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, 2013, 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_MIN_STRIPE_SIZE) {
752 LCONSOLE_INFO("Increasing default stripe size to "
754 LOV_DEFAULT_STRIPE_SIZE);
755 *val = LOV_DEFAULT_STRIPE_SIZE;
756 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
757 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
758 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
760 *val, LOV_MIN_STRIPE_SIZE);
764 void lov_fix_desc_stripe_count(__u32 *val)
770 void lov_fix_desc_pattern(__u32 *val)
772 /* from lov_setstripe */
773 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
774 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
779 void lov_fix_desc_qos_maxage(__u32 *val)
783 *val = QOS_DEFAULT_MAXAGE;
786 void lov_fix_desc(struct lov_desc *desc)
788 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
789 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
790 lov_fix_desc_pattern(&desc->ld_pattern);
791 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
794 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
796 struct lprocfs_static_vars lvars = { 0 };
797 struct lov_desc *desc;
798 struct lov_obd *lov = &obd->u.lov;
802 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
803 CERROR("LOV setup requires a descriptor\n");
807 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
809 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
810 CERROR("descriptor size wrong: %d > %d\n",
811 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
815 if (desc->ld_magic != LOV_DESC_MAGIC) {
816 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
817 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
818 obd->obd_name, desc);
819 lustre_swab_lov_desc(desc);
821 CERROR("%s: Bad lov desc magic: %#x\n",
822 obd->obd_name, desc->ld_magic);
829 desc->ld_active_tgt_count = 0;
831 lov->lov_tgt_size = 0;
833 mutex_init(&lov->lov_lock);
834 cfs_atomic_set(&lov->lov_refcount, 0);
835 lov->lov_sp_me = LUSTRE_SP_CLI;
837 init_rwsem(&lov->lov_notify_lock);
839 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
841 HASH_POOLS_BKT_BITS, 0,
844 &pool_hash_operations,
846 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
847 lov->lov_pool_count = 0;
848 rc = lov_ost_pool_init(&lov->lov_packed, 0);
852 lprocfs_lov_init_vars(&lvars);
853 lprocfs_obd_setup(obd, lvars.obd_vars);
858 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
859 0444, &lov_proc_target_fops, obd);
861 CWARN("Error adding the target_obd file\n");
864 lov->lov_pool_proc_entry = lprocfs_register("pools",
874 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
877 struct lov_obd *lov = &obd->u.lov;
882 case OBD_CLEANUP_EARLY: {
884 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
885 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
887 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
892 case OBD_CLEANUP_EXPORTS:
893 rc = obd_llog_finish(obd, 0);
895 CERROR("failed to cleanup llogging subsystems\n");
901 static int lov_cleanup(struct obd_device *obd)
903 struct lov_obd *lov = &obd->u.lov;
904 cfs_list_t *pos, *tmp;
905 struct pool_desc *pool;
908 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
909 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
910 /* free pool structs */
911 CDEBUG(D_INFO, "delete pool %p\n", pool);
912 /* In the function below, .hs_keycmp resolves to
913 * pool_hashkey_keycmp() */
914 /* coverity[overrun-buffer-val] */
915 lov_pool_del(obd, pool->pool_name);
917 cfs_hash_putref(lov->lov_pools_hash_body);
918 lov_ost_pool_free(&lov->lov_packed);
920 lprocfs_obd_cleanup(obd);
924 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
925 if (!lov->lov_tgts[i])
928 /* Inactive targets may never have connected */
929 if (lov->lov_tgts[i]->ltd_active ||
930 cfs_atomic_read(&lov->lov_refcount))
931 /* We should never get here - these
932 should have been removed in the
934 CERROR("lov tgt %d not cleaned!"
935 " deathrow=%d, lovrc=%d\n",
936 i, lov->lov_death_row,
937 cfs_atomic_read(&lov->lov_refcount));
938 lov_del_target(obd, i, 0, 0);
941 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
943 lov->lov_tgt_size = 0;
948 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
949 __u32 *indexp, int *genp)
951 struct obd_uuid obd_uuid;
956 switch(cmd = lcfg->lcfg_command) {
957 case LCFG_LOV_ADD_OBD:
958 case LCFG_LOV_ADD_INA:
959 case LCFG_LOV_DEL_OBD: {
962 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
963 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
964 GOTO(out, rc = -EINVAL);
966 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
968 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
969 GOTO(out, rc = -EINVAL);
970 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
971 GOTO(out, rc = -EINVAL);
974 if (cmd == LCFG_LOV_ADD_OBD)
975 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
976 else if (cmd == LCFG_LOV_ADD_INA)
977 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
979 rc = lov_del_target(obd, index, &obd_uuid, gen);
983 struct lprocfs_static_vars lvars = { 0 };
984 struct lov_desc *desc = &(obd->u.lov.desc);
987 GOTO(out, rc = -EINVAL);
989 lprocfs_lov_init_vars(&lvars);
991 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1004 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1005 GOTO(out, rc = -EINVAL);
1013 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1014 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1016 struct lov_stripe_md *obj_mdp, *lsm;
1017 struct lov_obd *lov = &exp->exp_obd->u.lov;
1022 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1023 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1025 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1026 if (obj_mdp == NULL)
1029 ost_idx = src_oa->o_nlink;
1032 GOTO(out, rc = -EINVAL);
1033 if (ost_idx >= lov->desc.ld_tgt_count ||
1034 !lov->lov_tgts[ost_idx])
1035 GOTO(out, rc = -EINVAL);
1037 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1038 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1039 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1040 ostid_id(&src_oa->o_oi))
1041 GOTO(out, rc = -EINVAL);
1045 if (i == lsm->lsm_stripe_count)
1046 GOTO(out, rc = -EINVAL);
1048 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1049 src_oa, &obj_mdp, oti);
1051 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1055 /* the LOV expects oa->o_id to be set to the LOV object id */
1056 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1057 struct obdo *src_oa, struct lov_stripe_md **ea,
1058 struct obd_trans_info *oti)
1060 struct lov_obd *lov;
1064 LASSERT(ea != NULL);
1068 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1069 src_oa->o_flags == OBD_FL_DELORPHAN) {
1070 /* should be used with LOV anymore */
1074 lov = &exp->exp_obd->u.lov;
1075 if (!lov->desc.ld_active_tgt_count)
1078 obd_getref(exp->exp_obd);
1079 /* Recreate a specific object id at the given OST index */
1080 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1081 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1082 rc = lov_recreate(exp, src_oa, ea, oti);
1085 obd_putref(exp->exp_obd);
1089 #define ASSERT_LSM_MAGIC(lsmp) \
1091 LASSERT((lsmp) != NULL); \
1092 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1093 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1094 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1097 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1098 struct obdo *oa, struct lov_stripe_md *lsm,
1099 struct obd_trans_info *oti, struct obd_export *md_exp,
1102 struct lov_request_set *set;
1103 struct obd_info oinfo;
1104 struct lov_request *req;
1106 struct lov_obd *lov;
1107 int rc = 0, err = 0;
1110 ASSERT_LSM_MAGIC(lsm);
1112 if (!exp || !exp->exp_obd)
1115 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1117 LASSERT(oti->oti_logcookies);
1120 lov = &exp->exp_obd->u.lov;
1121 obd_getref(exp->exp_obd);
1122 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1126 cfs_list_for_each (pos, &set->set_list) {
1127 req = cfs_list_entry(pos, struct lov_request, rq_link);
1129 if (oa->o_valid & OBD_MD_FLCOOKIE)
1130 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1132 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1133 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1134 err = lov_update_common_set(set, req, err);
1136 CERROR("%s: destroying objid "DOSTID" subobj "
1137 DOSTID" on OST idx %d: rc = %d\n",
1138 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1139 POSTID(&req->rq_oi.oi_oa->o_oi),
1147 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1148 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1150 err = lov_fini_destroy_set(set);
1152 obd_putref(exp->exp_obd);
1153 RETURN(rc ? rc : err);
1156 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1157 struct obd_info *oinfo)
1159 struct lov_request_set *set;
1160 struct lov_request *req;
1162 struct lov_obd *lov;
1163 int err = 0, rc = 0;
1167 ASSERT_LSM_MAGIC(oinfo->oi_md);
1169 if (!exp || !exp->exp_obd)
1172 lov = &exp->exp_obd->u.lov;
1174 rc = lov_prep_getattr_set(exp, oinfo, &set);
1178 cfs_list_for_each (pos, &set->set_list) {
1179 req = cfs_list_entry(pos, struct lov_request, rq_link);
1181 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1182 " %u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1183 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1185 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1187 err = lov_update_common_set(set, req, rc);
1189 CERROR("%s: getattr objid "DOSTID" subobj "
1190 DOSTID" on OST idx %d: rc = %d\n",
1191 exp->exp_obd->obd_name,
1192 POSTID(&oinfo->oi_oa->o_oi),
1193 POSTID(&req->rq_oi.oi_oa->o_oi),
1199 rc = lov_fini_getattr_set(set);
1205 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1208 struct lov_request_set *lovset = (struct lov_request_set *)data;
1212 /* don't do attribute merge if this aysnc op failed */
1214 cfs_atomic_set(&lovset->set_completes, 0);
1215 err = lov_fini_getattr_set(lovset);
1216 RETURN(rc ? rc : err);
1219 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1220 struct ptlrpc_request_set *rqset)
1222 struct lov_request_set *lovset;
1223 struct lov_obd *lov;
1225 struct lov_request *req;
1230 ASSERT_LSM_MAGIC(oinfo->oi_md);
1232 if (!exp || !exp->exp_obd)
1235 lov = &exp->exp_obd->u.lov;
1237 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1241 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1242 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1243 oinfo->oi_md->lsm_stripe_size);
1245 cfs_list_for_each(pos, &lovset->set_list) {
1246 req = cfs_list_entry(pos, struct lov_request, rq_link);
1248 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1249 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1250 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1251 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1252 &req->rq_oi, rqset);
1254 CERROR("%s: getattr objid "DOSTID" subobj"
1255 DOSTID" on OST idx %d: rc = %d\n",
1256 exp->exp_obd->obd_name,
1257 POSTID(&oinfo->oi_oa->o_oi),
1258 POSTID(&req->rq_oi.oi_oa->o_oi),
1264 if (!cfs_list_empty(&rqset->set_requests)) {
1266 LASSERT (rqset->set_interpret == NULL);
1267 rqset->set_interpret = lov_getattr_interpret;
1268 rqset->set_arg = (void *)lovset;
1273 cfs_atomic_set(&lovset->set_completes, 0);
1274 err = lov_fini_getattr_set(lovset);
1275 RETURN(rc ? rc : err);
1278 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1279 struct obd_info *oinfo, struct obd_trans_info *oti)
1281 struct lov_request_set *set;
1282 struct lov_obd *lov;
1284 struct lov_request *req;
1285 int err = 0, rc = 0;
1289 ASSERT_LSM_MAGIC(oinfo->oi_md);
1291 if (!exp || !exp->exp_obd)
1294 /* for now, we only expect the following updates here */
1295 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1296 OBD_MD_FLMODE | OBD_MD_FLATIME |
1297 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1298 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1299 OBD_MD_FLGROUP | OBD_MD_FLUID |
1300 OBD_MD_FLGID | OBD_MD_FLFID |
1302 lov = &exp->exp_obd->u.lov;
1303 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1307 cfs_list_for_each (pos, &set->set_list) {
1308 req = cfs_list_entry(pos, struct lov_request, rq_link);
1310 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1312 err = lov_update_setattr_set(set, req, rc);
1314 CERROR("%s: setattr objid "DOSTID" subobj "
1315 DOSTID" on OST idx %d: rc = %d\n",
1316 exp->exp_obd->obd_name,
1317 POSTID(&set->set_oi->oi_oa->o_oi),
1318 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1324 err = lov_fini_setattr_set(set);
1330 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1333 struct lov_request_set *lovset = (struct lov_request_set *)data;
1338 cfs_atomic_set(&lovset->set_completes, 0);
1339 err = lov_fini_setattr_set(lovset);
1340 RETURN(rc ? rc : err);
1343 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1344 needed. Otherwise, a client is waiting for responses. */
1345 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1346 struct obd_trans_info *oti,
1347 struct ptlrpc_request_set *rqset)
1349 struct lov_request_set *set;
1350 struct lov_request *req;
1352 struct lov_obd *lov;
1357 ASSERT_LSM_MAGIC(oinfo->oi_md);
1358 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1360 LASSERT(oti->oti_logcookies);
1363 if (!exp || !exp->exp_obd)
1366 lov = &exp->exp_obd->u.lov;
1367 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1371 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1372 POSTID(&oinfo->oi_md->lsm_oi),
1373 oinfo->oi_md->lsm_stripe_count,
1374 oinfo->oi_md->lsm_stripe_size);
1376 cfs_list_for_each(pos, &set->set_list) {
1377 req = cfs_list_entry(pos, struct lov_request, rq_link);
1379 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1380 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1382 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1383 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1384 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1386 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1387 &req->rq_oi, oti, rqset);
1389 CERROR("error: setattr objid "DOSTID" subobj"
1390 DOSTID" on OST idx %d: rc = %d\n",
1391 POSTID(&set->set_oi->oi_oa->o_oi),
1392 POSTID(&req->rq_oi.oi_oa->o_oi),
1398 /* If we are not waiting for responses on async requests, return. */
1399 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1402 cfs_atomic_set(&set->set_completes, 0);
1403 err = lov_fini_setattr_set(set);
1404 RETURN(rc ? rc : err);
1407 LASSERT(rqset->set_interpret == NULL);
1408 rqset->set_interpret = lov_setattr_interpret;
1409 rqset->set_arg = (void *)set;
1414 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1417 struct lov_request_set *lovset = (struct lov_request_set *)data;
1422 cfs_atomic_set(&lovset->set_completes, 0);
1423 err = lov_fini_punch_set(lovset);
1424 RETURN(rc ? rc : err);
1427 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1428 * we can send this 'punch' to just the authoritative node and the nodes
1429 * that the punch will affect. */
1430 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1431 struct obd_info *oinfo, struct obd_trans_info *oti,
1432 struct ptlrpc_request_set *rqset)
1434 struct lov_request_set *set;
1435 struct lov_obd *lov;
1437 struct lov_request *req;
1442 ASSERT_LSM_MAGIC(oinfo->oi_md);
1444 if (!exp || !exp->exp_obd)
1447 lov = &exp->exp_obd->u.lov;
1448 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1452 cfs_list_for_each (pos, &set->set_list) {
1453 req = cfs_list_entry(pos, struct lov_request, rq_link);
1455 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1456 &req->rq_oi, NULL, rqset);
1458 CERROR("%s: punch objid "DOSTID" subobj "DOSTID
1459 " on OST idx %d: rc = %d\n",
1460 exp->exp_obd->obd_name,
1461 POSTID(&set->set_oi->oi_oa->o_oi),
1462 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, rc);
1467 if (rc || cfs_list_empty(&rqset->set_requests)) {
1469 err = lov_fini_punch_set(set);
1470 RETURN(rc ? rc : err);
1473 LASSERT(rqset->set_interpret == NULL);
1474 rqset->set_interpret = lov_punch_interpret;
1475 rqset->set_arg = (void *)set;
1480 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1483 struct lov_request_set *lovset = data;
1488 cfs_atomic_set(&lovset->set_completes, 0);
1489 err = lov_fini_sync_set(lovset);
1493 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1494 struct obd_info *oinfo, obd_off start, obd_off end,
1495 struct ptlrpc_request_set *rqset)
1497 struct lov_request_set *set = NULL;
1498 struct lov_obd *lov;
1500 struct lov_request *req;
1504 ASSERT_LSM_MAGIC(oinfo->oi_md);
1505 LASSERT(rqset != NULL);
1510 lov = &exp->exp_obd->u.lov;
1511 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1515 CDEBUG(D_INFO, "fsync objid "DOSTID" ["LPX64", "LPX64"]\n",
1516 POSTID(&set->set_oi->oi_oa->o_oi), start, end);
1518 cfs_list_for_each (pos, &set->set_list) {
1519 req = cfs_list_entry(pos, struct lov_request, rq_link);
1521 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1522 &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1523 req->rq_oi.oi_policy.l_extent.end, rqset);
1525 CERROR("%s: fsync objid "DOSTID" subobj "DOSTID
1526 " on OST idx %d: rc = %d\n",
1527 exp->exp_obd->obd_name,
1528 POSTID(&set->set_oi->oi_oa->o_oi),
1529 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1535 /* If we are not waiting for responses on async requests, return. */
1536 if (rc || cfs_list_empty(&rqset->set_requests)) {
1537 int err = lov_fini_sync_set(set);
1542 LASSERT(rqset->set_interpret == NULL);
1543 rqset->set_interpret = lov_sync_interpret;
1544 rqset->set_arg = (void *)set;
1549 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1550 obd_count oa_bufs, struct brw_page *pga)
1552 struct obd_info oinfo = { { { 0 } } };
1555 oinfo.oi_oa = lov_oinfo->oi_oa;
1557 /* The caller just wants to know if there's a chance that this
1558 * I/O can succeed */
1559 for (i = 0; i < oa_bufs; i++) {
1560 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1561 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1564 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1565 pga[i].off + pga[i].count - 1,
1569 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1570 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1574 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1582 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1583 obd_count oa_bufs, struct brw_page *pga,
1584 struct obd_trans_info *oti)
1586 struct lov_request_set *set;
1587 struct lov_request *req;
1589 struct lov_obd *lov = &exp->exp_obd->u.lov;
1593 ASSERT_LSM_MAGIC(oinfo->oi_md);
1595 if (cmd == OBD_BRW_CHECK) {
1596 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1600 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1604 cfs_list_for_each (pos, &set->set_list) {
1605 struct obd_export *sub_exp;
1606 struct brw_page *sub_pga;
1607 req = cfs_list_entry(pos, struct lov_request, rq_link);
1609 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1610 sub_pga = set->set_pga + req->rq_pgaidx;
1611 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1615 lov_update_common_set(set, req, rc);
1618 err = lov_fini_brw_set(set);
1624 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1627 struct lov_request_set *lovset = (struct lov_request_set *)data;
1629 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1633 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1634 struct ldlm_enqueue_info *einfo,
1635 struct ptlrpc_request_set *rqset)
1637 ldlm_mode_t mode = einfo->ei_mode;
1638 struct lov_request_set *set;
1639 struct lov_request *req;
1641 struct lov_obd *lov;
1646 ASSERT_LSM_MAGIC(oinfo->oi_md);
1647 LASSERT(mode == (mode & -mode));
1649 /* we should never be asked to replay a lock this way. */
1650 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1652 if (!exp || !exp->exp_obd)
1655 lov = &exp->exp_obd->u.lov;
1656 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1660 cfs_list_for_each (pos, &set->set_list) {
1661 req = cfs_list_entry(pos, struct lov_request, rq_link);
1663 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1664 &req->rq_oi, einfo, rqset);
1669 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1671 LASSERT(rqset->set_interpret == NULL);
1672 rqset->set_interpret = lov_enqueue_interpret;
1673 rqset->set_arg = (void *)set;
1677 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1681 static int lov_change_cbdata(struct obd_export *exp,
1682 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1685 struct lov_obd *lov;
1689 ASSERT_LSM_MAGIC(lsm);
1691 if (!exp || !exp->exp_obd)
1694 lov = &exp->exp_obd->u.lov;
1695 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1696 struct lov_stripe_md submd;
1697 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1699 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1700 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1704 submd.lsm_oi = loi->loi_oi;
1705 submd.lsm_stripe_count = 0;
1706 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1712 /* find any ldlm lock of the inode in lov
1716 static int lov_find_cbdata(struct obd_export *exp,
1717 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1720 struct lov_obd *lov;
1724 ASSERT_LSM_MAGIC(lsm);
1726 if (!exp || !exp->exp_obd)
1729 lov = &exp->exp_obd->u.lov;
1730 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1731 struct lov_stripe_md submd;
1732 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1734 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1735 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1738 submd.lsm_oi = loi->loi_oi;
1739 submd.lsm_stripe_count = 0;
1740 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1748 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1749 __u32 mode, struct lustre_handle *lockh)
1751 struct lov_request_set *set;
1752 struct obd_info oinfo;
1753 struct lov_request *req;
1755 struct lov_obd *lov;
1756 struct lustre_handle *lov_lockhp;
1757 int err = 0, rc = 0;
1760 ASSERT_LSM_MAGIC(lsm);
1762 if (!exp || !exp->exp_obd)
1766 lov = &exp->exp_obd->u.lov;
1767 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1771 cfs_list_for_each(pos, &set->set_list) {
1772 req = cfs_list_entry(pos, struct lov_request, rq_link);
1773 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1775 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1776 req->rq_oi.oi_md, mode, lov_lockhp);
1777 rc = lov_update_common_set(set, req, rc);
1779 CERROR("%s: cancel objid "DOSTID" subobj "
1780 DOSTID" on OST idx %d: rc = %d\n",
1781 exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1782 POSTID(&req->rq_oi.oi_md->lsm_oi),
1788 lov_fini_cancel_set(set);
1792 static int lov_cancel_unused(struct obd_export *exp,
1793 struct lov_stripe_md *lsm,
1794 ldlm_cancel_flags_t flags, void *opaque)
1796 struct lov_obd *lov;
1800 if (!exp || !exp->exp_obd)
1803 lov = &exp->exp_obd->u.lov;
1805 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1807 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1810 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1818 ASSERT_LSM_MAGIC(lsm);
1820 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1821 struct lov_stripe_md submd;
1822 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1823 int idx = loi->loi_ost_idx;
1826 if (!lov->lov_tgts[idx]) {
1827 CDEBUG(D_HA, "lov idx %d NULL\n", idx);
1831 if (!lov->lov_tgts[idx]->ltd_active)
1832 CDEBUG(D_HA, "lov idx %d inactive\n", idx);
1834 submd.lsm_oi = loi->loi_oi;
1835 submd.lsm_stripe_count = 0;
1836 err = obd_cancel_unused(lov->lov_tgts[idx]->ltd_exp,
1837 &submd, flags, opaque);
1838 if (err && lov->lov_tgts[idx]->ltd_active) {
1839 CERROR("%s: cancel unused objid "DOSTID
1840 " subobj "DOSTID" on OST idx %d: rc = %d\n",
1841 exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1842 POSTID(&loi->loi_oi), idx, err);
1850 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1852 struct lov_request_set *lovset = (struct lov_request_set *)data;
1857 cfs_atomic_set(&lovset->set_completes, 0);
1859 err = lov_fini_statfs_set(lovset);
1860 RETURN(rc ? rc : err);
1863 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1864 __u64 max_age, struct ptlrpc_request_set *rqset)
1866 struct obd_device *obd = class_exp2obd(exp);
1867 struct lov_request_set *set;
1868 struct lov_request *req;
1870 struct lov_obd *lov;
1874 LASSERT(oinfo != NULL);
1875 LASSERT(oinfo->oi_osfs != NULL);
1878 rc = lov_prep_statfs_set(obd, oinfo, &set);
1882 cfs_list_for_each (pos, &set->set_list) {
1883 req = cfs_list_entry(pos, struct lov_request, rq_link);
1884 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1885 &req->rq_oi, max_age, rqset);
1890 if (rc || cfs_list_empty(&rqset->set_requests)) {
1893 cfs_atomic_set(&set->set_completes, 0);
1894 err = lov_fini_statfs_set(set);
1895 RETURN(rc ? rc : err);
1898 LASSERT(rqset->set_interpret == NULL);
1899 rqset->set_interpret = lov_statfs_interpret;
1900 rqset->set_arg = (void *)set;
1904 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1905 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1907 struct ptlrpc_request_set *set = NULL;
1908 struct obd_info oinfo = { { { 0 } } };
1913 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1914 * statfs requests */
1915 set = ptlrpc_prep_set();
1919 oinfo.oi_osfs = osfs;
1920 oinfo.oi_flags = flags;
1921 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1923 rc = ptlrpc_set_wait(set);
1924 ptlrpc_set_destroy(set);
1929 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1930 void *karg, void *uarg)
1932 struct obd_device *obddev = class_exp2obd(exp);
1933 struct lov_obd *lov = &obddev->u.lov;
1934 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1935 struct obd_uuid *uuidp;
1939 case IOC_OBD_STATFS: {
1940 struct obd_ioctl_data *data = karg;
1941 struct obd_device *osc_obd;
1942 struct obd_statfs stat_buf = {0};
1946 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1947 if ((index >= count))
1950 if (!lov->lov_tgts[index])
1951 /* Try again with the next index */
1953 if (!lov->lov_tgts[index]->ltd_active)
1956 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1961 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1962 min((int)data->ioc_plen2,
1963 (int)sizeof(struct obd_uuid))))
1966 flags = uarg ? *(__u32*)uarg : 0;
1967 /* got statfs data */
1968 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1969 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1973 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1974 min((int) data->ioc_plen1,
1975 (int) sizeof(stat_buf))))
1979 case OBD_IOC_LOV_GET_CONFIG: {
1980 struct obd_ioctl_data *data;
1981 struct lov_desc *desc;
1986 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1989 data = (struct obd_ioctl_data *)buf;
1991 if (sizeof(*desc) > data->ioc_inllen1) {
1992 obd_ioctl_freedata(buf, len);
1996 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1997 obd_ioctl_freedata(buf, len);
2001 if (sizeof(__u32) * count > data->ioc_inllen3) {
2002 obd_ioctl_freedata(buf, len);
2006 desc = (struct lov_desc *)data->ioc_inlbuf1;
2007 memcpy(desc, &(lov->desc), sizeof(*desc));
2009 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2010 genp = (__u32 *)data->ioc_inlbuf3;
2011 /* the uuid will be empty for deleted OSTs */
2012 for (i = 0; i < count; i++, uuidp++, genp++) {
2013 if (!lov->lov_tgts[i])
2015 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2016 *genp = lov->lov_tgts[i]->ltd_gen;
2019 if (copy_to_user((void *)uarg, buf, len))
2021 obd_ioctl_freedata(buf, len);
2024 case LL_IOC_LOV_SETSTRIPE:
2025 rc = lov_setstripe(exp, len, karg, uarg);
2027 case LL_IOC_LOV_GETSTRIPE:
2028 rc = lov_getstripe(exp, karg, uarg);
2030 case LL_IOC_LOV_SETEA:
2031 rc = lov_setea(exp, karg, uarg);
2033 case OBD_IOC_QUOTACTL: {
2034 struct if_quotactl *qctl = karg;
2035 struct lov_tgt_desc *tgt = NULL;
2036 struct obd_quotactl *oqctl;
2038 if (qctl->qc_valid == QC_OSTIDX) {
2039 if (count <= qctl->qc_idx)
2042 tgt = lov->lov_tgts[qctl->qc_idx];
2043 if (!tgt || !tgt->ltd_exp)
2045 } else if (qctl->qc_valid == QC_UUID) {
2046 for (i = 0; i < count; i++) {
2047 tgt = lov->lov_tgts[i];
2049 !obd_uuid_equals(&tgt->ltd_uuid,
2053 if (tgt->ltd_exp == NULL)
2065 LASSERT(tgt && tgt->ltd_exp);
2066 OBD_ALLOC_PTR(oqctl);
2070 QCTL_COPY(oqctl, qctl);
2071 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2073 QCTL_COPY(qctl, oqctl);
2074 qctl->qc_valid = QC_OSTIDX;
2075 qctl->obd_uuid = tgt->ltd_uuid;
2077 OBD_FREE_PTR(oqctl);
2086 for (i = 0; i < count; i++) {
2088 struct obd_device *osc_obd;
2090 /* OST was disconnected */
2091 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2094 /* ll_umount_begin() sets force flag but for lov, not
2095 * osc. Let's pass it through */
2096 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2097 osc_obd->obd_force = obddev->obd_force;
2098 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2100 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2103 if (lov->lov_tgts[i]->ltd_active) {
2104 CDEBUG(err == -ENOTTY ?
2105 D_IOCTL : D_WARNING,
2106 "iocontrol OSC %s on OST "
2107 "idx %d cmd %x: err = %d\n",
2108 lov_uuid2str(lov, i),
2125 #define FIEMAP_BUFFER_SIZE 4096
2128 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2129 * call. The local end offset and the device are sent in the first
2130 * fm_extent. This function calculates the stripe number from the index.
2131 * This function returns a stripe_no on which mapping is to be restarted.
2133 * This function returns fm_end_offset which is the in-OST offset at which
2134 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2135 * will re-calculate proper offset in next stripe.
2136 * Note that the first extent is passed to lov_get_info via the value field.
2138 * \param fiemap fiemap request header
2139 * \param lsm striping information for the file
2140 * \param fm_start logical start of mapping
2141 * \param fm_end logical end of mapping
2142 * \param start_stripe starting stripe will be returned in this
2144 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2145 struct lov_stripe_md *lsm, obd_size fm_start,
2146 obd_size fm_end, int *start_stripe)
2148 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2149 obd_off lun_start, lun_end;
2150 obd_size fm_end_offset;
2151 int stripe_no = -1, i;
2153 if (fiemap->fm_extent_count == 0 ||
2154 fiemap->fm_extents[0].fe_logical == 0)
2157 /* Find out stripe_no from ost_index saved in the fe_device */
2158 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2159 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2160 fiemap->fm_extents[0].fe_device) {
2165 if (stripe_no == -1)
2168 /* If we have finished mapping on previous device, shift logical
2169 * offset to start of next device */
2170 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2171 &lun_start, &lun_end)) != 0 &&
2172 local_end < lun_end) {
2173 fm_end_offset = local_end;
2174 *start_stripe = stripe_no;
2176 /* This is a special value to indicate that caller should
2177 * calculate offset in next stripe. */
2179 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2182 return fm_end_offset;
2186 * We calculate on which OST the mapping will end. If the length of mapping
2187 * is greater than (stripe_size * stripe_count) then the last_stripe will
2188 * will be one just before start_stripe. Else we check if the mapping
2189 * intersects each OST and find last_stripe.
2190 * This function returns the last_stripe and also sets the stripe_count
2191 * over which the mapping is spread
2193 * \param lsm striping information for the file
2194 * \param fm_start logical start of mapping
2195 * \param fm_end logical end of mapping
2196 * \param start_stripe starting stripe of the mapping
2197 * \param stripe_count the number of stripes across which to map is returned
2199 * \retval last_stripe return the last stripe of the mapping
2201 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2202 obd_size fm_end, int start_stripe,
2206 obd_off obd_start, obd_end;
2209 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2210 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2212 *stripe_count = lsm->lsm_stripe_count;
2214 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2215 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2216 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2217 &obd_start, &obd_end)) == 0)
2221 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2228 * Set fe_device and copy extents from local buffer into main return buffer.
2230 * \param fiemap fiemap request header
2231 * \param lcl_fm_ext array of local fiemap extents to be copied
2232 * \param ost_index OST index to be written into the fm_device field for each
2234 * \param ext_count number of extents to be copied
2235 * \param current_extent where to start copying in main extent array
2237 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2238 struct ll_fiemap_extent *lcl_fm_ext,
2239 int ost_index, unsigned int ext_count,
2245 for (ext = 0; ext < ext_count; ext++) {
2246 lcl_fm_ext[ext].fe_device = ost_index;
2247 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2250 /* Copy fm_extent's from fm_local to return buffer */
2251 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2252 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2256 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2257 * This also handles the restarting of FIEMAP calls in case mapping overflows
2258 * the available number of extents in single call.
2260 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2261 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2263 struct ll_fiemap_info_key *fm_key = key;
2264 struct ll_user_fiemap *fiemap = val;
2265 struct ll_user_fiemap *fm_local = NULL;
2266 struct ll_fiemap_extent *lcl_fm_ext;
2268 unsigned int get_num_extents = 0;
2269 int ost_index = 0, actual_start_stripe, start_stripe;
2270 obd_size fm_start, fm_end, fm_length, fm_end_offset;
2272 int current_extent = 0, rc = 0, i;
2273 int ost_eof = 0; /* EOF for object */
2274 int ost_done = 0; /* done with required mapping for this OST? */
2276 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2277 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2279 if (!lsm_has_objects(lsm))
2282 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2283 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2285 OBD_ALLOC_LARGE(fm_local, buffer_size);
2286 if (fm_local == NULL)
2287 GOTO(out, rc = -ENOMEM);
2288 lcl_fm_ext = &fm_local->fm_extents[0];
2290 count_local = fiemap_size_to_count(buffer_size);
2292 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2293 fm_start = fiemap->fm_start;
2294 fm_length = fiemap->fm_length;
2295 /* Calculate start stripe, last stripe and length of mapping */
2296 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2297 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2298 fm_start + fm_length - 1);
2299 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2300 if (fm_end > fm_key->oa.o_size)
2301 fm_end = fm_key->oa.o_size;
2303 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2304 actual_start_stripe, &stripe_count);
2306 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2307 fm_end, &start_stripe);
2308 if (fm_end_offset == -EINVAL)
2309 GOTO(out, rc = -EINVAL);
2311 if (fiemap->fm_extent_count == 0) {
2312 get_num_extents = 1;
2316 /* Check each stripe */
2317 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2318 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2319 obd_size req_fm_len; /* Stores length of required mapping */
2320 obd_size len_mapped_single_call;
2321 obd_off lun_start, lun_end, obd_object_end;
2322 unsigned int ext_count;
2324 cur_stripe_wrap = cur_stripe;
2326 /* Find out range of mapping on this stripe */
2327 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2328 &lun_start, &obd_object_end)) == 0)
2331 /* If this is a continuation FIEMAP call and we are on
2332 * starting stripe then lun_start needs to be set to
2334 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2335 lun_start = fm_end_offset;
2337 if (fm_length != ~0ULL) {
2338 /* Handle fm_start + fm_length overflow */
2339 if (fm_start + fm_length < fm_start)
2340 fm_length = ~0ULL - fm_start;
2341 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2347 if (lun_start == lun_end)
2350 req_fm_len = obd_object_end - lun_start;
2351 fm_local->fm_length = 0;
2352 len_mapped_single_call = 0;
2354 /* If the output buffer is very large and the objects have many
2355 * extents we may need to loop on a single OST repeatedly */
2359 if (get_num_extents == 0) {
2360 /* Don't get too many extents. */
2361 if (current_extent + count_local >
2362 fiemap->fm_extent_count)
2363 count_local = fiemap->fm_extent_count -
2367 lun_start += len_mapped_single_call;
2368 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2369 req_fm_len = fm_local->fm_length;
2370 fm_local->fm_extent_count = count_local;
2371 fm_local->fm_mapped_extents = 0;
2372 fm_local->fm_flags = fiemap->fm_flags;
2374 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
2375 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2377 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2378 GOTO(out, rc = -EINVAL);
2380 /* If OST is inactive, return extent with UNKNOWN flag */
2381 if (!lov->lov_tgts[ost_index]->ltd_active) {
2382 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2383 fm_local->fm_mapped_extents = 1;
2385 lcl_fm_ext[0].fe_logical = lun_start;
2386 lcl_fm_ext[0].fe_length = obd_object_end -
2388 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2393 fm_local->fm_start = lun_start;
2394 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2395 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2396 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2397 rc = obd_get_info(NULL,
2398 lov->lov_tgts[ost_index]->ltd_exp,
2399 keylen, key, vallen, fm_local, lsm);
2404 ext_count = fm_local->fm_mapped_extents;
2405 if (ext_count == 0) {
2407 /* If last stripe has hole at the end,
2408 * then we need to return */
2409 if (cur_stripe_wrap == last_stripe) {
2410 fiemap->fm_mapped_extents = 0;
2416 /* If we just need num of extents then go to next device */
2417 if (get_num_extents) {
2418 current_extent += ext_count;
2422 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2423 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2425 /* Have we finished mapping on this device? */
2426 if (req_fm_len <= len_mapped_single_call)
2429 /* Clear the EXTENT_LAST flag which can be present on
2431 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2432 lcl_fm_ext[ext_count - 1].fe_flags &=
2433 ~FIEMAP_EXTENT_LAST;
2435 curr_loc = lov_stripe_size(lsm,
2436 lcl_fm_ext[ext_count - 1].fe_logical+
2437 lcl_fm_ext[ext_count - 1].fe_length,
2439 if (curr_loc >= fm_key->oa.o_size)
2442 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2443 ost_index, ext_count,
2446 current_extent += ext_count;
2448 /* Ran out of available extents? */
2449 if (current_extent >= fiemap->fm_extent_count)
2451 } while (ost_done == 0 && ost_eof == 0);
2453 if (cur_stripe_wrap == last_stripe)
2458 /* Indicate that we are returning device offsets unless file just has
2460 if (lsm->lsm_stripe_count > 1)
2461 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2463 if (get_num_extents)
2464 goto skip_last_device_calc;
2466 /* Check if we have reached the last stripe and whether mapping for that
2467 * stripe is done. */
2468 if (cur_stripe_wrap == last_stripe) {
2469 if (ost_done || ost_eof)
2470 fiemap->fm_extents[current_extent - 1].fe_flags |=
2474 skip_last_device_calc:
2475 fiemap->fm_mapped_extents = current_extent;
2478 OBD_FREE_LARGE(fm_local, buffer_size);
2482 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2483 __u32 keylen, void *key, __u32 *vallen, void *val,
2484 struct lov_stripe_md *lsm)
2486 struct obd_device *obddev = class_exp2obd(exp);
2487 struct lov_obd *lov = &obddev->u.lov;
2491 if (!vallen || !val)
2496 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2499 struct ldlm_lock *lock;
2501 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2502 struct lov_oinfo *loi;
2503 __u32 *stripe = val;
2505 if (*vallen < sizeof(*stripe))
2506 GOTO(out, rc = -EFAULT);
2507 *vallen = sizeof(*stripe);
2509 /* XXX This is another one of those bits that will need to
2510 * change if we ever actually support nested LOVs. It uses
2511 * the lock's export to find out which stripe it is. */
2512 /* XXX - it's assumed all the locks for deleted OSTs have
2513 * been cancelled. Also, the export for deleted OSTs will
2514 * be NULL and won't match the lock's export. */
2515 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2516 loi = lsm->lsm_oinfo[i];
2517 if (!lov->lov_tgts[loi->loi_ost_idx])
2519 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2520 data->lock->l_conn_export &&
2521 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2526 LDLM_ERROR(data->lock, "lock on inode without such object");
2527 dump_lsm(D_ERROR, lsm);
2528 GOTO(out, rc = -ENXIO);
2529 } else if (KEY_IS(KEY_LAST_ID)) {
2530 struct obd_id_info *info = val;
2531 __u32 size = sizeof(obd_id);
2532 struct lov_tgt_desc *tgt;
2534 LASSERT(*vallen == sizeof(struct obd_id_info));
2535 tgt = lov->lov_tgts[info->idx];
2537 if (!tgt || !tgt->ltd_active)
2538 GOTO(out, rc = -ESRCH);
2540 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2541 &size, info->data, NULL);
2543 } else if (KEY_IS(KEY_LOVDESC)) {
2544 struct lov_desc *desc_ret = val;
2545 *desc_ret = lov->desc;
2548 } else if (KEY_IS(KEY_FIEMAP)) {
2549 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2551 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2552 struct lov_tgt_desc *tgt;
2553 __u64 ost_idx = *((__u64*)val);
2555 LASSERT(*vallen == sizeof(__u64));
2556 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2557 tgt = lov->lov_tgts[ost_idx];
2559 if (!tgt || !tgt->ltd_exp)
2560 GOTO(out, rc = -ESRCH);
2562 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2564 } else if (KEY_IS(KEY_TGT_COUNT)) {
2565 *((int *)val) = lov->desc.ld_tgt_count;
2576 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2577 obd_count keylen, void *key, obd_count vallen,
2578 void *val, struct ptlrpc_request_set *set)
2580 struct obd_device *obddev = class_exp2obd(exp);
2581 struct lov_obd *lov = &obddev->u.lov;
2584 struct lov_tgt_desc *tgt;
2585 unsigned incr, check_uuid,
2586 do_inactive, no_set;
2587 unsigned next_id = 0, mds_con = 0, capa = 0;
2590 incr = check_uuid = do_inactive = no_set = 0;
2593 set = ptlrpc_prep_set();
2599 count = lov->desc.ld_tgt_count;
2601 if (KEY_IS(KEY_NEXT_ID)) {
2602 count = vallen / sizeof(struct obd_id_info);
2603 vallen = sizeof(obd_id);
2604 incr = sizeof(struct obd_id_info);
2607 } else if (KEY_IS(KEY_CHECKSUM)) {
2609 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2610 /* use defaults: do_inactive = incr = 0; */
2611 } else if (KEY_IS(KEY_MDS_CONN)) {
2613 } else if (KEY_IS(KEY_CAPA_KEY)) {
2615 } else if (KEY_IS(KEY_CACHE_SET)) {
2616 LASSERT(lov->lov_cache == NULL);
2617 lov->lov_cache = val;
2621 for (i = 0; i < count; i++, val = (char *)val + incr) {
2623 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2625 tgt = lov->lov_tgts[i];
2627 /* OST was disconnected */
2628 if (!tgt || !tgt->ltd_exp)
2631 /* OST is inactive and we don't want inactive OSCs */
2632 if (!tgt->ltd_active && !do_inactive)
2636 struct mds_group_info *mgi;
2638 LASSERT(vallen == sizeof(*mgi));
2639 mgi = (struct mds_group_info *)val;
2641 /* Only want a specific OSC */
2642 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2646 err = obd_set_info_async(env, tgt->ltd_exp,
2647 keylen, key, sizeof(int),
2649 } else if (next_id) {
2650 err = obd_set_info_async(env, tgt->ltd_exp,
2651 keylen, key, vallen,
2652 ((struct obd_id_info*)val)->data, set);
2654 struct mds_capa_info *info = (struct mds_capa_info*)val;
2656 LASSERT(vallen == sizeof(*info));
2658 /* Only want a specific OSC */
2660 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2663 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2664 key, sizeof(*info->capa),
2667 /* Only want a specific OSC */
2669 !obd_uuid_equals(val, &tgt->ltd_uuid))
2672 err = obd_set_info_async(env, tgt->ltd_exp,
2673 keylen, key, vallen, val, set);
2682 err = ptlrpc_set_wait(set);
2685 ptlrpc_set_destroy(set);
2690 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2691 int cmd, __u64 *offset)
2693 __u32 ssize = lsm->lsm_stripe_size;
2697 lov_do_div64(start, ssize);
2698 start = start * ssize;
2700 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2701 ", end "LPU64"\n", *offset, ssize, start,
2703 if (cmd == OBD_CALC_STRIPE_END) {
2704 *offset = start + ssize - 1;
2705 } else if (cmd == OBD_CALC_STRIPE_START) {
2714 void lov_stripe_lock(struct lov_stripe_md *md)
2716 LASSERT(md->lsm_lock_owner != current_pid());
2717 spin_lock(&md->lsm_lock);
2718 LASSERT(md->lsm_lock_owner == 0);
2719 md->lsm_lock_owner = current_pid();
2721 EXPORT_SYMBOL(lov_stripe_lock);
2723 void lov_stripe_unlock(struct lov_stripe_md *md)
2725 LASSERT(md->lsm_lock_owner == current_pid());
2726 md->lsm_lock_owner = 0;
2727 spin_unlock(&md->lsm_lock);
2729 EXPORT_SYMBOL(lov_stripe_unlock);
2731 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2732 struct obd_quotactl *oqctl)
2734 struct lov_obd *lov = &obd->u.lov;
2735 struct lov_tgt_desc *tgt;
2737 __u64 bhardlimit = 0;
2741 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2742 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2743 oqctl->qc_cmd != Q_GETOQUOTA &&
2744 oqctl->qc_cmd != Q_INITQUOTA &&
2745 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2746 oqctl->qc_cmd != Q_FINVALIDATE) {
2747 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2753 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2756 tgt = lov->lov_tgts[i];
2761 if (!tgt->ltd_active || tgt->ltd_reap) {
2762 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2763 lov->lov_tgts[i]->ltd_activate) {
2765 CERROR("ost %d is inactive\n", i);
2767 CDEBUG(D_HA, "ost %d is inactive\n", i);
2772 err = obd_quotactl(tgt->ltd_exp, oqctl);
2774 if (tgt->ltd_active && !rc)
2779 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2780 curspace += oqctl->qc_dqblk.dqb_curspace;
2781 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2786 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2787 oqctl->qc_dqblk.dqb_curspace = curspace;
2788 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2793 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2794 struct obd_quotactl *oqctl)
2796 struct lov_obd *lov = &obd->u.lov;
2802 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2803 if (!lov->lov_tgts[i])
2806 /* Skip quota check on the administratively disabled OSTs. */
2807 if (!lov->lov_tgts[i]->ltd_activate) {
2808 CWARN("lov idx %d was administratively disabled, "
2809 "skip quotacheck on it.\n", i);
2813 if (!lov->lov_tgts[i]->ltd_active) {
2814 CERROR("lov idx %d inactive\n", i);
2820 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2823 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2826 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2837 struct obd_ops lov_obd_ops = {
2838 .o_owner = THIS_MODULE,
2839 .o_setup = lov_setup,
2840 .o_precleanup = lov_precleanup,
2841 .o_cleanup = lov_cleanup,
2842 //.o_process_config = lov_process_config,
2843 .o_connect = lov_connect,
2844 .o_disconnect = lov_disconnect,
2845 .o_statfs = lov_statfs,
2846 .o_statfs_async = lov_statfs_async,
2847 .o_packmd = lov_packmd,
2848 .o_unpackmd = lov_unpackmd,
2849 .o_create = lov_create,
2850 .o_destroy = lov_destroy,
2851 .o_getattr = lov_getattr,
2852 .o_getattr_async = lov_getattr_async,
2853 .o_setattr = lov_setattr,
2854 .o_setattr_async = lov_setattr_async,
2856 .o_merge_lvb = lov_merge_lvb,
2857 .o_adjust_kms = lov_adjust_kms,
2858 .o_punch = lov_punch,
2860 .o_enqueue = lov_enqueue,
2861 .o_change_cbdata = lov_change_cbdata,
2862 .o_find_cbdata = lov_find_cbdata,
2863 .o_cancel = lov_cancel,
2864 .o_cancel_unused = lov_cancel_unused,
2865 .o_iocontrol = lov_iocontrol,
2866 .o_get_info = lov_get_info,
2867 .o_set_info_async = lov_set_info_async,
2868 .o_extent_calc = lov_extent_calc,
2869 .o_llog_init = lov_llog_init,
2870 .o_llog_finish = lov_llog_finish,
2871 .o_notify = lov_notify,
2872 .o_pool_new = lov_pool_new,
2873 .o_pool_rem = lov_pool_remove,
2874 .o_pool_add = lov_pool_add,
2875 .o_pool_del = lov_pool_del,
2876 .o_getref = lov_getref,
2877 .o_putref = lov_putref,
2878 .o_quotactl = lov_quotactl,
2879 .o_quotacheck = lov_quotacheck,
2882 struct kmem_cache *lov_oinfo_slab;
2884 extern struct lu_kmem_descr lov_caches[];
2886 int __init lov_init(void)
2888 struct lprocfs_static_vars lvars = { 0 };
2892 /* print an address of _any_ initialized kernel symbol from this
2893 * module, to allow debugging with gdb that doesn't support data
2894 * symbols from modules.*/
2895 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2897 rc = lu_kmem_init(lov_caches);
2901 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2902 sizeof(struct lov_oinfo), 0,
2903 SLAB_HWCACHE_ALIGN, NULL);
2904 if (lov_oinfo_slab == NULL) {
2905 lu_kmem_fini(lov_caches);
2908 lprocfs_lov_init_vars(&lvars);
2910 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2911 LUSTRE_LOV_NAME, &lov_device_type);
2914 kmem_cache_destroy(lov_oinfo_slab);
2915 lu_kmem_fini(lov_caches);
2922 static void /*__exit*/ lov_exit(void)
2924 class_unregister_type(LUSTRE_LOV_NAME);
2925 kmem_cache_destroy(lov_oinfo_slab);
2926 lu_kmem_fini(lov_caches);
2929 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2930 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2931 MODULE_LICENSE("GPL");
2933 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);