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, Intel Corporation.
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 <lclient.h> /* for cl_client_lru */
65 #include <lustre/ll_fiemap.h>
66 #include <lustre_log.h>
67 #include <lustre_fid.h>
69 #include "lov_internal.h"
71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
72 Any function that expects lov_tgts to remain stationary must take a ref. */
73 static void lov_getref(struct obd_device *obd)
75 struct lov_obd *lov = &obd->u.lov;
77 /* nobody gets through here until lov_putref is done */
78 mutex_lock(&lov->lov_lock);
79 cfs_atomic_inc(&lov->lov_refcount);
80 mutex_unlock(&lov->lov_lock);
84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
86 static void lov_putref(struct obd_device *obd)
88 struct lov_obd *lov = &obd->u.lov;
90 mutex_lock(&lov->lov_lock);
91 /* ok to dec to 0 more than once -- ltd_exp's will be null */
92 if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
95 struct lov_tgt_desc *tgt, *n;
96 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
98 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
99 tgt = lov->lov_tgts[i];
101 if (!tgt || !tgt->ltd_reap)
103 cfs_list_add(&tgt->ltd_kill, &kill);
104 /* XXX - right now there is a dependency on ld_tgt_count
105 * being the maximum tgt index for computing the
106 * mds_max_easize. So we can't shrink it. */
107 lov_ost_pool_remove(&lov->lov_packed, i);
108 lov->lov_tgts[i] = NULL;
109 lov->lov_death_row--;
111 mutex_unlock(&lov->lov_lock);
113 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
114 cfs_list_del(&tgt->ltd_kill);
116 __lov_del_obd(obd, tgt);
119 mutex_unlock(&lov->lov_lock);
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
124 enum obd_notify_event ev);
125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
126 enum obd_notify_event ev, void *data);
129 #define MAX_STRING_SIZE 128
130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
131 struct obd_connect_data *data)
133 struct lov_obd *lov = &obd->u.lov;
134 struct obd_uuid *tgt_uuid;
135 struct obd_device *tgt_obd;
136 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
137 struct obd_import *imp;
139 cfs_proc_dir_entry_t *lov_proc_dir;
144 if (!lov->lov_tgts[index])
147 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
148 tgt_obd = lov->lov_tgts[index]->ltd_obd;
150 if (!tgt_obd->obd_set_up) {
151 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
155 /* override the sp_me from lov */
156 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
158 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
159 data->ocd_index = index;
162 * Divine LOV knows that OBDs under it are OSCs.
164 imp = tgt_obd->u.cli.cl_import;
167 tgt_obd->obd_no_recov = 0;
168 /* FIXME this is probably supposed to be
169 ptlrpc_set_import_active. Horrible naming. */
170 ptlrpc_activate_import(imp);
173 rc = obd_register_observer(tgt_obd, obd);
175 CERROR("Target %s register_observer error %d\n",
176 obd_uuid2str(tgt_uuid), rc);
181 if (imp->imp_invalid) {
182 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
183 "disabled\n", obd_uuid2str(tgt_uuid));
187 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
188 &lov_osc_uuid, data, NULL);
189 if (rc || !lov->lov_tgts[index]->ltd_exp) {
190 CERROR("Target %s connect error %d\n",
191 obd_uuid2str(tgt_uuid), rc);
195 lov->lov_tgts[index]->ltd_reap = 0;
197 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
198 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
201 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
203 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
204 cfs_proc_dir_entry_t *osc_symlink;
206 LASSERT(osc_obd != NULL);
207 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
208 LASSERT(osc_obd->obd_type->typ_name != NULL);
210 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
213 osc_obd->obd_type->typ_name,
215 if (osc_symlink == NULL) {
216 CERROR("could not register LOV target "
217 "/proc/fs/lustre/%s/%s/target_obds/%s.",
218 obd->obd_type->typ_name, obd->obd_name,
220 lprocfs_remove(&lov_proc_dir);
228 static int lov_connect(const struct lu_env *env,
229 struct obd_export **exp, struct obd_device *obd,
230 struct obd_uuid *cluuid, struct obd_connect_data *data,
233 struct lov_obd *lov = &obd->u.lov;
234 struct lov_tgt_desc *tgt;
235 struct lustre_handle conn;
239 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
241 rc = class_connect(&conn, obd, cluuid);
245 *exp = class_conn2export(&conn);
247 /* Why should there ever be more than 1 connect? */
249 LASSERT(lov->lov_connects == 1);
251 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
253 lov->lov_ocd = *data;
256 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
257 tgt = lov->lov_tgts[i];
258 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
260 /* Flags will be lowest common denominator */
261 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
263 CERROR("%s: lov connect tgt %d failed: %d\n",
264 obd->obd_name, i, rc);
267 /* connect to administrative disabled ost */
268 if (!lov->lov_tgts[i]->ltd_exp)
271 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
272 OBD_NOTIFY_CONNECT, (void *)&i);
274 CERROR("%s error sending notify %d\n",
283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
285 cfs_proc_dir_entry_t *lov_proc_dir;
286 struct lov_obd *lov = &obd->u.lov;
287 struct obd_device *osc_obd;
291 osc_obd = class_exp2obd(tgt->ltd_exp);
292 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
293 obd->obd_name, osc_obd->obd_name);
295 if (tgt->ltd_active) {
297 lov->desc.ld_active_tgt_count--;
298 tgt->ltd_exp->exp_obd->obd_inactive = 1;
301 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
303 cfs_proc_dir_entry_t *osc_symlink;
305 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
307 lprocfs_remove(&osc_symlink);
309 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
310 obd->obd_type->typ_name, obd->obd_name,
316 /* Pass it on to our clients.
317 * XXX This should be an argument to disconnect,
318 * XXX not a back-door flag on the OBD. Ah well.
320 osc_obd->obd_force = obd->obd_force;
321 osc_obd->obd_fail = obd->obd_fail;
322 osc_obd->obd_no_recov = obd->obd_no_recov;
325 obd_register_observer(osc_obd, NULL);
327 rc = obd_disconnect(tgt->ltd_exp);
329 CERROR("Target %s disconnect error %d\n",
330 tgt->ltd_uuid.uuid, rc);
338 static int lov_disconnect(struct obd_export *exp)
340 struct obd_device *obd = class_exp2obd(exp);
341 struct lov_obd *lov = &obd->u.lov;
348 /* Only disconnect the underlying layers on the final disconnect. */
350 if (lov->lov_connects != 0) {
351 /* why should there be more than 1 connect? */
352 CERROR("disconnect #%d\n", lov->lov_connects);
356 /* Let's hold another reference so lov_del_obd doesn't spin through
360 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
361 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
362 /* Disconnection is the last we know about an obd */
363 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
369 rc = class_disconnect(exp); /* bz 9811 */
375 * -EINVAL : UUID can't be found in the LOV's target list
376 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
377 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
378 * any >= 0 : is log target index
380 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
381 enum obd_notify_event ev)
383 struct lov_obd *lov = &obd->u.lov;
384 struct lov_tgt_desc *tgt;
385 int index, activate, active;
388 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
389 lov, uuid->uuid, ev);
392 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
393 tgt = lov->lov_tgts[index];
394 if (!tgt || !tgt->ltd_exp)
397 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
398 index, obd_uuid2str(&tgt->ltd_uuid),
399 tgt->ltd_exp->exp_handle.h_cookie);
400 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
404 if (index == lov->desc.ld_tgt_count)
405 GOTO(out, index = -EINVAL);
407 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
408 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
410 if (lov->lov_tgts[index]->ltd_activate == activate) {
411 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
412 uuid->uuid, activate ? "" : "de");
414 lov->lov_tgts[index]->ltd_activate = activate;
415 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
416 activate ? "" : "de", obd_uuid2str(uuid));
419 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
420 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
422 if (lov->lov_tgts[index]->ltd_active == active) {
423 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
424 uuid->uuid, active ? "" : "in");
427 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
428 obd_uuid2str(uuid), active ? "" : "in");
431 lov->lov_tgts[index]->ltd_active = active;
433 lov->desc.ld_active_tgt_count++;
434 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
436 lov->desc.ld_active_tgt_count--;
437 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
440 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
448 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
449 enum obd_notify_event ev, void *data)
454 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
455 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
456 struct obd_uuid *uuid;
460 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
461 CERROR("unexpected notification of %s %s!\n",
462 watched->obd_type->typ_name,
466 uuid = &watched->u.cli.cl_target_uuid;
468 /* Set OSC as active before notifying the observer, so the
469 * observer can use the OSC normally.
471 rc = lov_set_osc_active(obd, uuid, ev);
473 CERROR("event(%d) of %s failed: %d\n", ev,
474 obd_uuid2str(uuid), rc);
477 /* active event should be pass lov target index as data */
481 /* Pass the notification up the chain. */
483 rc = obd_notify_observer(obd, watched, ev, data);
485 /* NULL watched means all osc's in the lov (only for syncs) */
486 /* sync event should be send lov idx as data */
487 struct lov_obd *lov = &obd->u.lov;
491 is_sync = (ev == OBD_NOTIFY_SYNC) ||
492 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
495 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
496 if (!lov->lov_tgts[i])
499 /* don't send sync event if target not
500 * connected/activated */
501 if (is_sync && !lov->lov_tgts[i]->ltd_active)
504 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
507 CERROR("%s: notify %s of %s failed %d\n",
509 obd->obd_observer->obd_name,
510 lov->lov_tgts[i]->ltd_obd->obd_name,
520 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
521 __u32 index, int gen, int active)
523 struct lov_obd *lov = &obd->u.lov;
524 struct lov_tgt_desc *tgt;
525 struct obd_device *tgt_obd;
529 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
530 uuidp->uuid, index, gen, active);
533 CERROR("request to add OBD %s with invalid generation: %d\n",
538 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
543 mutex_lock(&lov->lov_lock);
545 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
546 tgt = lov->lov_tgts[index];
547 CERROR("UUID %s already assigned at LOV target index %d\n",
548 obd_uuid2str(&tgt->ltd_uuid), index);
549 mutex_unlock(&lov->lov_lock);
553 if (index >= lov->lov_tgt_size) {
554 /* We need to reallocate the lov target array. */
555 struct lov_tgt_desc **newtgts, **old = NULL;
556 __u32 newsize, oldsize = 0;
558 newsize = max(lov->lov_tgt_size, (__u32)2);
559 while (newsize < index + 1)
560 newsize = newsize << 1;
561 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
562 if (newtgts == NULL) {
563 mutex_unlock(&lov->lov_lock);
567 if (lov->lov_tgt_size) {
568 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
571 oldsize = lov->lov_tgt_size;
574 lov->lov_tgts = newtgts;
575 lov->lov_tgt_size = newsize;
578 OBD_FREE(old, sizeof(*old) * oldsize);
580 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
581 lov->lov_tgts, lov->lov_tgt_size);
586 mutex_unlock(&lov->lov_lock);
590 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
592 mutex_unlock(&lov->lov_lock);
597 tgt->ltd_uuid = *uuidp;
598 tgt->ltd_obd = tgt_obd;
599 /* XXX - add a sanity check on the generation number. */
601 tgt->ltd_index = index;
602 tgt->ltd_activate = active;
603 lov->lov_tgts[index] = tgt;
604 if (index >= lov->desc.ld_tgt_count)
605 lov->desc.ld_tgt_count = index + 1;
607 mutex_unlock(&lov->lov_lock);
609 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
610 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
612 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
614 if (lov->lov_connects == 0) {
615 /* lov_connect hasn't been called yet. We'll do the
616 lov_connect_obd on this target when that fn first runs,
617 because we don't know the connect flags yet. */
623 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
627 /* connect to administrative disabled ost */
631 if (lov->lov_cache != NULL) {
632 rc = obd_set_info_async(NULL, tgt->ltd_exp,
633 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
634 sizeof(struct cl_client_cache), lov->lov_cache,
640 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
641 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
646 CERROR("add failed (%d), deleting %s\n", rc,
647 obd_uuid2str(&tgt->ltd_uuid));
648 lov_del_target(obd, index, 0, 0);
654 /* Schedule a target for deletion */
655 int lov_del_target(struct obd_device *obd, __u32 index,
656 struct obd_uuid *uuidp, int gen)
658 struct lov_obd *lov = &obd->u.lov;
659 int count = lov->desc.ld_tgt_count;
663 if (index >= count) {
664 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
671 if (!lov->lov_tgts[index]) {
672 CERROR("LOV target at index %d is not setup.\n", index);
673 GOTO(out, rc = -EINVAL);
676 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
677 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
678 lov_uuid2str(lov, index), index,
679 obd_uuid2str(uuidp));
680 GOTO(out, rc = -EINVAL);
683 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
684 lov_uuid2str(lov, index), index,
685 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
686 lov->lov_tgts[index]->ltd_active);
688 lov->lov_tgts[index]->ltd_reap = 1;
689 lov->lov_death_row++;
690 /* we really delete it from obd_putref */
697 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
699 struct obd_device *osc_obd;
702 LASSERT(tgt->ltd_reap);
704 osc_obd = class_exp2obd(tgt->ltd_exp);
706 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
708 osc_obd ? osc_obd->obd_name : "<no obd>");
711 lov_disconnect_obd(obd, tgt);
715 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
716 do it ourselves. And we can't do it from lov_cleanup,
717 because we just lost our only reference to it. */
719 class_manual_cleanup(osc_obd);
722 void lov_fix_desc_stripe_size(__u64 *val)
724 if (*val < PTLRPC_MAX_BRW_SIZE) {
725 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
726 PTLRPC_MAX_BRW_SIZE);
727 *val = PTLRPC_MAX_BRW_SIZE;
728 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
729 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
730 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
732 *val, LOV_MIN_STRIPE_SIZE);
736 void lov_fix_desc_stripe_count(__u32 *val)
742 void lov_fix_desc_pattern(__u32 *val)
744 /* from lov_setstripe */
745 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
746 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
751 void lov_fix_desc_qos_maxage(__u32 *val)
755 *val = QOS_DEFAULT_MAXAGE;
758 void lov_fix_desc(struct lov_desc *desc)
760 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
761 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
762 lov_fix_desc_pattern(&desc->ld_pattern);
763 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
766 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
768 struct lprocfs_static_vars lvars = { 0 };
769 struct lov_desc *desc;
770 struct lov_obd *lov = &obd->u.lov;
774 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
775 CERROR("LOV setup requires a descriptor\n");
779 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
781 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
782 CERROR("descriptor size wrong: %d > %d\n",
783 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
787 if (desc->ld_magic != LOV_DESC_MAGIC) {
788 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
789 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
790 obd->obd_name, desc);
791 lustre_swab_lov_desc(desc);
793 CERROR("%s: Bad lov desc magic: %#x\n",
794 obd->obd_name, desc->ld_magic);
801 desc->ld_active_tgt_count = 0;
803 lov->lov_tgt_size = 0;
805 mutex_init(&lov->lov_lock);
806 cfs_atomic_set(&lov->lov_refcount, 0);
807 lov->lov_sp_me = LUSTRE_SP_CLI;
809 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
811 HASH_POOLS_BKT_BITS, 0,
814 &pool_hash_operations,
816 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
817 lov->lov_pool_count = 0;
818 rc = lov_ost_pool_init(&lov->lov_packed, 0);
822 lprocfs_lov_init_vars(&lvars);
823 lprocfs_obd_setup(obd, lvars.obd_vars);
828 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
829 0444, &lov_proc_target_fops, obd);
831 CWARN("Error adding the target_obd file\n");
834 lov->lov_pool_proc_entry = lprocfs_register("pools",
844 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
847 struct lov_obd *lov = &obd->u.lov;
852 case OBD_CLEANUP_EARLY: {
854 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
855 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
857 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
862 case OBD_CLEANUP_EXPORTS:
863 rc = obd_llog_finish(obd, 0);
865 CERROR("failed to cleanup llogging subsystems\n");
871 static int lov_cleanup(struct obd_device *obd)
873 struct lov_obd *lov = &obd->u.lov;
874 cfs_list_t *pos, *tmp;
875 struct pool_desc *pool;
878 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
879 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
880 /* free pool structs */
881 CDEBUG(D_INFO, "delete pool %p\n", pool);
882 /* In the function below, .hs_keycmp resolves to
883 * pool_hashkey_keycmp() */
884 /* coverity[overrun-buffer-val] */
885 lov_pool_del(obd, pool->pool_name);
887 cfs_hash_putref(lov->lov_pools_hash_body);
888 lov_ost_pool_free(&lov->lov_packed);
890 lprocfs_obd_cleanup(obd);
894 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
895 if (!lov->lov_tgts[i])
898 /* Inactive targets may never have connected */
899 if (lov->lov_tgts[i]->ltd_active ||
900 cfs_atomic_read(&lov->lov_refcount))
901 /* We should never get here - these
902 should have been removed in the
904 CERROR("lov tgt %d not cleaned!"
905 " deathrow=%d, lovrc=%d\n",
906 i, lov->lov_death_row,
907 cfs_atomic_read(&lov->lov_refcount));
908 lov_del_target(obd, i, 0, 0);
911 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
913 lov->lov_tgt_size = 0;
918 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
919 __u32 *indexp, int *genp)
921 struct obd_uuid obd_uuid;
926 switch(cmd = lcfg->lcfg_command) {
927 case LCFG_LOV_ADD_OBD:
928 case LCFG_LOV_ADD_INA:
929 case LCFG_LOV_DEL_OBD: {
932 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
933 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
934 GOTO(out, rc = -EINVAL);
936 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
938 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
939 GOTO(out, rc = -EINVAL);
940 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
941 GOTO(out, rc = -EINVAL);
944 if (cmd == LCFG_LOV_ADD_OBD)
945 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
946 else if (cmd == LCFG_LOV_ADD_INA)
947 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
949 rc = lov_del_target(obd, index, &obd_uuid, gen);
953 struct lprocfs_static_vars lvars = { 0 };
954 struct lov_desc *desc = &(obd->u.lov.desc);
957 GOTO(out, rc = -EINVAL);
959 lprocfs_lov_init_vars(&lvars);
961 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
974 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
975 GOTO(out, rc = -EINVAL);
983 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
984 struct lov_stripe_md **ea, struct obd_trans_info *oti)
986 struct lov_stripe_md *obj_mdp, *lsm;
987 struct lov_obd *lov = &exp->exp_obd->u.lov;
992 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
993 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
995 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
999 ost_idx = src_oa->o_nlink;
1002 GOTO(out, rc = -EINVAL);
1003 if (ost_idx >= lov->desc.ld_tgt_count ||
1004 !lov->lov_tgts[ost_idx])
1005 GOTO(out, rc = -EINVAL);
1007 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1008 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1009 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1010 GOTO(out, rc = -EINVAL);
1014 if (i == lsm->lsm_stripe_count)
1015 GOTO(out, rc = -EINVAL);
1017 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1018 src_oa, &obj_mdp, oti);
1020 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1024 /* the LOV expects oa->o_id to be set to the LOV object id */
1025 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1026 struct obdo *src_oa, struct lov_stripe_md **ea,
1027 struct obd_trans_info *oti)
1029 struct lov_obd *lov;
1033 LASSERT(ea != NULL);
1037 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1038 src_oa->o_flags == OBD_FL_DELORPHAN) {
1039 /* should be used with LOV anymore */
1043 lov = &exp->exp_obd->u.lov;
1044 if (!lov->desc.ld_active_tgt_count)
1047 obd_getref(exp->exp_obd);
1048 /* Recreate a specific object id at the given OST index */
1049 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1050 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1051 rc = lov_recreate(exp, src_oa, ea, oti);
1054 obd_putref(exp->exp_obd);
1058 #define ASSERT_LSM_MAGIC(lsmp) \
1060 LASSERT((lsmp) != NULL); \
1061 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1062 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1063 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1066 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1067 struct obdo *oa, struct lov_stripe_md *lsm,
1068 struct obd_trans_info *oti, struct obd_export *md_exp,
1071 struct lov_request_set *set;
1072 struct obd_info oinfo;
1073 struct lov_request *req;
1075 struct lov_obd *lov;
1076 int rc = 0, err = 0;
1079 ASSERT_LSM_MAGIC(lsm);
1081 if (!exp || !exp->exp_obd)
1084 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1086 LASSERT(oti->oti_logcookies);
1089 lov = &exp->exp_obd->u.lov;
1090 obd_getref(exp->exp_obd);
1091 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1095 cfs_list_for_each (pos, &set->set_list) {
1096 req = cfs_list_entry(pos, struct lov_request, rq_link);
1098 if (oa->o_valid & OBD_MD_FLCOOKIE)
1099 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1101 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1102 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1103 err = lov_update_common_set(set, req, err);
1105 CERROR("error: destroying objid "LPX64" subobj "
1106 LPX64" on OST idx %d: rc = %d\n",
1107 oa->o_id, req->rq_oi.oi_oa->o_id,
1115 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1116 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1118 err = lov_fini_destroy_set(set);
1120 obd_putref(exp->exp_obd);
1121 RETURN(rc ? rc : err);
1124 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1125 struct obd_info *oinfo)
1127 struct lov_request_set *set;
1128 struct lov_request *req;
1130 struct lov_obd *lov;
1131 int err = 0, rc = 0;
1135 ASSERT_LSM_MAGIC(oinfo->oi_md);
1137 if (!exp || !exp->exp_obd)
1140 lov = &exp->exp_obd->u.lov;
1142 rc = lov_prep_getattr_set(exp, oinfo, &set);
1146 cfs_list_for_each (pos, &set->set_list) {
1147 req = cfs_list_entry(pos, struct lov_request, rq_link);
1149 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1150 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1151 req->rq_oi.oi_oa->o_id, req->rq_idx);
1153 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1155 err = lov_update_common_set(set, req, rc);
1157 CERROR("error: getattr objid "LPX64" subobj "
1158 LPX64" on OST idx %d: rc = %d\n",
1159 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1165 rc = lov_fini_getattr_set(set);
1171 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1174 struct lov_request_set *lovset = (struct lov_request_set *)data;
1178 /* don't do attribute merge if this aysnc op failed */
1180 cfs_atomic_set(&lovset->set_completes, 0);
1181 err = lov_fini_getattr_set(lovset);
1182 RETURN(rc ? rc : err);
1185 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1186 struct ptlrpc_request_set *rqset)
1188 struct lov_request_set *lovset;
1189 struct lov_obd *lov;
1191 struct lov_request *req;
1196 ASSERT_LSM_MAGIC(oinfo->oi_md);
1198 if (!exp || !exp->exp_obd)
1201 lov = &exp->exp_obd->u.lov;
1203 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1207 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1208 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1209 oinfo->oi_md->lsm_stripe_size);
1211 cfs_list_for_each (pos, &lovset->set_list) {
1212 req = cfs_list_entry(pos, struct lov_request, rq_link);
1214 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1215 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1216 req->rq_oi.oi_oa->o_id, req->rq_idx);
1217 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1218 &req->rq_oi, rqset);
1220 CERROR("error: getattr objid "LPX64" subobj "
1221 LPX64" on OST idx %d: rc = %d\n",
1222 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1228 if (!cfs_list_empty(&rqset->set_requests)) {
1230 LASSERT (rqset->set_interpret == NULL);
1231 rqset->set_interpret = lov_getattr_interpret;
1232 rqset->set_arg = (void *)lovset;
1237 cfs_atomic_set(&lovset->set_completes, 0);
1238 err = lov_fini_getattr_set(lovset);
1239 RETURN(rc ? rc : err);
1242 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1243 struct obd_info *oinfo, struct obd_trans_info *oti)
1245 struct lov_request_set *set;
1246 struct lov_obd *lov;
1248 struct lov_request *req;
1249 int err = 0, rc = 0;
1253 ASSERT_LSM_MAGIC(oinfo->oi_md);
1255 if (!exp || !exp->exp_obd)
1258 /* for now, we only expect the following updates here */
1259 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1260 OBD_MD_FLMODE | OBD_MD_FLATIME |
1261 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1262 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1263 OBD_MD_FLGROUP | OBD_MD_FLUID |
1264 OBD_MD_FLGID | OBD_MD_FLFID |
1266 lov = &exp->exp_obd->u.lov;
1267 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1271 cfs_list_for_each (pos, &set->set_list) {
1272 req = cfs_list_entry(pos, struct lov_request, rq_link);
1274 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1276 err = lov_update_setattr_set(set, req, rc);
1278 CERROR("error: setattr objid "LPX64" subobj "
1279 LPX64" on OST idx %d: rc = %d\n",
1280 set->set_oi->oi_oa->o_id,
1281 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1286 err = lov_fini_setattr_set(set);
1292 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1295 struct lov_request_set *lovset = (struct lov_request_set *)data;
1300 cfs_atomic_set(&lovset->set_completes, 0);
1301 err = lov_fini_setattr_set(lovset);
1302 RETURN(rc ? rc : err);
1305 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1306 needed. Otherwise, a client is waiting for responses. */
1307 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1308 struct obd_trans_info *oti,
1309 struct ptlrpc_request_set *rqset)
1311 struct lov_request_set *set;
1312 struct lov_request *req;
1314 struct lov_obd *lov;
1319 ASSERT_LSM_MAGIC(oinfo->oi_md);
1320 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1322 LASSERT(oti->oti_logcookies);
1325 if (!exp || !exp->exp_obd)
1328 lov = &exp->exp_obd->u.lov;
1329 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1333 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1334 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1335 oinfo->oi_md->lsm_stripe_size);
1337 cfs_list_for_each (pos, &set->set_list) {
1338 req = cfs_list_entry(pos, struct lov_request, rq_link);
1340 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1341 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1343 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1344 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1345 req->rq_oi.oi_oa->o_id, req->rq_idx);
1347 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1348 &req->rq_oi, oti, rqset);
1350 CERROR("error: setattr objid "LPX64" subobj "
1351 LPX64" on OST idx %d: rc = %d\n",
1352 set->set_oi->oi_oa->o_id,
1353 req->rq_oi.oi_oa->o_id,
1359 /* If we are not waiting for responses on async requests, return. */
1360 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1363 cfs_atomic_set(&set->set_completes, 0);
1364 err = lov_fini_setattr_set(set);
1365 RETURN(rc ? rc : err);
1368 LASSERT(rqset->set_interpret == NULL);
1369 rqset->set_interpret = lov_setattr_interpret;
1370 rqset->set_arg = (void *)set;
1375 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1378 struct lov_request_set *lovset = (struct lov_request_set *)data;
1383 cfs_atomic_set(&lovset->set_completes, 0);
1384 err = lov_fini_punch_set(lovset);
1385 RETURN(rc ? rc : err);
1388 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1389 * we can send this 'punch' to just the authoritative node and the nodes
1390 * that the punch will affect. */
1391 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1392 struct obd_info *oinfo, struct obd_trans_info *oti,
1393 struct ptlrpc_request_set *rqset)
1395 struct lov_request_set *set;
1396 struct lov_obd *lov;
1398 struct lov_request *req;
1403 ASSERT_LSM_MAGIC(oinfo->oi_md);
1405 if (!exp || !exp->exp_obd)
1408 lov = &exp->exp_obd->u.lov;
1409 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1413 cfs_list_for_each (pos, &set->set_list) {
1414 req = cfs_list_entry(pos, struct lov_request, rq_link);
1416 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1417 &req->rq_oi, NULL, rqset);
1419 CERROR("error: punch objid "LPX64" subobj "LPX64
1420 " on OST idx %d: rc = %d\n",
1421 set->set_oi->oi_oa->o_id,
1422 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1427 if (rc || cfs_list_empty(&rqset->set_requests)) {
1429 err = lov_fini_punch_set(set);
1430 RETURN(rc ? rc : err);
1433 LASSERT(rqset->set_interpret == NULL);
1434 rqset->set_interpret = lov_punch_interpret;
1435 rqset->set_arg = (void *)set;
1440 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1443 struct lov_request_set *lovset = data;
1448 cfs_atomic_set(&lovset->set_completes, 0);
1449 err = lov_fini_sync_set(lovset);
1453 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1454 struct obd_info *oinfo, obd_off start, obd_off end,
1455 struct ptlrpc_request_set *rqset)
1457 struct lov_request_set *set = NULL;
1458 struct lov_obd *lov;
1460 struct lov_request *req;
1464 ASSERT_LSM_MAGIC(oinfo->oi_md);
1465 LASSERT(rqset != NULL);
1470 lov = &exp->exp_obd->u.lov;
1471 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1475 CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1476 set->set_oi->oi_oa->o_id, start, end);
1478 cfs_list_for_each (pos, &set->set_list) {
1479 req = cfs_list_entry(pos, struct lov_request, rq_link);
1481 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1482 &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1483 req->rq_oi.oi_policy.l_extent.end, rqset);
1485 CERROR("error: fsync objid "LPX64" subobj "LPX64
1486 " on OST idx %d: rc = %d\n",
1487 set->set_oi->oi_oa->o_id,
1488 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1493 /* If we are not waiting for responses on async requests, return. */
1494 if (rc || cfs_list_empty(&rqset->set_requests)) {
1495 int err = lov_fini_sync_set(set);
1500 LASSERT(rqset->set_interpret == NULL);
1501 rqset->set_interpret = lov_sync_interpret;
1502 rqset->set_arg = (void *)set;
1507 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1508 obd_count oa_bufs, struct brw_page *pga)
1510 struct obd_info oinfo = { { { 0 } } };
1513 oinfo.oi_oa = lov_oinfo->oi_oa;
1515 /* The caller just wants to know if there's a chance that this
1516 * I/O can succeed */
1517 for (i = 0; i < oa_bufs; i++) {
1518 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1519 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1522 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1523 pga[i].off + pga[i].count - 1,
1527 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1528 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1532 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1540 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1541 obd_count oa_bufs, struct brw_page *pga,
1542 struct obd_trans_info *oti)
1544 struct lov_request_set *set;
1545 struct lov_request *req;
1547 struct lov_obd *lov = &exp->exp_obd->u.lov;
1551 ASSERT_LSM_MAGIC(oinfo->oi_md);
1553 if (cmd == OBD_BRW_CHECK) {
1554 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1558 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1562 cfs_list_for_each (pos, &set->set_list) {
1563 struct obd_export *sub_exp;
1564 struct brw_page *sub_pga;
1565 req = cfs_list_entry(pos, struct lov_request, rq_link);
1567 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1568 sub_pga = set->set_pga + req->rq_pgaidx;
1569 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1573 lov_update_common_set(set, req, rc);
1576 err = lov_fini_brw_set(set);
1582 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1585 struct lov_request_set *lovset = (struct lov_request_set *)data;
1587 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1591 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1592 struct ldlm_enqueue_info *einfo,
1593 struct ptlrpc_request_set *rqset)
1595 ldlm_mode_t mode = einfo->ei_mode;
1596 struct lov_request_set *set;
1597 struct lov_request *req;
1599 struct lov_obd *lov;
1604 ASSERT_LSM_MAGIC(oinfo->oi_md);
1605 LASSERT(mode == (mode & -mode));
1607 /* we should never be asked to replay a lock this way. */
1608 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1610 if (!exp || !exp->exp_obd)
1613 lov = &exp->exp_obd->u.lov;
1614 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1618 cfs_list_for_each (pos, &set->set_list) {
1619 req = cfs_list_entry(pos, struct lov_request, rq_link);
1621 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1622 &req->rq_oi, einfo, rqset);
1627 if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1629 LASSERT(rqset->set_interpret == NULL);
1630 rqset->set_interpret = lov_enqueue_interpret;
1631 rqset->set_arg = (void *)set;
1635 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1639 static int lov_change_cbdata(struct obd_export *exp,
1640 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1643 struct lov_obd *lov;
1647 ASSERT_LSM_MAGIC(lsm);
1649 if (!exp || !exp->exp_obd)
1652 lov = &exp->exp_obd->u.lov;
1653 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1654 struct lov_stripe_md submd;
1655 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1657 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1658 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1662 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1663 submd.lsm_object_id = loi->loi_id;
1664 submd.lsm_object_seq = loi->loi_seq;
1665 submd.lsm_stripe_count = 0;
1666 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1672 /* find any ldlm lock of the inode in lov
1676 static int lov_find_cbdata(struct obd_export *exp,
1677 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1680 struct lov_obd *lov;
1684 ASSERT_LSM_MAGIC(lsm);
1686 if (!exp || !exp->exp_obd)
1689 lov = &exp->exp_obd->u.lov;
1690 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1691 struct lov_stripe_md submd;
1692 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1694 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1695 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1699 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1700 submd.lsm_object_id = loi->loi_id;
1701 submd.lsm_object_seq = loi->loi_seq;
1702 submd.lsm_stripe_count = 0;
1703 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1711 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1712 __u32 mode, struct lustre_handle *lockh)
1714 struct lov_request_set *set;
1715 struct obd_info oinfo;
1716 struct lov_request *req;
1718 struct lov_obd *lov;
1719 struct lustre_handle *lov_lockhp;
1720 int err = 0, rc = 0;
1723 ASSERT_LSM_MAGIC(lsm);
1725 if (!exp || !exp->exp_obd)
1728 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1730 lov = &exp->exp_obd->u.lov;
1731 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1735 cfs_list_for_each (pos, &set->set_list) {
1736 req = cfs_list_entry(pos, struct lov_request, rq_link);
1737 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1739 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1740 req->rq_oi.oi_md, mode, lov_lockhp);
1741 rc = lov_update_common_set(set, req, rc);
1743 CERROR("error: cancel objid "LPX64" subobj "
1744 LPX64" on OST idx %d: rc = %d\n",
1746 req->rq_oi.oi_md->lsm_object_id,
1752 lov_fini_cancel_set(set);
1756 static int lov_cancel_unused(struct obd_export *exp,
1757 struct lov_stripe_md *lsm,
1758 ldlm_cancel_flags_t flags, void *opaque)
1760 struct lov_obd *lov;
1764 if (!exp || !exp->exp_obd)
1767 lov = &exp->exp_obd->u.lov;
1769 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1771 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1774 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1782 ASSERT_LSM_MAGIC(lsm);
1784 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1785 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1786 struct lov_stripe_md submd;
1787 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1790 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1791 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1795 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1796 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1798 submd.lsm_object_id = loi->loi_id;
1799 submd.lsm_object_seq = loi->loi_seq;
1800 submd.lsm_stripe_count = 0;
1801 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1802 &submd, flags, opaque);
1803 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1804 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1805 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1806 loi->loi_id, loi->loi_ost_idx, err);
1814 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1816 struct lov_request_set *lovset = (struct lov_request_set *)data;
1821 cfs_atomic_set(&lovset->set_completes, 0);
1823 err = lov_fini_statfs_set(lovset);
1824 RETURN(rc ? rc : err);
1827 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1828 __u64 max_age, struct ptlrpc_request_set *rqset)
1830 struct obd_device *obd = class_exp2obd(exp);
1831 struct lov_request_set *set;
1832 struct lov_request *req;
1834 struct lov_obd *lov;
1838 LASSERT(oinfo != NULL);
1839 LASSERT(oinfo->oi_osfs != NULL);
1842 rc = lov_prep_statfs_set(obd, oinfo, &set);
1846 cfs_list_for_each (pos, &set->set_list) {
1847 req = cfs_list_entry(pos, struct lov_request, rq_link);
1848 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1849 &req->rq_oi, max_age, rqset);
1854 if (rc || cfs_list_empty(&rqset->set_requests)) {
1857 cfs_atomic_set(&set->set_completes, 0);
1858 err = lov_fini_statfs_set(set);
1859 RETURN(rc ? rc : err);
1862 LASSERT(rqset->set_interpret == NULL);
1863 rqset->set_interpret = lov_statfs_interpret;
1864 rqset->set_arg = (void *)set;
1868 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1869 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1871 struct ptlrpc_request_set *set = NULL;
1872 struct obd_info oinfo = { { { 0 } } };
1877 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1878 * statfs requests */
1879 set = ptlrpc_prep_set();
1883 oinfo.oi_osfs = osfs;
1884 oinfo.oi_flags = flags;
1885 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1887 rc = ptlrpc_set_wait(set);
1888 ptlrpc_set_destroy(set);
1893 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1894 void *karg, void *uarg)
1896 struct obd_device *obddev = class_exp2obd(exp);
1897 struct lov_obd *lov = &obddev->u.lov;
1898 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1899 struct obd_uuid *uuidp;
1903 case IOC_OBD_STATFS: {
1904 struct obd_ioctl_data *data = karg;
1905 struct obd_device *osc_obd;
1906 struct obd_statfs stat_buf = {0};
1910 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1911 if ((index >= count))
1914 if (!lov->lov_tgts[index])
1915 /* Try again with the next index */
1917 if (!lov->lov_tgts[index]->ltd_active)
1920 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1925 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1926 min((int) data->ioc_plen2,
1927 (int) sizeof(struct obd_uuid))))
1930 flags = uarg ? *(__u32*)uarg : 0;
1931 /* got statfs data */
1932 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1933 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1937 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1938 min((int) data->ioc_plen1,
1939 (int) sizeof(stat_buf))))
1943 case OBD_IOC_LOV_GET_CONFIG: {
1944 struct obd_ioctl_data *data;
1945 struct lov_desc *desc;
1950 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1953 data = (struct obd_ioctl_data *)buf;
1955 if (sizeof(*desc) > data->ioc_inllen1) {
1956 obd_ioctl_freedata(buf, len);
1960 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1961 obd_ioctl_freedata(buf, len);
1965 if (sizeof(__u32) * count > data->ioc_inllen3) {
1966 obd_ioctl_freedata(buf, len);
1970 desc = (struct lov_desc *)data->ioc_inlbuf1;
1971 memcpy(desc, &(lov->desc), sizeof(*desc));
1973 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1974 genp = (__u32 *)data->ioc_inlbuf3;
1975 /* the uuid will be empty for deleted OSTs */
1976 for (i = 0; i < count; i++, uuidp++, genp++) {
1977 if (!lov->lov_tgts[i])
1979 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1980 *genp = lov->lov_tgts[i]->ltd_gen;
1983 if (cfs_copy_to_user((void *)uarg, buf, len))
1985 obd_ioctl_freedata(buf, len);
1988 case LL_IOC_LOV_SETSTRIPE:
1989 rc = lov_setstripe(exp, len, karg, uarg);
1991 case LL_IOC_LOV_GETSTRIPE:
1992 rc = lov_getstripe(exp, karg, uarg);
1994 case LL_IOC_LOV_SETEA:
1995 rc = lov_setea(exp, karg, uarg);
1997 case OBD_IOC_QUOTACTL: {
1998 struct if_quotactl *qctl = karg;
1999 struct lov_tgt_desc *tgt = NULL;
2000 struct obd_quotactl *oqctl;
2002 if (qctl->qc_valid == QC_OSTIDX) {
2003 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2006 tgt = lov->lov_tgts[qctl->qc_idx];
2007 if (!tgt || !tgt->ltd_exp)
2009 } else if (qctl->qc_valid == QC_UUID) {
2010 for (i = 0; i < count; i++) {
2011 tgt = lov->lov_tgts[i];
2013 !obd_uuid_equals(&tgt->ltd_uuid,
2017 if (tgt->ltd_exp == NULL)
2029 LASSERT(tgt && tgt->ltd_exp);
2030 OBD_ALLOC_PTR(oqctl);
2034 QCTL_COPY(oqctl, qctl);
2035 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2037 QCTL_COPY(qctl, oqctl);
2038 qctl->qc_valid = QC_OSTIDX;
2039 qctl->obd_uuid = tgt->ltd_uuid;
2041 OBD_FREE_PTR(oqctl);
2050 for (i = 0; i < count; i++) {
2052 struct obd_device *osc_obd;
2054 /* OST was disconnected */
2055 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2058 /* ll_umount_begin() sets force flag but for lov, not
2059 * osc. Let's pass it through */
2060 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2061 osc_obd->obd_force = obddev->obd_force;
2062 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2064 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2067 if (lov->lov_tgts[i]->ltd_active) {
2068 CDEBUG(err == -ENOTTY ?
2069 D_IOCTL : D_WARNING,
2070 "iocontrol OSC %s on OST "
2071 "idx %d cmd %x: err = %d\n",
2072 lov_uuid2str(lov, i),
2089 #define FIEMAP_BUFFER_SIZE 4096
2092 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2093 * call. The local end offset and the device are sent in the first
2094 * fm_extent. This function calculates the stripe number from the index.
2095 * This function returns a stripe_no on which mapping is to be restarted.
2097 * This function returns fm_end_offset which is the in-OST offset at which
2098 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2099 * will re-calculate proper offset in next stripe.
2100 * Note that the first extent is passed to lov_get_info via the value field.
2102 * \param fiemap fiemap request header
2103 * \param lsm striping information for the file
2104 * \param fm_start logical start of mapping
2105 * \param fm_end logical end of mapping
2106 * \param start_stripe starting stripe will be returned in this
2108 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2109 struct lov_stripe_md *lsm, obd_size fm_start,
2110 obd_size fm_end, int *start_stripe)
2112 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2113 obd_off lun_start, lun_end;
2114 obd_size fm_end_offset;
2115 int stripe_no = -1, i;
2117 if (fiemap->fm_extent_count == 0 ||
2118 fiemap->fm_extents[0].fe_logical == 0)
2121 /* Find out stripe_no from ost_index saved in the fe_device */
2122 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2123 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2124 fiemap->fm_extents[0].fe_device) {
2129 if (stripe_no == -1)
2132 /* If we have finished mapping on previous device, shift logical
2133 * offset to start of next device */
2134 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2135 &lun_start, &lun_end)) != 0 &&
2136 local_end < lun_end) {
2137 fm_end_offset = local_end;
2138 *start_stripe = stripe_no;
2140 /* This is a special value to indicate that caller should
2141 * calculate offset in next stripe. */
2143 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2146 return fm_end_offset;
2150 * We calculate on which OST the mapping will end. If the length of mapping
2151 * is greater than (stripe_size * stripe_count) then the last_stripe will
2152 * will be one just before start_stripe. Else we check if the mapping
2153 * intersects each OST and find last_stripe.
2154 * This function returns the last_stripe and also sets the stripe_count
2155 * over which the mapping is spread
2157 * \param lsm striping information for the file
2158 * \param fm_start logical start of mapping
2159 * \param fm_end logical end of mapping
2160 * \param start_stripe starting stripe of the mapping
2161 * \param stripe_count the number of stripes across which to map is returned
2163 * \retval last_stripe return the last stripe of the mapping
2165 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2166 obd_size fm_end, int start_stripe,
2170 obd_off obd_start, obd_end;
2173 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2174 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2176 *stripe_count = lsm->lsm_stripe_count;
2178 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2179 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2180 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2181 &obd_start, &obd_end)) == 0)
2185 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2192 * Set fe_device and copy extents from local buffer into main return buffer.
2194 * \param fiemap fiemap request header
2195 * \param lcl_fm_ext array of local fiemap extents to be copied
2196 * \param ost_index OST index to be written into the fm_device field for each
2198 * \param ext_count number of extents to be copied
2199 * \param current_extent where to start copying in main extent array
2201 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2202 struct ll_fiemap_extent *lcl_fm_ext,
2203 int ost_index, unsigned int ext_count,
2209 for (ext = 0; ext < ext_count; ext++) {
2210 lcl_fm_ext[ext].fe_device = ost_index;
2211 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2214 /* Copy fm_extent's from fm_local to return buffer */
2215 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2216 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2220 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2221 * This also handles the restarting of FIEMAP calls in case mapping overflows
2222 * the available number of extents in single call.
2224 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2225 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2227 struct ll_fiemap_info_key *fm_key = key;
2228 struct ll_user_fiemap *fiemap = val;
2229 struct ll_user_fiemap *fm_local = NULL;
2230 struct ll_fiemap_extent *lcl_fm_ext;
2232 unsigned int get_num_extents = 0;
2233 int ost_index = 0, actual_start_stripe, start_stripe;
2234 obd_size fm_start, fm_end, fm_length, fm_end_offset;
2236 int current_extent = 0, rc = 0, i;
2237 int ost_eof = 0; /* EOF for object */
2238 int ost_done = 0; /* done with required mapping for this OST? */
2240 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2241 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2246 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2247 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2249 OBD_ALLOC_LARGE(fm_local, buffer_size);
2250 if (fm_local == NULL)
2251 GOTO(out, rc = -ENOMEM);
2252 lcl_fm_ext = &fm_local->fm_extents[0];
2254 count_local = fiemap_size_to_count(buffer_size);
2256 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2257 fm_start = fiemap->fm_start;
2258 fm_length = fiemap->fm_length;
2259 /* Calculate start stripe, last stripe and length of mapping */
2260 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2261 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2262 fm_start + fm_length - 1);
2263 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2264 if (fm_end > fm_key->oa.o_size)
2265 fm_end = fm_key->oa.o_size;
2267 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2268 actual_start_stripe, &stripe_count);
2270 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2271 fm_end, &start_stripe);
2272 if (fm_end_offset == -EINVAL)
2273 GOTO(out, rc = -EINVAL);
2275 if (fiemap->fm_extent_count == 0) {
2276 get_num_extents = 1;
2280 /* Check each stripe */
2281 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2282 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2283 obd_size req_fm_len; /* Stores length of required mapping */
2284 obd_size len_mapped_single_call;
2285 obd_off lun_start, lun_end, obd_object_end;
2286 unsigned int ext_count;
2288 cur_stripe_wrap = cur_stripe;
2290 /* Find out range of mapping on this stripe */
2291 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2292 &lun_start, &obd_object_end)) == 0)
2295 /* If this is a continuation FIEMAP call and we are on
2296 * starting stripe then lun_start needs to be set to
2298 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2299 lun_start = fm_end_offset;
2301 if (fm_length != ~0ULL) {
2302 /* Handle fm_start + fm_length overflow */
2303 if (fm_start + fm_length < fm_start)
2304 fm_length = ~0ULL - fm_start;
2305 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2311 if (lun_start == lun_end)
2314 req_fm_len = obd_object_end - lun_start;
2315 fm_local->fm_length = 0;
2316 len_mapped_single_call = 0;
2318 /* If the output buffer is very large and the objects have many
2319 * extents we may need to loop on a single OST repeatedly */
2323 if (get_num_extents == 0) {
2324 /* Don't get too many extents. */
2325 if (current_extent + count_local >
2326 fiemap->fm_extent_count)
2327 count_local = fiemap->fm_extent_count -
2331 lun_start += len_mapped_single_call;
2332 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2333 req_fm_len = fm_local->fm_length;
2334 fm_local->fm_extent_count = count_local;
2335 fm_local->fm_mapped_extents = 0;
2336 fm_local->fm_flags = fiemap->fm_flags;
2338 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2339 fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2340 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2342 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2343 GOTO(out, rc = -EINVAL);
2345 /* If OST is inactive, return extent with UNKNOWN flag */
2346 if (!lov->lov_tgts[ost_index]->ltd_active) {
2347 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2348 fm_local->fm_mapped_extents = 1;
2350 lcl_fm_ext[0].fe_logical = lun_start;
2351 lcl_fm_ext[0].fe_length = obd_object_end -
2353 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2358 fm_local->fm_start = lun_start;
2359 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2360 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2361 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2362 rc = obd_get_info(NULL,
2363 lov->lov_tgts[ost_index]->ltd_exp,
2364 keylen, key, vallen, fm_local, lsm);
2369 ext_count = fm_local->fm_mapped_extents;
2370 if (ext_count == 0) {
2372 /* If last stripe has hole at the end,
2373 * then we need to return */
2374 if (cur_stripe_wrap == last_stripe) {
2375 fiemap->fm_mapped_extents = 0;
2381 /* If we just need num of extents then go to next device */
2382 if (get_num_extents) {
2383 current_extent += ext_count;
2387 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2388 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2390 /* Have we finished mapping on this device? */
2391 if (req_fm_len <= len_mapped_single_call)
2394 /* Clear the EXTENT_LAST flag which can be present on
2396 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2397 lcl_fm_ext[ext_count - 1].fe_flags &=
2398 ~FIEMAP_EXTENT_LAST;
2400 curr_loc = lov_stripe_size(lsm,
2401 lcl_fm_ext[ext_count - 1].fe_logical+
2402 lcl_fm_ext[ext_count - 1].fe_length,
2404 if (curr_loc >= fm_key->oa.o_size)
2407 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2408 ost_index, ext_count,
2411 current_extent += ext_count;
2413 /* Ran out of available extents? */
2414 if (current_extent >= fiemap->fm_extent_count)
2416 } while (ost_done == 0 && ost_eof == 0);
2418 if (cur_stripe_wrap == last_stripe)
2423 /* Indicate that we are returning device offsets unless file just has
2425 if (lsm->lsm_stripe_count > 1)
2426 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2428 if (get_num_extents)
2429 goto skip_last_device_calc;
2431 /* Check if we have reached the last stripe and whether mapping for that
2432 * stripe is done. */
2433 if (cur_stripe_wrap == last_stripe) {
2434 if (ost_done || ost_eof)
2435 fiemap->fm_extents[current_extent - 1].fe_flags |=
2439 skip_last_device_calc:
2440 fiemap->fm_mapped_extents = current_extent;
2443 OBD_FREE_LARGE(fm_local, buffer_size);
2447 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2448 __u32 keylen, void *key, __u32 *vallen, void *val,
2449 struct lov_stripe_md *lsm)
2451 struct obd_device *obddev = class_exp2obd(exp);
2452 struct lov_obd *lov = &obddev->u.lov;
2456 if (!vallen || !val)
2461 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2464 struct ldlm_lock *lock;
2466 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2467 struct lov_oinfo *loi;
2468 __u32 *stripe = val;
2470 if (*vallen < sizeof(*stripe))
2471 GOTO(out, rc = -EFAULT);
2472 *vallen = sizeof(*stripe);
2474 /* XXX This is another one of those bits that will need to
2475 * change if we ever actually support nested LOVs. It uses
2476 * the lock's export to find out which stripe it is. */
2477 /* XXX - it's assumed all the locks for deleted OSTs have
2478 * been cancelled. Also, the export for deleted OSTs will
2479 * be NULL and won't match the lock's export. */
2480 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2481 loi = lsm->lsm_oinfo[i];
2482 if (!lov->lov_tgts[loi->loi_ost_idx])
2484 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2485 data->lock->l_conn_export &&
2486 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2491 LDLM_ERROR(data->lock, "lock on inode without such object");
2492 dump_lsm(D_ERROR, lsm);
2493 GOTO(out, rc = -ENXIO);
2494 } else if (KEY_IS(KEY_LAST_ID)) {
2495 struct obd_id_info *info = val;
2496 __u32 size = sizeof(obd_id);
2497 struct lov_tgt_desc *tgt;
2499 LASSERT(*vallen == sizeof(struct obd_id_info));
2500 tgt = lov->lov_tgts[info->idx];
2502 if (!tgt || !tgt->ltd_active)
2503 GOTO(out, rc = -ESRCH);
2505 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2506 &size, info->data, NULL);
2508 } else if (KEY_IS(KEY_LOVDESC)) {
2509 struct lov_desc *desc_ret = val;
2510 *desc_ret = lov->desc;
2513 } else if (KEY_IS(KEY_FIEMAP)) {
2514 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2516 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2517 struct lov_tgt_desc *tgt;
2518 __u64 ost_idx = *((__u64*)val);
2520 LASSERT(*vallen == sizeof(__u64));
2521 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2522 tgt = lov->lov_tgts[ost_idx];
2524 if (!tgt || !tgt->ltd_exp)
2525 GOTO(out, rc = -ESRCH);
2527 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2529 } else if (KEY_IS(KEY_TGT_COUNT)) {
2530 *((int *)val) = lov->desc.ld_tgt_count;
2541 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2542 obd_count keylen, void *key, obd_count vallen,
2543 void *val, struct ptlrpc_request_set *set)
2545 struct obd_device *obddev = class_exp2obd(exp);
2546 struct lov_obd *lov = &obddev->u.lov;
2549 struct lov_tgt_desc *tgt;
2550 unsigned incr, check_uuid,
2551 do_inactive, no_set;
2552 unsigned next_id = 0, mds_con = 0, capa = 0;
2555 incr = check_uuid = do_inactive = no_set = 0;
2558 set = ptlrpc_prep_set();
2564 count = lov->desc.ld_tgt_count;
2566 if (KEY_IS(KEY_NEXT_ID)) {
2567 count = vallen / sizeof(struct obd_id_info);
2568 vallen = sizeof(obd_id);
2569 incr = sizeof(struct obd_id_info);
2572 } else if (KEY_IS(KEY_CHECKSUM)) {
2574 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2575 /* use defaults: do_inactive = incr = 0; */
2576 } else if (KEY_IS(KEY_MDS_CONN)) {
2578 } else if (KEY_IS(KEY_CAPA_KEY)) {
2580 } else if (KEY_IS(KEY_CACHE_SET)) {
2581 LASSERT(lov->lov_cache == NULL);
2582 lov->lov_cache = val;
2586 for (i = 0; i < count; i++, val = (char *)val + incr) {
2588 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2590 tgt = lov->lov_tgts[i];
2592 /* OST was disconnected */
2593 if (!tgt || !tgt->ltd_exp)
2596 /* OST is inactive and we don't want inactive OSCs */
2597 if (!tgt->ltd_active && !do_inactive)
2601 struct mds_group_info *mgi;
2603 LASSERT(vallen == sizeof(*mgi));
2604 mgi = (struct mds_group_info *)val;
2606 /* Only want a specific OSC */
2607 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2611 err = obd_set_info_async(env, tgt->ltd_exp,
2612 keylen, key, sizeof(int),
2614 } else if (next_id) {
2615 err = obd_set_info_async(env, tgt->ltd_exp,
2616 keylen, key, vallen,
2617 ((struct obd_id_info*)val)->data, set);
2619 struct mds_capa_info *info = (struct mds_capa_info*)val;
2621 LASSERT(vallen == sizeof(*info));
2623 /* Only want a specific OSC */
2625 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2628 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2629 key, sizeof(*info->capa),
2632 /* Only want a specific OSC */
2634 !obd_uuid_equals(val, &tgt->ltd_uuid))
2637 err = obd_set_info_async(env, tgt->ltd_exp,
2638 keylen, key, vallen, val, set);
2647 err = ptlrpc_set_wait(set);
2650 ptlrpc_set_destroy(set);
2655 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2656 int cmd, __u64 *offset)
2658 __u32 ssize = lsm->lsm_stripe_size;
2662 lov_do_div64(start, ssize);
2663 start = start * ssize;
2665 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2666 ", end "LPU64"\n", *offset, ssize, start,
2668 if (cmd == OBD_CALC_STRIPE_END) {
2669 *offset = start + ssize - 1;
2670 } else if (cmd == OBD_CALC_STRIPE_START) {
2679 void lov_stripe_lock(struct lov_stripe_md *md)
2681 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2682 spin_lock(&md->lsm_lock);
2683 LASSERT(md->lsm_lock_owner == 0);
2684 md->lsm_lock_owner = cfs_curproc_pid();
2686 EXPORT_SYMBOL(lov_stripe_lock);
2688 void lov_stripe_unlock(struct lov_stripe_md *md)
2690 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2691 md->lsm_lock_owner = 0;
2692 spin_unlock(&md->lsm_lock);
2694 EXPORT_SYMBOL(lov_stripe_unlock);
2696 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2697 struct obd_quotactl *oqctl)
2699 struct lov_obd *lov = &obd->u.lov;
2700 struct lov_tgt_desc *tgt;
2702 __u64 bhardlimit = 0;
2706 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2707 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2708 oqctl->qc_cmd != Q_GETOQUOTA &&
2709 oqctl->qc_cmd != Q_INITQUOTA &&
2710 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2711 oqctl->qc_cmd != Q_FINVALIDATE) {
2712 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2718 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2721 tgt = lov->lov_tgts[i];
2726 if (!tgt->ltd_active || tgt->ltd_reap) {
2727 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2728 lov->lov_tgts[i]->ltd_activate) {
2730 CERROR("ost %d is inactive\n", i);
2732 CDEBUG(D_HA, "ost %d is inactive\n", i);
2737 err = obd_quotactl(tgt->ltd_exp, oqctl);
2739 if (tgt->ltd_active && !rc)
2744 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2745 curspace += oqctl->qc_dqblk.dqb_curspace;
2746 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2751 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2752 oqctl->qc_dqblk.dqb_curspace = curspace;
2753 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2758 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2759 struct obd_quotactl *oqctl)
2761 struct lov_obd *lov = &obd->u.lov;
2767 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2768 if (!lov->lov_tgts[i])
2771 /* Skip quota check on the administratively disabled OSTs. */
2772 if (!lov->lov_tgts[i]->ltd_activate) {
2773 CWARN("lov idx %d was administratively disabled, "
2774 "skip quotacheck on it.\n", i);
2778 if (!lov->lov_tgts[i]->ltd_active) {
2779 CERROR("lov idx %d inactive\n", i);
2785 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2788 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2791 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2802 struct obd_ops lov_obd_ops = {
2803 .o_owner = THIS_MODULE,
2804 .o_setup = lov_setup,
2805 .o_precleanup = lov_precleanup,
2806 .o_cleanup = lov_cleanup,
2807 //.o_process_config = lov_process_config,
2808 .o_connect = lov_connect,
2809 .o_disconnect = lov_disconnect,
2810 .o_statfs = lov_statfs,
2811 .o_statfs_async = lov_statfs_async,
2812 .o_packmd = lov_packmd,
2813 .o_unpackmd = lov_unpackmd,
2814 .o_create = lov_create,
2815 .o_destroy = lov_destroy,
2816 .o_getattr = lov_getattr,
2817 .o_getattr_async = lov_getattr_async,
2818 .o_setattr = lov_setattr,
2819 .o_setattr_async = lov_setattr_async,
2821 .o_merge_lvb = lov_merge_lvb,
2822 .o_adjust_kms = lov_adjust_kms,
2823 .o_punch = lov_punch,
2825 .o_enqueue = lov_enqueue,
2826 .o_change_cbdata = lov_change_cbdata,
2827 .o_find_cbdata = lov_find_cbdata,
2828 .o_cancel = lov_cancel,
2829 .o_cancel_unused = lov_cancel_unused,
2830 .o_iocontrol = lov_iocontrol,
2831 .o_get_info = lov_get_info,
2832 .o_set_info_async = lov_set_info_async,
2833 .o_extent_calc = lov_extent_calc,
2834 .o_llog_init = lov_llog_init,
2835 .o_llog_finish = lov_llog_finish,
2836 .o_notify = lov_notify,
2837 .o_pool_new = lov_pool_new,
2838 .o_pool_rem = lov_pool_remove,
2839 .o_pool_add = lov_pool_add,
2840 .o_pool_del = lov_pool_del,
2841 .o_getref = lov_getref,
2842 .o_putref = lov_putref,
2843 .o_quotactl = lov_quotactl,
2844 .o_quotacheck = lov_quotacheck,
2847 cfs_mem_cache_t *lov_oinfo_slab;
2849 extern struct lu_kmem_descr lov_caches[];
2851 int __init lov_init(void)
2853 struct lprocfs_static_vars lvars = { 0 };
2857 /* print an address of _any_ initialized kernel symbol from this
2858 * module, to allow debugging with gdb that doesn't support data
2859 * symbols from modules.*/
2860 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2862 rc = lu_kmem_init(lov_caches);
2866 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2867 sizeof(struct lov_oinfo),
2868 0, CFS_SLAB_HWCACHE_ALIGN);
2869 if (lov_oinfo_slab == NULL) {
2870 lu_kmem_fini(lov_caches);
2873 lprocfs_lov_init_vars(&lvars);
2875 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2876 LUSTRE_LOV_NAME, &lov_device_type);
2879 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2881 lu_kmem_fini(lov_caches);
2888 static void /*__exit*/ lov_exit(void)
2892 class_unregister_type(LUSTRE_LOV_NAME);
2893 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2896 lu_kmem_fini(lov_caches);
2899 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2900 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2901 MODULE_LICENSE("GPL");
2903 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);