4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_obd.c
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_LOV
46 #include <libcfs/libcfs.h>
48 #include <liblustre.h>
51 #include <obd_support.h>
52 #include <lustre_lib.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_dlm.h>
56 #include <lustre_mds.h>
57 #include <lustre_debug.h>
58 #include <obd_class.h>
61 #include <lprocfs_status.h>
62 #include <lustre_param.h>
63 #include <cl_object.h>
64 #include <lustre/ll_fiemap.h>
65 #include <lustre_log.h>
67 #include "lov_internal.h"
69 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
70 Any function that expects lov_tgts to remain stationary must take a ref. */
71 static void lov_getref(struct obd_device *obd)
73 struct lov_obd *lov = &obd->u.lov;
75 /* nobody gets through here until lov_putref is done */
76 cfs_mutex_lock(&lov->lov_lock);
77 cfs_atomic_inc(&lov->lov_refcount);
78 cfs_mutex_unlock(&lov->lov_lock);
82 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
84 static void lov_putref(struct obd_device *obd)
86 struct lov_obd *lov = &obd->u.lov;
88 cfs_mutex_lock(&lov->lov_lock);
89 /* ok to dec to 0 more than once -- ltd_exp's will be null */
90 if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
93 struct lov_tgt_desc *tgt, *n;
94 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
96 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
97 tgt = lov->lov_tgts[i];
99 if (!tgt || !tgt->ltd_reap)
101 cfs_list_add(&tgt->ltd_kill, &kill);
102 /* XXX - right now there is a dependency on ld_tgt_count
103 * being the maximum tgt index for computing the
104 * mds_max_easize. So we can't shrink it. */
105 lov_ost_pool_remove(&lov->lov_packed, i);
106 lov->lov_tgts[i] = NULL;
107 lov->lov_death_row--;
109 cfs_mutex_unlock(&lov->lov_lock);
111 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
112 cfs_list_del(&tgt->ltd_kill);
114 __lov_del_obd(obd, tgt);
117 cfs_mutex_unlock(&lov->lov_lock);
121 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
122 enum obd_notify_event ev);
123 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
124 enum obd_notify_event ev, void *data);
127 #define MAX_STRING_SIZE 128
128 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
129 struct obd_connect_data *data)
131 struct lov_obd *lov = &obd->u.lov;
132 struct obd_uuid *tgt_uuid;
133 struct obd_device *tgt_obd;
134 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
135 struct obd_import *imp;
137 cfs_proc_dir_entry_t *lov_proc_dir;
142 if (!lov->lov_tgts[index])
145 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
146 tgt_obd = lov->lov_tgts[index]->ltd_obd;
148 if (!tgt_obd->obd_set_up) {
149 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
153 /* override the sp_me from lov */
154 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
156 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
157 data->ocd_index = index;
160 * Divine LOV knows that OBDs under it are OSCs.
162 imp = tgt_obd->u.cli.cl_import;
165 tgt_obd->obd_no_recov = 0;
166 /* FIXME this is probably supposed to be
167 ptlrpc_set_import_active. Horrible naming. */
168 ptlrpc_activate_import(imp);
171 rc = obd_register_observer(tgt_obd, obd);
173 CERROR("Target %s register_observer error %d\n",
174 obd_uuid2str(tgt_uuid), rc);
179 if (imp->imp_invalid) {
180 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
181 "disabled\n", obd_uuid2str(tgt_uuid));
185 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
186 &lov_osc_uuid, data, NULL);
187 if (rc || !lov->lov_tgts[index]->ltd_exp) {
188 CERROR("Target %s connect error %d\n",
189 obd_uuid2str(tgt_uuid), rc);
193 lov->lov_tgts[index]->ltd_reap = 0;
195 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
196 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
199 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
201 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
202 cfs_proc_dir_entry_t *osc_symlink;
204 LASSERT(osc_obd != NULL);
205 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
206 LASSERT(osc_obd->obd_type->typ_name != NULL);
208 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
211 osc_obd->obd_type->typ_name,
213 if (osc_symlink == NULL) {
214 CERROR("could not register LOV target "
215 "/proc/fs/lustre/%s/%s/target_obds/%s.",
216 obd->obd_type->typ_name, obd->obd_name,
218 lprocfs_remove(&lov_proc_dir);
223 rc = qos_add_tgt(obd, index);
225 CERROR("qos_add_tgt failed %d\n", rc);
230 static int lov_connect(const struct lu_env *env,
231 struct obd_export **exp, struct obd_device *obd,
232 struct obd_uuid *cluuid, struct obd_connect_data *data,
235 struct lov_obd *lov = &obd->u.lov;
236 struct lov_tgt_desc *tgt;
237 struct lustre_handle conn;
241 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
243 rc = class_connect(&conn, obd, cluuid);
247 *exp = class_conn2export(&conn);
249 /* Why should there ever be more than 1 connect? */
251 LASSERT(lov->lov_connects == 1);
253 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
255 lov->lov_ocd = *data;
258 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
259 tgt = lov->lov_tgts[i];
260 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
262 /* Flags will be lowest common denominator */
263 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
265 CERROR("%s: lov connect tgt %d failed: %d\n",
266 obd->obd_name, i, rc);
269 /* connect to administrative disabled ost */
270 if (!lov->lov_tgts[i]->ltd_exp)
273 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
274 OBD_NOTIFY_CONNECT, (void *)&i);
276 CERROR("%s error sending notify %d\n",
285 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
287 cfs_proc_dir_entry_t *lov_proc_dir;
288 struct lov_obd *lov = &obd->u.lov;
289 struct obd_device *osc_obd;
293 osc_obd = class_exp2obd(tgt->ltd_exp);
294 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
295 obd->obd_name, osc_obd->obd_name);
297 if (tgt->ltd_active) {
299 lov->desc.ld_active_tgt_count--;
300 tgt->ltd_exp->exp_obd->obd_inactive = 1;
303 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
305 cfs_proc_dir_entry_t *osc_symlink;
307 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
309 lprocfs_remove(&osc_symlink);
311 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
312 obd->obd_type->typ_name, obd->obd_name,
318 /* Pass it on to our clients.
319 * XXX This should be an argument to disconnect,
320 * XXX not a back-door flag on the OBD. Ah well.
322 osc_obd->obd_force = obd->obd_force;
323 osc_obd->obd_fail = obd->obd_fail;
324 osc_obd->obd_no_recov = obd->obd_no_recov;
327 obd_register_observer(osc_obd, NULL);
329 rc = obd_disconnect(tgt->ltd_exp);
331 CERROR("Target %s disconnect error %d\n",
332 tgt->ltd_uuid.uuid, rc);
336 qos_del_tgt(obd, tgt);
342 static int lov_disconnect(struct obd_export *exp)
344 struct obd_device *obd = class_exp2obd(exp);
345 struct lov_obd *lov = &obd->u.lov;
352 /* Only disconnect the underlying layers on the final disconnect. */
354 if (lov->lov_connects != 0) {
355 /* why should there be more than 1 connect? */
356 CERROR("disconnect #%d\n", lov->lov_connects);
360 /* Let's hold another reference so lov_del_obd doesn't spin through
364 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
365 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
366 /* Disconnection is the last we know about an obd */
367 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
373 rc = class_disconnect(exp); /* bz 9811 */
379 * -EINVAL : UUID can't be found in the LOV's target list
380 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
381 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
382 * any >= 0 : is log target index
384 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
385 enum obd_notify_event ev)
387 struct lov_obd *lov = &obd->u.lov;
388 struct lov_tgt_desc *tgt;
389 int index, activate, active;
392 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
393 lov, uuid->uuid, ev);
396 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
397 tgt = lov->lov_tgts[index];
398 if (!tgt || !tgt->ltd_exp)
401 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
402 index, obd_uuid2str(&tgt->ltd_uuid),
403 tgt->ltd_exp->exp_handle.h_cookie);
404 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
408 if (index == lov->desc.ld_tgt_count)
409 GOTO(out, index = -EINVAL);
411 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
412 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
414 if (lov->lov_tgts[index]->ltd_activate == activate) {
415 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
416 uuid->uuid, activate ? "" : "de");
418 lov->lov_tgts[index]->ltd_activate = activate;
419 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
420 activate ? "" : "de", obd_uuid2str(uuid));
423 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
424 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
426 if (lov->lov_tgts[index]->ltd_active == active) {
427 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
428 uuid->uuid, active ? "" : "in");
431 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
432 obd_uuid2str(uuid), active ? "" : "in");
435 lov->lov_tgts[index]->ltd_active = active;
437 lov->desc.ld_active_tgt_count++;
438 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
440 lov->desc.ld_active_tgt_count--;
441 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
443 /* remove any old qos penalty */
444 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
446 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
454 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
455 enum obd_notify_event ev, void *data)
460 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
461 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
462 struct obd_uuid *uuid;
466 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
467 CERROR("unexpected notification of %s %s!\n",
468 watched->obd_type->typ_name,
472 uuid = &watched->u.cli.cl_target_uuid;
474 /* Set OSC as active before notifying the observer, so the
475 * observer can use the OSC normally.
477 rc = lov_set_osc_active(obd, uuid, ev);
479 CERROR("event(%d) of %s failed: %d\n", ev,
480 obd_uuid2str(uuid), rc);
483 /* active event should be pass lov target index as data */
487 /* Pass the notification up the chain. */
489 rc = obd_notify_observer(obd, watched, ev, data);
491 /* NULL watched means all osc's in the lov (only for syncs) */
492 /* sync event should be send lov idx as data */
493 struct lov_obd *lov = &obd->u.lov;
497 is_sync = (ev == OBD_NOTIFY_SYNC) ||
498 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
501 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
502 if (!lov->lov_tgts[i])
505 /* don't send sync event if target not
506 * connected/activated */
507 if (is_sync && !lov->lov_tgts[i]->ltd_active)
510 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
513 CERROR("%s: notify %s of %s failed %d\n",
515 obd->obd_observer->obd_name,
516 lov->lov_tgts[i]->ltd_obd->obd_name,
526 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
527 __u32 index, int gen, int active)
529 struct lov_obd *lov = &obd->u.lov;
530 struct lov_tgt_desc *tgt;
531 struct obd_device *tgt_obd;
535 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
536 uuidp->uuid, index, gen, active);
539 CERROR("request to add OBD %s with invalid generation: %d\n",
544 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
549 cfs_mutex_lock(&lov->lov_lock);
551 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
552 tgt = lov->lov_tgts[index];
553 CERROR("UUID %s already assigned at LOV target index %d\n",
554 obd_uuid2str(&tgt->ltd_uuid), index);
555 cfs_mutex_unlock(&lov->lov_lock);
559 if (index >= lov->lov_tgt_size) {
560 /* We need to reallocate the lov target array. */
561 struct lov_tgt_desc **newtgts, **old = NULL;
562 __u32 newsize, oldsize = 0;
564 newsize = max(lov->lov_tgt_size, (__u32)2);
565 while (newsize < index + 1)
566 newsize = newsize << 1;
567 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
568 if (newtgts == NULL) {
569 cfs_mutex_unlock(&lov->lov_lock);
573 if (lov->lov_tgt_size) {
574 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
577 oldsize = lov->lov_tgt_size;
580 lov->lov_tgts = newtgts;
581 lov->lov_tgt_size = newsize;
586 OBD_FREE(old, sizeof(*old) * oldsize);
588 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
589 lov->lov_tgts, lov->lov_tgt_size);
594 cfs_mutex_unlock(&lov->lov_lock);
598 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
600 cfs_mutex_unlock(&lov->lov_lock);
605 tgt->ltd_uuid = *uuidp;
606 tgt->ltd_obd = tgt_obd;
607 /* XXX - add a sanity check on the generation number. */
609 tgt->ltd_index = index;
610 tgt->ltd_activate = active;
611 lov->lov_tgts[index] = tgt;
612 if (index >= lov->desc.ld_tgt_count)
613 lov->desc.ld_tgt_count = index + 1;
615 cfs_mutex_unlock(&lov->lov_lock);
617 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
618 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
620 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
622 if (lov->lov_connects == 0) {
623 /* lov_connect hasn't been called yet. We'll do the
624 lov_connect_obd on this target when that fn first runs,
625 because we don't know the connect flags yet. */
631 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
635 /* connect to administrative disabled ost */
639 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
640 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
645 CERROR("add failed (%d), deleting %s\n", rc,
646 obd_uuid2str(&tgt->ltd_uuid));
647 lov_del_target(obd, index, 0, 0);
653 /* Schedule a target for deletion */
654 int lov_del_target(struct obd_device *obd, __u32 index,
655 struct obd_uuid *uuidp, int gen)
657 struct lov_obd *lov = &obd->u.lov;
658 int count = lov->desc.ld_tgt_count;
662 if (index >= count) {
663 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
670 if (!lov->lov_tgts[index]) {
671 CERROR("LOV target at index %d is not setup.\n", index);
672 GOTO(out, rc = -EINVAL);
675 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
676 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
677 lov_uuid2str(lov, index), index,
678 obd_uuid2str(uuidp));
679 GOTO(out, rc = -EINVAL);
682 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
683 lov_uuid2str(lov, index), index,
684 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
685 lov->lov_tgts[index]->ltd_active);
687 lov->lov_tgts[index]->ltd_reap = 1;
688 lov->lov_death_row++;
689 /* we really delete it from obd_putref */
696 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
698 struct obd_device *osc_obd;
701 LASSERT(tgt->ltd_reap);
703 osc_obd = class_exp2obd(tgt->ltd_exp);
705 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
707 osc_obd ? osc_obd->obd_name : "<no obd>");
710 lov_disconnect_obd(obd, tgt);
714 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
715 do it ourselves. And we can't do it from lov_cleanup,
716 because we just lost our only reference to it. */
718 class_manual_cleanup(osc_obd);
721 void lov_fix_desc_stripe_size(__u64 *val)
723 if (*val < PTLRPC_MAX_BRW_SIZE) {
724 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
725 PTLRPC_MAX_BRW_SIZE);
726 *val = PTLRPC_MAX_BRW_SIZE;
727 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
728 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
729 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
731 *val, LOV_MIN_STRIPE_SIZE);
735 void lov_fix_desc_stripe_count(__u32 *val)
741 void lov_fix_desc_pattern(__u32 *val)
743 /* from lov_setstripe */
744 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
745 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
750 void lov_fix_desc_qos_maxage(__u32 *val)
754 *val = QOS_DEFAULT_MAXAGE;
757 void lov_fix_desc(struct lov_desc *desc)
759 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
760 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
761 lov_fix_desc_pattern(&desc->ld_pattern);
762 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
765 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
767 struct lprocfs_static_vars lvars = { 0 };
768 struct lov_desc *desc;
769 struct lov_obd *lov = &obd->u.lov;
773 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
774 CERROR("LOV setup requires a descriptor\n");
778 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
780 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
781 CERROR("descriptor size wrong: %d > %d\n",
782 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
786 if (desc->ld_magic != LOV_DESC_MAGIC) {
787 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
788 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
789 obd->obd_name, desc);
790 lustre_swab_lov_desc(desc);
792 CERROR("%s: Bad lov desc magic: %#x\n",
793 obd->obd_name, desc->ld_magic);
800 desc->ld_active_tgt_count = 0;
802 lov->lov_tgt_size = 0;
804 cfs_mutex_init(&lov->lov_lock);
805 cfs_atomic_set(&lov->lov_refcount, 0);
806 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
807 cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
808 lov->lov_sp_me = LUSTRE_SP_CLI;
809 lov->lov_qos.lq_dirty = 1;
810 lov->lov_qos.lq_rr.lqr_dirty = 1;
811 lov->lov_qos.lq_reset = 1;
812 /* Default priority is toward free space balance */
813 lov->lov_qos.lq_prio_free = 232;
814 /* Default threshold for rr (roughly 17%) */
815 lov->lov_qos.lq_threshold_rr = 43;
816 /* Init statfs fields */
817 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
818 if (NULL == lov->lov_qos.lq_statfs_data)
820 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
822 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
824 HASH_POOLS_BKT_BITS, 0,
827 &pool_hash_operations,
829 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
830 lov->lov_pool_count = 0;
831 rc = lov_ost_pool_init(&lov->lov_packed, 0);
833 GOTO(out_free_statfs, rc);
834 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
836 GOTO(out_free_lov_packed, rc);
838 lprocfs_lov_init_vars(&lvars);
839 lprocfs_obd_setup(obd, lvars.obd_vars);
844 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
845 0444, &lov_proc_target_fops, obd);
847 CWARN("Error adding the target_obd file\n");
850 lov->lov_pool_proc_entry = lprocfs_register("pools",
857 lov_ost_pool_free(&lov->lov_packed);
859 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
863 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
866 struct lov_obd *lov = &obd->u.lov;
871 case OBD_CLEANUP_EARLY: {
873 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
874 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
876 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
881 case OBD_CLEANUP_EXPORTS:
882 rc = obd_llog_finish(obd, 0);
884 CERROR("failed to cleanup llogging subsystems\n");
890 static int lov_cleanup(struct obd_device *obd)
892 struct lov_obd *lov = &obd->u.lov;
893 cfs_list_t *pos, *tmp;
894 struct pool_desc *pool;
897 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
898 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
899 /* free pool structs */
900 CDEBUG(D_INFO, "delete pool %p\n", pool);
901 lov_pool_del(obd, pool->pool_name);
903 cfs_hash_putref(lov->lov_pools_hash_body);
904 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
905 lov_ost_pool_free(&lov->lov_packed);
907 lprocfs_obd_cleanup(obd);
911 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
912 if (!lov->lov_tgts[i])
915 /* Inactive targets may never have connected */
916 if (lov->lov_tgts[i]->ltd_active ||
917 cfs_atomic_read(&lov->lov_refcount))
918 /* We should never get here - these
919 should have been removed in the
921 CERROR("lov tgt %d not cleaned!"
922 " deathrow=%d, lovrc=%d\n",
923 i, lov->lov_death_row,
924 cfs_atomic_read(&lov->lov_refcount));
925 lov_del_target(obd, i, 0, 0);
928 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
930 lov->lov_tgt_size = 0;
932 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
936 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
937 __u32 *indexp, int *genp)
939 struct obd_uuid obd_uuid;
944 switch(cmd = lcfg->lcfg_command) {
945 case LCFG_LOV_ADD_OBD:
946 case LCFG_LOV_ADD_INA:
947 case LCFG_LOV_DEL_OBD: {
950 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
951 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
952 GOTO(out, rc = -EINVAL);
954 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
956 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
957 GOTO(out, rc = -EINVAL);
958 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
959 GOTO(out, rc = -EINVAL);
962 if (cmd == LCFG_LOV_ADD_OBD)
963 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
964 else if (cmd == LCFG_LOV_ADD_INA)
965 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
967 rc = lov_del_target(obd, index, &obd_uuid, gen);
971 struct lprocfs_static_vars lvars = { 0 };
972 struct lov_desc *desc = &(obd->u.lov.desc);
975 GOTO(out, rc = -EINVAL);
977 lprocfs_lov_init_vars(&lvars);
979 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
992 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
993 GOTO(out, rc = -EINVAL);
1002 #define log2(n) cfs_ffz(~(n))
1005 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1006 struct lov_stripe_md **ea,
1007 struct obd_trans_info *oti)
1009 struct lov_obd *lov;
1010 struct obdo *tmp_oa;
1011 struct obd_uuid *ost_uuid = NULL;
1015 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1016 src_oa->o_flags == OBD_FL_DELORPHAN);
1018 lov = &export->exp_obd->u.lov;
1024 if (oti->oti_ost_uuid) {
1025 ost_uuid = oti->oti_ost_uuid;
1026 CDEBUG(D_HA, "clearing orphans only for %s\n",
1030 obd_getref(export->exp_obd);
1031 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1032 struct lov_stripe_md obj_md;
1033 struct lov_stripe_md *obj_mdp = &obj_md;
1034 struct lov_tgt_desc *tgt;
1037 tgt = lov->lov_tgts[i];
1041 /* if called for a specific target, we don't
1042 care if it is not active. */
1043 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1044 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1048 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1051 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1052 obd_uuid2str(ost_uuid));
1054 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1056 LASSERT(lov->lov_tgts[i]->ltd_exp);
1057 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1058 err = obd_create(NULL, lov->lov_tgts[i]->ltd_exp,
1059 tmp_oa, &obj_mdp, oti);
1061 /* This export will be disabled until it is recovered,
1062 and then orphan recovery will be completed. */
1063 CERROR("error in orphan recovery on OST idx %d/%d: "
1064 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1071 obd_putref(export->exp_obd);
1077 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1078 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1080 struct lov_stripe_md *obj_mdp, *lsm;
1081 struct lov_obd *lov = &exp->exp_obd->u.lov;
1086 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1087 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1089 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1090 if (obj_mdp == NULL)
1093 ost_idx = src_oa->o_nlink;
1096 GOTO(out, rc = -EINVAL);
1097 if (ost_idx >= lov->desc.ld_tgt_count ||
1098 !lov->lov_tgts[ost_idx])
1099 GOTO(out, rc = -EINVAL);
1101 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1102 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1103 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1104 GOTO(out, rc = -EINVAL);
1108 if (i == lsm->lsm_stripe_count)
1109 GOTO(out, rc = -EINVAL);
1111 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1112 src_oa, &obj_mdp, oti);
1114 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1118 /* the LOV expects oa->o_id to be set to the LOV object id */
1119 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1120 struct obdo *src_oa, struct lov_stripe_md **ea,
1121 struct obd_trans_info *oti)
1123 struct lov_obd *lov;
1124 struct obd_info oinfo;
1125 struct lov_request_set *set = NULL;
1126 struct lov_request *req;
1127 struct l_wait_info lwi = { 0 };
1131 LASSERT(ea != NULL);
1135 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1136 src_oa->o_flags == OBD_FL_DELORPHAN) {
1137 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1141 lov = &exp->exp_obd->u.lov;
1142 if (!lov->desc.ld_active_tgt_count)
1145 obd_getref(exp->exp_obd);
1146 /* Recreate a specific object id at the given OST index */
1147 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1148 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1149 rc = lov_recreate(exp, src_oa, ea, oti);
1153 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1154 * later in alloc_qos(), we will wait for those rpcs to complete if
1155 * the osfs age is older than 2 * qos_maxage */
1156 qos_statfs_update(exp->exp_obd,
1157 cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1158 OBD_STATFS_CACHE_SECONDS),
1161 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1165 cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1166 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1167 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1168 &req->rq_oi, &req->rq_oi.oi_md, oti);
1171 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1172 * longer then this */
1173 l_wait_event(set->set_waitq, lov_set_finished(set, 1), &lwi);
1175 /* we not have ptlrpc set for assign set->interpret and should
1176 * be call interpret function himself. calling from cb_create_update
1177 * not permited because lov_fini_create_set can sleep for long time,
1178 * but we must avoid sleeping in ptlrpcd interpret function. */
1179 rc = lov_fini_create_set(set, ea);
1181 obd_putref(exp->exp_obd);
1185 #define ASSERT_LSM_MAGIC(lsmp) \
1187 LASSERT((lsmp) != NULL); \
1188 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1189 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1190 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1193 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1194 struct obdo *oa, struct lov_stripe_md *lsm,
1195 struct obd_trans_info *oti, struct obd_export *md_exp,
1198 struct lov_request_set *set;
1199 struct obd_info oinfo;
1200 struct lov_request *req;
1202 struct lov_obd *lov;
1203 int rc = 0, err = 0;
1206 ASSERT_LSM_MAGIC(lsm);
1208 if (!exp || !exp->exp_obd)
1211 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1213 LASSERT(oti->oti_logcookies);
1216 lov = &exp->exp_obd->u.lov;
1217 obd_getref(exp->exp_obd);
1218 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1222 cfs_list_for_each (pos, &set->set_list) {
1223 req = cfs_list_entry(pos, struct lov_request, rq_link);
1225 if (oa->o_valid & OBD_MD_FLCOOKIE)
1226 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1228 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1229 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1230 err = lov_update_common_set(set, req, err);
1232 CERROR("error: destroying objid "LPX64" subobj "
1233 LPX64" on OST idx %d: rc = %d\n",
1234 oa->o_id, req->rq_oi.oi_oa->o_id,
1242 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1243 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1245 err = lov_fini_destroy_set(set);
1247 obd_putref(exp->exp_obd);
1248 RETURN(rc ? rc : err);
1251 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1252 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(env, 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(const struct lu_env *env, struct obd_export *exp,
1370 struct obd_info *oinfo, 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(env, 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(const struct lu_env *env, struct obd_export *exp,
1519 struct obd_info *oinfo, 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(env, 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(const struct lu_env *env, struct obd_export *exp,
1581 struct obd_info *oinfo, 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(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1609 &req->rq_oi, 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_export *exp, struct obd_info *oinfo,
1955 __u64 max_age, struct ptlrpc_request_set *rqset)
1957 struct obd_device *obd = class_exp2obd(exp);
1958 struct lov_request_set *set;
1959 struct lov_request *req;
1961 struct lov_obd *lov;
1965 LASSERT(oinfo != NULL);
1966 LASSERT(oinfo->oi_osfs != NULL);
1969 rc = lov_prep_statfs_set(obd, oinfo, &set);
1973 cfs_list_for_each (pos, &set->set_list) {
1974 req = cfs_list_entry(pos, struct lov_request, rq_link);
1975 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1976 &req->rq_oi, max_age, rqset);
1981 if (rc || cfs_list_empty(&rqset->set_requests)) {
1984 cfs_atomic_set(&set->set_completes, 0);
1985 err = lov_fini_statfs_set(set);
1986 RETURN(rc ? rc : err);
1989 LASSERT(rqset->set_interpret == NULL);
1990 rqset->set_interpret = lov_statfs_interpret;
1991 rqset->set_arg = (void *)set;
1995 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1996 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1998 struct ptlrpc_request_set *set = NULL;
1999 struct obd_info oinfo = { { { 0 } } };
2004 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2005 * statfs requests */
2006 set = ptlrpc_prep_set();
2010 oinfo.oi_osfs = osfs;
2011 oinfo.oi_flags = flags;
2012 rc = lov_statfs_async(exp, &oinfo, max_age, set);
2014 rc = ptlrpc_set_wait(set);
2015 ptlrpc_set_destroy(set);
2020 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2021 void *karg, void *uarg)
2023 struct obd_device *obddev = class_exp2obd(exp);
2024 struct lov_obd *lov = &obddev->u.lov;
2025 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
2026 struct obd_uuid *uuidp;
2030 case IOC_OBD_STATFS: {
2031 struct obd_ioctl_data *data = karg;
2032 struct obd_device *osc_obd;
2033 struct obd_statfs stat_buf = {0};
2036 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2037 if ((index >= count))
2040 if (!lov->lov_tgts[index])
2041 /* Try again with the next index */
2043 if (!lov->lov_tgts[index]->ltd_active)
2046 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2051 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2052 min((int) data->ioc_plen2,
2053 (int) sizeof(struct obd_uuid))))
2056 /* got statfs data */
2057 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
2058 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2062 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2063 min((int) data->ioc_plen1,
2064 (int) sizeof(stat_buf))))
2068 case OBD_IOC_LOV_GET_CONFIG: {
2069 struct obd_ioctl_data *data;
2070 struct lov_desc *desc;
2075 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2078 data = (struct obd_ioctl_data *)buf;
2080 if (sizeof(*desc) > data->ioc_inllen1) {
2081 obd_ioctl_freedata(buf, len);
2085 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2086 obd_ioctl_freedata(buf, len);
2090 if (sizeof(__u32) * count > data->ioc_inllen3) {
2091 obd_ioctl_freedata(buf, len);
2095 desc = (struct lov_desc *)data->ioc_inlbuf1;
2096 memcpy(desc, &(lov->desc), sizeof(*desc));
2098 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2099 genp = (__u32 *)data->ioc_inlbuf3;
2100 /* the uuid will be empty for deleted OSTs */
2101 for (i = 0; i < count; i++, uuidp++, genp++) {
2102 if (!lov->lov_tgts[i])
2104 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2105 *genp = lov->lov_tgts[i]->ltd_gen;
2108 if (cfs_copy_to_user((void *)uarg, buf, len))
2110 obd_ioctl_freedata(buf, len);
2113 case LL_IOC_LOV_SETSTRIPE:
2114 rc = lov_setstripe(exp, len, karg, uarg);
2116 case LL_IOC_LOV_GETSTRIPE:
2117 rc = lov_getstripe(exp, karg, uarg);
2119 case LL_IOC_LOV_SETEA:
2120 rc = lov_setea(exp, karg, uarg);
2122 case OBD_IOC_QUOTACTL: {
2123 struct if_quotactl *qctl = karg;
2124 struct lov_tgt_desc *tgt = NULL;
2125 struct obd_quotactl *oqctl;
2127 if (qctl->qc_valid == QC_OSTIDX) {
2128 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2131 tgt = lov->lov_tgts[qctl->qc_idx];
2132 if (!tgt || !tgt->ltd_exp)
2134 } else if (qctl->qc_valid == QC_UUID) {
2135 for (i = 0; i < count; i++) {
2136 tgt = lov->lov_tgts[i];
2138 !obd_uuid_equals(&tgt->ltd_uuid,
2142 if (tgt->ltd_exp == NULL)
2154 LASSERT(tgt && tgt->ltd_exp);
2155 OBD_ALLOC_PTR(oqctl);
2159 QCTL_COPY(oqctl, qctl);
2160 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2162 QCTL_COPY(qctl, oqctl);
2163 qctl->qc_valid = QC_OSTIDX;
2164 qctl->obd_uuid = tgt->ltd_uuid;
2166 OBD_FREE_PTR(oqctl);
2175 for (i = 0; i < count; i++) {
2177 struct obd_device *osc_obd;
2179 /* OST was disconnected */
2180 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2183 /* ll_umount_begin() sets force flag but for lov, not
2184 * osc. Let's pass it through */
2185 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2186 osc_obd->obd_force = obddev->obd_force;
2187 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2189 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2192 if (lov->lov_tgts[i]->ltd_active) {
2193 CDEBUG(err == -ENOTTY ?
2194 D_IOCTL : D_WARNING,
2195 "iocontrol OSC %s on OST "
2196 "idx %d cmd %x: err = %d\n",
2197 lov_uuid2str(lov, i),
2214 #define FIEMAP_BUFFER_SIZE 4096
2217 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2218 * call. The local end offset and the device are sent in the first
2219 * fm_extent. This function calculates the stripe number from the index.
2220 * This function returns a stripe_no on which mapping is to be restarted.
2222 * This function returns fm_end_offset which is the in-OST offset at which
2223 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2224 * will re-calculate proper offset in next stripe.
2225 * Note that the first extent is passed to lov_get_info via the value field.
2227 * \param fiemap fiemap request header
2228 * \param lsm striping information for the file
2229 * \param fm_start logical start of mapping
2230 * \param fm_end logical end of mapping
2231 * \param start_stripe starting stripe will be returned in this
2233 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2234 struct lov_stripe_md *lsm, obd_size fm_start,
2235 obd_size fm_end, int *start_stripe)
2237 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2238 obd_off lun_start, lun_end;
2239 obd_size fm_end_offset;
2240 int stripe_no = -1, i;
2242 if (fiemap->fm_extent_count == 0 ||
2243 fiemap->fm_extents[0].fe_logical == 0)
2246 /* Find out stripe_no from ost_index saved in the fe_device */
2247 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2248 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2249 fiemap->fm_extents[0].fe_device) {
2254 if (stripe_no == -1)
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;
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,
2396 fm_end, &start_stripe);
2397 if (fm_end_offset == -EINVAL)
2400 if (fiemap->fm_extent_count == 0) {
2401 get_num_extents = 1;
2405 /* Check each stripe */
2406 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2407 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2408 obd_size req_fm_len; /* Stores length of required mapping */
2409 obd_size len_mapped_single_call;
2410 obd_off lun_start, lun_end, obd_object_end;
2411 unsigned int ext_count;
2413 cur_stripe_wrap = cur_stripe;
2415 /* Find out range of mapping on this stripe */
2416 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2417 &lun_start, &obd_object_end)) == 0)
2420 /* If this is a continuation FIEMAP call and we are on
2421 * starting stripe then lun_start needs to be set to
2423 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2424 lun_start = fm_end_offset;
2426 if (fm_length != ~0ULL) {
2427 /* Handle fm_start + fm_length overflow */
2428 if (fm_start + fm_length < fm_start)
2429 fm_length = ~0ULL - fm_start;
2430 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2436 if (lun_start == lun_end)
2439 req_fm_len = obd_object_end - lun_start;
2440 fm_local->fm_length = 0;
2441 len_mapped_single_call = 0;
2443 /* If the output buffer is very large and the objects have many
2444 * extents we may need to loop on a single OST repeatedly */
2448 if (get_num_extents == 0) {
2449 /* Don't get too many extents. */
2450 if (current_extent + count_local >
2451 fiemap->fm_extent_count)
2452 count_local = fiemap->fm_extent_count -
2456 lun_start += len_mapped_single_call;
2457 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2458 req_fm_len = fm_local->fm_length;
2459 fm_local->fm_extent_count = count_local;
2460 fm_local->fm_mapped_extents = 0;
2461 fm_local->fm_flags = fiemap->fm_flags;
2463 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2464 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2465 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2467 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2468 GOTO(out, rc = -EINVAL);
2470 /* If OST is inactive, return extent with UNKNOWN flag */
2471 if (!lov->lov_tgts[ost_index]->ltd_active) {
2472 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2473 fm_local->fm_mapped_extents = 1;
2475 lcl_fm_ext[0].fe_logical = lun_start;
2476 lcl_fm_ext[0].fe_length = obd_object_end -
2478 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2483 fm_local->fm_start = lun_start;
2484 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2485 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2486 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2487 rc = obd_get_info(NULL,
2488 lov->lov_tgts[ost_index]->ltd_exp,
2489 keylen, key, vallen, fm_local, lsm);
2494 ext_count = fm_local->fm_mapped_extents;
2495 if (ext_count == 0) {
2497 /* If last stripe has hole at the end,
2498 * then we need to return */
2499 if (cur_stripe_wrap == last_stripe) {
2500 fiemap->fm_mapped_extents = 0;
2506 /* If we just need num of extents then go to next device */
2507 if (get_num_extents) {
2508 current_extent += ext_count;
2512 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2513 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2515 /* Have we finished mapping on this device? */
2516 if (req_fm_len <= len_mapped_single_call)
2519 /* Clear the EXTENT_LAST flag which can be present on
2521 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2522 lcl_fm_ext[ext_count - 1].fe_flags &=
2523 ~FIEMAP_EXTENT_LAST;
2525 curr_loc = lov_stripe_size(lsm,
2526 lcl_fm_ext[ext_count - 1].fe_logical+
2527 lcl_fm_ext[ext_count - 1].fe_length,
2529 if (curr_loc >= fm_key->oa.o_size)
2532 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2533 ost_index, ext_count,
2536 current_extent += ext_count;
2538 /* Ran out of available extents? */
2539 if (current_extent >= fiemap->fm_extent_count)
2541 } while (ost_done == 0 && ost_eof == 0);
2543 if (cur_stripe_wrap == last_stripe)
2548 /* Indicate that we are returning device offsets unless file just has
2550 if (lsm->lsm_stripe_count > 1)
2551 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2553 if (get_num_extents)
2554 goto skip_last_device_calc;
2556 /* Check if we have reached the last stripe and whether mapping for that
2557 * stripe is done. */
2558 if (cur_stripe_wrap == last_stripe) {
2559 if (ost_done || ost_eof)
2560 fiemap->fm_extents[current_extent - 1].fe_flags |=
2564 skip_last_device_calc:
2565 fiemap->fm_mapped_extents = current_extent;
2568 OBD_FREE_LARGE(fm_local, buffer_size);
2572 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2573 __u32 keylen, void *key, __u32 *vallen, void *val,
2574 struct lov_stripe_md *lsm)
2576 struct obd_device *obddev = class_exp2obd(exp);
2577 struct lov_obd *lov = &obddev->u.lov;
2581 if (!vallen || !val)
2586 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2589 struct ldlm_lock *lock;
2591 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2592 struct lov_oinfo *loi;
2593 __u32 *stripe = val;
2595 if (*vallen < sizeof(*stripe))
2596 GOTO(out, rc = -EFAULT);
2597 *vallen = sizeof(*stripe);
2599 /* XXX This is another one of those bits that will need to
2600 * change if we ever actually support nested LOVs. It uses
2601 * the lock's export to find out which stripe it is. */
2602 /* XXX - it's assumed all the locks for deleted OSTs have
2603 * been cancelled. Also, the export for deleted OSTs will
2604 * be NULL and won't match the lock's export. */
2605 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2606 loi = lsm->lsm_oinfo[i];
2607 if (!lov->lov_tgts[loi->loi_ost_idx])
2609 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2610 data->lock->l_conn_export &&
2611 osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2616 LDLM_ERROR(data->lock, "lock on inode without such object");
2617 dump_lsm(D_ERROR, lsm);
2618 GOTO(out, rc = -ENXIO);
2619 } else if (KEY_IS(KEY_LAST_ID)) {
2620 struct obd_id_info *info = val;
2621 __u32 size = sizeof(obd_id);
2622 struct lov_tgt_desc *tgt;
2624 LASSERT(*vallen == sizeof(struct obd_id_info));
2625 tgt = lov->lov_tgts[info->idx];
2627 if (!tgt || !tgt->ltd_active)
2628 GOTO(out, rc = -ESRCH);
2630 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2631 &size, info->data, NULL);
2633 } else if (KEY_IS(KEY_LOVDESC)) {
2634 struct lov_desc *desc_ret = val;
2635 *desc_ret = lov->desc;
2638 } else if (KEY_IS(KEY_FIEMAP)) {
2639 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2641 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2642 struct lov_tgt_desc *tgt;
2643 __u64 ost_idx = *((__u64*)val);
2645 LASSERT(*vallen == sizeof(__u64));
2646 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2647 tgt = lov->lov_tgts[ost_idx];
2649 if (!tgt || !tgt->ltd_exp)
2650 GOTO(out, rc = -ESRCH);
2652 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2654 } else if (KEY_IS(KEY_TGT_COUNT)) {
2655 *((int *)val) = lov->desc.ld_tgt_count;
2666 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2667 obd_count keylen, void *key, obd_count vallen,
2668 void *val, struct ptlrpc_request_set *set)
2670 struct obd_device *obddev = class_exp2obd(exp);
2671 struct lov_obd *lov = &obddev->u.lov;
2674 struct lov_tgt_desc *tgt;
2675 unsigned incr, check_uuid,
2676 do_inactive, no_set;
2677 unsigned next_id = 0, mds_con = 0, capa = 0;
2680 incr = check_uuid = do_inactive = no_set = 0;
2683 set = ptlrpc_prep_set();
2689 count = lov->desc.ld_tgt_count;
2691 if (KEY_IS(KEY_NEXT_ID)) {
2692 count = vallen / sizeof(struct obd_id_info);
2693 vallen = sizeof(obd_id);
2694 incr = sizeof(struct obd_id_info);
2697 } else if (KEY_IS(KEY_CHECKSUM)) {
2699 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2700 /* use defaults: do_inactive = incr = 0; */
2701 } else if (KEY_IS(KEY_MDS_CONN)) {
2703 } else if (KEY_IS(KEY_CAPA_KEY)) {
2707 for (i = 0; i < count; i++, val = (char *)val + incr) {
2709 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2711 tgt = lov->lov_tgts[i];
2713 /* OST was disconnected */
2714 if (!tgt || !tgt->ltd_exp)
2717 /* OST is inactive and we don't want inactive OSCs */
2718 if (!tgt->ltd_active && !do_inactive)
2722 struct mds_group_info *mgi;
2724 LASSERT(vallen == sizeof(*mgi));
2725 mgi = (struct mds_group_info *)val;
2727 /* Only want a specific OSC */
2728 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2732 err = obd_set_info_async(env, tgt->ltd_exp,
2733 keylen, key, sizeof(int),
2735 } else if (next_id) {
2736 err = obd_set_info_async(env, tgt->ltd_exp,
2737 keylen, key, vallen,
2738 ((struct obd_id_info*)val)->data, set);
2740 struct mds_capa_info *info = (struct mds_capa_info*)val;
2742 LASSERT(vallen == sizeof(*info));
2744 /* Only want a specific OSC */
2746 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2749 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2750 key, sizeof(*info->capa),
2753 /* Only want a specific OSC */
2755 !obd_uuid_equals(val, &tgt->ltd_uuid))
2758 err = obd_set_info_async(env, tgt->ltd_exp,
2759 keylen, key, vallen, val, set);
2768 err = ptlrpc_set_wait(set);
2771 ptlrpc_set_destroy(set);
2776 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2777 int cmd, __u64 *offset)
2779 __u32 ssize = lsm->lsm_stripe_size;
2783 lov_do_div64(start, ssize);
2784 start = start * ssize;
2786 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2787 ", end "LPU64"\n", *offset, ssize, start,
2789 if (cmd == OBD_CALC_STRIPE_END) {
2790 *offset = start + ssize - 1;
2791 } else if (cmd == OBD_CALC_STRIPE_START) {
2800 void lov_stripe_lock(struct lov_stripe_md *md)
2802 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2803 cfs_spin_lock(&md->lsm_lock);
2804 LASSERT(md->lsm_lock_owner == 0);
2805 md->lsm_lock_owner = cfs_curproc_pid();
2807 EXPORT_SYMBOL(lov_stripe_lock);
2809 void lov_stripe_unlock(struct lov_stripe_md *md)
2811 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2812 md->lsm_lock_owner = 0;
2813 cfs_spin_unlock(&md->lsm_lock);
2815 EXPORT_SYMBOL(lov_stripe_unlock);
2817 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2818 struct obd_quotactl *oqctl)
2820 struct lov_obd *lov = &obd->u.lov;
2821 struct lov_tgt_desc *tgt;
2823 __u64 bhardlimit = 0;
2827 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2828 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2829 oqctl->qc_cmd != Q_GETOQUOTA &&
2830 oqctl->qc_cmd != Q_INITQUOTA &&
2831 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2832 oqctl->qc_cmd != Q_FINVALIDATE) {
2833 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2839 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2842 tgt = lov->lov_tgts[i];
2847 if (!tgt->ltd_active || tgt->ltd_reap) {
2848 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2849 lov->lov_tgts[i]->ltd_activate) {
2851 CERROR("ost %d is inactive\n", i);
2853 CDEBUG(D_HA, "ost %d is inactive\n", i);
2858 err = obd_quotactl(tgt->ltd_exp, oqctl);
2860 if (tgt->ltd_active && !rc)
2865 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2866 curspace += oqctl->qc_dqblk.dqb_curspace;
2867 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2872 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2873 oqctl->qc_dqblk.dqb_curspace = curspace;
2874 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2879 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2880 struct obd_quotactl *oqctl)
2882 struct lov_obd *lov = &obd->u.lov;
2888 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2889 if (!lov->lov_tgts[i])
2892 /* Skip quota check on the administratively disabled OSTs. */
2893 if (!lov->lov_tgts[i]->ltd_activate) {
2894 CWARN("lov idx %d was administratively disabled, "
2895 "skip quotacheck on it.\n", i);
2899 if (!lov->lov_tgts[i]->ltd_active) {
2900 CERROR("lov idx %d inactive\n", i);
2906 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2909 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2912 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2923 int lov_quota_adjust_qunit(struct obd_export *exp,
2924 struct quota_adjust_qunit *oqaq,
2925 struct lustre_quota_ctxt *qctxt,
2926 struct ptlrpc_request_set *rqset)
2928 struct obd_device *obd = class_exp2obd(exp);
2929 struct lov_obd *lov = &obd->u.lov;
2931 unsigned no_set = 0;
2934 if (!QAQ_IS_ADJBLK(oqaq)) {
2935 CERROR("bad qaq_flags %x for lov obd.\n", oqaq->qaq_flags);
2939 if (rqset == NULL) {
2940 rqset = ptlrpc_prep_set();
2947 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2949 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
2950 CDEBUG(D_HA, "ost %d is inactive\n", i);
2954 err = obd_quota_adjust_qunit(lov->lov_tgts[i]->ltd_exp, oqaq,
2957 if (lov->lov_tgts[i]->ltd_active && !rc)
2963 err = ptlrpc_set_wait(rqset);
2967 /* Destroy the set if none was provided by the caller */
2969 ptlrpc_set_destroy(rqset);
2975 struct obd_ops lov_obd_ops = {
2976 .o_owner = THIS_MODULE,
2977 .o_setup = lov_setup,
2978 .o_precleanup = lov_precleanup,
2979 .o_cleanup = lov_cleanup,
2980 //.o_process_config = lov_process_config,
2981 .o_connect = lov_connect,
2982 .o_disconnect = lov_disconnect,
2983 .o_statfs = lov_statfs,
2984 .o_statfs_async = lov_statfs_async,
2985 .o_packmd = lov_packmd,
2986 .o_unpackmd = lov_unpackmd,
2987 .o_create = lov_create,
2988 .o_destroy = lov_destroy,
2989 .o_getattr = lov_getattr,
2990 .o_getattr_async = lov_getattr_async,
2991 .o_setattr = lov_setattr,
2992 .o_setattr_async = lov_setattr_async,
2994 .o_merge_lvb = lov_merge_lvb,
2995 .o_adjust_kms = lov_adjust_kms,
2996 .o_punch = lov_punch,
2998 .o_enqueue = lov_enqueue,
2999 .o_change_cbdata = lov_change_cbdata,
3000 .o_find_cbdata = lov_find_cbdata,
3001 .o_cancel = lov_cancel,
3002 .o_cancel_unused = lov_cancel_unused,
3003 .o_iocontrol = lov_iocontrol,
3004 .o_get_info = lov_get_info,
3005 .o_set_info_async = lov_set_info_async,
3006 .o_extent_calc = lov_extent_calc,
3007 .o_llog_init = lov_llog_init,
3008 .o_llog_finish = lov_llog_finish,
3009 .o_notify = lov_notify,
3010 .o_pool_new = lov_pool_new,
3011 .o_pool_rem = lov_pool_remove,
3012 .o_pool_add = lov_pool_add,
3013 .o_pool_del = lov_pool_del,
3014 .o_getref = lov_getref,
3015 .o_putref = lov_putref,
3016 .o_quotactl = lov_quotactl,
3017 .o_quotacheck = lov_quotacheck,
3018 .o_quota_adjust_qunit = lov_quota_adjust_qunit,
3021 static quota_interface_t *quota_interface;
3022 extern quota_interface_t lov_quota_interface;
3024 cfs_mem_cache_t *lov_oinfo_slab;
3026 extern struct lu_kmem_descr lov_caches[];
3028 int __init lov_init(void)
3030 struct lprocfs_static_vars lvars = { 0 };
3034 /* print an address of _any_ initialized kernel symbol from this
3035 * module, to allow debugging with gdb that doesn't support data
3036 * symbols from modules.*/
3037 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
3039 rc = lu_kmem_init(lov_caches);
3043 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3044 sizeof(struct lov_oinfo),
3045 0, CFS_SLAB_HWCACHE_ALIGN);
3046 if (lov_oinfo_slab == NULL) {
3047 lu_kmem_fini(lov_caches);
3050 lprocfs_lov_init_vars(&lvars);
3052 cfs_request_module("lquota");
3053 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3054 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3056 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3057 LUSTRE_LOV_NAME, &lov_device_type);
3060 if (quota_interface)
3061 PORTAL_SYMBOL_PUT(lov_quota_interface);
3062 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3064 lu_kmem_fini(lov_caches);
3071 static void /*__exit*/ lov_exit(void)
3075 if (quota_interface)
3076 PORTAL_SYMBOL_PUT(lov_quota_interface);
3078 class_unregister_type(LUSTRE_LOV_NAME);
3079 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3082 lu_kmem_fini(lov_caches);
3085 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3086 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3087 MODULE_LICENSE("GPL");
3089 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);