1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/lov/lov_obd.c
40 * Author: Phil Schwan <phil@clusterfs.com>
41 * Author: Peter Braam <braam@clusterfs.com>
42 * Author: Mike Shaver <shaver@clusterfs.com>
43 * Author: Nathan Rutman <nathan@clusterfs.com>
47 # define EXPORT_SYMTAB
49 #define DEBUG_SUBSYSTEM S_LOV
51 #include <libcfs/libcfs.h>
53 #include <liblustre.h>
56 #include <obd_support.h>
57 #include <lustre_lib.h>
58 #include <lustre_net.h>
59 #include <lustre/lustre_idl.h>
60 #include <lustre_dlm.h>
61 #include <lustre_mds.h>
62 #include <lustre_debug.h>
63 #include <obd_class.h>
66 #include <lprocfs_status.h>
67 #include <lustre_param.h>
68 #include <cl_object.h>
69 #include <lustre/ll_fiemap.h>
71 #include "lov_internal.h"
73 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
74 Any function that expects lov_tgts to remain stationary must take a ref. */
75 static void lov_getref(struct obd_device *obd)
77 struct lov_obd *lov = &obd->u.lov;
79 /* nobody gets through here until lov_putref is done */
80 cfs_mutex_lock(&lov->lov_lock);
81 cfs_atomic_inc(&lov->lov_refcount);
82 cfs_mutex_unlock(&lov->lov_lock);
86 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
88 static void lov_putref(struct obd_device *obd)
90 struct lov_obd *lov = &obd->u.lov;
92 cfs_mutex_lock(&lov->lov_lock);
93 /* ok to dec to 0 more than once -- ltd_exp's will be null */
94 if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
97 struct lov_tgt_desc *tgt, *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 cfs_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 cfs_mutex_unlock(&lov->lov_lock);
115 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
116 cfs_list_del(&tgt->ltd_kill);
118 __lov_del_obd(obd, tgt);
121 cfs_mutex_unlock(&lov->lov_lock);
125 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
126 enum obd_notify_event ev);
127 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
128 enum obd_notify_event ev, void *data);
131 #define MAX_STRING_SIZE 128
132 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
133 struct obd_connect_data *data)
135 struct lov_obd *lov = &obd->u.lov;
136 struct obd_uuid *tgt_uuid;
137 struct obd_device *tgt_obd;
138 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
139 struct obd_import *imp;
141 cfs_proc_dir_entry_t *lov_proc_dir;
146 if (!lov->lov_tgts[index])
149 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
150 tgt_obd = lov->lov_tgts[index]->ltd_obd;
152 if (!tgt_obd->obd_set_up) {
153 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
157 /* override the sp_me from lov */
158 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
160 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
161 data->ocd_index = index;
164 * Divine LOV knows that OBDs under it are OSCs.
166 imp = tgt_obd->u.cli.cl_import;
169 tgt_obd->obd_no_recov = 0;
170 /* FIXME this is probably supposed to be
171 ptlrpc_set_import_active. Horrible naming. */
172 ptlrpc_activate_import(imp);
175 rc = obd_register_observer(tgt_obd, obd);
177 CERROR("Target %s register_observer error %d\n",
178 obd_uuid2str(tgt_uuid), rc);
183 if (imp->imp_invalid) {
184 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
185 "disabled\n", obd_uuid2str(tgt_uuid));
189 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
190 &lov_osc_uuid, data, NULL);
191 if (rc || !lov->lov_tgts[index]->ltd_exp) {
192 CERROR("Target %s connect error %d\n",
193 obd_uuid2str(tgt_uuid), rc);
197 lov->lov_tgts[index]->ltd_reap = 0;
199 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
200 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
203 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
205 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
206 cfs_proc_dir_entry_t *osc_symlink;
208 LASSERT(osc_obd != NULL);
209 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
210 LASSERT(osc_obd->obd_type->typ_name != NULL);
212 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
215 osc_obd->obd_type->typ_name,
217 if (osc_symlink == NULL) {
218 CERROR("could not register LOV target "
219 "/proc/fs/lustre/%s/%s/target_obds/%s.",
220 obd->obd_type->typ_name, obd->obd_name,
222 lprocfs_remove(&lov_proc_dir);
227 rc = qos_add_tgt(obd, index);
229 CERROR("qos_add_tgt failed %d\n", rc);
234 static int lov_connect(const struct lu_env *env,
235 struct obd_export **exp, struct obd_device *obd,
236 struct obd_uuid *cluuid, struct obd_connect_data *data,
239 struct lov_obd *lov = &obd->u.lov;
240 struct lov_tgt_desc *tgt;
241 struct lustre_handle conn;
245 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
247 rc = class_connect(&conn, obd, cluuid);
251 *exp = class_conn2export(&conn);
253 /* Why should there ever be more than 1 connect? */
255 LASSERT(lov->lov_connects == 1);
257 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
259 lov->lov_ocd = *data;
262 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
263 tgt = lov->lov_tgts[i];
264 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
266 /* Flags will be lowest common denominator */
267 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
269 CERROR("%s: lov connect tgt %d failed: %d\n",
270 obd->obd_name, i, rc);
273 /* connect to administrative disabled ost */
274 if (!lov->lov_tgts[i]->ltd_exp)
277 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
278 OBD_NOTIFY_CONNECT, (void *)&i);
280 CERROR("%s error sending notify %d\n",
289 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
291 cfs_proc_dir_entry_t *lov_proc_dir;
292 struct lov_obd *lov = &obd->u.lov;
293 struct obd_device *osc_obd;
297 osc_obd = class_exp2obd(tgt->ltd_exp);
298 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
299 obd->obd_name, osc_obd->obd_name);
301 if (tgt->ltd_active) {
303 lov->desc.ld_active_tgt_count--;
304 tgt->ltd_exp->exp_obd->obd_inactive = 1;
307 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
309 cfs_proc_dir_entry_t *osc_symlink;
311 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
313 lprocfs_remove(&osc_symlink);
315 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
316 obd->obd_type->typ_name, obd->obd_name,
322 /* Pass it on to our clients.
323 * XXX This should be an argument to disconnect,
324 * XXX not a back-door flag on the OBD. Ah well.
326 osc_obd->obd_force = obd->obd_force;
327 osc_obd->obd_fail = obd->obd_fail;
328 osc_obd->obd_no_recov = obd->obd_no_recov;
331 obd_register_observer(osc_obd, NULL);
333 rc = obd_disconnect(tgt->ltd_exp);
335 CERROR("Target %s disconnect error %d\n",
336 tgt->ltd_uuid.uuid, rc);
340 qos_del_tgt(obd, tgt);
346 static int lov_disconnect(struct obd_export *exp)
348 struct obd_device *obd = class_exp2obd(exp);
349 struct lov_obd *lov = &obd->u.lov;
356 /* Only disconnect the underlying layers on the final disconnect. */
358 if (lov->lov_connects != 0) {
359 /* why should there be more than 1 connect? */
360 CERROR("disconnect #%d\n", lov->lov_connects);
364 /* Let's hold another reference so lov_del_obd doesn't spin through
368 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
369 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
370 /* Disconnection is the last we know about an obd */
371 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
377 rc = class_disconnect(exp); /* bz 9811 */
383 * -EINVAL : UUID can't be found in the LOV's target list
384 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
385 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
386 * any >= 0 : is log target index
388 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
389 enum obd_notify_event ev)
391 struct lov_obd *lov = &obd->u.lov;
392 struct lov_tgt_desc *tgt;
393 int index, activate, active;
396 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
397 lov, uuid->uuid, ev);
400 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
401 tgt = lov->lov_tgts[index];
402 if (!tgt || !tgt->ltd_exp)
405 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
406 index, obd_uuid2str(&tgt->ltd_uuid),
407 tgt->ltd_exp->exp_handle.h_cookie);
408 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
412 if (index == lov->desc.ld_tgt_count)
413 GOTO(out, index = -EINVAL);
415 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
416 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
418 if (lov->lov_tgts[index]->ltd_activate == activate) {
419 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
420 uuid->uuid, activate ? "" : "de");
422 lov->lov_tgts[index]->ltd_activate = activate;
423 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
424 activate ? "" : "de", obd_uuid2str(uuid));
427 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
428 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
430 if (lov->lov_tgts[index]->ltd_active == active) {
431 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
432 uuid->uuid, active ? "" : "in");
435 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
436 obd_uuid2str(uuid), active ? "" : "in");
439 lov->lov_tgts[index]->ltd_active = active;
441 lov->desc.ld_active_tgt_count++;
442 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
444 lov->desc.ld_active_tgt_count--;
445 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
447 /* remove any old qos penalty */
448 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
450 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
458 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
459 enum obd_notify_event ev, void *data)
464 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
465 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
466 struct obd_uuid *uuid;
470 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
471 CERROR("unexpected notification of %s %s!\n",
472 watched->obd_type->typ_name,
476 uuid = &watched->u.cli.cl_target_uuid;
478 /* Set OSC as active before notifying the observer, so the
479 * observer can use the OSC normally.
481 rc = lov_set_osc_active(obd, uuid, ev);
483 CERROR("event(%d) of %s failed: %d\n", ev,
484 obd_uuid2str(uuid), rc);
487 /* active event should be pass lov target index as data */
491 /* Pass the notification up the chain. */
493 rc = obd_notify_observer(obd, watched, ev, data);
495 /* NULL watched means all osc's in the lov (only for syncs) */
496 /* sync event should be send lov idx as data */
497 struct lov_obd *lov = &obd->u.lov;
501 is_sync = (ev == OBD_NOTIFY_SYNC) ||
502 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
505 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
506 if (!lov->lov_tgts[i])
509 /* don't send sync event if target not
510 * connected/activated */
511 if (is_sync && !lov->lov_tgts[i]->ltd_active)
514 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
517 CERROR("%s: notify %s of %s failed %d\n",
519 obd->obd_observer->obd_name,
520 lov->lov_tgts[i]->ltd_obd->obd_name,
530 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
531 __u32 index, int gen, int active)
533 struct lov_obd *lov = &obd->u.lov;
534 struct lov_tgt_desc *tgt;
535 struct obd_device *tgt_obd;
539 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
540 uuidp->uuid, index, gen, active);
543 CERROR("request to add OBD %s with invalid generation: %d\n",
548 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
553 cfs_mutex_lock(&lov->lov_lock);
555 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
556 tgt = lov->lov_tgts[index];
557 CERROR("UUID %s already assigned at LOV target index %d\n",
558 obd_uuid2str(&tgt->ltd_uuid), index);
559 cfs_mutex_unlock(&lov->lov_lock);
563 if (index >= lov->lov_tgt_size) {
564 /* We need to reallocate the lov target array. */
565 struct lov_tgt_desc **newtgts, **old = NULL;
566 __u32 newsize, oldsize = 0;
568 newsize = max(lov->lov_tgt_size, (__u32)2);
569 while (newsize < index + 1)
570 newsize = newsize << 1;
571 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
572 if (newtgts == NULL) {
573 cfs_mutex_unlock(&lov->lov_lock);
577 if (lov->lov_tgt_size) {
578 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
581 oldsize = lov->lov_tgt_size;
584 lov->lov_tgts = newtgts;
585 lov->lov_tgt_size = newsize;
590 OBD_FREE(old, sizeof(*old) * oldsize);
592 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
593 lov->lov_tgts, lov->lov_tgt_size);
598 cfs_mutex_unlock(&lov->lov_lock);
602 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
604 cfs_mutex_unlock(&lov->lov_lock);
609 tgt->ltd_uuid = *uuidp;
610 tgt->ltd_obd = tgt_obd;
611 /* XXX - add a sanity check on the generation number. */
613 tgt->ltd_index = index;
614 tgt->ltd_activate = active;
615 lov->lov_tgts[index] = tgt;
616 if (index >= lov->desc.ld_tgt_count)
617 lov->desc.ld_tgt_count = index + 1;
619 cfs_mutex_unlock(&lov->lov_lock);
621 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
622 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
624 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
626 if (lov->lov_connects == 0) {
627 /* lov_connect hasn't been called yet. We'll do the
628 lov_connect_obd on this target when that fn first runs,
629 because we don't know the connect flags yet. */
635 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
639 /* connect to administrative disabled ost */
643 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
644 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
649 CERROR("add failed (%d), deleting %s\n", rc,
650 obd_uuid2str(&tgt->ltd_uuid));
651 lov_del_target(obd, index, 0, 0);
657 /* Schedule a target for deletion */
658 int lov_del_target(struct obd_device *obd, __u32 index,
659 struct obd_uuid *uuidp, int gen)
661 struct lov_obd *lov = &obd->u.lov;
662 int count = lov->desc.ld_tgt_count;
666 if (index >= count) {
667 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
674 if (!lov->lov_tgts[index]) {
675 CERROR("LOV target at index %d is not setup.\n", index);
676 GOTO(out, rc = -EINVAL);
679 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
680 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
681 lov_uuid2str(lov, index), index,
682 obd_uuid2str(uuidp));
683 GOTO(out, rc = -EINVAL);
686 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
687 lov_uuid2str(lov, index), index,
688 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
689 lov->lov_tgts[index]->ltd_active);
691 lov->lov_tgts[index]->ltd_reap = 1;
692 lov->lov_death_row++;
693 /* we really delete it from obd_putref */
700 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
702 struct obd_device *osc_obd;
705 LASSERT(tgt->ltd_reap);
707 osc_obd = class_exp2obd(tgt->ltd_exp);
709 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
711 osc_obd ? osc_obd->obd_name : "<no obd>");
714 lov_disconnect_obd(obd, tgt);
718 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
719 do it ourselves. And we can't do it from lov_cleanup,
720 because we just lost our only reference to it. */
722 class_manual_cleanup(osc_obd);
725 void lov_fix_desc_stripe_size(__u64 *val)
727 if (*val < PTLRPC_MAX_BRW_SIZE) {
728 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
729 PTLRPC_MAX_BRW_SIZE);
730 *val = PTLRPC_MAX_BRW_SIZE;
731 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
732 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
733 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
735 *val, LOV_MIN_STRIPE_SIZE);
739 void lov_fix_desc_stripe_count(__u32 *val)
745 void lov_fix_desc_pattern(__u32 *val)
747 /* from lov_setstripe */
748 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
749 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
754 void lov_fix_desc_qos_maxage(__u32 *val)
758 *val = QOS_DEFAULT_MAXAGE;
761 void lov_fix_desc(struct lov_desc *desc)
763 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
764 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
765 lov_fix_desc_pattern(&desc->ld_pattern);
766 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
769 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
771 struct lprocfs_static_vars lvars = { 0 };
772 struct lov_desc *desc;
773 struct lov_obd *lov = &obd->u.lov;
777 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
778 CERROR("LOV setup requires a descriptor\n");
782 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
784 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
785 CERROR("descriptor size wrong: %d > %d\n",
786 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
790 if (desc->ld_magic != LOV_DESC_MAGIC) {
791 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
792 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
793 obd->obd_name, desc);
794 lustre_swab_lov_desc(desc);
796 CERROR("%s: Bad lov desc magic: %#x\n",
797 obd->obd_name, desc->ld_magic);
804 desc->ld_active_tgt_count = 0;
806 lov->lov_tgt_size = 0;
808 cfs_mutex_init(&lov->lov_lock);
809 cfs_atomic_set(&lov->lov_refcount, 0);
810 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
811 cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
812 lov->lov_sp_me = LUSTRE_SP_CLI;
813 lov->lov_qos.lq_dirty = 1;
814 lov->lov_qos.lq_rr.lqr_dirty = 1;
815 lov->lov_qos.lq_reset = 1;
816 /* Default priority is toward free space balance */
817 lov->lov_qos.lq_prio_free = 232;
818 /* Default threshold for rr (roughly 17%) */
819 lov->lov_qos.lq_threshold_rr = 43;
820 /* Init statfs fields */
821 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
822 if (NULL == lov->lov_qos.lq_statfs_data)
824 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
826 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
828 HASH_POOLS_BKT_BITS, 0,
831 &pool_hash_operations,
833 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
834 lov->lov_pool_count = 0;
835 rc = lov_ost_pool_init(&lov->lov_packed, 0);
837 GOTO(out_free_statfs, rc);
838 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
840 GOTO(out_free_lov_packed, rc);
842 lprocfs_lov_init_vars(&lvars);
843 lprocfs_obd_setup(obd, lvars.obd_vars);
848 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
849 0444, &lov_proc_target_fops, obd);
851 CWARN("Error adding the target_obd file\n");
854 lov->lov_pool_proc_entry = lprocfs_register("pools",
861 lov_ost_pool_free(&lov->lov_packed);
863 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
867 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
870 struct lov_obd *lov = &obd->u.lov;
875 case OBD_CLEANUP_EARLY: {
877 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
878 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
880 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
885 case OBD_CLEANUP_EXPORTS:
886 lprocfs_obd_cleanup(obd);
887 rc = obd_llog_finish(obd, 0);
889 CERROR("failed to cleanup llogging subsystems\n");
895 static int lov_cleanup(struct obd_device *obd)
897 struct lov_obd *lov = &obd->u.lov;
898 cfs_list_t *pos, *tmp;
899 struct pool_desc *pool;
902 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
903 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
904 /* free pool structs */
905 CDEBUG(D_INFO, "delete pool %p\n", pool);
906 lov_pool_del(obd, pool->pool_name);
908 cfs_hash_putref(lov->lov_pools_hash_body);
909 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
910 lov_ost_pool_free(&lov->lov_packed);
915 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
916 if (!lov->lov_tgts[i])
919 /* Inactive targets may never have connected */
920 if (lov->lov_tgts[i]->ltd_active ||
921 cfs_atomic_read(&lov->lov_refcount))
922 /* We should never get here - these
923 should have been removed in the
925 CERROR("lov tgt %d not cleaned!"
926 " deathrow=%d, lovrc=%d\n",
927 i, lov->lov_death_row,
928 cfs_atomic_read(&lov->lov_refcount));
929 lov_del_target(obd, i, 0, 0);
932 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
934 lov->lov_tgt_size = 0;
936 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
940 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
941 __u32 *indexp, int *genp)
943 struct obd_uuid obd_uuid;
948 switch(cmd = lcfg->lcfg_command) {
949 case LCFG_LOV_ADD_OBD:
950 case LCFG_LOV_ADD_INA:
951 case LCFG_LOV_DEL_OBD: {
954 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
955 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
956 GOTO(out, rc = -EINVAL);
958 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
960 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
961 GOTO(out, rc = -EINVAL);
962 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
963 GOTO(out, rc = -EINVAL);
966 if (cmd == LCFG_LOV_ADD_OBD)
967 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
968 else if (cmd == LCFG_LOV_ADD_INA)
969 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
971 rc = lov_del_target(obd, index, &obd_uuid, gen);
975 struct lprocfs_static_vars lvars = { 0 };
976 struct lov_desc *desc = &(obd->u.lov.desc);
979 GOTO(out, rc = -EINVAL);
981 lprocfs_lov_init_vars(&lvars);
983 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
996 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
997 GOTO(out, rc = -EINVAL);
1006 #define log2(n) cfs_ffz(~(n))
1009 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1010 struct lov_stripe_md **ea,
1011 struct obd_trans_info *oti)
1013 struct lov_obd *lov;
1014 struct obdo *tmp_oa;
1015 struct obd_uuid *ost_uuid = NULL;
1019 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1020 src_oa->o_flags == OBD_FL_DELORPHAN);
1022 lov = &export->exp_obd->u.lov;
1028 if (oti->oti_ost_uuid) {
1029 ost_uuid = oti->oti_ost_uuid;
1030 CDEBUG(D_HA, "clearing orphans only for %s\n",
1034 obd_getref(export->exp_obd);
1035 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1036 struct lov_stripe_md obj_md;
1037 struct lov_stripe_md *obj_mdp = &obj_md;
1038 struct lov_tgt_desc *tgt;
1041 tgt = lov->lov_tgts[i];
1045 /* if called for a specific target, we don't
1046 care if it is not active. */
1047 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1048 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1052 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1055 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1056 obd_uuid2str(ost_uuid));
1058 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1060 LASSERT(lov->lov_tgts[i]->ltd_exp);
1061 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1062 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1063 tmp_oa, &obj_mdp, oti);
1065 /* This export will be disabled until it is recovered,
1066 and then orphan recovery will be completed. */
1067 CERROR("error in orphan recovery on OST idx %d/%d: "
1068 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1075 obd_putref(export->exp_obd);
1081 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1082 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1084 struct lov_stripe_md *obj_mdp, *lsm;
1085 struct lov_obd *lov = &exp->exp_obd->u.lov;
1090 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1091 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1093 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1094 if (obj_mdp == NULL)
1097 ost_idx = src_oa->o_nlink;
1100 GOTO(out, rc = -EINVAL);
1101 if (ost_idx >= lov->desc.ld_tgt_count ||
1102 !lov->lov_tgts[ost_idx])
1103 GOTO(out, rc = -EINVAL);
1105 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1106 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1107 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1108 GOTO(out, rc = -EINVAL);
1112 if (i == lsm->lsm_stripe_count)
1113 GOTO(out, rc = -EINVAL);
1115 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1117 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1121 /* the LOV expects oa->o_id to be set to the LOV object id */
1122 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1123 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1125 struct lov_obd *lov;
1126 struct obd_info oinfo;
1127 struct lov_request_set *set = NULL;
1128 struct lov_request *req;
1129 struct l_wait_info lwi = { 0 };
1133 LASSERT(ea != NULL);
1137 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1138 src_oa->o_flags == OBD_FL_DELORPHAN) {
1139 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1143 lov = &exp->exp_obd->u.lov;
1144 if (!lov->desc.ld_active_tgt_count)
1147 obd_getref(exp->exp_obd);
1148 /* Recreate a specific object id at the given OST index */
1149 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1150 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1151 rc = lov_recreate(exp, src_oa, ea, oti);
1155 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1156 * later in alloc_qos(), we will wait for those rpcs to complete if
1157 * the osfs age is older than 2 * qos_maxage */
1158 qos_statfs_update(exp->exp_obd,
1159 cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1160 OBD_STATFS_CACHE_SECONDS),
1163 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1167 cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1168 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1169 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1170 &req->rq_oi, &req->rq_oi.oi_md, oti);
1173 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1174 * longer then this */
1175 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1177 /* we not have ptlrpc set for assign set->interpret and should
1178 * be call interpret function himself. calling from cb_create_update
1179 * not permited because lov_fini_create_set can sleep for long time,
1180 * but we must avoid sleeping in ptlrpcd interpret function. */
1181 rc = lov_fini_create_set(set, ea);
1183 obd_putref(exp->exp_obd);
1187 #define ASSERT_LSM_MAGIC(lsmp) \
1189 LASSERT((lsmp) != NULL); \
1190 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1191 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1192 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1195 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1196 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1197 struct obd_export *md_exp, void *capa)
1199 struct lov_request_set *set;
1200 struct obd_info oinfo;
1201 struct lov_request *req;
1203 struct lov_obd *lov;
1204 int rc = 0, err = 0;
1207 ASSERT_LSM_MAGIC(lsm);
1209 if (!exp || !exp->exp_obd)
1212 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1214 LASSERT(oti->oti_logcookies);
1217 lov = &exp->exp_obd->u.lov;
1218 obd_getref(exp->exp_obd);
1219 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1223 cfs_list_for_each (pos, &set->set_list) {
1224 req = cfs_list_entry(pos, struct lov_request, rq_link);
1226 if (oa->o_valid & OBD_MD_FLCOOKIE)
1227 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1229 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1230 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1231 err = lov_update_common_set(set, req, err);
1233 CERROR("error: destroying objid "LPX64" subobj "
1234 LPX64" on OST idx %d: rc = %d\n",
1235 oa->o_id, req->rq_oi.oi_oa->o_id,
1243 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1244 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1246 err = lov_fini_destroy_set(set);
1248 obd_putref(exp->exp_obd);
1249 RETURN(rc ? rc : err);
1252 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1254 struct lov_request_set *set;
1255 struct lov_request *req;
1257 struct lov_obd *lov;
1258 int err = 0, rc = 0;
1262 ASSERT_LSM_MAGIC(oinfo->oi_md);
1264 if (!exp || !exp->exp_obd)
1267 lov = &exp->exp_obd->u.lov;
1269 rc = lov_prep_getattr_set(exp, oinfo, &set);
1273 cfs_list_for_each (pos, &set->set_list) {
1274 req = cfs_list_entry(pos, struct lov_request, rq_link);
1276 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1277 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1278 req->rq_oi.oi_oa->o_id, req->rq_idx);
1280 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1282 err = lov_update_common_set(set, req, rc);
1284 CERROR("error: getattr objid "LPX64" subobj "
1285 LPX64" on OST idx %d: rc = %d\n",
1286 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1292 rc = lov_fini_getattr_set(set);
1298 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1301 struct lov_request_set *lovset = (struct lov_request_set *)data;
1305 /* don't do attribute merge if this aysnc op failed */
1307 cfs_atomic_set(&lovset->set_completes, 0);
1308 err = lov_fini_getattr_set(lovset);
1309 RETURN(rc ? rc : err);
1312 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1313 struct ptlrpc_request_set *rqset)
1315 struct lov_request_set *lovset;
1316 struct lov_obd *lov;
1318 struct lov_request *req;
1323 ASSERT_LSM_MAGIC(oinfo->oi_md);
1325 if (!exp || !exp->exp_obd)
1328 lov = &exp->exp_obd->u.lov;
1330 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1334 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1335 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1336 oinfo->oi_md->lsm_stripe_size);
1338 cfs_list_for_each (pos, &lovset->set_list) {
1339 req = cfs_list_entry(pos, struct lov_request, rq_link);
1341 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1342 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1343 req->rq_oi.oi_oa->o_id, req->rq_idx);
1344 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1345 &req->rq_oi, rqset);
1347 CERROR("error: getattr objid "LPX64" subobj "
1348 LPX64" on OST idx %d: rc = %d\n",
1349 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1355 if (!cfs_list_empty(&rqset->set_requests)) {
1357 LASSERT (rqset->set_interpret == NULL);
1358 rqset->set_interpret = lov_getattr_interpret;
1359 rqset->set_arg = (void *)lovset;
1364 cfs_atomic_set(&lovset->set_completes, 0);
1365 err = lov_fini_getattr_set(lovset);
1366 RETURN(rc ? rc : err);
1369 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1370 struct obd_trans_info *oti)
1372 struct lov_request_set *set;
1373 struct lov_obd *lov;
1375 struct lov_request *req;
1376 int err = 0, rc = 0;
1380 ASSERT_LSM_MAGIC(oinfo->oi_md);
1382 if (!exp || !exp->exp_obd)
1385 /* for now, we only expect the following updates here */
1386 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1387 OBD_MD_FLMODE | OBD_MD_FLATIME |
1388 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1389 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1390 OBD_MD_FLGROUP | OBD_MD_FLUID |
1391 OBD_MD_FLGID | OBD_MD_FLFID |
1393 lov = &exp->exp_obd->u.lov;
1394 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1398 cfs_list_for_each (pos, &set->set_list) {
1399 req = cfs_list_entry(pos, struct lov_request, rq_link);
1401 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1403 err = lov_update_setattr_set(set, req, rc);
1405 CERROR("error: setattr objid "LPX64" subobj "
1406 LPX64" on OST idx %d: rc = %d\n",
1407 set->set_oi->oi_oa->o_id,
1408 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1413 err = lov_fini_setattr_set(set);
1419 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1422 struct lov_request_set *lovset = (struct lov_request_set *)data;
1427 cfs_atomic_set(&lovset->set_completes, 0);
1428 err = lov_fini_setattr_set(lovset);
1429 RETURN(rc ? rc : err);
1432 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1433 needed. Otherwise, a client is waiting for responses. */
1434 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1435 struct obd_trans_info *oti,
1436 struct ptlrpc_request_set *rqset)
1438 struct lov_request_set *set;
1439 struct lov_request *req;
1441 struct lov_obd *lov;
1446 ASSERT_LSM_MAGIC(oinfo->oi_md);
1447 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1449 LASSERT(oti->oti_logcookies);
1452 if (!exp || !exp->exp_obd)
1455 lov = &exp->exp_obd->u.lov;
1456 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1460 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1461 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1462 oinfo->oi_md->lsm_stripe_size);
1464 cfs_list_for_each (pos, &set->set_list) {
1465 req = cfs_list_entry(pos, struct lov_request, rq_link);
1467 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1468 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1470 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1471 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1472 req->rq_oi.oi_oa->o_id, req->rq_idx);
1474 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1475 &req->rq_oi, oti, rqset);
1477 CERROR("error: setattr objid "LPX64" subobj "
1478 LPX64" on OST idx %d: rc = %d\n",
1479 set->set_oi->oi_oa->o_id,
1480 req->rq_oi.oi_oa->o_id,
1486 /* If we are not waiting for responses on async requests, return. */
1487 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1490 cfs_atomic_set(&set->set_completes, 0);
1491 err = lov_fini_setattr_set(set);
1492 RETURN(rc ? rc : err);
1495 LASSERT(rqset->set_interpret == NULL);
1496 rqset->set_interpret = lov_setattr_interpret;
1497 rqset->set_arg = (void *)set;
1502 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1505 struct lov_request_set *lovset = (struct lov_request_set *)data;
1510 cfs_atomic_set(&lovset->set_completes, 0);
1511 err = lov_fini_punch_set(lovset);
1512 RETURN(rc ? rc : err);
1515 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1516 * we can send this 'punch' to just the authoritative node and the nodes
1517 * that the punch will affect. */
1518 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1519 struct obd_trans_info *oti,
1520 struct ptlrpc_request_set *rqset)
1522 struct lov_request_set *set;
1523 struct lov_obd *lov;
1525 struct lov_request *req;
1530 ASSERT_LSM_MAGIC(oinfo->oi_md);
1532 if (!exp || !exp->exp_obd)
1535 lov = &exp->exp_obd->u.lov;
1536 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1540 cfs_list_for_each (pos, &set->set_list) {
1541 req = cfs_list_entry(pos, struct lov_request, rq_link);
1543 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1544 &req->rq_oi, NULL, rqset);
1546 CERROR("error: punch objid "LPX64" subobj "LPX64
1547 " on OST idx %d: rc = %d\n",
1548 set->set_oi->oi_oa->o_id,
1549 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1554 if (rc || cfs_list_empty(&rqset->set_requests)) {
1556 err = lov_fini_punch_set(set);
1557 RETURN(rc ? rc : err);
1560 LASSERT(rqset->set_interpret == NULL);
1561 rqset->set_interpret = lov_punch_interpret;
1562 rqset->set_arg = (void *)set;
1567 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1570 struct lov_request_set *lovset = data;
1575 cfs_atomic_set(&lovset->set_completes, 0);
1576 err = lov_fini_sync_set(lovset);
1580 static int lov_sync(struct obd_export *exp, struct obd_info *oinfo,
1581 obd_off start, obd_off end,
1582 struct ptlrpc_request_set *rqset)
1584 struct lov_request_set *set = NULL;
1585 struct lov_obd *lov;
1587 struct lov_request *req;
1591 ASSERT_LSM_MAGIC(oinfo->oi_md);
1592 LASSERT(rqset != NULL);
1597 lov = &exp->exp_obd->u.lov;
1598 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1602 CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1603 set->set_oi->oi_oa->o_id, start, end);
1605 cfs_list_for_each (pos, &set->set_list) {
1606 req = cfs_list_entry(pos, struct lov_request, rq_link);
1608 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi,
1609 req->rq_oi.oi_policy.l_extent.start,
1610 req->rq_oi.oi_policy.l_extent.end, rqset);
1612 CERROR("error: fsync objid "LPX64" subobj "LPX64
1613 " on OST idx %d: rc = %d\n",
1614 set->set_oi->oi_oa->o_id,
1615 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1620 /* If we are not waiting for responses on async requests, return. */
1621 if (rc || cfs_list_empty(&rqset->set_requests)) {
1622 int err = lov_fini_sync_set(set);
1627 LASSERT(rqset->set_interpret == NULL);
1628 rqset->set_interpret = lov_sync_interpret;
1629 rqset->set_arg = (void *)set;
1634 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1635 obd_count oa_bufs, struct brw_page *pga)
1637 struct obd_info oinfo = { { { 0 } } };
1640 oinfo.oi_oa = lov_oinfo->oi_oa;
1642 /* The caller just wants to know if there's a chance that this
1643 * I/O can succeed */
1644 for (i = 0; i < oa_bufs; i++) {
1645 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1646 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1649 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1650 pga[i].off + pga[i].count - 1,
1654 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1655 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1659 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1667 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1668 obd_count oa_bufs, struct brw_page *pga,
1669 struct obd_trans_info *oti)
1671 struct lov_request_set *set;
1672 struct lov_request *req;
1674 struct lov_obd *lov = &exp->exp_obd->u.lov;
1678 ASSERT_LSM_MAGIC(oinfo->oi_md);
1680 if (cmd == OBD_BRW_CHECK) {
1681 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1685 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1689 cfs_list_for_each (pos, &set->set_list) {
1690 struct obd_export *sub_exp;
1691 struct brw_page *sub_pga;
1692 req = cfs_list_entry(pos, struct lov_request, rq_link);
1694 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1695 sub_pga = set->set_pga + req->rq_pgaidx;
1696 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1700 lov_update_common_set(set, req, rc);
1703 err = lov_fini_brw_set(set);
1709 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1712 struct lov_request_set *lovset = (struct lov_request_set *)data;
1714 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1718 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1719 struct ldlm_enqueue_info *einfo,
1720 struct ptlrpc_request_set *rqset)
1722 ldlm_mode_t mode = einfo->ei_mode;
1723 struct lov_request_set *set;
1724 struct lov_request *req;
1726 struct lov_obd *lov;
1731 ASSERT_LSM_MAGIC(oinfo->oi_md);
1732 LASSERT(mode == (mode & -mode));
1734 /* we should never be asked to replay a lock this way. */
1735 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1737 if (!exp || !exp->exp_obd)
1740 lov = &exp->exp_obd->u.lov;
1741 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1745 cfs_list_for_each (pos, &set->set_list) {
1746 req = cfs_list_entry(pos, struct lov_request, rq_link);
1748 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1749 &req->rq_oi, einfo, rqset);
1754 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1756 LASSERT(rqset->set_interpret == NULL);
1757 rqset->set_interpret = lov_enqueue_interpret;
1758 rqset->set_arg = (void *)set;
1762 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1766 static int lov_change_cbdata(struct obd_export *exp,
1767 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1770 struct lov_obd *lov;
1774 ASSERT_LSM_MAGIC(lsm);
1776 if (!exp || !exp->exp_obd)
1779 lov = &exp->exp_obd->u.lov;
1780 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1781 struct lov_stripe_md submd;
1782 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1784 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1785 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1789 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1790 submd.lsm_object_id = loi->loi_id;
1791 submd.lsm_object_seq = loi->loi_seq;
1792 submd.lsm_stripe_count = 0;
1793 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1799 /* find any ldlm lock of the inode in lov
1803 static int lov_find_cbdata(struct obd_export *exp,
1804 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1807 struct lov_obd *lov;
1811 ASSERT_LSM_MAGIC(lsm);
1813 if (!exp || !exp->exp_obd)
1816 lov = &exp->exp_obd->u.lov;
1817 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1818 struct lov_stripe_md submd;
1819 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1821 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1822 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1826 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1827 submd.lsm_object_id = loi->loi_id;
1828 submd.lsm_object_seq = loi->loi_seq;
1829 submd.lsm_stripe_count = 0;
1830 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1838 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1839 __u32 mode, struct lustre_handle *lockh)
1841 struct lov_request_set *set;
1842 struct obd_info oinfo;
1843 struct lov_request *req;
1845 struct lov_obd *lov;
1846 struct lustre_handle *lov_lockhp;
1847 int err = 0, rc = 0;
1850 ASSERT_LSM_MAGIC(lsm);
1852 if (!exp || !exp->exp_obd)
1855 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1857 lov = &exp->exp_obd->u.lov;
1858 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1862 cfs_list_for_each (pos, &set->set_list) {
1863 req = cfs_list_entry(pos, struct lov_request, rq_link);
1864 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1866 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1867 req->rq_oi.oi_md, mode, lov_lockhp);
1868 rc = lov_update_common_set(set, req, rc);
1870 CERROR("error: cancel objid "LPX64" subobj "
1871 LPX64" on OST idx %d: rc = %d\n",
1873 req->rq_oi.oi_md->lsm_object_id,
1879 lov_fini_cancel_set(set);
1883 static int lov_cancel_unused(struct obd_export *exp,
1884 struct lov_stripe_md *lsm,
1885 ldlm_cancel_flags_t flags, void *opaque)
1887 struct lov_obd *lov;
1891 if (!exp || !exp->exp_obd)
1894 lov = &exp->exp_obd->u.lov;
1896 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1898 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1901 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1909 ASSERT_LSM_MAGIC(lsm);
1911 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1912 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1913 struct lov_stripe_md submd;
1914 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1917 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1918 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1922 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1923 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1925 submd.lsm_object_id = loi->loi_id;
1926 submd.lsm_object_seq = loi->loi_seq;
1927 submd.lsm_stripe_count = 0;
1928 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1929 &submd, flags, opaque);
1930 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1931 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1932 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1933 loi->loi_id, loi->loi_ost_idx, err);
1941 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1943 struct lov_request_set *lovset = (struct lov_request_set *)data;
1948 cfs_atomic_set(&lovset->set_completes, 0);
1950 err = lov_fini_statfs_set(lovset);
1951 RETURN(rc ? rc : err);
1954 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1955 __u64 max_age, struct ptlrpc_request_set *rqset)
1957 struct lov_request_set *set;
1958 struct lov_request *req;
1960 struct lov_obd *lov;
1964 LASSERT(oinfo != NULL);
1965 LASSERT(oinfo->oi_osfs != NULL);
1968 rc = lov_prep_statfs_set(obd, oinfo, &set);
1972 cfs_list_for_each (pos, &set->set_list) {
1973 struct obd_device *osc_obd;
1975 req = cfs_list_entry(pos, struct lov_request, rq_link);
1977 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1978 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1983 if (rc || cfs_list_empty(&rqset->set_requests)) {
1986 cfs_atomic_set(&set->set_completes, 0);
1987 err = lov_fini_statfs_set(set);
1988 RETURN(rc ? rc : err);
1991 LASSERT(rqset->set_interpret == NULL);
1992 rqset->set_interpret = lov_statfs_interpret;
1993 rqset->set_arg = (void *)set;
1997 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1998 __u64 max_age, __u32 flags)
2000 struct ptlrpc_request_set *set = NULL;
2001 struct obd_info oinfo = { { { 0 } } };
2006 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2007 * statfs requests */
2008 set = ptlrpc_prep_set();
2012 oinfo.oi_osfs = osfs;
2013 oinfo.oi_flags = flags;
2014 rc = lov_statfs_async(obd, &oinfo, max_age, set);
2016 rc = ptlrpc_set_wait(set);
2017 ptlrpc_set_destroy(set);
2022 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2023 void *karg, void *uarg)
2025 struct obd_device *obddev = class_exp2obd(exp);
2026 struct lov_obd *lov = &obddev->u.lov;
2027 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
2028 struct obd_uuid *uuidp;
2032 case IOC_OBD_STATFS: {
2033 struct obd_ioctl_data *data = karg;
2034 struct obd_device *osc_obd;
2035 struct obd_statfs stat_buf = {0};
2038 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2039 if ((index >= count))
2042 if (!lov->lov_tgts[index])
2043 /* Try again with the next index */
2045 if (!lov->lov_tgts[index]->ltd_active)
2048 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2053 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2054 min((int) data->ioc_plen2,
2055 (int) sizeof(struct obd_uuid))))
2058 /* got statfs data */
2059 rc = obd_statfs(osc_obd, &stat_buf,
2060 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2064 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2065 min((int) data->ioc_plen1,
2066 (int) sizeof(stat_buf))))
2070 case OBD_IOC_LOV_GET_CONFIG: {
2071 struct obd_ioctl_data *data;
2072 struct lov_desc *desc;
2077 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2080 data = (struct obd_ioctl_data *)buf;
2082 if (sizeof(*desc) > data->ioc_inllen1) {
2083 obd_ioctl_freedata(buf, len);
2087 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2088 obd_ioctl_freedata(buf, len);
2092 if (sizeof(__u32) * count > data->ioc_inllen3) {
2093 obd_ioctl_freedata(buf, len);
2097 desc = (struct lov_desc *)data->ioc_inlbuf1;
2098 memcpy(desc, &(lov->desc), sizeof(*desc));
2100 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2101 genp = (__u32 *)data->ioc_inlbuf3;
2102 /* the uuid will be empty for deleted OSTs */
2103 for (i = 0; i < count; i++, uuidp++, genp++) {
2104 if (!lov->lov_tgts[i])
2106 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2107 *genp = lov->lov_tgts[i]->ltd_gen;
2110 if (cfs_copy_to_user((void *)uarg, buf, len))
2112 obd_ioctl_freedata(buf, len);
2115 case LL_IOC_LOV_SETSTRIPE:
2116 rc = lov_setstripe(exp, len, karg, uarg);
2118 case LL_IOC_LOV_GETSTRIPE:
2119 rc = lov_getstripe(exp, karg, uarg);
2121 case LL_IOC_LOV_SETEA:
2122 rc = lov_setea(exp, karg, uarg);
2124 case OBD_IOC_QUOTACTL: {
2125 struct if_quotactl *qctl = karg;
2126 struct lov_tgt_desc *tgt = NULL;
2127 struct obd_quotactl *oqctl;
2129 if (qctl->qc_valid == QC_OSTIDX) {
2130 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2133 tgt = lov->lov_tgts[qctl->qc_idx];
2134 if (!tgt || !tgt->ltd_exp)
2136 } else if (qctl->qc_valid == QC_UUID) {
2137 for (i = 0; i < count; i++) {
2138 tgt = lov->lov_tgts[i];
2140 !obd_uuid_equals(&tgt->ltd_uuid,
2144 if (tgt->ltd_exp == NULL)
2156 LASSERT(tgt && tgt->ltd_exp);
2157 OBD_ALLOC_PTR(oqctl);
2161 QCTL_COPY(oqctl, qctl);
2162 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2164 QCTL_COPY(qctl, oqctl);
2165 qctl->qc_valid = QC_OSTIDX;
2166 qctl->obd_uuid = tgt->ltd_uuid;
2168 OBD_FREE_PTR(oqctl);
2177 for (i = 0; i < count; i++) {
2179 struct obd_device *osc_obd;
2181 /* OST was disconnected */
2182 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2185 /* ll_umount_begin() sets force flag but for lov, not
2186 * osc. Let's pass it through */
2187 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2188 osc_obd->obd_force = obddev->obd_force;
2189 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2191 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2194 if (lov->lov_tgts[i]->ltd_active) {
2195 CDEBUG(err == -ENOTTY ?
2196 D_IOCTL : D_WARNING,
2197 "iocontrol OSC %s on OST "
2198 "idx %d cmd %x: err = %d\n",
2199 lov_uuid2str(lov, i),
2216 #define FIEMAP_BUFFER_SIZE 4096
2219 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2220 * call. The local end offset and the device are sent in the first
2221 * fm_extent. This function calculates the stripe number from the index.
2222 * This function returns a stripe_no on which mapping is to be restarted.
2224 * This function returns fm_end_offset which is the in-OST offset at which
2225 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2226 * will re-calculate proper offset in next stripe.
2227 * Note that the first extent is passed to lov_get_info via the value field.
2229 * \param fiemap fiemap request header
2230 * \param lsm striping information for the file
2231 * \param fm_start logical start of mapping
2232 * \param fm_end logical end of mapping
2233 * \param start_stripe starting stripe will be returned in this
2235 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2236 struct lov_stripe_md *lsm, obd_size fm_start,
2237 obd_size fm_end, int *start_stripe)
2239 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2240 obd_off lun_start, lun_end;
2241 obd_size fm_end_offset;
2242 int stripe_no = -1, i;
2244 if (fiemap->fm_extent_count == 0 ||
2245 fiemap->fm_extents[0].fe_logical == 0)
2248 /* Find out stripe_no from ost_index saved in the fe_device */
2249 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2250 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2251 fiemap->fm_extents[0].fe_device) {
2257 /* If we have finished mapping on previous device, shift logical
2258 * offset to start of next device */
2259 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2260 &lun_start, &lun_end)) != 0 &&
2261 local_end < lun_end) {
2262 fm_end_offset = local_end;
2263 *start_stripe = stripe_no;
2265 /* This is a special value to indicate that caller should
2266 * calculate offset in next stripe. */
2268 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2271 return fm_end_offset;
2275 * We calculate on which OST the mapping will end. If the length of mapping
2276 * is greater than (stripe_size * stripe_count) then the last_stripe will
2277 * will be one just before start_stripe. Else we check if the mapping
2278 * intersects each OST and find last_stripe.
2279 * This function returns the last_stripe and also sets the stripe_count
2280 * over which the mapping is spread
2282 * \param lsm striping information for the file
2283 * \param fm_start logical start of mapping
2284 * \param fm_end logical end of mapping
2285 * \param start_stripe starting stripe of the mapping
2286 * \param stripe_count the number of stripes across which to map is returned
2288 * \retval last_stripe return the last stripe of the mapping
2290 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2291 obd_size fm_end, int start_stripe,
2295 obd_off obd_start, obd_end;
2298 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2299 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2301 *stripe_count = lsm->lsm_stripe_count;
2303 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2304 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2305 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2306 &obd_start, &obd_end)) == 0)
2310 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2317 * Set fe_device and copy extents from local buffer into main return buffer.
2319 * \param fiemap fiemap request header
2320 * \param lcl_fm_ext array of local fiemap extents to be copied
2321 * \param ost_index OST index to be written into the fm_device field for each
2323 * \param ext_count number of extents to be copied
2324 * \param current_extent where to start copying in main extent array
2326 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2327 struct ll_fiemap_extent *lcl_fm_ext,
2328 int ost_index, unsigned int ext_count,
2334 for (ext = 0; ext < ext_count; ext++) {
2335 lcl_fm_ext[ext].fe_device = ost_index;
2336 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2339 /* Copy fm_extent's from fm_local to return buffer */
2340 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2341 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2345 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2346 * This also handles the restarting of FIEMAP calls in case mapping overflows
2347 * the available number of extents in single call.
2349 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2350 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2352 struct ll_fiemap_info_key *fm_key = key;
2353 struct ll_user_fiemap *fiemap = val;
2354 struct ll_user_fiemap *fm_local = NULL;
2355 struct ll_fiemap_extent *lcl_fm_ext;
2357 unsigned int get_num_extents = 0;
2358 int ost_index = 0, actual_start_stripe, start_stripe;
2359 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2361 int current_extent = 0, rc = 0, i;
2362 int ost_eof = 0; /* EOF for object */
2363 int ost_done = 0; /* done with required mapping for this OST? */
2365 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2366 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2371 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2372 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2374 OBD_ALLOC_LARGE(fm_local, buffer_size);
2375 if (fm_local == NULL)
2376 GOTO(out, rc = -ENOMEM);
2377 lcl_fm_ext = &fm_local->fm_extents[0];
2379 count_local = fiemap_size_to_count(buffer_size);
2381 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2382 fm_start = fiemap->fm_start;
2383 fm_length = fiemap->fm_length;
2384 /* Calculate start stripe, last stripe and length of mapping */
2385 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2386 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2387 fm_start + fm_length - 1);
2388 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2389 if (fm_end > fm_key->oa.o_size)
2390 fm_end = fm_key->oa.o_size;
2392 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2393 actual_start_stripe, &stripe_count);
2395 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2398 if (fiemap->fm_extent_count == 0) {
2399 get_num_extents = 1;
2403 /* Check each stripe */
2404 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2405 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2406 obd_size req_fm_len; /* Stores length of required mapping */
2407 obd_size len_mapped_single_call;
2408 obd_off lun_start, lun_end, obd_object_end;
2409 unsigned int ext_count;
2411 cur_stripe_wrap = cur_stripe;
2413 /* Find out range of mapping on this stripe */
2414 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2415 &lun_start, &obd_object_end)) == 0)
2418 /* If this is a continuation FIEMAP call and we are on
2419 * starting stripe then lun_start needs to be set to
2421 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2422 lun_start = fm_end_offset;
2424 if (fm_length != ~0ULL) {
2425 /* Handle fm_start + fm_length overflow */
2426 if (fm_start + fm_length < fm_start)
2427 fm_length = ~0ULL - fm_start;
2428 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2434 if (lun_start == lun_end)
2437 req_fm_len = obd_object_end - lun_start;
2438 fm_local->fm_length = 0;
2439 len_mapped_single_call = 0;
2441 /* If the output buffer is very large and the objects have many
2442 * extents we may need to loop on a single OST repeatedly */
2446 if (get_num_extents == 0) {
2447 /* Don't get too many extents. */
2448 if (current_extent + count_local >
2449 fiemap->fm_extent_count)
2450 count_local = fiemap->fm_extent_count -
2454 lun_start += len_mapped_single_call;
2455 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2456 req_fm_len = fm_local->fm_length;
2457 fm_local->fm_extent_count = count_local;
2458 fm_local->fm_mapped_extents = 0;
2459 fm_local->fm_flags = fiemap->fm_flags;
2461 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2462 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2463 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2465 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2466 GOTO(out, rc = -EINVAL);
2468 /* If OST is inactive, return extent with UNKNOWN flag */
2469 if (!lov->lov_tgts[ost_index]->ltd_active) {
2470 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2471 fm_local->fm_mapped_extents = 1;
2473 lcl_fm_ext[0].fe_logical = lun_start;
2474 lcl_fm_ext[0].fe_length = obd_object_end -
2476 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2481 fm_local->fm_start = lun_start;
2482 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2483 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2484 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2485 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2486 keylen, key, vallen, fm_local, lsm);
2491 ext_count = fm_local->fm_mapped_extents;
2492 if (ext_count == 0) {
2494 /* If last stripe has hole at the end,
2495 * then we need to return */
2496 if (cur_stripe_wrap == last_stripe) {
2497 fiemap->fm_mapped_extents = 0;
2503 /* If we just need num of extents then go to next device */
2504 if (get_num_extents) {
2505 current_extent += ext_count;
2509 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2510 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2512 /* Have we finished mapping on this device? */
2513 if (req_fm_len <= len_mapped_single_call)
2516 /* Clear the EXTENT_LAST flag which can be present on
2518 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2519 lcl_fm_ext[ext_count - 1].fe_flags &=
2520 ~FIEMAP_EXTENT_LAST;
2522 curr_loc = lov_stripe_size(lsm,
2523 lcl_fm_ext[ext_count - 1].fe_logical+
2524 lcl_fm_ext[ext_count - 1].fe_length,
2526 if (curr_loc >= fm_key->oa.o_size)
2529 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2530 ost_index, ext_count,
2533 current_extent += ext_count;
2535 /* Ran out of available extents? */
2536 if (current_extent >= fiemap->fm_extent_count)
2538 } while (ost_done == 0 && ost_eof == 0);
2540 if (cur_stripe_wrap == last_stripe)
2545 /* Indicate that we are returning device offsets unless file just has
2547 if (lsm->lsm_stripe_count > 1)
2548 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2550 if (get_num_extents)
2551 goto skip_last_device_calc;
2553 /* Check if we have reached the last stripe and whether mapping for that
2554 * stripe is done. */
2555 if (cur_stripe_wrap == last_stripe) {
2556 if (ost_done || ost_eof)
2557 fiemap->fm_extents[current_extent - 1].fe_flags |=
2561 skip_last_device_calc:
2562 fiemap->fm_mapped_extents = current_extent;
2565 OBD_FREE_LARGE(fm_local, buffer_size);
2569 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2570 void *key, __u32 *vallen, void *val,
2571 struct lov_stripe_md *lsm)
2573 struct obd_device *obddev = class_exp2obd(exp);
2574 struct lov_obd *lov = &obddev->u.lov;
2578 if (!vallen || !val)
2583 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2586 struct ldlm_lock *lock;
2588 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2589 struct lov_oinfo *loi;
2590 __u32 *stripe = val;
2592 if (*vallen < sizeof(*stripe))
2593 GOTO(out, rc = -EFAULT);
2594 *vallen = sizeof(*stripe);
2596 /* XXX This is another one of those bits that will need to
2597 * change if we ever actually support nested LOVs. It uses
2598 * the lock's export to find out which stripe it is. */
2599 /* XXX - it's assumed all the locks for deleted OSTs have
2600 * been cancelled. Also, the export for deleted OSTs will
2601 * be NULL and won't match the lock's export. */
2602 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2603 loi = lsm->lsm_oinfo[i];
2604 if (!lov->lov_tgts[loi->loi_ost_idx])
2606 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2607 data->lock->l_conn_export &&
2608 osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2613 LDLM_ERROR(data->lock, "lock on inode without such object");
2614 dump_lsm(D_ERROR, lsm);
2615 GOTO(out, rc = -ENXIO);
2616 } else if (KEY_IS(KEY_LAST_ID)) {
2617 struct obd_id_info *info = val;
2618 __u32 size = sizeof(obd_id);
2619 struct lov_tgt_desc *tgt;
2621 LASSERT(*vallen == sizeof(struct obd_id_info));
2622 tgt = lov->lov_tgts[info->idx];
2624 if (!tgt || !tgt->ltd_active)
2625 GOTO(out, rc = -ESRCH);
2627 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2629 } else if (KEY_IS(KEY_LOVDESC)) {
2630 struct lov_desc *desc_ret = val;
2631 *desc_ret = lov->desc;
2634 } else if (KEY_IS(KEY_FIEMAP)) {
2635 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2637 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2638 struct lov_tgt_desc *tgt;
2639 __u64 ost_idx = *((__u64*)val);
2641 LASSERT(*vallen == sizeof(__u64));
2642 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2643 tgt = lov->lov_tgts[ost_idx];
2645 if (!tgt || !tgt->ltd_exp)
2646 GOTO(out, rc = -ESRCH);
2648 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2650 } else if (KEY_IS(KEY_TGT_COUNT)) {
2651 *((int *)val) = lov->desc.ld_tgt_count;
2662 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2663 void *key, obd_count vallen, void *val,
2664 struct ptlrpc_request_set *set)
2666 struct obd_device *obddev = class_exp2obd(exp);
2667 struct lov_obd *lov = &obddev->u.lov;
2670 struct lov_tgt_desc *tgt;
2671 unsigned incr, check_uuid,
2672 do_inactive, no_set;
2673 unsigned next_id = 0, mds_con = 0, capa = 0;
2676 incr = check_uuid = do_inactive = no_set = 0;
2679 set = ptlrpc_prep_set();
2685 count = lov->desc.ld_tgt_count;
2687 if (KEY_IS(KEY_NEXT_ID)) {
2688 count = vallen / sizeof(struct obd_id_info);
2689 vallen = sizeof(obd_id);
2690 incr = sizeof(struct obd_id_info);
2693 } else if (KEY_IS(KEY_CHECKSUM)) {
2695 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2696 /* use defaults: do_inactive = incr = 0; */
2697 } else if (KEY_IS(KEY_MDS_CONN)) {
2699 } else if (KEY_IS(KEY_CAPA_KEY)) {
2703 for (i = 0; i < count; i++, val = (char *)val + incr) {
2705 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2707 tgt = lov->lov_tgts[i];
2709 /* OST was disconnected */
2710 if (!tgt || !tgt->ltd_exp)
2713 /* OST is inactive and we don't want inactive OSCs */
2714 if (!tgt->ltd_active && !do_inactive)
2718 struct mds_group_info *mgi;
2720 LASSERT(vallen == sizeof(*mgi));
2721 mgi = (struct mds_group_info *)val;
2723 /* Only want a specific OSC */
2724 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2728 err = obd_set_info_async(tgt->ltd_exp,
2729 keylen, key, sizeof(int),
2731 } else if (next_id) {
2732 err = obd_set_info_async(tgt->ltd_exp,
2733 keylen, key, vallen,
2734 ((struct obd_id_info*)val)->data, set);
2736 struct mds_capa_info *info = (struct mds_capa_info*)val;
2738 LASSERT(vallen == sizeof(*info));
2740 /* Only want a specific OSC */
2742 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2745 err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2746 sizeof(*info->capa),
2749 /* Only want a specific OSC */
2751 !obd_uuid_equals(val, &tgt->ltd_uuid))
2754 err = obd_set_info_async(tgt->ltd_exp,
2755 keylen, key, vallen, val, set);
2764 err = ptlrpc_set_wait(set);
2767 ptlrpc_set_destroy(set);
2772 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2777 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2778 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2779 if (loi->loi_ar.ar_rc && !rc)
2780 rc = loi->loi_ar.ar_rc;
2781 loi->loi_ar.ar_rc = 0;
2785 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2788 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2789 int cmd, __u64 *offset)
2791 __u32 ssize = lsm->lsm_stripe_size;
2795 do_div(start, ssize);
2796 start = start * ssize;
2798 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2799 ", end "LPU64"\n", *offset, ssize, start,
2801 if (cmd == OBD_CALC_STRIPE_END) {
2802 *offset = start + ssize - 1;
2803 } else if (cmd == OBD_CALC_STRIPE_START) {
2812 void lov_stripe_lock(struct lov_stripe_md *md)
2814 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2815 cfs_spin_lock(&md->lsm_lock);
2816 LASSERT(md->lsm_lock_owner == 0);
2817 md->lsm_lock_owner = cfs_curproc_pid();
2819 EXPORT_SYMBOL(lov_stripe_lock);
2821 void lov_stripe_unlock(struct lov_stripe_md *md)
2823 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2824 md->lsm_lock_owner = 0;
2825 cfs_spin_unlock(&md->lsm_lock);
2827 EXPORT_SYMBOL(lov_stripe_unlock);
2829 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2830 struct obd_quotactl *oqctl)
2832 struct lov_obd *lov = &obd->u.lov;
2833 struct lov_tgt_desc *tgt;
2835 __u64 bhardlimit = 0;
2839 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2840 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2841 oqctl->qc_cmd != Q_GETOQUOTA &&
2842 oqctl->qc_cmd != Q_INITQUOTA &&
2843 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2844 oqctl->qc_cmd != Q_FINVALIDATE) {
2845 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2851 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2854 tgt = lov->lov_tgts[i];
2859 if (!tgt->ltd_active || tgt->ltd_reap) {
2860 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2861 lov->lov_tgts[i]->ltd_activate) {
2863 CERROR("ost %d is inactive\n", i);
2865 CDEBUG(D_HA, "ost %d is inactive\n", i);
2870 err = obd_quotactl(tgt->ltd_exp, oqctl);
2872 if (tgt->ltd_active && !rc)
2877 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2878 curspace += oqctl->qc_dqblk.dqb_curspace;
2879 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2884 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2885 oqctl->qc_dqblk.dqb_curspace = curspace;
2886 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2891 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2892 struct obd_quotactl *oqctl)
2894 struct lov_obd *lov = &obd->u.lov;
2900 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2901 if (!lov->lov_tgts[i])
2904 /* Skip quota check on the administratively disabled OSTs. */
2905 if (!lov->lov_tgts[i]->ltd_activate) {
2906 CWARN("lov idx %d was administratively disabled, "
2907 "skip quotacheck on it.\n", i);
2911 if (!lov->lov_tgts[i]->ltd_active) {
2912 CERROR("lov idx %d inactive\n", i);
2918 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2921 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2924 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2935 int lov_quota_adjust_qunit(struct obd_export *exp,
2936 struct quota_adjust_qunit *oqaq,
2937 struct lustre_quota_ctxt *qctxt,
2938 struct ptlrpc_request_set *rqset)
2940 struct obd_device *obd = class_exp2obd(exp);
2941 struct lov_obd *lov = &obd->u.lov;
2943 unsigned no_set = 0;
2946 if (!QAQ_IS_ADJBLK(oqaq)) {
2947 CERROR("bad qaq_flags %x for lov obd.\n", oqaq->qaq_flags);
2951 if (rqset == NULL) {
2952 rqset = ptlrpc_prep_set();
2959 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2961 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
2962 CDEBUG(D_HA, "ost %d is inactive\n", i);
2966 err = obd_quota_adjust_qunit(lov->lov_tgts[i]->ltd_exp, oqaq,
2969 if (lov->lov_tgts[i]->ltd_active && !rc)
2975 err = ptlrpc_set_wait(rqset);
2979 /* Destroy the set if none was provided by the caller */
2981 ptlrpc_set_destroy(rqset);
2987 struct obd_ops lov_obd_ops = {
2988 .o_owner = THIS_MODULE,
2989 .o_setup = lov_setup,
2990 .o_precleanup = lov_precleanup,
2991 .o_cleanup = lov_cleanup,
2992 //.o_process_config = lov_process_config,
2993 .o_connect = lov_connect,
2994 .o_disconnect = lov_disconnect,
2995 .o_statfs = lov_statfs,
2996 .o_statfs_async = lov_statfs_async,
2997 .o_packmd = lov_packmd,
2998 .o_unpackmd = lov_unpackmd,
2999 .o_create = lov_create,
3000 .o_destroy = lov_destroy,
3001 .o_getattr = lov_getattr,
3002 .o_getattr_async = lov_getattr_async,
3003 .o_setattr = lov_setattr,
3004 .o_setattr_async = lov_setattr_async,
3006 .o_merge_lvb = lov_merge_lvb,
3007 .o_adjust_kms = lov_adjust_kms,
3008 .o_punch = lov_punch,
3010 .o_enqueue = lov_enqueue,
3011 .o_change_cbdata = lov_change_cbdata,
3012 .o_find_cbdata = lov_find_cbdata,
3013 .o_cancel = lov_cancel,
3014 .o_cancel_unused = lov_cancel_unused,
3015 .o_iocontrol = lov_iocontrol,
3016 .o_get_info = lov_get_info,
3017 .o_set_info_async = lov_set_info_async,
3018 .o_extent_calc = lov_extent_calc,
3019 .o_llog_init = lov_llog_init,
3020 .o_llog_finish = lov_llog_finish,
3021 .o_notify = lov_notify,
3022 .o_pool_new = lov_pool_new,
3023 .o_pool_rem = lov_pool_remove,
3024 .o_pool_add = lov_pool_add,
3025 .o_pool_del = lov_pool_del,
3026 .o_getref = lov_getref,
3027 .o_putref = lov_putref,
3028 .o_quotactl = lov_quotactl,
3029 .o_quotacheck = lov_quotacheck,
3030 .o_quota_adjust_qunit = lov_quota_adjust_qunit,
3033 static quota_interface_t *quota_interface;
3034 extern quota_interface_t lov_quota_interface;
3036 cfs_mem_cache_t *lov_oinfo_slab;
3038 extern struct lu_kmem_descr lov_caches[];
3040 int __init lov_init(void)
3042 struct lprocfs_static_vars lvars = { 0 };
3046 /* print an address of _any_ initialized kernel symbol from this
3047 * module, to allow debugging with gdb that doesn't support data
3048 * symbols from modules.*/
3049 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
3051 rc = lu_kmem_init(lov_caches);
3055 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3056 sizeof(struct lov_oinfo),
3057 0, CFS_SLAB_HWCACHE_ALIGN);
3058 if (lov_oinfo_slab == NULL) {
3059 lu_kmem_fini(lov_caches);
3062 lprocfs_lov_init_vars(&lvars);
3064 cfs_request_module("lquota");
3065 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3066 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3068 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3069 LUSTRE_LOV_NAME, &lov_device_type);
3072 if (quota_interface)
3073 PORTAL_SYMBOL_PUT(lov_quota_interface);
3074 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3076 lu_kmem_fini(lov_caches);
3083 static void /*__exit*/ lov_exit(void)
3087 lu_device_type_fini(&lov_device_type);
3088 lu_kmem_fini(lov_caches);
3090 if (quota_interface)
3091 PORTAL_SYMBOL_PUT(lov_quota_interface);
3093 class_unregister_type(LUSTRE_LOV_NAME);
3094 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3098 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3099 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3100 MODULE_LICENSE("GPL");
3102 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);