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;
479 struct obd_device *tgt_obd;
482 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
483 /* don't send sync event if target not
484 * connected/activated */
485 if (!lov->lov_tgts[i] ||
486 !lov->lov_tgts[i]->ltd_active)
489 if ((ev == OBD_NOTIFY_SYNC) ||
490 (ev == OBD_NOTIFY_SYNC_NONBLOCK))
493 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
495 rc = obd_notify_observer(obd, tgt_obd, ev, data);
497 CERROR("%s: notify %s of %s failed %d\n",
499 obd->obd_observer->obd_name,
500 tgt_obd->obd_name, rc);
510 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
511 __u32 index, int gen, int active)
513 struct lov_obd *lov = &obd->u.lov;
514 struct lov_tgt_desc *tgt;
515 struct obd_device *tgt_obd;
519 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
520 uuidp->uuid, index, gen, active);
523 CERROR("request to add OBD %s with invalid generation: %d\n",
528 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
533 cfs_mutex_down(&lov->lov_lock);
535 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
536 tgt = lov->lov_tgts[index];
537 CERROR("UUID %s already assigned at LOV target index %d\n",
538 obd_uuid2str(&tgt->ltd_uuid), index);
539 cfs_mutex_up(&lov->lov_lock);
543 if (index >= lov->lov_tgt_size) {
544 /* We need to reallocate the lov target array. */
545 struct lov_tgt_desc **newtgts, **old = NULL;
546 __u32 newsize, oldsize = 0;
548 newsize = max(lov->lov_tgt_size, (__u32)2);
549 while (newsize < index + 1)
550 newsize = newsize << 1;
551 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
552 if (newtgts == NULL) {
553 cfs_mutex_up(&lov->lov_lock);
557 if (lov->lov_tgt_size) {
558 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
561 oldsize = lov->lov_tgt_size;
564 lov->lov_tgts = newtgts;
565 lov->lov_tgt_size = newsize;
570 OBD_FREE(old, sizeof(*old) * oldsize);
572 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
573 lov->lov_tgts, lov->lov_tgt_size);
578 cfs_mutex_up(&lov->lov_lock);
582 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
584 cfs_mutex_up(&lov->lov_lock);
589 memset(tgt, 0, sizeof(*tgt));
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 if (lov->lov_connects == 0) {
606 /* lov_connect hasn't been called yet. We'll do the
607 lov_connect_obd on this target when that fn first runs,
608 because we don't know the connect flags yet. */
614 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
618 /* connect to administrative disabled ost */
622 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
623 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
628 CERROR("add failed (%d), deleting %s\n", rc,
629 obd_uuid2str(&tgt->ltd_uuid));
630 lov_del_target(obd, index, 0, 0);
636 /* Schedule a target for deletion */
637 int lov_del_target(struct obd_device *obd, __u32 index,
638 struct obd_uuid *uuidp, int gen)
640 struct lov_obd *lov = &obd->u.lov;
641 int count = lov->desc.ld_tgt_count;
645 if (index >= count) {
646 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
653 if (!lov->lov_tgts[index]) {
654 CERROR("LOV target at index %d is not setup.\n", index);
655 GOTO(out, rc = -EINVAL);
658 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
659 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
660 lov_uuid2str(lov, index), index,
661 obd_uuid2str(uuidp));
662 GOTO(out, rc = -EINVAL);
665 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
666 lov_uuid2str(lov, index), index,
667 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
668 lov->lov_tgts[index]->ltd_active);
670 lov->lov_tgts[index]->ltd_reap = 1;
671 lov->lov_death_row++;
672 /* we really delete it from obd_putref */
679 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
681 struct obd_device *osc_obd;
684 LASSERT(tgt->ltd_reap);
686 osc_obd = class_exp2obd(tgt->ltd_exp);
688 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
690 osc_obd ? osc_obd->obd_name : "<no obd>");
693 lov_disconnect_obd(obd, tgt);
697 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
698 do it ourselves. And we can't do it from lov_cleanup,
699 because we just lost our only reference to it. */
701 class_manual_cleanup(osc_obd);
704 void lov_fix_desc_stripe_size(__u64 *val)
706 if (*val < PTLRPC_MAX_BRW_SIZE) {
707 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
708 PTLRPC_MAX_BRW_SIZE);
709 *val = PTLRPC_MAX_BRW_SIZE;
710 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
711 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
712 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
714 *val, LOV_MIN_STRIPE_SIZE);
718 void lov_fix_desc_stripe_count(__u32 *val)
724 void lov_fix_desc_pattern(__u32 *val)
726 /* from lov_setstripe */
727 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
728 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
733 void lov_fix_desc_qos_maxage(__u32 *val)
737 *val = QOS_DEFAULT_MAXAGE;
740 void lov_fix_desc(struct lov_desc *desc)
742 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
743 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
744 lov_fix_desc_pattern(&desc->ld_pattern);
745 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
748 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
750 struct lprocfs_static_vars lvars = { 0 };
751 struct lov_desc *desc;
752 struct lov_obd *lov = &obd->u.lov;
756 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
757 CERROR("LOV setup requires a descriptor\n");
761 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
763 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
764 CERROR("descriptor size wrong: %d > %d\n",
765 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
769 if (desc->ld_magic != LOV_DESC_MAGIC) {
770 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
771 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
772 obd->obd_name, desc);
773 lustre_swab_lov_desc(desc);
775 CERROR("%s: Bad lov desc magic: %#x\n",
776 obd->obd_name, desc->ld_magic);
783 desc->ld_active_tgt_count = 0;
785 lov->lov_tgt_size = 0;
787 cfs_sema_init(&lov->lov_lock, 1);
788 cfs_atomic_set(&lov->lov_refcount, 0);
789 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
790 cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
791 lov->lov_sp_me = LUSTRE_SP_CLI;
792 lov->lov_qos.lq_dirty = 1;
793 lov->lov_qos.lq_rr.lqr_dirty = 1;
794 lov->lov_qos.lq_reset = 1;
795 /* Default priority is toward free space balance */
796 lov->lov_qos.lq_prio_free = 232;
797 /* Default threshold for rr (roughly 17%) */
798 lov->lov_qos.lq_threshold_rr = 43;
799 /* Init statfs fields */
800 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
801 if (NULL == lov->lov_qos.lq_statfs_data)
803 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
805 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
807 &pool_hash_operations, CFS_HASH_REHASH);
808 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
809 lov->lov_pool_count = 0;
810 rc = lov_ost_pool_init(&lov->lov_packed, 0);
813 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
815 lov_ost_pool_free(&lov->lov_packed);
819 lprocfs_lov_init_vars(&lvars);
820 lprocfs_obd_setup(obd, lvars.obd_vars);
825 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
826 0444, &lov_proc_target_fops, obd);
828 CWARN("Error adding the target_obd file\n");
831 lov->lov_pool_proc_entry = lprocfs_register("pools",
838 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
841 struct lov_obd *lov = &obd->u.lov;
846 case OBD_CLEANUP_EARLY: {
848 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
849 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
851 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
856 case OBD_CLEANUP_EXPORTS:
857 rc = obd_llog_finish(obd, 0);
859 CERROR("failed to cleanup llogging subsystems\n");
865 static int lov_cleanup(struct obd_device *obd)
867 struct lov_obd *lov = &obd->u.lov;
868 cfs_list_t *pos, *tmp;
869 struct pool_desc *pool;
871 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
872 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
873 /* free pool structs */
874 CDEBUG(D_INFO, "delete pool %p\n", pool);
875 lov_pool_del(obd, pool->pool_name);
877 cfs_hash_putref(lov->lov_pools_hash_body);
878 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
879 lov_ost_pool_free(&lov->lov_packed);
884 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
885 if (!lov->lov_tgts[i])
888 /* Inactive targets may never have connected */
889 if (lov->lov_tgts[i]->ltd_active ||
890 cfs_atomic_read(&lov->lov_refcount))
891 /* We should never get here - these
892 should have been removed in the
894 CERROR("lov tgt %d not cleaned!"
895 " deathrow=%d, lovrc=%d\n",
896 i, lov->lov_death_row,
897 cfs_atomic_read(&lov->lov_refcount));
898 lov_del_target(obd, i, 0, 0);
901 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
903 lov->lov_tgt_size = 0;
906 /* clear pools parent proc entry only after all pools is killed */
907 lprocfs_obd_cleanup(obd);
909 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
913 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
914 __u32 *indexp, int *genp)
916 struct obd_uuid obd_uuid;
921 switch(cmd = lcfg->lcfg_command) {
922 case LCFG_LOV_ADD_OBD:
923 case LCFG_LOV_ADD_INA:
924 case LCFG_LOV_DEL_OBD: {
927 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
928 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
929 GOTO(out, rc = -EINVAL);
931 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
933 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
934 GOTO(out, rc = -EINVAL);
935 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
936 GOTO(out, rc = -EINVAL);
939 if (cmd == LCFG_LOV_ADD_OBD)
940 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
941 else if (cmd == LCFG_LOV_ADD_INA)
942 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
944 rc = lov_del_target(obd, index, &obd_uuid, gen);
948 struct lprocfs_static_vars lvars = { 0 };
949 struct lov_desc *desc = &(obd->u.lov.desc);
952 GOTO(out, rc = -EINVAL);
954 lprocfs_lov_init_vars(&lvars);
956 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
969 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
970 GOTO(out, rc = -EINVAL);
979 #define log2(n) cfs_ffz(~(n))
982 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
983 struct lov_stripe_md **ea,
984 struct obd_trans_info *oti)
988 struct obd_uuid *ost_uuid = NULL;
992 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
993 src_oa->o_flags == OBD_FL_DELORPHAN);
995 lov = &export->exp_obd->u.lov;
1001 if (oti->oti_ost_uuid) {
1002 ost_uuid = oti->oti_ost_uuid;
1003 CDEBUG(D_HA, "clearing orphans only for %s\n",
1007 obd_getref(export->exp_obd);
1008 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1009 struct lov_stripe_md obj_md;
1010 struct lov_stripe_md *obj_mdp = &obj_md;
1011 struct lov_tgt_desc *tgt;
1014 tgt = lov->lov_tgts[i];
1018 /* if called for a specific target, we don't
1019 care if it is not active. */
1020 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1021 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1025 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1028 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1029 obd_uuid2str(ost_uuid));
1031 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1033 LASSERT(lov->lov_tgts[i]->ltd_exp);
1034 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1035 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1036 tmp_oa, &obj_mdp, oti);
1038 /* This export will be disabled until it is recovered,
1039 and then orphan recovery will be completed. */
1040 CERROR("error in orphan recovery on OST idx %d/%d: "
1041 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1048 obd_putref(export->exp_obd);
1054 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1055 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1057 struct lov_stripe_md *obj_mdp, *lsm;
1058 struct lov_obd *lov = &exp->exp_obd->u.lov;
1063 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1064 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1066 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1067 if (obj_mdp == NULL)
1070 ost_idx = src_oa->o_nlink;
1073 GOTO(out, rc = -EINVAL);
1074 if (ost_idx >= lov->desc.ld_tgt_count ||
1075 !lov->lov_tgts[ost_idx])
1076 GOTO(out, rc = -EINVAL);
1078 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1079 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1080 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1081 GOTO(out, rc = -EINVAL);
1085 if (i == lsm->lsm_stripe_count)
1086 GOTO(out, rc = -EINVAL);
1088 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1090 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1094 /* the LOV expects oa->o_id to be set to the LOV object id */
1095 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1096 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1098 struct lov_obd *lov;
1099 struct obd_info oinfo;
1100 struct lov_request_set *set = NULL;
1101 struct lov_request *req;
1102 struct l_wait_info lwi = { 0 };
1106 LASSERT(ea != NULL);
1110 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1111 src_oa->o_flags == OBD_FL_DELORPHAN) {
1112 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1116 lov = &exp->exp_obd->u.lov;
1117 if (!lov->desc.ld_active_tgt_count)
1120 obd_getref(exp->exp_obd);
1121 /* Recreate a specific object id at the given OST index */
1122 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1123 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1124 rc = lov_recreate(exp, src_oa, ea, oti);
1128 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1129 * later in alloc_qos(), we will wait for those rpcs to complete if
1130 * the osfs age is older than 2 * qos_maxage */
1131 qos_statfs_update(exp->exp_obd,
1132 cfs_time_shift_64(-lov->desc.ld_qos_maxage) + CFS_HZ,
1135 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1139 cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1140 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1141 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1142 &req->rq_oi, &req->rq_oi.oi_md, oti);
1145 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1146 * longer then this */
1147 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1149 /* we not have ptlrpc set for assign set->interpret and should
1150 * be call interpret function himself. calling from cb_create_update
1151 * not permited because lov_fini_create_set can sleep for long time,
1152 * but we must avoid sleeping in ptlrpcd interpret function. */
1153 rc = lov_fini_create_set(set, ea);
1155 obd_putref(exp->exp_obd);
1159 #define ASSERT_LSM_MAGIC(lsmp) \
1161 LASSERT((lsmp) != NULL); \
1162 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1163 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1164 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1167 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1168 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1169 struct obd_export *md_exp, void *capa)
1171 struct lov_request_set *set;
1172 struct obd_info oinfo;
1173 struct lov_request *req;
1175 struct lov_obd *lov;
1176 int rc = 0, err = 0;
1179 ASSERT_LSM_MAGIC(lsm);
1181 if (!exp || !exp->exp_obd)
1184 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1186 LASSERT(oti->oti_logcookies);
1189 lov = &exp->exp_obd->u.lov;
1190 obd_getref(exp->exp_obd);
1191 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1195 cfs_list_for_each (pos, &set->set_list) {
1196 req = cfs_list_entry(pos, struct lov_request, rq_link);
1198 if (oa->o_valid & OBD_MD_FLCOOKIE)
1199 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1201 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1202 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1203 err = lov_update_common_set(set, req, err);
1205 CERROR("error: destroying objid "LPX64" subobj "
1206 LPX64" on OST idx %d: rc = %d\n",
1207 oa->o_id, req->rq_oi.oi_oa->o_id,
1215 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1216 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1218 err = lov_fini_destroy_set(set);
1220 obd_putref(exp->exp_obd);
1221 RETURN(rc ? rc : err);
1224 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1226 struct lov_request_set *set;
1227 struct lov_request *req;
1229 struct lov_obd *lov;
1230 int err = 0, rc = 0;
1234 ASSERT_LSM_MAGIC(oinfo->oi_md);
1236 if (!exp || !exp->exp_obd)
1239 lov = &exp->exp_obd->u.lov;
1241 rc = lov_prep_getattr_set(exp, oinfo, &set);
1245 cfs_list_for_each (pos, &set->set_list) {
1246 req = cfs_list_entry(pos, struct lov_request, rq_link);
1248 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1249 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1250 req->rq_oi.oi_oa->o_id, req->rq_idx);
1252 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1254 err = lov_update_common_set(set, req, rc);
1256 CERROR("error: getattr objid "LPX64" subobj "
1257 LPX64" on OST idx %d: rc = %d\n",
1258 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1264 rc = lov_fini_getattr_set(set);
1270 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1273 struct lov_request_set *lovset = (struct lov_request_set *)data;
1277 /* don't do attribute merge if this aysnc op failed */
1279 lovset->set_completes = 0;
1280 err = lov_fini_getattr_set(lovset);
1281 RETURN(rc ? rc : err);
1284 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1285 struct ptlrpc_request_set *rqset)
1287 struct lov_request_set *lovset;
1288 struct lov_obd *lov;
1290 struct lov_request *req;
1295 ASSERT_LSM_MAGIC(oinfo->oi_md);
1297 if (!exp || !exp->exp_obd)
1300 lov = &exp->exp_obd->u.lov;
1302 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1306 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1307 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1308 oinfo->oi_md->lsm_stripe_size);
1310 cfs_list_for_each (pos, &lovset->set_list) {
1311 req = cfs_list_entry(pos, struct lov_request, rq_link);
1313 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1314 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1315 req->rq_oi.oi_oa->o_id, req->rq_idx);
1316 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1317 &req->rq_oi, rqset);
1319 CERROR("error: getattr objid "LPX64" subobj "
1320 LPX64" on OST idx %d: rc = %d\n",
1321 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1327 if (!cfs_list_empty(&rqset->set_requests)) {
1329 LASSERT (rqset->set_interpret == NULL);
1330 rqset->set_interpret = lov_getattr_interpret;
1331 rqset->set_arg = (void *)lovset;
1336 lovset->set_completes = 0;
1337 err = lov_fini_getattr_set(lovset);
1338 RETURN(rc ? rc : err);
1341 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1342 struct obd_trans_info *oti)
1344 struct lov_request_set *set;
1345 struct lov_obd *lov;
1347 struct lov_request *req;
1348 int err = 0, rc = 0;
1352 ASSERT_LSM_MAGIC(oinfo->oi_md);
1354 if (!exp || !exp->exp_obd)
1357 /* for now, we only expect the following updates here */
1358 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1359 OBD_MD_FLMODE | OBD_MD_FLATIME |
1360 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1361 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1362 OBD_MD_FLGROUP | OBD_MD_FLUID |
1363 OBD_MD_FLGID | OBD_MD_FLFID |
1365 lov = &exp->exp_obd->u.lov;
1366 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1370 cfs_list_for_each (pos, &set->set_list) {
1371 req = cfs_list_entry(pos, struct lov_request, rq_link);
1373 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1375 err = lov_update_setattr_set(set, req, rc);
1377 CERROR("error: setattr objid "LPX64" subobj "
1378 LPX64" on OST idx %d: rc = %d\n",
1379 set->set_oi->oi_oa->o_id,
1380 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1385 err = lov_fini_setattr_set(set);
1391 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1394 struct lov_request_set *lovset = (struct lov_request_set *)data;
1399 lovset->set_completes = 0;
1400 err = lov_fini_setattr_set(lovset);
1401 RETURN(rc ? rc : err);
1404 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1405 needed. Otherwise, a client is waiting for responses. */
1406 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1407 struct obd_trans_info *oti,
1408 struct ptlrpc_request_set *rqset)
1410 struct lov_request_set *set;
1411 struct lov_request *req;
1413 struct lov_obd *lov;
1418 ASSERT_LSM_MAGIC(oinfo->oi_md);
1419 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1421 LASSERT(oti->oti_logcookies);
1424 if (!exp || !exp->exp_obd)
1427 lov = &exp->exp_obd->u.lov;
1428 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1432 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1433 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1434 oinfo->oi_md->lsm_stripe_size);
1436 cfs_list_for_each (pos, &set->set_list) {
1437 req = cfs_list_entry(pos, struct lov_request, rq_link);
1439 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1440 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1442 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1443 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1444 req->rq_oi.oi_oa->o_id, req->rq_idx);
1446 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1447 &req->rq_oi, oti, rqset);
1449 CERROR("error: setattr objid "LPX64" subobj "
1450 LPX64" on OST idx %d: rc = %d\n",
1451 set->set_oi->oi_oa->o_id,
1452 req->rq_oi.oi_oa->o_id,
1458 /* If we are not waiting for responses on async requests, return. */
1459 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1462 set->set_completes = 0;
1463 err = lov_fini_setattr_set(set);
1464 RETURN(rc ? rc : err);
1467 LASSERT(rqset->set_interpret == NULL);
1468 rqset->set_interpret = lov_setattr_interpret;
1469 rqset->set_arg = (void *)set;
1474 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1477 struct lov_request_set *lovset = (struct lov_request_set *)data;
1482 lovset->set_completes = 0;
1483 err = lov_fini_punch_set(lovset);
1484 RETURN(rc ? rc : err);
1487 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1488 * we can send this 'punch' to just the authoritative node and the nodes
1489 * that the punch will affect. */
1490 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1491 struct obd_trans_info *oti,
1492 struct ptlrpc_request_set *rqset)
1494 struct lov_request_set *set;
1495 struct lov_obd *lov;
1497 struct lov_request *req;
1502 ASSERT_LSM_MAGIC(oinfo->oi_md);
1504 if (!exp || !exp->exp_obd)
1507 lov = &exp->exp_obd->u.lov;
1508 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1512 cfs_list_for_each (pos, &set->set_list) {
1513 req = cfs_list_entry(pos, struct lov_request, rq_link);
1515 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1516 &req->rq_oi, NULL, rqset);
1518 CERROR("error: punch objid "LPX64" subobj "LPX64
1519 " on OST idx %d: rc = %d\n",
1520 set->set_oi->oi_oa->o_id,
1521 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1526 if (rc || cfs_list_empty(&rqset->set_requests)) {
1528 err = lov_fini_punch_set(set);
1529 RETURN(rc ? rc : err);
1532 LASSERT(rqset->set_interpret == NULL);
1533 rqset->set_interpret = lov_punch_interpret;
1534 rqset->set_arg = (void *)set;
1539 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1540 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1543 struct lov_request_set *set;
1544 struct obd_info oinfo;
1545 struct lov_obd *lov;
1547 struct lov_request *req;
1548 int err = 0, rc = 0;
1551 ASSERT_LSM_MAGIC(lsm);
1556 lov = &exp->exp_obd->u.lov;
1557 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1561 cfs_list_for_each (pos, &set->set_list) {
1562 req = cfs_list_entry(pos, struct lov_request, rq_link);
1564 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1565 req->rq_oi.oi_oa, NULL,
1566 req->rq_oi.oi_policy.l_extent.start,
1567 req->rq_oi.oi_policy.l_extent.end, capa);
1568 err = lov_update_common_set(set, req, rc);
1570 CERROR("error: fsync objid "LPX64" subobj "LPX64
1571 " on OST idx %d: rc = %d\n",
1572 set->set_oi->oi_oa->o_id,
1573 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1578 err = lov_fini_sync_set(set);
1584 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1585 obd_count oa_bufs, struct brw_page *pga)
1587 struct obd_info oinfo = { { { 0 } } };
1590 oinfo.oi_oa = lov_oinfo->oi_oa;
1592 /* The caller just wants to know if there's a chance that this
1593 * I/O can succeed */
1594 for (i = 0; i < oa_bufs; i++) {
1595 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1596 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1599 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1600 pga[i].off + pga[i].count - 1,
1604 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1605 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1609 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1617 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1618 obd_count oa_bufs, struct brw_page *pga,
1619 struct obd_trans_info *oti)
1621 struct lov_request_set *set;
1622 struct lov_request *req;
1624 struct lov_obd *lov = &exp->exp_obd->u.lov;
1628 ASSERT_LSM_MAGIC(oinfo->oi_md);
1630 if (cmd == OBD_BRW_CHECK) {
1631 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1635 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1639 cfs_list_for_each (pos, &set->set_list) {
1640 struct obd_export *sub_exp;
1641 struct brw_page *sub_pga;
1642 req = cfs_list_entry(pos, struct lov_request, rq_link);
1644 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1645 sub_pga = set->set_pga + req->rq_pgaidx;
1646 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1650 lov_update_common_set(set, req, rc);
1653 err = lov_fini_brw_set(set);
1659 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1662 struct lov_request_set *lovset = (struct lov_request_set *)data;
1664 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1668 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1669 struct ldlm_enqueue_info *einfo,
1670 struct ptlrpc_request_set *rqset)
1672 ldlm_mode_t mode = einfo->ei_mode;
1673 struct lov_request_set *set;
1674 struct lov_request *req;
1676 struct lov_obd *lov;
1681 ASSERT_LSM_MAGIC(oinfo->oi_md);
1682 LASSERT(mode == (mode & -mode));
1684 /* we should never be asked to replay a lock this way. */
1685 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1687 if (!exp || !exp->exp_obd)
1690 lov = &exp->exp_obd->u.lov;
1691 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1695 cfs_list_for_each (pos, &set->set_list) {
1696 req = cfs_list_entry(pos, struct lov_request, rq_link);
1698 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1699 &req->rq_oi, einfo, rqset);
1704 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1706 LASSERT(rqset->set_interpret == NULL);
1707 rqset->set_interpret = lov_enqueue_interpret;
1708 rqset->set_arg = (void *)set;
1712 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1716 static int lov_change_cbdata(struct obd_export *exp,
1717 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1720 struct lov_obd *lov;
1724 ASSERT_LSM_MAGIC(lsm);
1726 if (!exp || !exp->exp_obd)
1729 lov = &exp->exp_obd->u.lov;
1730 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1731 struct lov_stripe_md submd;
1732 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1734 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1735 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1739 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1740 submd.lsm_object_id = loi->loi_id;
1741 submd.lsm_object_seq = loi->loi_seq;
1742 submd.lsm_stripe_count = 0;
1743 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1749 /* find any ldlm lock of the inode in lov
1753 static int lov_find_cbdata(struct obd_export *exp,
1754 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1757 struct lov_obd *lov;
1761 ASSERT_LSM_MAGIC(lsm);
1763 if (!exp || !exp->exp_obd)
1766 lov = &exp->exp_obd->u.lov;
1767 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1768 struct lov_stripe_md submd;
1769 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1771 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1772 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1776 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1777 submd.lsm_object_id = loi->loi_id;
1778 submd.lsm_object_seq = loi->loi_seq;
1779 submd.lsm_stripe_count = 0;
1780 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1788 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1789 __u32 mode, struct lustre_handle *lockh)
1791 struct lov_request_set *set;
1792 struct obd_info oinfo;
1793 struct lov_request *req;
1795 struct lov_obd *lov;
1796 struct lustre_handle *lov_lockhp;
1797 int err = 0, rc = 0;
1800 ASSERT_LSM_MAGIC(lsm);
1802 if (!exp || !exp->exp_obd)
1805 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1807 lov = &exp->exp_obd->u.lov;
1808 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1812 cfs_list_for_each (pos, &set->set_list) {
1813 req = cfs_list_entry(pos, struct lov_request, rq_link);
1814 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1816 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1817 req->rq_oi.oi_md, mode, lov_lockhp);
1818 rc = lov_update_common_set(set, req, rc);
1820 CERROR("error: cancel objid "LPX64" subobj "
1821 LPX64" on OST idx %d: rc = %d\n",
1823 req->rq_oi.oi_md->lsm_object_id,
1829 lov_fini_cancel_set(set);
1833 static int lov_cancel_unused(struct obd_export *exp,
1834 struct lov_stripe_md *lsm,
1835 ldlm_cancel_flags_t flags, void *opaque)
1837 struct lov_obd *lov;
1841 if (!exp || !exp->exp_obd)
1844 lov = &exp->exp_obd->u.lov;
1846 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1848 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1851 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1859 ASSERT_LSM_MAGIC(lsm);
1861 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1862 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1863 struct lov_stripe_md submd;
1864 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1867 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1868 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1872 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1873 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1875 submd.lsm_object_id = loi->loi_id;
1876 submd.lsm_object_seq = loi->loi_seq;
1877 submd.lsm_stripe_count = 0;
1878 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1879 &submd, flags, opaque);
1880 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1881 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1882 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1883 loi->loi_id, loi->loi_ost_idx, err);
1891 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1893 struct lov_request_set *lovset = (struct lov_request_set *)data;
1898 lovset->set_completes = 0;
1900 err = lov_fini_statfs_set(lovset);
1901 RETURN(rc ? rc : err);
1904 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1905 __u64 max_age, struct ptlrpc_request_set *rqset)
1907 struct lov_request_set *set;
1908 struct lov_request *req;
1910 struct lov_obd *lov;
1914 LASSERT(oinfo != NULL);
1915 LASSERT(oinfo->oi_osfs != NULL);
1918 rc = lov_prep_statfs_set(obd, oinfo, &set);
1922 cfs_list_for_each (pos, &set->set_list) {
1923 struct obd_device *osc_obd;
1925 req = cfs_list_entry(pos, struct lov_request, rq_link);
1927 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1928 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1933 if (rc || cfs_list_empty(&rqset->set_requests)) {
1936 set->set_completes = 0;
1937 err = lov_fini_statfs_set(set);
1938 RETURN(rc ? rc : err);
1941 LASSERT(rqset->set_interpret == NULL);
1942 rqset->set_interpret = lov_statfs_interpret;
1943 rqset->set_arg = (void *)set;
1947 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1948 __u64 max_age, __u32 flags)
1950 struct ptlrpc_request_set *set = NULL;
1951 struct obd_info oinfo = { { { 0 } } };
1956 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1957 * statfs requests */
1958 set = ptlrpc_prep_set();
1962 oinfo.oi_osfs = osfs;
1963 oinfo.oi_flags = flags;
1964 rc = lov_statfs_async(obd, &oinfo, max_age, set);
1966 rc = ptlrpc_set_wait(set);
1967 ptlrpc_set_destroy(set);
1972 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1973 void *karg, void *uarg)
1975 struct obd_device *obddev = class_exp2obd(exp);
1976 struct lov_obd *lov = &obddev->u.lov;
1977 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1978 struct obd_uuid *uuidp;
1982 case IOC_OBD_STATFS: {
1983 struct obd_ioctl_data *data = karg;
1984 struct obd_device *osc_obd;
1985 struct obd_statfs stat_buf = {0};
1988 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1989 if ((index >= count))
1992 if (!lov->lov_tgts[index])
1993 /* Try again with the next index */
1995 if (!lov->lov_tgts[index]->ltd_active)
1998 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2003 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2004 min((int) data->ioc_plen2,
2005 (int) sizeof(struct obd_uuid))))
2008 /* got statfs data */
2009 rc = obd_statfs(osc_obd, &stat_buf,
2010 cfs_time_current_64() - CFS_HZ, 0);
2013 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2014 min((int) data->ioc_plen1,
2015 (int) sizeof(stat_buf))))
2019 case OBD_IOC_LOV_GET_CONFIG: {
2020 struct obd_ioctl_data *data;
2021 struct lov_desc *desc;
2026 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2029 data = (struct obd_ioctl_data *)buf;
2031 if (sizeof(*desc) > data->ioc_inllen1) {
2032 obd_ioctl_freedata(buf, len);
2036 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2037 obd_ioctl_freedata(buf, len);
2041 if (sizeof(__u32) * count > data->ioc_inllen3) {
2042 obd_ioctl_freedata(buf, len);
2046 desc = (struct lov_desc *)data->ioc_inlbuf1;
2047 memcpy(desc, &(lov->desc), sizeof(*desc));
2049 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2050 genp = (__u32 *)data->ioc_inlbuf3;
2051 /* the uuid will be empty for deleted OSTs */
2052 for (i = 0; i < count; i++, uuidp++, genp++) {
2053 if (!lov->lov_tgts[i])
2055 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2056 *genp = lov->lov_tgts[i]->ltd_gen;
2059 if (cfs_copy_to_user((void *)uarg, buf, len))
2061 obd_ioctl_freedata(buf, len);
2064 case LL_IOC_LOV_SETSTRIPE:
2065 rc = lov_setstripe(exp, len, karg, uarg);
2067 case LL_IOC_LOV_GETSTRIPE:
2068 rc = lov_getstripe(exp, karg, uarg);
2070 case LL_IOC_LOV_SETEA:
2071 rc = lov_setea(exp, karg, uarg);
2073 case OBD_IOC_QUOTACTL: {
2074 struct if_quotactl *qctl = karg;
2075 struct lov_tgt_desc *tgt = NULL;
2076 struct obd_quotactl *oqctl;
2078 if (qctl->qc_valid == QC_OSTIDX) {
2079 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2082 tgt = lov->lov_tgts[qctl->qc_idx];
2083 if (!tgt || !tgt->ltd_exp)
2085 } else if (qctl->qc_valid == QC_UUID) {
2086 for (i = 0; i < count; i++) {
2087 tgt = lov->lov_tgts[i];
2089 !obd_uuid_equals(&tgt->ltd_uuid,
2093 if (tgt->ltd_exp == NULL)
2105 LASSERT(tgt && tgt->ltd_exp);
2106 OBD_ALLOC_PTR(oqctl);
2110 QCTL_COPY(oqctl, qctl);
2111 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2113 QCTL_COPY(qctl, oqctl);
2114 qctl->qc_valid = QC_OSTIDX;
2115 qctl->obd_uuid = tgt->ltd_uuid;
2117 OBD_FREE_PTR(oqctl);
2126 for (i = 0; i < count; i++) {
2129 /* OST was disconnected */
2130 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2133 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2135 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2138 if (lov->lov_tgts[i]->ltd_active) {
2139 CDEBUG(err == -ENOTTY ?
2140 D_IOCTL : D_WARNING,
2141 "iocontrol OSC %s on OST "
2142 "idx %d cmd %x: err = %d\n",
2143 lov_uuid2str(lov, i),
2160 #define FIEMAP_BUFFER_SIZE 4096
2163 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2164 * call. The local end offset and the device are sent in the first
2165 * fm_extent. This function calculates the stripe number from the index.
2166 * This function returns a stripe_no on which mapping is to be restarted.
2168 * This function returns fm_end_offset which is the in-OST offset at which
2169 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2170 * will re-calculate proper offset in next stripe.
2171 * Note that the first extent is passed to lov_get_info via the value field.
2173 * \param fiemap fiemap request header
2174 * \param lsm striping information for the file
2175 * \param fm_start logical start of mapping
2176 * \param fm_end logical end of mapping
2177 * \param start_stripe starting stripe will be returned in this
2179 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2180 struct lov_stripe_md *lsm, obd_size fm_start,
2181 obd_size fm_end, int *start_stripe)
2183 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2184 obd_off lun_start, lun_end;
2185 obd_size fm_end_offset;
2186 int stripe_no = -1, i;
2188 if (fiemap->fm_extent_count == 0 ||
2189 fiemap->fm_extents[0].fe_logical == 0)
2192 /* Find out stripe_no from ost_index saved in the fe_device */
2193 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2194 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2195 fiemap->fm_extents[0].fe_device) {
2201 /* If we have finished mapping on previous device, shift logical
2202 * offset to start of next device */
2203 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2204 &lun_start, &lun_end)) != 0 &&
2205 local_end < lun_end) {
2206 fm_end_offset = local_end;
2207 *start_stripe = stripe_no;
2209 /* This is a special value to indicate that caller should
2210 * calculate offset in next stripe. */
2212 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2215 return fm_end_offset;
2219 * We calculate on which OST the mapping will end. If the length of mapping
2220 * is greater than (stripe_size * stripe_count) then the last_stripe will
2221 * will be one just before start_stripe. Else we check if the mapping
2222 * intersects each OST and find last_stripe.
2223 * This function returns the last_stripe and also sets the stripe_count
2224 * over which the mapping is spread
2226 * \param lsm striping information for the file
2227 * \param fm_start logical start of mapping
2228 * \param fm_end logical end of mapping
2229 * \param start_stripe starting stripe of the mapping
2230 * \param stripe_count the number of stripes across which to map is returned
2232 * \retval last_stripe return the last stripe of the mapping
2234 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2235 obd_size fm_end, int start_stripe,
2239 obd_off obd_start, obd_end;
2242 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2243 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2245 *stripe_count = lsm->lsm_stripe_count;
2247 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2248 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2249 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2250 &obd_start, &obd_end)) == 0)
2254 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2261 * Set fe_device and copy extents from local buffer into main return buffer.
2263 * \param fiemap fiemap request header
2264 * \param lcl_fm_ext array of local fiemap extents to be copied
2265 * \param ost_index OST index to be written into the fm_device field for each
2267 * \param ext_count number of extents to be copied
2268 * \param current_extent where to start copying in main extent array
2270 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2271 struct ll_fiemap_extent *lcl_fm_ext,
2272 int ost_index, unsigned int ext_count,
2278 for (ext = 0; ext < ext_count; ext++) {
2279 lcl_fm_ext[ext].fe_device = ost_index;
2280 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2283 /* Copy fm_extent's from fm_local to return buffer */
2284 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2285 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2289 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2290 * This also handles the restarting of FIEMAP calls in case mapping overflows
2291 * the available number of extents in single call.
2293 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2294 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2296 struct ll_fiemap_info_key *fm_key = key;
2297 struct ll_user_fiemap *fiemap = val;
2298 struct ll_user_fiemap *fm_local = NULL;
2299 struct ll_fiemap_extent *lcl_fm_ext;
2301 unsigned int get_num_extents = 0;
2302 int ost_index = 0, actual_start_stripe, start_stripe;
2303 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2305 int current_extent = 0, rc = 0, i;
2306 int ost_eof = 0; /* EOF for object */
2307 int ost_done = 0; /* done with required mapping for this OST? */
2309 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2310 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2315 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2316 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2318 OBD_ALLOC(fm_local, buffer_size);
2319 if (fm_local == NULL)
2320 GOTO(out, rc = -ENOMEM);
2321 lcl_fm_ext = &fm_local->fm_extents[0];
2323 count_local = fiemap_size_to_count(buffer_size);
2325 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2326 fm_start = fiemap->fm_start;
2327 fm_length = fiemap->fm_length;
2328 /* Calculate start stripe, last stripe and length of mapping */
2329 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2330 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2331 fm_start + fm_length - 1);
2332 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2333 if (fm_end > fm_key->oa.o_size)
2334 fm_end = fm_key->oa.o_size;
2336 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2337 actual_start_stripe, &stripe_count);
2339 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2342 if (fiemap->fm_extent_count == 0) {
2343 get_num_extents = 1;
2347 /* Check each stripe */
2348 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2349 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2350 obd_size req_fm_len; /* Stores length of required mapping */
2351 obd_size len_mapped_single_call;
2352 obd_off lun_start, lun_end, obd_object_end;
2353 unsigned int ext_count;
2355 cur_stripe_wrap = cur_stripe;
2357 /* Find out range of mapping on this stripe */
2358 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2359 &lun_start, &obd_object_end)) == 0)
2362 /* If this is a continuation FIEMAP call and we are on
2363 * starting stripe then lun_start needs to be set to
2365 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2366 lun_start = fm_end_offset;
2368 if (fm_length != ~0ULL) {
2369 /* Handle fm_start + fm_length overflow */
2370 if (fm_start + fm_length < fm_start)
2371 fm_length = ~0ULL - fm_start;
2372 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2378 if (lun_start == lun_end)
2381 req_fm_len = obd_object_end - lun_start;
2382 fm_local->fm_length = 0;
2383 len_mapped_single_call = 0;
2385 /* If the output buffer is very large and the objects have many
2386 * extents we may need to loop on a single OST repeatedly */
2390 if (get_num_extents == 0) {
2391 /* Don't get too many extents. */
2392 if (current_extent + count_local >
2393 fiemap->fm_extent_count)
2394 count_local = fiemap->fm_extent_count -
2398 lun_start += len_mapped_single_call;
2399 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2400 req_fm_len = fm_local->fm_length;
2401 fm_local->fm_extent_count = count_local;
2402 fm_local->fm_mapped_extents = 0;
2403 fm_local->fm_flags = fiemap->fm_flags;
2405 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2406 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2407 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2409 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2410 GOTO(out, rc = -EINVAL);
2412 /* If OST is inactive, return extent with UNKNOWN flag */
2413 if (!lov->lov_tgts[ost_index]->ltd_active) {
2414 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2415 fm_local->fm_mapped_extents = 1;
2417 lcl_fm_ext[0].fe_logical = lun_start;
2418 lcl_fm_ext[0].fe_length = obd_object_end -
2420 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2425 fm_local->fm_start = lun_start;
2426 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2427 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2428 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2429 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2430 keylen, key, vallen, fm_local, lsm);
2435 ext_count = fm_local->fm_mapped_extents;
2436 if (ext_count == 0) {
2438 /* If last stripe has hole at the end,
2439 * then we need to return */
2440 if (cur_stripe_wrap == last_stripe) {
2441 fiemap->fm_mapped_extents = 0;
2447 /* If we just need num of extents then go to next device */
2448 if (get_num_extents) {
2449 current_extent += ext_count;
2453 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2454 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2456 /* Have we finished mapping on this device? */
2457 if (req_fm_len <= len_mapped_single_call)
2460 /* Clear the EXTENT_LAST flag which can be present on
2462 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2463 lcl_fm_ext[ext_count - 1].fe_flags &=
2464 ~FIEMAP_EXTENT_LAST;
2466 curr_loc = lov_stripe_size(lsm,
2467 lcl_fm_ext[ext_count - 1].fe_logical+
2468 lcl_fm_ext[ext_count - 1].fe_length,
2470 if (curr_loc >= fm_key->oa.o_size)
2473 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2474 ost_index, ext_count,
2477 current_extent += ext_count;
2479 /* Ran out of available extents? */
2480 if (current_extent >= fiemap->fm_extent_count)
2482 } while (ost_done == 0 && ost_eof == 0);
2484 if (cur_stripe_wrap == last_stripe)
2489 /* Indicate that we are returning device offsets unless file just has
2491 if (lsm->lsm_stripe_count > 1)
2492 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2494 if (get_num_extents)
2495 goto skip_last_device_calc;
2497 /* Check if we have reached the last stripe and whether mapping for that
2498 * stripe is done. */
2499 if (cur_stripe_wrap == last_stripe) {
2500 if (ost_done || ost_eof)
2501 fiemap->fm_extents[current_extent - 1].fe_flags |=
2505 skip_last_device_calc:
2506 fiemap->fm_mapped_extents = current_extent;
2509 OBD_FREE(fm_local, buffer_size);
2513 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2514 void *key, __u32 *vallen, void *val,
2515 struct lov_stripe_md *lsm)
2517 struct obd_device *obddev = class_exp2obd(exp);
2518 struct lov_obd *lov = &obddev->u.lov;
2522 if (!vallen || !val)
2527 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2530 struct ldlm_lock *lock;
2532 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2533 struct lov_oinfo *loi;
2534 __u32 *stripe = val;
2536 if (*vallen < sizeof(*stripe))
2537 GOTO(out, rc = -EFAULT);
2538 *vallen = sizeof(*stripe);
2540 /* XXX This is another one of those bits that will need to
2541 * change if we ever actually support nested LOVs. It uses
2542 * the lock's export to find out which stripe it is. */
2543 /* XXX - it's assumed all the locks for deleted OSTs have
2544 * been cancelled. Also, the export for deleted OSTs will
2545 * be NULL and won't match the lock's export. */
2546 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2547 loi = lsm->lsm_oinfo[i];
2548 if (!lov->lov_tgts[loi->loi_ost_idx])
2550 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2551 data->lock->l_conn_export &&
2552 osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2557 LDLM_ERROR(data->lock, "lock on inode without such object");
2558 dump_lsm(D_ERROR, lsm);
2559 GOTO(out, rc = -ENXIO);
2560 } else if (KEY_IS(KEY_LAST_ID)) {
2561 struct obd_id_info *info = val;
2562 __u32 size = sizeof(obd_id);
2563 struct lov_tgt_desc *tgt;
2565 LASSERT(*vallen == sizeof(struct obd_id_info));
2566 tgt = lov->lov_tgts[info->idx];
2568 if (!tgt || !tgt->ltd_active)
2569 GOTO(out, rc = -ESRCH);
2571 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2573 } else if (KEY_IS(KEY_LOVDESC)) {
2574 struct lov_desc *desc_ret = val;
2575 *desc_ret = lov->desc;
2578 } else if (KEY_IS(KEY_FIEMAP)) {
2579 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2581 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2582 struct lov_tgt_desc *tgt;
2583 __u64 ost_idx = *((__u64*)val);
2585 LASSERT(*vallen == sizeof(__u64));
2586 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2587 tgt = lov->lov_tgts[ost_idx];
2589 if (!tgt || !tgt->ltd_exp)
2590 GOTO(out, rc = -ESRCH);
2592 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2603 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2604 void *key, obd_count vallen, void *val,
2605 struct ptlrpc_request_set *set)
2607 struct obd_device *obddev = class_exp2obd(exp);
2608 struct lov_obd *lov = &obddev->u.lov;
2611 struct lov_tgt_desc *tgt;
2612 unsigned incr, check_uuid,
2613 do_inactive, no_set;
2614 unsigned next_id = 0, mds_con = 0, capa = 0;
2617 incr = check_uuid = do_inactive = no_set = 0;
2620 set = ptlrpc_prep_set();
2626 count = lov->desc.ld_tgt_count;
2628 if (KEY_IS(KEY_NEXT_ID)) {
2629 count = vallen / sizeof(struct obd_id_info);
2630 vallen = sizeof(obd_id);
2631 incr = sizeof(struct obd_id_info);
2634 } else if (KEY_IS(KEY_CHECKSUM)) {
2636 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2637 /* use defaults: do_inactive = incr = 0; */
2638 } else if (KEY_IS(KEY_MDS_CONN)) {
2640 } else if (KEY_IS(KEY_CAPA_KEY)) {
2644 for (i = 0; i < count; i++, val = (char *)val + incr) {
2646 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2648 tgt = lov->lov_tgts[i];
2650 /* OST was disconnected */
2651 if (!tgt || !tgt->ltd_exp)
2654 /* OST is inactive and we don't want inactive OSCs */
2655 if (!tgt->ltd_active && !do_inactive)
2659 struct mds_group_info *mgi;
2661 LASSERT(vallen == sizeof(*mgi));
2662 mgi = (struct mds_group_info *)val;
2664 /* Only want a specific OSC */
2665 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2669 err = obd_set_info_async(tgt->ltd_exp,
2670 keylen, key, sizeof(int),
2672 } else if (next_id) {
2673 err = obd_set_info_async(tgt->ltd_exp,
2674 keylen, key, vallen,
2675 ((struct obd_id_info*)val)->data, set);
2677 struct mds_capa_info *info = (struct mds_capa_info*)val;
2679 LASSERT(vallen == sizeof(*info));
2681 /* Only want a specific OSC */
2683 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2686 err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2687 sizeof(*info->capa),
2691 /* Only want a specific OSC */
2693 !obd_uuid_equals(val, &tgt->ltd_uuid))
2696 err = obd_set_info_async(tgt->ltd_exp,
2697 keylen, key, vallen, val, set);
2706 err = ptlrpc_set_wait(set);
2709 ptlrpc_set_destroy(set);
2714 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2719 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2720 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2721 if (loi->loi_ar.ar_rc && !rc)
2722 rc = loi->loi_ar.ar_rc;
2723 loi->loi_ar.ar_rc = 0;
2727 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2730 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2731 int cmd, __u64 *offset)
2733 __u32 ssize = lsm->lsm_stripe_size;
2737 do_div(start, ssize);
2738 start = start * ssize;
2740 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2741 ", end "LPU64"\n", *offset, ssize, start,
2743 if (cmd == OBD_CALC_STRIPE_END) {
2744 *offset = start + ssize - 1;
2745 } else if (cmd == OBD_CALC_STRIPE_START) {
2754 void lov_stripe_lock(struct lov_stripe_md *md)
2756 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2757 cfs_spin_lock(&md->lsm_lock);
2758 LASSERT(md->lsm_lock_owner == 0);
2759 md->lsm_lock_owner = cfs_curproc_pid();
2761 EXPORT_SYMBOL(lov_stripe_lock);
2763 void lov_stripe_unlock(struct lov_stripe_md *md)
2765 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2766 md->lsm_lock_owner = 0;
2767 cfs_spin_unlock(&md->lsm_lock);
2769 EXPORT_SYMBOL(lov_stripe_unlock);
2772 struct obd_ops lov_obd_ops = {
2773 .o_owner = THIS_MODULE,
2774 .o_setup = lov_setup,
2775 .o_precleanup = lov_precleanup,
2776 .o_cleanup = lov_cleanup,
2777 //.o_process_config = lov_process_config,
2778 .o_connect = lov_connect,
2779 .o_disconnect = lov_disconnect,
2780 .o_statfs = lov_statfs,
2781 .o_statfs_async = lov_statfs_async,
2782 .o_packmd = lov_packmd,
2783 .o_unpackmd = lov_unpackmd,
2784 .o_create = lov_create,
2785 .o_destroy = lov_destroy,
2786 .o_getattr = lov_getattr,
2787 .o_getattr_async = lov_getattr_async,
2788 .o_setattr = lov_setattr,
2789 .o_setattr_async = lov_setattr_async,
2791 .o_merge_lvb = lov_merge_lvb,
2792 .o_adjust_kms = lov_adjust_kms,
2793 .o_punch = lov_punch,
2795 .o_enqueue = lov_enqueue,
2796 .o_change_cbdata = lov_change_cbdata,
2797 .o_find_cbdata = lov_find_cbdata,
2798 .o_cancel = lov_cancel,
2799 .o_cancel_unused = lov_cancel_unused,
2800 .o_iocontrol = lov_iocontrol,
2801 .o_get_info = lov_get_info,
2802 .o_set_info_async = lov_set_info_async,
2803 .o_extent_calc = lov_extent_calc,
2804 .o_llog_init = lov_llog_init,
2805 .o_llog_finish = lov_llog_finish,
2806 .o_notify = lov_notify,
2807 .o_pool_new = lov_pool_new,
2808 .o_pool_rem = lov_pool_remove,
2809 .o_pool_add = lov_pool_add,
2810 .o_pool_del = lov_pool_del,
2811 .o_getref = lov_getref,
2812 .o_putref = lov_putref,
2815 static quota_interface_t *quota_interface;
2816 extern quota_interface_t lov_quota_interface;
2818 cfs_mem_cache_t *lov_oinfo_slab;
2820 extern struct lu_kmem_descr lov_caches[];
2822 int __init lov_init(void)
2824 struct lprocfs_static_vars lvars = { 0 };
2828 /* print an address of _any_ initialized kernel symbol from this
2829 * module, to allow debugging with gdb that doesn't support data
2830 * symbols from modules.*/
2831 CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2833 rc = lu_kmem_init(lov_caches);
2837 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2838 sizeof(struct lov_oinfo),
2839 0, CFS_SLAB_HWCACHE_ALIGN);
2840 if (lov_oinfo_slab == NULL) {
2841 lu_kmem_fini(lov_caches);
2844 lprocfs_lov_init_vars(&lvars);
2846 cfs_request_module("lquota");
2847 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2848 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2850 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2851 LUSTRE_LOV_NAME, &lov_device_type);
2854 if (quota_interface)
2855 PORTAL_SYMBOL_PUT(lov_quota_interface);
2856 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2858 lu_kmem_fini(lov_caches);
2865 static void /*__exit*/ lov_exit(void)
2869 lu_device_type_fini(&lov_device_type);
2870 lu_kmem_fini(lov_caches);
2872 if (quota_interface)
2873 PORTAL_SYMBOL_PUT(lov_quota_interface);
2875 class_unregister_type(LUSTRE_LOV_NAME);
2876 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2880 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2881 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2882 MODULE_LICENSE("GPL");
2884 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);