR("event(%d) of %s failed: %d\n", ev,
496 obd_uuid2str(uuid), rc);
499 /* active event should be pass lov target index as data */
503 /* Pass the notification up the chain. */
505 rc = obd_notify_observer(obd, watched, ev, data);
507 /* NULL watched means all osc's in the lov (only for syncs) */
508 /* sync event should be send lov idx as data */
509 struct lov_obd *lov = &obd->u.lov;
513 is_sync = (ev == OBD_NOTIFY_SYNC) ||
514 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
517 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
518 if (!lov->lov_tgts[i])
521 /* don't send sync event if target not
522 * connected/activated */
523 if (is_sync && !lov->lov_tgts[i]->ltd_active)
526 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
529 CERROR("%s: notify %s of %s failed %d\n",
531 obd->obd_observer->obd_name,
532 lov->lov_tgts[i]->ltd_obd->obd_name,
539 up_read(&lov->lov_notify_lock);
543 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
544 __u32 index, int gen, int active)
546 struct lov_obd *lov = &obd->u.lov;
547 struct lov_tgt_desc *tgt;
548 struct obd_device *tgt_obd;
552 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
553 uuidp->uuid, index, gen, active);
556 CERROR("request to add OBD %s with invalid generation: %d\n",
561 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
566 mutex_lock(&lov->lov_lock);
568 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
569 tgt = lov->lov_tgts[index];
570 CERROR("UUID %s already assigned at LOV target index %d\n",
571 obd_uuid2str(&tgt->ltd_uuid), index);
572 mutex_unlock(&lov->lov_lock);
576 if (index >= lov->lov_tgt_size) {
577 /* We need to reallocate the lov target array. */
578 struct lov_tgt_desc **newtgts, **old = NULL;
579 __u32 newsize, oldsize = 0;
581 newsize = max(lov->lov_tgt_size, (__u32)2);
582 while (newsize < index + 1)
583 newsize = newsize << 1;
584 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
585 if (newtgts == NULL) {
586 mutex_unlock(&lov->lov_lock);
590 if (lov->lov_tgt_size) {
591 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
594 oldsize = lov->lov_tgt_size;
597 lov->lov_tgts = newtgts;
598 lov->lov_tgt_size = newsize;
601 OBD_FREE(old, sizeof(*old) * oldsize);
603 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
604 lov->lov_tgts, lov->lov_tgt_size);
609 mutex_unlock(&lov->lov_lock);
613 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
615 mutex_unlock(&lov->lov_lock);
620 tgt->ltd_uuid = *uuidp;
621 tgt->ltd_obd = tgt_obd;
622 /* XXX - add a sanity check on the generation number. */
624 tgt->ltd_index = index;
625 tgt->ltd_activate = active;
626 lov->lov_tgts[index] = tgt;
627 if (index >= lov->desc.ld_tgt_count)
628 lov->desc.ld_tgt_count = index + 1;
630 mutex_unlock(&lov->lov_lock);
632 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
633 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
635 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
637 if (lov->lov_connects == 0) {
638 /* lov_connect hasn't been called yet. We'll do the
639 lov_connect_obd on this target when that fn first runs,
640 because we don't know the connect flags yet. */
646 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
650 /* connect to administrative disabled ost */
654 if (lov->lov_cache != NULL) {
655 rc = obd_set_info_async(NULL, tgt->ltd_exp,
656 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
657 sizeof(struct cl_client_cache), lov->lov_cache,
663 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
664 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
669 CERROR("add failed (%d), deleting %s\n", rc,
670 obd_uuid2str(&tgt->ltd_uuid));
671 lov_del_target(obd, index, 0, 0);
677 /* Schedule a target for deletion */
678 int lov_del_target(struct obd_device *obd, __u32 index,
679 struct obd_uuid *uuidp, int gen)
681 struct lov_obd *lov = &obd->u.lov;
682 int count = lov->desc.ld_tgt_count;
686 if (index >= count) {
687 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
692 /* to make sure there's no ongoing lov_notify() now */
693 down_write(&lov->lov_notify_lock);
696 if (!lov->lov_tgts[index]) {
697 CERROR("LOV target at index %d is not setup.\n", index);
698 GOTO(out, rc = -EINVAL);
701 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
702 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
703 lov_uuid2str(lov, index), index,
704 obd_uuid2str(uuidp));
705 GOTO(out, rc = -EINVAL);
708 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
709 lov_uuid2str(lov, index), index,
710 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
711 lov->lov_tgts[index]->ltd_active);
713 lov->lov_tgts[index]->ltd_reap = 1;
714 lov->lov_death_row++;
715 /* we really delete it from obd_putref */
718 up_write(&lov->lov_notify_lock);
723 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
725 struct obd_device *osc_obd;
728 LASSERT(tgt->ltd_reap);
730 osc_obd = class_exp2obd(tgt->ltd_exp);
732 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
734 osc_obd ? osc_obd->obd_name : "<no obd>");
737 lov_disconnect_obd(obd, tgt);
741 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
742 do it ourselves. And we can't do it from lov_cleanup,
743 because we just lost our only reference to it. */
745 class_manual_cleanup(osc_obd);
748 void lov_fix_desc_stripe_size(__u64 *val)
750 if (*val < LOV_DEFAULT_STRIPE_SIZE) {
751 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
752 LOV_DEFAULT_STRIPE_SIZE);
753 *val = LOV_DEFAULT_STRIPE_SIZE;
754 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
755 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
756 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
758 *val, LOV_MIN_STRIPE_SIZE);
762 void lov_fix_desc_stripe_count(__u32 *val)
768 void lov_fix_desc_pattern(__u32 *val)
770 /* from lov_setstripe */
771 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
772 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
777 void lov_fix_desc_qos_maxage(__u32 *val)
781 *val = QOS_DEFAULT_MAXAGE;
784 void lov_fix_desc(struct lov_desc *desc)
786 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
787 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
788 lov_fix_desc_pattern(&desc->ld_pattern);
789 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
792 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
794 struct lprocfs_static_vars lvars = { 0 };
795 struct lov_desc *desc;
796 struct lov_obd *lov = &obd->u.lov;
800 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
801 CERROR("LOV setup requires a descriptor\n");
805 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
807 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
808 CERROR("descriptor size wrong: %d > %d\n",
809 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
813 if (desc->ld_magic != LOV_DESC_MAGIC) {
814 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
815 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
816 obd->obd_name, desc);
817 lustre_swab_lov_desc(desc);
819 CERROR("%s: Bad lov desc magic: %#x\n",
820 obd->obd_name, desc->ld_magic);
827 desc->ld_active_tgt_count = 0;
829 lov->lov_tgt_size = 0;
831 mutex_init(&lov->lov_lock);
832 cfs_atomic_set(&lov->lov_refcount, 0);
833 lov->lov_sp_me = LUSTRE_SP_CLI;
835 init_rwsem(&lov->lov_notify_lock);
837 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
839 HASH_POOLS_BKT_BITS, 0,
842 &pool_hash_operations,
844 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
845 lov->lov_pool_count = 0;
846 rc = lov_ost_pool_init(&lov->lov_packed, 0);
850 lprocfs_lov_init_vars(&lvars);
851 lprocfs_obd_setup(obd, lvars.obd_vars);
856 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
857 0444, &lov_proc_target_fops, obd);
859 CWARN("Error adding the target_obd file\n");
862 lov->lov_pool_proc_entry = lprocfs_register("pools",
872 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
875 struct lov_obd *lov = &obd->u.lov;
880 case OBD_CLEANUP_EARLY: {
882 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
883 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
885 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
890 case OBD_CLEANUP_EXPORTS:
891 rc = obd_llog_finish(obd, 0);
893 CERROR("failed to cleanup llogging subsystems\n");
899 static int lov_cleanup(struct obd_device *obd)
901 struct lov_obd *lov = &obd->u.lov;
902 cfs_list_t *pos, *tmp;
903 struct pool_desc *pool;
906 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
907 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
908 /* free pool structs */
909 CDEBUG(D_INFO, "delete pool %p\n", pool);
910 /* In the function below, .hs_keycmp resolves to
911 * pool_hashkey_keycmp() */
912 /* coverity[overrun-buffer-val] */
913 lov_pool_del(obd, pool->pool_name);
915 cfs_hash_putref(lov->lov_pools_hash_body);
916 lov_ost_pool_free(&lov->lov_packed);
918 lprocfs_obd_cleanup(obd);
922 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
923 if (!lov->lov_tgts[i])
926 /* Inactive targets may never have connected */
927 if (lov->lov_tgts[i]->ltd_active ||
928 cfs_atomic_read(&lov->lov_refcount))
929 /* We should never get here - these
930 should have been removed in the
932 CERROR("lov tgt %d not cleaned!"
933 " deathrow=%d, lovrc=%d\n",
934 i, lov->lov_death_row,
935 cfs_atomic_read(&lov->lov_refcount));
936 lov_del_target(obd, i, 0, 0);
939 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
941 lov->lov_tgt_size = 0;
946 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
947 __u32 *indexp, int *genp)
949 struct obd_uuid obd_uuid;
954 switch(cmd = lcfg->lcfg_command) {
955 case LCFG_LOV_ADD_OBD:
956 case LCFG_LOV_ADD_INA:
957 case LCFG_LOV_DEL_OBD: {
960 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
961 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
962 GOTO(out, rc = -EINVAL);
964 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
966 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
967 GOTO(out, rc = -EINVAL);
968 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
969 GOTO(out, rc = -EINVAL);
972 if (cmd == LCFG_LOV_ADD_OBD)
973 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
974 else if (cmd == LCFG_LOV_ADD_INA)
975 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
977 rc = lov_del_target(obd, index, &obd_uuid, gen);
981 struct lprocfs_static_vars lvars = { 0 };
982 struct lov_desc *desc = &(obd->u.lov.desc);
985 GOTO(out, rc = -EINVAL);
987 lprocfs_lov_init_vars(&lvars);
989 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1002 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1003 GOTO(out, rc = -EINVAL);
1011 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1012 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1014 struct lov_stripe_md *obj_mdp, *lsm;
1015 struct lov_obd *lov = &exp->exp_obd->u.lov;
1020 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1021 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1023 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1024 if (obj_mdp == NULL)
1027 ost_idx = src_oa->o_nlink;
1030 GOTO(out, rc = -EINVAL);
1031 if (ost_idx >= lov->desc.ld_tgt_count ||
1032 !lov->lov_tgts[ost_idx])
1033 GOTO(out, rc = -EINVAL);
1035 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1036 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1037 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1038 GOTO(out, rc = -EINVAL);
1042 if (i == lsm->lsm_stripe_count)
1043 GOTO(out, rc = -EINVAL);
1045 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1046 src_oa, &obj_mdp, oti);
1048 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1052 /* the LOV expects oa->o_id to be set to the LOV object id */
1053 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1054 struct obdo *src_oa, struct lov_stripe_md **ea,
1055 struct obd_trans_info *oti)
1057 struct lov_obd *lov;
1061 LASSERT(ea != NULL);
1065 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1066 src_oa->o_flags == OBD_FL_DELORPHAN) {
1067 /* should be used with LOV anymore */
1071 lov = &exp->exp_obd->u.lov;
1072 if (!lov->desc.ld_active_tgt_count)
1075 obd_getref(exp->exp_obd);
1076 /* Recreate a specific object id at the given OST index */
1077 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1078 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1079 rc = lov_recreate(exp, src_oa, ea, oti);
1082 obd_putref(exp->exp_obd);
1086 #define ASSERT_LSM_MAGIC(lsmp) \
1088 LASSERT((lsmp) != NULL); \
1089 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1090 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1091 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1094 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1095 struct obdo *oa, struct lov_stripe_md *lsm,
1096 struct obd_trans_info *oti, struct obd_export *md_exp,
1099 struct lov_request_set *set;
1100 struct obd_info oinfo;
1101 struct lov_request *req;
1103 struct lov_obd *lov;
1104 int rc = 0, err = 0;
1107 ASSERT_LSM_MAGIC(lsm);
1109 if (!exp || !exp->exp_obd)
1112 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1114 LASSERT(oti->oti_logcookies);
1117 lov = &exp->exp_obd->u.lov;
1118 obd_getref(exp->exp_obd);
1119 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1123 cfs_list_for_each (pos, &set->set_list) {
1124 req = cfs_list_entry(pos, struct lov_request, rq_link);
1126 if (oa->o_valid & OBD_MD_FLCOOKIE)
1127 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1129 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1130 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1131 err = lov_update_common_set(set, req, err);
1133 CERROR("error: destroying objid "LPX64" subobj "
1134 LPX64" on OST idx %d: rc = %d\n",
1135 oa->o_id, req->rq_oi.oi_oa->o_id,
1143 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1144 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1146 err = lov_fini_destroy_set(set);
1148 obd_putref(exp->exp_obd);
1149 RETURN(rc ? rc : err);
1152 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1153 struct obd_info *oinfo)
1155 struct lov_request_set *set;
1156 struct lov_request *req;
1158 struct lov_obd *lov;
1159 int err = 0, rc = 0;
1163 ASSERT_LSM_MAGIC(oinfo->oi_md);
1165 if (!exp || !exp->exp_obd)
1168 lov = &exp->exp_obd->u.lov;
1170 rc = lov_prep_getattr_set(exp, oinfo, &set);
1174 cfs_list_for_each (pos, &set->set_list) {
1175 req = cfs_list_entry(pos, struct lov_request, rq_link);
1177 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1178 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1179 req->rq_oi.oi_oa->o_id, req->rq_idx);
1181 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1183 err = lov_update_common_set(set, req, rc);
1185 CERROR("error: getattr objid "LPX64" subobj "
1186 LPX64" on OST idx %d: rc = %d\n",
1187 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1193 rc = lov_fini_getattr_set(set);
1199 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1202 struct lov_request_set *lovset = (struct lov_request_set *)data;
1206 /* don't do attribute merge if this aysnc op failed */
1208 cfs_atomic_set(&lovset->set_completes, 0);
1209 err = lov_fini_getattr_set(lovset);
1210 RETURN(rc ? rc : err);
1213 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1214 struct ptlrpc_request_set *rqset)
1216 struct lov_request_set *lovset;
1217 struct lov_obd *lov;
1219 struct lov_request *req;
1224 ASSERT_LSM_MAGIC(oinfo->oi_md);
1226 if (!exp || !exp->exp_obd)
1229 lov = &exp->exp_obd->u.lov;
1231 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1235 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1236 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1237 oinfo->oi_md->lsm_stripe_size);
1239 cfs_list_for_each (pos, &lovset->set_list) {
1240 req = cfs_list_entry(pos, struct lov_request, rq_link);
1242 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1243 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1244 req->rq_oi.oi_oa->o_id, req->rq_idx);
1245 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1246 &req->rq_oi, rqset);
1248 CERROR("error: getattr objid "LPX64" subobj "
1249 LPX64" on OST idx %d: rc = %d\n",
1250 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1256 if (!cfs_list_empty(&rqset->set_requests)) {
1258 LASSERT (rqset->set_interpret == NULL);
1259 rqset->set_interpret = lov_getattr_interpret;
1260 rqset->set_arg = (void *)lovset;
1265 cfs_atomic_set(&lovset->set_completes, 0);
1266 err = lov_fini_getattr_set(lovset);
1267 RETURN(rc ? rc : err);
1270 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1271 struct obd_info *oinfo, struct obd_trans_info *oti)
1273 struct lov_request_set *set;
1274 struct lov_obd *lov;
1276 struct lov_request *req;
1277 int err = 0, rc = 0;
1281 ASSERT_LSM_MAGIC(oinfo->oi_md);
1283 if (!exp || !exp->exp_obd)
1286 /* for now, we only expect the following updates here */
1287 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1288 OBD_MD_FLMODE | OBD_MD_FLATIME |
1289 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1290 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1291 OBD_MD_FLGROUP | OBD_MD_FLUID |
1292 OBD_MD_FLGID | OBD_MD_FLFID |
1294 lov = &exp->exp_obd->u.lov;
1295 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1299 cfs_list_for_each (pos, &set->set_list) {
1300 req = cfs_list_entry(pos, struct lov_request, rq_link);
1302 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1304 err = lov_update_setattr_set(set, req, rc);
1306 CERROR("error: setattr objid "LPX64" subobj "
1307 LPX64" on OST idx %d: rc = %d\n",
1308 set->set_oi->oi_oa->o_id,
1309 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1314 err = lov_fini_setattr_set(set);
1320 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1323 struct lov_request_set *lovset = (struct lov_request_set *)data;
1328 cfs_atomic_set(&lovset->set_completes, 0);
1329 err = lov_fini_setattr_set(lovset);
1330 RETURN(rc ? rc : err);
1333 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1334 needed. Otherwise, a client is waiting for responses. */
1335 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1336 struct obd_trans_info *oti,
1337 struct ptlrpc_request_set *rqset)
1339 struct lov_request_set *set;
1340 struct lov_request *req;
1342 struct lov_obd *lov;
1347 ASSERT_LSM_MAGIC(oinfo->oi_md);
1348 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1350 LASSERT(oti->oti_logcookies);
1353 if (!exp || !exp->exp_obd)
1356 lov = &exp->exp_obd->u.lov;
1357 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1361 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1362 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1363 oinfo->oi_md->lsm_stripe_size);
1365 cfs_list_for_each (pos, &set->set_list) {
1366 req = cfs_list_entry(pos, struct lov_request, rq_link);
1368 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1369 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1371 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1372 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1373 req->rq_oi.oi_oa->o_id, req->rq_idx);
1375 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1376 &req->rq_oi, oti, rqset);
1378 CERROR("error: setattr objid "LPX64" subobj "
1379 LPX64" on OST idx %d: rc = %d\n",
1380 set->set_oi->oi_oa->o_id,
1381 req->rq_oi.oi_oa->o_id,
1387 /* If we are not waiting for responses on async requests, return. */
1388 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1391 cfs_atomic_set(&set->set_completes, 0);
1392 err = lov_fini_setattr_set(set);
1393 RETURN(rc ? rc : err);
1396 LASSERT(rqset->set_interpret == NULL);
1397 rqset->set_interpret = lov_setattr_interpret;
1398 rqset->set_arg = (void *)set;
1403 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1406 struct lov_request_set *lovset = (struct lov_request_set *)data;
1411 cfs_atomic_set(&lovset->set_completes, 0);
1412 err = lov_fini_punch_set(lovset);
1413 RETURN(rc ? rc : err);
1416 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1417 * we can send this 'punch' to just the authoritative node and the nodes
1418 * that the punch will affect. */
1419 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1420 struct obd_info *oinfo, struct obd_trans_info *oti,
1421 struct ptlrpc_request_set *rqset)
1423 struct lov_request_set *set;
1424 struct lov_obd *lov;
1426 struct lov_request *req;
1431 ASSERT_LSM_MAGIC(oinfo->oi_md);
1433 if (!exp || !exp->exp_obd)
1436 lov = &exp->exp_obd->u.lov;
1437 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1441 cfs_list_for_each (pos, &set->set_list) {
1442 req = cfs_list_entry(pos, struct lov_request, rq_link);
1444 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1445 &req->rq_oi, NULL, rqset);
1447 CERROR("error: punch objid "LPX64" subobj "LPX64
1448 " on OST idx %d: rc = %d\n",
1449 set->set_oi->oi_oa->o_id,
1450 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1455 if (rc || cfs_list_empty(&rqset->set_requests)) {
1457 err = lov_fini_punch_set(set);
1458 RETURN(rc ? rc : err);
1461 LASSERT(rqset->set_interpret == NULL);
1462 rqset->set_interpret = lov_punch_interpret;
1463 rqset->set_arg = (void *)set;
1468 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1471 struct lov_request_set *lovset = data;
1476 cfs_atomic_set(&lovset->set_completes, 0);
1477 err = lov_fini_sync_set(lovset);
1481 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1482 struct obd_info *oinfo, obd_off start, obd_off end,
1483 struct ptlrpc_request_set *rqset)
1485 struct lov_request_set *set = NULL;
1486 struct lov_obd *lov;
1488 struct lov_request *req;
1492 ASSERT_LSM_MAGIC(oinfo->oi_md);
1493 LASSERT(rqset != NULL);
1498 lov = &exp->exp_obd->u.lov;
1499 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1503 CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1504 set->set_oi->oi_oa->o_id, start, end);
1506 cfs_list_for_each (pos, &set->set_list) {
1507 req = cfs_list_entry(pos, struct lov_request, rq_link);
1509 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1510 &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1511 req->rq_oi.oi_policy.l_extent.end, rqset);
1513 CERROR("error: fsync objid "LPX64" subobj "LPX64
1514 " on OST idx %d: rc = %d\n",
1515 set->set_oi->oi_oa->o_id,
1516 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1521 /* If we are not waiting for responses on async requests, return. */
1522 if (rc || cfs_list_empty(&rqset->set_requests)) {
1523 int err = lov_fini_sync_set(set);
1528 LASSERT(rqset->set_interpret == NULL);
1529 rqset->set_interpret = lov_sync_interpret;
1530 rqset->set_arg = (void *)set;
1535 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1536 obd_count oa_bufs, struct brw_page *pga)
1538 struct obd_info oinfo = { { { 0 } } };
1541 oinfo.oi_oa = lov_oinfo->oi_oa;
1543 /* The caller just wants to know if there's a chance that this
1544 * I/O can succeed */
1545 for (i = 0; i < oa_bufs; i++) {
1546 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1547 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1550 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1551 pga[i].off + pga[i].count - 1,
1555 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1556 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1560 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1568 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1569 obd_count oa_bufs, struct brw_page *pga,
1570 struct obd_trans_info *oti)
1572 struct lov_request_set *set;
1573 struct lov_request *req;
1575 struct lov_obd *lov = &exp->exp_obd->u.lov;
1579 ASSERT_LSM_MAGIC(oinfo->oi_md);
1581 if (cmd == OBD_BRW_CHECK) {
1582 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1586 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1590 cfs_list_for_each (pos, &set->set_list) {
1591 struct obd_export *sub_exp;
1592 struct brw_page *sub_pga;
1593 req = cfs_list_entry(pos, struct lov_request, rq_link);
1595 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1596 sub_pga = set->set_pga + req->rq_pgaidx;
1597 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1601 lov_update_common_set(set, req, rc);
1604 err = lov_fini_brw_set(set);
1610 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1613 struct lov_request_set *lovset = (struct lov_request_set *)data;
1615 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1619 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1620 struct ldlm_enqueue_info *einfo,
1621 struct ptlrpc_request_set *rqset)
1623 ldlm_mode_t mode = einfo->ei_mode;
1624 struct lov_request_set *set;
1625 struct lov_request *req;
1627 struct lov_obd *lov;
1632 ASSERT_LSM_MAGIC(oinfo->oi_md);
1633 LASSERT(mode == (mode & -mode));
1635 /* we should never be asked to replay a lock this way. */
1636 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1638 if (!exp || !exp->exp_obd)
1641 lov = &exp->exp_obd->u.lov;
1642 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1646 cfs_list_for_each (pos, &set->set_list) {
1647 req = cfs_list_entry(pos, struct lov_request, rq_link);
1649 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1650 &req->rq_oi, einfo, rqset);
1655 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1657 LASSERT(rqset->set_interpret == NULL);
1658 rqset->set_interpret = lov_enqueue_interpret;
1659 rqset->set_arg = (void *)set;
1663 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1667 static int lov_change_cbdata(struct obd_export *exp,
1668 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1671 struct lov_obd *lov;
1675 ASSERT_LSM_MAGIC(lsm);
1677 if (!exp || !exp->exp_obd)
1680 lov = &exp->exp_obd->u.lov;
1681 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1682 struct lov_stripe_md submd;
1683 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1685 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1686 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1690 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1691 submd.lsm_object_id = loi->loi_id;
1692 submd.lsm_object_seq = loi->loi_seq;
1693 submd.lsm_stripe_count = 0;
1694 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1700 /* find any ldlm lock of the inode in lov
1704 static int lov_find_cbdata(struct obd_export *exp,
1705 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1708 struct lov_obd *lov;
1712 ASSERT_LSM_MAGIC(lsm);
1714 if (!exp || !exp->exp_obd)
1717 lov = &exp->exp_obd->u.lov;
1718 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1719 struct lov_stripe_md submd;
1720 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1722 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1723 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1727 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1728 submd.lsm_object_id = loi->loi_id;
1729 submd.lsm_object_seq = loi->loi_seq;
1730 submd.lsm_stripe_count = 0;
1731 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1739 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1740 __u32 mode, struct lustre_handle *lockh)
1742 struct lov_request_set *set;
1743 struct obd_info oinfo;
1744 struct lov_request *req;
1746 struct lov_obd *lov;
1747 struct lustre_handle *lov_lockhp;
1748 int err = 0, rc = 0;
1751 ASSERT_LSM_MAGIC(lsm);
1753 if (!exp || !exp->exp_obd)
1756 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1758 lov = &exp->exp_obd->u.lov;
1759 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1763 cfs_list_for_each (pos, &set->set_list) {
1764 req = cfs_list_entry(pos, struct lov_request, rq_link);
1765 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1767 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1768 req->rq_oi.oi_md, mode, lov_lockhp);
1769 rc = lov_update_common_set(set, req, rc);
1771 CERROR("error: cancel objid "LPX64" subobj "
1772 LPX64" on OST idx %d: rc = %d\n",
1774 req->rq_oi.oi_md->lsm_object_id,
1780 lov_fini_cancel_set(set);
1784 static int lov_cancel_unused(struct obd_export *exp,
1785 struct lov_stripe_md *lsm,
1786 ldlm_cancel_flags_t flags, void *opaque)
1788 struct lov_obd *lov;
1792 if (!exp || !exp->exp_obd)
1795 lov = &exp->exp_obd->u.lov;
1797 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1799 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1802 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1810 ASSERT_LSM_MAGIC(lsm);
1812 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1813 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1814 struct lov_stripe_md submd;
1815 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1818 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1819 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1823 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1824 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1826 submd.lsm_object_id = loi->loi_id;
1827 submd.lsm_object_seq = loi->loi_seq;
1828 submd.lsm_stripe_count = 0;
1829 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1830 &submd, flags, opaque);
1831 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1832 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1833 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1834 loi->loi_id, loi->loi_ost_idx, err);
1842 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1844 struct lov_request_set *lovset = (struct lov_request_set *)data;
1849 cfs_atomic_set(&lovset->set_completes, 0);
1851 err = lov_fini_statfs_set(lovset);
1852 RETURN(rc ? rc : err);
1855 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1856 __u64 max_age, struct ptlrpc_request_set *rqset)
1858 struct obd_device *obd = class_exp2obd(exp);
1859 struct lov_request_set *set;
1860 struct lov_request *req;
1862 struct lov_obd *lov;
1866 LASSERT(oinfo != NULL);
1867 LASSERT(oinfo->oi_osfs != NULL);
1870 rc = lov_prep_statfs_set(obd, oinfo, &set);
1874 cfs_list_for_each (pos, &set->set_list) {
1875 req = cfs_list_entry(pos, struct lov_request, rq_link);
1876 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1877 &req->rq_oi, max_age, rqset);
1882 if (rc || cfs_list_empty(&rqset->set_requests)) {
1885 cfs_atomic_set(&set->set_completes, 0);
1886 err = lov_fini_statfs_set(set);
1887 RETURN(rc ? rc : err);
1890 LASSERT(rqset->set_interpret == NULL);
1891 rqset->set_interpret = lov_statfs_interpret;
1892 rqset->set_arg = (void *)set;
1896 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1897 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1899 struct ptlrpc_request_set *set = NULL;
1900 struct obd_info oinfo = { { { 0 } } };
1905 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1906 * statfs requests */
1907 set = ptlrpc_prep_set();
1911 oinfo.oi_osfs = osfs;
1912 oinfo.oi_flags = flags;
1913 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1915 rc = ptlrpc_set_wait(set);
1916 ptlrpc_set_destroy(set);
1921 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1922 void *karg, void *uarg)
1924 struct obd_device *obddev = class_exp2obd(exp);
1925 struct lov_obd *lov = &obddev->u.lov;
1926 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1927 struct obd_uuid *uuidp;
1931 case IOC_OBD_STATFS: {
1932 struct obd_ioctl_data *data = karg;
1933 struct obd_device *osc_obd;
1934 struct obd_statfs stat_buf = {0};
1938 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1939 if ((index >= count))
1942 if (!lov->lov_tgts[index])
1943 /* Try again with the next index */
1945 if (!lov->lov_tgts[index]->ltd_active)
1948 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1953 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1954 min((int) data->ioc_plen2,
1955 (int) sizeof(struct obd_uuid))))
1958 flags = uarg ? *(__u32*)uarg : 0;
1959 /* got statfs data */
1960 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1961 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1965 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1966 min((int) data->ioc_plen1,
1967 (int) sizeof(stat_buf))))
1971 case OBD_IOC_LOV_GET_CONFIG: {
1972 struct obd_ioctl_data *data;
1973 struct lov_desc *desc;
1978 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1981 data = (struct obd_ioctl_data *)buf;
1983 if (sizeof(*desc) > data->ioc_inllen1) {
1984 obd_ioctl_freedata(buf, len);
1988 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1989 obd_ioctl_freedata(buf, len);
1993 if (sizeof(__u32) * count > data->ioc_inllen3) {
1994 obd_ioctl_freedata(buf, len);
1998 desc = (struct lov_desc *)data->ioc_inlbuf1;
1999 memcpy(desc, &(lov->desc), sizeof(*desc));
2001 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2002 genp = (__u32 *)data->ioc_inlbuf3;
2003 /* the uuid will be empty for deleted OSTs */
2004 for (i = 0; i < count; i++, uuidp++, genp++) {
2005 if (!lov->lov_tgts[i])
2007 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2008 *genp = lov->lov_tgts[i]->ltd_gen;
2011 if (cfs_copy_to_user((void *)uarg, buf, len))
2013 obd_ioctl_freedata(buf, len);
2016 case LL_IOC_LOV_SETSTRIPE:
2017 rc = lov_setstripe(exp, len, karg, uarg);
2019 case LL_IOC_LOV_GETSTRIPE:
2020 rc = lov_getstripe(exp, karg, uarg);
2022 case LL_IOC_LOV_SETEA:
2023 rc = lov_setea(exp, karg, uarg);
2025 case OBD_IOC_QUOTACTL: {
2026 struct if_quotactl *qctl = karg;
2027 struct lov_tgt_desc *tgt = NULL;
2028 struct obd_quotactl *oqctl;
2030 if (qctl->qc_valid == QC_OSTIDX) {
2031 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2034 tgt = lov->lov_tgts[qctl->qc_idx];
2035 if (!tgt || !tgt->ltd_exp)
2037 } else if (qctl->qc_valid == QC_UUID) {
2038 for (i = 0; i < count; i++) {
2039 tgt = lov->lov_tgts[i];
2041 !obd_uuid_equals(&tgt->ltd_uuid,
2045 if (tgt->ltd_exp == NULL)
2057 LASSERT(tgt && tgt->ltd_exp);
2058 OBD_ALLOC_PTR(oqctl);
2062 QCTL_COPY(oqctl, qctl);
2063 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2065 QCTL_COPY(qctl, oqctl);
2066 qctl->qc_valid = QC_OSTIDX;
2067 qctl->obd_uuid = tgt->ltd_uuid;
2069 OBD_FREE_PTR(oqctl);
2078 for (i = 0; i < count; i++) {
2080 struct obd_device *osc_obd;
2082 /* OST was disconnected */
2083 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2086 /* ll_umount_begin() sets force flag but for lov, not
2087 * osc. Let's pass it through */
2088 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2089 osc_obd->obd_force = obddev->obd_force;
2090 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2092 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2095 if (lov->lov_tgts[i]->ltd_active) {
2096 CDEBUG(err == -ENOTTY ?
2097 D_IOCTL : D_WARNING,
2098 "iocontrol OSC %s on OST "
2099 "idx %d cmd %x: err = %d\n",
2100 lov_uuid2str(lov, i),
2117 #define FIEMAP_BUFFER_SIZE 4096
2120 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2121 * call. The local end offset and the device are sent in the first
2122 * fm_extent. This function calculates the stripe number from the index.
2123 * This function returns a stripe_no on which mapping is to be restarted.
2125 * This function returns fm_end_offset which is the in-OST offset at which
2126 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2127 * will re-calculate proper offset in next stripe.
2128 * Note that the first extent is passed to lov_get_info via the value field.
2130 * \param fiemap fiemap request header
2131 * \param lsm striping information for the file
2132 * \param fm_start logical start of mapping
2133 * \param fm_end logical end of mapping
2134 * \param start_stripe starting stripe will be returned in this
2136 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2137 struct lov_stripe_md *lsm, obd_size fm_start,
2138 obd_size fm_end, int *start_stripe)
2140 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2141 obd_off lun_start, lun_end;
2142 obd_size fm_end_offset;
2143 int stripe_no = -1, i;
2145 if (fiemap->fm_extent_count == 0 ||
2146 fiemap->fm_extents[0].fe_logical == 0)
2149 /* Find out stripe_no from ost_index saved in the fe_device */
2150 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2151 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2152 fiemap->fm_extents[0].fe_device) {
2157 if (stripe_no == -1)
2160 /* If we have finished mapping on previous device, shift logical
2161 * offset to start of next device */
2162 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2163 &lun_start, &lun_end)) != 0 &&
2164 local_end < lun_end) {
2165 fm_end_offset = local_end;
2166 *start_stripe = stripe_no;
2168 /* This is a special value to indicate that caller should
2169 * calculate offset in next stripe. */
2171 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2174 return fm_end_offset;
2178 * We calculate on which OST the mapping will end. If the length of mapping
2179 * is greater than (stripe_size * stripe_count) then the last_stripe will
2180 * will be one just before start_stripe. Else we check if the mapping
2181 * intersects each OST and find last_stripe.
2182 * This function returns the last_stripe and also sets the stripe_count
2183 * over which the mapping is spread
2185 * \param lsm striping information for the file
2186 * \param fm_start logical start of mapping
2187 * \param fm_end logical end of mapping
2188 * \param start_stripe starting stripe of the mapping
2189 * \param stripe_count the number of stripes across which to map is returned
2191 * \retval last_stripe return the last stripe of the mapping
2193 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2194 obd_size fm_end, int start_stripe,
2198 obd_off obd_start, obd_end;
2201 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2202 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2204 *stripe_count = lsm->lsm_stripe_count;
2206 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2207 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2208 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2209 &obd_start, &obd_end)) == 0)
2213 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2220 * Set fe_device and copy extents from local buffer into main return buffer.
2222 * \param fiemap fiemap request header
2223 * \param lcl_fm_ext array of local fiemap extents to be copied
2224 * \param ost_index OST index to be written into the fm_device field for each
2226 * \param ext_count number of extents to be copied
2227 * \param current_extent where to start copying in main extent array
2229 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2230 struct ll_fiemap_extent *lcl_fm_ext,
2231 int ost_index, unsigned int ext_count,
2237 for (ext = 0; ext < ext_count; ext++) {
2238 lcl_fm_ext[ext].fe_device = ost_index;
2239 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2242 /* Copy fm_extent's from fm_local to return buffer */
2243 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2244 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2248 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2249 * This also handles the restarting of FIEMAP calls in case mapping overflows
2250 * the available number of extents in single call.
2252 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2253 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2255 struct ll_fiemap_info_key *fm_key = key;
2256 struct ll_user_fiemap *fiemap = val;
2257 struct ll_user_fiemap *fm_local = NULL;
2258 struct ll_fiemap_extent *lcl_fm_ext;
2260 unsigned int get_num_extents = 0;
2261 int ost_index = 0, actual_start_stripe, start_stripe;
2262 obd_size fm_start, fm_end, fm_length, fm_end_offset;
2264 int current_extent = 0, rc = 0, i;
2265 int ost_eof = 0; /* EOF for object */
2266 int ost_done = 0; /* done with required mapping for this OST? */
2268 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2269 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2274 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2275 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2277 OBD_ALLOC_LARGE(fm_local, buffer_size);
2278 if (fm_local == NULL)
2279 GOTO(out, rc = -ENOMEM);
2280 lcl_fm_ext = &fm_local->fm_extents[0];
2282 count_local = fiemap_size_to_count(buffer_size);
2284 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2285 fm_start = fiemap->fm_start;
2286 fm_length = fiemap->fm_length;
2287 /* Calculate start stripe, last stripe and length of mapping */
2288 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2289 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2290 fm_start + fm_length - 1);
2291 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2292 if (fm_end > fm_key->oa.o_size)
2293 fm_end = fm_key->oa.o_size;
2295 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2296 actual_start_stripe, &stripe_count);
2298 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2299 fm_end, &start_stripe);
2300 if (fm_end_offset == -EINVAL)
2301 GOTO(out, rc = -EINVAL);
2303 if (fiemap->fm_extent_count == 0) {
2304 get_num_extents = 1;
2308 /* Check each stripe */
2309 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2310 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2311 obd_size req_fm_len; /* Stores length of required mapping */
2312 obd_size len_mapped_single_call;
2313 obd_off lun_start, lun_end, obd_object_end;
2314 unsigned int ext_count;
2316 cur_stripe_wrap = cur_stripe;
2318 /* Find out range of mapping on this stripe */
2319 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2320 &lun_start, &obd_object_end)) == 0)
2323 /* If this is a continuation FIEMAP call and we are on
2324 * starting stripe then lun_start needs to be set to
2326 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2327 lun_start = fm_end_offset;
2329 if (fm_length != ~0ULL) {
2330 /* Handle fm_start + fm_length overflow */
2331 if (fm_start + fm_length < fm_start)
2332 fm_length = ~0ULL - fm_start;
2333 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2339 if (lun_start == lun_end)
2342 req_fm_len = obd_object_end - lun_start;
2343 fm_local->fm_length = 0;
2344 len_mapped_single_call = 0;
2346 /* If the output buffer is very large and the objects have many
2347 * extents we may need to loop on a single OST repeatedly */
2351 if (get_num_extents == 0) {
2352 /* Don't get too many extents. */
2353 if (current_extent + count_local >
2354 fiemap->fm_extent_count)
2355 count_local = fiemap->fm_extent_count -
2359 lun_start += len_mapped_single_call;
2360 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2361 req_fm_len = fm_local->fm_length;
2362 fm_local->fm_extent_count = count_local;
2363 fm_local->fm_mapped_extents = 0;
2364 fm_local->fm_flags = fiemap->fm_flags;
2366 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2367 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2368 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2370 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2371 GOTO(out, rc = -EINVAL);
2373 /* If OST is inactive, return extent with UNKNOWN flag */
2374 if (!lov->lov_tgts[ost_index]->ltd_active) {
2375 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2376 fm_local->fm_mapped_extents = 1;
2378 lcl_fm_ext[0].fe_logical = lun_start;
2379 lcl_fm_ext[0].fe_length = obd_object_end -
2381 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2386 fm_local->fm_start = lun_start;
2387 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2388 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2389 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2390 rc = obd_get_info(NULL,
2391 lov->lov_tgts[ost_index]->ltd_exp,
2392 keylen, key, vallen, fm_local, lsm);
2397 ext_count = fm_local->fm_mapped_extents;
2398 if (ext_count == 0) {
2400 /* If last stripe has hole at the end,
2401 * then we need to return */
2402 if (cur_stripe_wrap == last_stripe) {
2403 fiemap->fm_mapped_extents = 0;
2409 /* If we just need num of extents then go to next device */
2410 if (get_num_extents) {
2411 current_extent += ext_count;
2415 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2416 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2418 /* Have we finished mapping on this device? */
2419 if (req_fm_len <= len_mapped_single_call)
2422 /* Clear the EXTENT_LAST flag which can be present on
2424 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2425 lcl_fm_ext[ext_count - 1].fe_flags &=
2426 ~FIEMAP_EXTENT_LAST;
2428 curr_loc = lov_stripe_size(lsm,
2429 lcl_fm_ext[ext_count - 1].fe_logical+
2430 lcl_fm_ext[ext_count - 1].fe_length,
2432 if (curr_loc >= fm_key->oa.o_size)
2435 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2436 ost_index, ext_count,
2439 current_extent += ext_count;
2441 /* Ran out of available extents? */
2442 if (current_extent >= fiemap->fm_extent_count)
2444 } while (ost_done == 0 && ost_eof == 0);
2446 if (cur_stripe_wrap == last_stripe)
2451 /* Indicate that we are returning device offsets unless file just has
2453 if (lsm->lsm_stripe_count > 1)
2454 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2456 if (get_num_extents)
2457 goto skip_last_device_calc;
2459 /* Check if we have reached the last stripe and whether mapping for that
2460 * stripe is done. */
2461 if (cur_stripe_wrap == last_stripe) {
2462 if (ost_done || ost_eof)
2463 fiemap->fm_extents[current_extent - 1].fe_flags |=
2467 skip_last_device_calc:
2468 fiemap->fm_mapped_extents = current_extent;
2471 OBD_FREE_LARGE(fm_local, buffer_size);
2475 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2476 __u32 keylen, void *key, __u32 *vallen, void *val,
2477 struct lov_stripe_md *lsm)
2479 struct obd_device *obddev = class_exp2obd(exp);
2480 struct lov_obd *lov = &obddev->u.lov;
2484 if (!vallen || !val)
2489 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2492 struct ldlm_lock *lock;
2494 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2495 struct lov_oinfo *loi;
2496 __u32 *stripe = val;
2498 if (*vallen < sizeof(*stripe))
2499 GOTO(out, rc = -EFAULT);
2500 *vallen = sizeof(*stripe);
2502 /* XXX This is another one of those bits that will need to
2503 * change if we ever actually support nested LOVs. It uses
2504 * the lock's export to find out which stripe it is. */
2505 /* XXX - it's assumed all the locks for deleted OSTs have
2506 * been cancelled. Also, the export for deleted OSTs will
2507 * be NULL and won't match the lock's export. */
2508 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2509 loi = lsm->lsm_oinfo[i];
2510 if (!lov->lov_tgts[loi->loi_ost_idx])
2512 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2513 data->lock->l_conn_export &&
2514 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2519 LDLM_ERROR(data->lock, "lock on inode without such object");
2520 dump_lsm(D_ERROR, lsm);
2521 GOTO(out, rc = -ENXIO);
2522 } else if (KEY_IS(KEY_LAST_ID)) {
2523 struct obd_id_info *info = val;
2524 __u32 size = sizeof(obd_id);
2525 struct lov_tgt_desc *tgt;
2527 LASSERT(*vallen == sizeof(struct obd_id_info));
2528 tgt = lov->lov_tgts[info->idx];
2530 if (!tgt || !tgt->ltd_active)
2531 GOTO(out, rc = -ESRCH);
2533 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2534 &size, info->data, NULL);
2536 } else if (KEY_IS(KEY_LOVDESC)) {
2537 struct lov_desc *desc_ret = val;
2538 *desc_ret = lov->desc;
2541 } else if (KEY_IS(KEY_FIEMAP)) {
2542 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2544 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2545 struct lov_tgt_desc *tgt;
2546 __u64 ost_idx = *((__u64*)val);
2548 LASSERT(*vallen == sizeof(__u64));
2549 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2550 tgt = lov->lov_tgts[ost_idx];
2552 if (!tgt || !tgt->ltd_exp)
2553 GOTO(out, rc = -ESRCH);
2555 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2557 } else if (KEY_IS(KEY_TGT_COUNT)) {
2558 *((int *)val) = lov->desc.ld_tgt_count;
2569 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2570 obd_count keylen, void *key, obd_count vallen,
2571 void *val, struct ptlrpc_request_set *set)
2573 struct obd_device *obddev = class_exp2obd(exp);
2574 struct lov_obd *lov = &obddev->u.lov;
2577 struct lov_tgt_desc *tgt;
2578 unsigned incr, check_uuid,
2579 do_inactive, no_set;
2580 unsigned next_id = 0, mds_con = 0, capa = 0;
2583 incr = check_uuid = do_inactive = no_set = 0;
2586 set = ptlrpc_prep_set();
2592 count = lov->desc.ld_tgt_count;
2594 if (KEY_IS(KEY_NEXT_ID)) {
2595 count = vallen / sizeof(struct obd_id_info);
2596 vallen = sizeof(obd_id);
2597 incr = sizeof(struct obd_id_info);
2600 } else if (KEY_IS(KEY_CHECKSUM)) {
2602 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2603 /* use defaults: do_inactive = incr = 0; */
2604 } else if (KEY_IS(KEY_MDS_CONN)) {
2606 } else if (KEY_IS(KEY_CAPA_KEY)) {
2608 } else if (KEY_IS(KEY_CACHE_SET)) {
2609 LASSERT(lov->lov_cache == NULL);
2610 lov->lov_cache = val;
2614 for (i = 0; i < count; i++, val = (char *)val + incr) {
2616 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2618 tgt = lov->lov_tgts[i];
2620 /* OST was disconnected */
2621 if (!tgt || !tgt->ltd_exp)
2624 /* OST is inactive and we don't want inactive OSCs */
2625 if (!tgt->ltd_active && !do_inactive)
2629 struct mds_group_info *mgi;
2631 LASSERT(vallen == sizeof(*mgi));
2632 mgi = (struct mds_group_info *)val;
2634 /* Only want a specific OSC */
2635 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2639 err = obd_set_info_async(env, tgt->ltd_exp,
2640 keylen, key, sizeof(int),
2642 } else if (next_id) {
2643 err = obd_set_info_async(env, tgt->ltd_exp,
2644 keylen, key, vallen,
2645 ((struct obd_id_info*)val)->data, set);
2647 struct mds_capa_info *info = (struct mds_capa_info*)val;
2649 LASSERT(vallen == sizeof(*info));
2651 /* Only want a specific OSC */
2653 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2656 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2657 key, sizeof(*info->capa),
2660 /* Only want a specific OSC */
2662 !obd_uuid_equals(val, &tgt->ltd_uuid))
2665 err = obd_set_info_async(env, tgt->ltd_exp,
2666 keylen, key, vallen, val, set);
2675 err = ptlrpc_set_wait(set);
2678 ptlrpc_set_destroy(set);
2683 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2684 int cmd, __u64 *offset)
2686 __u32 ssize = lsm->lsm_stripe_size;
2690 lov_do_div64(start, ssize);
2691 start = start * ssize;
2693 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2694 ", end "LPU64"\n", *offset, ssize, start,
2696 if (cmd == OBD_CALC_STRIPE_END) {
2697 *offset = start + ssize - 1;
2698 } else if (cmd == OBD_CALC_STRIPE_START) {
2707 void lov_stripe_lock(struct lov_stripe_md *md)
2709 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2710 spin_lock(&md->lsm_lock);
2711 LASSERT(md->lsm_lock_owner == 0);
2712 md->lsm_lock_owner = cfs_curproc_pid();
2714 EXPORT_SYMBOL(lov_stripe_lock);
2716 void lov_stripe_unlock(struct lov_stripe_md *md)
2718 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2719 md->lsm_lock_owner = 0;
2720 spin_unlock(&md->lsm_lock);
2722 EXPORT_SYMBOL(lov_stripe_unlock);
2724 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2725 struct obd_quotactl *oqctl)
2727 struct lov_obd *lov = &obd->u.lov;
2728 struct lov_tgt_desc *tgt;
2730 __u64 bhardlimit = 0;
2734 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2735 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2736 oqctl->qc_cmd != Q_GETOQUOTA &&
2737 oqctl->qc_cmd != Q_INITQUOTA &&
2738 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2739 oqctl->qc_cmd != Q_FINVALIDATE) {
2740 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2746 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2749 tgt = lov->lov_tgts[i];
2754 if (!tgt->ltd_active || tgt->ltd_reap) {
2755 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2756 lov->lov_tgts[i]->ltd_activate) {
2758 CERROR("ost %d is inactive\n", i);
2760 CDEBUG(D_HA, "ost %d is inactive\n", i);
2765 err = obd_quotactl(tgt->ltd_exp, oqctl);
2767 if (tgt->ltd_active && !rc)
2772 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2773 curspace += oqctl->qc_dqblk.dqb_curspace;
2774 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2779 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2780 oqctl->qc_dqblk.dqb_curspace = curspace;
2781 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2786 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2787 struct obd_quotactl *oqctl)
2789 struct lov_obd *lov = &obd->u.lov;
2795 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2796 if (!lov->lov_tgts[i])
2799 /* Skip quota check on the administratively disabled OSTs. */
2800 if (!lov->lov_tgts[i]->ltd_activate) {
2801 CWARN("lov idx %d was administratively disabled, "
2802 "skip quotacheck on it.\n", i);
2806 if (!lov->lov_tgts[i]->ltd_active) {
2807 CERROR("lov idx %d inactive\n", i);
2813 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2816 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2819 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2830 struct obd_ops lov_obd_ops = {
2831 .o_owner = THIS_MODULE,
2832 .o_setup = lov_setup,
2833 .o_precleanup = lov_precleanup,
2834 .o_cleanup = lov_cleanup,
2835 //.o_process_config = lov_process_config,
2836 .o_connect = lov_connect,
2837 .o_disconnect = lov_disconnect,
2838 .o_statfs = lov_statfs,
2839 .o_statfs_async = lov_statfs_async,
2840 .o_packmd = lov_packmd,
2841 .o_unpackmd = lov_unpackmd,
2842 .o_create = lov_create,
2843 .o_destroy = lov_destroy,
2844 .o_getattr = lov_getattr,
2845 .o_getattr_async = lov_getattr_async,
2846 .o_setattr = lov_setattr,
2847 .o_setattr_async = lov_setattr_async,
2849 .o_merge_lvb = lov_merge_lvb,
2850 .o_adjust_kms = lov_adjust_kms,
2851 .o_punch = lov_punch,
2853 .o_enqueue = lov_enqueue,
2854 .o_change_cbdata = lov_change_cbdata,
2855 .o_find_cbdata = lov_find_cbdata,
2856 .o_cancel = lov_cancel,
2857 .o_cancel_unused = lov_cancel_unused,
2858 .o_iocontrol = lov_iocontrol,
2859 .o_get_info = lov_get_info,
2860 .o_set_info_async = lov_set_info_async,
2861 .o_extent_calc = lov_extent_calc,
2862 .o_llog_init = lov_llog_init,
2863 .o_llog_finish = lov_llog_finish,
2864 .o_notify = lov_notify,
2865 .o_pool_new = lov_pool_new,
2866 .o_pool_rem = lov_pool_remove,
2867 .o_pool_add = lov_pool_add,
2868 .o_pool_del = lov_pool_del,
2869 .o_getref = lov_getref,
2870 .o_putref = lov_putref,
2871 .o_quotactl = lov_quotactl,
2872 .o_quotacheck = lov_quotacheck,
2875 cfs_mem_cache_t *lov_oinfo_slab;
2877 extern struct lu_kmem_descr lov_caches[];
2879 int __init lov_init(void)
2881 struct lprocfs_static_vars lvars = { 0 };
2885 /* print an address of _any_ initialized kernel symbol from this
2886 * module, to allow debugging with gdb that doesn't support data
2887 * symbols from modules.*/
2888 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2890 rc = lu_kmem_init(lov_caches);
2894 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2895 sizeof(struct lov_oinfo),
2896 0, CFS_SLAB_HWCACHE_ALIGN);
2897 if (lov_oinfo_slab == NULL) {
2898 lu_kmem_fini(lov_caches);
2901 lprocfs_lov_init_vars(&lvars);
2903 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2904 LUSTRE_LOV_NAME, &lov_device_type);
2907 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2909 lu_kmem_fini(lov_caches);
2916 static void /*__exit*/ lov_exit(void)
2920 class_unregister_type(LUSTRE_LOV_NAME);
2921 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2924 lu_kmem_fini(lov_caches);
2927 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2928 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2929 MODULE_LICENSE("GPL");
2931 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);