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>
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 struct proc_dir_entry *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 = obd->obd_proc_private;
203 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
204 struct proc_dir_entry *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);
221 obd->obd_proc_private = NULL;
229 static int lov_connect(const struct lu_env *env,
230 struct obd_export **exp, struct obd_device *obd,
231 struct obd_uuid *cluuid, struct obd_connect_data *data,
234 struct lov_obd *lov = &obd->u.lov;
235 struct lov_tgt_desc *tgt;
236 struct lustre_handle conn;
240 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
242 rc = class_connect(&conn, obd, cluuid);
246 *exp = class_conn2export(&conn);
248 /* Why should there ever be more than 1 connect? */
250 LASSERT(lov->lov_connects == 1);
252 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
254 lov->lov_ocd = *data;
257 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
258 tgt = lov->lov_tgts[i];
259 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
261 /* Flags will be lowest common denominator */
262 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
264 CERROR("%s: lov connect tgt %d failed: %d\n",
265 obd->obd_name, i, rc);
268 /* connect to administrative disabled ost */
269 if (!lov->lov_tgts[i]->ltd_exp)
272 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
273 OBD_NOTIFY_CONNECT, (void *)&i);
275 CERROR("%s error sending notify %d\n",
284 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
286 struct proc_dir_entry *lov_proc_dir;
287 struct lov_obd *lov = &obd->u.lov;
288 struct obd_device *osc_obd;
292 osc_obd = class_exp2obd(tgt->ltd_exp);
293 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
294 obd->obd_name, osc_obd->obd_name);
296 if (tgt->ltd_active) {
298 lov->desc.ld_active_tgt_count--;
299 tgt->ltd_exp->exp_obd->obd_inactive = 1;
302 lov_proc_dir = obd->obd_proc_private;
304 lprocfs_remove_proc_entry(osc_obd->obd_name, lov_proc_dir);
307 /* Pass it on to our clients.
308 * XXX This should be an argument to disconnect,
309 * XXX not a back-door flag on the OBD. Ah well.
311 osc_obd->obd_force = obd->obd_force;
312 osc_obd->obd_fail = obd->obd_fail;
313 osc_obd->obd_no_recov = obd->obd_no_recov;
316 obd_register_observer(osc_obd, NULL);
318 rc = obd_disconnect(tgt->ltd_exp);
320 CERROR("Target %s disconnect error %d\n",
321 tgt->ltd_uuid.uuid, rc);
329 static int lov_disconnect(struct obd_export *exp)
331 struct obd_device *obd = class_exp2obd(exp);
332 struct lov_obd *lov = &obd->u.lov;
339 /* Only disconnect the underlying layers on the final disconnect. */
341 if (lov->lov_connects != 0) {
342 /* why should there be more than 1 connect? */
343 CERROR("disconnect #%d\n", lov->lov_connects);
347 /* Let's hold another reference so lov_del_obd doesn't spin through
351 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
352 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
353 /* Disconnection is the last we know about an obd */
354 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
360 rc = class_disconnect(exp); /* bz 9811 */
366 * -EINVAL : UUID can't be found in the LOV's target list
367 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
368 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
369 * any >= 0 : is log target index
371 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
372 enum obd_notify_event ev)
374 struct lov_obd *lov = &obd->u.lov;
375 struct lov_tgt_desc *tgt;
376 int index, activate, active;
379 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
380 lov, uuid->uuid, ev);
383 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
384 tgt = lov->lov_tgts[index];
388 * LU-642, initially inactive OSC could miss the obd_connect,
389 * we make up for it here.
391 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
392 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
393 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
395 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
396 &lov_osc_uuid, &lov->lov_ocd, NULL);
401 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
402 index, obd_uuid2str(&tgt->ltd_uuid),
403 tgt->ltd_exp->exp_handle.h_cookie);
404 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
408 if (index == lov->desc.ld_tgt_count)
409 GOTO(out, index = -EINVAL);
411 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
412 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
414 if (lov->lov_tgts[index]->ltd_activate == activate) {
415 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
416 uuid->uuid, activate ? "" : "de");
418 lov->lov_tgts[index]->ltd_activate = activate;
419 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
420 activate ? "" : "de", obd_uuid2str(uuid));
423 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
424 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
426 if (lov->lov_tgts[index]->ltd_active == active) {
427 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
428 uuid->uuid, active ? "" : "in");
431 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
432 obd_uuid2str(uuid), active ? "" : "in");
435 lov->lov_tgts[index]->ltd_active = active;
437 lov->desc.ld_active_tgt_count++;
438 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
440 lov->desc.ld_active_tgt_count--;
441 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
444 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
452 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
453 enum obd_notify_event ev, void *data)
456 struct lov_obd *lov = &obd->u.lov;
459 down_read(&lov->lov_notify_lock);
460 if (!lov->lov_connects) {
461 up_read(&lov->lov_notify_lock);
465 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
466 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
467 struct obd_uuid *uuid;
471 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
472 up_read(&lov->lov_notify_lock);
473 CERROR("unexpected notification of %s %s!\n",
474 watched->obd_type->typ_name,
478 uuid = &watched->u.cli.cl_target_uuid;
480 /* Set OSC as active before notifying the observer, so the
481 * observer can use the OSC normally.
483 rc = lov_set_osc_active(obd, uuid, ev);
485 up_read(&lov->lov_notify_lock);
486 CERROR("event(%d) of %s failed: %d\n", ev,
487 obd_uuid2str(uuid), rc);
490 /* active event should be pass lov target index as data */
494 /* Pass the notification up the chain. */
496 rc = obd_notify_observer(obd, watched, ev, data);
498 /* NULL watched means all osc's in the lov (only for syncs) */
499 /* sync event should be send lov idx as data */
500 struct lov_obd *lov = &obd->u.lov;
504 is_sync = (ev == OBD_NOTIFY_SYNC) ||
505 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
508 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
509 if (!lov->lov_tgts[i])
512 /* don't send sync event if target not
513 * connected/activated */
514 if (is_sync && !lov->lov_tgts[i]->ltd_active)
517 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
520 CERROR("%s: notify %s of %s failed %d\n",
522 obd->obd_observer->obd_name,
523 lov->lov_tgts[i]->ltd_obd->obd_name,
530 up_read(&lov->lov_notify_lock);
534 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
535 __u32 index, int gen, int active)
537 struct lov_obd *lov = &obd->u.lov;
538 struct lov_tgt_desc *tgt;
539 struct obd_device *tgt_obd;
543 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
544 uuidp->uuid, index, gen, active);
547 CERROR("request to add OBD %s with invalid generation: %d\n",
552 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
557 mutex_lock(&lov->lov_lock);
559 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
560 tgt = lov->lov_tgts[index];
561 CERROR("UUID %s already assigned at LOV target index %d\n",
562 obd_uuid2str(&tgt->ltd_uuid), index);
563 mutex_unlock(&lov->lov_lock);
567 if (index >= lov->lov_tgt_size) {
568 /* We need to reallocate the lov target array. */
569 struct lov_tgt_desc **newtgts, **old = NULL;
570 __u32 newsize, oldsize = 0;
572 newsize = max(lov->lov_tgt_size, (__u32)2);
573 while (newsize < index + 1)
574 newsize = newsize << 1;
575 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
576 if (newtgts == NULL) {
577 mutex_unlock(&lov->lov_lock);
581 if (lov->lov_tgt_size) {
582 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
585 oldsize = lov->lov_tgt_size;
588 lov->lov_tgts = newtgts;
589 lov->lov_tgt_size = newsize;
592 OBD_FREE(old, sizeof(*old) * oldsize);
594 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
595 lov->lov_tgts, lov->lov_tgt_size);
600 mutex_unlock(&lov->lov_lock);
604 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
606 mutex_unlock(&lov->lov_lock);
611 tgt->ltd_uuid = *uuidp;
612 tgt->ltd_obd = tgt_obd;
613 /* XXX - add a sanity check on the generation number. */
615 tgt->ltd_index = index;
616 tgt->ltd_activate = active;
617 lov->lov_tgts[index] = tgt;
618 if (index >= lov->desc.ld_tgt_count)
619 lov->desc.ld_tgt_count = index + 1;
621 mutex_unlock(&lov->lov_lock);
623 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
624 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
626 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
628 if (lov->lov_connects == 0) {
629 /* lov_connect hasn't been called yet. We'll do the
630 lov_connect_obd on this target when that fn first runs,
631 because we don't know the connect flags yet. */
637 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
641 /* connect to administrative disabled ost */
645 if (lov->lov_cache != NULL) {
646 rc = obd_set_info_async(NULL, tgt->ltd_exp,
647 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
648 sizeof(struct cl_client_cache), lov->lov_cache,
654 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
655 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
660 CERROR("add failed (%d), deleting %s\n", rc,
661 obd_uuid2str(&tgt->ltd_uuid));
662 lov_del_target(obd, index, 0, 0);
668 /* Schedule a target for deletion */
669 int lov_del_target(struct obd_device *obd, __u32 index,
670 struct obd_uuid *uuidp, int gen)
672 struct lov_obd *lov = &obd->u.lov;
673 int count = lov->desc.ld_tgt_count;
677 if (index >= count) {
678 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
683 /* to make sure there's no ongoing lov_notify() now */
684 down_write(&lov->lov_notify_lock);
687 if (!lov->lov_tgts[index]) {
688 CERROR("LOV target at index %d is not setup.\n", index);
689 GOTO(out, rc = -EINVAL);
692 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
693 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
694 lov_uuid2str(lov, index), index,
695 obd_uuid2str(uuidp));
696 GOTO(out, rc = -EINVAL);
699 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
700 lov_uuid2str(lov, index), index,
701 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
702 lov->lov_tgts[index]->ltd_active);
704 lov->lov_tgts[index]->ltd_reap = 1;
705 lov->lov_death_row++;
706 /* we really delete it from obd_putref */
709 up_write(&lov->lov_notify_lock);
714 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
716 struct obd_device *osc_obd;
719 LASSERT(tgt->ltd_reap);
721 osc_obd = class_exp2obd(tgt->ltd_exp);
723 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
725 osc_obd ? osc_obd->obd_name : "<no obd>");
728 lov_disconnect_obd(obd, tgt);
732 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
733 do it ourselves. And we can't do it from lov_cleanup,
734 because we just lost our only reference to it. */
736 class_manual_cleanup(osc_obd);
739 void lov_fix_desc_stripe_size(__u64 *val)
741 if (*val < LOV_MIN_STRIPE_SIZE) {
743 LCONSOLE_INFO("Increasing default stripe size to "
745 LOV_DEFAULT_STRIPE_SIZE);
746 *val = LOV_DEFAULT_STRIPE_SIZE;
747 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
748 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
749 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
751 *val, LOV_MIN_STRIPE_SIZE);
755 void lov_fix_desc_stripe_count(__u32 *val)
761 void lov_fix_desc_pattern(__u32 *val)
763 /* from lov_setstripe */
764 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
765 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
770 void lov_fix_desc_qos_maxage(__u32 *val)
774 *val = QOS_DEFAULT_MAXAGE;
777 void lov_fix_desc(struct lov_desc *desc)
779 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
780 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
781 lov_fix_desc_pattern(&desc->ld_pattern);
782 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
785 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
787 struct lov_desc *desc;
788 struct lov_obd *lov = &obd->u.lov;
792 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
793 CERROR("LOV setup requires a descriptor\n");
797 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
799 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
800 CERROR("descriptor size wrong: %d > %d\n",
801 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
805 if (desc->ld_magic != LOV_DESC_MAGIC) {
806 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
807 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
808 obd->obd_name, desc);
809 lustre_swab_lov_desc(desc);
811 CERROR("%s: Bad lov desc magic: %#x\n",
812 obd->obd_name, desc->ld_magic);
819 desc->ld_active_tgt_count = 0;
821 lov->lov_tgt_size = 0;
823 mutex_init(&lov->lov_lock);
824 cfs_atomic_set(&lov->lov_refcount, 0);
825 lov->lov_sp_me = LUSTRE_SP_CLI;
827 init_rwsem(&lov->lov_notify_lock);
829 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
831 HASH_POOLS_BKT_BITS, 0,
834 &pool_hash_operations,
836 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
837 lov->lov_pool_count = 0;
838 rc = lov_ost_pool_init(&lov->lov_packed, 0);
843 obd->obd_vars = lprocfs_lov_obd_vars;
844 lprocfs_seq_obd_setup(obd);
845 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", 0444,
846 &lov_proc_target_fops, obd);
848 CWARN("Error adding the target_obd file\n");
850 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
860 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
863 struct lov_obd *lov = &obd->u.lov;
868 case OBD_CLEANUP_EARLY: {
870 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
871 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
873 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
878 case OBD_CLEANUP_EXPORTS:
879 rc = obd_llog_finish(obd, 0);
881 CERROR("failed to cleanup llogging subsystems\n");
887 static int lov_cleanup(struct obd_device *obd)
889 struct lov_obd *lov = &obd->u.lov;
890 cfs_list_t *pos, *tmp;
891 struct pool_desc *pool;
894 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
895 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
896 /* free pool structs */
897 CDEBUG(D_INFO, "delete pool %p\n", pool);
898 /* In the function below, .hs_keycmp resolves to
899 * pool_hashkey_keycmp() */
900 /* coverity[overrun-buffer-val] */
901 lov_pool_del(obd, pool->pool_name);
903 cfs_hash_putref(lov->lov_pools_hash_body);
904 lov_ost_pool_free(&lov->lov_packed);
906 lprocfs_obd_cleanup(obd);
910 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
911 if (!lov->lov_tgts[i])
914 /* Inactive targets may never have connected */
915 if (lov->lov_tgts[i]->ltd_active ||
916 cfs_atomic_read(&lov->lov_refcount))
917 /* We should never get here - these
918 should have been removed in the
920 CERROR("lov tgt %d not cleaned!"
921 " deathrow=%d, lovrc=%d\n",
922 i, lov->lov_death_row,
923 cfs_atomic_read(&lov->lov_refcount));
924 lov_del_target(obd, i, 0, 0);
927 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
929 lov->lov_tgt_size = 0;
934 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
935 __u32 *indexp, int *genp)
937 struct obd_uuid obd_uuid;
942 switch(cmd = lcfg->lcfg_command) {
943 case LCFG_LOV_ADD_OBD:
944 case LCFG_LOV_ADD_INA:
945 case LCFG_LOV_DEL_OBD: {
948 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
949 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
950 GOTO(out, rc = -EINVAL);
952 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
954 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
955 GOTO(out, rc = -EINVAL);
956 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
957 GOTO(out, rc = -EINVAL);
960 if (cmd == LCFG_LOV_ADD_OBD)
961 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
962 else if (cmd == LCFG_LOV_ADD_INA)
963 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
965 rc = lov_del_target(obd, index, &obd_uuid, gen);
969 struct lov_desc *desc = &(obd->u.lov.desc);
972 GOTO(out, rc = -EINVAL);
974 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
987 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
988 GOTO(out, rc = -EINVAL);
996 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
997 struct lov_stripe_md **ea, struct obd_trans_info *oti)
999 struct lov_stripe_md *obj_mdp, *lsm;
1000 struct lov_obd *lov = &exp->exp_obd->u.lov;
1005 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1006 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1008 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1009 if (obj_mdp == NULL)
1012 ost_idx = src_oa->o_nlink;
1015 GOTO(out, rc = -EINVAL);
1016 if (ost_idx >= lov->desc.ld_tgt_count ||
1017 !lov->lov_tgts[ost_idx])
1018 GOTO(out, rc = -EINVAL);
1020 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1021 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1022 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1023 ostid_id(&src_oa->o_oi))
1024 GOTO(out, rc = -EINVAL);
1028 if (i == lsm->lsm_stripe_count)
1029 GOTO(out, rc = -EINVAL);
1031 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1032 src_oa, &obj_mdp, oti);
1034 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1038 /* the LOV expects oa->o_id to be set to the LOV object id */
1039 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1040 struct obdo *src_oa, struct lov_stripe_md **ea,
1041 struct obd_trans_info *oti)
1043 struct lov_obd *lov;
1047 LASSERT(ea != NULL);
1051 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1052 src_oa->o_flags == OBD_FL_DELORPHAN) {
1053 /* should be used with LOV anymore */
1057 lov = &exp->exp_obd->u.lov;
1058 if (!lov->desc.ld_active_tgt_count)
1061 obd_getref(exp->exp_obd);
1062 /* Recreate a specific object id at the given OST index */
1063 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1064 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1065 rc = lov_recreate(exp, src_oa, ea, oti);
1068 obd_putref(exp->exp_obd);
1072 #define ASSERT_LSM_MAGIC(lsmp) \
1074 LASSERT((lsmp) != NULL); \
1075 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1076 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1077 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1080 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1081 struct obdo *oa, struct lov_stripe_md *lsm,
1082 struct obd_trans_info *oti, struct obd_export *md_exp,
1085 struct lov_request_set *set;
1086 struct obd_info oinfo;
1087 struct lov_request *req;
1089 struct lov_obd *lov;
1090 int rc = 0, err = 0;
1093 ASSERT_LSM_MAGIC(lsm);
1095 if (!exp || !exp->exp_obd)
1098 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1100 LASSERT(oti->oti_logcookies);
1103 lov = &exp->exp_obd->u.lov;
1104 obd_getref(exp->exp_obd);
1105 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1109 cfs_list_for_each (pos, &set->set_list) {
1110 req = cfs_list_entry(pos, struct lov_request, rq_link);
1112 if (oa->o_valid & OBD_MD_FLCOOKIE)
1113 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1115 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1116 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1117 err = lov_update_common_set(set, req, err);
1119 CERROR("%s: destroying objid "DOSTID" subobj "
1120 DOSTID" on OST idx %d: rc = %d\n",
1121 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1122 POSTID(&req->rq_oi.oi_oa->o_oi),
1130 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1131 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1133 err = lov_fini_destroy_set(set);
1135 obd_putref(exp->exp_obd);
1136 RETURN(rc ? rc : err);
1139 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1140 struct obd_info *oinfo)
1142 struct lov_request_set *set;
1143 struct lov_request *req;
1145 struct lov_obd *lov;
1146 int err = 0, rc = 0;
1150 ASSERT_LSM_MAGIC(oinfo->oi_md);
1152 if (!exp || !exp->exp_obd)
1155 lov = &exp->exp_obd->u.lov;
1157 rc = lov_prep_getattr_set(exp, oinfo, &set);
1161 cfs_list_for_each (pos, &set->set_list) {
1162 req = cfs_list_entry(pos, struct lov_request, rq_link);
1164 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1165 " %u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1166 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1168 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1170 err = lov_update_common_set(set, req, rc);
1172 CERROR("%s: getattr objid "DOSTID" subobj "
1173 DOSTID" on OST idx %d: rc = %d\n",
1174 exp->exp_obd->obd_name,
1175 POSTID(&oinfo->oi_oa->o_oi),
1176 POSTID(&req->rq_oi.oi_oa->o_oi),
1182 rc = lov_fini_getattr_set(set);
1188 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1191 struct lov_request_set *lovset = (struct lov_request_set *)data;
1195 /* don't do attribute merge if this aysnc op failed */
1197 cfs_atomic_set(&lovset->set_completes, 0);
1198 err = lov_fini_getattr_set(lovset);
1199 RETURN(rc ? rc : err);
1202 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1203 struct ptlrpc_request_set *rqset)
1205 struct lov_request_set *lovset;
1206 struct lov_obd *lov;
1208 struct lov_request *req;
1213 ASSERT_LSM_MAGIC(oinfo->oi_md);
1215 if (!exp || !exp->exp_obd)
1218 lov = &exp->exp_obd->u.lov;
1220 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1224 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1225 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1226 oinfo->oi_md->lsm_stripe_size);
1228 cfs_list_for_each(pos, &lovset->set_list) {
1229 req = cfs_list_entry(pos, struct lov_request, rq_link);
1231 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1232 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1233 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1234 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1235 &req->rq_oi, rqset);
1237 CERROR("%s: getattr objid "DOSTID" subobj"
1238 DOSTID" on OST idx %d: rc = %d\n",
1239 exp->exp_obd->obd_name,
1240 POSTID(&oinfo->oi_oa->o_oi),
1241 POSTID(&req->rq_oi.oi_oa->o_oi),
1247 if (!cfs_list_empty(&rqset->set_requests)) {
1249 LASSERT (rqset->set_interpret == NULL);
1250 rqset->set_interpret = lov_getattr_interpret;
1251 rqset->set_arg = (void *)lovset;
1256 cfs_atomic_set(&lovset->set_completes, 0);
1257 err = lov_fini_getattr_set(lovset);
1258 RETURN(rc ? rc : err);
1261 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1262 struct obd_info *oinfo, struct obd_trans_info *oti)
1264 struct lov_request_set *set;
1265 struct lov_obd *lov;
1267 struct lov_request *req;
1268 int err = 0, rc = 0;
1272 ASSERT_LSM_MAGIC(oinfo->oi_md);
1274 if (!exp || !exp->exp_obd)
1277 /* for now, we only expect the following updates here */
1278 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1279 OBD_MD_FLMODE | OBD_MD_FLATIME |
1280 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1281 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1282 OBD_MD_FLGROUP | OBD_MD_FLUID |
1283 OBD_MD_FLGID | OBD_MD_FLFID |
1285 lov = &exp->exp_obd->u.lov;
1286 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1290 cfs_list_for_each (pos, &set->set_list) {
1291 req = cfs_list_entry(pos, struct lov_request, rq_link);
1293 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1295 err = lov_update_setattr_set(set, req, rc);
1297 CERROR("%s: setattr objid "DOSTID" subobj "
1298 DOSTID" on OST idx %d: rc = %d\n",
1299 exp->exp_obd->obd_name,
1300 POSTID(&set->set_oi->oi_oa->o_oi),
1301 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1307 err = lov_fini_setattr_set(set);
1313 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1316 struct lov_request_set *lovset = (struct lov_request_set *)data;
1321 cfs_atomic_set(&lovset->set_completes, 0);
1322 err = lov_fini_setattr_set(lovset);
1323 RETURN(rc ? rc : err);
1326 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1327 needed. Otherwise, a client is waiting for responses. */
1328 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1329 struct obd_trans_info *oti,
1330 struct ptlrpc_request_set *rqset)
1332 struct lov_request_set *set;
1333 struct lov_request *req;
1335 struct lov_obd *lov;
1340 ASSERT_LSM_MAGIC(oinfo->oi_md);
1341 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1343 LASSERT(oti->oti_logcookies);
1346 if (!exp || !exp->exp_obd)
1349 lov = &exp->exp_obd->u.lov;
1350 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1354 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1355 POSTID(&oinfo->oi_md->lsm_oi),
1356 oinfo->oi_md->lsm_stripe_count,
1357 oinfo->oi_md->lsm_stripe_size);
1359 cfs_list_for_each(pos, &set->set_list) {
1360 req = cfs_list_entry(pos, struct lov_request, rq_link);
1362 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1363 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1365 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1366 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1367 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1369 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1370 &req->rq_oi, oti, rqset);
1372 CERROR("error: setattr objid "DOSTID" subobj"
1373 DOSTID" on OST idx %d: rc = %d\n",
1374 POSTID(&set->set_oi->oi_oa->o_oi),
1375 POSTID(&req->rq_oi.oi_oa->o_oi),
1381 /* If we are not waiting for responses on async requests, return. */
1382 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1385 cfs_atomic_set(&set->set_completes, 0);
1386 err = lov_fini_setattr_set(set);
1387 RETURN(rc ? rc : err);
1390 LASSERT(rqset->set_interpret == NULL);
1391 rqset->set_interpret = lov_setattr_interpret;
1392 rqset->set_arg = (void *)set;
1397 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1400 struct lov_request_set *lovset = (struct lov_request_set *)data;
1405 cfs_atomic_set(&lovset->set_completes, 0);
1406 err = lov_fini_punch_set(lovset);
1407 RETURN(rc ? rc : err);
1410 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1411 * we can send this 'punch' to just the authoritative node and the nodes
1412 * that the punch will affect. */
1413 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1414 struct obd_info *oinfo, struct obd_trans_info *oti,
1415 struct ptlrpc_request_set *rqset)
1417 struct lov_request_set *set;
1418 struct lov_obd *lov;
1420 struct lov_request *req;
1425 ASSERT_LSM_MAGIC(oinfo->oi_md);
1427 if (!exp || !exp->exp_obd)
1430 lov = &exp->exp_obd->u.lov;
1431 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1435 cfs_list_for_each (pos, &set->set_list) {
1436 req = cfs_list_entry(pos, struct lov_request, rq_link);
1438 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1439 &req->rq_oi, NULL, rqset);
1441 CERROR("%s: punch objid "DOSTID" subobj "DOSTID
1442 " on OST idx %d: rc = %d\n",
1443 exp->exp_obd->obd_name,
1444 POSTID(&set->set_oi->oi_oa->o_oi),
1445 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, rc);
1450 if (rc || cfs_list_empty(&rqset->set_requests)) {
1452 err = lov_fini_punch_set(set);
1453 RETURN(rc ? rc : err);
1456 LASSERT(rqset->set_interpret == NULL);
1457 rqset->set_interpret = lov_punch_interpret;
1458 rqset->set_arg = (void *)set;
1463 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1466 struct lov_request_set *lovset = data;
1471 cfs_atomic_set(&lovset->set_completes, 0);
1472 err = lov_fini_sync_set(lovset);
1476 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1477 struct obd_info *oinfo, obd_off start, obd_off end,
1478 struct ptlrpc_request_set *rqset)
1480 struct lov_request_set *set = NULL;
1481 struct lov_obd *lov;
1483 struct lov_request *req;
1487 ASSERT_LSM_MAGIC(oinfo->oi_md);
1488 LASSERT(rqset != NULL);
1493 lov = &exp->exp_obd->u.lov;
1494 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1498 CDEBUG(D_INFO, "fsync objid "DOSTID" ["LPX64", "LPX64"]\n",
1499 POSTID(&set->set_oi->oi_oa->o_oi), start, end);
1501 cfs_list_for_each (pos, &set->set_list) {
1502 req = cfs_list_entry(pos, struct lov_request, rq_link);
1504 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1505 &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1506 req->rq_oi.oi_policy.l_extent.end, rqset);
1508 CERROR("%s: fsync objid "DOSTID" subobj "DOSTID
1509 " on OST idx %d: rc = %d\n",
1510 exp->exp_obd->obd_name,
1511 POSTID(&set->set_oi->oi_oa->o_oi),
1512 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1518 /* If we are not waiting for responses on async requests, return. */
1519 if (rc || cfs_list_empty(&rqset->set_requests)) {
1520 int err = lov_fini_sync_set(set);
1525 LASSERT(rqset->set_interpret == NULL);
1526 rqset->set_interpret = lov_sync_interpret;
1527 rqset->set_arg = (void *)set;
1532 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1533 obd_count oa_bufs, struct brw_page *pga)
1535 struct obd_info oinfo = { { { 0 } } };
1538 oinfo.oi_oa = lov_oinfo->oi_oa;
1540 /* The caller just wants to know if there's a chance that this
1541 * I/O can succeed */
1542 for (i = 0; i < oa_bufs; i++) {
1543 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1544 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1547 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1548 pga[i].off + pga[i].count - 1,
1552 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1553 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1557 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1565 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1566 obd_count oa_bufs, struct brw_page *pga,
1567 struct obd_trans_info *oti)
1569 struct lov_request_set *set;
1570 struct lov_request *req;
1572 struct lov_obd *lov = &exp->exp_obd->u.lov;
1576 ASSERT_LSM_MAGIC(oinfo->oi_md);
1578 if (cmd == OBD_BRW_CHECK) {
1579 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1583 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1587 cfs_list_for_each (pos, &set->set_list) {
1588 struct obd_export *sub_exp;
1589 struct brw_page *sub_pga;
1590 req = cfs_list_entry(pos, struct lov_request, rq_link);
1592 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1593 sub_pga = set->set_pga + req->rq_pgaidx;
1594 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1598 lov_update_common_set(set, req, rc);
1601 err = lov_fini_brw_set(set);
1607 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1610 struct lov_request_set *lovset = (struct lov_request_set *)data;
1612 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1616 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1617 struct ldlm_enqueue_info *einfo,
1618 struct ptlrpc_request_set *rqset)
1620 ldlm_mode_t mode = einfo->ei_mode;
1621 struct lov_request_set *set;
1622 struct lov_request *req;
1624 struct lov_obd *lov;
1629 ASSERT_LSM_MAGIC(oinfo->oi_md);
1630 LASSERT(mode == (mode & -mode));
1632 /* we should never be asked to replay a lock this way. */
1633 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1635 if (!exp || !exp->exp_obd)
1638 lov = &exp->exp_obd->u.lov;
1639 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1643 cfs_list_for_each (pos, &set->set_list) {
1644 req = cfs_list_entry(pos, struct lov_request, rq_link);
1646 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1647 &req->rq_oi, einfo, rqset);
1652 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1654 LASSERT(rqset->set_interpret == NULL);
1655 rqset->set_interpret = lov_enqueue_interpret;
1656 rqset->set_arg = (void *)set;
1660 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1664 static int lov_change_cbdata(struct obd_export *exp,
1665 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1668 struct lov_obd *lov;
1672 ASSERT_LSM_MAGIC(lsm);
1674 if (!exp || !exp->exp_obd)
1677 lov = &exp->exp_obd->u.lov;
1678 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1679 struct lov_stripe_md submd;
1680 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1682 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1683 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1687 submd.lsm_oi = loi->loi_oi;
1688 submd.lsm_stripe_count = 0;
1689 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1695 /* find any ldlm lock of the inode in lov
1699 static int lov_find_cbdata(struct obd_export *exp,
1700 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1703 struct lov_obd *lov;
1707 ASSERT_LSM_MAGIC(lsm);
1709 if (!exp || !exp->exp_obd)
1712 lov = &exp->exp_obd->u.lov;
1713 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1714 struct lov_stripe_md submd;
1715 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1717 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1718 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1721 submd.lsm_oi = loi->loi_oi;
1722 submd.lsm_stripe_count = 0;
1723 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1731 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1732 __u32 mode, struct lustre_handle *lockh)
1734 struct lov_request_set *set;
1735 struct obd_info oinfo;
1736 struct lov_request *req;
1738 struct lov_obd *lov;
1739 struct lustre_handle *lov_lockhp;
1740 int err = 0, rc = 0;
1743 ASSERT_LSM_MAGIC(lsm);
1745 if (!exp || !exp->exp_obd)
1749 lov = &exp->exp_obd->u.lov;
1750 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1754 cfs_list_for_each(pos, &set->set_list) {
1755 req = cfs_list_entry(pos, struct lov_request, rq_link);
1756 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1758 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1759 req->rq_oi.oi_md, mode, lov_lockhp);
1760 rc = lov_update_common_set(set, req, rc);
1762 CERROR("%s: cancel objid "DOSTID" subobj "
1763 DOSTID" on OST idx %d: rc = %d\n",
1764 exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1765 POSTID(&req->rq_oi.oi_md->lsm_oi),
1771 lov_fini_cancel_set(set);
1775 static int lov_cancel_unused(struct obd_export *exp,
1776 struct lov_stripe_md *lsm,
1777 ldlm_cancel_flags_t flags, void *opaque)
1779 struct lov_obd *lov;
1783 if (!exp || !exp->exp_obd)
1786 lov = &exp->exp_obd->u.lov;
1788 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1790 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1793 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1801 ASSERT_LSM_MAGIC(lsm);
1803 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1804 struct lov_stripe_md submd;
1805 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1806 int idx = loi->loi_ost_idx;
1809 if (!lov->lov_tgts[idx]) {
1810 CDEBUG(D_HA, "lov idx %d NULL\n", idx);
1814 if (!lov->lov_tgts[idx]->ltd_active)
1815 CDEBUG(D_HA, "lov idx %d inactive\n", idx);
1817 submd.lsm_oi = loi->loi_oi;
1818 submd.lsm_stripe_count = 0;
1819 err = obd_cancel_unused(lov->lov_tgts[idx]->ltd_exp,
1820 &submd, flags, opaque);
1821 if (err && lov->lov_tgts[idx]->ltd_active) {
1822 CERROR("%s: cancel unused objid "DOSTID
1823 " subobj "DOSTID" on OST idx %d: rc = %d\n",
1824 exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1825 POSTID(&loi->loi_oi), idx, err);
1833 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1835 struct lov_request_set *lovset = (struct lov_request_set *)data;
1840 cfs_atomic_set(&lovset->set_completes, 0);
1842 err = lov_fini_statfs_set(lovset);
1843 RETURN(rc ? rc : err);
1846 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1847 __u64 max_age, struct ptlrpc_request_set *rqset)
1849 struct obd_device *obd = class_exp2obd(exp);
1850 struct lov_request_set *set;
1851 struct lov_request *req;
1853 struct lov_obd *lov;
1857 LASSERT(oinfo != NULL);
1858 LASSERT(oinfo->oi_osfs != NULL);
1861 rc = lov_prep_statfs_set(obd, oinfo, &set);
1865 cfs_list_for_each (pos, &set->set_list) {
1866 req = cfs_list_entry(pos, struct lov_request, rq_link);
1867 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1868 &req->rq_oi, max_age, rqset);
1873 if (rc || cfs_list_empty(&rqset->set_requests)) {
1876 cfs_atomic_set(&set->set_completes, 0);
1877 err = lov_fini_statfs_set(set);
1878 RETURN(rc ? rc : err);
1881 LASSERT(rqset->set_interpret == NULL);
1882 rqset->set_interpret = lov_statfs_interpret;
1883 rqset->set_arg = (void *)set;
1887 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1888 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1890 struct ptlrpc_request_set *set = NULL;
1891 struct obd_info oinfo = { { { 0 } } };
1896 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1897 * statfs requests */
1898 set = ptlrpc_prep_set();
1902 oinfo.oi_osfs = osfs;
1903 oinfo.oi_flags = flags;
1904 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1906 rc = ptlrpc_set_wait(set);
1907 ptlrpc_set_destroy(set);
1912 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1913 void *karg, void *uarg)
1915 struct obd_device *obddev = class_exp2obd(exp);
1916 struct lov_obd *lov = &obddev->u.lov;
1917 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1918 struct obd_uuid *uuidp;
1922 case IOC_OBD_STATFS: {
1923 struct obd_ioctl_data *data = karg;
1924 struct obd_device *osc_obd;
1925 struct obd_statfs stat_buf = {0};
1929 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1930 if ((index >= count))
1933 if (!lov->lov_tgts[index])
1934 /* Try again with the next index */
1936 if (!lov->lov_tgts[index]->ltd_active)
1939 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1944 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1945 min((int)data->ioc_plen2,
1946 (int)sizeof(struct obd_uuid))))
1949 flags = uarg ? *(__u32*)uarg : 0;
1950 /* got statfs data */
1951 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1952 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1956 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1957 min((int) data->ioc_plen1,
1958 (int) sizeof(stat_buf))))
1962 case OBD_IOC_LOV_GET_CONFIG: {
1963 struct obd_ioctl_data *data;
1964 struct lov_desc *desc;
1969 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1972 data = (struct obd_ioctl_data *)buf;
1974 if (sizeof(*desc) > data->ioc_inllen1) {
1975 obd_ioctl_freedata(buf, len);
1979 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1980 obd_ioctl_freedata(buf, len);
1984 if (sizeof(__u32) * count > data->ioc_inllen3) {
1985 obd_ioctl_freedata(buf, len);
1989 desc = (struct lov_desc *)data->ioc_inlbuf1;
1990 memcpy(desc, &(lov->desc), sizeof(*desc));
1992 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1993 genp = (__u32 *)data->ioc_inlbuf3;
1994 /* the uuid will be empty for deleted OSTs */
1995 for (i = 0; i < count; i++, uuidp++, genp++) {
1996 if (!lov->lov_tgts[i])
1998 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1999 *genp = lov->lov_tgts[i]->ltd_gen;
2002 if (copy_to_user((void *)uarg, buf, len))
2004 obd_ioctl_freedata(buf, len);
2007 case LL_IOC_LOV_SETSTRIPE:
2008 rc = lov_setstripe(exp, len, karg, uarg);
2010 case LL_IOC_LOV_GETSTRIPE:
2011 rc = lov_getstripe(exp, karg, uarg);
2013 case LL_IOC_LOV_SETEA:
2014 rc = lov_setea(exp, karg, uarg);
2016 case OBD_IOC_QUOTACTL: {
2017 struct if_quotactl *qctl = karg;
2018 struct lov_tgt_desc *tgt = NULL;
2019 struct obd_quotactl *oqctl;
2021 if (qctl->qc_valid == QC_OSTIDX) {
2022 if (count <= qctl->qc_idx)
2025 tgt = lov->lov_tgts[qctl->qc_idx];
2026 if (!tgt || !tgt->ltd_exp)
2028 } else if (qctl->qc_valid == QC_UUID) {
2029 for (i = 0; i < count; i++) {
2030 tgt = lov->lov_tgts[i];
2032 !obd_uuid_equals(&tgt->ltd_uuid,
2036 if (tgt->ltd_exp == NULL)
2048 LASSERT(tgt && tgt->ltd_exp);
2049 OBD_ALLOC_PTR(oqctl);
2053 QCTL_COPY(oqctl, qctl);
2054 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2056 QCTL_COPY(qctl, oqctl);
2057 qctl->qc_valid = QC_OSTIDX;
2058 qctl->obd_uuid = tgt->ltd_uuid;
2060 OBD_FREE_PTR(oqctl);
2069 for (i = 0; i < count; i++) {
2071 struct obd_device *osc_obd;
2073 /* OST was disconnected */
2074 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2077 /* ll_umount_begin() sets force flag but for lov, not
2078 * osc. Let's pass it through */
2079 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2080 osc_obd->obd_force = obddev->obd_force;
2081 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2083 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2086 if (lov->lov_tgts[i]->ltd_active) {
2087 CDEBUG(err == -ENOTTY ?
2088 D_IOCTL : D_WARNING,
2089 "iocontrol OSC %s on OST "
2090 "idx %d cmd %x: err = %d\n",
2091 lov_uuid2str(lov, i),
2108 #define FIEMAP_BUFFER_SIZE 4096
2111 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2112 * call. The local end offset and the device are sent in the first
2113 * fm_extent. This function calculates the stripe number from the index.
2114 * This function returns a stripe_no on which mapping is to be restarted.
2116 * This function returns fm_end_offset which is the in-OST offset at which
2117 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2118 * will re-calculate proper offset in next stripe.
2119 * Note that the first extent is passed to lov_get_info via the value field.
2121 * \param fiemap fiemap request header
2122 * \param lsm striping information for the file
2123 * \param fm_start logical start of mapping
2124 * \param fm_end logical end of mapping
2125 * \param start_stripe starting stripe will be returned in this
2127 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2128 struct lov_stripe_md *lsm, obd_size fm_start,
2129 obd_size fm_end, int *start_stripe)
2131 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2132 obd_off lun_start, lun_end;
2133 obd_size fm_end_offset;
2134 int stripe_no = -1, i;
2136 if (fiemap->fm_extent_count == 0 ||
2137 fiemap->fm_extents[0].fe_logical == 0)
2140 /* Find out stripe_no from ost_index saved in the fe_device */
2141 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2142 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2143 fiemap->fm_extents[0].fe_device) {
2148 if (stripe_no == -1)
2151 /* If we have finished mapping on previous device, shift logical
2152 * offset to start of next device */
2153 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2154 &lun_start, &lun_end)) != 0 &&
2155 local_end < lun_end) {
2156 fm_end_offset = local_end;
2157 *start_stripe = stripe_no;
2159 /* This is a special value to indicate that caller should
2160 * calculate offset in next stripe. */
2162 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2165 return fm_end_offset;
2169 * We calculate on which OST the mapping will end. If the length of mapping
2170 * is greater than (stripe_size * stripe_count) then the last_stripe will
2171 * will be one just before start_stripe. Else we check if the mapping
2172 * intersects each OST and find last_stripe.
2173 * This function returns the last_stripe and also sets the stripe_count
2174 * over which the mapping is spread
2176 * \param lsm striping information for the file
2177 * \param fm_start logical start of mapping
2178 * \param fm_end logical end of mapping
2179 * \param start_stripe starting stripe of the mapping
2180 * \param stripe_count the number of stripes across which to map is returned
2182 * \retval last_stripe return the last stripe of the mapping
2184 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2185 obd_size fm_end, int start_stripe,
2189 obd_off obd_start, obd_end;
2192 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2193 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2195 *stripe_count = lsm->lsm_stripe_count;
2197 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2198 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2199 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2200 &obd_start, &obd_end)) == 0)
2204 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2211 * Set fe_device and copy extents from local buffer into main return buffer.
2213 * \param fiemap fiemap request header
2214 * \param lcl_fm_ext array of local fiemap extents to be copied
2215 * \param ost_index OST index to be written into the fm_device field for each
2217 * \param ext_count number of extents to be copied
2218 * \param current_extent where to start copying in main extent array
2220 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2221 struct ll_fiemap_extent *lcl_fm_ext,
2222 int ost_index, unsigned int ext_count,
2228 for (ext = 0; ext < ext_count; ext++) {
2229 lcl_fm_ext[ext].fe_device = ost_index;
2230 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2233 /* Copy fm_extent's from fm_local to return buffer */
2234 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2235 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2239 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2240 * This also handles the restarting of FIEMAP calls in case mapping overflows
2241 * the available number of extents in single call.
2243 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2244 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2246 struct ll_fiemap_info_key *fm_key = key;
2247 struct ll_user_fiemap *fiemap = val;
2248 struct ll_user_fiemap *fm_local = NULL;
2249 struct ll_fiemap_extent *lcl_fm_ext;
2251 unsigned int get_num_extents = 0;
2252 int ost_index = 0, actual_start_stripe, start_stripe;
2253 obd_size fm_start, fm_end, fm_length, fm_end_offset;
2255 int current_extent = 0, rc = 0, i;
2256 int ost_eof = 0; /* EOF for object */
2257 int ost_done = 0; /* done with required mapping for this OST? */
2259 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2260 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2262 if (!lsm_has_objects(lsm)) {
2263 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
2264 fm_key->oa.o_size)) {
2265 /* released file, return a minimal FIEMAP if
2266 * request fits in file-size.
2268 fiemap->fm_mapped_extents = 1;
2269 fiemap->fm_extents[0].fe_logical =
2270 fm_key->fiemap.fm_start;
2271 if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
2273 fiemap->fm_extents[0].fe_length =
2274 fm_key->fiemap.fm_length;
2276 fiemap->fm_extents[0].fe_length =
2278 fm_key->fiemap.fm_start;
2279 fiemap->fm_extents[0].fe_flags |=
2280 (FIEMAP_EXTENT_UNKNOWN |
2281 FIEMAP_EXTENT_LAST);
2286 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2287 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2289 OBD_ALLOC_LARGE(fm_local, buffer_size);
2290 if (fm_local == NULL)
2291 GOTO(out, rc = -ENOMEM);
2292 lcl_fm_ext = &fm_local->fm_extents[0];
2294 count_local = fiemap_size_to_count(buffer_size);
2296 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2297 fm_start = fiemap->fm_start;
2298 fm_length = fiemap->fm_length;
2299 /* Calculate start stripe, last stripe and length of mapping */
2300 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2301 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2302 fm_start + fm_length - 1);
2303 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2304 if (fm_end > fm_key->oa.o_size)
2305 fm_end = fm_key->oa.o_size;
2307 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2308 actual_start_stripe, &stripe_count);
2310 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2311 fm_end, &start_stripe);
2312 if (fm_end_offset == -EINVAL)
2313 GOTO(out, rc = -EINVAL);
2315 if (fiemap->fm_extent_count == 0) {
2316 get_num_extents = 1;
2320 /* Check each stripe */
2321 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2322 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2323 obd_size req_fm_len; /* Stores length of required mapping */
2324 obd_size len_mapped_single_call;
2325 obd_off lun_start, lun_end, obd_object_end;
2326 unsigned int ext_count;
2328 cur_stripe_wrap = cur_stripe;
2330 /* Find out range of mapping on this stripe */
2331 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2332 &lun_start, &obd_object_end)) == 0)
2335 /* If this is a continuation FIEMAP call and we are on
2336 * starting stripe then lun_start needs to be set to
2338 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2339 lun_start = fm_end_offset;
2341 if (fm_length != ~0ULL) {
2342 /* Handle fm_start + fm_length overflow */
2343 if (fm_start + fm_length < fm_start)
2344 fm_length = ~0ULL - fm_start;
2345 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2351 if (lun_start == lun_end)
2354 req_fm_len = obd_object_end - lun_start;
2355 fm_local->fm_length = 0;
2356 len_mapped_single_call = 0;
2358 /* If the output buffer is very large and the objects have many
2359 * extents we may need to loop on a single OST repeatedly */
2363 if (get_num_extents == 0) {
2364 /* Don't get too many extents. */
2365 if (current_extent + count_local >
2366 fiemap->fm_extent_count)
2367 count_local = fiemap->fm_extent_count -
2371 lun_start += len_mapped_single_call;
2372 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2373 req_fm_len = fm_local->fm_length;
2374 fm_local->fm_extent_count = count_local;
2375 fm_local->fm_mapped_extents = 0;
2376 fm_local->fm_flags = fiemap->fm_flags;
2378 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
2379 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2381 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2382 GOTO(out, rc = -EINVAL);
2384 /* If OST is inactive, return extent with UNKNOWN flag */
2385 if (!lov->lov_tgts[ost_index]->ltd_active) {
2386 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2387 fm_local->fm_mapped_extents = 1;
2389 lcl_fm_ext[0].fe_logical = lun_start;
2390 lcl_fm_ext[0].fe_length = obd_object_end -
2392 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2397 fm_local->fm_start = lun_start;
2398 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2399 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2400 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2401 rc = obd_get_info(NULL,
2402 lov->lov_tgts[ost_index]->ltd_exp,
2403 keylen, key, vallen, fm_local, lsm);
2408 ext_count = fm_local->fm_mapped_extents;
2409 if (ext_count == 0) {
2411 /* If last stripe has hole at the end,
2412 * then we need to return */
2413 if (cur_stripe_wrap == last_stripe) {
2414 fiemap->fm_mapped_extents = 0;
2420 /* If we just need num of extents then go to next device */
2421 if (get_num_extents) {
2422 current_extent += ext_count;
2426 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2427 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2429 /* Have we finished mapping on this device? */
2430 if (req_fm_len <= len_mapped_single_call)
2433 /* Clear the EXTENT_LAST flag which can be present on
2435 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2436 lcl_fm_ext[ext_count - 1].fe_flags &=
2437 ~FIEMAP_EXTENT_LAST;
2439 curr_loc = lov_stripe_size(lsm,
2440 lcl_fm_ext[ext_count - 1].fe_logical+
2441 lcl_fm_ext[ext_count - 1].fe_length,
2443 if (curr_loc >= fm_key->oa.o_size)
2446 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2447 ost_index, ext_count,
2450 current_extent += ext_count;
2452 /* Ran out of available extents? */
2453 if (current_extent >= fiemap->fm_extent_count)
2455 } while (ost_done == 0 && ost_eof == 0);
2457 if (cur_stripe_wrap == last_stripe)
2462 /* Indicate that we are returning device offsets unless file just has
2464 if (lsm->lsm_stripe_count > 1)
2465 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2467 if (get_num_extents)
2468 goto skip_last_device_calc;
2470 /* Check if we have reached the last stripe and whether mapping for that
2471 * stripe is done. */
2472 if (cur_stripe_wrap == last_stripe) {
2473 if (ost_done || ost_eof)
2474 fiemap->fm_extents[current_extent - 1].fe_flags |=
2478 skip_last_device_calc:
2479 fiemap->fm_mapped_extents = current_extent;
2483 OBD_FREE_LARGE(fm_local, buffer_size);
2487 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2488 __u32 keylen, void *key, __u32 *vallen, void *val,
2489 struct lov_stripe_md *lsm)
2491 struct obd_device *obddev = class_exp2obd(exp);
2492 struct lov_obd *lov = &obddev->u.lov;
2496 if (!vallen || !val)
2501 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2504 struct ldlm_lock *lock;
2506 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2507 struct lov_oinfo *loi;
2508 __u32 *stripe = val;
2510 if (*vallen < sizeof(*stripe))
2511 GOTO(out, rc = -EFAULT);
2512 *vallen = sizeof(*stripe);
2514 /* XXX This is another one of those bits that will need to
2515 * change if we ever actually support nested LOVs. It uses
2516 * the lock's export to find out which stripe it is. */
2517 /* XXX - it's assumed all the locks for deleted OSTs have
2518 * been cancelled. Also, the export for deleted OSTs will
2519 * be NULL and won't match the lock's export. */
2520 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2521 loi = lsm->lsm_oinfo[i];
2522 if (!lov->lov_tgts[loi->loi_ost_idx])
2524 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2525 data->lock->l_conn_export &&
2526 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2531 LDLM_ERROR(data->lock, "lock on inode without such object");
2532 dump_lsm(D_ERROR, lsm);
2533 GOTO(out, rc = -ENXIO);
2534 } else if (KEY_IS(KEY_LAST_ID)) {
2535 struct obd_id_info *info = val;
2536 __u32 size = sizeof(obd_id);
2537 struct lov_tgt_desc *tgt;
2539 LASSERT(*vallen == sizeof(struct obd_id_info));
2540 tgt = lov->lov_tgts[info->idx];
2542 if (!tgt || !tgt->ltd_active)
2543 GOTO(out, rc = -ESRCH);
2545 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2546 &size, info->data, NULL);
2548 } else if (KEY_IS(KEY_LOVDESC)) {
2549 struct lov_desc *desc_ret = val;
2550 *desc_ret = lov->desc;
2553 } else if (KEY_IS(KEY_FIEMAP)) {
2554 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2556 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2557 struct lov_tgt_desc *tgt;
2558 __u64 ost_idx = *((__u64*)val);
2560 LASSERT(*vallen == sizeof(__u64));
2561 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2562 tgt = lov->lov_tgts[ost_idx];
2564 if (!tgt || !tgt->ltd_exp)
2565 GOTO(out, rc = -ESRCH);
2567 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2569 } else if (KEY_IS(KEY_TGT_COUNT)) {
2570 *((int *)val) = lov->desc.ld_tgt_count;
2581 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2582 obd_count keylen, void *key, obd_count vallen,
2583 void *val, struct ptlrpc_request_set *set)
2585 struct obd_device *obddev = class_exp2obd(exp);
2586 struct lov_obd *lov = &obddev->u.lov;
2589 struct lov_tgt_desc *tgt;
2590 unsigned incr, check_uuid,
2591 do_inactive, no_set;
2592 unsigned next_id = 0, mds_con = 0, capa = 0;
2595 incr = check_uuid = do_inactive = no_set = 0;
2598 set = ptlrpc_prep_set();
2604 count = lov->desc.ld_tgt_count;
2606 if (KEY_IS(KEY_NEXT_ID)) {
2607 count = vallen / sizeof(struct obd_id_info);
2608 vallen = sizeof(obd_id);
2609 incr = sizeof(struct obd_id_info);
2612 } else if (KEY_IS(KEY_CHECKSUM)) {
2614 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2615 /* use defaults: do_inactive = incr = 0; */
2616 } else if (KEY_IS(KEY_MDS_CONN)) {
2618 } else if (KEY_IS(KEY_CAPA_KEY)) {
2620 } else if (KEY_IS(KEY_CACHE_SET)) {
2621 LASSERT(lov->lov_cache == NULL);
2622 lov->lov_cache = val;
2626 for (i = 0; i < count; i++, val = (char *)val + incr) {
2628 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2630 tgt = lov->lov_tgts[i];
2632 /* OST was disconnected */
2633 if (!tgt || !tgt->ltd_exp)
2636 /* OST is inactive and we don't want inactive OSCs */
2637 if (!tgt->ltd_active && !do_inactive)
2641 struct mds_group_info *mgi;
2643 LASSERT(vallen == sizeof(*mgi));
2644 mgi = (struct mds_group_info *)val;
2646 /* Only want a specific OSC */
2647 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2651 err = obd_set_info_async(env, tgt->ltd_exp,
2652 keylen, key, sizeof(int),
2654 } else if (next_id) {
2655 err = obd_set_info_async(env, tgt->ltd_exp,
2656 keylen, key, vallen,
2657 ((struct obd_id_info*)val)->data, set);
2659 struct mds_capa_info *info = (struct mds_capa_info*)val;
2661 LASSERT(vallen == sizeof(*info));
2663 /* Only want a specific OSC */
2665 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2668 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2669 key, sizeof(*info->capa),
2672 /* Only want a specific OSC */
2674 !obd_uuid_equals(val, &tgt->ltd_uuid))
2677 err = obd_set_info_async(env, tgt->ltd_exp,
2678 keylen, key, vallen, val, set);
2687 err = ptlrpc_set_wait(set);
2690 ptlrpc_set_destroy(set);
2695 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2696 int cmd, __u64 *offset)
2698 __u32 ssize = lsm->lsm_stripe_size;
2702 lov_do_div64(start, ssize);
2703 start = start * ssize;
2705 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2706 ", end "LPU64"\n", *offset, ssize, start,
2708 if (cmd == OBD_CALC_STRIPE_END) {
2709 *offset = start + ssize - 1;
2710 } else if (cmd == OBD_CALC_STRIPE_START) {
2719 void lov_stripe_lock(struct lov_stripe_md *md)
2721 LASSERT(md->lsm_lock_owner != current_pid());
2722 spin_lock(&md->lsm_lock);
2723 LASSERT(md->lsm_lock_owner == 0);
2724 md->lsm_lock_owner = current_pid();
2726 EXPORT_SYMBOL(lov_stripe_lock);
2728 void lov_stripe_unlock(struct lov_stripe_md *md)
2730 LASSERT(md->lsm_lock_owner == current_pid());
2731 md->lsm_lock_owner = 0;
2732 spin_unlock(&md->lsm_lock);
2734 EXPORT_SYMBOL(lov_stripe_unlock);
2736 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2737 struct obd_quotactl *oqctl)
2739 struct lov_obd *lov = &obd->u.lov;
2740 struct lov_tgt_desc *tgt;
2742 __u64 bhardlimit = 0;
2746 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2747 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2748 oqctl->qc_cmd != Q_GETOQUOTA &&
2749 oqctl->qc_cmd != Q_INITQUOTA &&
2750 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2751 oqctl->qc_cmd != Q_FINVALIDATE) {
2752 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2758 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2761 tgt = lov->lov_tgts[i];
2766 if (!tgt->ltd_active || tgt->ltd_reap) {
2767 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2768 lov->lov_tgts[i]->ltd_activate) {
2770 CERROR("ost %d is inactive\n", i);
2772 CDEBUG(D_HA, "ost %d is inactive\n", i);
2777 err = obd_quotactl(tgt->ltd_exp, oqctl);
2779 if (tgt->ltd_active && !rc)
2784 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2785 curspace += oqctl->qc_dqblk.dqb_curspace;
2786 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2791 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2792 oqctl->qc_dqblk.dqb_curspace = curspace;
2793 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2798 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2799 struct obd_quotactl *oqctl)
2801 struct lov_obd *lov = &obd->u.lov;
2807 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2808 if (!lov->lov_tgts[i])
2811 /* Skip quota check on the administratively disabled OSTs. */
2812 if (!lov->lov_tgts[i]->ltd_activate) {
2813 CWARN("lov idx %d was administratively disabled, "
2814 "skip quotacheck on it.\n", i);
2818 if (!lov->lov_tgts[i]->ltd_active) {
2819 CERROR("lov idx %d inactive\n", i);
2825 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2828 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2831 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2842 struct obd_ops lov_obd_ops = {
2843 .o_owner = THIS_MODULE,
2844 .o_setup = lov_setup,
2845 .o_precleanup = lov_precleanup,
2846 .o_cleanup = lov_cleanup,
2847 //.o_process_config = lov_process_config,
2848 .o_connect = lov_connect,
2849 .o_disconnect = lov_disconnect,
2850 .o_statfs = lov_statfs,
2851 .o_statfs_async = lov_statfs_async,
2852 .o_packmd = lov_packmd,
2853 .o_unpackmd = lov_unpackmd,
2854 .o_create = lov_create,
2855 .o_destroy = lov_destroy,
2856 .o_getattr = lov_getattr,
2857 .o_getattr_async = lov_getattr_async,
2858 .o_setattr = lov_setattr,
2859 .o_setattr_async = lov_setattr_async,
2861 .o_merge_lvb = lov_merge_lvb,
2862 .o_adjust_kms = lov_adjust_kms,
2863 .o_punch = lov_punch,
2865 .o_enqueue = lov_enqueue,
2866 .o_change_cbdata = lov_change_cbdata,
2867 .o_find_cbdata = lov_find_cbdata,
2868 .o_cancel = lov_cancel,
2869 .o_cancel_unused = lov_cancel_unused,
2870 .o_iocontrol = lov_iocontrol,
2871 .o_get_info = lov_get_info,
2872 .o_set_info_async = lov_set_info_async,
2873 .o_extent_calc = lov_extent_calc,
2874 .o_llog_init = lov_llog_init,
2875 .o_llog_finish = lov_llog_finish,
2876 .o_notify = lov_notify,
2877 .o_pool_new = lov_pool_new,
2878 .o_pool_rem = lov_pool_remove,
2879 .o_pool_add = lov_pool_add,
2880 .o_pool_del = lov_pool_del,
2881 .o_getref = lov_getref,
2882 .o_putref = lov_putref,
2883 .o_quotactl = lov_quotactl,
2884 .o_quotacheck = lov_quotacheck,
2887 struct kmem_cache *lov_oinfo_slab;
2889 extern struct lu_kmem_descr lov_caches[];
2891 int __init lov_init(void)
2896 /* print an address of _any_ initialized kernel symbol from this
2897 * module, to allow debugging with gdb that doesn't support data
2898 * symbols from modules.*/
2899 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2901 rc = lu_kmem_init(lov_caches);
2905 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2906 sizeof(struct lov_oinfo), 0,
2907 SLAB_HWCACHE_ALIGN, NULL);
2908 if (lov_oinfo_slab == NULL) {
2909 lu_kmem_fini(lov_caches);
2913 rc = class_register_type(&lov_obd_ops, NULL, NULL,
2914 #ifndef HAVE_ONLY_PROCFS_SEQ
2917 LUSTRE_LOV_NAME, &lov_device_type);
2920 kmem_cache_destroy(lov_oinfo_slab);
2921 lu_kmem_fini(lov_caches);
2928 static void /*__exit*/ lov_exit(void)
2930 class_unregister_type(LUSTRE_LOV_NAME);
2931 kmem_cache_destroy(lov_oinfo_slab);
2932 lu_kmem_fini(lov_caches);
2935 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2936 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2937 MODULE_LICENSE("GPL");
2939 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);