4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_obd.c
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_LOV
45 #include "../../include/linux/libcfs/libcfs.h"
47 #include "../include/obd_support.h"
48 #include "../include/lustre_lib.h"
49 #include "../include/lustre_net.h"
50 #include "../include/lustre/lustre_idl.h"
51 #include "../include/lustre_dlm.h"
52 #include "../include/lustre_mds.h"
53 #include "../include/obd_class.h"
54 #include "../include/lprocfs_status.h"
55 #include "../include/lustre_param.h"
56 #include "../include/cl_object.h"
57 #include "../include/lclient.h" /* for cl_client_lru */
58 #include "../include/lustre/ll_fiemap.h"
59 #include "../include/lustre_fid.h"
61 #include "lov_internal.h"
63 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
64 Any function that expects lov_tgts to remain stationary must take a ref. */
65 static void lov_getref(struct obd_device *obd)
67 struct lov_obd *lov = &obd->u.lov;
69 /* nobody gets through here until lov_putref is done */
70 mutex_lock(&lov->lov_lock);
71 atomic_inc(&lov->lov_refcount);
72 mutex_unlock(&lov->lov_lock);
76 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
78 static void lov_putref(struct obd_device *obd)
80 struct lov_obd *lov = &obd->u.lov;
82 mutex_lock(&lov->lov_lock);
83 /* ok to dec to 0 more than once -- ltd_exp's will be null */
84 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
87 struct lov_tgt_desc *tgt, *n;
88 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
90 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
91 tgt = lov->lov_tgts[i];
93 if (!tgt || !tgt->ltd_reap)
95 list_add(&tgt->ltd_kill, &kill);
96 /* XXX - right now there is a dependency on ld_tgt_count
97 * being the maximum tgt index for computing the
98 * mds_max_easize. So we can't shrink it. */
99 lov_ost_pool_remove(&lov->lov_packed, i);
100 lov->lov_tgts[i] = NULL;
101 lov->lov_death_row--;
103 mutex_unlock(&lov->lov_lock);
105 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
106 list_del(&tgt->ltd_kill);
108 __lov_del_obd(obd, tgt);
111 mutex_unlock(&lov->lov_lock);
115 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
116 enum obd_notify_event ev);
117 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
118 enum obd_notify_event ev, void *data);
121 #define MAX_STRING_SIZE 128
122 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
123 struct obd_connect_data *data)
125 struct lov_obd *lov = &obd->u.lov;
126 struct obd_uuid *tgt_uuid;
127 struct obd_device *tgt_obd;
128 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
129 struct obd_import *imp;
130 struct proc_dir_entry *lov_proc_dir;
133 if (!lov->lov_tgts[index])
136 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
137 tgt_obd = lov->lov_tgts[index]->ltd_obd;
139 if (!tgt_obd->obd_set_up) {
140 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
144 /* override the sp_me from lov */
145 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
147 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
148 data->ocd_index = index;
151 * Divine LOV knows that OBDs under it are OSCs.
153 imp = tgt_obd->u.cli.cl_import;
156 tgt_obd->obd_no_recov = 0;
157 /* FIXME this is probably supposed to be
158 ptlrpc_set_import_active. Horrible naming. */
159 ptlrpc_activate_import(imp);
162 rc = obd_register_observer(tgt_obd, obd);
164 CERROR("Target %s register_observer error %d\n",
165 obd_uuid2str(tgt_uuid), rc);
170 if (imp->imp_invalid) {
171 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
172 obd_uuid2str(tgt_uuid));
176 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
177 &lov_osc_uuid, data, NULL);
178 if (rc || !lov->lov_tgts[index]->ltd_exp) {
179 CERROR("Target %s connect error %d\n",
180 obd_uuid2str(tgt_uuid), rc);
184 lov->lov_tgts[index]->ltd_reap = 0;
186 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
187 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
189 lov_proc_dir = obd->obd_proc_private;
191 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
192 struct proc_dir_entry *osc_symlink;
194 LASSERT(osc_obd != NULL);
195 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
196 LASSERT(osc_obd->obd_type->typ_name != NULL);
198 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
201 osc_obd->obd_type->typ_name,
203 if (osc_symlink == NULL) {
204 CERROR("could not register LOV target /proc/fs/lustre/%s/%s/target_obds/%s.",
205 obd->obd_type->typ_name, obd->obd_name,
207 lprocfs_remove(&lov_proc_dir);
208 obd->obd_proc_private = NULL;
215 static int lov_connect(const struct lu_env *env,
216 struct obd_export **exp, struct obd_device *obd,
217 struct obd_uuid *cluuid, struct obd_connect_data *data,
220 struct lov_obd *lov = &obd->u.lov;
221 struct lov_tgt_desc *tgt;
222 struct lustre_handle conn;
225 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
227 rc = class_connect(&conn, obd, cluuid);
231 *exp = class_conn2export(&conn);
233 /* Why should there ever be more than 1 connect? */
235 LASSERT(lov->lov_connects == 1);
237 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
239 lov->lov_ocd = *data;
242 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
243 tgt = lov->lov_tgts[i];
244 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
246 /* Flags will be lowest common denominator */
247 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
249 CERROR("%s: lov connect tgt %d failed: %d\n",
250 obd->obd_name, i, rc);
253 /* connect to administrative disabled ost */
254 if (!lov->lov_tgts[i]->ltd_exp)
257 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
258 OBD_NOTIFY_CONNECT, (void *)&i);
260 CERROR("%s error sending notify %d\n",
269 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
271 struct proc_dir_entry *lov_proc_dir;
272 struct lov_obd *lov = &obd->u.lov;
273 struct obd_device *osc_obd;
276 osc_obd = class_exp2obd(tgt->ltd_exp);
277 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
278 obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
280 if (tgt->ltd_active) {
282 lov->desc.ld_active_tgt_count--;
283 tgt->ltd_exp->exp_obd->obd_inactive = 1;
287 lov_proc_dir = obd->obd_proc_private;
289 lprocfs_remove_proc_entry(osc_obd->obd_name, lov_proc_dir);
291 /* Pass it on to our clients.
292 * XXX This should be an argument to disconnect,
293 * XXX not a back-door flag on the OBD. Ah well.
295 osc_obd->obd_force = obd->obd_force;
296 osc_obd->obd_fail = obd->obd_fail;
297 osc_obd->obd_no_recov = obd->obd_no_recov;
300 obd_register_observer(osc_obd, NULL);
302 rc = obd_disconnect(tgt->ltd_exp);
304 CERROR("Target %s disconnect error %d\n",
305 tgt->ltd_uuid.uuid, rc);
313 static int lov_disconnect(struct obd_export *exp)
315 struct obd_device *obd = class_exp2obd(exp);
316 struct lov_obd *lov = &obd->u.lov;
322 /* Only disconnect the underlying layers on the final disconnect. */
324 if (lov->lov_connects != 0) {
325 /* why should there be more than 1 connect? */
326 CERROR("disconnect #%d\n", lov->lov_connects);
330 /* Let's hold another reference so lov_del_obd doesn't spin through
334 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
335 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
336 /* Disconnection is the last we know about an obd */
337 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
343 rc = class_disconnect(exp); /* bz 9811 */
349 * -EINVAL : UUID can't be found in the LOV's target list
350 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
351 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
352 * any >= 0 : is log target index
354 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
355 enum obd_notify_event ev)
357 struct lov_obd *lov = &obd->u.lov;
358 struct lov_tgt_desc *tgt;
359 int index, activate, active;
361 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
362 lov, uuid->uuid, ev);
365 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
366 tgt = lov->lov_tgts[index];
370 * LU-642, initially inactive OSC could miss the obd_connect,
371 * we make up for it here.
373 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
374 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
375 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
377 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
378 &lov_osc_uuid, &lov->lov_ocd, NULL);
383 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
384 index, obd_uuid2str(&tgt->ltd_uuid),
385 tgt->ltd_exp->exp_handle.h_cookie);
386 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
390 if (index == lov->desc.ld_tgt_count) {
395 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
396 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
398 if (lov->lov_tgts[index]->ltd_activate == activate) {
399 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
400 uuid->uuid, activate ? "" : "de");
402 lov->lov_tgts[index]->ltd_activate = activate;
403 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
404 activate ? "" : "de", obd_uuid2str(uuid));
407 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
408 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
410 if (lov->lov_tgts[index]->ltd_active == active) {
411 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
412 uuid->uuid, active ? "" : "in");
415 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
416 obd_uuid2str(uuid), active ? "" : "in");
419 lov->lov_tgts[index]->ltd_active = active;
421 lov->desc.ld_active_tgt_count++;
422 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
424 lov->desc.ld_active_tgt_count--;
425 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
428 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
436 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
437 enum obd_notify_event ev, void *data)
440 struct lov_obd *lov = &obd->u.lov;
442 down_read(&lov->lov_notify_lock);
443 if (!lov->lov_connects) {
444 up_read(&lov->lov_notify_lock);
448 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
449 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
450 struct obd_uuid *uuid;
454 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
455 up_read(&lov->lov_notify_lock);
456 CERROR("unexpected notification of %s %s!\n",
457 watched->obd_type->typ_name,
461 uuid = &watched->u.cli.cl_target_uuid;
463 /* Set OSC as active before notifying the observer, so the
464 * observer can use the OSC normally.
466 rc = lov_set_osc_active(obd, uuid, ev);
468 up_read(&lov->lov_notify_lock);
469 CERROR("event(%d) of %s failed: %d\n", ev,
470 obd_uuid2str(uuid), rc);
473 /* active event should be pass lov target index as data */
477 /* Pass the notification up the chain. */
479 rc = obd_notify_observer(obd, watched, ev, data);
481 /* NULL watched means all osc's in the lov (only for syncs) */
482 /* sync event should be send lov idx as data */
483 struct lov_obd *lov = &obd->u.lov;
487 is_sync = (ev == OBD_NOTIFY_SYNC) ||
488 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
491 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
492 if (!lov->lov_tgts[i])
495 /* don't send sync event if target not
496 * connected/activated */
497 if (is_sync && !lov->lov_tgts[i]->ltd_active)
500 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
503 CERROR("%s: notify %s of %s failed %d\n",
505 obd->obd_observer->obd_name,
506 lov->lov_tgts[i]->ltd_obd->obd_name,
513 up_read(&lov->lov_notify_lock);
517 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
518 __u32 index, int gen, int active)
520 struct lov_obd *lov = &obd->u.lov;
521 struct lov_tgt_desc *tgt;
522 struct obd_device *tgt_obd;
525 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
526 uuidp->uuid, index, gen, active);
529 CERROR("request to add OBD %s with invalid generation: %d\n",
534 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
539 mutex_lock(&lov->lov_lock);
541 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
542 tgt = lov->lov_tgts[index];
543 CERROR("UUID %s already assigned at LOV target index %d\n",
544 obd_uuid2str(&tgt->ltd_uuid), index);
545 mutex_unlock(&lov->lov_lock);
549 if (index >= lov->lov_tgt_size) {
550 /* We need to reallocate the lov target array. */
551 struct lov_tgt_desc **newtgts, **old = NULL;
552 __u32 newsize, oldsize = 0;
554 newsize = max_t(__u32, lov->lov_tgt_size, 2);
555 while (newsize < index + 1)
557 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
558 if (newtgts == NULL) {
559 mutex_unlock(&lov->lov_lock);
563 if (lov->lov_tgt_size) {
564 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
567 oldsize = lov->lov_tgt_size;
570 lov->lov_tgts = newtgts;
571 lov->lov_tgt_size = newsize;
576 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
577 lov->lov_tgts, lov->lov_tgt_size);
580 tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
582 mutex_unlock(&lov->lov_lock);
586 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
588 mutex_unlock(&lov->lov_lock);
593 tgt->ltd_uuid = *uuidp;
594 tgt->ltd_obd = tgt_obd;
595 /* XXX - add a sanity check on the generation number. */
597 tgt->ltd_index = index;
598 tgt->ltd_activate = active;
599 lov->lov_tgts[index] = tgt;
600 if (index >= lov->desc.ld_tgt_count)
601 lov->desc.ld_tgt_count = index + 1;
603 mutex_unlock(&lov->lov_lock);
605 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
606 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
608 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
610 if (lov->lov_connects == 0) {
611 /* lov_connect hasn't been called yet. We'll do the
612 lov_connect_obd on this target when that fn first runs,
613 because we don't know the connect flags yet. */
619 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
623 /* connect to administrative disabled ost */
629 if (lov->lov_cache != NULL) {
630 rc = obd_set_info_async(NULL, tgt->ltd_exp,
631 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
632 sizeof(struct cl_client_cache), lov->lov_cache,
638 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
639 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
644 CERROR("add failed (%d), deleting %s\n", rc,
645 obd_uuid2str(&tgt->ltd_uuid));
646 lov_del_target(obd, index, NULL, 0);
652 /* Schedule a target for deletion */
653 int lov_del_target(struct obd_device *obd, __u32 index,
654 struct obd_uuid *uuidp, int gen)
656 struct lov_obd *lov = &obd->u.lov;
657 int count = lov->desc.ld_tgt_count;
660 if (index >= count) {
661 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
666 /* to make sure there's no ongoing lov_notify() now */
667 down_write(&lov->lov_notify_lock);
670 if (!lov->lov_tgts[index]) {
671 CERROR("LOV target at index %d is not setup.\n", index);
676 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
677 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
678 lov_uuid2str(lov, index), index,
679 obd_uuid2str(uuidp));
684 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
685 lov_uuid2str(lov, index), index,
686 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
687 lov->lov_tgts[index]->ltd_active);
689 lov->lov_tgts[index]->ltd_reap = 1;
690 lov->lov_death_row++;
691 /* we really delete it from obd_putref */
694 up_write(&lov->lov_notify_lock);
699 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
701 struct obd_device *osc_obd;
704 LASSERT(tgt->ltd_reap);
706 osc_obd = class_exp2obd(tgt->ltd_exp);
708 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
710 osc_obd ? osc_obd->obd_name : "<no obd>");
713 lov_disconnect_obd(obd, tgt);
717 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
718 do it ourselves. And we can't do it from lov_cleanup,
719 because we just lost our only reference to it. */
721 class_manual_cleanup(osc_obd);
724 void lov_fix_desc_stripe_size(__u64 *val)
726 if (*val < LOV_MIN_STRIPE_SIZE) {
728 LCONSOLE_INFO("Increasing default stripe size to minimum %u\n",
729 LOV_DESC_STRIPE_SIZE_DEFAULT);
730 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
731 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
732 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
733 LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n",
734 *val, LOV_MIN_STRIPE_SIZE);
738 void lov_fix_desc_stripe_count(__u32 *val)
744 void lov_fix_desc_pattern(__u32 *val)
746 /* from lov_setstripe */
747 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
748 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
753 void lov_fix_desc_qos_maxage(__u32 *val)
756 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
759 void lov_fix_desc(struct lov_desc *desc)
761 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
762 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
763 lov_fix_desc_pattern(&desc->ld_pattern);
764 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
767 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
769 struct lprocfs_static_vars lvars = { NULL };
770 struct lov_desc *desc;
771 struct lov_obd *lov = &obd->u.lov;
774 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
775 CERROR("LOV setup requires a descriptor\n");
779 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
781 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
782 CERROR("descriptor size wrong: %d > %d\n",
783 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
787 if (desc->ld_magic != LOV_DESC_MAGIC) {
788 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
789 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
790 obd->obd_name, desc);
791 lustre_swab_lov_desc(desc);
793 CERROR("%s: Bad lov desc magic: %#x\n",
794 obd->obd_name, desc->ld_magic);
801 desc->ld_active_tgt_count = 0;
803 lov->lov_tgt_size = 0;
805 mutex_init(&lov->lov_lock);
806 atomic_set(&lov->lov_refcount, 0);
807 lov->lov_sp_me = LUSTRE_SP_CLI;
809 init_rwsem(&lov->lov_notify_lock);
811 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
813 HASH_POOLS_BKT_BITS, 0,
816 &pool_hash_operations,
818 INIT_LIST_HEAD(&lov->lov_pool_list);
819 lov->lov_pool_count = 0;
820 rc = lov_ost_pool_init(&lov->lov_packed, 0);
824 lprocfs_lov_init_vars(&lvars);
825 lprocfs_obd_setup(obd, lvars.obd_vars);
826 #if defined (CONFIG_PROC_FS)
830 rc1 = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
831 0444, &lov_proc_target_fops, obd);
833 CWARN("Error adding the target_obd file\n");
836 lov->lov_pool_proc_entry = lprocfs_register("pools",
846 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
848 struct lov_obd *lov = &obd->u.lov;
851 case OBD_CLEANUP_EARLY: {
853 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
854 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
856 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
868 static int lov_cleanup(struct obd_device *obd)
870 struct lov_obd *lov = &obd->u.lov;
871 struct list_head *pos, *tmp;
872 struct pool_desc *pool;
874 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
875 pool = list_entry(pos, struct pool_desc, pool_list);
876 /* free pool structs */
877 CDEBUG(D_INFO, "delete pool %p\n", pool);
878 /* In the function below, .hs_keycmp resolves to
879 * pool_hashkey_keycmp() */
880 /* coverity[overrun-buffer-val] */
881 lov_pool_del(obd, pool->pool_name);
883 cfs_hash_putref(lov->lov_pools_hash_body);
884 lov_ost_pool_free(&lov->lov_packed);
886 lprocfs_obd_cleanup(obd);
890 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
891 if (!lov->lov_tgts[i])
894 /* Inactive targets may never have connected */
895 if (lov->lov_tgts[i]->ltd_active ||
896 atomic_read(&lov->lov_refcount))
897 /* We should never get here - these
898 should have been removed in the
900 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
901 i, lov->lov_death_row,
902 atomic_read(&lov->lov_refcount));
903 lov_del_target(obd, i, NULL, 0);
906 kfree(lov->lov_tgts);
907 lov->lov_tgt_size = 0;
912 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
913 __u32 *indexp, int *genp)
915 struct obd_uuid obd_uuid;
919 switch (cmd = lcfg->lcfg_command) {
920 case LCFG_LOV_ADD_OBD:
921 case LCFG_LOV_ADD_INA:
922 case LCFG_LOV_DEL_OBD: {
925 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
926 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
931 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
933 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1) {
937 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1) {
943 if (cmd == LCFG_LOV_ADD_OBD)
944 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
945 else if (cmd == LCFG_LOV_ADD_INA)
946 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
948 rc = lov_del_target(obd, index, &obd_uuid, gen);
952 struct lprocfs_static_vars lvars = { NULL };
953 struct lov_desc *desc = &(obd->u.lov.desc);
960 lprocfs_lov_init_vars(&lvars);
962 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
975 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
985 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
986 struct lov_stripe_md **ea, struct obd_trans_info *oti)
988 struct lov_stripe_md *obj_mdp, *lsm;
989 struct lov_obd *lov = &exp->exp_obd->u.lov;
993 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
994 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
996 obj_mdp = kzalloc(sizeof(*obj_mdp), GFP_NOFS);
1000 ost_idx = src_oa->o_nlink;
1006 if (ost_idx >= lov->desc.ld_tgt_count ||
1007 !lov->lov_tgts[ost_idx]) {
1012 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1013 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1015 if (lov_oinfo_is_dummy(loi))
1018 if (loi->loi_ost_idx == ost_idx) {
1019 if (ostid_id(&loi->loi_oi) != ostid_id(&src_oa->o_oi)) {
1026 if (i == lsm->lsm_stripe_count) {
1031 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1032 src_oa, &obj_mdp, oti);
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;
1046 LASSERT(ea != NULL);
1050 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1051 src_oa->o_flags == OBD_FL_DELORPHAN) {
1052 /* should be used with LOV anymore */
1056 lov = &exp->exp_obd->u.lov;
1057 if (!lov->desc.ld_active_tgt_count)
1060 obd_getref(exp->exp_obd);
1061 /* Recreate a specific object id at the given OST index */
1062 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1063 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1064 rc = lov_recreate(exp, src_oa, ea, oti);
1067 obd_putref(exp->exp_obd);
1071 #define ASSERT_LSM_MAGIC(lsmp) \
1073 LASSERT((lsmp) != NULL); \
1074 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1075 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1076 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1079 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1080 struct obdo *oa, struct lov_stripe_md *lsm,
1081 struct obd_trans_info *oti, struct obd_export *md_exp,
1084 struct lov_request_set *set;
1085 struct obd_info oinfo;
1086 struct lov_request *req;
1087 struct list_head *pos;
1088 struct lov_obd *lov;
1089 int rc = 0, err = 0;
1091 ASSERT_LSM_MAGIC(lsm);
1093 if (!exp || !exp->exp_obd)
1096 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1098 LASSERT(oti->oti_logcookies);
1101 lov = &exp->exp_obd->u.lov;
1102 obd_getref(exp->exp_obd);
1103 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1107 list_for_each(pos, &set->set_list) {
1108 req = list_entry(pos, struct lov_request, rq_link);
1110 if (oa->o_valid & OBD_MD_FLCOOKIE)
1111 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1113 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1114 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1115 err = lov_update_common_set(set, req, err);
1117 CERROR("%s: destroying objid "DOSTID" subobj "
1118 DOSTID" on OST idx %d: rc = %d\n",
1119 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1120 POSTID(&req->rq_oi.oi_oa->o_oi),
1128 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1129 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1131 err = lov_fini_destroy_set(set);
1133 obd_putref(exp->exp_obd);
1134 return rc ? rc : err;
1137 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1140 struct lov_request_set *lovset = (struct lov_request_set *)data;
1143 /* don't do attribute merge if this async op failed */
1145 atomic_set(&lovset->set_completes, 0);
1146 err = lov_fini_getattr_set(lovset);
1147 return rc ? rc : err;
1150 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1151 struct ptlrpc_request_set *rqset)
1153 struct lov_request_set *lovset;
1154 struct lov_obd *lov;
1155 struct list_head *pos;
1156 struct lov_request *req;
1160 ASSERT_LSM_MAGIC(oinfo->oi_md);
1162 if (!exp || !exp->exp_obd)
1165 lov = &exp->exp_obd->u.lov;
1167 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1171 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1172 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1173 oinfo->oi_md->lsm_stripe_size);
1175 list_for_each(pos, &lovset->set_list) {
1176 req = list_entry(pos, struct lov_request, rq_link);
1178 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1179 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1180 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1181 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1182 &req->rq_oi, rqset);
1184 CERROR("%s: getattr objid "DOSTID" subobj"
1185 DOSTID" on OST idx %d: rc = %d\n",
1186 exp->exp_obd->obd_name,
1187 POSTID(&oinfo->oi_oa->o_oi),
1188 POSTID(&req->rq_oi.oi_oa->o_oi),
1194 if (!list_empty(&rqset->set_requests)) {
1196 LASSERT(rqset->set_interpret == NULL);
1197 rqset->set_interpret = lov_getattr_interpret;
1198 rqset->set_arg = (void *)lovset;
1203 atomic_set(&lovset->set_completes, 0);
1204 err = lov_fini_getattr_set(lovset);
1205 return rc ? rc : err;
1208 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1211 struct lov_request_set *lovset = (struct lov_request_set *)data;
1215 atomic_set(&lovset->set_completes, 0);
1216 err = lov_fini_setattr_set(lovset);
1217 return rc ? rc : err;
1220 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1221 needed. Otherwise, a client is waiting for responses. */
1222 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1223 struct obd_trans_info *oti,
1224 struct ptlrpc_request_set *rqset)
1226 struct lov_request_set *set;
1227 struct lov_request *req;
1228 struct list_head *pos;
1229 struct lov_obd *lov;
1233 ASSERT_LSM_MAGIC(oinfo->oi_md);
1234 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1236 LASSERT(oti->oti_logcookies);
1239 if (!exp || !exp->exp_obd)
1242 lov = &exp->exp_obd->u.lov;
1243 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1247 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1248 POSTID(&oinfo->oi_md->lsm_oi),
1249 oinfo->oi_md->lsm_stripe_count,
1250 oinfo->oi_md->lsm_stripe_size);
1252 list_for_each(pos, &set->set_list) {
1253 req = list_entry(pos, struct lov_request, rq_link);
1255 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1256 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1258 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1259 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1260 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1262 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1263 &req->rq_oi, oti, rqset);
1265 CERROR("error: setattr objid "DOSTID" subobj"
1266 DOSTID" on OST idx %d: rc = %d\n",
1267 POSTID(&set->set_oi->oi_oa->o_oi),
1268 POSTID(&req->rq_oi.oi_oa->o_oi),
1274 /* If we are not waiting for responses on async requests, return. */
1275 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1278 atomic_set(&set->set_completes, 0);
1279 err = lov_fini_setattr_set(set);
1280 return rc ? rc : err;
1283 LASSERT(rqset->set_interpret == NULL);
1284 rqset->set_interpret = lov_setattr_interpret;
1285 rqset->set_arg = (void *)set;
1290 /* find any ldlm lock of the inode in lov
1294 static int lov_find_cbdata(struct obd_export *exp,
1295 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1298 struct lov_obd *lov;
1301 ASSERT_LSM_MAGIC(lsm);
1303 if (!exp || !exp->exp_obd)
1306 lov = &exp->exp_obd->u.lov;
1307 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1308 struct lov_stripe_md submd;
1309 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1311 if (lov_oinfo_is_dummy(loi))
1314 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1315 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1319 submd.lsm_oi = loi->loi_oi;
1320 submd.lsm_stripe_count = 0;
1321 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1329 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1331 struct lov_request_set *lovset = (struct lov_request_set *)data;
1335 atomic_set(&lovset->set_completes, 0);
1337 err = lov_fini_statfs_set(lovset);
1338 return rc ? rc : err;
1341 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1342 __u64 max_age, struct ptlrpc_request_set *rqset)
1344 struct obd_device *obd = class_exp2obd(exp);
1345 struct lov_request_set *set;
1346 struct lov_request *req;
1347 struct list_head *pos;
1348 struct lov_obd *lov;
1351 LASSERT(oinfo != NULL);
1352 LASSERT(oinfo->oi_osfs != NULL);
1355 rc = lov_prep_statfs_set(obd, oinfo, &set);
1359 list_for_each(pos, &set->set_list) {
1360 req = list_entry(pos, struct lov_request, rq_link);
1361 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1362 &req->rq_oi, max_age, rqset);
1367 if (rc || list_empty(&rqset->set_requests)) {
1370 atomic_set(&set->set_completes, 0);
1371 err = lov_fini_statfs_set(set);
1372 return rc ? rc : err;
1375 LASSERT(rqset->set_interpret == NULL);
1376 rqset->set_interpret = lov_statfs_interpret;
1377 rqset->set_arg = (void *)set;
1381 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1382 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1384 struct ptlrpc_request_set *set = NULL;
1385 struct obd_info oinfo = { { { 0 } } };
1388 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1389 * statfs requests */
1390 set = ptlrpc_prep_set();
1394 oinfo.oi_osfs = osfs;
1395 oinfo.oi_flags = flags;
1396 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1398 rc = ptlrpc_set_wait(set);
1399 ptlrpc_set_destroy(set);
1404 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1405 void *karg, void *uarg)
1407 struct obd_device *obddev = class_exp2obd(exp);
1408 struct lov_obd *lov = &obddev->u.lov;
1409 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1410 struct obd_uuid *uuidp;
1413 case IOC_OBD_STATFS: {
1414 struct obd_ioctl_data *data = karg;
1415 struct obd_device *osc_obd;
1416 struct obd_statfs stat_buf = {0};
1420 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1421 if ((index >= count))
1424 if (!lov->lov_tgts[index])
1425 /* Try again with the next index */
1427 if (!lov->lov_tgts[index]->ltd_active)
1430 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1435 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1436 min((int) data->ioc_plen2,
1437 (int) sizeof(struct obd_uuid))))
1440 flags = uarg ? *(__u32 *)uarg : 0;
1441 /* got statfs data */
1442 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1443 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1447 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1448 min((int) data->ioc_plen1,
1449 (int) sizeof(stat_buf))))
1453 case OBD_IOC_LOV_GET_CONFIG: {
1454 struct obd_ioctl_data *data;
1455 struct lov_desc *desc;
1460 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1463 data = (struct obd_ioctl_data *)buf;
1465 if (sizeof(*desc) > data->ioc_inllen1) {
1466 obd_ioctl_freedata(buf, len);
1470 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1471 obd_ioctl_freedata(buf, len);
1475 if (sizeof(__u32) * count > data->ioc_inllen3) {
1476 obd_ioctl_freedata(buf, len);
1480 desc = (struct lov_desc *)data->ioc_inlbuf1;
1481 memcpy(desc, &(lov->desc), sizeof(*desc));
1483 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1484 genp = (__u32 *)data->ioc_inlbuf3;
1485 /* the uuid will be empty for deleted OSTs */
1486 for (i = 0; i < count; i++, uuidp++, genp++) {
1487 if (!lov->lov_tgts[i])
1489 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1490 *genp = lov->lov_tgts[i]->ltd_gen;
1493 if (copy_to_user((void *)uarg, buf, len))
1495 obd_ioctl_freedata(buf, len);
1498 case LL_IOC_LOV_GETSTRIPE:
1499 rc = lov_getstripe(exp, karg, uarg);
1501 case OBD_IOC_QUOTACTL: {
1502 struct if_quotactl *qctl = karg;
1503 struct lov_tgt_desc *tgt = NULL;
1504 struct obd_quotactl *oqctl;
1506 if (qctl->qc_valid == QC_OSTIDX) {
1507 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
1510 tgt = lov->lov_tgts[qctl->qc_idx];
1511 if (!tgt || !tgt->ltd_exp)
1513 } else if (qctl->qc_valid == QC_UUID) {
1514 for (i = 0; i < count; i++) {
1515 tgt = lov->lov_tgts[i];
1517 !obd_uuid_equals(&tgt->ltd_uuid,
1521 if (tgt->ltd_exp == NULL)
1533 LASSERT(tgt && tgt->ltd_exp);
1534 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1538 QCTL_COPY(oqctl, qctl);
1539 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1541 QCTL_COPY(qctl, oqctl);
1542 qctl->qc_valid = QC_OSTIDX;
1543 qctl->obd_uuid = tgt->ltd_uuid;
1554 for (i = 0; i < count; i++) {
1556 struct obd_device *osc_obd;
1558 /* OST was disconnected */
1559 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1562 /* ll_umount_begin() sets force flag but for lov, not
1563 * osc. Let's pass it through */
1564 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1565 osc_obd->obd_force = obddev->obd_force;
1566 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1568 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1571 if (lov->lov_tgts[i]->ltd_active) {
1572 CDEBUG(err == -ENOTTY ?
1573 D_IOCTL : D_WARNING,
1574 "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1575 lov_uuid2str(lov, i),
1592 #define FIEMAP_BUFFER_SIZE 4096
1595 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1596 * call. The local end offset and the device are sent in the first
1597 * fm_extent. This function calculates the stripe number from the index.
1598 * This function returns a stripe_no on which mapping is to be restarted.
1600 * This function returns fm_end_offset which is the in-OST offset at which
1601 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1602 * will re-calculate proper offset in next stripe.
1603 * Note that the first extent is passed to lov_get_info via the value field.
1605 * \param fiemap fiemap request header
1606 * \param lsm striping information for the file
1607 * \param fm_start logical start of mapping
1608 * \param fm_end logical end of mapping
1609 * \param start_stripe starting stripe will be returned in this
1611 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1612 struct lov_stripe_md *lsm, u64 fm_start,
1613 u64 fm_end, int *start_stripe)
1615 u64 local_end = fiemap->fm_extents[0].fe_logical;
1616 u64 lun_start, lun_end;
1618 int stripe_no = -1, i;
1620 if (fiemap->fm_extent_count == 0 ||
1621 fiemap->fm_extents[0].fe_logical == 0)
1624 /* Find out stripe_no from ost_index saved in the fe_device */
1625 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1626 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1628 if (lov_oinfo_is_dummy(oinfo))
1631 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1636 if (stripe_no == -1)
1639 /* If we have finished mapping on previous device, shift logical
1640 * offset to start of next device */
1641 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1642 &lun_start, &lun_end)) != 0 &&
1643 local_end < lun_end) {
1644 fm_end_offset = local_end;
1645 *start_stripe = stripe_no;
1647 /* This is a special value to indicate that caller should
1648 * calculate offset in next stripe. */
1650 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1653 return fm_end_offset;
1657 * We calculate on which OST the mapping will end. If the length of mapping
1658 * is greater than (stripe_size * stripe_count) then the last_stripe will
1659 * will be one just before start_stripe. Else we check if the mapping
1660 * intersects each OST and find last_stripe.
1661 * This function returns the last_stripe and also sets the stripe_count
1662 * over which the mapping is spread
1664 * \param lsm striping information for the file
1665 * \param fm_start logical start of mapping
1666 * \param fm_end logical end of mapping
1667 * \param start_stripe starting stripe of the mapping
1668 * \param stripe_count the number of stripes across which to map is returned
1670 * \retval last_stripe return the last stripe of the mapping
1672 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1673 u64 fm_end, int start_stripe,
1677 u64 obd_start, obd_end;
1680 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1681 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1683 *stripe_count = lsm->lsm_stripe_count;
1685 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1686 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1687 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1688 &obd_start, &obd_end)) == 0)
1692 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1699 * Set fe_device and copy extents from local buffer into main return buffer.
1701 * \param fiemap fiemap request header
1702 * \param lcl_fm_ext array of local fiemap extents to be copied
1703 * \param ost_index OST index to be written into the fm_device field for each
1705 * \param ext_count number of extents to be copied
1706 * \param current_extent where to start copying in main extent array
1708 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1709 struct ll_fiemap_extent *lcl_fm_ext,
1710 int ost_index, unsigned int ext_count,
1716 for (ext = 0; ext < ext_count; ext++) {
1717 lcl_fm_ext[ext].fe_device = ost_index;
1718 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1721 /* Copy fm_extent's from fm_local to return buffer */
1722 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1723 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1727 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1728 * This also handles the restarting of FIEMAP calls in case mapping overflows
1729 * the available number of extents in single call.
1731 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1732 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1734 struct ll_fiemap_info_key *fm_key = key;
1735 struct ll_user_fiemap *fiemap = val;
1736 struct ll_user_fiemap *fm_local = NULL;
1737 struct ll_fiemap_extent *lcl_fm_ext;
1739 unsigned int get_num_extents = 0;
1740 int ost_index = 0, actual_start_stripe, start_stripe;
1741 u64 fm_start, fm_end, fm_length, fm_end_offset;
1743 int current_extent = 0, rc = 0, i;
1744 int ost_eof = 0; /* EOF for object */
1745 int ost_done = 0; /* done with required mapping for this OST? */
1747 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1748 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1750 if (!lsm_has_objects(lsm)) {
1755 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1756 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1758 OBD_ALLOC_LARGE(fm_local, buffer_size);
1759 if (fm_local == NULL) {
1763 lcl_fm_ext = &fm_local->fm_extents[0];
1765 count_local = fiemap_size_to_count(buffer_size);
1767 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1768 fm_start = fiemap->fm_start;
1769 fm_length = fiemap->fm_length;
1770 /* Calculate start stripe, last stripe and length of mapping */
1771 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1772 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1773 fm_start + fm_length - 1);
1774 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1775 if (fm_end > fm_key->oa.o_size)
1776 fm_end = fm_key->oa.o_size;
1778 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1779 actual_start_stripe, &stripe_count);
1781 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1782 fm_end, &start_stripe);
1783 if (fm_end_offset == -EINVAL) {
1788 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1789 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1790 if (fiemap->fm_extent_count == 0) {
1791 get_num_extents = 1;
1794 /* Check each stripe */
1795 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1796 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1797 u64 req_fm_len; /* Stores length of required mapping */
1798 u64 len_mapped_single_call;
1799 u64 lun_start, lun_end, obd_object_end;
1800 unsigned int ext_count;
1802 cur_stripe_wrap = cur_stripe;
1804 /* Find out range of mapping on this stripe */
1805 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1806 &lun_start, &obd_object_end)) == 0)
1809 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1814 /* If this is a continuation FIEMAP call and we are on
1815 * starting stripe then lun_start needs to be set to
1817 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1818 lun_start = fm_end_offset;
1820 if (fm_length != ~0ULL) {
1821 /* Handle fm_start + fm_length overflow */
1822 if (fm_start + fm_length < fm_start)
1823 fm_length = ~0ULL - fm_start;
1824 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1830 if (lun_start == lun_end)
1833 req_fm_len = obd_object_end - lun_start;
1834 fm_local->fm_length = 0;
1835 len_mapped_single_call = 0;
1837 /* If the output buffer is very large and the objects have many
1838 * extents we may need to loop on a single OST repeatedly */
1842 if (get_num_extents == 0) {
1843 /* Don't get too many extents. */
1844 if (current_extent + count_local >
1845 fiemap->fm_extent_count)
1846 count_local = fiemap->fm_extent_count -
1850 lun_start += len_mapped_single_call;
1851 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1852 req_fm_len = fm_local->fm_length;
1853 fm_local->fm_extent_count = count_local;
1854 fm_local->fm_mapped_extents = 0;
1855 fm_local->fm_flags = fiemap->fm_flags;
1857 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1858 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1860 if (ost_index < 0 ||
1861 ost_index >= lov->desc.ld_tgt_count) {
1866 /* If OST is inactive, return extent with UNKNOWN flag */
1867 if (!lov->lov_tgts[ost_index]->ltd_active) {
1868 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1869 fm_local->fm_mapped_extents = 1;
1871 lcl_fm_ext[0].fe_logical = lun_start;
1872 lcl_fm_ext[0].fe_length = obd_object_end -
1874 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1879 fm_local->fm_start = lun_start;
1880 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1881 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1882 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1883 rc = obd_get_info(NULL,
1884 lov->lov_tgts[ost_index]->ltd_exp,
1885 keylen, key, vallen, fm_local, lsm);
1890 ext_count = fm_local->fm_mapped_extents;
1891 if (ext_count == 0) {
1893 /* If last stripe has hole at the end,
1894 * then we need to return */
1895 if (cur_stripe_wrap == last_stripe) {
1896 fiemap->fm_mapped_extents = 0;
1902 /* If we just need num of extents then go to next device */
1903 if (get_num_extents) {
1904 current_extent += ext_count;
1908 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1909 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1911 /* Have we finished mapping on this device? */
1912 if (req_fm_len <= len_mapped_single_call)
1915 /* Clear the EXTENT_LAST flag which can be present on
1917 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1918 lcl_fm_ext[ext_count - 1].fe_flags &=
1919 ~FIEMAP_EXTENT_LAST;
1921 curr_loc = lov_stripe_size(lsm,
1922 lcl_fm_ext[ext_count - 1].fe_logical+
1923 lcl_fm_ext[ext_count - 1].fe_length,
1925 if (curr_loc >= fm_key->oa.o_size)
1928 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1929 ost_index, ext_count,
1932 current_extent += ext_count;
1934 /* Ran out of available extents? */
1935 if (current_extent >= fiemap->fm_extent_count)
1937 } while (ost_done == 0 && ost_eof == 0);
1939 if (cur_stripe_wrap == last_stripe)
1944 /* Indicate that we are returning device offsets unless file just has
1946 if (lsm->lsm_stripe_count > 1)
1947 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1949 if (get_num_extents)
1950 goto skip_last_device_calc;
1952 /* Check if we have reached the last stripe and whether mapping for that
1953 * stripe is done. */
1954 if (cur_stripe_wrap == last_stripe) {
1955 if (ost_done || ost_eof)
1956 fiemap->fm_extents[current_extent - 1].fe_flags |=
1960 skip_last_device_calc:
1961 fiemap->fm_mapped_extents = current_extent;
1964 OBD_FREE_LARGE(fm_local, buffer_size);
1968 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1969 __u32 keylen, void *key, __u32 *vallen, void *val,
1970 struct lov_stripe_md *lsm)
1972 struct obd_device *obddev = class_exp2obd(exp);
1973 struct lov_obd *lov = &obddev->u.lov;
1976 if (!vallen || !val)
1981 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
1984 struct ldlm_lock *lock;
1986 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
1987 struct lov_oinfo *loi;
1988 __u32 *stripe = val;
1990 if (*vallen < sizeof(*stripe)) {
1994 *vallen = sizeof(*stripe);
1996 /* XXX This is another one of those bits that will need to
1997 * change if we ever actually support nested LOVs. It uses
1998 * the lock's export to find out which stripe it is. */
1999 /* XXX - it's assumed all the locks for deleted OSTs have
2000 * been cancelled. Also, the export for deleted OSTs will
2001 * be NULL and won't match the lock's export. */
2002 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2003 loi = lsm->lsm_oinfo[i];
2004 if (lov_oinfo_is_dummy(loi))
2007 if (!lov->lov_tgts[loi->loi_ost_idx])
2009 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2010 data->lock->l_conn_export &&
2011 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2017 LDLM_ERROR(data->lock, "lock on inode without such object");
2018 dump_lsm(D_ERROR, lsm);
2021 } else if (KEY_IS(KEY_LAST_ID)) {
2022 struct obd_id_info *info = val;
2023 __u32 size = sizeof(u64);
2024 struct lov_tgt_desc *tgt;
2026 LASSERT(*vallen == sizeof(struct obd_id_info));
2027 tgt = lov->lov_tgts[info->idx];
2029 if (!tgt || !tgt->ltd_active) {
2034 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2035 &size, info->data, NULL);
2038 } else if (KEY_IS(KEY_LOVDESC)) {
2039 struct lov_desc *desc_ret = val;
2040 *desc_ret = lov->desc;
2044 } else if (KEY_IS(KEY_FIEMAP)) {
2045 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2047 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2048 struct lov_tgt_desc *tgt;
2049 __u64 ost_idx = *((__u64 *)val);
2051 LASSERT(*vallen == sizeof(__u64));
2052 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2053 tgt = lov->lov_tgts[ost_idx];
2055 if (!tgt || !tgt->ltd_exp) {
2060 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2063 } else if (KEY_IS(KEY_TGT_COUNT)) {
2064 *((int *)val) = lov->desc.ld_tgt_count;
2076 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2077 u32 keylen, void *key, u32 vallen,
2078 void *val, struct ptlrpc_request_set *set)
2080 struct obd_device *obddev = class_exp2obd(exp);
2081 struct lov_obd *lov = &obddev->u.lov;
2084 struct lov_tgt_desc *tgt;
2085 unsigned incr, check_uuid,
2086 do_inactive, no_set;
2087 unsigned next_id = 0, mds_con = 0, capa = 0;
2089 incr = check_uuid = do_inactive = no_set = 0;
2092 set = ptlrpc_prep_set();
2098 count = lov->desc.ld_tgt_count;
2100 if (KEY_IS(KEY_NEXT_ID)) {
2101 count = vallen / sizeof(struct obd_id_info);
2102 vallen = sizeof(u64);
2103 incr = sizeof(struct obd_id_info);
2106 } else if (KEY_IS(KEY_CHECKSUM)) {
2108 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2109 /* use defaults: do_inactive = incr = 0; */
2110 } else if (KEY_IS(KEY_MDS_CONN)) {
2112 } else if (KEY_IS(KEY_CAPA_KEY)) {
2114 } else if (KEY_IS(KEY_CACHE_SET)) {
2115 LASSERT(lov->lov_cache == NULL);
2116 lov->lov_cache = val;
2120 for (i = 0; i < count; i++, val = (char *)val + incr) {
2122 tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx];
2124 tgt = lov->lov_tgts[i];
2126 /* OST was disconnected */
2127 if (!tgt || !tgt->ltd_exp)
2130 /* OST is inactive and we don't want inactive OSCs */
2131 if (!tgt->ltd_active && !do_inactive)
2135 struct mds_group_info *mgi;
2137 LASSERT(vallen == sizeof(*mgi));
2138 mgi = (struct mds_group_info *)val;
2140 /* Only want a specific OSC */
2141 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2145 err = obd_set_info_async(env, tgt->ltd_exp,
2146 keylen, key, sizeof(int),
2148 } else if (next_id) {
2149 err = obd_set_info_async(env, tgt->ltd_exp,
2150 keylen, key, vallen,
2151 ((struct obd_id_info *)val)->data, set);
2153 struct mds_capa_info *info = (struct mds_capa_info *)val;
2155 LASSERT(vallen == sizeof(*info));
2157 /* Only want a specific OSC */
2159 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2162 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2163 key, sizeof(*info->capa),
2166 /* Only want a specific OSC */
2168 !obd_uuid_equals(val, &tgt->ltd_uuid))
2171 err = obd_set_info_async(env, tgt->ltd_exp,
2172 keylen, key, vallen, val, set);
2181 err = ptlrpc_set_wait(set);
2184 ptlrpc_set_destroy(set);
2189 void lov_stripe_lock(struct lov_stripe_md *md)
2190 __acquires(&md->lsm_lock)
2192 LASSERT(md->lsm_lock_owner != current_pid());
2193 spin_lock(&md->lsm_lock);
2194 LASSERT(md->lsm_lock_owner == 0);
2195 md->lsm_lock_owner = current_pid();
2197 EXPORT_SYMBOL(lov_stripe_lock);
2199 void lov_stripe_unlock(struct lov_stripe_md *md)
2200 __releases(&md->lsm_lock)
2202 LASSERT(md->lsm_lock_owner == current_pid());
2203 md->lsm_lock_owner = 0;
2204 spin_unlock(&md->lsm_lock);
2206 EXPORT_SYMBOL(lov_stripe_unlock);
2208 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2209 struct obd_quotactl *oqctl)
2211 struct lov_obd *lov = &obd->u.lov;
2212 struct lov_tgt_desc *tgt;
2214 __u64 bhardlimit = 0;
2217 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2218 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2219 oqctl->qc_cmd != Q_GETOQUOTA &&
2220 oqctl->qc_cmd != Q_INITQUOTA &&
2221 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2222 oqctl->qc_cmd != Q_FINVALIDATE) {
2223 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2229 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2232 tgt = lov->lov_tgts[i];
2237 if (!tgt->ltd_active || tgt->ltd_reap) {
2238 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2239 lov->lov_tgts[i]->ltd_activate) {
2241 CERROR("ost %d is inactive\n", i);
2243 CDEBUG(D_HA, "ost %d is inactive\n", i);
2248 err = obd_quotactl(tgt->ltd_exp, oqctl);
2250 if (tgt->ltd_active && !rc)
2255 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2256 curspace += oqctl->qc_dqblk.dqb_curspace;
2257 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2262 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2263 oqctl->qc_dqblk.dqb_curspace = curspace;
2264 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2269 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2270 struct obd_quotactl *oqctl)
2272 struct lov_obd *lov = &obd->u.lov;
2277 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2278 if (!lov->lov_tgts[i])
2281 /* Skip quota check on the administratively disabled OSTs. */
2282 if (!lov->lov_tgts[i]->ltd_activate) {
2283 CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
2288 if (!lov->lov_tgts[i]->ltd_active) {
2289 CERROR("lov idx %d inactive\n", i);
2295 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2298 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2301 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2312 static struct obd_ops lov_obd_ops = {
2313 .o_owner = THIS_MODULE,
2314 .o_setup = lov_setup,
2315 .o_precleanup = lov_precleanup,
2316 .o_cleanup = lov_cleanup,
2317 /*.o_process_config = lov_process_config,*/
2318 .o_connect = lov_connect,
2319 .o_disconnect = lov_disconnect,
2320 .o_statfs = lov_statfs,
2321 .o_statfs_async = lov_statfs_async,
2322 .o_packmd = lov_packmd,
2323 .o_unpackmd = lov_unpackmd,
2324 .o_create = lov_create,
2325 .o_destroy = lov_destroy,
2326 .o_getattr_async = lov_getattr_async,
2327 .o_setattr_async = lov_setattr_async,
2328 .o_adjust_kms = lov_adjust_kms,
2329 .o_find_cbdata = lov_find_cbdata,
2330 .o_iocontrol = lov_iocontrol,
2331 .o_get_info = lov_get_info,
2332 .o_set_info_async = lov_set_info_async,
2333 .o_notify = lov_notify,
2334 .o_pool_new = lov_pool_new,
2335 .o_pool_rem = lov_pool_remove,
2336 .o_pool_add = lov_pool_add,
2337 .o_pool_del = lov_pool_del,
2338 .o_getref = lov_getref,
2339 .o_putref = lov_putref,
2340 .o_quotactl = lov_quotactl,
2341 .o_quotacheck = lov_quotacheck,
2344 struct kmem_cache *lov_oinfo_slab;
2346 static int __init lov_init(void)
2348 struct lprocfs_static_vars lvars = { NULL };
2351 /* print an address of _any_ initialized kernel symbol from this
2352 * module, to allow debugging with gdb that doesn't support data
2353 * symbols from modules.*/
2354 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2356 rc = lu_kmem_init(lov_caches);
2360 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2361 sizeof(struct lov_oinfo),
2362 0, SLAB_HWCACHE_ALIGN, NULL);
2363 if (lov_oinfo_slab == NULL) {
2364 lu_kmem_fini(lov_caches);
2367 lprocfs_lov_init_vars(&lvars);
2369 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2370 LUSTRE_LOV_NAME, &lov_device_type);
2373 kmem_cache_destroy(lov_oinfo_slab);
2374 lu_kmem_fini(lov_caches);
2380 static void /*__exit*/ lov_exit(void)
2382 class_unregister_type(LUSTRE_LOV_NAME);
2383 kmem_cache_destroy(lov_oinfo_slab);
2385 lu_kmem_fini(lov_caches);
2388 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2389 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2390 MODULE_LICENSE("GPL");
2391 MODULE_VERSION(LUSTRE_VERSION_STRING);
2393 module_init(lov_init);
2394 module_exit(lov_exit);