1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_obd.c
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
45 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_LOV
49 #include <libcfs/libcfs.h>
51 #include <liblustre.h>
54 #include <obd_support.h>
55 #include <lustre_lib.h>
56 #include <lustre_net.h>
57 #include <lustre/lustre_idl.h>
58 #include <lustre_dlm.h>
59 #include <lustre_mds.h>
60 #include <lustre_debug.h>
61 #include <obd_class.h>
64 #include <lprocfs_status.h>
65 #include <lustre_param.h>
66 #include <cl_object.h>
67 #include <lustre/ll_fiemap.h>
69 #include "lov_internal.h"
71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
72 Any function that expects lov_tgts to remain stationary must take a ref. */
73 static void lov_getref(struct obd_device *obd)
75 struct lov_obd *lov = &obd->u.lov;
77 /* nobody gets through here until lov_putref is done */
78 mutex_down(&lov->lov_lock);
79 atomic_inc(&lov->lov_refcount);
80 mutex_up(&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_down(&lov->lov_lock);
91 /* ok to dec to 0 more than once -- ltd_exp's will be null */
92 if (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 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_up(&lov->lov_lock);
113 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
114 list_del(&tgt->ltd_kill);
116 __lov_del_obd(obd, tgt);
119 mutex_up(&lov->lov_lock);
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
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;
140 cfs_proc_dir_entry_t *lov_proc_dir;
145 if (!lov->lov_tgts[index])
148 tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
149 tgt_obd = lov->lov_tgts[index]->ltd_obd;
151 if (!tgt_obd->obd_set_up) {
152 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
156 /* override the sp_me from lov */
157 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
159 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
160 data->ocd_index = index;
163 * Divine LOV knows that OBDs under it are OSCs.
165 imp = tgt_obd->u.cli.cl_import;
168 tgt_obd->obd_no_recov = 0;
169 /* FIXME this is probably supposed to be
170 ptlrpc_set_import_active. Horrible naming. */
171 ptlrpc_activate_import(imp);
174 rc = obd_register_observer(tgt_obd, obd);
176 CERROR("Target %s register_observer error %d\n",
177 obd_uuid2str(&tgt_uuid), rc);
182 if (imp->imp_invalid) {
183 CERROR("not connecting OSC %s; administratively "
184 "disabled\n", obd_uuid2str(&tgt_uuid));
188 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
189 &lov_osc_uuid, data, NULL);
190 if (rc || !lov->lov_tgts[index]->ltd_exp) {
191 CERROR("Target %s connect error %d\n",
192 obd_uuid2str(&tgt_uuid), rc);
196 lov->lov_tgts[index]->ltd_reap = 0;
198 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
199 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
202 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
204 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
205 cfs_proc_dir_entry_t *osc_symlink;
206 char name[MAX_STRING_SIZE];
208 LASSERT(osc_obd != NULL);
209 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
210 LASSERT(osc_obd->obd_type->typ_name != NULL);
211 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
212 osc_obd->obd_type->typ_name,
214 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
216 if (osc_symlink == NULL) {
217 CERROR("could not register LOV target "
218 "/proc/fs/lustre/%s/%s/target_obds/%s.",
219 obd->obd_type->typ_name, obd->obd_name,
221 lprocfs_remove(&lov_proc_dir);
226 rc = qos_add_tgt(obd, index);
228 CERROR("qos_add_tgt failed %d\n", rc);
233 static int lov_connect(const struct lu_env *env,
234 struct obd_export **exp, struct obd_device *obd,
235 struct obd_uuid *cluuid, struct obd_connect_data *data,
238 struct lov_obd *lov = &obd->u.lov;
239 struct lov_tgt_desc *tgt;
240 struct lustre_handle conn;
244 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
246 rc = class_connect(&conn, obd, cluuid);
250 *exp = class_conn2export(&conn);
252 /* Why should there ever be more than 1 connect? */
254 LASSERT(lov->lov_connects == 1);
256 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
258 lov->lov_ocd = *data;
261 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
262 tgt = lov->lov_tgts[i];
263 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
265 /* Flags will be lowest common denominator */
266 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
268 CERROR("%s: lov connect tgt %d failed: %d\n",
269 obd->obd_name, i, rc);
272 /* connect to administrative disabled ost */
273 if (!lov->lov_tgts[i]->ltd_exp)
276 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
277 OBD_NOTIFY_CONNECT, (void *)&i);
279 CERROR("%s error sending notify %d\n",
288 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
290 cfs_proc_dir_entry_t *lov_proc_dir;
291 struct lov_obd *lov = &obd->u.lov;
292 struct obd_device *osc_obd;
296 osc_obd = class_exp2obd(tgt->ltd_exp);
297 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
298 obd->obd_name, osc_obd->obd_name);
300 if (tgt->ltd_active) {
302 lov->desc.ld_active_tgt_count--;
303 tgt->ltd_exp->exp_obd->obd_inactive = 1;
306 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
308 cfs_proc_dir_entry_t *osc_symlink;
310 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
312 lprocfs_remove(&osc_symlink);
314 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
315 obd->obd_type->typ_name, obd->obd_name,
321 /* Pass it on to our clients.
322 * XXX This should be an argument to disconnect,
323 * XXX not a back-door flag on the OBD. Ah well.
325 osc_obd->obd_force = obd->obd_force;
326 osc_obd->obd_fail = obd->obd_fail;
327 osc_obd->obd_no_recov = obd->obd_no_recov;
330 obd_register_observer(osc_obd, NULL);
332 rc = obd_disconnect(tgt->ltd_exp);
334 CERROR("Target %s disconnect error %d\n",
335 tgt->ltd_uuid.uuid, rc);
339 qos_del_tgt(obd, tgt);
345 static int lov_disconnect(struct obd_export *exp)
347 struct obd_device *obd = class_exp2obd(exp);
348 struct lov_obd *lov = &obd->u.lov;
355 /* Only disconnect the underlying layers on the final disconnect. */
357 if (lov->lov_connects != 0) {
358 /* why should there be more than 1 connect? */
359 CERROR("disconnect #%d\n", lov->lov_connects);
363 /* Let's hold another reference so lov_del_obd doesn't spin through
367 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
368 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
369 /* Disconnection is the last we know about an obd */
370 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
376 rc = class_disconnect(exp); /* bz 9811 */
382 * -EINVAL : UUID can't be found in the LOV's target list
383 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
384 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
385 * any >= 0 : is log target index
387 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
390 struct lov_obd *lov = &obd->u.lov;
391 struct lov_tgt_desc *tgt;
395 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
396 lov, uuid->uuid, activate);
399 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
400 tgt = lov->lov_tgts[index];
401 if (!tgt || !tgt->ltd_exp)
404 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
405 index, obd_uuid2str(&tgt->ltd_uuid),
406 tgt->ltd_exp->exp_handle.h_cookie);
407 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
411 if (index == lov->desc.ld_tgt_count)
412 GOTO(out, index = -EINVAL);
414 if (lov->lov_tgts[index]->ltd_active == activate) {
415 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
416 activate ? "" : "in");
420 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
421 activate ? "" : "in");
423 lov->lov_tgts[index]->ltd_active = activate;
426 lov->desc.ld_active_tgt_count++;
427 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
429 lov->desc.ld_active_tgt_count--;
430 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
432 /* remove any old qos penalty */
433 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
440 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
441 enum obd_notify_event ev, void *data)
446 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
447 struct obd_uuid *uuid;
451 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
452 CERROR("unexpected notification of %s %s!\n",
453 watched->obd_type->typ_name,
457 uuid = &watched->u.cli.cl_target_uuid;
459 /* Set OSC as active before notifying the observer, so the
460 * observer can use the OSC normally.
462 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
464 CERROR("%sactivation of %s failed: %d\n",
465 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
466 obd_uuid2str(uuid), rc);
469 /* active event should be pass lov target index as data */
473 /* Pass the notification up the chain. */
475 rc = obd_notify_observer(obd, watched, ev, data);
477 /* NULL watched means all osc's in the lov (only for syncs) */
478 /* sync event should be send lov idx as data */
479 struct lov_obd *lov = &obd->u.lov;
480 struct obd_device *tgt_obd;
483 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
484 /* don't send sync event if target not
485 * connected/activated */
486 if (!lov->lov_tgts[i] ||
487 !lov->lov_tgts[i]->ltd_active)
490 if ((ev == OBD_NOTIFY_SYNC) ||
491 (ev == OBD_NOTIFY_SYNC_NONBLOCK))
494 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
496 rc = obd_notify_observer(obd, tgt_obd, ev, data);
498 CERROR("%s: notify %s of %s failed %d\n",
500 obd->obd_observer->obd_name,
501 tgt_obd->obd_name, rc);
511 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
512 __u32 index, int gen, int active)
514 struct lov_obd *lov = &obd->u.lov;
515 struct lov_tgt_desc *tgt;
516 struct obd_device *tgt_obd;
520 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
521 uuidp->uuid, index, gen, active);
524 CERROR("request to add OBD %s with invalid generation: %d\n",
529 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
534 mutex_down(&lov->lov_lock);
536 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
537 tgt = lov->lov_tgts[index];
538 CERROR("UUID %s already assigned at LOV target index %d\n",
539 obd_uuid2str(&tgt->ltd_uuid), index);
540 mutex_up(&lov->lov_lock);
544 if (index >= lov->lov_tgt_size) {
545 /* We need to reallocate the lov target array. */
546 struct lov_tgt_desc **newtgts, **old = NULL;
547 __u32 newsize, oldsize = 0;
549 newsize = max(lov->lov_tgt_size, (__u32)2);
550 while (newsize < index + 1)
551 newsize = newsize << 1;
552 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
553 if (newtgts == NULL) {
554 mutex_up(&lov->lov_lock);
558 if (lov->lov_tgt_size) {
559 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
562 oldsize = lov->lov_tgt_size;
565 lov->lov_tgts = newtgts;
566 lov->lov_tgt_size = newsize;
571 OBD_FREE(old, sizeof(*old) * oldsize);
573 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
574 lov->lov_tgts, lov->lov_tgt_size);
579 mutex_up(&lov->lov_lock);
583 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
585 mutex_up(&lov->lov_lock);
590 memset(tgt, 0, sizeof(*tgt));
591 tgt->ltd_uuid = *uuidp;
592 tgt->ltd_obd = tgt_obd;
593 /* XXX - add a sanity check on the generation number. */
595 tgt->ltd_index = index;
596 tgt->ltd_activate = active;
597 lov->lov_tgts[index] = tgt;
598 if (index >= lov->desc.ld_tgt_count)
599 lov->desc.ld_tgt_count = index + 1;
601 mutex_up(&lov->lov_lock);
603 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
604 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
606 if (lov->lov_connects == 0) {
607 /* lov_connect hasn't been called yet. We'll do the
608 lov_connect_obd on this target when that fn first runs,
609 because we don't know the connect flags yet. */
615 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
619 /* connect to administrative disabled ost */
623 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
624 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
629 CERROR("add failed (%d), deleting %s\n", rc,
630 obd_uuid2str(&tgt->ltd_uuid));
631 lov_del_target(obd, index, 0, 0);
637 /* Schedule a target for deletion */
638 int lov_del_target(struct obd_device *obd, __u32 index,
639 struct obd_uuid *uuidp, int gen)
641 struct lov_obd *lov = &obd->u.lov;
642 int count = lov->desc.ld_tgt_count;
646 if (index >= count) {
647 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
654 if (!lov->lov_tgts[index]) {
655 CERROR("LOV target at index %d is not setup.\n", index);
656 GOTO(out, rc = -EINVAL);
659 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
660 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
661 lov_uuid2str(lov, index), index,
662 obd_uuid2str(uuidp));
663 GOTO(out, rc = -EINVAL);
666 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
667 lov_uuid2str(lov, index), index,
668 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
669 lov->lov_tgts[index]->ltd_active);
671 lov->lov_tgts[index]->ltd_reap = 1;
672 lov->lov_death_row++;
673 /* we really delete it from obd_putref */
680 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
682 struct obd_device *osc_obd;
685 LASSERT(tgt->ltd_reap);
687 osc_obd = class_exp2obd(tgt->ltd_exp);
689 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
691 osc_obd ? osc_obd->obd_name : "<no obd>");
694 lov_disconnect_obd(obd, tgt);
698 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
699 do it ourselves. And we can't do it from lov_cleanup,
700 because we just lost our only reference to it. */
702 class_manual_cleanup(osc_obd);
705 void lov_fix_desc_stripe_size(__u64 *val)
707 if (*val < PTLRPC_MAX_BRW_SIZE) {
708 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
709 PTLRPC_MAX_BRW_SIZE);
710 *val = PTLRPC_MAX_BRW_SIZE;
711 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
712 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
713 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
715 *val, LOV_MIN_STRIPE_SIZE);
719 void lov_fix_desc_stripe_count(__u32 *val)
725 void lov_fix_desc_pattern(__u32 *val)
727 /* from lov_setstripe */
728 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
729 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
734 void lov_fix_desc_qos_maxage(__u32 *val)
738 *val = QOS_DEFAULT_MAXAGE;
741 void lov_fix_desc(struct lov_desc *desc)
743 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
744 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
745 lov_fix_desc_pattern(&desc->ld_pattern);
746 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
749 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
751 struct lprocfs_static_vars lvars = { 0 };
752 struct lov_desc *desc;
753 struct lov_obd *lov = &obd->u.lov;
757 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
758 CERROR("LOV setup requires a descriptor\n");
762 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
764 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
765 CERROR("descriptor size wrong: %d > %d\n",
766 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
770 if (desc->ld_magic != LOV_DESC_MAGIC) {
771 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
772 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
773 obd->obd_name, desc);
774 lustre_swab_lov_desc(desc);
776 CERROR("%s: Bad lov desc magic: %#x\n",
777 obd->obd_name, desc->ld_magic);
784 desc->ld_active_tgt_count = 0;
786 lov->lov_tgt_size = 0;
788 sema_init(&lov->lov_lock, 1);
789 atomic_set(&lov->lov_refcount, 0);
790 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
791 init_rwsem(&lov->lov_qos.lq_rw_sem);
792 lov->lov_sp_me = LUSTRE_SP_CLI;
793 lov->lov_qos.lq_dirty = 1;
794 lov->lov_qos.lq_rr.lqr_dirty = 1;
795 lov->lov_qos.lq_reset = 1;
796 /* Default priority is toward free space balance */
797 lov->lov_qos.lq_prio_free = 232;
798 /* Default threshold for rr (roughly 17%) */
799 lov->lov_qos.lq_threshold_rr = 43;
800 /* Init statfs fields */
801 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
802 if (NULL == lov->lov_qos.lq_statfs_data)
804 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
806 lov->lov_pools_hash_body = lustre_hash_init("POOLS",
809 &pool_hash_operations, 0);
810 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
811 lov->lov_pool_count = 0;
812 rc = lov_ost_pool_init(&lov->lov_packed, 0);
815 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
817 lov_ost_pool_free(&lov->lov_packed);
821 lprocfs_lov_init_vars(&lvars);
822 lprocfs_obd_setup(obd, lvars.obd_vars);
827 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
828 0444, &lov_proc_target_fops, obd);
830 CWARN("Error adding the target_obd file\n");
833 lov->lov_pool_proc_entry = lprocfs_register("pools",
840 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
843 struct lov_obd *lov = &obd->u.lov;
848 case OBD_CLEANUP_EARLY: {
850 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
851 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
853 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
858 case OBD_CLEANUP_EXPORTS:
859 rc = obd_llog_finish(obd, 0);
861 CERROR("failed to cleanup llogging subsystems\n");
867 static int lov_cleanup(struct obd_device *obd)
869 struct lov_obd *lov = &obd->u.lov;
870 struct list_head *pos, *tmp;
871 struct pool_desc *pool;
873 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
874 pool = list_entry(pos, struct pool_desc, pool_list);
875 /* free pool structs */
876 CDEBUG(D_INFO, "delete pool %p\n", pool);
877 lov_pool_del(obd, pool->pool_name);
879 lustre_hash_exit(lov->lov_pools_hash_body);
880 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
881 lov_ost_pool_free(&lov->lov_packed);
886 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
887 if (!lov->lov_tgts[i])
890 /* Inactive targets may never have connected */
891 if (lov->lov_tgts[i]->ltd_active ||
892 atomic_read(&lov->lov_refcount))
893 /* We should never get here - these
894 should have been removed in the
896 CERROR("lov tgt %d not cleaned!"
897 " deathrow=%d, lovrc=%d\n",
898 i, lov->lov_death_row,
899 atomic_read(&lov->lov_refcount));
900 lov_del_target(obd, i, 0, 0);
903 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
905 lov->lov_tgt_size = 0;
908 /* clear pools parent proc entry only after all pools is killed */
909 lprocfs_obd_cleanup(obd);
911 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
915 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
916 __u32 *indexp, int *genp)
918 struct obd_uuid obd_uuid;
923 switch(cmd = lcfg->lcfg_command) {
924 case LCFG_LOV_ADD_OBD:
925 case LCFG_LOV_ADD_INA:
926 case LCFG_LOV_DEL_OBD: {
929 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
930 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
931 GOTO(out, rc = -EINVAL);
933 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
935 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
936 GOTO(out, rc = -EINVAL);
937 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
938 GOTO(out, rc = -EINVAL);
941 if (cmd == LCFG_LOV_ADD_OBD)
942 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
943 else if (cmd == LCFG_LOV_ADD_INA)
944 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
946 rc = lov_del_target(obd, index, &obd_uuid, gen);
950 struct lprocfs_static_vars lvars = { 0 };
951 struct lov_desc *desc = &(obd->u.lov.desc);
954 GOTO(out, rc = -EINVAL);
956 lprocfs_lov_init_vars(&lvars);
958 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
971 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
972 GOTO(out, rc = -EINVAL);
981 #define log2(n) ffz(~(n))
984 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
985 struct lov_stripe_md **ea,
986 struct obd_trans_info *oti)
990 struct obd_uuid *ost_uuid = NULL;
994 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
995 src_oa->o_flags == OBD_FL_DELORPHAN);
997 lov = &export->exp_obd->u.lov;
1003 if (oti->oti_ost_uuid) {
1004 ost_uuid = oti->oti_ost_uuid;
1005 CDEBUG(D_HA, "clearing orphans only for %s\n",
1009 obd_getref(export->exp_obd);
1010 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1011 struct lov_stripe_md obj_md;
1012 struct lov_stripe_md *obj_mdp = &obj_md;
1013 struct lov_tgt_desc *tgt;
1016 tgt = lov->lov_tgts[i];
1020 /* if called for a specific target, we don't
1021 care if it is not active. */
1022 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1023 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1027 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1030 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1031 obd_uuid2str(ost_uuid));
1033 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1035 LASSERT(lov->lov_tgts[i]->ltd_exp);
1036 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1037 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1038 tmp_oa, &obj_mdp, oti);
1040 /* This export will be disabled until it is recovered,
1041 and then orphan recovery will be completed. */
1042 CERROR("error in orphan recovery on OST idx %d/%d: "
1043 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1050 obd_putref(export->exp_obd);
1056 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1057 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1059 struct lov_stripe_md *obj_mdp, *lsm;
1060 struct lov_obd *lov = &exp->exp_obd->u.lov;
1065 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1066 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1068 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1069 if (obj_mdp == NULL)
1072 ost_idx = src_oa->o_nlink;
1075 GOTO(out, rc = -EINVAL);
1076 if (ost_idx >= lov->desc.ld_tgt_count ||
1077 !lov->lov_tgts[ost_idx])
1078 GOTO(out, rc = -EINVAL);
1080 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1081 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1082 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1083 GOTO(out, rc = -EINVAL);
1087 if (i == lsm->lsm_stripe_count)
1088 GOTO(out, rc = -EINVAL);
1090 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1092 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1096 /* the LOV expects oa->o_id to be set to the LOV object id */
1097 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1098 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1100 struct lov_obd *lov;
1101 struct obd_info oinfo;
1102 struct lov_request_set *set = NULL;
1103 struct lov_request *req;
1107 LASSERT(ea != NULL);
1111 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1112 src_oa->o_flags == OBD_FL_DELORPHAN) {
1113 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1117 lov = &exp->exp_obd->u.lov;
1118 if (!lov->desc.ld_active_tgt_count)
1121 obd_getref(exp->exp_obd);
1122 /* Recreate a specific object id at the given OST index */
1123 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1124 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1125 rc = lov_recreate(exp, src_oa, ea, oti);
1129 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1130 * later in alloc_qos(), we will wait for those rpcs to complete if
1131 * the osfs age is older than 2 * qos_maxage */
1132 qos_statfs_update(exp->exp_obd,
1133 cfs_time_shift_64(-lov->desc.ld_qos_maxage) + HZ, 0);
1135 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1139 list_for_each_entry(req, &set->set_list, rq_link) {
1140 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1141 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1142 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1143 lov_update_create_set(set, req, rc);
1145 rc = lov_fini_create_set(set, ea);
1147 obd_putref(exp->exp_obd);
1151 #define ASSERT_LSM_MAGIC(lsmp) \
1153 LASSERT((lsmp) != NULL); \
1154 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1155 (lsmp)->lsm_magic == LOV_MAGIC_V3 || \
1156 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1157 (lsmp), (lsmp)->lsm_magic); \
1160 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1161 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1162 struct obd_export *md_exp, void *capa)
1164 struct lov_request_set *set;
1165 struct obd_info oinfo;
1166 struct lov_request *req;
1167 struct list_head *pos;
1168 struct lov_obd *lov;
1169 int rc = 0, err = 0;
1172 ASSERT_LSM_MAGIC(lsm);
1174 if (!exp || !exp->exp_obd)
1177 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1179 LASSERT(oti->oti_logcookies);
1182 lov = &exp->exp_obd->u.lov;
1183 obd_getref(exp->exp_obd);
1184 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1188 list_for_each (pos, &set->set_list) {
1189 req = list_entry(pos, struct lov_request, rq_link);
1191 if (oa->o_valid & OBD_MD_FLCOOKIE)
1192 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1194 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1195 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1196 err = lov_update_common_set(set, req, err);
1198 CERROR("error: destroying objid "LPX64" subobj "
1199 LPX64" on OST idx %d: rc = %d\n",
1200 oa->o_id, req->rq_oi.oi_oa->o_id,
1208 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1209 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1211 err = lov_fini_destroy_set(set);
1213 obd_putref(exp->exp_obd);
1214 RETURN(rc ? rc : err);
1217 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1219 struct lov_request_set *set;
1220 struct lov_request *req;
1221 struct list_head *pos;
1222 struct lov_obd *lov;
1223 int err = 0, rc = 0;
1227 ASSERT_LSM_MAGIC(oinfo->oi_md);
1229 if (!exp || !exp->exp_obd)
1232 lov = &exp->exp_obd->u.lov;
1234 rc = lov_prep_getattr_set(exp, oinfo, &set);
1238 list_for_each (pos, &set->set_list) {
1239 req = list_entry(pos, struct lov_request, rq_link);
1241 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1242 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1243 req->rq_oi.oi_oa->o_id, req->rq_idx);
1245 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1247 err = lov_update_common_set(set, req, rc);
1249 CERROR("error: getattr objid "LPX64" subobj "
1250 LPX64" on OST idx %d: rc = %d\n",
1251 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1257 rc = lov_fini_getattr_set(set);
1263 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1266 struct lov_request_set *lovset = (struct lov_request_set *)data;
1270 /* don't do attribute merge if this aysnc op failed */
1272 lovset->set_completes = 0;
1273 err = lov_fini_getattr_set(lovset);
1274 RETURN(rc ? rc : err);
1277 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1278 struct ptlrpc_request_set *rqset)
1280 struct lov_request_set *lovset;
1281 struct lov_obd *lov;
1282 struct list_head *pos;
1283 struct lov_request *req;
1288 ASSERT_LSM_MAGIC(oinfo->oi_md);
1290 if (!exp || !exp->exp_obd)
1293 lov = &exp->exp_obd->u.lov;
1295 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1299 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1300 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1301 oinfo->oi_md->lsm_stripe_size);
1303 list_for_each (pos, &lovset->set_list) {
1304 req = list_entry(pos, struct lov_request, rq_link);
1306 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1307 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1308 req->rq_oi.oi_oa->o_id, req->rq_idx);
1309 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1310 &req->rq_oi, rqset);
1312 CERROR("error: getattr objid "LPX64" subobj "
1313 LPX64" on OST idx %d: rc = %d\n",
1314 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1320 if (!list_empty(&rqset->set_requests)) {
1322 LASSERT (rqset->set_interpret == NULL);
1323 rqset->set_interpret = lov_getattr_interpret;
1324 rqset->set_arg = (void *)lovset;
1329 lovset->set_completes = 0;
1330 err = lov_fini_getattr_set(lovset);
1331 RETURN(rc ? rc : err);
1334 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1335 struct obd_trans_info *oti)
1337 struct lov_request_set *set;
1338 struct lov_obd *lov;
1339 struct list_head *pos;
1340 struct lov_request *req;
1341 int err = 0, rc = 0;
1345 ASSERT_LSM_MAGIC(oinfo->oi_md);
1347 if (!exp || !exp->exp_obd)
1350 /* for now, we only expect the following updates here */
1351 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1352 OBD_MD_FLMODE | OBD_MD_FLATIME |
1353 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1354 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1355 OBD_MD_FLGROUP | OBD_MD_FLUID |
1356 OBD_MD_FLGID | OBD_MD_FLFID |
1358 lov = &exp->exp_obd->u.lov;
1359 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1363 list_for_each (pos, &set->set_list) {
1364 req = list_entry(pos, struct lov_request, rq_link);
1366 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1368 err = lov_update_setattr_set(set, req, rc);
1370 CERROR("error: setattr objid "LPX64" subobj "
1371 LPX64" on OST idx %d: rc = %d\n",
1372 set->set_oi->oi_oa->o_id,
1373 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1378 err = lov_fini_setattr_set(set);
1384 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1387 struct lov_request_set *lovset = (struct lov_request_set *)data;
1392 lovset->set_completes = 0;
1393 err = lov_fini_setattr_set(lovset);
1394 RETURN(rc ? rc : err);
1397 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1398 needed. Otherwise, a client is waiting for responses. */
1399 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1400 struct obd_trans_info *oti,
1401 struct ptlrpc_request_set *rqset)
1403 struct lov_request_set *set;
1404 struct lov_request *req;
1405 struct list_head *pos;
1406 struct lov_obd *lov;
1411 ASSERT_LSM_MAGIC(oinfo->oi_md);
1412 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1414 LASSERT(oti->oti_logcookies);
1417 if (!exp || !exp->exp_obd)
1420 lov = &exp->exp_obd->u.lov;
1421 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1425 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1426 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1427 oinfo->oi_md->lsm_stripe_size);
1429 list_for_each (pos, &set->set_list) {
1430 req = list_entry(pos, struct lov_request, rq_link);
1432 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1433 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1435 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1436 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1437 req->rq_oi.oi_oa->o_id, req->rq_idx);
1439 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1440 &req->rq_oi, oti, rqset);
1442 CERROR("error: setattr objid "LPX64" subobj "
1443 LPX64" on OST idx %d: rc = %d\n",
1444 set->set_oi->oi_oa->o_id,
1445 req->rq_oi.oi_oa->o_id,
1451 /* If we are not waiting for responses on async requests, return. */
1452 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1455 set->set_completes = 0;
1456 err = lov_fini_setattr_set(set);
1457 RETURN(rc ? rc : err);
1460 LASSERT(rqset->set_interpret == NULL);
1461 rqset->set_interpret = lov_setattr_interpret;
1462 rqset->set_arg = (void *)set;
1467 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1470 struct lov_request_set *lovset = (struct lov_request_set *)data;
1475 lovset->set_completes = 0;
1476 err = lov_fini_punch_set(lovset);
1477 RETURN(rc ? rc : err);
1480 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1481 * we can send this 'punch' to just the authoritative node and the nodes
1482 * that the punch will affect. */
1483 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1484 struct obd_trans_info *oti,
1485 struct ptlrpc_request_set *rqset)
1487 struct lov_request_set *set;
1488 struct lov_obd *lov;
1489 struct list_head *pos;
1490 struct lov_request *req;
1495 ASSERT_LSM_MAGIC(oinfo->oi_md);
1497 if (!exp || !exp->exp_obd)
1500 lov = &exp->exp_obd->u.lov;
1501 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1505 list_for_each (pos, &set->set_list) {
1506 req = list_entry(pos, struct lov_request, rq_link);
1508 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1509 &req->rq_oi, NULL, rqset);
1511 CERROR("error: punch objid "LPX64" subobj "LPX64
1512 " on OST idx %d: rc = %d\n",
1513 set->set_oi->oi_oa->o_id,
1514 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1519 if (rc || list_empty(&rqset->set_requests)) {
1521 err = lov_fini_punch_set(set);
1522 RETURN(rc ? rc : err);
1525 LASSERT(rqset->set_interpret == NULL);
1526 rqset->set_interpret = lov_punch_interpret;
1527 rqset->set_arg = (void *)set;
1532 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1533 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1536 struct lov_request_set *set;
1537 struct obd_info oinfo;
1538 struct lov_obd *lov;
1539 struct list_head *pos;
1540 struct lov_request *req;
1541 int err = 0, rc = 0;
1544 ASSERT_LSM_MAGIC(lsm);
1549 lov = &exp->exp_obd->u.lov;
1550 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1554 list_for_each (pos, &set->set_list) {
1555 req = list_entry(pos, struct lov_request, rq_link);
1557 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1558 req->rq_oi.oi_oa, NULL,
1559 req->rq_oi.oi_policy.l_extent.start,
1560 req->rq_oi.oi_policy.l_extent.end, capa);
1561 err = lov_update_common_set(set, req, rc);
1563 CERROR("error: fsync objid "LPX64" subobj "LPX64
1564 " on OST idx %d: rc = %d\n",
1565 set->set_oi->oi_oa->o_id,
1566 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1571 err = lov_fini_sync_set(set);
1577 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1578 obd_count oa_bufs, struct brw_page *pga)
1580 struct obd_info oinfo = { { { 0 } } };
1583 oinfo.oi_oa = lov_oinfo->oi_oa;
1585 /* The caller just wants to know if there's a chance that this
1586 * I/O can succeed */
1587 for (i = 0; i < oa_bufs; i++) {
1588 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1589 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1592 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1593 pga[i].off + pga[i].count - 1,
1597 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1598 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1602 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1610 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1611 obd_count oa_bufs, struct brw_page *pga,
1612 struct obd_trans_info *oti)
1614 struct lov_request_set *set;
1615 struct lov_request *req;
1616 struct list_head *pos;
1617 struct lov_obd *lov = &exp->exp_obd->u.lov;
1621 ASSERT_LSM_MAGIC(oinfo->oi_md);
1623 if (cmd == OBD_BRW_CHECK) {
1624 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1628 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1632 list_for_each (pos, &set->set_list) {
1633 struct obd_export *sub_exp;
1634 struct brw_page *sub_pga;
1635 req = list_entry(pos, struct lov_request, rq_link);
1637 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1638 sub_pga = set->set_pga + req->rq_pgaidx;
1639 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1643 lov_update_common_set(set, req, rc);
1646 err = lov_fini_brw_set(set);
1652 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1655 struct lov_request_set *lovset = (struct lov_request_set *)data;
1657 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1661 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1662 struct ldlm_enqueue_info *einfo,
1663 struct ptlrpc_request_set *rqset)
1665 ldlm_mode_t mode = einfo->ei_mode;
1666 struct lov_request_set *set;
1667 struct lov_request *req;
1668 struct list_head *pos;
1669 struct lov_obd *lov;
1674 ASSERT_LSM_MAGIC(oinfo->oi_md);
1675 LASSERT(mode == (mode & -mode));
1677 /* we should never be asked to replay a lock this way. */
1678 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1680 if (!exp || !exp->exp_obd)
1683 lov = &exp->exp_obd->u.lov;
1684 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1688 list_for_each (pos, &set->set_list) {
1689 req = list_entry(pos, struct lov_request, rq_link);
1691 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1692 &req->rq_oi, einfo, rqset);
1697 if (rqset && !list_empty(&rqset->set_requests)) {
1699 LASSERT(rqset->set_interpret == NULL);
1700 rqset->set_interpret = lov_enqueue_interpret;
1701 rqset->set_arg = (void *)set;
1705 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1709 static int lov_change_cbdata(struct obd_export *exp,
1710 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1713 struct lov_obd *lov;
1717 ASSERT_LSM_MAGIC(lsm);
1719 if (!exp || !exp->exp_obd)
1722 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1724 lov = &exp->exp_obd->u.lov;
1725 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1726 struct lov_stripe_md submd;
1727 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1729 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1730 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1734 submd.lsm_object_id = loi->loi_id;
1735 submd.lsm_object_gr = lsm->lsm_object_gr;
1736 submd.lsm_stripe_count = 0;
1737 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1743 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1744 __u32 mode, struct lustre_handle *lockh)
1746 struct lov_request_set *set;
1747 struct obd_info oinfo;
1748 struct lov_request *req;
1749 struct list_head *pos;
1750 struct lov_obd *lov;
1751 struct lustre_handle *lov_lockhp;
1752 int err = 0, rc = 0;
1755 ASSERT_LSM_MAGIC(lsm);
1757 if (!exp || !exp->exp_obd)
1760 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1762 lov = &exp->exp_obd->u.lov;
1763 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1767 list_for_each (pos, &set->set_list) {
1768 req = list_entry(pos, struct lov_request, rq_link);
1769 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1771 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1772 req->rq_oi.oi_md, mode, lov_lockhp);
1773 rc = lov_update_common_set(set, req, rc);
1775 CERROR("error: cancel objid "LPX64" subobj "
1776 LPX64" on OST idx %d: rc = %d\n",
1778 req->rq_oi.oi_md->lsm_object_id,
1784 lov_fini_cancel_set(set);
1788 static int lov_cancel_unused(struct obd_export *exp,
1789 struct lov_stripe_md *lsm,
1790 int flags, void *opaque)
1792 struct lov_obd *lov;
1796 if (!exp || !exp->exp_obd)
1799 lov = &exp->exp_obd->u.lov;
1801 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1803 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1806 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1814 ASSERT_LSM_MAGIC(lsm);
1816 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
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];
1822 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1823 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1827 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1828 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1830 submd.lsm_object_id = loi->loi_id;
1831 submd.lsm_object_gr = lsm->lsm_object_gr;
1832 submd.lsm_stripe_count = 0;
1833 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1834 &submd, flags, opaque);
1835 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1836 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1837 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1838 loi->loi_id, loi->loi_ost_idx, err);
1846 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1848 struct lov_request_set *lovset = (struct lov_request_set *)data;
1853 lovset->set_completes = 0;
1855 err = lov_fini_statfs_set(lovset);
1856 RETURN(rc ? rc : err);
1859 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1860 __u64 max_age, struct ptlrpc_request_set *rqset)
1862 struct lov_request_set *set;
1863 struct lov_request *req;
1864 struct list_head *pos;
1865 struct lov_obd *lov;
1869 LASSERT(oinfo != NULL);
1870 LASSERT(oinfo->oi_osfs != NULL);
1873 rc = lov_prep_statfs_set(obd, oinfo, &set);
1877 list_for_each (pos, &set->set_list) {
1878 struct obd_device *osc_obd;
1880 req = list_entry(pos, struct lov_request, rq_link);
1882 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1883 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1888 if (rc || list_empty(&rqset->set_requests)) {
1891 set->set_completes = 0;
1892 err = lov_fini_statfs_set(set);
1893 RETURN(rc ? rc : err);
1896 LASSERT(rqset->set_interpret == NULL);
1897 rqset->set_interpret = lov_statfs_interpret;
1898 rqset->set_arg = (void *)set;
1902 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1903 __u64 max_age, __u32 flags)
1905 struct ptlrpc_request_set *set = NULL;
1906 struct obd_info oinfo = { { { 0 } } };
1911 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1912 * statfs requests */
1913 set = ptlrpc_prep_set();
1917 oinfo.oi_osfs = osfs;
1918 oinfo.oi_flags = flags;
1919 rc = lov_statfs_async(obd, &oinfo, max_age, set);
1921 rc = ptlrpc_set_wait(set);
1922 ptlrpc_set_destroy(set);
1927 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1928 void *karg, void *uarg)
1930 struct obd_device *obddev = class_exp2obd(exp);
1931 struct lov_obd *lov = &obddev->u.lov;
1932 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1933 struct obd_uuid *uuidp;
1937 case IOC_OBD_STATFS: {
1938 struct obd_ioctl_data *data = karg;
1939 struct obd_device *osc_obd;
1940 struct obd_statfs stat_buf = {0};
1943 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1944 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1946 if ((index >= count))
1949 if (!lov->lov_tgts[index])
1950 /* Try again with the next index */
1952 if (!lov->lov_tgts[index]->ltd_active)
1955 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1959 /* got statfs data */
1960 rc = obd_statfs(osc_obd, &stat_buf,
1961 cfs_time_current_64() - HZ, 0);
1964 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1967 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1972 case OBD_IOC_LOV_GET_CONFIG: {
1973 struct obd_ioctl_data *data;
1974 struct lov_desc *desc;
1979 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1982 data = (struct obd_ioctl_data *)buf;
1984 if (sizeof(*desc) > data->ioc_inllen1) {
1985 obd_ioctl_freedata(buf, len);
1989 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1990 obd_ioctl_freedata(buf, len);
1994 if (sizeof(__u32) * count > data->ioc_inllen3) {
1995 obd_ioctl_freedata(buf, len);
1999 desc = (struct lov_desc *)data->ioc_inlbuf1;
2000 memcpy(desc, &(lov->desc), sizeof(*desc));
2002 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2003 genp = (__u32 *)data->ioc_inlbuf3;
2004 /* the uuid will be empty for deleted OSTs */
2005 for (i = 0; i < count; i++, uuidp++, genp++) {
2006 if (!lov->lov_tgts[i])
2008 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2009 *genp = lov->lov_tgts[i]->ltd_gen;
2012 if (copy_to_user((void *)uarg, buf, len))
2014 obd_ioctl_freedata(buf, len);
2017 case LL_IOC_LOV_SETSTRIPE:
2018 rc = lov_setstripe(exp, len, karg, uarg);
2020 case LL_IOC_LOV_GETSTRIPE:
2021 rc = lov_getstripe(exp, karg, uarg);
2023 case LL_IOC_LOV_SETEA:
2024 rc = lov_setea(exp, karg, uarg);
2026 case OBD_IOC_QUOTACTL: {
2027 struct if_quotactl *qctl = karg;
2028 struct lov_tgt_desc *tgt = NULL;
2029 struct obd_quotactl *oqctl;
2031 if (qctl->qc_valid == QC_OSTIDX) {
2032 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2035 tgt = lov->lov_tgts[qctl->qc_idx];
2036 if (!tgt || !tgt->ltd_exp)
2038 } else if (qctl->qc_valid == QC_UUID) {
2039 for (i = 0; i < count; i++) {
2040 tgt = lov->lov_tgts[i];
2042 !obd_uuid_equals(&tgt->ltd_uuid,
2046 if (tgt->ltd_exp == NULL)
2058 LASSERT(tgt && tgt->ltd_exp);
2059 OBD_ALLOC_PTR(oqctl);
2063 QCTL_COPY(oqctl, qctl);
2064 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2066 QCTL_COPY(qctl, oqctl);
2067 qctl->qc_valid = QC_OSTIDX;
2068 qctl->obd_uuid = tgt->ltd_uuid;
2070 OBD_FREE_PTR(oqctl);
2079 for (i = 0; i < count; i++) {
2082 /* OST was disconnected */
2083 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2086 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2088 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2091 if (lov->lov_tgts[i]->ltd_active) {
2092 CDEBUG(err == -ENOTTY ?
2093 D_IOCTL : D_WARNING,
2094 "iocontrol OSC %s on OST "
2095 "idx %d cmd %x: err = %d\n",
2096 lov_uuid2str(lov, i),
2113 #define FIEMAP_BUFFER_SIZE 4096
2116 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2117 * call. The local end offset and the device are sent in the first
2118 * fm_extent. This function calculates the stripe number from the index.
2119 * This function returns a stripe_no on which mapping is to be restarted.
2121 * This function returns fm_end_offset which is the in-OST offset at which
2122 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2123 * will re-calculate proper offset in next stripe.
2124 * Note that the first extent is passed to lov_get_info via the value field.
2126 * \param fiemap fiemap request header
2127 * \param lsm striping information for the file
2128 * \param fm_start logical start of mapping
2129 * \param fm_end logical end of mapping
2130 * \param start_stripe starting stripe will be returned in this
2132 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2133 struct lov_stripe_md *lsm, obd_size fm_start,
2134 obd_size fm_end, int *start_stripe)
2136 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2137 obd_off lun_start, lun_end;
2138 obd_size fm_end_offset;
2139 int stripe_no = -1, i;
2141 if (fiemap->fm_extent_count == 0 ||
2142 fiemap->fm_extents[0].fe_logical == 0)
2145 /* Find out stripe_no from ost_index saved in the fe_device */
2146 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2147 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2148 fiemap->fm_extents[0].fe_device) {
2154 /* If we have finished mapping on previous device, shift logical
2155 * offset to start of next device */
2156 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2157 &lun_start, &lun_end)) != 0 &&
2158 local_end < lun_end) {
2159 fm_end_offset = local_end;
2160 *start_stripe = stripe_no;
2162 /* This is a special value to indicate that caller should
2163 * calculate offset in next stripe. */
2165 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2168 return fm_end_offset;
2172 * We calculate on which OST the mapping will end. If the length of mapping
2173 * is greater than (stripe_size * stripe_count) then the last_stripe will
2174 * will be one just before start_stripe. Else we check if the mapping
2175 * intersects each OST and find last_stripe.
2176 * This function returns the last_stripe and also sets the stripe_count
2177 * over which the mapping is spread
2179 * \param lsm striping information for the file
2180 * \param fm_start logical start of mapping
2181 * \param fm_end logical end of mapping
2182 * \param start_stripe starting stripe of the mapping
2183 * \param stripe_count the number of stripes across which to map is returned
2185 * \retval last_stripe return the last stripe of the mapping
2187 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2188 obd_size fm_end, int start_stripe,
2192 obd_off obd_start, obd_end;
2195 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2196 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2198 *stripe_count = lsm->lsm_stripe_count;
2200 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2201 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2202 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2203 &obd_start, &obd_end)) == 0)
2207 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2214 * Set fe_device and copy extents from local buffer into main return buffer.
2216 * \param fiemap fiemap request header
2217 * \param lcl_fm_ext array of local fiemap extents to be copied
2218 * \param ost_index OST index to be written into the fm_device field for each
2220 * \param ext_count number of extents to be copied
2221 * \param current_extent where to start copying in main extent array
2223 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2224 struct ll_fiemap_extent *lcl_fm_ext,
2225 int ost_index, unsigned int ext_count,
2231 for (ext = 0; ext < ext_count; ext++) {
2232 lcl_fm_ext[ext].fe_device = ost_index;
2233 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2236 /* Copy fm_extent's from fm_local to return buffer */
2237 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2238 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2242 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2243 * This also handles the restarting of FIEMAP calls in case mapping overflows
2244 * the available number of extents in single call.
2246 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2247 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2249 struct ll_fiemap_info_key *fm_key = key;
2250 struct ll_user_fiemap *fiemap = val;
2251 struct ll_user_fiemap *fm_local = NULL;
2252 struct ll_fiemap_extent *lcl_fm_ext;
2254 unsigned int get_num_extents = 0;
2255 int ost_index = 0, actual_start_stripe, start_stripe;
2256 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2258 int current_extent = 0, rc = 0, i;
2259 int ost_eof = 0; /* EOF for object */
2260 int ost_done = 0; /* done with required mapping for this OST? */
2262 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2263 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2268 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2269 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2271 OBD_ALLOC(fm_local, buffer_size);
2272 if (fm_local == NULL)
2273 GOTO(out, rc = -ENOMEM);
2274 lcl_fm_ext = &fm_local->fm_extents[0];
2276 count_local = fiemap_size_to_count(buffer_size);
2278 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2279 fm_start = fiemap->fm_start;
2280 fm_length = fiemap->fm_length;
2281 /* Calculate start stripe, last stripe and length of mapping */
2282 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2283 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2284 fm_start + fm_length - 1);
2285 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2286 if (fm_end > fm_key->oa.o_size)
2287 fm_end = fm_key->oa.o_size;
2289 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2290 actual_start_stripe, &stripe_count);
2292 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2295 if (fiemap->fm_extent_count == 0) {
2296 get_num_extents = 1;
2300 /* Check each stripe */
2301 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2302 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2303 obd_size req_fm_len; /* Stores length of required mapping */
2304 obd_size len_mapped_single_call;
2305 obd_off lun_start, lun_end, obd_object_end;
2306 unsigned int ext_count;
2308 cur_stripe_wrap = cur_stripe;
2310 /* Find out range of mapping on this stripe */
2311 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2312 &lun_start, &obd_object_end)) == 0)
2315 /* If this is a continuation FIEMAP call and we are on
2316 * starting stripe then lun_start needs to be set to
2318 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2319 lun_start = fm_end_offset;
2321 if (fm_length != ~0ULL) {
2322 /* Handle fm_start + fm_length overflow */
2323 if (fm_start + fm_length < fm_start)
2324 fm_length = ~0ULL - fm_start;
2325 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2331 if (lun_start == lun_end)
2334 req_fm_len = obd_object_end - lun_start;
2335 fm_local->fm_length = 0;
2336 len_mapped_single_call = 0;
2338 /* If the output buffer is very large and the objects have many
2339 * extents we may need to loop on a single OST repeatedly */
2343 if (get_num_extents == 0) {
2344 /* Don't get too many extents. */
2345 if (current_extent + count_local >
2346 fiemap->fm_extent_count)
2347 count_local = fiemap->fm_extent_count -
2351 lun_start += len_mapped_single_call;
2352 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2353 req_fm_len = fm_local->fm_length;
2354 fm_local->fm_extent_count = count_local;
2355 fm_local->fm_mapped_extents = 0;
2356 fm_local->fm_flags = fiemap->fm_flags;
2358 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2359 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2361 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2362 GOTO(out, rc = -EINVAL);
2364 /* If OST is inactive, return extent with UNKNOWN flag */
2365 if (!lov->lov_tgts[ost_index]->ltd_active) {
2366 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2367 fm_local->fm_mapped_extents = 1;
2369 lcl_fm_ext[0].fe_logical = lun_start;
2370 lcl_fm_ext[0].fe_length = obd_object_end -
2372 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2377 fm_local->fm_start = lun_start;
2378 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2379 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2380 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2381 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2382 keylen, key, vallen, fm_local, lsm);
2387 ext_count = fm_local->fm_mapped_extents;
2388 if (ext_count == 0) {
2390 /* If last stripe has hole at the end,
2391 * then we need to return */
2392 if (cur_stripe_wrap == last_stripe) {
2393 fiemap->fm_mapped_extents = 0;
2399 /* If we just need num of extents then go to next device */
2400 if (get_num_extents) {
2401 current_extent += ext_count;
2405 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2406 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2408 /* Have we finished mapping on this device? */
2409 if (req_fm_len <= len_mapped_single_call)
2412 /* Clear the EXTENT_LAST flag which can be present on
2414 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2415 lcl_fm_ext[ext_count - 1].fe_flags &=
2416 ~FIEMAP_EXTENT_LAST;
2418 curr_loc = lov_stripe_size(lsm,
2419 lcl_fm_ext[ext_count - 1].fe_logical+
2420 lcl_fm_ext[ext_count - 1].fe_length,
2422 if (curr_loc >= fm_key->oa.o_size)
2425 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2426 ost_index, ext_count,
2429 current_extent += ext_count;
2431 /* Ran out of available extents? */
2432 if (current_extent >= fiemap->fm_extent_count)
2434 } while (ost_done == 0 && ost_eof == 0);
2436 if (cur_stripe_wrap == last_stripe)
2441 /* Indicate that we are returning device offsets unless file just has
2443 if (lsm->lsm_stripe_count > 1)
2444 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2446 if (get_num_extents)
2447 goto skip_last_device_calc;
2449 /* Check if we have reached the last stripe and whether mapping for that
2450 * stripe is done. */
2451 if (cur_stripe_wrap == last_stripe) {
2452 if (ost_done || ost_eof)
2453 fiemap->fm_extents[current_extent - 1].fe_flags |=
2457 skip_last_device_calc:
2458 fiemap->fm_mapped_extents = current_extent;
2461 OBD_FREE(fm_local, buffer_size);
2465 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2466 void *key, __u32 *vallen, void *val,
2467 struct lov_stripe_md *lsm)
2469 struct obd_device *obddev = class_exp2obd(exp);
2470 struct lov_obd *lov = &obddev->u.lov;
2474 if (!vallen || !val)
2479 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2482 struct ldlm_lock *lock;
2484 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2485 struct lov_oinfo *loi;
2486 __u32 *stripe = val;
2488 if (*vallen < sizeof(*stripe))
2489 GOTO(out, rc = -EFAULT);
2490 *vallen = sizeof(*stripe);
2492 /* XXX This is another one of those bits that will need to
2493 * change if we ever actually support nested LOVs. It uses
2494 * the lock's export to find out which stripe it is. */
2495 /* XXX - it's assumed all the locks for deleted OSTs have
2496 * been cancelled. Also, the export for deleted OSTs will
2497 * be NULL and won't match the lock's export. */
2498 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2499 loi = lsm->lsm_oinfo[i];
2500 if (!lov->lov_tgts[loi->loi_ost_idx])
2502 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2503 data->lock->l_conn_export &&
2504 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2509 LDLM_ERROR(data->lock, "lock on inode without such object");
2510 dump_lsm(D_ERROR, lsm);
2511 GOTO(out, rc = -ENXIO);
2512 } else if (KEY_IS(KEY_LAST_ID)) {
2513 struct obd_id_info *info = val;
2514 __u32 size = sizeof(obd_id);
2515 struct lov_tgt_desc *tgt;
2517 LASSERT(*vallen == sizeof(struct obd_id_info));
2518 tgt = lov->lov_tgts[info->idx];
2520 if (!tgt || !tgt->ltd_active)
2521 GOTO(out, rc = -ESRCH);
2523 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2525 } else if (KEY_IS(KEY_LOVDESC)) {
2526 struct lov_desc *desc_ret = val;
2527 *desc_ret = lov->desc;
2530 } else if (KEY_IS(KEY_FIEMAP)) {
2531 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2542 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2543 void *key, obd_count vallen, void *val,
2544 struct ptlrpc_request_set *set)
2546 struct obd_device *obddev = class_exp2obd(exp);
2547 struct lov_obd *lov = &obddev->u.lov;
2550 struct lov_tgt_desc *tgt;
2551 unsigned incr, check_uuid,
2552 do_inactive, no_set;
2553 unsigned next_id = 0, mds_con = 0, capa = 0;
2556 incr = check_uuid = do_inactive = no_set = 0;
2559 set = ptlrpc_prep_set();
2565 count = lov->desc.ld_tgt_count;
2567 if (KEY_IS(KEY_NEXT_ID)) {
2568 count = vallen / sizeof(struct obd_id_info);
2569 vallen = sizeof(obd_id);
2570 incr = sizeof(struct obd_id_info);
2573 } else if (KEY_IS(KEY_CHECKSUM)) {
2575 } else if (KEY_IS(KEY_UNLINKED)) {
2576 check_uuid = val ? 1 : 0;
2577 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2578 /* use defaults: do_inactive = incr = 0; */
2579 } else if (KEY_IS(KEY_MDS_CONN)) {
2581 } else if (KEY_IS(KEY_CAPA_KEY)) {
2585 for (i = 0; i < count; i++, val = (char *)val + incr) {
2587 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2589 tgt = lov->lov_tgts[i];
2591 /* OST was disconnected */
2592 if (!tgt || !tgt->ltd_exp)
2595 /* OST is inactive and we don't want inactive OSCs */
2596 if (!tgt->ltd_active && !do_inactive)
2600 struct mds_group_info *mgi;
2602 LASSERT(vallen == sizeof(*mgi));
2603 mgi = (struct mds_group_info *)val;
2605 /* Only want a specific OSC */
2606 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2610 err = obd_set_info_async(tgt->ltd_exp,
2611 keylen, key, sizeof(int),
2613 } else if (next_id) {
2614 err = obd_set_info_async(tgt->ltd_exp,
2615 keylen, key, vallen,
2616 ((struct obd_id_info*)val)->data, set);
2618 struct mds_capa_info *info = (struct mds_capa_info*)val;
2620 LASSERT(vallen == sizeof(*info));
2622 /* Only want a specific OSC */
2624 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2627 err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2628 sizeof(*info->capa),
2632 /* Only want a specific OSC */
2634 !obd_uuid_equals(val, &tgt->ltd_uuid))
2637 err = obd_set_info_async(tgt->ltd_exp,
2638 keylen, key, vallen, val, set);
2647 err = ptlrpc_set_wait(set);
2650 ptlrpc_set_destroy(set);
2655 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2656 struct lov_stripe_md *lsm)
2664 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2665 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2670 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2675 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2676 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2677 if (loi->loi_ar.ar_rc && !rc)
2678 rc = loi->loi_ar.ar_rc;
2679 loi->loi_ar.ar_rc = 0;
2683 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2686 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2687 int cmd, __u64 *offset)
2689 __u32 ssize = lsm->lsm_stripe_size;
2693 do_div(start, ssize);
2694 start = start * ssize;
2696 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2697 ", end "LPU64"\n", *offset, ssize, start,
2699 if (cmd == OBD_CALC_STRIPE_END) {
2700 *offset = start + ssize - 1;
2701 } else if (cmd == OBD_CALC_STRIPE_START) {
2710 void lov_stripe_lock(struct lov_stripe_md *md)
2712 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2713 spin_lock(&md->lsm_lock);
2714 LASSERT(md->lsm_lock_owner == 0);
2715 md->lsm_lock_owner = cfs_curproc_pid();
2717 EXPORT_SYMBOL(lov_stripe_lock);
2719 void lov_stripe_unlock(struct lov_stripe_md *md)
2721 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2722 md->lsm_lock_owner = 0;
2723 spin_unlock(&md->lsm_lock);
2725 EXPORT_SYMBOL(lov_stripe_unlock);
2728 struct obd_ops lov_obd_ops = {
2729 .o_owner = THIS_MODULE,
2730 .o_setup = lov_setup,
2731 .o_precleanup = lov_precleanup,
2732 .o_cleanup = lov_cleanup,
2733 //.o_process_config = lov_process_config,
2734 .o_connect = lov_connect,
2735 .o_disconnect = lov_disconnect,
2736 .o_statfs = lov_statfs,
2737 .o_statfs_async = lov_statfs_async,
2738 .o_packmd = lov_packmd,
2739 .o_unpackmd = lov_unpackmd,
2740 .o_checkmd = lov_checkmd,
2741 .o_create = lov_create,
2742 .o_destroy = lov_destroy,
2743 .o_getattr = lov_getattr,
2744 .o_getattr_async = lov_getattr_async,
2745 .o_setattr = lov_setattr,
2746 .o_setattr_async = lov_setattr_async,
2748 .o_merge_lvb = lov_merge_lvb,
2749 .o_adjust_kms = lov_adjust_kms,
2750 .o_punch = lov_punch,
2752 .o_enqueue = lov_enqueue,
2753 .o_change_cbdata = lov_change_cbdata,
2754 .o_cancel = lov_cancel,
2755 .o_cancel_unused = lov_cancel_unused,
2756 .o_iocontrol = lov_iocontrol,
2757 .o_get_info = lov_get_info,
2758 .o_set_info_async = lov_set_info_async,
2759 .o_extent_calc = lov_extent_calc,
2760 .o_llog_init = lov_llog_init,
2761 .o_llog_finish = lov_llog_finish,
2762 .o_notify = lov_notify,
2763 .o_pool_new = lov_pool_new,
2764 .o_pool_rem = lov_pool_remove,
2765 .o_pool_add = lov_pool_add,
2766 .o_pool_del = lov_pool_del,
2767 .o_getref = lov_getref,
2768 .o_putref = lov_putref,
2771 static quota_interface_t *quota_interface;
2772 extern quota_interface_t lov_quota_interface;
2774 cfs_mem_cache_t *lov_oinfo_slab;
2776 extern struct lu_kmem_descr lov_caches[];
2778 int __init lov_init(void)
2780 struct lprocfs_static_vars lvars = { 0 };
2784 /* print an address of _any_ initialized kernel symbol from this
2785 * module, to allow debugging with gdb that doesn't support data
2786 * symbols from modules.*/
2787 CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2789 rc = lu_kmem_init(lov_caches);
2793 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2794 sizeof(struct lov_oinfo),
2795 0, SLAB_HWCACHE_ALIGN);
2796 if (lov_oinfo_slab == NULL) {
2797 lu_kmem_fini(lov_caches);
2800 lprocfs_lov_init_vars(&lvars);
2802 request_module("lquota");
2803 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2804 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2806 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2807 LUSTRE_LOV_NAME, &lov_device_type);
2810 if (quota_interface)
2811 PORTAL_SYMBOL_PUT(lov_quota_interface);
2812 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2814 lu_kmem_fini(lov_caches);
2821 static void /*__exit*/ lov_exit(void)
2825 lu_device_type_fini(&lov_device_type);
2826 lu_kmem_fini(lov_caches);
2828 if (quota_interface)
2829 PORTAL_SYMBOL_PUT(lov_quota_interface);
2831 class_unregister_type(LUSTRE_LOV_NAME);
2832 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2836 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2837 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2838 MODULE_LICENSE("GPL");
2840 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);