1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
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>
45 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_LOV
49 #include <libcfs/libcfs.h>
51 #include <liblustre.h>
54 #include <obd_support.h>
55 #include <lustre_lib.h>
56 #include <lustre_net.h>
57 #include <lustre/lustre_idl.h>
58 #include <lustre_dlm.h>
59 #include <lustre_mds.h>
60 #include <lustre_debug.h>
61 #include <obd_class.h>
64 #include <lprocfs_status.h>
65 #include <lustre_param.h>
66 #include <cl_object.h>
67 #include <lustre/ll_fiemap.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 cfs_mutex_down(&lov->lov_lock);
79 cfs_atomic_inc(&lov->lov_refcount);
80 cfs_mutex_up(&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 cfs_mutex_down(&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 cfs_mutex_up(&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 cfs_mutex_up(&lov->lov_lock);
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
126 enum obd_notify_event ev, void *data);
129 #define MAX_STRING_SIZE 128
130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
131 struct obd_connect_data *data)
133 struct lov_obd *lov = &obd->u.lov;
134 struct obd_uuid *tgt_uuid;
135 struct obd_device *tgt_obd;
136 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
137 struct obd_import *imp;
139 cfs_proc_dir_entry_t *lov_proc_dir;
144 if (!lov->lov_tgts[index])
147 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
148 tgt_obd = lov->lov_tgts[index]->ltd_obd;
150 if (!tgt_obd->obd_set_up) {
151 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
155 /* override the sp_me from lov */
156 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
158 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
159 data->ocd_index = index;
162 * Divine LOV knows that OBDs under it are OSCs.
164 imp = tgt_obd->u.cli.cl_import;
167 tgt_obd->obd_no_recov = 0;
168 /* FIXME this is probably supposed to be
169 ptlrpc_set_import_active. Horrible naming. */
170 ptlrpc_activate_import(imp);
173 rc = obd_register_observer(tgt_obd, obd);
175 CERROR("Target %s register_observer error %d\n",
176 obd_uuid2str(tgt_uuid), rc);
181 if (imp->imp_invalid) {
182 CERROR("not connecting OSC %s; administratively "
183 "disabled\n", obd_uuid2str(tgt_uuid));
187 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
188 &lov_osc_uuid, data, NULL);
189 if (rc || !lov->lov_tgts[index]->ltd_exp) {
190 CERROR("Target %s connect error %d\n",
191 obd_uuid2str(tgt_uuid), rc);
195 lov->lov_tgts[index]->ltd_reap = 0;
197 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
198 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
201 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
203 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
204 cfs_proc_dir_entry_t *osc_symlink;
206 LASSERT(osc_obd != NULL);
207 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
208 LASSERT(osc_obd->obd_type->typ_name != NULL);
210 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
213 osc_obd->obd_type->typ_name,
215 if (osc_symlink == NULL) {
216 CERROR("could not register LOV target "
217 "/proc/fs/lustre/%s/%s/target_obds/%s.",
218 obd->obd_type->typ_name, obd->obd_name,
220 lprocfs_remove(&lov_proc_dir);
225 rc = qos_add_tgt(obd, index);
227 CERROR("qos_add_tgt failed %d\n", rc);
232 static int lov_connect(const struct lu_env *env,
233 struct obd_export **exp, struct obd_device *obd,
234 struct obd_uuid *cluuid, struct obd_connect_data *data,
237 struct lov_obd *lov = &obd->u.lov;
238 struct lov_tgt_desc *tgt;
239 struct lustre_handle conn;
243 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
245 rc = class_connect(&conn, obd, cluuid);
249 *exp = class_conn2export(&conn);
251 /* Why should there ever be more than 1 connect? */
253 LASSERT(lov->lov_connects == 1);
255 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
257 lov->lov_ocd = *data;
260 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
261 tgt = lov->lov_tgts[i];
262 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
264 /* Flags will be lowest common denominator */
265 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
267 CERROR("%s: lov connect tgt %d failed: %d\n",
268 obd->obd_name, i, rc);
271 /* connect to administrative disabled ost */
272 if (!lov->lov_tgts[i]->ltd_exp)
275 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
276 OBD_NOTIFY_CONNECT, (void *)&i);
278 CERROR("%s error sending notify %d\n",
287 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
289 cfs_proc_dir_entry_t *lov_proc_dir;
290 struct lov_obd *lov = &obd->u.lov;
291 struct obd_device *osc_obd;
295 osc_obd = class_exp2obd(tgt->ltd_exp);
296 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
297 obd->obd_name, osc_obd->obd_name);
299 if (tgt->ltd_active) {
301 lov->desc.ld_active_tgt_count--;
302 tgt->ltd_exp->exp_obd->obd_inactive = 1;
305 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
307 cfs_proc_dir_entry_t *osc_symlink;
309 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
311 lprocfs_remove(&osc_symlink);
313 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
314 obd->obd_type->typ_name, obd->obd_name,
320 /* Pass it on to our clients.
321 * XXX This should be an argument to disconnect,
322 * XXX not a back-door flag on the OBD. Ah well.
324 osc_obd->obd_force = obd->obd_force;
325 osc_obd->obd_fail = obd->obd_fail;
326 osc_obd->obd_no_recov = obd->obd_no_recov;
329 obd_register_observer(osc_obd, NULL);
331 rc = obd_disconnect(tgt->ltd_exp);
333 CERROR("Target %s disconnect error %d\n",
334 tgt->ltd_uuid.uuid, rc);
338 qos_del_tgt(obd, tgt);
344 static int lov_disconnect(struct obd_export *exp)
346 struct obd_device *obd = class_exp2obd(exp);
347 struct lov_obd *lov = &obd->u.lov;
354 /* Only disconnect the underlying layers on the final disconnect. */
356 if (lov->lov_connects != 0) {
357 /* why should there be more than 1 connect? */
358 CERROR("disconnect #%d\n", lov->lov_connects);
362 /* Let's hold another reference so lov_del_obd doesn't spin through
366 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
367 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
368 /* Disconnection is the last we know about an obd */
369 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
375 rc = class_disconnect(exp); /* bz 9811 */
381 * -EINVAL : UUID can't be found in the LOV's target list
382 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
383 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
384 * any >= 0 : is log target index
386 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
389 struct lov_obd *lov = &obd->u.lov;
390 struct lov_tgt_desc *tgt;
394 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
395 lov, uuid->uuid, activate);
398 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
399 tgt = lov->lov_tgts[index];
400 if (!tgt || !tgt->ltd_exp)
403 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
404 index, obd_uuid2str(&tgt->ltd_uuid),
405 tgt->ltd_exp->exp_handle.h_cookie);
406 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
410 if (index == lov->desc.ld_tgt_count)
411 GOTO(out, index = -EINVAL);
413 if (lov->lov_tgts[index]->ltd_active == activate) {
414 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
415 activate ? "" : "in");
419 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
420 activate ? "" : "in");
422 lov->lov_tgts[index]->ltd_active = activate;
425 lov->desc.ld_active_tgt_count++;
426 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
428 lov->desc.ld_active_tgt_count--;
429 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
431 /* remove any old qos penalty */
432 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
439 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
440 enum obd_notify_event ev, void *data)
445 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
446 struct obd_uuid *uuid;
450 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
451 CERROR("unexpected notification of %s %s!\n",
452 watched->obd_type->typ_name,
456 uuid = &watched->u.cli.cl_target_uuid;
458 /* Set OSC as active before notifying the observer, so the
459 * observer can use the OSC normally.
461 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
463 CERROR("%sactivation of %s failed: %d\n",
464 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
465 obd_uuid2str(uuid), rc);
468 /* active event should be pass lov target index as data */
472 /* Pass the notification up the chain. */
474 rc = obd_notify_observer(obd, watched, ev, data);
476 /* NULL watched means all osc's in the lov (only for syncs) */
477 /* sync event should be send lov idx as data */
478 struct lov_obd *lov = &obd->u.lov;
482 is_sync = (ev == OBD_NOTIFY_SYNC) ||
483 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
486 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
487 if (!lov->lov_tgts[i])
490 /* don't send sync event if target not
491 * connected/activated */
492 if (is_sync && !lov->lov_tgts[i]->ltd_active)
495 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
498 CERROR("%s: notify %s of %s failed %d\n",
500 obd->obd_observer->obd_name,
501 lov->lov_tgts[i]->ltd_obd->obd_name,
511 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
512 __u32 index, int gen, int active)
514 struct lov_obd *lov = &obd->u.lov;
515 struct lov_tgt_desc *tgt;
516 struct obd_device *tgt_obd;
520 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
521 uuidp->uuid, index, gen, active);
524 CERROR("request to add OBD %s with invalid generation: %d\n",
529 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
534 cfs_mutex_down(&lov->lov_lock);
536 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
537 tgt = lov->lov_tgts[index];
538 CERROR("UUID %s already assigned at LOV target index %d\n",
539 obd_uuid2str(&tgt->ltd_uuid), index);
540 cfs_mutex_up(&lov->lov_lock);
544 if (index >= lov->lov_tgt_size) {
545 /* We need to reallocate the lov target array. */
546 struct lov_tgt_desc **newtgts, **old = NULL;
547 __u32 newsize, oldsize = 0;
549 newsize = max(lov->lov_tgt_size, (__u32)2);
550 while (newsize < index + 1)
551 newsize = newsize << 1;
552 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
553 if (newtgts == NULL) {
554 cfs_mutex_up(&lov->lov_lock);
558 if (lov->lov_tgt_size) {
559 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
562 oldsize = lov->lov_tgt_size;
565 lov->lov_tgts = newtgts;
566 lov->lov_tgt_size = newsize;
571 OBD_FREE(old, sizeof(*old) * oldsize);
573 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
574 lov->lov_tgts, lov->lov_tgt_size);
579 cfs_mutex_up(&lov->lov_lock);
583 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
585 cfs_mutex_up(&lov->lov_lock);
590 tgt->ltd_uuid = *uuidp;
591 tgt->ltd_obd = tgt_obd;
592 /* XXX - add a sanity check on the generation number. */
594 tgt->ltd_index = index;
595 tgt->ltd_activate = active;
596 lov->lov_tgts[index] = tgt;
597 if (index >= lov->desc.ld_tgt_count)
598 lov->desc.ld_tgt_count = index + 1;
600 cfs_mutex_up(&lov->lov_lock);
602 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
603 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
605 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
607 if (lov->lov_connects == 0) {
608 /* lov_connect hasn't been called yet. We'll do the
609 lov_connect_obd on this target when that fn first runs,
610 because we don't know the connect flags yet. */
616 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
620 /* connect to administrative disabled ost */
624 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
625 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
630 CERROR("add failed (%d), deleting %s\n", rc,
631 obd_uuid2str(&tgt->ltd_uuid));
632 lov_del_target(obd, index, 0, 0);
638 /* Schedule a target for deletion */
639 int lov_del_target(struct obd_device *obd, __u32 index,
640 struct obd_uuid *uuidp, int gen)
642 struct lov_obd *lov = &obd->u.lov;
643 int count = lov->desc.ld_tgt_count;
647 if (index >= count) {
648 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
655 if (!lov->lov_tgts[index]) {
656 CERROR("LOV target at index %d is not setup.\n", index);
657 GOTO(out, rc = -EINVAL);
660 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
661 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
662 lov_uuid2str(lov, index), index,
663 obd_uuid2str(uuidp));
664 GOTO(out, rc = -EINVAL);
667 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
668 lov_uuid2str(lov, index), index,
669 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
670 lov->lov_tgts[index]->ltd_active);
672 lov->lov_tgts[index]->ltd_reap = 1;
673 lov->lov_death_row++;
674 /* we really delete it from obd_putref */
681 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
683 struct obd_device *osc_obd;
686 LASSERT(tgt->ltd_reap);
688 osc_obd = class_exp2obd(tgt->ltd_exp);
690 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
692 osc_obd ? osc_obd->obd_name : "<no obd>");
695 lov_disconnect_obd(obd, tgt);
699 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
700 do it ourselves. And we can't do it from lov_cleanup,
701 because we just lost our only reference to it. */
703 class_manual_cleanup(osc_obd);
706 void lov_fix_desc_stripe_size(__u64 *val)
708 if (*val < PTLRPC_MAX_BRW_SIZE) {
709 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
710 PTLRPC_MAX_BRW_SIZE);
711 *val = PTLRPC_MAX_BRW_SIZE;
712 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
713 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
714 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
716 *val, LOV_MIN_STRIPE_SIZE);
720 void lov_fix_desc_stripe_count(__u32 *val)
726 void lov_fix_desc_pattern(__u32 *val)
728 /* from lov_setstripe */
729 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
730 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
735 void lov_fix_desc_qos_maxage(__u32 *val)
739 *val = QOS_DEFAULT_MAXAGE;
742 void lov_fix_desc(struct lov_desc *desc)
744 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
745 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
746 lov_fix_desc_pattern(&desc->ld_pattern);
747 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
750 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
752 struct lprocfs_static_vars lvars = { 0 };
753 struct lov_desc *desc;
754 struct lov_obd *lov = &obd->u.lov;
758 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
759 CERROR("LOV setup requires a descriptor\n");
763 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
765 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
766 CERROR("descriptor size wrong: %d > %d\n",
767 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
771 if (desc->ld_magic != LOV_DESC_MAGIC) {
772 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
773 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
774 obd->obd_name, desc);
775 lustre_swab_lov_desc(desc);
777 CERROR("%s: Bad lov desc magic: %#x\n",
778 obd->obd_name, desc->ld_magic);
785 desc->ld_active_tgt_count = 0;
787 lov->lov_tgt_size = 0;
789 cfs_sema_init(&lov->lov_lock, 1);
790 cfs_atomic_set(&lov->lov_refcount, 0);
791 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
792 cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
793 lov->lov_sp_me = LUSTRE_SP_CLI;
794 lov->lov_qos.lq_dirty = 1;
795 lov->lov_qos.lq_rr.lqr_dirty = 1;
796 lov->lov_qos.lq_reset = 1;
797 /* Default priority is toward free space balance */
798 lov->lov_qos.lq_prio_free = 232;
799 /* Default threshold for rr (roughly 17%) */
800 lov->lov_qos.lq_threshold_rr = 43;
801 /* Init statfs fields */
802 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
803 if (NULL == lov->lov_qos.lq_statfs_data)
805 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
807 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
809 HASH_POOLS_BKT_BITS, 0,
812 &pool_hash_operations,
814 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
815 lov->lov_pool_count = 0;
816 rc = lov_ost_pool_init(&lov->lov_packed, 0);
819 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
821 lov_ost_pool_free(&lov->lov_packed);
825 lprocfs_lov_init_vars(&lvars);
826 lprocfs_obd_setup(obd, lvars.obd_vars);
831 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
832 0444, &lov_proc_target_fops, obd);
834 CWARN("Error adding the target_obd file\n");
837 lov->lov_pool_proc_entry = lprocfs_register("pools",
844 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
847 struct lov_obd *lov = &obd->u.lov;
852 case OBD_CLEANUP_EARLY: {
854 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
855 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
857 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
862 case OBD_CLEANUP_EXPORTS:
863 rc = obd_llog_finish(obd, 0);
865 CERROR("failed to cleanup llogging subsystems\n");
871 static int lov_cleanup(struct obd_device *obd)
873 struct lov_obd *lov = &obd->u.lov;
874 cfs_list_t *pos, *tmp;
875 struct pool_desc *pool;
877 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
878 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
879 /* free pool structs */
880 CDEBUG(D_INFO, "delete pool %p\n", pool);
881 lov_pool_del(obd, pool->pool_name);
883 cfs_hash_putref(lov->lov_pools_hash_body);
884 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
885 lov_ost_pool_free(&lov->lov_packed);
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 cfs_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!"
901 " deathrow=%d, lovrc=%d\n",
902 i, lov->lov_death_row,
903 cfs_atomic_read(&lov->lov_refcount));
904 lov_del_target(obd, i, 0, 0);
907 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
909 lov->lov_tgt_size = 0;
912 /* clear pools parent proc entry only after all pools is killed */
913 lprocfs_obd_cleanup(obd);
915 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
919 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
920 __u32 *indexp, int *genp)
922 struct obd_uuid obd_uuid;
927 switch(cmd = lcfg->lcfg_command) {
928 case LCFG_LOV_ADD_OBD:
929 case LCFG_LOV_ADD_INA:
930 case LCFG_LOV_DEL_OBD: {
933 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
934 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
935 GOTO(out, rc = -EINVAL);
937 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
939 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
940 GOTO(out, rc = -EINVAL);
941 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
942 GOTO(out, rc = -EINVAL);
945 if (cmd == LCFG_LOV_ADD_OBD)
946 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
947 else if (cmd == LCFG_LOV_ADD_INA)
948 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
950 rc = lov_del_target(obd, index, &obd_uuid, gen);
954 struct lprocfs_static_vars lvars = { 0 };
955 struct lov_desc *desc = &(obd->u.lov.desc);
958 GOTO(out, rc = -EINVAL);
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);
976 GOTO(out, rc = -EINVAL);
985 #define log2(n) cfs_ffz(~(n))
988 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
989 struct lov_stripe_md **ea,
990 struct obd_trans_info *oti)
994 struct obd_uuid *ost_uuid = NULL;
998 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
999 src_oa->o_flags == OBD_FL_DELORPHAN);
1001 lov = &export->exp_obd->u.lov;
1007 if (oti->oti_ost_uuid) {
1008 ost_uuid = oti->oti_ost_uuid;
1009 CDEBUG(D_HA, "clearing orphans only for %s\n",
1013 obd_getref(export->exp_obd);
1014 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1015 struct lov_stripe_md obj_md;
1016 struct lov_stripe_md *obj_mdp = &obj_md;
1017 struct lov_tgt_desc *tgt;
1020 tgt = lov->lov_tgts[i];
1024 /* if called for a specific target, we don't
1025 care if it is not active. */
1026 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1027 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1031 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1034 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1035 obd_uuid2str(ost_uuid));
1037 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1039 LASSERT(lov->lov_tgts[i]->ltd_exp);
1040 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1041 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1042 tmp_oa, &obj_mdp, oti);
1044 /* This export will be disabled until it is recovered,
1045 and then orphan recovery will be completed. */
1046 CERROR("error in orphan recovery on OST idx %d/%d: "
1047 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1054 obd_putref(export->exp_obd);
1060 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1061 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1063 struct lov_stripe_md *obj_mdp, *lsm;
1064 struct lov_obd *lov = &exp->exp_obd->u.lov;
1069 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1070 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1072 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1073 if (obj_mdp == NULL)
1076 ost_idx = src_oa->o_nlink;
1079 GOTO(out, rc = -EINVAL);
1080 if (ost_idx >= lov->desc.ld_tgt_count ||
1081 !lov->lov_tgts[ost_idx])
1082 GOTO(out, rc = -EINVAL);
1084 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1085 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1086 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1087 GOTO(out, rc = -EINVAL);
1091 if (i == lsm->lsm_stripe_count)
1092 GOTO(out, rc = -EINVAL);
1094 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1096 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1100 /* the LOV expects oa->o_id to be set to the LOV object id */
1101 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1102 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1104 struct lov_obd *lov;
1105 struct obd_info oinfo;
1106 struct lov_request_set *set = NULL;
1107 struct lov_request *req;
1108 struct l_wait_info lwi = { 0 };
1112 LASSERT(ea != NULL);
1116 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1117 src_oa->o_flags == OBD_FL_DELORPHAN) {
1118 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1122 lov = &exp->exp_obd->u.lov;
1123 if (!lov->desc.ld_active_tgt_count)
1126 obd_getref(exp->exp_obd);
1127 /* Recreate a specific object id at the given OST index */
1128 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1129 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1130 rc = lov_recreate(exp, src_oa, ea, oti);
1134 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1135 * later in alloc_qos(), we will wait for those rpcs to complete if
1136 * the osfs age is older than 2 * qos_maxage */
1137 qos_statfs_update(exp->exp_obd,
1138 cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1139 OBD_STATFS_CACHE_SECONDS),
1142 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1146 cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1147 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1148 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1149 &req->rq_oi, &req->rq_oi.oi_md, oti);
1152 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1153 * longer then this */
1154 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1156 /* we not have ptlrpc set for assign set->interpret and should
1157 * be call interpret function himself. calling from cb_create_update
1158 * not permited because lov_fini_create_set can sleep for long time,
1159 * but we must avoid sleeping in ptlrpcd interpret function. */
1160 rc = lov_fini_create_set(set, ea);
1162 obd_putref(exp->exp_obd);
1166 #define ASSERT_LSM_MAGIC(lsmp) \
1168 LASSERT((lsmp) != NULL); \
1169 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1170 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1171 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1174 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1175 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1176 struct obd_export *md_exp, void *capa)
1178 struct lov_request_set *set;
1179 struct obd_info oinfo;
1180 struct lov_request *req;
1182 struct lov_obd *lov;
1183 int rc = 0, err = 0;
1186 ASSERT_LSM_MAGIC(lsm);
1188 if (!exp || !exp->exp_obd)
1191 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1193 LASSERT(oti->oti_logcookies);
1196 lov = &exp->exp_obd->u.lov;
1197 obd_getref(exp->exp_obd);
1198 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1202 cfs_list_for_each (pos, &set->set_list) {
1203 req = cfs_list_entry(pos, struct lov_request, rq_link);
1205 if (oa->o_valid & OBD_MD_FLCOOKIE)
1206 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1208 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1209 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1210 err = lov_update_common_set(set, req, err);
1212 CERROR("error: destroying objid "LPX64" subobj "
1213 LPX64" on OST idx %d: rc = %d\n",
1214 oa->o_id, req->rq_oi.oi_oa->o_id,
1222 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1223 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1225 err = lov_fini_destroy_set(set);
1227 obd_putref(exp->exp_obd);
1228 RETURN(rc ? rc : err);
1231 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1233 struct lov_request_set *set;
1234 struct lov_request *req;
1236 struct lov_obd *lov;
1237 int err = 0, rc = 0;
1241 ASSERT_LSM_MAGIC(oinfo->oi_md);
1243 if (!exp || !exp->exp_obd)
1246 lov = &exp->exp_obd->u.lov;
1248 rc = lov_prep_getattr_set(exp, oinfo, &set);
1252 cfs_list_for_each (pos, &set->set_list) {
1253 req = cfs_list_entry(pos, struct lov_request, rq_link);
1255 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1256 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1257 req->rq_oi.oi_oa->o_id, req->rq_idx);
1259 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1261 err = lov_update_common_set(set, req, rc);
1263 CERROR("error: getattr objid "LPX64" subobj "
1264 LPX64" on OST idx %d: rc = %d\n",
1265 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1271 rc = lov_fini_getattr_set(set);
1277 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1280 struct lov_request_set *lovset = (struct lov_request_set *)data;
1284 /* don't do attribute merge if this aysnc op failed */
1286 lovset->set_completes = 0;
1287 err = lov_fini_getattr_set(lovset);
1288 RETURN(rc ? rc : err);
1291 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1292 struct ptlrpc_request_set *rqset)
1294 struct lov_request_set *lovset;
1295 struct lov_obd *lov;
1297 struct lov_request *req;
1302 ASSERT_LSM_MAGIC(oinfo->oi_md);
1304 if (!exp || !exp->exp_obd)
1307 lov = &exp->exp_obd->u.lov;
1309 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1313 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1314 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1315 oinfo->oi_md->lsm_stripe_size);
1317 cfs_list_for_each (pos, &lovset->set_list) {
1318 req = cfs_list_entry(pos, struct lov_request, rq_link);
1320 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1321 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1322 req->rq_oi.oi_oa->o_id, req->rq_idx);
1323 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1324 &req->rq_oi, rqset);
1326 CERROR("error: getattr objid "LPX64" subobj "
1327 LPX64" on OST idx %d: rc = %d\n",
1328 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1334 if (!cfs_list_empty(&rqset->set_requests)) {
1336 LASSERT (rqset->set_interpret == NULL);
1337 rqset->set_interpret = lov_getattr_interpret;
1338 rqset->set_arg = (void *)lovset;
1343 lovset->set_completes = 0;
1344 err = lov_fini_getattr_set(lovset);
1345 RETURN(rc ? rc : err);
1348 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1349 struct obd_trans_info *oti)
1351 struct lov_request_set *set;
1352 struct lov_obd *lov;
1354 struct lov_request *req;
1355 int err = 0, rc = 0;
1359 ASSERT_LSM_MAGIC(oinfo->oi_md);
1361 if (!exp || !exp->exp_obd)
1364 /* for now, we only expect the following updates here */
1365 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1366 OBD_MD_FLMODE | OBD_MD_FLATIME |
1367 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1368 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1369 OBD_MD_FLGROUP | OBD_MD_FLUID |
1370 OBD_MD_FLGID | OBD_MD_FLFID |
1372 lov = &exp->exp_obd->u.lov;
1373 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1377 cfs_list_for_each (pos, &set->set_list) {
1378 req = cfs_list_entry(pos, struct lov_request, rq_link);
1380 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1382 err = lov_update_setattr_set(set, req, rc);
1384 CERROR("error: setattr objid "LPX64" subobj "
1385 LPX64" on OST idx %d: rc = %d\n",
1386 set->set_oi->oi_oa->o_id,
1387 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1392 err = lov_fini_setattr_set(set);
1398 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1401 struct lov_request_set *lovset = (struct lov_request_set *)data;
1406 lovset->set_completes = 0;
1407 err = lov_fini_setattr_set(lovset);
1408 RETURN(rc ? rc : err);
1411 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1412 needed. Otherwise, a client is waiting for responses. */
1413 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1414 struct obd_trans_info *oti,
1415 struct ptlrpc_request_set *rqset)
1417 struct lov_request_set *set;
1418 struct lov_request *req;
1420 struct lov_obd *lov;
1425 ASSERT_LSM_MAGIC(oinfo->oi_md);
1426 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1428 LASSERT(oti->oti_logcookies);
1431 if (!exp || !exp->exp_obd)
1434 lov = &exp->exp_obd->u.lov;
1435 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1439 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1440 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1441 oinfo->oi_md->lsm_stripe_size);
1443 cfs_list_for_each (pos, &set->set_list) {
1444 req = cfs_list_entry(pos, struct lov_request, rq_link);
1446 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1447 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1449 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1450 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1451 req->rq_oi.oi_oa->o_id, req->rq_idx);
1453 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1454 &req->rq_oi, oti, rqset);
1456 CERROR("error: setattr objid "LPX64" subobj "
1457 LPX64" on OST idx %d: rc = %d\n",
1458 set->set_oi->oi_oa->o_id,
1459 req->rq_oi.oi_oa->o_id,
1465 /* If we are not waiting for responses on async requests, return. */
1466 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1469 set->set_completes = 0;
1470 err = lov_fini_setattr_set(set);
1471 RETURN(rc ? rc : err);
1474 LASSERT(rqset->set_interpret == NULL);
1475 rqset->set_interpret = lov_setattr_interpret;
1476 rqset->set_arg = (void *)set;
1481 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1484 struct lov_request_set *lovset = (struct lov_request_set *)data;
1489 lovset->set_completes = 0;
1490 err = lov_fini_punch_set(lovset);
1491 RETURN(rc ? rc : err);
1494 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1495 * we can send this 'punch' to just the authoritative node and the nodes
1496 * that the punch will affect. */
1497 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1498 struct obd_trans_info *oti,
1499 struct ptlrpc_request_set *rqset)
1501 struct lov_request_set *set;
1502 struct lov_obd *lov;
1504 struct lov_request *req;
1509 ASSERT_LSM_MAGIC(oinfo->oi_md);
1511 if (!exp || !exp->exp_obd)
1514 lov = &exp->exp_obd->u.lov;
1515 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1519 cfs_list_for_each (pos, &set->set_list) {
1520 req = cfs_list_entry(pos, struct lov_request, rq_link);
1522 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1523 &req->rq_oi, NULL, rqset);
1525 CERROR("error: punch objid "LPX64" subobj "LPX64
1526 " on OST idx %d: rc = %d\n",
1527 set->set_oi->oi_oa->o_id,
1528 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1533 if (rc || cfs_list_empty(&rqset->set_requests)) {
1535 err = lov_fini_punch_set(set);
1536 RETURN(rc ? rc : err);
1539 LASSERT(rqset->set_interpret == NULL);
1540 rqset->set_interpret = lov_punch_interpret;
1541 rqset->set_arg = (void *)set;
1546 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1547 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1550 struct lov_request_set *set;
1551 struct obd_info oinfo;
1552 struct lov_obd *lov;
1554 struct lov_request *req;
1555 int err = 0, rc = 0;
1558 ASSERT_LSM_MAGIC(lsm);
1563 lov = &exp->exp_obd->u.lov;
1564 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1568 cfs_list_for_each (pos, &set->set_list) {
1569 req = cfs_list_entry(pos, struct lov_request, rq_link);
1571 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1572 req->rq_oi.oi_oa, NULL,
1573 req->rq_oi.oi_policy.l_extent.start,
1574 req->rq_oi.oi_policy.l_extent.end, capa);
1575 err = lov_update_common_set(set, req, rc);
1577 CERROR("error: fsync objid "LPX64" subobj "LPX64
1578 " on OST idx %d: rc = %d\n",
1579 set->set_oi->oi_oa->o_id,
1580 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1585 err = lov_fini_sync_set(set);
1591 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1592 obd_count oa_bufs, struct brw_page *pga)
1594 struct obd_info oinfo = { { { 0 } } };
1597 oinfo.oi_oa = lov_oinfo->oi_oa;
1599 /* The caller just wants to know if there's a chance that this
1600 * I/O can succeed */
1601 for (i = 0; i < oa_bufs; i++) {
1602 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1603 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1606 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1607 pga[i].off + pga[i].count - 1,
1611 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1612 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1616 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1624 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1625 obd_count oa_bufs, struct brw_page *pga,
1626 struct obd_trans_info *oti)
1628 struct lov_request_set *set;
1629 struct lov_request *req;
1631 struct lov_obd *lov = &exp->exp_obd->u.lov;
1635 ASSERT_LSM_MAGIC(oinfo->oi_md);
1637 if (cmd == OBD_BRW_CHECK) {
1638 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1642 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1646 cfs_list_for_each (pos, &set->set_list) {
1647 struct obd_export *sub_exp;
1648 struct brw_page *sub_pga;
1649 req = cfs_list_entry(pos, struct lov_request, rq_link);
1651 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1652 sub_pga = set->set_pga + req->rq_pgaidx;
1653 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1657 lov_update_common_set(set, req, rc);
1660 err = lov_fini_brw_set(set);
1666 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1669 struct lov_request_set *lovset = (struct lov_request_set *)data;
1671 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1675 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1676 struct ldlm_enqueue_info *einfo,
1677 struct ptlrpc_request_set *rqset)
1679 ldlm_mode_t mode = einfo->ei_mode;
1680 struct lov_request_set *set;
1681 struct lov_request *req;
1683 struct lov_obd *lov;
1688 ASSERT_LSM_MAGIC(oinfo->oi_md);
1689 LASSERT(mode == (mode & -mode));
1691 /* we should never be asked to replay a lock this way. */
1692 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1694 if (!exp || !exp->exp_obd)
1697 lov = &exp->exp_obd->u.lov;
1698 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1702 cfs_list_for_each (pos, &set->set_list) {
1703 req = cfs_list_entry(pos, struct lov_request, rq_link);
1705 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1706 &req->rq_oi, einfo, rqset);
1711 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1713 LASSERT(rqset->set_interpret == NULL);
1714 rqset->set_interpret = lov_enqueue_interpret;
1715 rqset->set_arg = (void *)set;
1719 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1723 static int lov_change_cbdata(struct obd_export *exp,
1724 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1727 struct lov_obd *lov;
1731 ASSERT_LSM_MAGIC(lsm);
1733 if (!exp || !exp->exp_obd)
1736 lov = &exp->exp_obd->u.lov;
1737 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1738 struct lov_stripe_md submd;
1739 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1741 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1742 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1746 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1747 submd.lsm_object_id = loi->loi_id;
1748 submd.lsm_object_seq = loi->loi_seq;
1749 submd.lsm_stripe_count = 0;
1750 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1756 /* find any ldlm lock of the inode in lov
1760 static int lov_find_cbdata(struct obd_export *exp,
1761 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1764 struct lov_obd *lov;
1768 ASSERT_LSM_MAGIC(lsm);
1770 if (!exp || !exp->exp_obd)
1773 lov = &exp->exp_obd->u.lov;
1774 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1775 struct lov_stripe_md submd;
1776 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1778 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1779 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1783 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1784 submd.lsm_object_id = loi->loi_id;
1785 submd.lsm_object_seq = loi->loi_seq;
1786 submd.lsm_stripe_count = 0;
1787 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1795 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1796 __u32 mode, struct lustre_handle *lockh)
1798 struct lov_request_set *set;
1799 struct obd_info oinfo;
1800 struct lov_request *req;
1802 struct lov_obd *lov;
1803 struct lustre_handle *lov_lockhp;
1804 int err = 0, rc = 0;
1807 ASSERT_LSM_MAGIC(lsm);
1809 if (!exp || !exp->exp_obd)
1812 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1814 lov = &exp->exp_obd->u.lov;
1815 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1819 cfs_list_for_each (pos, &set->set_list) {
1820 req = cfs_list_entry(pos, struct lov_request, rq_link);
1821 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1823 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1824 req->rq_oi.oi_md, mode, lov_lockhp);
1825 rc = lov_update_common_set(set, req, rc);
1827 CERROR("error: cancel objid "LPX64" subobj "
1828 LPX64" on OST idx %d: rc = %d\n",
1830 req->rq_oi.oi_md->lsm_object_id,
1836 lov_fini_cancel_set(set);
1840 static int lov_cancel_unused(struct obd_export *exp,
1841 struct lov_stripe_md *lsm,
1842 ldlm_cancel_flags_t flags, void *opaque)
1844 struct lov_obd *lov;
1848 if (!exp || !exp->exp_obd)
1851 lov = &exp->exp_obd->u.lov;
1853 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1855 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1858 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1866 ASSERT_LSM_MAGIC(lsm);
1868 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1869 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1870 struct lov_stripe_md submd;
1871 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1874 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1875 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1879 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1880 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1882 submd.lsm_object_id = loi->loi_id;
1883 submd.lsm_object_seq = loi->loi_seq;
1884 submd.lsm_stripe_count = 0;
1885 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1886 &submd, flags, opaque);
1887 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1888 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1889 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1890 loi->loi_id, loi->loi_ost_idx, err);
1898 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1900 struct lov_request_set *lovset = (struct lov_request_set *)data;
1905 lovset->set_completes = 0;
1907 err = lov_fini_statfs_set(lovset);
1908 RETURN(rc ? rc : err);
1911 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1912 __u64 max_age, struct ptlrpc_request_set *rqset)
1914 struct lov_request_set *set;
1915 struct lov_request *req;
1917 struct lov_obd *lov;
1921 LASSERT(oinfo != NULL);
1922 LASSERT(oinfo->oi_osfs != NULL);
1925 rc = lov_prep_statfs_set(obd, oinfo, &set);
1929 cfs_list_for_each (pos, &set->set_list) {
1930 struct obd_device *osc_obd;
1932 req = cfs_list_entry(pos, struct lov_request, rq_link);
1934 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1935 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1940 if (rc || cfs_list_empty(&rqset->set_requests)) {
1943 set->set_completes = 0;
1944 err = lov_fini_statfs_set(set);
1945 RETURN(rc ? rc : err);
1948 LASSERT(rqset->set_interpret == NULL);
1949 rqset->set_interpret = lov_statfs_interpret;
1950 rqset->set_arg = (void *)set;
1954 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1955 __u64 max_age, __u32 flags)
1957 struct ptlrpc_request_set *set = NULL;
1958 struct obd_info oinfo = { { { 0 } } };
1963 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1964 * statfs requests */
1965 set = ptlrpc_prep_set();
1969 oinfo.oi_osfs = osfs;
1970 oinfo.oi_flags = flags;
1971 rc = lov_statfs_async(obd, &oinfo, max_age, set);
1973 rc = ptlrpc_set_wait(set);
1974 ptlrpc_set_destroy(set);
1979 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1980 void *karg, void *uarg)
1982 struct obd_device *obddev = class_exp2obd(exp);
1983 struct lov_obd *lov = &obddev->u.lov;
1984 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1985 struct obd_uuid *uuidp;
1989 case IOC_OBD_STATFS: {
1990 struct obd_ioctl_data *data = karg;
1991 struct obd_device *osc_obd;
1992 struct obd_statfs stat_buf = {0};
1995 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1996 if ((index >= count))
1999 if (!lov->lov_tgts[index])
2000 /* Try again with the next index */
2002 if (!lov->lov_tgts[index]->ltd_active)
2005 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2010 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2011 min((int) data->ioc_plen2,
2012 (int) sizeof(struct obd_uuid))))
2015 /* got statfs data */
2016 rc = obd_statfs(osc_obd, &stat_buf,
2017 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2021 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2022 min((int) data->ioc_plen1,
2023 (int) sizeof(stat_buf))))
2027 case OBD_IOC_LOV_GET_CONFIG: {
2028 struct obd_ioctl_data *data;
2029 struct lov_desc *desc;
2034 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2037 data = (struct obd_ioctl_data *)buf;
2039 if (sizeof(*desc) > data->ioc_inllen1) {
2040 obd_ioctl_freedata(buf, len);
2044 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2045 obd_ioctl_freedata(buf, len);
2049 if (sizeof(__u32) * count > data->ioc_inllen3) {
2050 obd_ioctl_freedata(buf, len);
2054 desc = (struct lov_desc *)data->ioc_inlbuf1;
2055 memcpy(desc, &(lov->desc), sizeof(*desc));
2057 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2058 genp = (__u32 *)data->ioc_inlbuf3;
2059 /* the uuid will be empty for deleted OSTs */
2060 for (i = 0; i < count; i++, uuidp++, genp++) {
2061 if (!lov->lov_tgts[i])
2063 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2064 *genp = lov->lov_tgts[i]->ltd_gen;
2067 if (cfs_copy_to_user((void *)uarg, buf, len))
2069 obd_ioctl_freedata(buf, len);
2072 case LL_IOC_LOV_SETSTRIPE:
2073 rc = lov_setstripe(exp, len, karg, uarg);
2075 case LL_IOC_LOV_GETSTRIPE:
2076 rc = lov_getstripe(exp, karg, uarg);
2078 case LL_IOC_LOV_SETEA:
2079 rc = lov_setea(exp, karg, uarg);
2081 case OBD_IOC_QUOTACTL: {
2082 struct if_quotactl *qctl = karg;
2083 struct lov_tgt_desc *tgt = NULL;
2084 struct obd_quotactl *oqctl;
2086 if (qctl->qc_valid == QC_OSTIDX) {
2087 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2090 tgt = lov->lov_tgts[qctl->qc_idx];
2091 if (!tgt || !tgt->ltd_exp)
2093 } else if (qctl->qc_valid == QC_UUID) {
2094 for (i = 0; i < count; i++) {
2095 tgt = lov->lov_tgts[i];
2097 !obd_uuid_equals(&tgt->ltd_uuid,
2101 if (tgt->ltd_exp == NULL)
2113 LASSERT(tgt && tgt->ltd_exp);
2114 OBD_ALLOC_PTR(oqctl);
2118 QCTL_COPY(oqctl, qctl);
2119 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2121 QCTL_COPY(qctl, oqctl);
2122 qctl->qc_valid = QC_OSTIDX;
2123 qctl->obd_uuid = tgt->ltd_uuid;
2125 OBD_FREE_PTR(oqctl);
2134 for (i = 0; i < count; i++) {
2137 /* OST was disconnected */
2138 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2141 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2143 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2146 if (lov->lov_tgts[i]->ltd_active) {
2147 CDEBUG(err == -ENOTTY ?
2148 D_IOCTL : D_WARNING,
2149 "iocontrol OSC %s on OST "
2150 "idx %d cmd %x: err = %d\n",
2151 lov_uuid2str(lov, i),
2168 #define FIEMAP_BUFFER_SIZE 4096
2171 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2172 * call. The local end offset and the device are sent in the first
2173 * fm_extent. This function calculates the stripe number from the index.
2174 * This function returns a stripe_no on which mapping is to be restarted.
2176 * This function returns fm_end_offset which is the in-OST offset at which
2177 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2178 * will re-calculate proper offset in next stripe.
2179 * Note that the first extent is passed to lov_get_info via the value field.
2181 * \param fiemap fiemap request header
2182 * \param lsm striping information for the file
2183 * \param fm_start logical start of mapping
2184 * \param fm_end logical end of mapping
2185 * \param start_stripe starting stripe will be returned in this
2187 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2188 struct lov_stripe_md *lsm, obd_size fm_start,
2189 obd_size fm_end, int *start_stripe)
2191 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2192 obd_off lun_start, lun_end;
2193 obd_size fm_end_offset;
2194 int stripe_no = -1, i;
2196 if (fiemap->fm_extent_count == 0 ||
2197 fiemap->fm_extents[0].fe_logical == 0)
2200 /* Find out stripe_no from ost_index saved in the fe_device */
2201 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2202 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2203 fiemap->fm_extents[0].fe_device) {
2209 /* If we have finished mapping on previous device, shift logical
2210 * offset to start of next device */
2211 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2212 &lun_start, &lun_end)) != 0 &&
2213 local_end < lun_end) {
2214 fm_end_offset = local_end;
2215 *start_stripe = stripe_no;
2217 /* This is a special value to indicate that caller should
2218 * calculate offset in next stripe. */
2220 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2223 return fm_end_offset;
2227 * We calculate on which OST the mapping will end. If the length of mapping
2228 * is greater than (stripe_size * stripe_count) then the last_stripe will
2229 * will be one just before start_stripe. Else we check if the mapping
2230 * intersects each OST and find last_stripe.
2231 * This function returns the last_stripe and also sets the stripe_count
2232 * over which the mapping is spread
2234 * \param lsm striping information for the file
2235 * \param fm_start logical start of mapping
2236 * \param fm_end logical end of mapping
2237 * \param start_stripe starting stripe of the mapping
2238 * \param stripe_count the number of stripes across which to map is returned
2240 * \retval last_stripe return the last stripe of the mapping
2242 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2243 obd_size fm_end, int start_stripe,
2247 obd_off obd_start, obd_end;
2250 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2251 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2253 *stripe_count = lsm->lsm_stripe_count;
2255 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2256 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2257 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2258 &obd_start, &obd_end)) == 0)
2262 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2269 * Set fe_device and copy extents from local buffer into main return buffer.
2271 * \param fiemap fiemap request header
2272 * \param lcl_fm_ext array of local fiemap extents to be copied
2273 * \param ost_index OST index to be written into the fm_device field for each
2275 * \param ext_count number of extents to be copied
2276 * \param current_extent where to start copying in main extent array
2278 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2279 struct ll_fiemap_extent *lcl_fm_ext,
2280 int ost_index, unsigned int ext_count,
2286 for (ext = 0; ext < ext_count; ext++) {
2287 lcl_fm_ext[ext].fe_device = ost_index;
2288 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2291 /* Copy fm_extent's from fm_local to return buffer */
2292 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2293 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2297 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2298 * This also handles the restarting of FIEMAP calls in case mapping overflows
2299 * the available number of extents in single call.
2301 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2302 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2304 struct ll_fiemap_info_key *fm_key = key;
2305 struct ll_user_fiemap *fiemap = val;
2306 struct ll_user_fiemap *fm_local = NULL;
2307 struct ll_fiemap_extent *lcl_fm_ext;
2309 unsigned int get_num_extents = 0;
2310 int ost_index = 0, actual_start_stripe, start_stripe;
2311 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2313 int current_extent = 0, rc = 0, i;
2314 int ost_eof = 0; /* EOF for object */
2315 int ost_done = 0; /* done with required mapping for this OST? */
2317 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2318 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2323 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2324 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2326 OBD_ALLOC(fm_local, buffer_size);
2327 if (fm_local == NULL)
2328 GOTO(out, rc = -ENOMEM);
2329 lcl_fm_ext = &fm_local->fm_extents[0];
2331 count_local = fiemap_size_to_count(buffer_size);
2333 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2334 fm_start = fiemap->fm_start;
2335 fm_length = fiemap->fm_length;
2336 /* Calculate start stripe, last stripe and length of mapping */
2337 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2338 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2339 fm_start + fm_length - 1);
2340 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2341 if (fm_end > fm_key->oa.o_size)
2342 fm_end = fm_key->oa.o_size;
2344 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2345 actual_start_stripe, &stripe_count);
2347 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2350 if (fiemap->fm_extent_count == 0) {
2351 get_num_extents = 1;
2355 /* Check each stripe */
2356 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2357 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2358 obd_size req_fm_len; /* Stores length of required mapping */
2359 obd_size len_mapped_single_call;
2360 obd_off lun_start, lun_end, obd_object_end;
2361 unsigned int ext_count;
2363 cur_stripe_wrap = cur_stripe;
2365 /* Find out range of mapping on this stripe */
2366 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2367 &lun_start, &obd_object_end)) == 0)
2370 /* If this is a continuation FIEMAP call and we are on
2371 * starting stripe then lun_start needs to be set to
2373 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2374 lun_start = fm_end_offset;
2376 if (fm_length != ~0ULL) {
2377 /* Handle fm_start + fm_length overflow */
2378 if (fm_start + fm_length < fm_start)
2379 fm_length = ~0ULL - fm_start;
2380 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2386 if (lun_start == lun_end)
2389 req_fm_len = obd_object_end - lun_start;
2390 fm_local->fm_length = 0;
2391 len_mapped_single_call = 0;
2393 /* If the output buffer is very large and the objects have many
2394 * extents we may need to loop on a single OST repeatedly */
2398 if (get_num_extents == 0) {
2399 /* Don't get too many extents. */
2400 if (current_extent + count_local >
2401 fiemap->fm_extent_count)
2402 count_local = fiemap->fm_extent_count -
2406 lun_start += len_mapped_single_call;
2407 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2408 req_fm_len = fm_local->fm_length;
2409 fm_local->fm_extent_count = count_local;
2410 fm_local->fm_mapped_extents = 0;
2411 fm_local->fm_flags = fiemap->fm_flags;
2413 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2414 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2415 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2417 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2418 GOTO(out, rc = -EINVAL);
2420 /* If OST is inactive, return extent with UNKNOWN flag */
2421 if (!lov->lov_tgts[ost_index]->ltd_active) {
2422 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2423 fm_local->fm_mapped_extents = 1;
2425 lcl_fm_ext[0].fe_logical = lun_start;
2426 lcl_fm_ext[0].fe_length = obd_object_end -
2428 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2433 fm_local->fm_start = lun_start;
2434 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2435 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2436 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2437 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2438 keylen, key, vallen, fm_local, lsm);
2443 ext_count = fm_local->fm_mapped_extents;
2444 if (ext_count == 0) {
2446 /* If last stripe has hole at the end,
2447 * then we need to return */
2448 if (cur_stripe_wrap == last_stripe) {
2449 fiemap->fm_mapped_extents = 0;
2455 /* If we just need num of extents then go to next device */
2456 if (get_num_extents) {
2457 current_extent += ext_count;
2461 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2462 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2464 /* Have we finished mapping on this device? */
2465 if (req_fm_len <= len_mapped_single_call)
2468 /* Clear the EXTENT_LAST flag which can be present on
2470 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2471 lcl_fm_ext[ext_count - 1].fe_flags &=
2472 ~FIEMAP_EXTENT_LAST;
2474 curr_loc = lov_stripe_size(lsm,
2475 lcl_fm_ext[ext_count - 1].fe_logical+
2476 lcl_fm_ext[ext_count - 1].fe_length,
2478 if (curr_loc >= fm_key->oa.o_size)
2481 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2482 ost_index, ext_count,
2485 current_extent += ext_count;
2487 /* Ran out of available extents? */
2488 if (current_extent >= fiemap->fm_extent_count)
2490 } while (ost_done == 0 && ost_eof == 0);
2492 if (cur_stripe_wrap == last_stripe)
2497 /* Indicate that we are returning device offsets unless file just has
2499 if (lsm->lsm_stripe_count > 1)
2500 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2502 if (get_num_extents)
2503 goto skip_last_device_calc;
2505 /* Check if we have reached the last stripe and whether mapping for that
2506 * stripe is done. */
2507 if (cur_stripe_wrap == last_stripe) {
2508 if (ost_done || ost_eof)
2509 fiemap->fm_extents[current_extent - 1].fe_flags |=
2513 skip_last_device_calc:
2514 fiemap->fm_mapped_extents = current_extent;
2517 OBD_FREE(fm_local, buffer_size);
2521 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2522 void *key, __u32 *vallen, void *val,
2523 struct lov_stripe_md *lsm)
2525 struct obd_device *obddev = class_exp2obd(exp);
2526 struct lov_obd *lov = &obddev->u.lov;
2530 if (!vallen || !val)
2535 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2538 struct ldlm_lock *lock;
2540 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2541 struct lov_oinfo *loi;
2542 __u32 *stripe = val;
2544 if (*vallen < sizeof(*stripe))
2545 GOTO(out, rc = -EFAULT);
2546 *vallen = sizeof(*stripe);
2548 /* XXX This is another one of those bits that will need to
2549 * change if we ever actually support nested LOVs. It uses
2550 * the lock's export to find out which stripe it is. */
2551 /* XXX - it's assumed all the locks for deleted OSTs have
2552 * been cancelled. Also, the export for deleted OSTs will
2553 * be NULL and won't match the lock's export. */
2554 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2555 loi = lsm->lsm_oinfo[i];
2556 if (!lov->lov_tgts[loi->loi_ost_idx])
2558 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2559 data->lock->l_conn_export &&
2560 osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2565 LDLM_ERROR(data->lock, "lock on inode without such object");
2566 dump_lsm(D_ERROR, lsm);
2567 GOTO(out, rc = -ENXIO);
2568 } else if (KEY_IS(KEY_LAST_ID)) {
2569 struct obd_id_info *info = val;
2570 __u32 size = sizeof(obd_id);
2571 struct lov_tgt_desc *tgt;
2573 LASSERT(*vallen == sizeof(struct obd_id_info));
2574 tgt = lov->lov_tgts[info->idx];
2576 if (!tgt || !tgt->ltd_active)
2577 GOTO(out, rc = -ESRCH);
2579 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2581 } else if (KEY_IS(KEY_LOVDESC)) {
2582 struct lov_desc *desc_ret = val;
2583 *desc_ret = lov->desc;
2586 } else if (KEY_IS(KEY_FIEMAP)) {
2587 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2589 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2590 struct lov_tgt_desc *tgt;
2591 __u64 ost_idx = *((__u64*)val);
2593 LASSERT(*vallen == sizeof(__u64));
2594 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2595 tgt = lov->lov_tgts[ost_idx];
2597 if (!tgt || !tgt->ltd_exp)
2598 GOTO(out, rc = -ESRCH);
2600 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2611 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2612 void *key, obd_count vallen, void *val,
2613 struct ptlrpc_request_set *set)
2615 struct obd_device *obddev = class_exp2obd(exp);
2616 struct lov_obd *lov = &obddev->u.lov;
2619 struct lov_tgt_desc *tgt;
2620 unsigned incr, check_uuid,
2621 do_inactive, no_set;
2622 unsigned next_id = 0, mds_con = 0, capa = 0;
2625 incr = check_uuid = do_inactive = no_set = 0;
2628 set = ptlrpc_prep_set();
2634 count = lov->desc.ld_tgt_count;
2636 if (KEY_IS(KEY_NEXT_ID)) {
2637 count = vallen / sizeof(struct obd_id_info);
2638 vallen = sizeof(obd_id);
2639 incr = sizeof(struct obd_id_info);
2642 } else if (KEY_IS(KEY_CHECKSUM)) {
2644 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2645 /* use defaults: do_inactive = incr = 0; */
2646 } else if (KEY_IS(KEY_MDS_CONN)) {
2648 } else if (KEY_IS(KEY_CAPA_KEY)) {
2652 for (i = 0; i < count; i++, val = (char *)val + incr) {
2654 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2656 tgt = lov->lov_tgts[i];
2658 /* OST was disconnected */
2659 if (!tgt || !tgt->ltd_exp)
2662 /* OST is inactive and we don't want inactive OSCs */
2663 if (!tgt->ltd_active && !do_inactive)
2667 struct mds_group_info *mgi;
2669 LASSERT(vallen == sizeof(*mgi));
2670 mgi = (struct mds_group_info *)val;
2672 /* Only want a specific OSC */
2673 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2677 err = obd_set_info_async(tgt->ltd_exp,
2678 keylen, key, sizeof(int),
2680 } else if (next_id) {
2681 err = obd_set_info_async(tgt->ltd_exp,
2682 keylen, key, vallen,
2683 ((struct obd_id_info*)val)->data, set);
2685 struct mds_capa_info *info = (struct mds_capa_info*)val;
2687 LASSERT(vallen == sizeof(*info));
2689 /* Only want a specific OSC */
2691 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2694 err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2695 sizeof(*info->capa),
2699 /* Only want a specific OSC */
2701 !obd_uuid_equals(val, &tgt->ltd_uuid))
2704 err = obd_set_info_async(tgt->ltd_exp,
2705 keylen, key, vallen, val, set);
2714 err = ptlrpc_set_wait(set);
2717 ptlrpc_set_destroy(set);
2722 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2727 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2728 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2729 if (loi->loi_ar.ar_rc && !rc)
2730 rc = loi->loi_ar.ar_rc;
2731 loi->loi_ar.ar_rc = 0;
2735 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2738 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2739 int cmd, __u64 *offset)
2741 __u32 ssize = lsm->lsm_stripe_size;
2745 do_div(start, ssize);
2746 start = start * ssize;
2748 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2749 ", end "LPU64"\n", *offset, ssize, start,
2751 if (cmd == OBD_CALC_STRIPE_END) {
2752 *offset = start + ssize - 1;
2753 } else if (cmd == OBD_CALC_STRIPE_START) {
2762 void lov_stripe_lock(struct lov_stripe_md *md)
2764 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2765 cfs_spin_lock(&md->lsm_lock);
2766 LASSERT(md->lsm_lock_owner == 0);
2767 md->lsm_lock_owner = cfs_curproc_pid();
2769 EXPORT_SYMBOL(lov_stripe_lock);
2771 void lov_stripe_unlock(struct lov_stripe_md *md)
2773 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2774 md->lsm_lock_owner = 0;
2775 cfs_spin_unlock(&md->lsm_lock);
2777 EXPORT_SYMBOL(lov_stripe_unlock);
2780 struct obd_ops lov_obd_ops = {
2781 .o_owner = THIS_MODULE,
2782 .o_setup = lov_setup,
2783 .o_precleanup = lov_precleanup,
2784 .o_cleanup = lov_cleanup,
2785 //.o_process_config = lov_process_config,
2786 .o_connect = lov_connect,
2787 .o_disconnect = lov_disconnect,
2788 .o_statfs = lov_statfs,
2789 .o_statfs_async = lov_statfs_async,
2790 .o_packmd = lov_packmd,
2791 .o_unpackmd = lov_unpackmd,
2792 .o_create = lov_create,
2793 .o_destroy = lov_destroy,
2794 .o_getattr = lov_getattr,
2795 .o_getattr_async = lov_getattr_async,
2796 .o_setattr = lov_setattr,
2797 .o_setattr_async = lov_setattr_async,
2799 .o_merge_lvb = lov_merge_lvb,
2800 .o_adjust_kms = lov_adjust_kms,
2801 .o_punch = lov_punch,
2803 .o_enqueue = lov_enqueue,
2804 .o_change_cbdata = lov_change_cbdata,
2805 .o_find_cbdata = lov_find_cbdata,
2806 .o_cancel = lov_cancel,
2807 .o_cancel_unused = lov_cancel_unused,
2808 .o_iocontrol = lov_iocontrol,
2809 .o_get_info = lov_get_info,
2810 .o_set_info_async = lov_set_info_async,
2811 .o_extent_calc = lov_extent_calc,
2812 .o_llog_init = lov_llog_init,
2813 .o_llog_finish = lov_llog_finish,
2814 .o_notify = lov_notify,
2815 .o_pool_new = lov_pool_new,
2816 .o_pool_rem = lov_pool_remove,
2817 .o_pool_add = lov_pool_add,
2818 .o_pool_del = lov_pool_del,
2819 .o_getref = lov_getref,
2820 .o_putref = lov_putref,
2823 static quota_interface_t *quota_interface;
2824 extern quota_interface_t lov_quota_interface;
2826 cfs_mem_cache_t *lov_oinfo_slab;
2828 extern struct lu_kmem_descr lov_caches[];
2830 int __init lov_init(void)
2832 struct lprocfs_static_vars lvars = { 0 };
2836 /* print an address of _any_ initialized kernel symbol from this
2837 * module, to allow debugging with gdb that doesn't support data
2838 * symbols from modules.*/
2839 CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2841 rc = lu_kmem_init(lov_caches);
2845 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2846 sizeof(struct lov_oinfo),
2847 0, CFS_SLAB_HWCACHE_ALIGN);
2848 if (lov_oinfo_slab == NULL) {
2849 lu_kmem_fini(lov_caches);
2852 lprocfs_lov_init_vars(&lvars);
2854 cfs_request_module("lquota");
2855 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2856 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2858 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2859 LUSTRE_LOV_NAME, &lov_device_type);
2862 if (quota_interface)
2863 PORTAL_SYMBOL_PUT(lov_quota_interface);
2864 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2866 lu_kmem_fini(lov_caches);
2873 static void /*__exit*/ lov_exit(void)
2877 lu_device_type_fini(&lov_device_type);
2878 lu_kmem_fini(lov_caches);
2880 if (quota_interface)
2881 PORTAL_SYMBOL_PUT(lov_quota_interface);
2883 class_unregister_type(LUSTRE_LOV_NAME);
2884 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2888 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2889 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2890 MODULE_LICENSE("GPL");
2892 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);