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 2008 Sun Microsystems, Inc. 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 <lustre_cache.h>
67 #include <lustre/ll_fiemap.h>
69 #include "lov_internal.h"
72 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
73 Any function that expects lov_tgts to remain stationary must take a ref. */
74 static void lov_getref(struct obd_device *obd)
76 struct lov_obd *lov = &obd->u.lov;
78 /* nobody gets through here until lov_putref is done */
79 mutex_down(&lov->lov_lock);
80 atomic_inc(&lov->lov_refcount);
81 mutex_up(&lov->lov_lock);
85 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
87 static void lov_putref(struct obd_device *obd)
89 struct lov_obd *lov = &obd->u.lov;
91 struct lov_tgt_desc *tgt;
93 mutex_down(&lov->lov_lock);
94 /* ok to dec to 0 more than once -- ltd_exp's will be null */
95 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
97 struct lov_tgt_desc *n;
98 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
100 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
101 tgt = lov->lov_tgts[i];
103 if (!tgt || !tgt->ltd_reap)
105 list_add(&tgt->ltd_kill, &kill);
106 /* XXX - right now there is a dependency on ld_tgt_count
107 * being the maximum tgt index for computing the
108 * mds_max_easize. So we can't shrink it. */
109 lov_ost_pool_remove(&lov->lov_packed, i);
110 lov->lov_tgts[i] = NULL;
111 lov->lov_death_row--;
113 mutex_up(&lov->lov_lock);
115 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
116 list_del(&tgt->ltd_kill);
118 __lov_del_obd(obd, tgt);
121 mutex_up(&lov->lov_lock);
125 static int lov_obd_register_page_removal_cb(struct obd_device *obd,
126 obd_page_removal_cb_t func,
127 obd_pin_extent_cb pin_cb)
129 struct lov_obd *lov = &obd->u.lov;
131 if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
134 if (lov->lov_page_pin_cb && lov->lov_page_pin_cb != pin_cb)
137 lov->lov_page_removal_cb = func;
138 lov->lov_page_pin_cb = pin_cb;
142 static int lov_obd_unregister_page_removal_cb(struct obd_device *obd,
143 obd_page_removal_cb_t func)
145 struct lov_obd *lov = &obd->u.lov;
148 if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
151 lov->lov_page_removal_cb = NULL;
152 lov->lov_page_pin_cb = NULL;
155 for (i = 0; i < lov->desc.ld_tgt_count; ++i) {
156 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
158 rc |= obd_unregister_page_removal_cb(
159 lov->lov_tgts[i]->ltd_exp->exp_obd, func);
166 static int lov_obd_register_lock_cancel_cb(struct obd_device *obd,
167 obd_lock_cancel_cb func)
169 struct lov_obd *lov = &obd->u.lov;
171 if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
174 lov->lov_lock_cancel_cb = func;
179 static int lov_obd_unregister_lock_cancel_cb(struct obd_device *obd,
180 obd_lock_cancel_cb func)
182 struct lov_obd *lov = &obd->u.lov;
184 if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
187 lov->lov_lock_cancel_cb = NULL;
192 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
195 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
196 enum obd_notify_event ev, void *data)
201 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
204 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
205 CERROR("unexpected notification of %s %s!\n",
206 watched->obd_type->typ_name,
211 /* Set OSC as active before notifying the observer, so the
212 * observer can use the OSC normally.
214 rc = lov_set_osc_active(obd, &watched->u.cli.cl_target_uuid,
215 ev == OBD_NOTIFY_ACTIVE);
217 CERROR("%sactivation of %s failed: %d\n",
218 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
219 obd_uuid2str(&watched->u.cli.cl_target_uuid),
223 /* active event should be pass lov target index as data */
227 /* Pass the notification up the chain. */
229 rc = obd_notify_observer(obd, watched, ev, data);
231 /* NULL watched means all osc's in the lov (only for syncs) */
232 /* sync event should be send lov idx as data */
233 struct lov_obd *lov = &obd->u.lov;
237 is_sync = (ev == OBD_NOTIFY_SYNC) ||
238 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
241 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
243 if (!lov->lov_tgts[i])
245 /* don't send sync event if target not
246 * connected/activated */
247 if (is_sync && !lov->lov_tgts[i]->ltd_active)
250 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
253 CERROR("%s: notify %s of %s failed %d\n",
255 obd->obd_observer->obd_name,
256 lov->lov_tgts[i]->ltd_obd->obd_name, rc);
265 #define MAX_STRING_SIZE 128
266 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
267 struct obd_connect_data *data)
269 struct lov_obd *lov = &obd->u.lov;
270 struct obd_uuid tgt_uuid;
271 struct obd_device *tgt_obd;
272 struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
273 struct lustre_handle conn = {0, };
274 struct obd_import *imp;
276 cfs_proc_dir_entry_t *lov_proc_dir;
281 if (!lov->lov_tgts[index])
284 tgt_obd = lov->lov_tgts[index]->ltd_obd;
285 if (!tgt_obd->obd_set_up) {
286 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
290 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
291 data->ocd_index = index;
294 * Divine LOV knows that OBDs under it are OSCs.
296 imp = tgt_obd->u.cli.cl_import;
299 tgt_obd->obd_no_recov = 0;
300 /* FIXME this is probably supposed to be
301 ptlrpc_set_import_active. Horrible naming. */
302 ptlrpc_activate_import(imp);
305 rc = obd_register_observer(tgt_obd, obd);
307 CERROR("Target %s register_observer error %d; "
308 "will not be able to reactivate\n",
309 obd_uuid2str(&tgt_uuid), rc);
313 if (imp->imp_invalid) {
314 CERROR("not connecting OSC %s; administratively "
315 "disabled\n", obd_uuid2str(&tgt_uuid));
318 if (lov->lov_lock_cancel_cb)
319 rc = obd_register_lock_cancel_cb(tgt_obd, lov->lov_lock_cancel_cb);
323 if (lov->lov_page_removal_cb)
324 rc = obd_register_page_removal_cb(tgt_obd, lov->lov_page_removal_cb,
325 lov->lov_page_pin_cb);
327 GOTO(out_lock_cb, rc);
329 rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, data, &lov->lov_tgts[index]->ltd_exp);
330 if (rc || !lov->lov_tgts[index]->ltd_exp) {
331 CERROR("Target %s connect error %d\n",
332 obd_uuid2str(&tgt_uuid), rc);
333 GOTO(out_page_cb, rc);
337 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
339 struct obd_device *osc_obd = class_conn2obd(&conn);
340 cfs_proc_dir_entry_t *osc_symlink;
341 char name[MAX_STRING_SIZE];
343 LASSERT(osc_obd != NULL);
344 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
345 LASSERT(osc_obd->obd_type->typ_name != NULL);
346 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
347 osc_obd->obd_type->typ_name,
349 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
351 if (osc_symlink == NULL) {
352 CERROR("could not register LOV target "
353 "/proc/fs/lustre/%s/%s/target_obds/%s.",
354 obd->obd_type->typ_name, obd->obd_name,
356 lprocfs_remove(&lov_proc_dir);
360 rc = qos_add_tgt(obd, index);
362 CERROR("qos_add_tgt failed %d\n", rc);
367 obd_unregister_page_removal_cb(obd, lov->lov_page_removal_cb);
369 obd_unregister_lock_cancel_cb(obd, lov->lov_lock_cancel_cb);
374 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
375 struct obd_uuid *cluuid, struct obd_connect_data *data,
378 struct lov_obd *lov = &obd->u.lov;
379 struct lov_tgt_desc *tgt;
380 struct obd_export **exp = localdata;
384 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
386 rc = class_connect(conn, obd, cluuid);
390 *exp = class_conn2export(conn);
392 /* Why should there ever be more than 1 connect? */
394 LASSERT(lov->lov_connects == 1);
396 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
398 lov->lov_ocd = *data;
401 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
402 tgt = lov->lov_tgts[i];
403 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
405 /* Flags will be lowest common denominator */
406 rc = lov_connect_obd(obd, i, lov->lov_tgts[i]->ltd_activate,
409 CERROR("%s: lov connect tgt %d failed: %d\n",
410 obd->obd_name, i, rc);
413 /* connect to administrative disabled ost */
414 if (!lov->lov_tgts[i]->ltd_exp)
417 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
418 OBD_NOTIFY_CONNECT, (void *)&i);
420 CERROR("%s error sending notify %d\n",
429 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
431 cfs_proc_dir_entry_t *lov_proc_dir;
432 struct lov_obd *lov = &obd->u.lov;
433 struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
438 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
439 obd->obd_name, osc_obd->obd_name);
441 if (tgt->ltd_active) {
443 lov->desc.ld_active_tgt_count--;
444 tgt->ltd_exp->exp_obd->obd_inactive = 1;
447 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
449 cfs_proc_dir_entry_t *osc_symlink;
451 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
453 lprocfs_remove(&osc_symlink);
455 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
456 obd->obd_type->typ_name, obd->obd_name,
461 if (obd->obd_force) {
462 /* Pass it on to our clients.
463 * XXX This should be an argument to disconnect,
464 * XXX not a back-door flag on the OBD. Ah well.
467 osc_obd->obd_force = 1;
470 obd_register_observer(osc_obd, NULL);
472 obd_unregister_page_removal_cb(osc_obd, lov->lov_page_removal_cb);
473 obd_unregister_lock_cancel_cb(osc_obd, lov->lov_lock_cancel_cb);
475 rc = obd_disconnect(tgt->ltd_exp);
477 CERROR("Target %s disconnect error %d\n",
478 tgt->ltd_uuid.uuid, rc);
482 qos_del_tgt(obd, tgt);
489 static int lov_del_target(struct obd_device *obd, __u32 index,
490 struct obd_uuid *uuidp, int gen);
492 static int lov_disconnect(struct obd_export *exp)
494 struct obd_device *obd = class_exp2obd(exp);
495 struct lov_obd *lov = &obd->u.lov;
502 /* Only disconnect the underlying layers on the final disconnect. */
504 if (lov->lov_connects != 0) {
505 /* why should there be more than 1 connect? */
506 CERROR("disconnect #%d\n", lov->lov_connects);
510 /* Let's hold another reference so lov_del_obd doesn't spin through
513 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
514 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
515 /* Disconnection is the last we know about an obd */
516 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
522 rc = class_disconnect(exp); /* bz 9811 */
528 * -EINVAL : UUID can't be found in the LOV's target list
529 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
530 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
531 * - any above 0 is lov index
533 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
536 struct lov_obd *lov = &obd->u.lov;
537 struct lov_tgt_desc *tgt;
541 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
542 lov, uuid->uuid, activate);
545 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
546 tgt = lov->lov_tgts[i];
547 if (!tgt || !tgt->ltd_exp)
550 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
551 i, obd_uuid2str(&tgt->ltd_uuid),
552 tgt->ltd_exp->exp_handle.h_cookie);
553 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
557 if (i == lov->desc.ld_tgt_count)
558 GOTO(out, i = -EINVAL);
560 if (lov->lov_tgts[i]->ltd_active == activate) {
561 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
562 activate ? "" : "in");
566 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
567 activate ? "" : "in");
569 lov->lov_tgts[i]->ltd_active = activate;
572 lov->desc.ld_active_tgt_count++;
573 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
575 lov->desc.ld_active_tgt_count--;
576 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
578 /* remove any old qos penalty */
579 lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
587 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
588 __u32 index, int gen, int active)
590 struct obd_device *tgt_obd;
591 struct lov_obd *lov = &obd->u.lov;
592 struct lov_tgt_desc *tgt;
596 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
597 uuidp->uuid, index, gen, active);
600 CERROR("request to add OBD %s with invalid generation: %d\n",
606 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
609 CERROR("Target %s not attached\n", obd_uuid2str(uuidp));
613 mutex_down(&lov->lov_lock);
615 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
616 tgt = lov->lov_tgts[index];
617 CERROR("UUID %s already assigned at LOV target index %d\n",
618 obd_uuid2str(&tgt->ltd_uuid), index);
619 GOTO(err_unlock, rc = -EEXIST);
622 if (index >= lov->lov_tgt_size) {
623 /* We need to reallocate the lov target array. */
624 struct lov_tgt_desc **newtgts, **old = NULL;
625 __u32 newsize, oldsize = 0;
627 newsize = max(lov->lov_tgt_size, (__u32)2);
628 while (newsize < index + 1)
629 newsize = newsize << 1;
630 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
632 GOTO(err_unlock, rc = -ENOMEM);
634 if (lov->lov_tgt_size) {
635 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
638 oldsize = lov->lov_tgt_size;
641 lov->lov_tgts = newtgts;
642 lov->lov_tgt_size = newsize;
647 OBD_FREE(old, sizeof(*old) * oldsize);
649 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
650 lov->lov_tgts, lov->lov_tgt_size);
655 GOTO(err_unlock, rc = -ENOMEM);
657 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
659 GOTO(err_free_tgt, rc = -EEXIST);
661 memset(tgt, 0, sizeof(*tgt));
662 tgt->ltd_obd = tgt_obd;
663 tgt->ltd_uuid = *uuidp;
664 /* XXX - add a sanity check on the generation number. */
666 tgt->ltd_index = index;
667 tgt->ltd_activate = active;
669 lov->lov_tgts[index] = tgt;
670 if (index >= lov->desc.ld_tgt_count)
671 lov->desc.ld_tgt_count = index + 1;
673 mutex_up(&lov->lov_lock);
675 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
676 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
678 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
680 if (lov->lov_connects == 0) {
681 /* lov_connect hasn't been called yet. We'll do the
682 lov_connect_obd on this target when that fn first runs,
683 because we don't know the connect flags yet. */
689 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
693 /* connect to administrative disabled ost */
697 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
698 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
702 /* connect or notify failed - we can try connect later
703 * instead of complete delete target */
704 CERROR("connect or notify failed (%d) for %s\n", rc,
705 obd_uuid2str(&tgt->ltd_uuid));
713 mutex_up(&lov->lov_lock);
717 /* Schedule a target for deletion */
718 static int lov_del_target(struct obd_device *obd, __u32 index,
719 struct obd_uuid *uuidp, int gen)
721 struct lov_obd *lov = &obd->u.lov;
722 int count = lov->desc.ld_tgt_count;
726 if (index >= count) {
727 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
734 if (!lov->lov_tgts[index]) {
735 CERROR("LOV target at index %d is not setup.\n", index);
736 GOTO(out, rc = -EINVAL);
739 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
740 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
741 lov_uuid2str(lov, index), index,
742 obd_uuid2str(uuidp));
743 GOTO(out, rc = -EINVAL);
746 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
747 lov_uuid2str(lov, index), index,
748 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
749 lov->lov_tgts[index]->ltd_active);
751 lov->lov_tgts[index]->ltd_reap = 1;
752 lov->lov_death_row++;
753 /* we really delete it from obd_putref */
760 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
762 struct obd_device *osc_obd;
765 LASSERT(tgt->ltd_reap);
767 osc_obd = class_exp2obd(tgt->ltd_exp);
769 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
771 osc_obd ? osc_obd->obd_name : "<no obd>");
774 lov_disconnect_obd(obd, tgt);
778 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
779 do it ourselves. And we can't do it from lov_cleanup,
780 because we just lost our only reference to it. */
782 /* Use lov's force/fail flags. */
783 osc_obd->obd_force = obd->obd_force;
784 osc_obd->obd_fail = obd->obd_fail;
785 class_manual_cleanup(osc_obd);
789 void lov_fix_desc_stripe_size(__u64 *val)
791 if (*val < PTLRPC_MAX_BRW_SIZE) {
793 LCONSOLE_WARN("Increasing default stripe size from "
794 LPU64" to %u\n",*val,PTLRPC_MAX_BRW_SIZE);
795 *val = PTLRPC_MAX_BRW_SIZE;
796 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
797 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
798 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
800 *val, LOV_MIN_STRIPE_SIZE);
804 void lov_fix_desc_stripe_count(__u32 *val)
810 void lov_fix_desc_pattern(__u32 *val)
812 /* from lov_setstripe */
813 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
814 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
819 void lov_fix_desc_qos_maxage(__u32 *val)
823 *val = QOS_DEFAULT_MAXAGE;
826 void lov_fix_desc(struct lov_desc *desc)
828 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
829 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
830 lov_fix_desc_pattern(&desc->ld_pattern);
831 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
834 static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
836 struct lprocfs_static_vars lvars = { 0 };
837 struct lustre_cfg *lcfg = buf;
838 struct lov_desc *desc;
839 struct lov_obd *lov = &obd->u.lov;
843 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
844 CERROR("LOV setup requires a descriptor\n");
848 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
850 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
851 CERROR("descriptor size wrong: %d > %d\n",
852 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
856 if (desc->ld_magic != LOV_DESC_MAGIC) {
857 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
858 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
859 obd->obd_name, desc);
860 lustre_swab_lov_desc(desc);
862 CERROR("%s: Bad lov desc magic: %#x\n",
863 obd->obd_name, desc->ld_magic);
870 desc->ld_active_tgt_count = 0;
872 lov->lov_tgt_size = 0;
874 sema_init(&lov->lov_lock, 1);
875 atomic_set(&lov->lov_refcount, 0);
876 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
877 init_rwsem(&lov->lov_qos.lq_rw_sem);
878 lov->lov_qos.lq_dirty = 1;
879 lov->lov_qos.lq_rr.lqr_dirty = 1;
880 lov->lov_qos.lq_reset = 1;
881 /* Default priority is toward free space balance */
882 lov->lov_qos.lq_prio_free = 232;
883 /* Default threshold for rr (roughly 17%) */
884 lov->lov_qos.lq_threshold_rr = 43;
885 /* Init statfs fields */
886 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
887 if (NULL == lov->lov_qos.lq_statfs_data)
889 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
891 lov->lov_pools_hash_body = lustre_hash_init("POOLS",
894 &pool_hash_operations,
896 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
897 lov->lov_pool_count = 0;
898 rc = lov_ost_pool_init(&lov->lov_packed, 0);
901 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
903 lov_ost_pool_free(&lov->lov_packed);
907 lprocfs_lov_init_vars(&lvars);
908 lprocfs_obd_setup(obd, lvars.obd_vars);
911 cfs_proc_dir_entry_t *entry;
913 entry = create_proc_entry("target_obd", 0444,
914 obd->obd_proc_entry);
916 entry->proc_fops = &lov_proc_target_fops;
921 lov->lov_pool_proc_entry = lprocfs_register("pools",
928 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
934 case OBD_CLEANUP_EARLY: {
935 struct lov_obd *lov = &obd->u.lov;
937 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
938 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
940 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
945 case OBD_CLEANUP_EXPORTS:
947 case OBD_CLEANUP_SELF_EXP:
948 rc = obd_llog_finish(obd, 0);
950 CERROR("failed to cleanup llogging subsystems\n");
952 case OBD_CLEANUP_OBD:
958 static int lov_cleanup(struct obd_device *obd)
960 struct lov_obd *lov = &obd->u.lov;
961 struct list_head *pos, *tmp;
962 struct pool_desc *pool;
964 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
965 pool = list_entry(pos, struct pool_desc, pool_list);
966 /* free the pool structs */
967 CDEBUG(D_INFO, "delete pool %p\n", pool);
968 lov_pool_del(obd, pool->pool_name);
971 lustre_hash_exit(lov->lov_pools_hash_body);
973 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
974 lov_ost_pool_free(&lov->lov_packed);
979 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
980 if (lov->lov_tgts[i]) {
981 /* Inactive targets may never have connected */
982 if (lov->lov_tgts[i]->ltd_active ||
983 atomic_read(&lov->lov_refcount))
984 /* We should never get here - these
985 should have been removed in the
987 CERROR("lov tgt %d not cleaned!"
988 " deathrow=%d, lovrc=%d\n",
989 i, lov->lov_death_row,
990 atomic_read(&lov->lov_refcount));
991 lov_del_target(obd, i, 0, 0);
995 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
997 lov->lov_tgt_size = 0;
1000 /* clear pools parent proc entry only after all pools is killed */
1001 lprocfs_obd_cleanup(obd);
1003 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
1007 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
1009 struct lustre_cfg *lcfg = buf;
1010 struct obd_uuid obd_uuid;
1015 switch(cmd = lcfg->lcfg_command) {
1016 case LCFG_LOV_ADD_OBD:
1017 case LCFG_LOV_ADD_INA:
1018 case LCFG_LOV_DEL_OBD: {
1021 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
1022 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1023 GOTO(out, rc = -EINVAL);
1025 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1027 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1028 GOTO(out, rc = -EINVAL);
1029 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1030 GOTO(out, rc = -EINVAL);
1031 if (cmd == LCFG_LOV_ADD_OBD)
1032 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
1033 else if (cmd == LCFG_LOV_ADD_INA)
1034 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
1036 rc = lov_del_target(obd, index, &obd_uuid, gen);
1040 struct lprocfs_static_vars lvars = { 0 };
1041 struct lov_desc *desc = &(obd->u.lov.desc);
1044 GOTO(out, rc = -EINVAL);
1046 lprocfs_lov_init_vars(&lvars);
1048 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1059 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1060 GOTO(out, rc = -EINVAL);
1069 #define log2(n) ffz(~(n))
1072 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1073 struct lov_stripe_md **ea,
1074 struct obd_trans_info *oti)
1076 struct lov_obd *lov;
1077 struct obdo *tmp_oa;
1078 struct obd_uuid *ost_uuid = NULL;
1082 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1083 src_oa->o_flags == OBD_FL_DELORPHAN);
1085 lov = &export->exp_obd->u.lov;
1091 if (oti->oti_ost_uuid) {
1092 ost_uuid = oti->oti_ost_uuid;
1093 CDEBUG(D_HA, "clearing orphans only for %s\n",
1097 obd_getref(export->exp_obd);
1098 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1099 struct lov_stripe_md obj_md;
1100 struct lov_stripe_md *obj_mdp = &obj_md;
1101 struct lov_tgt_desc *tgt;
1104 tgt = lov->lov_tgts[i];
1108 /* if called for a specific target, we don't
1109 care if it is not active. */
1110 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1111 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1115 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1118 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1119 obd_uuid2str(ost_uuid));
1121 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1123 LASSERT(lov->lov_tgts[i]->ltd_exp);
1124 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1125 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1126 tmp_oa, &obj_mdp, oti);
1128 /* This export will be disabled until it is recovered,
1129 and then orphan recovery will be completed. */
1130 CERROR("error in orphan recovery on OST idx %d/%d: "
1131 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1138 obd_putref(export->exp_obd);
1144 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1145 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1147 struct lov_stripe_md *obj_mdp, *lsm;
1148 struct lov_obd *lov = &exp->exp_obd->u.lov;
1153 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1154 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1156 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1157 if (obj_mdp == NULL)
1160 ost_idx = src_oa->o_nlink;
1163 GOTO(out, rc = -EINVAL);
1164 if (ost_idx >= lov->desc.ld_tgt_count ||
1165 !lov->lov_tgts[ost_idx])
1166 GOTO(out, rc = -EINVAL);
1168 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1169 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1170 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1171 GOTO(out, rc = -EINVAL);
1175 if (i == lsm->lsm_stripe_count)
1176 GOTO(out, rc = -EINVAL);
1178 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1180 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1184 /* the LOV expects oa->o_id to be set to the LOV object id */
1185 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1186 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1188 struct lov_obd *lov;
1189 struct obd_info *oinfo;
1190 struct lov_request_set *set = NULL;
1191 struct lov_request *req;
1192 struct l_wait_info lwi = { 0 };
1196 LASSERT(ea != NULL);
1200 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1201 src_oa->o_flags == OBD_FL_DELORPHAN) {
1202 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1206 lov = &exp->exp_obd->u.lov;
1207 if (!lov->desc.ld_active_tgt_count)
1210 OBD_ALLOC_PTR(oinfo);
1214 obd_getref(exp->exp_obd);
1215 /* Recreate a specific object id at the given OST index */
1216 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1217 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1218 rc = lov_recreate(exp, src_oa, ea, oti);
1222 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1223 * later in alloc_qos(), we will wait for those rpcs to complete if
1224 * the osfs age is older than 2 * qos_maxage */
1225 qos_statfs_update(exp->exp_obd,
1226 cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1227 OBD_STATFS_CACHE_SECONDS),
1230 rc = lov_prep_create_set(exp, oinfo, ea, src_oa, oti, &set);
1234 list_for_each_entry(req, &set->set_list, rq_link) {
1235 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1236 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1237 &req->rq_oi, &req->rq_oi.oi_md, oti);
1240 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1241 * longer then this */
1242 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1244 /* we not have ptlrpc set for assign set->interpret and should
1245 * be call interpret function himself. calling from cb_create_update
1246 * not permited because lov_fini_create_set can sleep for long time,
1247 * but we must avoid sleeping in ptlrpcd interpret function. */
1248 rc = lov_fini_create_set(set, ea);
1250 obd_putref(exp->exp_obd);
1251 OBD_FREE_PTR(oinfo);
1255 #define ASSERT_LSM_MAGIC(lsmp) \
1257 LASSERT((lsmp) != NULL); \
1258 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1259 (lsmp)->lsm_magic == LOV_MAGIC_V3 || \
1260 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1261 (lsmp), (lsmp)->lsm_magic); \
1264 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1265 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1266 struct obd_export *md_exp)
1268 struct lov_request_set *set;
1269 struct obd_info *oinfo;
1270 struct lov_request *req;
1271 struct list_head *pos;
1272 struct lov_obd *lov;
1273 int rc = 0, err = 0;
1276 ASSERT_LSM_MAGIC(lsm);
1278 if (!exp || !exp->exp_obd)
1281 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1283 LASSERT(oti->oti_logcookies);
1286 OBD_ALLOC_PTR(oinfo);
1290 lov = &exp->exp_obd->u.lov;
1291 obd_getref(exp->exp_obd);
1292 rc = lov_prep_destroy_set(exp, oinfo, oa, lsm, oti, &set);
1296 list_for_each (pos, &set->set_list) {
1297 req = list_entry(pos, struct lov_request, rq_link);
1299 if (oa->o_valid & OBD_MD_FLCOOKIE)
1300 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1302 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1303 req->rq_oi.oi_oa, NULL, oti, NULL);
1304 err = lov_update_common_set(set, req, err);
1306 CERROR("error: destroying objid "LPX64" subobj "
1307 LPX64" on OST idx %d: rc = %d\n",
1308 oa->o_id, req->rq_oi.oi_oa->o_id,
1316 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1317 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1319 err = lov_fini_destroy_set(set);
1321 obd_putref(exp->exp_obd);
1322 OBD_FREE_PTR(oinfo);
1323 RETURN(rc ? rc : err);
1326 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1328 struct lov_request_set *set;
1329 struct lov_request *req;
1330 struct list_head *pos;
1331 struct lov_obd *lov;
1332 int err = 0, rc = 0;
1336 ASSERT_LSM_MAGIC(oinfo->oi_md);
1338 if (!exp || !exp->exp_obd)
1341 lov = &exp->exp_obd->u.lov;
1343 rc = lov_prep_getattr_set(exp, oinfo, &set);
1347 list_for_each (pos, &set->set_list) {
1348 req = list_entry(pos, struct lov_request, rq_link);
1350 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1351 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1352 req->rq_oi.oi_oa->o_id, req->rq_idx);
1354 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1356 err = lov_update_common_set(set, req, rc);
1358 CERROR("error: getattr objid "LPX64" subobj "
1359 LPX64" on OST idx %d: rc = %d\n",
1360 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1366 rc = lov_fini_getattr_set(set);
1372 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1375 struct lov_request_set *lovset = (struct lov_request_set *)data;
1379 /* don't do attribute merge if this aysnc op failed */
1381 lovset->set_completes = 0;
1382 err = lov_fini_getattr_set(lovset);
1383 RETURN(rc ? rc : err);
1386 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1387 struct ptlrpc_request_set *rqset)
1389 struct lov_request_set *lovset;
1390 struct lov_obd *lov;
1391 struct list_head *pos;
1392 struct lov_request *req;
1397 ASSERT_LSM_MAGIC(oinfo->oi_md);
1399 if (!exp || !exp->exp_obd)
1402 lov = &exp->exp_obd->u.lov;
1404 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1408 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1409 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1410 oinfo->oi_md->lsm_stripe_size);
1412 list_for_each (pos, &lovset->set_list) {
1413 req = list_entry(pos, struct lov_request, rq_link);
1415 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1416 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1417 req->rq_oi.oi_oa->o_id, req->rq_idx);
1418 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1419 &req->rq_oi, rqset);
1421 CERROR("error: getattr objid "LPX64" subobj "
1422 LPX64" on OST idx %d: rc = %d\n",
1423 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1429 if (!list_empty(&rqset->set_requests)) {
1431 LASSERT (rqset->set_interpret == NULL);
1432 rqset->set_interpret = lov_getattr_interpret;
1433 rqset->set_arg = (void *)lovset;
1438 lovset->set_completes = 0;
1439 err = lov_fini_getattr_set(lovset);
1440 RETURN(rc ? rc : err);
1443 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1444 struct obd_trans_info *oti)
1446 struct lov_request_set *set;
1447 struct lov_obd *lov;
1448 struct list_head *pos;
1449 struct lov_request *req;
1450 int err = 0, rc = 0;
1454 ASSERT_LSM_MAGIC(oinfo->oi_md);
1456 if (!exp || !exp->exp_obd)
1459 /* for now, we only expect the following updates here */
1460 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1461 OBD_MD_FLMODE | OBD_MD_FLATIME |
1462 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1463 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1464 OBD_MD_FLGROUP | OBD_MD_FLUID |
1465 OBD_MD_FLGID | OBD_MD_FLFID |
1467 lov = &exp->exp_obd->u.lov;
1468 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1472 list_for_each (pos, &set->set_list) {
1473 req = list_entry(pos, struct lov_request, rq_link);
1475 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1477 err = lov_update_setattr_set(set, req, rc);
1479 CERROR("error: setattr objid "LPX64" subobj "
1480 LPX64" on OST idx %d: rc = %d\n",
1481 set->set_oi->oi_oa->o_id,
1482 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1487 err = lov_fini_setattr_set(set);
1493 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1496 struct lov_request_set *lovset = (struct lov_request_set *)data;
1501 lovset->set_completes = 0;
1502 err = lov_fini_setattr_set(lovset);
1503 RETURN(rc ? rc : err);
1506 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1507 needed. Otherwise, a client is waiting for responses. */
1508 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1509 struct obd_trans_info *oti,
1510 struct ptlrpc_request_set *rqset)
1512 struct lov_request_set *set;
1513 struct lov_request *req;
1514 struct list_head *pos;
1515 struct lov_obd *lov;
1520 ASSERT_LSM_MAGIC(oinfo->oi_md);
1521 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1523 LASSERT(oti->oti_logcookies);
1526 if (!exp || !exp->exp_obd)
1529 lov = &exp->exp_obd->u.lov;
1530 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1534 CDEBUG(D_INFO, "objid "LPX64"@"LPX64": %ux%u byte stripes\n",
1535 oinfo->oi_md->lsm_object_id,
1536 oinfo->oi_md->lsm_object_gr,
1537 oinfo->oi_md->lsm_stripe_count,
1538 oinfo->oi_md->lsm_stripe_size);
1540 list_for_each (pos, &set->set_list) {
1541 req = list_entry(pos, struct lov_request, rq_link);
1543 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1544 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1546 CDEBUG(D_INFO, "objid "LPX64"@"LPX64"[%d] has subobj "LPX64
1547 " at idx %u\n", oinfo->oi_oa->o_id, oinfo->oi_oa->o_gr,
1548 req->rq_stripe, req->rq_oi.oi_oa->o_id, req->rq_idx);
1550 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1551 &req->rq_oi, oti, rqset);
1553 CERROR("error: setattr objid "LPX64" subobj "
1554 LPX64" on OST idx %d: rc = %d\n",
1555 set->set_oi->oi_oa->o_id,
1556 req->rq_oi.oi_oa->o_id,
1562 /* If we are not waiting for responses on async requests, return. */
1563 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1566 set->set_completes = 0;
1567 err = lov_fini_setattr_set(set);
1568 RETURN(rc ? rc : err);
1571 LASSERT(rqset->set_interpret == NULL);
1572 rqset->set_interpret = lov_setattr_interpret;
1573 rqset->set_arg = (void *)set;
1578 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1581 struct lov_request_set *lovset = (struct lov_request_set *)data;
1586 lovset->set_completes = 0;
1587 err = lov_fini_punch_set(lovset);
1588 RETURN(rc ? rc : err);
1591 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1592 * we can send this 'punch' to just the authoritative node and the nodes
1593 * that the punch will affect. */
1594 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1595 struct obd_trans_info *oti,
1596 struct ptlrpc_request_set *rqset)
1598 struct lov_request_set *set;
1599 struct lov_obd *lov;
1600 struct list_head *pos;
1601 struct lov_request *req;
1606 ASSERT_LSM_MAGIC(oinfo->oi_md);
1608 if (!exp || !exp->exp_obd)
1611 lov = &exp->exp_obd->u.lov;
1612 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1616 list_for_each (pos, &set->set_list) {
1617 req = list_entry(pos, struct lov_request, rq_link);
1619 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1620 &req->rq_oi, NULL, rqset);
1622 CERROR("error: punch objid "LPX64" subobj "LPX64
1623 " on OST idx %d: rc = %d\n",
1624 set->set_oi->oi_oa->o_id,
1625 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1630 if (rc || list_empty(&rqset->set_requests)) {
1632 err = lov_fini_punch_set(set);
1633 RETURN(rc ? rc : err);
1636 LASSERT(rqset->set_interpret == NULL);
1637 rqset->set_interpret = lov_punch_interpret;
1638 rqset->set_arg = (void *)set;
1643 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1646 struct lov_request_set *lovset = (struct lov_request_set *)data;
1651 lovset->set_completes = 0;
1652 err = lov_fini_sync_set(lovset);
1653 RETURN(rc ? rc : err);
1656 static int lov_sync(struct obd_export *exp, struct obd_info *oinfo,
1657 obd_off start, obd_off end,
1658 struct ptlrpc_request_set *rqset)
1660 struct lov_request_set *set = NULL;
1661 struct lov_obd *lov;
1662 struct list_head *pos;
1663 struct lov_request *req;
1667 ASSERT_LSM_MAGIC(oinfo->oi_md);
1668 LASSERT(rqset != NULL);
1673 lov = &exp->exp_obd->u.lov;
1674 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1678 list_for_each (pos, &set->set_list) {
1679 req = list_entry(pos, struct lov_request, rq_link);
1681 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi,
1682 req->rq_oi.oi_policy.l_extent.start,
1683 req->rq_oi.oi_policy.l_extent.end, rqset);
1685 CERROR("error: fsync objid "LPX64" subobj "LPX64
1686 " on OST idx %d: rc = %d\n",
1687 set->set_oi->oi_oa->o_id,
1688 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1693 /* If we are not waiting for responses on async requests, return. */
1694 if (rc || list_empty(&rqset->set_requests)) {
1695 int err = lov_fini_sync_set(set);
1696 RETURN(rc ? rc : err);
1699 LASSERT(rqset->set_interpret == NULL);
1700 rqset->set_interpret = lov_sync_interpret;
1701 rqset->set_arg = (void *)set;
1706 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1707 obd_count oa_bufs, struct brw_page *pga)
1709 struct obd_info oinfo = { { { 0 } } };
1712 oinfo.oi_oa = lov_oinfo->oi_oa;
1714 /* The caller just wants to know if there's a chance that this
1715 * I/O can succeed */
1716 for (i = 0; i < oa_bufs; i++) {
1717 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1718 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1721 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1722 pga[i].off + pga[i].count - 1,
1726 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1727 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1731 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1739 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1740 obd_count oa_bufs, struct brw_page *pga,
1741 struct obd_trans_info *oti)
1743 struct lov_request_set *set;
1744 struct lov_request *req;
1745 struct list_head *pos;
1746 struct lov_obd *lov = &exp->exp_obd->u.lov;
1750 ASSERT_LSM_MAGIC(oinfo->oi_md);
1752 if (cmd == OBD_BRW_CHECK) {
1753 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1757 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1761 list_for_each (pos, &set->set_list) {
1762 struct obd_export *sub_exp;
1763 struct brw_page *sub_pga;
1764 req = list_entry(pos, struct lov_request, rq_link);
1766 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1767 sub_pga = set->set_pga + req->rq_pgaidx;
1768 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1772 lov_update_common_set(set, req, rc);
1775 err = lov_fini_brw_set(set);
1781 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1784 struct lov_request_set *lovset = (struct lov_request_set *)data;
1788 lovset->set_completes = 0;
1789 lov_fini_brw_set(lovset);
1791 rc = lov_fini_brw_set(lovset);
1797 static int lov_brw_async(int cmd, struct obd_export *exp,
1798 struct obd_info *oinfo, obd_count oa_bufs,
1799 struct brw_page *pga, struct obd_trans_info *oti,
1800 struct ptlrpc_request_set *set, int pshift)
1802 struct lov_request_set *lovset;
1803 struct lov_request *req;
1804 struct list_head *pos;
1805 struct lov_obd *lov = &exp->exp_obd->u.lov;
1810 ASSERT_LSM_MAGIC(oinfo->oi_md);
1812 if (cmd == OBD_BRW_CHECK) {
1813 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1817 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1821 list_for_each (pos, &lovset->set_list) {
1822 struct obd_export *sub_exp;
1823 struct brw_page *sub_pga;
1824 req = list_entry(pos, struct lov_request, rq_link);
1826 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1827 sub_pga = lovset->set_pga + req->rq_pgaidx;
1828 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1829 sub_pga, oti, set, pshift);
1832 lov_update_common_set(lovset, req, rc);
1835 LASSERT(set->set_interpret == NULL);
1836 LASSERT(set->set_arg == NULL);
1837 rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1843 lov_fini_brw_set(lovset);
1847 static int lov_ap_make_ready(void *data, int cmd)
1849 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1851 return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1854 static int lov_ap_refresh_count(void *data, int cmd)
1856 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1858 return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1862 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1864 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1866 lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1867 /* XXX woah, shouldn't we be altering more here? size? */
1868 oa->o_id = lap->lap_loi_id;
1869 oa->o_stripe_idx = lap->lap_stripe;
1872 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1875 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1877 lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1880 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1882 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1884 /* in a raid1 regime this would down a count of many ios
1885 * in flight, onl calling the caller_ops completion when all
1886 * the raid1 ios are complete */
1887 rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1891 static struct obd_async_page_ops lov_async_page_ops = {
1892 .ap_make_ready = lov_ap_make_ready,
1893 .ap_refresh_count = lov_ap_refresh_count,
1894 .ap_fill_obdo = lov_ap_fill_obdo,
1895 .ap_update_obdo = lov_ap_update_obdo,
1896 .ap_completion = lov_ap_completion,
1899 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1900 struct lov_oinfo *loi, cfs_page_t *page,
1901 obd_off offset, struct obd_async_page_ops *ops,
1902 void *data, void **res, int flags,
1903 struct lustre_handle *lockh)
1905 struct lov_obd *lov = &exp->exp_obd->u.lov;
1906 struct lov_async_page *lap;
1907 struct lov_lock_handles *lov_lockh = NULL;
1913 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1914 Only because of this layering limitation will a client
1915 mount with no osts fail */
1916 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1917 !lov->lov_tgts[i]->ltd_exp) {
1919 if (i >= lov->desc.ld_tgt_count)
1922 rc = size_round(sizeof(*lap)) +
1923 obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1924 NULL, NULL, 0, NULL, NULL, NULL, 0,
1928 ASSERT_LSM_MAGIC(lsm);
1929 LASSERT(loi == NULL);
1932 lap->lap_magic = LOV_AP_MAGIC;
1933 lap->lap_caller_ops = ops;
1934 lap->lap_caller_data = data;
1936 /* for now only raid 0 which passes through */
1937 lap->lap_stripe = lov_stripe_number(lsm, offset);
1938 lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1939 loi = lsm->lsm_oinfo[lap->lap_stripe];
1941 /* so the callback doesn't need the lsm */
1942 lap->lap_loi_id = loi->loi_id;
1944 lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1946 if (lockh && lustre_handle_is_used(lockh) && !(flags & OBD_FAST_LOCK)) {
1947 lov_lockh = lov_handle2llh(lockh);
1949 lockh = lov_lockh->llh_handles + lap->lap_stripe;
1953 rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1954 lsm, loi, page, lap->lap_sub_offset,
1955 &lov_async_page_ops, lap,
1956 &lap->lap_sub_cookie, flags, lockh);
1958 lov_llh_put(lov_lockh);
1961 CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1962 lap->lap_sub_cookie, offset);
1966 static int lov_queue_async_io(struct obd_export *exp,
1967 struct lov_stripe_md *lsm,
1968 struct lov_oinfo *loi, void *cookie,
1969 int cmd, obd_off off, int count,
1970 obd_flag brw_flags, obd_flag async_flags)
1972 struct lov_obd *lov = &exp->exp_obd->u.lov;
1973 struct lov_async_page *lap;
1976 LASSERT(loi == NULL);
1978 ASSERT_LSM_MAGIC(lsm);
1980 lap = LAP_FROM_COOKIE(cookie);
1982 loi = lsm->lsm_oinfo[lap->lap_stripe];
1984 rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1985 loi, lap->lap_sub_cookie, cmd, off, count,
1986 brw_flags, async_flags);
1990 static int lov_set_async_flags(struct obd_export *exp,
1991 struct lov_stripe_md *lsm,
1992 struct lov_oinfo *loi, void *cookie,
1993 obd_flag async_flags)
1995 struct lov_obd *lov = &exp->exp_obd->u.lov;
1996 struct lov_async_page *lap;
1999 LASSERT(loi == NULL);
2001 ASSERT_LSM_MAGIC(lsm);
2003 lap = LAP_FROM_COOKIE(cookie);
2005 loi = lsm->lsm_oinfo[lap->lap_stripe];
2007 rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2008 lsm, loi, lap->lap_sub_cookie, async_flags);
2012 static int lov_queue_group_io(struct obd_export *exp,
2013 struct lov_stripe_md *lsm,
2014 struct lov_oinfo *loi,
2015 struct obd_io_group *oig, void *cookie,
2016 int cmd, obd_off off, int count,
2017 obd_flag brw_flags, obd_flag async_flags)
2019 struct lov_obd *lov = &exp->exp_obd->u.lov;
2020 struct lov_async_page *lap;
2023 LASSERT(loi == NULL);
2025 ASSERT_LSM_MAGIC(lsm);
2027 lap = LAP_FROM_COOKIE(cookie);
2029 loi = lsm->lsm_oinfo[lap->lap_stripe];
2031 rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
2032 loi, oig, lap->lap_sub_cookie, cmd, off, count,
2033 brw_flags, async_flags);
2037 /* this isn't exactly optimal. we may have queued sync io in oscs on
2038 * all stripes, but we don't record that fact at queue time. so we
2039 * trigger sync io on all stripes. */
2040 static int lov_trigger_group_io(struct obd_export *exp,
2041 struct lov_stripe_md *lsm,
2042 struct lov_oinfo *loi,
2043 struct obd_io_group *oig)
2045 struct lov_obd *lov = &exp->exp_obd->u.lov;
2048 LASSERT(loi == NULL);
2050 ASSERT_LSM_MAGIC(lsm);
2052 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2053 loi = lsm->lsm_oinfo[i];
2054 if (!lov->lov_tgts[loi->loi_ost_idx] ||
2055 !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2056 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2060 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2062 if (rc == 0 && err != 0)
2068 static int lov_teardown_async_page(struct obd_export *exp,
2069 struct lov_stripe_md *lsm,
2070 struct lov_oinfo *loi, void *cookie)
2072 struct lov_obd *lov = &exp->exp_obd->u.lov;
2073 struct lov_async_page *lap;
2076 LASSERT(loi == NULL);
2078 ASSERT_LSM_MAGIC(lsm);
2080 lap = LAP_FROM_COOKIE(cookie);
2082 loi = lsm->lsm_oinfo[lap->lap_stripe];
2084 rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2085 lsm, loi, lap->lap_sub_cookie);
2087 CERROR("unable to teardown sub cookie %p: %d\n",
2088 lap->lap_sub_cookie, rc);
2094 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2097 struct lov_request_set *lovset = (struct lov_request_set *)data;
2099 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2103 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2104 struct ldlm_enqueue_info *einfo,
2105 struct ptlrpc_request_set *rqset)
2107 ldlm_mode_t mode = einfo->ei_mode;
2108 struct lov_request_set *set;
2109 struct lov_request *req;
2110 struct list_head *pos;
2111 struct lov_obd *lov;
2116 ASSERT_LSM_MAGIC(oinfo->oi_md);
2117 LASSERT(mode == (mode & -mode));
2119 /* we should never be asked to replay a lock this way. */
2120 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2122 if (!exp || !exp->exp_obd)
2125 lov = &exp->exp_obd->u.lov;
2126 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2130 list_for_each (pos, &set->set_list) {
2131 req = list_entry(pos, struct lov_request, rq_link);
2133 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2134 &req->rq_oi, einfo, rqset);
2139 if (rqset && !list_empty(&rqset->set_requests)) {
2141 LASSERT(rqset->set_interpret == NULL);
2142 rqset->set_interpret = lov_enqueue_interpret;
2143 rqset->set_arg = (void *)set;
2147 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2151 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2152 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2153 int *flags, void *data, struct lustre_handle *lockh,
2156 struct lov_request_set *set;
2157 struct obd_info oinfo;
2158 struct lov_request *req;
2159 struct list_head *pos;
2160 struct lov_obd *lov;
2161 struct lustre_handle *lov_lockhp;
2162 int lov_flags, rc = 0;
2165 ASSERT_LSM_MAGIC(lsm);
2166 LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2168 if (!exp || !exp->exp_obd)
2171 lov = &exp->exp_obd->u.lov;
2172 rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2176 list_for_each (pos, &set->set_list) {
2177 ldlm_policy_data_t sub_policy;
2178 req = list_entry(pos, struct lov_request, rq_link);
2179 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2180 LASSERT(lov_lockhp);
2183 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2185 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2186 req->rq_oi.oi_md, type, &sub_policy,
2187 mode, &lov_flags, data, lov_lockhp,
2189 rc = lov_update_match_set(set, req, rc);
2193 lov_fini_match_set(set, mode, *flags);
2197 static int lov_change_cbdata(struct obd_export *exp,
2198 struct lov_stripe_md *lsm, ldlm_iterator_t it,
2201 struct lov_obd *lov;
2202 struct lov_oinfo *loi;
2206 ASSERT_LSM_MAGIC(lsm);
2208 if (!exp || !exp->exp_obd)
2211 lov = &exp->exp_obd->u.lov;
2212 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2213 struct lov_stripe_md submd;
2215 loi = lsm->lsm_oinfo[i];
2216 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2217 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2220 submd.lsm_object_id = loi->loi_id;
2221 submd.lsm_object_gr = loi->loi_gr;
2222 submd.lsm_stripe_count = 0;
2223 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2229 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2230 __u32 mode, struct lustre_handle *lockh, int flags,
2233 struct lov_request_set *set;
2234 struct obd_info oinfo;
2235 struct lov_request *req;
2236 struct list_head *pos;
2237 struct lov_obd *lov;
2238 struct lustre_handle *lov_lockhp;
2239 ldlm_mode_t this_mode;
2240 int err = 0, rc = 0;
2243 ASSERT_LSM_MAGIC(lsm);
2245 if (!exp || !exp->exp_obd)
2249 lov = &exp->exp_obd->u.lov;
2250 if (flags & OBD_FAST_LOCK) {
2251 int stripe = lov_stripe_number(lsm, end);
2252 RETURN(obd_cancel(lov->lov_tgts[lsm->lsm_oinfo[stripe]->
2253 loi_ost_idx]->ltd_exp, NULL, mode, lockh,
2257 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2261 list_for_each (pos, &set->set_list) {
2262 req = list_entry(pos, struct lov_request, rq_link);
2263 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2265 /* If this lock was used for a write or truncate, the object
2266 * will have been recreated by the OST, cancel the lock
2267 * (setting LCK_GROUP incidentally causes immediate cancel). */
2268 if (OST_LVB_IS_ERR(lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_blocks) &&
2269 (mode == LCK_PW || mode == LCK_CW))
2270 this_mode = LCK_GROUP;
2274 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2275 req->rq_oi.oi_md, this_mode, lov_lockhp, flags,
2277 rc = lov_update_common_set(set, req, rc);
2279 CERROR("error: cancel objid "LPX64" subobj "
2280 LPX64" on OST idx %d: rc = %d\n",
2282 req->rq_oi.oi_md->lsm_object_id,
2288 lov_fini_cancel_set(set);
2292 static int lov_cancel_unused(struct obd_export *exp,
2293 struct lov_stripe_md *lsm, int flags, void *opaque)
2295 struct lov_obd *lov;
2296 struct lov_oinfo *loi;
2300 if (!exp || !exp->exp_obd)
2303 lov = &exp->exp_obd->u.lov;
2305 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2307 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2310 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2318 ASSERT_LSM_MAGIC(lsm);
2320 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2321 struct lov_stripe_md submd;
2324 loi = lsm->lsm_oinfo[i];
2325 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2326 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2330 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2331 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2333 submd.lsm_object_id = loi->loi_id;
2334 submd.lsm_stripe_count = 0;
2335 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2336 &submd, flags, opaque);
2337 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2338 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2339 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2340 loi->loi_id, loi->loi_ost_idx, err);
2348 static int lov_join_lru(struct obd_export *exp,
2349 struct lov_stripe_md *lsm, int join)
2351 struct lov_obd *lov;
2352 struct lov_oinfo *loi;
2356 ASSERT_LSM_MAGIC(lsm);
2357 if (!exp || !exp->exp_obd)
2360 lov = &exp->exp_obd->u.lov;
2361 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2362 struct lov_stripe_md submd;
2365 loi = lsm->lsm_oinfo[i];
2366 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2367 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2371 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2372 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2374 submd.lsm_object_id = loi->loi_id;
2375 submd.lsm_stripe_count = 0;
2376 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2379 CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2380 " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2381 loi->loi_id, loi->loi_ost_idx, rc);
2390 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
2392 struct lov_request_set *lovset = (struct lov_request_set *)data;
2397 lovset->set_completes = 0;
2399 err = lov_fini_statfs_set(lovset);
2400 RETURN(rc ? rc : err);
2403 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2404 __u64 max_age, struct ptlrpc_request_set *rqset)
2406 struct lov_request_set *set;
2407 struct lov_request *req;
2408 struct list_head *pos;
2409 struct lov_obd *lov;
2413 LASSERT(oinfo != NULL);
2414 LASSERT(oinfo->oi_osfs != NULL);
2417 rc = lov_prep_statfs_set(obd, oinfo, &set);
2421 list_for_each (pos, &set->set_list) {
2422 struct obd_device *osc_obd;
2424 req = list_entry(pos, struct lov_request, rq_link);
2426 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2427 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2432 if (rc || list_empty(&rqset->set_requests)) {
2435 set->set_completes = 0;
2436 err = lov_fini_statfs_set(set);
2437 RETURN(rc ? rc : err);
2440 LASSERT(rqset->set_interpret == NULL);
2441 rqset->set_interpret = lov_statfs_interpret;
2442 rqset->set_arg = (void *)set;
2446 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2447 __u64 max_age, __u32 flags)
2449 struct ptlrpc_request_set *set = NULL;
2450 struct obd_info oinfo = { { { 0 } } };
2454 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2455 * statfs requests */
2456 set = ptlrpc_prep_set();
2460 oinfo.oi_osfs = osfs;
2461 oinfo.oi_flags = flags;
2462 rc = lov_statfs_async(obd, &oinfo, max_age, set);
2464 rc = ptlrpc_set_wait(set);
2465 ptlrpc_set_destroy(set);
2470 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2471 void *karg, void *uarg)
2473 struct obd_device *obddev = class_exp2obd(exp);
2474 struct lov_obd *lov = &obddev->u.lov;
2475 int i, rc, count = lov->desc.ld_tgt_count;
2476 struct obd_uuid *uuidp;
2480 case OBD_IOC_LOV_GET_CONFIG: {
2481 struct obd_ioctl_data *data;
2482 struct lov_desc *desc;
2487 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2490 data = (struct obd_ioctl_data *)buf;
2492 if (sizeof(*desc) > data->ioc_inllen1) {
2493 obd_ioctl_freedata(buf, len);
2497 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2498 obd_ioctl_freedata(buf, len);
2502 if (sizeof(__u32) * count > data->ioc_inllen3) {
2503 obd_ioctl_freedata(buf, len);
2507 desc = (struct lov_desc *)data->ioc_inlbuf1;
2508 memcpy(desc, &(lov->desc), sizeof(*desc));
2510 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2511 genp = (__u32 *)data->ioc_inlbuf3;
2512 /* the uuid will be empty for deleted OSTs */
2513 for (i = 0; i < count; i++, uuidp++, genp++) {
2514 if (!lov->lov_tgts[i])
2516 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2517 *genp = lov->lov_tgts[i]->ltd_gen;
2520 rc = copy_to_user((void *)uarg, buf, len);
2523 obd_ioctl_freedata(buf, len);
2526 case LL_IOC_LOV_SETSTRIPE:
2527 rc = lov_setstripe(exp, karg, uarg);
2529 case LL_IOC_LOV_GETSTRIPE:
2530 rc = lov_getstripe(exp, karg, uarg);
2532 case LL_IOC_LOV_SETEA:
2533 rc = lov_setea(exp, karg, uarg);
2542 for (i = 0; i < count; i++) {
2545 /* OST was disconnected */
2546 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2549 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2551 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2554 if (lov->lov_tgts[i]->ltd_active) {
2555 CDEBUG(err == -ENOTTY ?
2556 D_IOCTL : D_WARNING,
2557 "iocontrol OSC %s on OST "
2558 "idx %d cmd %x: err = %d\n",
2559 lov_uuid2str(lov, i),
2576 #define FIEMAP_BUFFER_SIZE 4096
2578 /* Non-zero fe_logical indicates that this is a continuation FIEMAP
2579 * call. The local end offset and the device are sent in the first
2580 * fm_extent. This function calculates the stripe number from the index.
2581 * This function returns a stripe_no on which mapping is to be restarted.
2583 * This function returns fm_end_offset which is the in-OST offset at which
2584 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2585 * will re-calculate proper offset in next stripe.
2586 * Note that the first extent is passed to lov_get_info via the value field */
2587 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2588 struct lov_stripe_md *lsm, obd_size fm_start,
2589 obd_size fm_end, int *start_stripe)
2591 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2592 obd_off lun_start, lun_end;
2593 obd_size fm_end_offset;
2594 int stripe_no = -1, i;
2596 if (fiemap->fm_extent_count == 0 ||
2597 fiemap->fm_extents[0].fe_logical == 0)
2600 /* Find out stripe_no from ost_index saved in the fe_device */
2601 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2602 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2603 fiemap->fm_extents[0].fe_device) {
2609 /* If we have finished mapping on previous device, shift logical
2610 * offset to start of next device */
2611 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2612 &lun_start, &lun_end)) != 0 &&
2613 local_end < lun_end) {
2614 fm_end_offset = local_end;
2615 *start_stripe = stripe_no;
2617 /* This is a special value to indicate that caller should
2618 * calculate offset in next stripe. */
2620 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2623 return fm_end_offset;
2626 /* We calculate on which OST the mapping will end. If the length of mapping
2627 * is greater than (stripe_size * stripe_count) then the last_stripe will
2628 * will be one just before start_stripe. Else we check if the mapping
2629 * intersects each OST and find last_stripe.
2630 * This function returns the last_stripe and also sets the stripe_count
2631 * over which the mapping is spread */
2632 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2633 obd_size fm_end, int start_stripe,
2637 obd_off obd_start, obd_end;
2640 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2641 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2643 *stripe_count = lsm->lsm_stripe_count;
2645 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2646 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2647 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2648 &obd_start, &obd_end)) == 0)
2652 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2658 /* Set fe_device and copy extents from local buffer into main return buffer */
2659 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2660 struct ll_fiemap_extent *lcl_fm_ext,
2661 int ost_index, unsigned int ext_count,
2667 for (ext = 0; ext < ext_count; ext++) {
2668 lcl_fm_ext[ext].fe_device = ost_index;
2669 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2672 /* Copy fm_extent's from fm_local to return buffer */
2673 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2674 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2677 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2678 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2680 struct ll_fiemap_info_key *fm_key = key;
2681 struct ll_user_fiemap *fiemap = val;
2682 struct ll_user_fiemap *fm_local = NULL;
2683 struct ll_fiemap_extent *lcl_fm_ext;
2685 unsigned int get_num_extents = 0;
2686 int ost_index = 0, actual_start_stripe, start_stripe;
2687 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2689 int current_extent = 0, rc = 0, i;
2690 int ost_eof = 0; /* EOF for object */
2691 int ost_done = 0; /* done with required mapping for this OST? */
2693 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2694 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2699 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2700 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2702 OBD_ALLOC(fm_local, buffer_size);
2703 if (fm_local == NULL)
2704 GOTO(out, rc = -ENOMEM);
2705 lcl_fm_ext = &fm_local->fm_extents[0];
2707 count_local = fiemap_size_to_count(buffer_size);
2709 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2710 fm_start = fiemap->fm_start;
2711 fm_length = fiemap->fm_length;
2712 /* Calculate start stripe, last stripe and length of mapping */
2713 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2714 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2715 fm_start + fm_length - 1);
2716 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2717 if (fm_end > fm_key->oa.o_size)
2718 fm_end = fm_key->oa.o_size;
2720 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2721 actual_start_stripe, &stripe_count);
2723 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2726 if (fiemap->fm_extent_count == 0) {
2727 get_num_extents = 1;
2731 /* Check each stripe */
2732 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2733 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2734 obd_size req_fm_len; /* Stores length of required mapping */
2735 obd_size len_mapped_single_call;
2736 obd_off lun_start, lun_end, obd_object_end;
2737 unsigned int ext_count;
2739 cur_stripe_wrap = cur_stripe;
2741 /* Find out range of mapping on this stripe */
2742 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2743 &lun_start, &obd_object_end)) == 0)
2746 /* If this is a continuation FIEMAP call and we are on
2747 * starting stripe then lun_start needs to be set to
2749 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2750 lun_start = fm_end_offset;
2752 if (fm_length != ~0ULL) {
2753 /* Handle fm_start + fm_length overflow */
2754 if (fm_start + fm_length < fm_start)
2755 fm_length = ~0ULL - fm_start;
2756 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2762 if (lun_start == lun_end)
2765 req_fm_len = obd_object_end - lun_start;
2766 fm_local->fm_length = 0;
2767 len_mapped_single_call = 0;
2769 /* If the output buffer is very large and the objects have many
2770 * extents we may need to loop on a single OST repeatedly */
2774 if (get_num_extents == 0) {
2775 /* Don't get too many extents. */
2776 if (current_extent + count_local >
2777 fiemap->fm_extent_count)
2778 count_local = fiemap->fm_extent_count -
2782 lun_start += len_mapped_single_call;
2783 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2784 req_fm_len = fm_local->fm_length;
2785 fm_local->fm_extent_count = count_local;
2786 fm_local->fm_mapped_extents = 0;
2787 fm_local->fm_flags = fiemap->fm_flags;
2789 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2790 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2792 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2793 GOTO(out, rc = -EINVAL);
2795 /* If OST is inactive, return extent with UNKNOWN flag */
2796 if (!lov->lov_tgts[ost_index]->ltd_active) {
2797 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2798 fm_local->fm_mapped_extents = 1;
2800 lcl_fm_ext[0].fe_logical = lun_start;
2801 lcl_fm_ext[0].fe_length = obd_object_end -
2803 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2808 fm_local->fm_start = lun_start;
2809 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2810 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2811 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2812 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2813 keylen, key, vallen, fm_local, lsm);
2818 ext_count = fm_local->fm_mapped_extents;
2819 if (ext_count == 0) {
2821 /* If last stripe has hole at the end,
2822 * then we need to return */
2823 if (cur_stripe_wrap == last_stripe) {
2824 fiemap->fm_mapped_extents = 0;
2830 /* If we just need num of extents then go to next device */
2831 if (get_num_extents) {
2832 current_extent += ext_count;
2836 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2837 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2839 /* Have we finished mapping on this device? */
2840 if (req_fm_len <= len_mapped_single_call)
2843 /* Clear the EXTENT_LAST flag which can be present on
2845 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2846 lcl_fm_ext[ext_count - 1].fe_flags &=
2847 ~FIEMAP_EXTENT_LAST;
2849 curr_loc = lov_stripe_size(lsm,
2850 lcl_fm_ext[ext_count - 1].fe_logical+
2851 lcl_fm_ext[ext_count - 1].fe_length,
2853 if (curr_loc >= fm_key->oa.o_size)
2856 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2857 ost_index, ext_count,
2860 current_extent += ext_count;
2862 /* Ran out of available extents? */
2863 if (current_extent >= fiemap->fm_extent_count)
2865 } while (ost_done == 0 && ost_eof == 0);
2867 if (cur_stripe_wrap == last_stripe)
2872 /* Indicate that we are returning device offsets unless file just has
2874 if (lsm->lsm_stripe_count > 1)
2875 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2877 if (get_num_extents)
2878 goto skip_last_device_calc;
2880 /* Check if we have reached the last stripe and whether mapping for that
2881 * stripe is done. */
2882 if (cur_stripe_wrap == last_stripe) {
2883 if (ost_done || ost_eof)
2884 fiemap->fm_extents[current_extent - 1].fe_flags |=
2888 skip_last_device_calc:
2889 fiemap->fm_mapped_extents = current_extent;
2892 OBD_FREE(fm_local, buffer_size);
2896 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2897 void *key, __u32 *vallen, void *val,
2898 struct lov_stripe_md *lsm)
2900 struct obd_device *obddev = class_exp2obd(exp);
2901 struct lov_obd *lov = &obddev->u.lov;
2905 if (!vallen || !val)
2910 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2913 struct ldlm_lock *lock;
2915 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2916 struct lov_oinfo *loi;
2917 __u32 *stripe = val;
2919 if (*vallen < sizeof(*stripe))
2920 GOTO(out, rc = -EFAULT);
2921 *vallen = sizeof(*stripe);
2923 /* XXX This is another one of those bits that will need to
2924 * change if we ever actually support nested LOVs. It uses
2925 * the lock's export to find out which stripe it is. */
2926 /* XXX - it's assumed all the locks for deleted OSTs have
2927 * been cancelled. Also, the export for deleted OSTs will
2928 * be NULL and won't match the lock's export. */
2929 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2930 loi = lsm->lsm_oinfo[i];
2931 if (!lov->lov_tgts[loi->loi_ost_idx])
2933 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2934 data->lock->l_conn_export &&
2935 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2940 LDLM_ERROR(data->lock, "lock on inode without such object");
2941 dump_lsm(D_ERROR, lsm);
2942 GOTO(out, rc = -ENXIO);
2943 } else if (KEY_IS(KEY_LAST_ID)) {
2944 struct obd_id_info *info = val;
2945 __u32 size = sizeof(obd_id);
2946 struct lov_tgt_desc *tgt;
2948 LASSERT(*vallen == sizeof(struct obd_id_info));
2949 tgt = lov->lov_tgts[info->idx];
2951 if (!tgt || !tgt->ltd_active)
2952 GOTO(out, rc = -ESRCH);
2954 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2956 } else if (KEY_IS(KEY_LOVDESC)) {
2957 struct lov_desc *desc_ret = val;
2958 *desc_ret = lov->desc;
2961 } else if (KEY_IS(KEY_FIEMAP)) {
2962 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2964 } else if (KEY_IS(KEY_OFF_RPCSIZE)) {
2965 __u64 *offset = val;
2966 struct lov_tgt_desc *tgt;
2967 struct lov_oinfo *loi;
2970 LASSERT(*vallen == sizeof(__u64));
2971 stripe = lov_stripe_number(lsm, *offset);
2972 loi = lsm->lsm_oinfo[stripe];
2973 tgt = lov->lov_tgts[loi->loi_ost_idx];
2974 if (!tgt || !tgt->ltd_active)
2975 GOTO(out, rc = -ESRCH);
2976 rc = obd_get_info(tgt->ltd_exp, keylen, key, vallen, val, NULL);
2978 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2979 struct lov_tgt_desc *tgt;
2980 __u64 ost_idx = *((__u64*)val);
2982 LASSERT(*vallen == sizeof(__u64));
2983 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2984 tgt = lov->lov_tgts[ost_idx];
2986 if (!tgt || !tgt->ltd_exp)
2987 GOTO(out, rc = -ESRCH);
2989 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2999 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
3000 void *key, obd_count vallen, void *val,
3001 struct ptlrpc_request_set *set)
3003 struct obd_device *obddev = class_exp2obd(exp);
3004 struct lov_obd *lov = &obddev->u.lov;
3006 int i, rc = 0, err, incr = 0, check_uuid = 0, do_inactive = 0;
3008 unsigned next_id = 0;
3009 struct lov_tgt_desc *tgt;
3014 set = ptlrpc_prep_set();
3020 count = lov->desc.ld_tgt_count;
3022 if (KEY_IS(KEY_NEXT_ID)) {
3023 count = vallen / sizeof(struct obd_id_info);
3024 vallen = sizeof(obd_id);
3025 incr = sizeof(struct obd_id_info);
3028 } else if (KEY_IS(KEY_CHECKSUM)) {
3030 } else if (KEY_IS(KEY_MDS_CONN)) {
3031 check_uuid = val ? 1 : 0;
3032 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
3034 do_inactive = incr = 0;
3038 for (i = 0; i < count; i++, val = (char *)val + incr) {
3040 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
3041 data = ((struct obd_id_info*)val)->data;
3043 tgt = lov->lov_tgts[i];
3046 /* OST was disconnected */
3047 if (!tgt || !tgt->ltd_exp)
3050 /* OST is inactive and we don't want inactive OSCs */
3051 if (!tgt->ltd_active && !do_inactive)
3054 /* Only want a specific OSC */
3056 !obd_uuid_equals(val, &tgt->ltd_uuid))
3059 err = obd_set_info_async(tgt->ltd_exp,
3060 keylen, key, vallen, data, set);
3066 err = ptlrpc_set_wait(set);
3069 ptlrpc_set_destroy(set);
3074 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3075 struct lov_stripe_md *lsm)
3083 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3084 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3089 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3091 struct lov_oinfo *loi;
3095 for (i = 0; i < lsm->lsm_stripe_count; i++) {
3096 loi = lsm->lsm_oinfo[i];
3097 if (loi->loi_ar.ar_rc && !rc)
3098 rc = loi->loi_ar.ar_rc;
3099 loi->loi_ar.ar_rc = 0;
3103 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3106 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3107 int cmd, __u64 *offset)
3110 __u32 ssize = lsm->lsm_stripe_size;
3112 if (cmd & OBD_CALC_STRIPE_RPC_ALIGN)
3113 ssize = ssize > PTLRPC_MAX_BRW_SIZE ?
3114 PTLRPC_MAX_BRW_SIZE : ssize;
3117 do_div(start, ssize);
3118 start = start * ssize;
3120 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3121 ", end "LPU64"\n", *offset, ssize, start, start + ssize - 1);
3122 if (cmd & OBD_CALC_STRIPE_END)
3123 *offset = start + ssize - 1;
3124 else if (cmd & OBD_CALC_STRIPE_START)
3133 struct lov_multi_wait {
3134 struct ldlm_lock *lock;
3140 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3141 struct lustre_handle *lockh)
3143 struct lov_lock_handles *lov_lockh = NULL;
3144 struct lustre_handle *lov_lockhp;
3145 struct lov_obd *lov;
3146 struct lov_oinfo *loi;
3147 struct lov_multi_wait *queues;
3151 ASSERT_LSM_MAGIC(lsm);
3153 if (!exp || !exp->exp_obd)
3156 LASSERT(lockh != NULL);
3157 if (lsm->lsm_stripe_count > 1) {
3158 lov_lockh = lov_handle2llh(lockh);
3159 if (lov_lockh == NULL) {
3160 CERROR("LOV: invalid lov lock handle %p\n", lockh);
3164 lov_lockhp = lov_lockh->llh_handles;
3169 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3171 GOTO(out, rc = -ENOMEM);
3173 lov = &exp->exp_obd->u.lov;
3174 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3175 i++, loi++, lov_lockhp++) {
3176 struct ldlm_lock *lock;
3177 struct obd_device *obd;
3179 lock = ldlm_handle2lock(lov_lockhp);
3181 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3182 loi->loi_ost_idx, loi->loi_id);
3183 queues[i].completed = 1;
3187 queues[i].lock = lock;
3188 init_waitqueue_entry(&(queues[i].wait), current);
3189 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3191 obd = class_exp2obd(lock->l_conn_export);
3193 imp = obd->u.cli.cl_import;
3195 spin_lock(&imp->imp_lock);
3196 queues[i].generation = imp->imp_generation;
3197 spin_unlock(&imp->imp_lock);
3201 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3202 interrupted_completion_wait, &lwd);
3203 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3205 for (i = 0; i < lsm->lsm_stripe_count; i++)
3206 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3208 if (rc == -EINTR || rc == -ETIMEDOUT) {
3214 if (lov_lockh != NULL)
3215 lov_llh_put(lov_lockh);
3220 void lov_stripe_lock(struct lov_stripe_md *md)
3222 LASSERT(md->lsm_lock_owner != cfs_current());
3223 spin_lock(&md->lsm_lock);
3224 LASSERT(md->lsm_lock_owner == NULL);
3225 md->lsm_lock_owner = cfs_current();
3227 EXPORT_SYMBOL(lov_stripe_lock);
3229 void lov_stripe_unlock(struct lov_stripe_md *md)
3231 LASSERT(md->lsm_lock_owner == cfs_current());
3232 md->lsm_lock_owner = NULL;
3233 spin_unlock(&md->lsm_lock);
3235 EXPORT_SYMBOL(lov_stripe_unlock);
3237 static int lov_get_lock(struct obd_export *exp, struct lov_stripe_md *lsm,
3238 void **res, int rw, obd_off start, obd_off end,
3239 struct lustre_handle *lockh, int flags)
3241 struct lov_async_page *l = *res;
3242 obd_off stripe_start, stripe_end = start;
3243 struct lov_lock_handles *lov_lockh = NULL;
3248 if (lockh && lustre_handle_is_used(lockh) &&
3249 !(flags & OBD_FAST_LOCK)) {
3250 lov_lockh = lov_handle2llh(lockh);
3251 if (lov_lockh == NULL) {
3252 CERROR("LOV: invalid lov lock handle %p\n", lockh);
3255 lockh = lov_lockh->llh_handles + l->lap_stripe;
3258 /* ensure we don't cross stripe boundaries */
3259 lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3260 if (stripe_end < end)
3263 /* map the region limits to the object limits */
3264 lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3265 lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3267 rc = obd_get_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3268 lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3269 ltd_exp, NULL, &l->lap_sub_cookie,
3270 rw, stripe_start, stripe_end, lockh, flags);
3272 if (lov_lockh != NULL)
3273 lov_llh_put(lov_lockh);
3277 struct obd_ops lov_obd_ops = {
3278 .o_owner = THIS_MODULE,
3279 .o_setup = lov_setup,
3280 .o_precleanup = lov_precleanup,
3281 .o_cleanup = lov_cleanup,
3282 .o_process_config = lov_process_config,
3283 .o_connect = lov_connect,
3284 .o_disconnect = lov_disconnect,
3285 .o_statfs = lov_statfs,
3286 .o_statfs_async = lov_statfs_async,
3287 .o_packmd = lov_packmd,
3288 .o_unpackmd = lov_unpackmd,
3289 .o_checkmd = lov_checkmd,
3290 .o_create = lov_create,
3291 .o_destroy = lov_destroy,
3292 .o_getattr = lov_getattr,
3293 .o_getattr_async = lov_getattr_async,
3294 .o_setattr = lov_setattr,
3295 .o_setattr_async = lov_setattr_async,
3297 .o_brw_async = lov_brw_async,
3298 .o_prep_async_page = lov_prep_async_page,
3299 .o_get_lock = lov_get_lock,
3300 .o_queue_async_io = lov_queue_async_io,
3301 .o_set_async_flags = lov_set_async_flags,
3302 .o_queue_group_io = lov_queue_group_io,
3303 .o_trigger_group_io = lov_trigger_group_io,
3304 .o_teardown_async_page = lov_teardown_async_page,
3305 .o_merge_lvb = lov_merge_lvb,
3306 .o_update_lvb = lov_update_lvb,
3307 .o_adjust_kms = lov_adjust_kms,
3308 .o_punch = lov_punch,
3310 .o_enqueue = lov_enqueue,
3311 .o_match = lov_match,
3312 .o_change_cbdata = lov_change_cbdata,
3313 .o_cancel = lov_cancel,
3314 .o_cancel_unused = lov_cancel_unused,
3315 .o_join_lru = lov_join_lru,
3316 .o_iocontrol = lov_iocontrol,
3317 .o_get_info = lov_get_info,
3318 .o_set_info_async = lov_set_info_async,
3319 .o_extent_calc = lov_extent_calc,
3320 .o_llog_init = lov_llog_init,
3321 .o_llog_finish = lov_llog_finish,
3322 .o_notify = lov_notify,
3323 .o_register_page_removal_cb = lov_obd_register_page_removal_cb,
3324 .o_unregister_page_removal_cb = lov_obd_unregister_page_removal_cb,
3325 .o_register_lock_cancel_cb = lov_obd_register_lock_cancel_cb,
3326 .o_unregister_lock_cancel_cb = lov_obd_unregister_lock_cancel_cb,
3327 .o_pool_new = lov_pool_new,
3328 .o_pool_rem = lov_pool_remove,
3329 .o_pool_add = lov_pool_add,
3330 .o_pool_del = lov_pool_del,
3331 .o_getref = lov_getref,
3332 .o_putref = lov_putref,
3335 static quota_interface_t *quota_interface;
3336 extern quota_interface_t lov_quota_interface;
3338 cfs_mem_cache_t *lov_oinfo_slab;
3340 int __init lov_init(void)
3342 struct lprocfs_static_vars lvars = { 0 };
3346 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3347 sizeof(struct lov_oinfo),
3348 0, SLAB_HWCACHE_ALIGN);
3349 if (lov_oinfo_slab == NULL)
3351 lprocfs_lov_init_vars(&lvars);
3353 request_module("lquota");
3354 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3355 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3357 rc = class_register_type(&lov_obd_ops, lvars.module_vars,
3360 if (quota_interface)
3361 PORTAL_SYMBOL_PUT(lov_quota_interface);
3362 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3370 static void /*__exit*/ lov_exit(void)
3374 if (quota_interface)
3375 PORTAL_SYMBOL_PUT(lov_quota_interface);
3377 class_unregister_type(LUSTRE_LOV_NAME);
3378 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3382 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3383 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3384 MODULE_LICENSE("GPL");
3386 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);