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 = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
808 &pool_hash_operations, 0);
809 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
810 lov->lov_pool_count = 0;
811 rc = lov_ost_pool_init(&lov->lov_packed, 0);
814 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
816 lov_ost_pool_free(&lov->lov_packed);
820 lprocfs_lov_init_vars(&lvars);
821 lprocfs_obd_setup(obd, lvars.obd_vars);
826 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
827 0444, &lov_proc_target_fops, obd);
829 CWARN("Error adding the target_obd file\n");
832 lov->lov_pool_proc_entry = lprocfs_register("pools",
839 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
842 struct lov_obd *lov = &obd->u.lov;
847 case OBD_CLEANUP_EARLY: {
849 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
850 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
852 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
857 case OBD_CLEANUP_EXPORTS:
858 rc = obd_llog_finish(obd, 0);
860 CERROR("failed to cleanup llogging subsystems\n");
866 static int lov_cleanup(struct obd_device *obd)
868 struct lov_obd *lov = &obd->u.lov;
869 struct list_head *pos, *tmp;
870 struct pool_desc *pool;
872 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
873 pool = list_entry(pos, struct pool_desc, pool_list);
874 /* free pool structs */
875 CDEBUG(D_INFO, "delete pool %p\n", pool);
876 lov_pool_del(obd, pool->pool_name);
878 cfs_hash_destroy(lov->lov_pools_hash_body);
879 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
880 lov_ost_pool_free(&lov->lov_packed);
885 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
886 if (!lov->lov_tgts[i])
889 /* Inactive targets may never have connected */
890 if (lov->lov_tgts[i]->ltd_active ||
891 atomic_read(&lov->lov_refcount))
892 /* We should never get here - these
893 should have been removed in the
895 CERROR("lov tgt %d not cleaned!"
896 " deathrow=%d, lovrc=%d\n",
897 i, lov->lov_death_row,
898 atomic_read(&lov->lov_refcount));
899 lov_del_target(obd, i, 0, 0);
902 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
904 lov->lov_tgt_size = 0;
907 /* clear pools parent proc entry only after all pools is killed */
908 lprocfs_obd_cleanup(obd);
910 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
914 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
915 __u32 *indexp, int *genp)
917 struct obd_uuid obd_uuid;
922 switch(cmd = lcfg->lcfg_command) {
923 case LCFG_LOV_ADD_OBD:
924 case LCFG_LOV_ADD_INA:
925 case LCFG_LOV_DEL_OBD: {
928 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
929 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
930 GOTO(out, rc = -EINVAL);
932 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
934 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
935 GOTO(out, rc = -EINVAL);
936 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
937 GOTO(out, rc = -EINVAL);
940 if (cmd == LCFG_LOV_ADD_OBD)
941 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
942 else if (cmd == LCFG_LOV_ADD_INA)
943 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
945 rc = lov_del_target(obd, index, &obd_uuid, gen);
949 struct lprocfs_static_vars lvars = { 0 };
950 struct lov_desc *desc = &(obd->u.lov.desc);
953 GOTO(out, rc = -EINVAL);
955 lprocfs_lov_init_vars(&lvars);
957 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
970 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
971 GOTO(out, rc = -EINVAL);
980 #define log2(n) ffz(~(n))
983 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
984 struct lov_stripe_md **ea,
985 struct obd_trans_info *oti)
989 struct obd_uuid *ost_uuid = NULL;
993 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
994 src_oa->o_flags == OBD_FL_DELORPHAN);
996 lov = &export->exp_obd->u.lov;
1002 if (oti->oti_ost_uuid) {
1003 ost_uuid = oti->oti_ost_uuid;
1004 CDEBUG(D_HA, "clearing orphans only for %s\n",
1008 obd_getref(export->exp_obd);
1009 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1010 struct lov_stripe_md obj_md;
1011 struct lov_stripe_md *obj_mdp = &obj_md;
1012 struct lov_tgt_desc *tgt;
1015 tgt = lov->lov_tgts[i];
1019 /* if called for a specific target, we don't
1020 care if it is not active. */
1021 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1022 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1026 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1029 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1030 obd_uuid2str(ost_uuid));
1032 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1034 LASSERT(lov->lov_tgts[i]->ltd_exp);
1035 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1036 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1037 tmp_oa, &obj_mdp, oti);
1039 /* This export will be disabled until it is recovered,
1040 and then orphan recovery will be completed. */
1041 CERROR("error in orphan recovery on OST idx %d/%d: "
1042 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1049 obd_putref(export->exp_obd);
1055 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1056 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1058 struct lov_stripe_md *obj_mdp, *lsm;
1059 struct lov_obd *lov = &exp->exp_obd->u.lov;
1064 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1065 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1067 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1068 if (obj_mdp == NULL)
1071 ost_idx = src_oa->o_nlink;
1074 GOTO(out, rc = -EINVAL);
1075 if (ost_idx >= lov->desc.ld_tgt_count ||
1076 !lov->lov_tgts[ost_idx])
1077 GOTO(out, rc = -EINVAL);
1079 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1080 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1081 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1082 GOTO(out, rc = -EINVAL);
1086 if (i == lsm->lsm_stripe_count)
1087 GOTO(out, rc = -EINVAL);
1089 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1091 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1095 /* the LOV expects oa->o_id to be set to the LOV object id */
1096 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1097 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1099 struct lov_obd *lov;
1100 struct obd_info oinfo;
1101 struct lov_request_set *set = NULL;
1102 struct lov_request *req;
1103 struct l_wait_info lwi = { 0 };
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_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1142 &req->rq_oi, &req->rq_oi.oi_md, oti);
1145 /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1146 * longer then this */
1147 l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1149 /* we not have ptlrpc set for assign set->interpret and should
1150 * be call interpret function himself. calling from cb_create_update
1151 * not permited because lov_fini_create_set can sleep for long time,
1152 * but we must avoid sleeping in ptlrpcd interpret function. */
1153 rc = lov_fini_create_set(set, ea);
1155 obd_putref(exp->exp_obd);
1159 #define ASSERT_LSM_MAGIC(lsmp) \
1161 LASSERT((lsmp) != NULL); \
1162 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1163 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1164 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1167 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1168 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1169 struct obd_export *md_exp, void *capa)
1171 struct lov_request_set *set;
1172 struct obd_info oinfo;
1173 struct lov_request *req;
1174 struct list_head *pos;
1175 struct lov_obd *lov;
1176 int rc = 0, err = 0;
1179 ASSERT_LSM_MAGIC(lsm);
1181 if (!exp || !exp->exp_obd)
1184 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1186 LASSERT(oti->oti_logcookies);
1189 lov = &exp->exp_obd->u.lov;
1190 obd_getref(exp->exp_obd);
1191 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1195 list_for_each (pos, &set->set_list) {
1196 req = list_entry(pos, struct lov_request, rq_link);
1198 if (oa->o_valid & OBD_MD_FLCOOKIE)
1199 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1201 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1202 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1203 err = lov_update_common_set(set, req, err);
1205 CERROR("error: destroying objid "LPX64" subobj "
1206 LPX64" on OST idx %d: rc = %d\n",
1207 oa->o_id, req->rq_oi.oi_oa->o_id,
1215 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1216 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1218 err = lov_fini_destroy_set(set);
1220 obd_putref(exp->exp_obd);
1221 RETURN(rc ? rc : err);
1224 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1226 struct lov_request_set *set;
1227 struct lov_request *req;
1228 struct list_head *pos;
1229 struct lov_obd *lov;
1230 int err = 0, rc = 0;
1234 ASSERT_LSM_MAGIC(oinfo->oi_md);
1236 if (!exp || !exp->exp_obd)
1239 lov = &exp->exp_obd->u.lov;
1241 rc = lov_prep_getattr_set(exp, oinfo, &set);
1245 list_for_each (pos, &set->set_list) {
1246 req = list_entry(pos, struct lov_request, rq_link);
1248 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1249 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1250 req->rq_oi.oi_oa->o_id, req->rq_idx);
1252 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1254 err = lov_update_common_set(set, req, rc);
1256 CERROR("error: getattr objid "LPX64" subobj "
1257 LPX64" on OST idx %d: rc = %d\n",
1258 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1264 rc = lov_fini_getattr_set(set);
1270 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1273 struct lov_request_set *lovset = (struct lov_request_set *)data;
1277 /* don't do attribute merge if this aysnc op failed */
1279 lovset->set_completes = 0;
1280 err = lov_fini_getattr_set(lovset);
1281 RETURN(rc ? rc : err);
1284 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1285 struct ptlrpc_request_set *rqset)
1287 struct lov_request_set *lovset;
1288 struct lov_obd *lov;
1289 struct list_head *pos;
1290 struct lov_request *req;
1295 ASSERT_LSM_MAGIC(oinfo->oi_md);
1297 if (!exp || !exp->exp_obd)
1300 lov = &exp->exp_obd->u.lov;
1302 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1306 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1307 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1308 oinfo->oi_md->lsm_stripe_size);
1310 list_for_each (pos, &lovset->set_list) {
1311 req = list_entry(pos, struct lov_request, rq_link);
1313 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1314 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1315 req->rq_oi.oi_oa->o_id, req->rq_idx);
1316 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1317 &req->rq_oi, rqset);
1319 CERROR("error: getattr objid "LPX64" subobj "
1320 LPX64" on OST idx %d: rc = %d\n",
1321 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1327 if (!list_empty(&rqset->set_requests)) {
1329 LASSERT (rqset->set_interpret == NULL);
1330 rqset->set_interpret = lov_getattr_interpret;
1331 rqset->set_arg = (void *)lovset;
1336 lovset->set_completes = 0;
1337 err = lov_fini_getattr_set(lovset);
1338 RETURN(rc ? rc : err);
1341 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1342 struct obd_trans_info *oti)
1344 struct lov_request_set *set;
1345 struct lov_obd *lov;
1346 struct list_head *pos;
1347 struct lov_request *req;
1348 int err = 0, rc = 0;
1352 ASSERT_LSM_MAGIC(oinfo->oi_md);
1354 if (!exp || !exp->exp_obd)
1357 /* for now, we only expect the following updates here */
1358 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1359 OBD_MD_FLMODE | OBD_MD_FLATIME |
1360 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1361 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1362 OBD_MD_FLGROUP | OBD_MD_FLUID |
1363 OBD_MD_FLGID | OBD_MD_FLFID |
1365 lov = &exp->exp_obd->u.lov;
1366 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1370 list_for_each (pos, &set->set_list) {
1371 req = list_entry(pos, struct lov_request, rq_link);
1373 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1375 err = lov_update_setattr_set(set, req, rc);
1377 CERROR("error: setattr objid "LPX64" subobj "
1378 LPX64" on OST idx %d: rc = %d\n",
1379 set->set_oi->oi_oa->o_id,
1380 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1385 err = lov_fini_setattr_set(set);
1391 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1394 struct lov_request_set *lovset = (struct lov_request_set *)data;
1399 lovset->set_completes = 0;
1400 err = lov_fini_setattr_set(lovset);
1401 RETURN(rc ? rc : err);
1404 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1405 needed. Otherwise, a client is waiting for responses. */
1406 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1407 struct obd_trans_info *oti,
1408 struct ptlrpc_request_set *rqset)
1410 struct lov_request_set *set;
1411 struct lov_request *req;
1412 struct list_head *pos;
1413 struct lov_obd *lov;
1418 ASSERT_LSM_MAGIC(oinfo->oi_md);
1419 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1421 LASSERT(oti->oti_logcookies);
1424 if (!exp || !exp->exp_obd)
1427 lov = &exp->exp_obd->u.lov;
1428 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1432 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1433 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1434 oinfo->oi_md->lsm_stripe_size);
1436 list_for_each (pos, &set->set_list) {
1437 req = list_entry(pos, struct lov_request, rq_link);
1439 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1440 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1442 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1443 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1444 req->rq_oi.oi_oa->o_id, req->rq_idx);
1446 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1447 &req->rq_oi, oti, rqset);
1449 CERROR("error: setattr objid "LPX64" subobj "
1450 LPX64" on OST idx %d: rc = %d\n",
1451 set->set_oi->oi_oa->o_id,
1452 req->rq_oi.oi_oa->o_id,
1458 /* If we are not waiting for responses on async requests, return. */
1459 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1462 set->set_completes = 0;
1463 err = lov_fini_setattr_set(set);
1464 RETURN(rc ? rc : err);
1467 LASSERT(rqset->set_interpret == NULL);
1468 rqset->set_interpret = lov_setattr_interpret;
1469 rqset->set_arg = (void *)set;
1474 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1477 struct lov_request_set *lovset = (struct lov_request_set *)data;
1482 lovset->set_completes = 0;
1483 err = lov_fini_punch_set(lovset);
1484 RETURN(rc ? rc : err);
1487 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1488 * we can send this 'punch' to just the authoritative node and the nodes
1489 * that the punch will affect. */
1490 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1491 struct obd_trans_info *oti,
1492 struct ptlrpc_request_set *rqset)
1494 struct lov_request_set *set;
1495 struct lov_obd *lov;
1496 struct list_head *pos;
1497 struct lov_request *req;
1502 ASSERT_LSM_MAGIC(oinfo->oi_md);
1504 if (!exp || !exp->exp_obd)
1507 lov = &exp->exp_obd->u.lov;
1508 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1512 list_for_each (pos, &set->set_list) {
1513 req = list_entry(pos, struct lov_request, rq_link);
1515 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1516 &req->rq_oi, NULL, rqset);
1518 CERROR("error: punch objid "LPX64" subobj "LPX64
1519 " on OST idx %d: rc = %d\n",
1520 set->set_oi->oi_oa->o_id,
1521 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1526 if (rc || list_empty(&rqset->set_requests)) {
1528 err = lov_fini_punch_set(set);
1529 RETURN(rc ? rc : err);
1532 LASSERT(rqset->set_interpret == NULL);
1533 rqset->set_interpret = lov_punch_interpret;
1534 rqset->set_arg = (void *)set;
1539 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1540 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1543 struct lov_request_set *set;
1544 struct obd_info oinfo;
1545 struct lov_obd *lov;
1546 struct list_head *pos;
1547 struct lov_request *req;
1548 int err = 0, rc = 0;
1551 ASSERT_LSM_MAGIC(lsm);
1556 lov = &exp->exp_obd->u.lov;
1557 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1561 list_for_each (pos, &set->set_list) {
1562 req = list_entry(pos, struct lov_request, rq_link);
1564 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1565 req->rq_oi.oi_oa, NULL,
1566 req->rq_oi.oi_policy.l_extent.start,
1567 req->rq_oi.oi_policy.l_extent.end, capa);
1568 err = lov_update_common_set(set, req, rc);
1570 CERROR("error: fsync objid "LPX64" subobj "LPX64
1571 " on OST idx %d: rc = %d\n",
1572 set->set_oi->oi_oa->o_id,
1573 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1578 err = lov_fini_sync_set(set);
1584 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1585 obd_count oa_bufs, struct brw_page *pga)
1587 struct obd_info oinfo = { { { 0 } } };
1590 oinfo.oi_oa = lov_oinfo->oi_oa;
1592 /* The caller just wants to know if there's a chance that this
1593 * I/O can succeed */
1594 for (i = 0; i < oa_bufs; i++) {
1595 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1596 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1599 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1600 pga[i].off + pga[i].count - 1,
1604 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1605 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1609 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1617 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1618 obd_count oa_bufs, struct brw_page *pga,
1619 struct obd_trans_info *oti)
1621 struct lov_request_set *set;
1622 struct lov_request *req;
1623 struct list_head *pos;
1624 struct lov_obd *lov = &exp->exp_obd->u.lov;
1628 ASSERT_LSM_MAGIC(oinfo->oi_md);
1630 if (cmd == OBD_BRW_CHECK) {
1631 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1635 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1639 list_for_each (pos, &set->set_list) {
1640 struct obd_export *sub_exp;
1641 struct brw_page *sub_pga;
1642 req = list_entry(pos, struct lov_request, rq_link);
1644 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1645 sub_pga = set->set_pga + req->rq_pgaidx;
1646 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1650 lov_update_common_set(set, req, rc);
1653 err = lov_fini_brw_set(set);
1659 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1662 struct lov_request_set *lovset = (struct lov_request_set *)data;
1664 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1668 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1669 struct ldlm_enqueue_info *einfo,
1670 struct ptlrpc_request_set *rqset)
1672 ldlm_mode_t mode = einfo->ei_mode;
1673 struct lov_request_set *set;
1674 struct lov_request *req;
1675 struct list_head *pos;
1676 struct lov_obd *lov;
1681 ASSERT_LSM_MAGIC(oinfo->oi_md);
1682 LASSERT(mode == (mode & -mode));
1684 /* we should never be asked to replay a lock this way. */
1685 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1687 if (!exp || !exp->exp_obd)
1690 lov = &exp->exp_obd->u.lov;
1691 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1695 list_for_each (pos, &set->set_list) {
1696 req = list_entry(pos, struct lov_request, rq_link);
1698 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1699 &req->rq_oi, einfo, rqset);
1704 if (rqset && !list_empty(&rqset->set_requests)) {
1706 LASSERT(rqset->set_interpret == NULL);
1707 rqset->set_interpret = lov_enqueue_interpret;
1708 rqset->set_arg = (void *)set;
1712 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1716 static int lov_change_cbdata(struct obd_export *exp,
1717 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1720 struct lov_obd *lov;
1724 ASSERT_LSM_MAGIC(lsm);
1726 if (!exp || !exp->exp_obd)
1729 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1731 lov = &exp->exp_obd->u.lov;
1732 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1733 struct lov_stripe_md submd;
1734 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1736 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1737 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1741 submd.lsm_object_id = loi->loi_id;
1742 submd.lsm_object_gr = lsm->lsm_object_gr;
1743 submd.lsm_stripe_count = 0;
1744 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1750 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1751 __u32 mode, struct lustre_handle *lockh)
1753 struct lov_request_set *set;
1754 struct obd_info oinfo;
1755 struct lov_request *req;
1756 struct list_head *pos;
1757 struct lov_obd *lov;
1758 struct lustre_handle *lov_lockhp;
1759 int err = 0, rc = 0;
1762 ASSERT_LSM_MAGIC(lsm);
1764 if (!exp || !exp->exp_obd)
1767 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1769 lov = &exp->exp_obd->u.lov;
1770 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1774 list_for_each (pos, &set->set_list) {
1775 req = list_entry(pos, struct lov_request, rq_link);
1776 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1778 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1779 req->rq_oi.oi_md, mode, lov_lockhp);
1780 rc = lov_update_common_set(set, req, rc);
1782 CERROR("error: cancel objid "LPX64" subobj "
1783 LPX64" on OST idx %d: rc = %d\n",
1785 req->rq_oi.oi_md->lsm_object_id,
1791 lov_fini_cancel_set(set);
1795 static int lov_cancel_unused(struct obd_export *exp,
1796 struct lov_stripe_md *lsm,
1797 int flags, void *opaque)
1799 struct lov_obd *lov;
1803 if (!exp || !exp->exp_obd)
1806 lov = &exp->exp_obd->u.lov;
1808 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1810 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1813 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1821 ASSERT_LSM_MAGIC(lsm);
1823 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1824 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1825 struct lov_stripe_md submd;
1826 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1829 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1830 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1834 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1835 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1837 submd.lsm_object_id = loi->loi_id;
1838 submd.lsm_object_gr = lsm->lsm_object_gr;
1839 submd.lsm_stripe_count = 0;
1840 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1841 &submd, flags, opaque);
1842 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1843 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1844 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1845 loi->loi_id, loi->loi_ost_idx, err);
1853 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1855 struct lov_request_set *lovset = (struct lov_request_set *)data;
1860 lovset->set_completes = 0;
1862 err = lov_fini_statfs_set(lovset);
1863 RETURN(rc ? rc : err);
1866 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1867 __u64 max_age, struct ptlrpc_request_set *rqset)
1869 struct lov_request_set *set;
1870 struct lov_request *req;
1871 struct list_head *pos;
1872 struct lov_obd *lov;
1876 LASSERT(oinfo != NULL);
1877 LASSERT(oinfo->oi_osfs != NULL);
1880 rc = lov_prep_statfs_set(obd, oinfo, &set);
1884 list_for_each (pos, &set->set_list) {
1885 struct obd_device *osc_obd;
1887 req = list_entry(pos, struct lov_request, rq_link);
1889 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1890 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1895 if (rc || list_empty(&rqset->set_requests)) {
1898 set->set_completes = 0;
1899 err = lov_fini_statfs_set(set);
1900 RETURN(rc ? rc : err);
1903 LASSERT(rqset->set_interpret == NULL);
1904 rqset->set_interpret = lov_statfs_interpret;
1905 rqset->set_arg = (void *)set;
1909 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1910 __u64 max_age, __u32 flags)
1912 struct ptlrpc_request_set *set = NULL;
1913 struct obd_info oinfo = { { { 0 } } };
1918 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1919 * statfs requests */
1920 set = ptlrpc_prep_set();
1924 oinfo.oi_osfs = osfs;
1925 oinfo.oi_flags = flags;
1926 rc = lov_statfs_async(obd, &oinfo, max_age, set);
1928 rc = ptlrpc_set_wait(set);
1929 ptlrpc_set_destroy(set);
1934 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1935 void *karg, void *uarg)
1937 struct obd_device *obddev = class_exp2obd(exp);
1938 struct lov_obd *lov = &obddev->u.lov;
1939 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1940 struct obd_uuid *uuidp;
1944 case IOC_OBD_STATFS: {
1945 struct obd_ioctl_data *data = karg;
1946 struct obd_device *osc_obd;
1947 struct obd_statfs stat_buf = {0};
1950 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1951 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1953 if ((index >= count))
1956 if (!lov->lov_tgts[index])
1957 /* Try again with the next index */
1959 if (!lov->lov_tgts[index]->ltd_active)
1962 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1966 /* got statfs data */
1967 rc = obd_statfs(osc_obd, &stat_buf,
1968 cfs_time_current_64() - HZ, 0);
1971 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1974 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1979 case OBD_IOC_LOV_GET_CONFIG: {
1980 struct obd_ioctl_data *data;
1981 struct lov_desc *desc;
1986 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1989 data = (struct obd_ioctl_data *)buf;
1991 if (sizeof(*desc) > data->ioc_inllen1) {
1992 obd_ioctl_freedata(buf, len);
1996 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1997 obd_ioctl_freedata(buf, len);
2001 if (sizeof(__u32) * count > data->ioc_inllen3) {
2002 obd_ioctl_freedata(buf, len);
2006 desc = (struct lov_desc *)data->ioc_inlbuf1;
2007 memcpy(desc, &(lov->desc), sizeof(*desc));
2009 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2010 genp = (__u32 *)data->ioc_inlbuf3;
2011 /* the uuid will be empty for deleted OSTs */
2012 for (i = 0; i < count; i++, uuidp++, genp++) {
2013 if (!lov->lov_tgts[i])
2015 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2016 *genp = lov->lov_tgts[i]->ltd_gen;
2019 if (copy_to_user((void *)uarg, buf, len))
2021 obd_ioctl_freedata(buf, len);
2024 case LL_IOC_LOV_SETSTRIPE:
2025 rc = lov_setstripe(exp, len, karg, uarg);
2027 case LL_IOC_LOV_GETSTRIPE:
2028 rc = lov_getstripe(exp, karg, uarg);
2030 case LL_IOC_LOV_SETEA:
2031 rc = lov_setea(exp, karg, uarg);
2033 case OBD_IOC_QUOTACTL: {
2034 struct if_quotactl *qctl = karg;
2035 struct lov_tgt_desc *tgt = NULL;
2036 struct obd_quotactl *oqctl;
2038 if (qctl->qc_valid == QC_OSTIDX) {
2039 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2042 tgt = lov->lov_tgts[qctl->qc_idx];
2043 if (!tgt || !tgt->ltd_exp)
2045 } else if (qctl->qc_valid == QC_UUID) {
2046 for (i = 0; i < count; i++) {
2047 tgt = lov->lov_tgts[i];
2049 !obd_uuid_equals(&tgt->ltd_uuid,
2053 if (tgt->ltd_exp == NULL)
2065 LASSERT(tgt && tgt->ltd_exp);
2066 OBD_ALLOC_PTR(oqctl);
2070 QCTL_COPY(oqctl, qctl);
2071 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2073 QCTL_COPY(qctl, oqctl);
2074 qctl->qc_valid = QC_OSTIDX;
2075 qctl->obd_uuid = tgt->ltd_uuid;
2077 OBD_FREE_PTR(oqctl);
2086 for (i = 0; i < count; i++) {
2089 /* OST was disconnected */
2090 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2093 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2095 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2098 if (lov->lov_tgts[i]->ltd_active) {
2099 CDEBUG(err == -ENOTTY ?
2100 D_IOCTL : D_WARNING,
2101 "iocontrol OSC %s on OST "
2102 "idx %d cmd %x: err = %d\n",
2103 lov_uuid2str(lov, i),
2120 #define FIEMAP_BUFFER_SIZE 4096
2123 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2124 * call. The local end offset and the device are sent in the first
2125 * fm_extent. This function calculates the stripe number from the index.
2126 * This function returns a stripe_no on which mapping is to be restarted.
2128 * This function returns fm_end_offset which is the in-OST offset at which
2129 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2130 * will re-calculate proper offset in next stripe.
2131 * Note that the first extent is passed to lov_get_info via the value field.
2133 * \param fiemap fiemap request header
2134 * \param lsm striping information for the file
2135 * \param fm_start logical start of mapping
2136 * \param fm_end logical end of mapping
2137 * \param start_stripe starting stripe will be returned in this
2139 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2140 struct lov_stripe_md *lsm, obd_size fm_start,
2141 obd_size fm_end, int *start_stripe)
2143 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2144 obd_off lun_start, lun_end;
2145 obd_size fm_end_offset;
2146 int stripe_no = -1, i;
2148 if (fiemap->fm_extent_count == 0 ||
2149 fiemap->fm_extents[0].fe_logical == 0)
2152 /* Find out stripe_no from ost_index saved in the fe_device */
2153 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2154 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2155 fiemap->fm_extents[0].fe_device) {
2161 /* If we have finished mapping on previous device, shift logical
2162 * offset to start of next device */
2163 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2164 &lun_start, &lun_end)) != 0 &&
2165 local_end < lun_end) {
2166 fm_end_offset = local_end;
2167 *start_stripe = stripe_no;
2169 /* This is a special value to indicate that caller should
2170 * calculate offset in next stripe. */
2172 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2175 return fm_end_offset;
2179 * We calculate on which OST the mapping will end. If the length of mapping
2180 * is greater than (stripe_size * stripe_count) then the last_stripe will
2181 * will be one just before start_stripe. Else we check if the mapping
2182 * intersects each OST and find last_stripe.
2183 * This function returns the last_stripe and also sets the stripe_count
2184 * over which the mapping is spread
2186 * \param lsm striping information for the file
2187 * \param fm_start logical start of mapping
2188 * \param fm_end logical end of mapping
2189 * \param start_stripe starting stripe of the mapping
2190 * \param stripe_count the number of stripes across which to map is returned
2192 * \retval last_stripe return the last stripe of the mapping
2194 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2195 obd_size fm_end, int start_stripe,
2199 obd_off obd_start, obd_end;
2202 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2203 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2205 *stripe_count = lsm->lsm_stripe_count;
2207 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2208 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2209 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2210 &obd_start, &obd_end)) == 0)
2214 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2221 * Set fe_device and copy extents from local buffer into main return buffer.
2223 * \param fiemap fiemap request header
2224 * \param lcl_fm_ext array of local fiemap extents to be copied
2225 * \param ost_index OST index to be written into the fm_device field for each
2227 * \param ext_count number of extents to be copied
2228 * \param current_extent where to start copying in main extent array
2230 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2231 struct ll_fiemap_extent *lcl_fm_ext,
2232 int ost_index, unsigned int ext_count,
2238 for (ext = 0; ext < ext_count; ext++) {
2239 lcl_fm_ext[ext].fe_device = ost_index;
2240 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2243 /* Copy fm_extent's from fm_local to return buffer */
2244 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2245 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2249 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2250 * This also handles the restarting of FIEMAP calls in case mapping overflows
2251 * the available number of extents in single call.
2253 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2254 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2256 struct ll_fiemap_info_key *fm_key = key;
2257 struct ll_user_fiemap *fiemap = val;
2258 struct ll_user_fiemap *fm_local = NULL;
2259 struct ll_fiemap_extent *lcl_fm_ext;
2261 unsigned int get_num_extents = 0;
2262 int ost_index = 0, actual_start_stripe, start_stripe;
2263 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2265 int current_extent = 0, rc = 0, i;
2266 int ost_eof = 0; /* EOF for object */
2267 int ost_done = 0; /* done with required mapping for this OST? */
2269 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2270 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2275 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2276 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2278 OBD_ALLOC(fm_local, buffer_size);
2279 if (fm_local == NULL)
2280 GOTO(out, rc = -ENOMEM);
2281 lcl_fm_ext = &fm_local->fm_extents[0];
2283 count_local = fiemap_size_to_count(buffer_size);
2285 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2286 fm_start = fiemap->fm_start;
2287 fm_length = fiemap->fm_length;
2288 /* Calculate start stripe, last stripe and length of mapping */
2289 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2290 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2291 fm_start + fm_length - 1);
2292 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2293 if (fm_end > fm_key->oa.o_size)
2294 fm_end = fm_key->oa.o_size;
2296 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2297 actual_start_stripe, &stripe_count);
2299 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2302 if (fiemap->fm_extent_count == 0) {
2303 get_num_extents = 1;
2307 /* Check each stripe */
2308 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2309 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2310 obd_size req_fm_len; /* Stores length of required mapping */
2311 obd_size len_mapped_single_call;
2312 obd_off lun_start, lun_end, obd_object_end;
2313 unsigned int ext_count;
2315 cur_stripe_wrap = cur_stripe;
2317 /* Find out range of mapping on this stripe */
2318 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2319 &lun_start, &obd_object_end)) == 0)
2322 /* If this is a continuation FIEMAP call and we are on
2323 * starting stripe then lun_start needs to be set to
2325 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2326 lun_start = fm_end_offset;
2328 if (fm_length != ~0ULL) {
2329 /* Handle fm_start + fm_length overflow */
2330 if (fm_start + fm_length < fm_start)
2331 fm_length = ~0ULL - fm_start;
2332 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2338 if (lun_start == lun_end)
2341 req_fm_len = obd_object_end - lun_start;
2342 fm_local->fm_length = 0;
2343 len_mapped_single_call = 0;
2345 /* If the output buffer is very large and the objects have many
2346 * extents we may need to loop on a single OST repeatedly */
2350 if (get_num_extents == 0) {
2351 /* Don't get too many extents. */
2352 if (current_extent + count_local >
2353 fiemap->fm_extent_count)
2354 count_local = fiemap->fm_extent_count -
2358 lun_start += len_mapped_single_call;
2359 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2360 req_fm_len = fm_local->fm_length;
2361 fm_local->fm_extent_count = count_local;
2362 fm_local->fm_mapped_extents = 0;
2363 fm_local->fm_flags = fiemap->fm_flags;
2365 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2366 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2368 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2369 GOTO(out, rc = -EINVAL);
2371 /* If OST is inactive, return extent with UNKNOWN flag */
2372 if (!lov->lov_tgts[ost_index]->ltd_active) {
2373 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2374 fm_local->fm_mapped_extents = 1;
2376 lcl_fm_ext[0].fe_logical = lun_start;
2377 lcl_fm_ext[0].fe_length = obd_object_end -
2379 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2384 fm_local->fm_start = lun_start;
2385 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2386 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2387 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2388 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2389 keylen, key, vallen, fm_local, lsm);
2394 ext_count = fm_local->fm_mapped_extents;
2395 if (ext_count == 0) {
2397 /* If last stripe has hole at the end,
2398 * then we need to return */
2399 if (cur_stripe_wrap == last_stripe) {
2400 fiemap->fm_mapped_extents = 0;
2406 /* If we just need num of extents then go to next device */
2407 if (get_num_extents) {
2408 current_extent += ext_count;
2412 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2413 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2415 /* Have we finished mapping on this device? */
2416 if (req_fm_len <= len_mapped_single_call)
2419 /* Clear the EXTENT_LAST flag which can be present on
2421 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2422 lcl_fm_ext[ext_count - 1].fe_flags &=
2423 ~FIEMAP_EXTENT_LAST;
2425 curr_loc = lov_stripe_size(lsm,
2426 lcl_fm_ext[ext_count - 1].fe_logical+
2427 lcl_fm_ext[ext_count - 1].fe_length,
2429 if (curr_loc >= fm_key->oa.o_size)
2432 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2433 ost_index, ext_count,
2436 current_extent += ext_count;
2438 /* Ran out of available extents? */
2439 if (current_extent >= fiemap->fm_extent_count)
2441 } while (ost_done == 0 && ost_eof == 0);
2443 if (cur_stripe_wrap == last_stripe)
2448 /* Indicate that we are returning device offsets unless file just has
2450 if (lsm->lsm_stripe_count > 1)
2451 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2453 if (get_num_extents)
2454 goto skip_last_device_calc;
2456 /* Check if we have reached the last stripe and whether mapping for that
2457 * stripe is done. */
2458 if (cur_stripe_wrap == last_stripe) {
2459 if (ost_done || ost_eof)
2460 fiemap->fm_extents[current_extent - 1].fe_flags |=
2464 skip_last_device_calc:
2465 fiemap->fm_mapped_extents = current_extent;
2468 OBD_FREE(fm_local, buffer_size);
2472 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2473 void *key, __u32 *vallen, void *val,
2474 struct lov_stripe_md *lsm)
2476 struct obd_device *obddev = class_exp2obd(exp);
2477 struct lov_obd *lov = &obddev->u.lov;
2481 if (!vallen || !val)
2486 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2489 struct ldlm_lock *lock;
2491 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2492 struct lov_oinfo *loi;
2493 __u32 *stripe = val;
2495 if (*vallen < sizeof(*stripe))
2496 GOTO(out, rc = -EFAULT);
2497 *vallen = sizeof(*stripe);
2499 /* XXX This is another one of those bits that will need to
2500 * change if we ever actually support nested LOVs. It uses
2501 * the lock's export to find out which stripe it is. */
2502 /* XXX - it's assumed all the locks for deleted OSTs have
2503 * been cancelled. Also, the export for deleted OSTs will
2504 * be NULL and won't match the lock's export. */
2505 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2506 loi = lsm->lsm_oinfo[i];
2507 if (!lov->lov_tgts[loi->loi_ost_idx])
2509 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2510 data->lock->l_conn_export &&
2511 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2516 LDLM_ERROR(data->lock, "lock on inode without such object");
2517 dump_lsm(D_ERROR, lsm);
2518 GOTO(out, rc = -ENXIO);
2519 } else if (KEY_IS(KEY_LAST_ID)) {
2520 struct obd_id_info *info = val;
2521 __u32 size = sizeof(obd_id);
2522 struct lov_tgt_desc *tgt;
2524 LASSERT(*vallen == sizeof(struct obd_id_info));
2525 tgt = lov->lov_tgts[info->idx];
2527 if (!tgt || !tgt->ltd_active)
2528 GOTO(out, rc = -ESRCH);
2530 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2532 } else if (KEY_IS(KEY_LOVDESC)) {
2533 struct lov_desc *desc_ret = val;
2534 *desc_ret = lov->desc;
2537 } else if (KEY_IS(KEY_FIEMAP)) {
2538 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2549 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2550 void *key, obd_count vallen, void *val,
2551 struct ptlrpc_request_set *set)
2553 struct obd_device *obddev = class_exp2obd(exp);
2554 struct lov_obd *lov = &obddev->u.lov;
2557 struct lov_tgt_desc *tgt;
2558 unsigned incr, check_uuid,
2559 do_inactive, no_set;
2560 unsigned next_id = 0, mds_con = 0, capa = 0;
2563 incr = check_uuid = do_inactive = no_set = 0;
2566 set = ptlrpc_prep_set();
2572 count = lov->desc.ld_tgt_count;
2574 if (KEY_IS(KEY_NEXT_ID)) {
2575 count = vallen / sizeof(struct obd_id_info);
2576 vallen = sizeof(obd_id);
2577 incr = sizeof(struct obd_id_info);
2580 } else if (KEY_IS(KEY_CHECKSUM)) {
2582 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2583 /* use defaults: do_inactive = incr = 0; */
2584 } else if (KEY_IS(KEY_MDS_CONN)) {
2586 } else if (KEY_IS(KEY_CAPA_KEY)) {
2590 for (i = 0; i < count; i++, val = (char *)val + incr) {
2592 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2594 tgt = lov->lov_tgts[i];
2596 /* OST was disconnected */
2597 if (!tgt || !tgt->ltd_exp)
2600 /* OST is inactive and we don't want inactive OSCs */
2601 if (!tgt->ltd_active && !do_inactive)
2605 struct mds_group_info *mgi;
2607 LASSERT(vallen == sizeof(*mgi));
2608 mgi = (struct mds_group_info *)val;
2610 /* Only want a specific OSC */
2611 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2615 err = obd_set_info_async(tgt->ltd_exp,
2616 keylen, key, sizeof(int),
2618 } else if (next_id) {
2619 err = obd_set_info_async(tgt->ltd_exp,
2620 keylen, key, vallen,
2621 ((struct obd_id_info*)val)->data, set);
2623 struct mds_capa_info *info = (struct mds_capa_info*)val;
2625 LASSERT(vallen == sizeof(*info));
2627 /* Only want a specific OSC */
2629 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2632 err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2633 sizeof(*info->capa),
2637 /* Only want a specific OSC */
2639 !obd_uuid_equals(val, &tgt->ltd_uuid))
2642 err = obd_set_info_async(tgt->ltd_exp,
2643 keylen, key, vallen, val, set);
2652 err = ptlrpc_set_wait(set);
2655 ptlrpc_set_destroy(set);
2660 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2665 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2666 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2667 if (loi->loi_ar.ar_rc && !rc)
2668 rc = loi->loi_ar.ar_rc;
2669 loi->loi_ar.ar_rc = 0;
2673 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2676 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2677 int cmd, __u64 *offset)
2679 __u32 ssize = lsm->lsm_stripe_size;
2683 do_div(start, ssize);
2684 start = start * ssize;
2686 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2687 ", end "LPU64"\n", *offset, ssize, start,
2689 if (cmd == OBD_CALC_STRIPE_END) {
2690 *offset = start + ssize - 1;
2691 } else if (cmd == OBD_CALC_STRIPE_START) {
2700 void lov_stripe_lock(struct lov_stripe_md *md)
2702 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2703 spin_lock(&md->lsm_lock);
2704 LASSERT(md->lsm_lock_owner == 0);
2705 md->lsm_lock_owner = cfs_curproc_pid();
2707 EXPORT_SYMBOL(lov_stripe_lock);
2709 void lov_stripe_unlock(struct lov_stripe_md *md)
2711 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2712 md->lsm_lock_owner = 0;
2713 spin_unlock(&md->lsm_lock);
2715 EXPORT_SYMBOL(lov_stripe_unlock);
2718 struct obd_ops lov_obd_ops = {
2719 .o_owner = THIS_MODULE,
2720 .o_setup = lov_setup,
2721 .o_precleanup = lov_precleanup,
2722 .o_cleanup = lov_cleanup,
2723 //.o_process_config = lov_process_config,
2724 .o_connect = lov_connect,
2725 .o_disconnect = lov_disconnect,
2726 .o_statfs = lov_statfs,
2727 .o_statfs_async = lov_statfs_async,
2728 .o_packmd = lov_packmd,
2729 .o_unpackmd = lov_unpackmd,
2730 .o_create = lov_create,
2731 .o_destroy = lov_destroy,
2732 .o_getattr = lov_getattr,
2733 .o_getattr_async = lov_getattr_async,
2734 .o_setattr = lov_setattr,
2735 .o_setattr_async = lov_setattr_async,
2737 .o_merge_lvb = lov_merge_lvb,
2738 .o_adjust_kms = lov_adjust_kms,
2739 .o_punch = lov_punch,
2741 .o_enqueue = lov_enqueue,
2742 .o_change_cbdata = lov_change_cbdata,
2743 .o_cancel = lov_cancel,
2744 .o_cancel_unused = lov_cancel_unused,
2745 .o_iocontrol = lov_iocontrol,
2746 .o_get_info = lov_get_info,
2747 .o_set_info_async = lov_set_info_async,
2748 .o_extent_calc = lov_extent_calc,
2749 .o_llog_init = lov_llog_init,
2750 .o_llog_finish = lov_llog_finish,
2751 .o_notify = lov_notify,
2752 .o_pool_new = lov_pool_new,
2753 .o_pool_rem = lov_pool_remove,
2754 .o_pool_add = lov_pool_add,
2755 .o_pool_del = lov_pool_del,
2756 .o_getref = lov_getref,
2757 .o_putref = lov_putref,
2760 static quota_interface_t *quota_interface;
2761 extern quota_interface_t lov_quota_interface;
2763 cfs_mem_cache_t *lov_oinfo_slab;
2765 extern struct lu_kmem_descr lov_caches[];
2767 int __init lov_init(void)
2769 struct lprocfs_static_vars lvars = { 0 };
2773 /* print an address of _any_ initialized kernel symbol from this
2774 * module, to allow debugging with gdb that doesn't support data
2775 * symbols from modules.*/
2776 CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2778 rc = lu_kmem_init(lov_caches);
2782 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2783 sizeof(struct lov_oinfo),
2784 0, SLAB_HWCACHE_ALIGN);
2785 if (lov_oinfo_slab == NULL) {
2786 lu_kmem_fini(lov_caches);
2789 lprocfs_lov_init_vars(&lvars);
2791 request_module("lquota");
2792 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2793 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2795 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2796 LUSTRE_LOV_NAME, &lov_device_type);
2799 if (quota_interface)
2800 PORTAL_SYMBOL_PUT(lov_quota_interface);
2801 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2803 lu_kmem_fini(lov_caches);
2810 static void /*__exit*/ lov_exit(void)
2814 lu_device_type_fini(&lov_device_type);
2815 lu_kmem_fini(lov_caches);
2817 if (quota_interface)
2818 PORTAL_SYMBOL_PUT(lov_quota_interface);
2820 class_unregister_type(LUSTRE_LOV_NAME);
2821 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2825 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2826 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2827 MODULE_LICENSE("GPL");
2829 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);