4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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_lock(&lov->lov_lock);
79 cfs_atomic_inc(&lov->lov_refcount);
80 cfs_mutex_unlock(&lov->lov_lock);
84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
86 static void lov_putref(struct obd_device *obd)
88 struct lov_obd *lov = &obd->u.lov;
90 cfs_mutex_lock(&lov->lov_lock);
91 /* ok to dec to 0 more than once -- ltd_exp's will be null */
92 if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
95 struct lov_tgt_desc *tgt, *n;
96 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
98 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
99 tgt = lov->lov_tgts[i];
101 if (!tgt || !tgt->ltd_reap)
103 cfs_list_add(&tgt->ltd_kill, &kill);
104 /* XXX - right now there is a dependency on ld_tgt_count
105 * being the maximum tgt index for computing the
106 * mds_max_easize. So we can't shrink it. */
107 lov_ost_pool_remove(&lov->lov_packed, i);
108 lov->lov_tgts[i] = NULL;
109 lov->lov_death_row--;
111 cfs_mutex_unlock(&lov->lov_lock);
113 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
114 cfs_list_del(&tgt->ltd_kill);
116 __lov_del_obd(obd, tgt);
119 cfs_mutex_unlock(&lov->lov_lock);
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
124 enum obd_notify_event ev);
125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
126 enum obd_notify_event ev, void *data);
129 #define MAX_STRING_SIZE 128
130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
131 struct obd_connect_data *data)
133 struct lov_obd *lov = &obd->u.lov;
134 struct obd_uuid *tgt_uuid;
135 struct obd_device *tgt_obd;
136 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
137 struct obd_import *imp;
139 cfs_proc_dir_entry_t *lov_proc_dir;
144 if (!lov->lov_tgts[index])
147 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
148 tgt_obd = lov->lov_tgts[index]->ltd_obd;
150 if (!tgt_obd->obd_set_up) {
151 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
155 /* override the sp_me from lov */
156 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
158 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
159 data->ocd_index = index;
162 * Divine LOV knows that OBDs under it are OSCs.
164 imp = tgt_obd->u.cli.cl_import;
167 tgt_obd->obd_no_recov = 0;
168 /* FIXME this is probably supposed to be
169 ptlrpc_set_import_active. Horrible naming. */
170 ptlrpc_activate_import(imp);
173 rc = obd_register_observer(tgt_obd, obd);
175 CERROR("Target %s register_observer error %d\n",
176 obd_uuid2str(tgt_uuid), rc);
181 if (imp->imp_invalid) {
182 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
183 "disabled\n", obd_uuid2str(tgt_uuid));
187 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
188 &lov_osc_uuid, data, NULL);
189 if (rc || !lov->lov_tgts[index]->ltd_exp) {
190 CERROR("Target %s connect error %d\n",
191 obd_uuid2str(tgt_uuid), rc);
195 lov->lov_tgts[index]->ltd_reap = 0;
197 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
198 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
201 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
203 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
204 cfs_proc_dir_entry_t *osc_symlink;
206 LASSERT(osc_obd != NULL);
207 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
208 LASSERT(osc_obd->obd_type->typ_name != NULL);
210 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
213 osc_obd->obd_type->typ_name,
215 if (osc_symlink == NULL) {
216 CERROR("could not register LOV target "
217 "/proc/fs/lustre/%s/%s/target_obds/%s.",
218 obd->obd_type->typ_name, obd->obd_name,
220 lprocfs_remove(&lov_proc_dir);
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,
387 enum obd_notify_event ev)
389 struct lov_obd *lov = &obd->u.lov;
390 struct lov_tgt_desc *tgt;
391 int index, activate, active;
394 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
395 lov, uuid->uuid, ev);
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 (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
414 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
416 if (lov->lov_tgts[index]->ltd_activate == activate) {
417 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
418 uuid->uuid, activate ? "" : "de");
420 lov->lov_tgts[index]->ltd_activate = activate;
421 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
422 activate ? "" : "de", obd_uuid2str(uuid));
425 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
426 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
428 if (lov->lov_tgts[index]->ltd_active == active) {
429 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
430 uuid->uuid, active ? "" : "in");
433 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
434 obd_uuid2str(uuid), active ? "" : "in");
437 lov->lov_tgts[index]->ltd_active = active;
439 lov->desc.ld_active_tgt_count++;
440 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
442 lov->desc.ld_active_tgt_count--;
443 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
445 /* remove any old qos penalty */
446 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
448 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
456 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
457 enum obd_notify_event ev, void *data)
462 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
463 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
464 struct obd_uuid *uuid;
468 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
469 CERROR("unexpected notification of %s %s!\n",
470 watched->obd_type->typ_name,
474 uuid = &watched->u.cli.cl_target_uuid;
476 /* Set OSC as active before notifying the observer, so the
477 * observer can use the OSC normally.
479 rc = lov_set_osc_active(obd, uuid, ev);
481 CERROR("event(%d) of %s failed: %d\n", ev,
482 obd_uuid2str(uuid), rc);
485 /* active event should be pass lov target index as data */
489 /* Pass the notification up the chain. */
491 rc = obd_notify_observer(obd, watched, ev, data);
493 /* NULL watched means all osc's in the lov (only for syncs) */
494 /* sync event should be send lov idx as data */
495 struct lov_obd *lov = &obd->u.lov;
499 is_sync = (ev == OBD_NOTIFY_SYNC) ||
500 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
503 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
504 if (!lov->lov_tgts[i])
507 /* don't send sync event if target not
508 * connected/activated */
509 if (is_sync && !lov->lov_tgts[i]->ltd_active)
512 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
515 CERROR("%s: notify %s of %s failed %d\n",
517 obd->obd_observer->obd_name,
518 lov->lov_tgts[i]->ltd_obd->obd_name,
528 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
529 __u32 index, int gen, int active)
531 struct lov_obd *lov = &obd->u.lov;
532 struct lov_tgt_desc *tgt;
533 struct obd_device *tgt_obd;
537 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
538 uuidp->uuid, index, gen, active);
541 CERROR("request to add OBD %s with invalid generation: %d\n",
546 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
551 cfs_mutex_lock(&lov->lov_lock);
553 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
554 tgt = lov->lov_tgts[index];
555 CERROR("UUID %s already assigned at LOV target index %d\n",
556 obd_uuid2str(&tgt->ltd_uuid), index);
557 cfs_mutex_unlock(&lov->lov_lock);
561 if (index >= lov->lov_tgt_size) {
562 /* We need to reallocate the lov target array. */
563 struct lov_tgt_desc **newtgts, **old = NULL;
564 __u32 newsize, oldsize = 0;
566 newsize = max(lov->lov_tgt_size, (__u32)2);
567 while (newsize < index + 1)
568 newsize = newsize << 1;
569 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
570 if (newtgts == NULL) {
571 cfs_mutex_unlock(&lov->lov_lock);
575 if (lov->lov_tgt_size) {
576 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
579 oldsize = lov->lov_tgt_size;
582 lov->lov_tgts = newtgts;
583 lov->lov_tgt_size = newsize;
588 OBD_FREE(old, sizeof(*old) * oldsize);
590 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
591 lov->lov_tgts, lov->lov_tgt_size);
596 cfs_mutex_unlock(&lov->lov_lock);
600 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
602 cfs_mutex_unlock(&lov->lov_lock);
607 tgt->ltd_uuid = *uuidp;
608 tgt->ltd_obd = tgt_obd;
609 /* XXX - add a sanity check on the generation number. */
611 tgt->ltd_index = index;
612 tgt->ltd_activate = active;
613 lov->lov_tgts[index] = tgt;
614 if (index >= lov->desc.ld_tgt_count)
615 lov->desc.ld_tgt_count = index + 1;
617 cfs_mutex_unlock(&lov->lov_lock);
619 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
620 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
622 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
624 if (lov->lov_connects == 0) {
625 /* lov_connect hasn't been called yet. We'll do the
626 lov_connect_obd on this target when that fn first runs,
627 because we don't know the connect flags yet. */
633 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
637 /* connect to administrative disabled ost */
641 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
642 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
647 CERROR("add failed (%d), deleting %s\n", rc,
648 obd_uuid2str(&tgt->ltd_uuid));
649 lov_del_target(obd, index, 0, 0);
655 /* Schedule a target for deletion */
656 int lov_del_target(struct obd_device *obd, __u32 index,
657 struct obd_uuid *uuidp, int gen)
659 struct lov_obd *lov = &obd->u.lov;
660 int count = lov->desc.ld_tgt_count;
664 if (index >= count) {
665 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
672 if (!lov->lov_tgts[index]) {
673 CERROR("LOV target at index %d is not setup.\n", index);
674 GOTO(out, rc = -EINVAL);
677 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
678 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
679 lov_uuid2str(lov, index), index,
680 obd_uuid2str(uuidp));
681 GOTO(out, rc = -EINVAL);
684 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
685 lov_uuid2str(lov, index), index,
686 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
687 lov->lov_tgts[index]->ltd_active);
689 lov->lov_tgts[index]->ltd_reap = 1;
690 lov->lov_death_row++;
691 /* we really delete it from obd_putref */
698 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
700 struct obd_device *osc_obd;
703 LASSERT(tgt->ltd_reap);
705 osc_obd = class_exp2obd(tgt->ltd_exp);
707 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
709 osc_obd ? osc_obd->obd_name : "<no obd>");
712 lov_disconnect_obd(obd, tgt);
716 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
717 do it ourselves. And we can't do it from lov_cleanup,
718 because we just lost our only reference to it. */
720 class_manual_cleanup(osc_obd);
723 void lov_fix_desc_stripe_size(__u64 *val)
725 if (*val < PTLRPC_MAX_BRW_SIZE) {
726 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
727 PTLRPC_MAX_BRW_SIZE);
728 *val = PTLRPC_MAX_BRW_SIZE;
729 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
730 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
731 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
733 *val, LOV_MIN_STRIPE_SIZE);
737 void lov_fix_desc_stripe_count(__u32 *val)
743 void lov_fix_desc_pattern(__u32 *val)
745 /* from lov_setstripe */
746 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
747 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
752 void lov_fix_desc_qos_maxage(__u32 *val)
756 *val = QOS_DEFAULT_MAXAGE;
759 void lov_fix_desc(struct lov_desc *desc)
761 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
762 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
763 lov_fix_desc_pattern(&desc->ld_pattern);
764 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
767 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
769 struct lprocfs_static_vars lvars = { 0 };
770 struct lov_desc *desc;
771 struct lov_obd *lov = &obd->u.lov;
775 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
776 CERROR("LOV setup requires a descriptor\n");
780 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
782 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
783 CERROR("descriptor size wrong: %d > %d\n",
784 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
788 if (desc->ld_magic != LOV_DESC_MAGIC) {
789 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
790 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
791 obd->obd_name, desc);
792 lustre_swab_lov_desc(desc);
794 CERROR("%s: Bad lov desc magic: %#x\n",
795 obd->obd_name, desc->ld_magic);
802 desc->ld_active_tgt_count = 0;
804 lov->lov_tgt_size = 0;
806 cfs_mutex_init(&lov->lov_lock);
807 cfs_atomic_set(&lov->lov_refcount, 0);
808 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
809 cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
810 lov->lov_sp_me = LUSTRE_SP_CLI;
811 lov->lov_qos.lq_dirty = 1;
812 lov->lov_qos.lq_rr.lqr_dirty = 1;
813 lov->lov_qos.lq_reset = 1;
814 /* Default priority is toward free space balance */
815 lov->lov_qos.lq_prio_free = 232;
816 /* Default threshold for rr (roughly 17%) */
817 lov->lov_qos.lq_threshold_rr = 43;
818 /* Init statfs fields */
819 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
820 if (NULL == lov->lov_qos.lq_statfs_data)
822 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
824 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
826 HASH_POOLS_BKT_BITS, 0,
829 &pool_hash_operations,
831 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
832 lov->lov_pool_count = 0;
833 rc = lov_ost_pool_init(&lov->lov_packed, 0);
835 GOTO(out_free_statfs, rc);
836 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
838 GOTO(out_free_lov_packed, rc);
840 lprocfs_lov_init_vars(&lvars);
841 lprocfs_obd_setup(obd, lvars.obd_vars);
846 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
847 0444, &lov_proc_target_fops, obd);
849 CWARN("Error adding the target_obd file\n");
852 lov->lov_pool_proc_entry = lprocfs_register("pools",
859 lov_ost_pool_free(&lov->lov_packed);
861 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
865 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
868 struct lov_obd *lov = &obd->u.lov;
873 case OBD_CLEANUP_EARLY: {
875 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
876 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
878 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
883 case OBD_CLEANUP_EXPORTS:
884 lprocfs_obd_cleanup(obd);
885 rc = obd_llog_finish(obd, 0);
887 CERROR("failed to cleanup llogging subsystems\n");
893 static int lov_cleanup(struct obd_device *obd)
895 struct lov_obd *lov = &obd->u.lov;
896 cfs_list_t *pos, *tmp;
897 struct pool_desc *pool;
900 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
901 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
902 /* free pool structs */
903 CDEBUG(D_INFO, "delete pool %p\n", pool);
904 lov_pool_del(obd, pool->pool_name);
906 cfs_hash_putref(lov->lov_pools_hash_body);
907 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
908 lov_ost_pool_free(&lov->lov_packed);
913 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
914 if (!lov->lov_tgts[i])
917 /* Inactive targets may never have connected */
918 if (lov->lov_tgts[i]->ltd_active ||
919 cfs_atomic_read(&lov->lov_refcount))
920 /* We should never get here - these
921 should have been removed in the
923 CERROR("lov tgt %d not cleaned!"
924 " deathrow=%d, lovrc=%d\n",
925 i, lov->lov_death_row,
926 cfs_atomic_read(&lov->lov_refcount));
927 lov_del_target(obd, i, 0, 0);
930 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
932 lov->lov_tgt_size = 0;
934 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
938 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
939 __u32 *indexp, int *genp)
941 struct obd_uuid obd_uuid;
946 switch(cmd = lcfg->lcfg_command) {
947 case LCFG_LOV_ADD_OBD:
948 case LCFG_LOV_ADD_INA:
949 case LCFG_LOV_DEL_OBD: {
952 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
953 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
954 GOTO(out, rc = -EINVAL);
956 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
958 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
959 GOTO(out, rc = -EINVAL);
960 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
961 GOTO(out, rc = -EINVAL);
964 if (cmd == LCFG_LOV_ADD_OBD)
965 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
966 else if (cmd == LCFG_LOV_ADD_INA)
967 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
969 rc = lov_del_target(obd, index, &obd_uuid, gen);
973 struct lprocfs_static_vars lvars = { 0 };
974 struct lov_desc *desc = &(obd->u.lov.desc);
977 GOTO(out, rc = -EINVAL);
979 lprocfs_lov_init_vars(&lvars);
981 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
994 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
995 GOTO(out, rc = -EINVAL);
1004 #define log2(n) cfs_ffz(~(n))
1007 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1008 struct lov_stripe_md **ea,
1009 struct obd_trans_info *oti)
1011 struct lov_obd *lov;
1012 struct obdo *tmp_oa;
1013 struct obd_uuid *ost_uuid = NULL;
1017 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1018 src_oa->o_flags == OBD_FL_DELORPHAN);
1020 lov = &export->exp_obd->u.lov;
1026 if (oti->oti_ost_uuid) {
1027 ost_uuid = oti->oti_ost_uuid;
1028 CDEBUG(D_HA, "clearing orphans only for %s\n",
1032 obd_getref(export->exp_obd);
1033 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1034 struct lov_stripe_md obj_md;
1035 struct lov_stripe_md *obj_mdp = &obj_md;
1036 struct lov_tgt_desc *tgt;
1039 tgt = lov->lov_tgts[i];
1043 /* if called for a specific target, we don't
1044 care if it is not active. */
1045 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1046 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1050 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1053 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1054 obd_uuid2str(ost_uuid));
1056 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1058 LASSERT(lov->lov_tgts[i]->ltd_exp);
1059 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1060 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1061 tmp_oa, &obj_mdp, oti);
1063 /* This export will be disabled until it is recovered,
1064 and then orphan recovery will be completed. */
1065 CERROR("error in orphan recovery on OST idx %d/%d: "
1066 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1073 obd_putref(export->exp_obd);
1079 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1080 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1082 struct lov_stripe_md *obj_mdp, *lsm;
1083 struct lov_obd *lov = &exp->exp_obd->u.lov;
1088 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1089 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1091 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1092 if (obj_mdp == NULL)
1095 ost_idx = src_oa->o_nlink;
1098 GOTO(out, rc = -EINVAL);
1099 if (ost_idx >= lov->desc.ld_tgt_count ||
1100 !lov->lov_tgts[ost_idx])
1101 GOTO(out, rc = -EINVAL);
1103 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1104 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1105 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1106 GOTO(out, rc = -EINVAL);
1110 if (i == lsm->lsm_stripe_count)
1111 GOTO(out, rc = -EINVAL);
1113 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1115 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1119 /* the LOV expects oa->o_id to be set to the LOV object id */
1120 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1121 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1123 struct lov_obd *lov;
1124 struct obd_info oinfo;
1125 struct lov_request_set *set = NULL;
1126 struct lov_request *req;
1127 struct l_wait_info lwi = { 0 };
1131 LASSERT(ea != NULL);
1135 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1136 src_oa->o_flags == OBD_FL_DELORPHAN) {
1137 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1141 lov = &exp->exp_obd->u.lov;
1142 if (!lov->desc.ld_active_tgt_count)
1145 obd_getref(exp->exp_obd);
1146 /* Recreate a specific object id at the given OST index */
1147 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1148 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1149 rc = lov_recreate(exp, src_oa, ea, oti);
1153 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1154 * later in alloc_qos(), we will wait for those rpcs to complete if
1155 * the osfs age is older than 2 * qos_maxage */
1156 qos_statfs_update(exp->exp_obd,
1157 cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1158 OBD_STATFS_CACHE_SECONDS),
1161 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1165 cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1166 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1167 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1168 &req->rq_oi, &req->rq_oi.oi_md, oti);
1171 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1172 * longer then this */
1173 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1175 /* we not have ptlrpc set for assign set->interpret and should
1176 * be call interpret function himself. calling from cb_create_update
1177 * not permited because lov_fini_create_set can sleep for long time,
1178 * but we must avoid sleeping in ptlrpcd interpret function. */
1179 rc = lov_fini_create_set(set, ea);
1181 obd_putref(exp->exp_obd);
1185 #define ASSERT_LSM_MAGIC(lsmp) \
1187 LASSERT((lsmp) != NULL); \
1188 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1189 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1190 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1193 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1194 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1195 struct obd_export *md_exp, void *capa)
1197 struct lov_request_set *set;
1198 struct obd_info oinfo;
1199 struct lov_request *req;
1201 struct lov_obd *lov;
1202 int rc = 0, err = 0;
1205 ASSERT_LSM_MAGIC(lsm);
1207 if (!exp || !exp->exp_obd)
1210 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1212 LASSERT(oti->oti_logcookies);
1215 lov = &exp->exp_obd->u.lov;
1216 obd_getref(exp->exp_obd);
1217 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1221 cfs_list_for_each (pos, &set->set_list) {
1222 req = cfs_list_entry(pos, struct lov_request, rq_link);
1224 if (oa->o_valid & OBD_MD_FLCOOKIE)
1225 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1227 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1228 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1229 err = lov_update_common_set(set, req, err);
1231 CERROR("error: destroying objid "LPX64" subobj "
1232 LPX64" on OST idx %d: rc = %d\n",
1233 oa->o_id, req->rq_oi.oi_oa->o_id,
1241 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1242 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1244 err = lov_fini_destroy_set(set);
1246 obd_putref(exp->exp_obd);
1247 RETURN(rc ? rc : err);
1250 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1252 struct lov_request_set *set;
1253 struct lov_request *req;
1255 struct lov_obd *lov;
1256 int err = 0, rc = 0;
1260 ASSERT_LSM_MAGIC(oinfo->oi_md);
1262 if (!exp || !exp->exp_obd)
1265 lov = &exp->exp_obd->u.lov;
1267 rc = lov_prep_getattr_set(exp, oinfo, &set);
1271 cfs_list_for_each (pos, &set->set_list) {
1272 req = cfs_list_entry(pos, struct lov_request, rq_link);
1274 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1275 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1276 req->rq_oi.oi_oa->o_id, req->rq_idx);
1278 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1280 err = lov_update_common_set(set, req, rc);
1282 CERROR("error: getattr objid "LPX64" subobj "
1283 LPX64" on OST idx %d: rc = %d\n",
1284 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1290 rc = lov_fini_getattr_set(set);
1296 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1299 struct lov_request_set *lovset = (struct lov_request_set *)data;
1303 /* don't do attribute merge if this aysnc op failed */
1305 cfs_atomic_set(&lovset->set_completes, 0);
1306 err = lov_fini_getattr_set(lovset);
1307 RETURN(rc ? rc : err);
1310 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1311 struct ptlrpc_request_set *rqset)
1313 struct lov_request_set *lovset;
1314 struct lov_obd *lov;
1316 struct lov_request *req;
1321 ASSERT_LSM_MAGIC(oinfo->oi_md);
1323 if (!exp || !exp->exp_obd)
1326 lov = &exp->exp_obd->u.lov;
1328 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1332 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1333 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1334 oinfo->oi_md->lsm_stripe_size);
1336 cfs_list_for_each (pos, &lovset->set_list) {
1337 req = cfs_list_entry(pos, struct lov_request, rq_link);
1339 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1340 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1341 req->rq_oi.oi_oa->o_id, req->rq_idx);
1342 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1343 &req->rq_oi, rqset);
1345 CERROR("error: getattr objid "LPX64" subobj "
1346 LPX64" on OST idx %d: rc = %d\n",
1347 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1353 if (!cfs_list_empty(&rqset->set_requests)) {
1355 LASSERT (rqset->set_interpret == NULL);
1356 rqset->set_interpret = lov_getattr_interpret;
1357 rqset->set_arg = (void *)lovset;
1362 cfs_atomic_set(&lovset->set_completes, 0);
1363 err = lov_fini_getattr_set(lovset);
1364 RETURN(rc ? rc : err);
1367 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1368 struct obd_trans_info *oti)
1370 struct lov_request_set *set;
1371 struct lov_obd *lov;
1373 struct lov_request *req;
1374 int err = 0, rc = 0;
1378 ASSERT_LSM_MAGIC(oinfo->oi_md);
1380 if (!exp || !exp->exp_obd)
1383 /* for now, we only expect the following updates here */
1384 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1385 OBD_MD_FLMODE | OBD_MD_FLATIME |
1386 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1387 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1388 OBD_MD_FLGROUP | OBD_MD_FLUID |
1389 OBD_MD_FLGID | OBD_MD_FLFID |
1391 lov = &exp->exp_obd->u.lov;
1392 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1396 cfs_list_for_each (pos, &set->set_list) {
1397 req = cfs_list_entry(pos, struct lov_request, rq_link);
1399 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1401 err = lov_update_setattr_set(set, req, rc);
1403 CERROR("error: setattr objid "LPX64" subobj "
1404 LPX64" on OST idx %d: rc = %d\n",
1405 set->set_oi->oi_oa->o_id,
1406 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1411 err = lov_fini_setattr_set(set);
1417 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1420 struct lov_request_set *lovset = (struct lov_request_set *)data;
1425 cfs_atomic_set(&lovset->set_completes, 0);
1426 err = lov_fini_setattr_set(lovset);
1427 RETURN(rc ? rc : err);
1430 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1431 needed. Otherwise, a client is waiting for responses. */
1432 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1433 struct obd_trans_info *oti,
1434 struct ptlrpc_request_set *rqset)
1436 struct lov_request_set *set;
1437 struct lov_request *req;
1439 struct lov_obd *lov;
1444 ASSERT_LSM_MAGIC(oinfo->oi_md);
1445 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1447 LASSERT(oti->oti_logcookies);
1450 if (!exp || !exp->exp_obd)
1453 lov = &exp->exp_obd->u.lov;
1454 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1458 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1459 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1460 oinfo->oi_md->lsm_stripe_size);
1462 cfs_list_for_each (pos, &set->set_list) {
1463 req = cfs_list_entry(pos, struct lov_request, rq_link);
1465 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1466 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1468 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1469 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1470 req->rq_oi.oi_oa->o_id, req->rq_idx);
1472 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1473 &req->rq_oi, oti, rqset);
1475 CERROR("error: setattr objid "LPX64" subobj "
1476 LPX64" on OST idx %d: rc = %d\n",
1477 set->set_oi->oi_oa->o_id,
1478 req->rq_oi.oi_oa->o_id,
1484 /* If we are not waiting for responses on async requests, return. */
1485 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1488 cfs_atomic_set(&set->set_completes, 0);
1489 err = lov_fini_setattr_set(set);
1490 RETURN(rc ? rc : err);
1493 LASSERT(rqset->set_interpret == NULL);
1494 rqset->set_interpret = lov_setattr_interpret;
1495 rqset->set_arg = (void *)set;
1500 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1503 struct lov_request_set *lovset = (struct lov_request_set *)data;
1508 cfs_atomic_set(&lovset->set_completes, 0);
1509 err = lov_fini_punch_set(lovset);
1510 RETURN(rc ? rc : err);
1513 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1514 * we can send this 'punch' to just the authoritative node and the nodes
1515 * that the punch will affect. */
1516 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1517 struct obd_trans_info *oti,
1518 struct ptlrpc_request_set *rqset)
1520 struct lov_request_set *set;
1521 struct lov_obd *lov;
1523 struct lov_request *req;
1528 ASSERT_LSM_MAGIC(oinfo->oi_md);
1530 if (!exp || !exp->exp_obd)
1533 lov = &exp->exp_obd->u.lov;
1534 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1538 cfs_list_for_each (pos, &set->set_list) {
1539 req = cfs_list_entry(pos, struct lov_request, rq_link);
1541 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1542 &req->rq_oi, NULL, rqset);
1544 CERROR("error: punch objid "LPX64" subobj "LPX64
1545 " on OST idx %d: rc = %d\n",
1546 set->set_oi->oi_oa->o_id,
1547 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1552 if (rc || cfs_list_empty(&rqset->set_requests)) {
1554 err = lov_fini_punch_set(set);
1555 RETURN(rc ? rc : err);
1558 LASSERT(rqset->set_interpret == NULL);
1559 rqset->set_interpret = lov_punch_interpret;
1560 rqset->set_arg = (void *)set;
1565 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1568 struct lov_request_set *lovset = data;
1573 cfs_atomic_set(&lovset->set_completes, 0);
1574 err = lov_fini_sync_set(lovset);
1578 static int lov_sync(struct obd_export *exp, struct obd_info *oinfo,
1579 obd_off start, obd_off end,
1580 struct ptlrpc_request_set *rqset)
1582 struct lov_request_set *set = NULL;
1583 struct lov_obd *lov;
1585 struct lov_request *req;
1589 ASSERT_LSM_MAGIC(oinfo->oi_md);
1590 LASSERT(rqset != NULL);
1595 lov = &exp->exp_obd->u.lov;
1596 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1600 CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1601 set->set_oi->oi_oa->o_id, start, end);
1603 cfs_list_for_each (pos, &set->set_list) {
1604 req = cfs_list_entry(pos, struct lov_request, rq_link);
1606 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi,
1607 req->rq_oi.oi_policy.l_extent.start,
1608 req->rq_oi.oi_policy.l_extent.end, rqset);
1610 CERROR("error: fsync objid "LPX64" subobj "LPX64
1611 " on OST idx %d: rc = %d\n",
1612 set->set_oi->oi_oa->o_id,
1613 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1618 /* If we are not waiting for responses on async requests, return. */
1619 if (rc || cfs_list_empty(&rqset->set_requests)) {
1620 int err = lov_fini_sync_set(set);
1625 LASSERT(rqset->set_interpret == NULL);
1626 rqset->set_interpret = lov_sync_interpret;
1627 rqset->set_arg = (void *)set;
1632 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1633 obd_count oa_bufs, struct brw_page *pga)
1635 struct obd_info oinfo = { { { 0 } } };
1638 oinfo.oi_oa = lov_oinfo->oi_oa;
1640 /* The caller just wants to know if there's a chance that this
1641 * I/O can succeed */
1642 for (i = 0; i < oa_bufs; i++) {
1643 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1644 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1647 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1648 pga[i].off + pga[i].count - 1,
1652 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1653 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1657 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1665 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1666 obd_count oa_bufs, struct brw_page *pga,
1667 struct obd_trans_info *oti)
1669 struct lov_request_set *set;
1670 struct lov_request *req;
1672 struct lov_obd *lov = &exp->exp_obd->u.lov;
1676 ASSERT_LSM_MAGIC(oinfo->oi_md);
1678 if (cmd == OBD_BRW_CHECK) {
1679 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1683 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1687 cfs_list_for_each (pos, &set->set_list) {
1688 struct obd_export *sub_exp;
1689 struct brw_page *sub_pga;
1690 req = cfs_list_entry(pos, struct lov_request, rq_link);
1692 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1693 sub_pga = set->set_pga + req->rq_pgaidx;
1694 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1698 lov_update_common_set(set, req, rc);
1701 err = lov_fini_brw_set(set);
1707 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1710 struct lov_request_set *lovset = (struct lov_request_set *)data;
1712 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1716 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1717 struct ldlm_enqueue_info *einfo,
1718 struct ptlrpc_request_set *rqset)
1720 ldlm_mode_t mode = einfo->ei_mode;
1721 struct lov_request_set *set;
1722 struct lov_request *req;
1724 struct lov_obd *lov;
1729 ASSERT_LSM_MAGIC(oinfo->oi_md);
1730 LASSERT(mode == (mode & -mode));
1732 /* we should never be asked to replay a lock this way. */
1733 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1735 if (!exp || !exp->exp_obd)
1738 lov = &exp->exp_obd->u.lov;
1739 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1743 cfs_list_for_each (pos, &set->set_list) {
1744 req = cfs_list_entry(pos, struct lov_request, rq_link);
1746 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1747 &req->rq_oi, einfo, rqset);
1752 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1754 LASSERT(rqset->set_interpret == NULL);
1755 rqset->set_interpret = lov_enqueue_interpret;
1756 rqset->set_arg = (void *)set;
1760 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1764 static int lov_change_cbdata(struct obd_export *exp,
1765 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1768 struct lov_obd *lov;
1772 ASSERT_LSM_MAGIC(lsm);
1774 if (!exp || !exp->exp_obd)
1777 lov = &exp->exp_obd->u.lov;
1778 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1779 struct lov_stripe_md submd;
1780 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1782 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1783 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1787 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1788 submd.lsm_object_id = loi->loi_id;
1789 submd.lsm_object_seq = loi->loi_seq;
1790 submd.lsm_stripe_count = 0;
1791 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1797 /* find any ldlm lock of the inode in lov
1801 static int lov_find_cbdata(struct obd_export *exp,
1802 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1805 struct lov_obd *lov;
1809 ASSERT_LSM_MAGIC(lsm);
1811 if (!exp || !exp->exp_obd)
1814 lov = &exp->exp_obd->u.lov;
1815 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1816 struct lov_stripe_md submd;
1817 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1819 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1820 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1824 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1825 submd.lsm_object_id = loi->loi_id;
1826 submd.lsm_object_seq = loi->loi_seq;
1827 submd.lsm_stripe_count = 0;
1828 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1836 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1837 __u32 mode, struct lustre_handle *lockh)
1839 struct lov_request_set *set;
1840 struct obd_info oinfo;
1841 struct lov_request *req;
1843 struct lov_obd *lov;
1844 struct lustre_handle *lov_lockhp;
1845 int err = 0, rc = 0;
1848 ASSERT_LSM_MAGIC(lsm);
1850 if (!exp || !exp->exp_obd)
1853 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1855 lov = &exp->exp_obd->u.lov;
1856 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1860 cfs_list_for_each (pos, &set->set_list) {
1861 req = cfs_list_entry(pos, struct lov_request, rq_link);
1862 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1864 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1865 req->rq_oi.oi_md, mode, lov_lockhp);
1866 rc = lov_update_common_set(set, req, rc);
1868 CERROR("error: cancel objid "LPX64" subobj "
1869 LPX64" on OST idx %d: rc = %d\n",
1871 req->rq_oi.oi_md->lsm_object_id,
1877 lov_fini_cancel_set(set);
1881 static int lov_cancel_unused(struct obd_export *exp,
1882 struct lov_stripe_md *lsm,
1883 ldlm_cancel_flags_t flags, void *opaque)
1885 struct lov_obd *lov;
1889 if (!exp || !exp->exp_obd)
1892 lov = &exp->exp_obd->u.lov;
1894 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1896 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1899 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1907 ASSERT_LSM_MAGIC(lsm);
1909 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1910 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1911 struct lov_stripe_md submd;
1912 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1915 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1916 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1920 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1921 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1923 submd.lsm_object_id = loi->loi_id;
1924 submd.lsm_object_seq = loi->loi_seq;
1925 submd.lsm_stripe_count = 0;
1926 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1927 &submd, flags, opaque);
1928 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1929 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1930 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1931 loi->loi_id, loi->loi_ost_idx, err);
1939 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1941 struct lov_request_set *lovset = (struct lov_request_set *)data;
1946 cfs_atomic_set(&lovset->set_completes, 0);
1948 err = lov_fini_statfs_set(lovset);
1949 RETURN(rc ? rc : err);
1952 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1953 __u64 max_age, struct ptlrpc_request_set *rqset)
1955 struct lov_request_set *set;
1956 struct lov_request *req;
1958 struct lov_obd *lov;
1962 LASSERT(oinfo != NULL);
1963 LASSERT(oinfo->oi_osfs != NULL);
1966 rc = lov_prep_statfs_set(obd, oinfo, &set);
1970 cfs_list_for_each (pos, &set->set_list) {
1971 struct obd_device *osc_obd;
1973 req = cfs_list_entry(pos, struct lov_request, rq_link);
1975 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1976 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1981 if (rc || cfs_list_empty(&rqset->set_requests)) {
1984 cfs_atomic_set(&set->set_completes, 0);
1985 err = lov_fini_statfs_set(set);
1986 RETURN(rc ? rc : err);
1989 LASSERT(rqset->set_interpret == NULL);
1990 rqset->set_interpret = lov_statfs_interpret;
1991 rqset->set_arg = (void *)set;
1995 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1996 __u64 max_age, __u32 flags)
1998 struct ptlrpc_request_set *set = NULL;
1999 struct obd_info oinfo = { { { 0 } } };
2004 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2005 * statfs requests */
2006 set = ptlrpc_prep_set();
2010 oinfo.oi_osfs = osfs;
2011 oinfo.oi_flags = flags;
2012 rc = lov_statfs_async(obd, &oinfo, max_age, set);
2014 rc = ptlrpc_set_wait(set);
2015 ptlrpc_set_destroy(set);
2020 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2021 void *karg, void *uarg)
2023 struct obd_device *obddev = class_exp2obd(exp);
2024 struct lov_obd *lov = &obddev->u.lov;
2025 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
2026 struct obd_uuid *uuidp;
2030 case IOC_OBD_STATFS: {
2031 struct obd_ioctl_data *data = karg;
2032 struct obd_device *osc_obd;
2033 struct obd_statfs stat_buf = {0};
2036 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2037 if ((index >= count))
2040 if (!lov->lov_tgts[index])
2041 /* Try again with the next index */
2043 if (!lov->lov_tgts[index]->ltd_active)
2046 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2051 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2052 min((int) data->ioc_plen2,
2053 (int) sizeof(struct obd_uuid))))
2056 /* got statfs data */
2057 rc = obd_statfs(osc_obd, &stat_buf,
2058 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2062 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2063 min((int) data->ioc_plen1,
2064 (int) sizeof(stat_buf))))
2068 case OBD_IOC_LOV_GET_CONFIG: {
2069 struct obd_ioctl_data *data;
2070 struct lov_desc *desc;
2075 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2078 data = (struct obd_ioctl_data *)buf;
2080 if (sizeof(*desc) > data->ioc_inllen1) {
2081 obd_ioctl_freedata(buf, len);
2085 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2086 obd_ioctl_freedata(buf, len);
2090 if (sizeof(__u32) * count > data->ioc_inllen3) {
2091 obd_ioctl_freedata(buf, len);
2095 desc = (struct lov_desc *)data->ioc_inlbuf1;
2096 memcpy(desc, &(lov->desc), sizeof(*desc));
2098 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2099 genp = (__u32 *)data->ioc_inlbuf3;
2100 /* the uuid will be empty for deleted OSTs */
2101 for (i = 0; i < count; i++, uuidp++, genp++) {
2102 if (!lov->lov_tgts[i])
2104 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2105 *genp = lov->lov_tgts[i]->ltd_gen;
2108 if (cfs_copy_to_user((void *)uarg, buf, len))
2110 obd_ioctl_freedata(buf, len);
2113 case LL_IOC_LOV_SETSTRIPE:
2114 rc = lov_setstripe(exp, len, karg, uarg);
2116 case LL_IOC_LOV_GETSTRIPE:
2117 rc = lov_getstripe(exp, karg, uarg);
2119 case LL_IOC_LOV_SETEA:
2120 rc = lov_setea(exp, karg, uarg);
2122 case OBD_IOC_QUOTACTL: {
2123 struct if_quotactl *qctl = karg;
2124 struct lov_tgt_desc *tgt = NULL;
2125 struct obd_quotactl *oqctl;
2127 if (qctl->qc_valid == QC_OSTIDX) {
2128 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2131 tgt = lov->lov_tgts[qctl->qc_idx];
2132 if (!tgt || !tgt->ltd_exp)
2134 } else if (qctl->qc_valid == QC_UUID) {
2135 for (i = 0; i < count; i++) {
2136 tgt = lov->lov_tgts[i];
2138 !obd_uuid_equals(&tgt->ltd_uuid,
2142 if (tgt->ltd_exp == NULL)
2154 LASSERT(tgt && tgt->ltd_exp);
2155 OBD_ALLOC_PTR(oqctl);
2159 QCTL_COPY(oqctl, qctl);
2160 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2162 QCTL_COPY(qctl, oqctl);
2163 qctl->qc_valid = QC_OSTIDX;
2164 qctl->obd_uuid = tgt->ltd_uuid;
2166 OBD_FREE_PTR(oqctl);
2175 for (i = 0; i < count; i++) {
2177 struct obd_device *osc_obd;
2179 /* OST was disconnected */
2180 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2183 /* ll_umount_begin() sets force flag but for lov, not
2184 * osc. Let's pass it through */
2185 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2186 osc_obd->obd_force = obddev->obd_force;
2187 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2189 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2192 if (lov->lov_tgts[i]->ltd_active) {
2193 CDEBUG(err == -ENOTTY ?
2194 D_IOCTL : D_WARNING,
2195 "iocontrol OSC %s on OST "
2196 "idx %d cmd %x: err = %d\n",
2197 lov_uuid2str(lov, i),
2214 #define FIEMAP_BUFFER_SIZE 4096
2217 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2218 * call. The local end offset and the device are sent in the first
2219 * fm_extent. This function calculates the stripe number from the index.
2220 * This function returns a stripe_no on which mapping is to be restarted.
2222 * This function returns fm_end_offset which is the in-OST offset at which
2223 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2224 * will re-calculate proper offset in next stripe.
2225 * Note that the first extent is passed to lov_get_info via the value field.
2227 * \param fiemap fiemap request header
2228 * \param lsm striping information for the file
2229 * \param fm_start logical start of mapping
2230 * \param fm_end logical end of mapping
2231 * \param start_stripe starting stripe will be returned in this
2233 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2234 struct lov_stripe_md *lsm, obd_size fm_start,
2235 obd_size fm_end, int *start_stripe)
2237 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2238 obd_off lun_start, lun_end;
2239 obd_size fm_end_offset;
2240 int stripe_no = -1, i;
2242 if (fiemap->fm_extent_count == 0 ||
2243 fiemap->fm_extents[0].fe_logical == 0)
2246 /* Find out stripe_no from ost_index saved in the fe_device */
2247 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2248 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2249 fiemap->fm_extents[0].fe_device) {
2255 /* If we have finished mapping on previous device, shift logical
2256 * offset to start of next device */
2257 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2258 &lun_start, &lun_end)) != 0 &&
2259 local_end < lun_end) {
2260 fm_end_offset = local_end;
2261 *start_stripe = stripe_no;
2263 /* This is a special value to indicate that caller should
2264 * calculate offset in next stripe. */
2266 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2269 return fm_end_offset;
2273 * We calculate on which OST the mapping will end. If the length of mapping
2274 * is greater than (stripe_size * stripe_count) then the last_stripe will
2275 * will be one just before start_stripe. Else we check if the mapping
2276 * intersects each OST and find last_stripe.
2277 * This function returns the last_stripe and also sets the stripe_count
2278 * over which the mapping is spread
2280 * \param lsm striping information for the file
2281 * \param fm_start logical start of mapping
2282 * \param fm_end logical end of mapping
2283 * \param start_stripe starting stripe of the mapping
2284 * \param stripe_count the number of stripes across which to map is returned
2286 * \retval last_stripe return the last stripe of the mapping
2288 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2289 obd_size fm_end, int start_stripe,
2293 obd_off obd_start, obd_end;
2296 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2297 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2299 *stripe_count = lsm->lsm_stripe_count;
2301 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2302 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2303 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2304 &obd_start, &obd_end)) == 0)
2308 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2315 * Set fe_device and copy extents from local buffer into main return buffer.
2317 * \param fiemap fiemap request header
2318 * \param lcl_fm_ext array of local fiemap extents to be copied
2319 * \param ost_index OST index to be written into the fm_device field for each
2321 * \param ext_count number of extents to be copied
2322 * \param current_extent where to start copying in main extent array
2324 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2325 struct ll_fiemap_extent *lcl_fm_ext,
2326 int ost_index, unsigned int ext_count,
2332 for (ext = 0; ext < ext_count; ext++) {
2333 lcl_fm_ext[ext].fe_device = ost_index;
2334 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2337 /* Copy fm_extent's from fm_local to return buffer */
2338 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2339 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2343 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2344 * This also handles the restarting of FIEMAP calls in case mapping overflows
2345 * the available number of extents in single call.
2347 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2348 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2350 struct ll_fiemap_info_key *fm_key = key;
2351 struct ll_user_fiemap *fiemap = val;
2352 struct ll_user_fiemap *fm_local = NULL;
2353 struct ll_fiemap_extent *lcl_fm_ext;
2355 unsigned int get_num_extents = 0;
2356 int ost_index = 0, actual_start_stripe, start_stripe;
2357 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2359 int current_extent = 0, rc = 0, i;
2360 int ost_eof = 0; /* EOF for object */
2361 int ost_done = 0; /* done with required mapping for this OST? */
2363 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2364 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2369 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2370 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2372 OBD_ALLOC_LARGE(fm_local, buffer_size);
2373 if (fm_local == NULL)
2374 GOTO(out, rc = -ENOMEM);
2375 lcl_fm_ext = &fm_local->fm_extents[0];
2377 count_local = fiemap_size_to_count(buffer_size);
2379 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2380 fm_start = fiemap->fm_start;
2381 fm_length = fiemap->fm_length;
2382 /* Calculate start stripe, last stripe and length of mapping */
2383 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2384 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2385 fm_start + fm_length - 1);
2386 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2387 if (fm_end > fm_key->oa.o_size)
2388 fm_end = fm_key->oa.o_size;
2390 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2391 actual_start_stripe, &stripe_count);
2393 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2396 if (fiemap->fm_extent_count == 0) {
2397 get_num_extents = 1;
2401 /* Check each stripe */
2402 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2403 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2404 obd_size req_fm_len; /* Stores length of required mapping */
2405 obd_size len_mapped_single_call;
2406 obd_off lun_start, lun_end, obd_object_end;
2407 unsigned int ext_count;
2409 cur_stripe_wrap = cur_stripe;
2411 /* Find out range of mapping on this stripe */
2412 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2413 &lun_start, &obd_object_end)) == 0)
2416 /* If this is a continuation FIEMAP call and we are on
2417 * starting stripe then lun_start needs to be set to
2419 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2420 lun_start = fm_end_offset;
2422 if (fm_length != ~0ULL) {
2423 /* Handle fm_start + fm_length overflow */
2424 if (fm_start + fm_length < fm_start)
2425 fm_length = ~0ULL - fm_start;
2426 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2432 if (lun_start == lun_end)
2435 req_fm_len = obd_object_end - lun_start;
2436 fm_local->fm_length = 0;
2437 len_mapped_single_call = 0;
2439 /* If the output buffer is very large and the objects have many
2440 * extents we may need to loop on a single OST repeatedly */
2444 if (get_num_extents == 0) {
2445 /* Don't get too many extents. */
2446 if (current_extent + count_local >
2447 fiemap->fm_extent_count)
2448 count_local = fiemap->fm_extent_count -
2452 lun_start += len_mapped_single_call;
2453 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2454 req_fm_len = fm_local->fm_length;
2455 fm_local->fm_extent_count = count_local;
2456 fm_local->fm_mapped_extents = 0;
2457 fm_local->fm_flags = fiemap->fm_flags;
2459 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2460 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2461 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2463 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2464 GOTO(out, rc = -EINVAL);
2466 /* If OST is inactive, return extent with UNKNOWN flag */
2467 if (!lov->lov_tgts[ost_index]->ltd_active) {
2468 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2469 fm_local->fm_mapped_extents = 1;
2471 lcl_fm_ext[0].fe_logical = lun_start;
2472 lcl_fm_ext[0].fe_length = obd_object_end -
2474 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2479 fm_local->fm_start = lun_start;
2480 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2481 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2482 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2483 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2484 keylen, key, vallen, fm_local, lsm);
2489 ext_count = fm_local->fm_mapped_extents;
2490 if (ext_count == 0) {
2492 /* If last stripe has hole at the end,
2493 * then we need to return */
2494 if (cur_stripe_wrap == last_stripe) {
2495 fiemap->fm_mapped_extents = 0;
2501 /* If we just need num of extents then go to next device */
2502 if (get_num_extents) {
2503 current_extent += ext_count;
2507 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2508 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2510 /* Have we finished mapping on this device? */
2511 if (req_fm_len <= len_mapped_single_call)
2514 /* Clear the EXTENT_LAST flag which can be present on
2516 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2517 lcl_fm_ext[ext_count - 1].fe_flags &=
2518 ~FIEMAP_EXTENT_LAST;
2520 curr_loc = lov_stripe_size(lsm,
2521 lcl_fm_ext[ext_count - 1].fe_logical+
2522 lcl_fm_ext[ext_count - 1].fe_length,
2524 if (curr_loc >= fm_key->oa.o_size)
2527 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2528 ost_index, ext_count,
2531 current_extent += ext_count;
2533 /* Ran out of available extents? */
2534 if (current_extent >= fiemap->fm_extent_count)
2536 } while (ost_done == 0 && ost_eof == 0);
2538 if (cur_stripe_wrap == last_stripe)
2543 /* Indicate that we are returning device offsets unless file just has
2545 if (lsm->lsm_stripe_count > 1)
2546 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2548 if (get_num_extents)
2549 goto skip_last_device_calc;
2551 /* Check if we have reached the last stripe and whether mapping for that
2552 * stripe is done. */
2553 if (cur_stripe_wrap == last_stripe) {
2554 if (ost_done || ost_eof)
2555 fiemap->fm_extents[current_extent - 1].fe_flags |=
2559 skip_last_device_calc:
2560 fiemap->fm_mapped_extents = current_extent;
2563 OBD_FREE_LARGE(fm_local, buffer_size);
2567 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2568 void *key, __u32 *vallen, void *val,
2569 struct lov_stripe_md *lsm)
2571 struct obd_device *obddev = class_exp2obd(exp);
2572 struct lov_obd *lov = &obddev->u.lov;
2576 if (!vallen || !val)
2581 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2584 struct ldlm_lock *lock;
2586 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2587 struct lov_oinfo *loi;
2588 __u32 *stripe = val;
2590 if (*vallen < sizeof(*stripe))
2591 GOTO(out, rc = -EFAULT);
2592 *vallen = sizeof(*stripe);
2594 /* XXX This is another one of those bits that will need to
2595 * change if we ever actually support nested LOVs. It uses
2596 * the lock's export to find out which stripe it is. */
2597 /* XXX - it's assumed all the locks for deleted OSTs have
2598 * been cancelled. Also, the export for deleted OSTs will
2599 * be NULL and won't match the lock's export. */
2600 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2601 loi = lsm->lsm_oinfo[i];
2602 if (!lov->lov_tgts[loi->loi_ost_idx])
2604 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2605 data->lock->l_conn_export &&
2606 osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2611 LDLM_ERROR(data->lock, "lock on inode without such object");
2612 dump_lsm(D_ERROR, lsm);
2613 GOTO(out, rc = -ENXIO);
2614 } else if (KEY_IS(KEY_LAST_ID)) {
2615 struct obd_id_info *info = val;
2616 __u32 size = sizeof(obd_id);
2617 struct lov_tgt_desc *tgt;
2619 LASSERT(*vallen == sizeof(struct obd_id_info));
2620 tgt = lov->lov_tgts[info->idx];
2622 if (!tgt || !tgt->ltd_active)
2623 GOTO(out, rc = -ESRCH);
2625 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2627 } else if (KEY_IS(KEY_LOVDESC)) {
2628 struct lov_desc *desc_ret = val;
2629 *desc_ret = lov->desc;
2632 } else if (KEY_IS(KEY_FIEMAP)) {
2633 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2635 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2636 struct lov_tgt_desc *tgt;
2637 __u64 ost_idx = *((__u64*)val);
2639 LASSERT(*vallen == sizeof(__u64));
2640 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2641 tgt = lov->lov_tgts[ost_idx];
2643 if (!tgt || !tgt->ltd_exp)
2644 GOTO(out, rc = -ESRCH);
2646 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2648 } else if (KEY_IS(KEY_TGT_COUNT)) {
2649 *((int *)val) = lov->desc.ld_tgt_count;
2660 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2661 void *key, obd_count vallen, void *val,
2662 struct ptlrpc_request_set *set)
2664 struct obd_device *obddev = class_exp2obd(exp);
2665 struct lov_obd *lov = &obddev->u.lov;
2668 struct lov_tgt_desc *tgt;
2669 unsigned incr, check_uuid,
2670 do_inactive, no_set;
2671 unsigned next_id = 0, mds_con = 0, capa = 0;
2674 incr = check_uuid = do_inactive = no_set = 0;
2677 set = ptlrpc_prep_set();
2683 count = lov->desc.ld_tgt_count;
2685 if (KEY_IS(KEY_NEXT_ID)) {
2686 count = vallen / sizeof(struct obd_id_info);
2687 vallen = sizeof(obd_id);
2688 incr = sizeof(struct obd_id_info);
2691 } else if (KEY_IS(KEY_CHECKSUM)) {
2693 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2694 /* use defaults: do_inactive = incr = 0; */
2695 } else if (KEY_IS(KEY_MDS_CONN)) {
2697 } else if (KEY_IS(KEY_CAPA_KEY)) {
2701 for (i = 0; i < count; i++, val = (char *)val + incr) {
2703 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2705 tgt = lov->lov_tgts[i];
2707 /* OST was disconnected */
2708 if (!tgt || !tgt->ltd_exp)
2711 /* OST is inactive and we don't want inactive OSCs */
2712 if (!tgt->ltd_active && !do_inactive)
2716 struct mds_group_info *mgi;
2718 LASSERT(vallen == sizeof(*mgi));
2719 mgi = (struct mds_group_info *)val;
2721 /* Only want a specific OSC */
2722 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2726 err = obd_set_info_async(tgt->ltd_exp,
2727 keylen, key, sizeof(int),
2729 } else if (next_id) {
2730 err = obd_set_info_async(tgt->ltd_exp,
2731 keylen, key, vallen,
2732 ((struct obd_id_info*)val)->data, set);
2734 struct mds_capa_info *info = (struct mds_capa_info*)val;
2736 LASSERT(vallen == sizeof(*info));
2738 /* Only want a specific OSC */
2740 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2743 err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2744 sizeof(*info->capa),
2747 /* Only want a specific OSC */
2749 !obd_uuid_equals(val, &tgt->ltd_uuid))
2752 err = obd_set_info_async(tgt->ltd_exp,
2753 keylen, key, vallen, val, set);
2762 err = ptlrpc_set_wait(set);
2765 ptlrpc_set_destroy(set);
2770 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2775 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2776 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2777 if (loi->loi_ar.ar_rc && !rc)
2778 rc = loi->loi_ar.ar_rc;
2779 loi->loi_ar.ar_rc = 0;
2783 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2786 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2787 int cmd, __u64 *offset)
2789 __u32 ssize = lsm->lsm_stripe_size;
2793 lov_do_div64(start, ssize);
2794 start = start * ssize;
2796 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2797 ", end "LPU64"\n", *offset, ssize, start,
2799 if (cmd == OBD_CALC_STRIPE_END) {
2800 *offset = start + ssize - 1;
2801 } else if (cmd == OBD_CALC_STRIPE_START) {
2810 void lov_stripe_lock(struct lov_stripe_md *md)
2812 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2813 cfs_spin_lock(&md->lsm_lock);
2814 LASSERT(md->lsm_lock_owner == 0);
2815 md->lsm_lock_owner = cfs_curproc_pid();
2817 EXPORT_SYMBOL(lov_stripe_lock);
2819 void lov_stripe_unlock(struct lov_stripe_md *md)
2821 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2822 md->lsm_lock_owner = 0;
2823 cfs_spin_unlock(&md->lsm_lock);
2825 EXPORT_SYMBOL(lov_stripe_unlock);
2827 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2828 struct obd_quotactl *oqctl)
2830 struct lov_obd *lov = &obd->u.lov;
2831 struct lov_tgt_desc *tgt;
2833 __u64 bhardlimit = 0;
2837 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2838 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2839 oqctl->qc_cmd != Q_GETOQUOTA &&
2840 oqctl->qc_cmd != Q_INITQUOTA &&
2841 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2842 oqctl->qc_cmd != Q_FINVALIDATE) {
2843 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2849 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2852 tgt = lov->lov_tgts[i];
2857 if (!tgt->ltd_active || tgt->ltd_reap) {
2858 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2859 lov->lov_tgts[i]->ltd_activate) {
2861 CERROR("ost %d is inactive\n", i);
2863 CDEBUG(D_HA, "ost %d is inactive\n", i);
2868 err = obd_quotactl(tgt->ltd_exp, oqctl);
2870 if (tgt->ltd_active && !rc)
2875 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2876 curspace += oqctl->qc_dqblk.dqb_curspace;
2877 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2882 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2883 oqctl->qc_dqblk.dqb_curspace = curspace;
2884 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2889 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2890 struct obd_quotactl *oqctl)
2892 struct lov_obd *lov = &obd->u.lov;
2898 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2899 if (!lov->lov_tgts[i])
2902 /* Skip quota check on the administratively disabled OSTs. */
2903 if (!lov->lov_tgts[i]->ltd_activate) {
2904 CWARN("lov idx %d was administratively disabled, "
2905 "skip quotacheck on it.\n", i);
2909 if (!lov->lov_tgts[i]->ltd_active) {
2910 CERROR("lov idx %d inactive\n", i);
2916 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2919 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2922 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2933 int lov_quota_adjust_qunit(struct obd_export *exp,
2934 struct quota_adjust_qunit *oqaq,
2935 struct lustre_quota_ctxt *qctxt,
2936 struct ptlrpc_request_set *rqset)
2938 struct obd_device *obd = class_exp2obd(exp);
2939 struct lov_obd *lov = &obd->u.lov;
2941 unsigned no_set = 0;
2944 if (!QAQ_IS_ADJBLK(oqaq)) {
2945 CERROR("bad qaq_flags %x for lov obd.\n", oqaq->qaq_flags);
2949 if (rqset == NULL) {
2950 rqset = ptlrpc_prep_set();
2957 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2959 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
2960 CDEBUG(D_HA, "ost %d is inactive\n", i);
2964 err = obd_quota_adjust_qunit(lov->lov_tgts[i]->ltd_exp, oqaq,
2967 if (lov->lov_tgts[i]->ltd_active && !rc)
2973 err = ptlrpc_set_wait(rqset);
2977 /* Destroy the set if none was provided by the caller */
2979 ptlrpc_set_destroy(rqset);
2985 struct obd_ops lov_obd_ops = {
2986 .o_owner = THIS_MODULE,
2987 .o_setup = lov_setup,
2988 .o_precleanup = lov_precleanup,
2989 .o_cleanup = lov_cleanup,
2990 //.o_process_config = lov_process_config,
2991 .o_connect = lov_connect,
2992 .o_disconnect = lov_disconnect,
2993 .o_statfs = lov_statfs,
2994 .o_statfs_async = lov_statfs_async,
2995 .o_packmd = lov_packmd,
2996 .o_unpackmd = lov_unpackmd,
2997 .o_create = lov_create,
2998 .o_destroy = lov_destroy,
2999 .o_getattr = lov_getattr,
3000 .o_getattr_async = lov_getattr_async,
3001 .o_setattr = lov_setattr,
3002 .o_setattr_async = lov_setattr_async,
3004 .o_merge_lvb = lov_merge_lvb,
3005 .o_adjust_kms = lov_adjust_kms,
3006 .o_punch = lov_punch,
3008 .o_enqueue = lov_enqueue,
3009 .o_change_cbdata = lov_change_cbdata,
3010 .o_find_cbdata = lov_find_cbdata,
3011 .o_cancel = lov_cancel,
3012 .o_cancel_unused = lov_cancel_unused,
3013 .o_iocontrol = lov_iocontrol,
3014 .o_get_info = lov_get_info,
3015 .o_set_info_async = lov_set_info_async,
3016 .o_extent_calc = lov_extent_calc,
3017 .o_llog_init = lov_llog_init,
3018 .o_llog_finish = lov_llog_finish,
3019 .o_notify = lov_notify,
3020 .o_pool_new = lov_pool_new,
3021 .o_pool_rem = lov_pool_remove,
3022 .o_pool_add = lov_pool_add,
3023 .o_pool_del = lov_pool_del,
3024 .o_getref = lov_getref,
3025 .o_putref = lov_putref,
3026 .o_quotactl = lov_quotactl,
3027 .o_quotacheck = lov_quotacheck,
3028 .o_quota_adjust_qunit = lov_quota_adjust_qunit,
3031 static quota_interface_t *quota_interface;
3032 extern quota_interface_t lov_quota_interface;
3034 cfs_mem_cache_t *lov_oinfo_slab;
3036 extern struct lu_kmem_descr lov_caches[];
3038 int __init lov_init(void)
3040 struct lprocfs_static_vars lvars = { 0 };
3044 /* print an address of _any_ initialized kernel symbol from this
3045 * module, to allow debugging with gdb that doesn't support data
3046 * symbols from modules.*/
3047 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
3049 rc = lu_kmem_init(lov_caches);
3053 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3054 sizeof(struct lov_oinfo),
3055 0, CFS_SLAB_HWCACHE_ALIGN);
3056 if (lov_oinfo_slab == NULL) {
3057 lu_kmem_fini(lov_caches);
3060 lprocfs_lov_init_vars(&lvars);
3062 cfs_request_module("lquota");
3063 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3064 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3066 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3067 LUSTRE_LOV_NAME, &lov_device_type);
3070 if (quota_interface)
3071 PORTAL_SYMBOL_PUT(lov_quota_interface);
3072 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3074 lu_kmem_fini(lov_caches);
3081 static void /*__exit*/ lov_exit(void)
3085 lu_device_type_fini(&lov_device_type);
3086 lu_kmem_fini(lov_caches);
3088 if (quota_interface)
3089 PORTAL_SYMBOL_PUT(lov_quota_interface);
3091 class_unregister_type(LUSTRE_LOV_NAME);
3092 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3096 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3097 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3098 MODULE_LICENSE("GPL");
3100 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);