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"
72 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
73 Any function that expects lov_tgts to remain stationary must take a ref. */
74 void lov_getref(struct obd_device *obd)
76 struct lov_obd *lov = &obd->u.lov;
78 /* nobody gets through here until lov_putref is done */
79 mutex_down(&lov->lov_lock);
80 atomic_inc(&lov->lov_refcount);
81 mutex_up(&lov->lov_lock);
85 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
87 void lov_putref(struct obd_device *obd)
89 struct lov_obd *lov = &obd->u.lov;
91 mutex_down(&lov->lov_lock);
92 /* ok to dec to 0 more than once -- ltd_exp's will be null */
93 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
96 struct lov_tgt_desc *tgt, *n;
97 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
99 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
100 tgt = lov->lov_tgts[i];
102 if (!tgt || !tgt->ltd_reap)
104 list_add(&tgt->ltd_kill, &kill);
105 /* XXX - right now there is a dependency on ld_tgt_count
106 * being the maximum tgt index for computing the
107 * mds_max_easize. So we can't shrink it. */
108 lov_ost_pool_remove(&lov->lov_packed, i);
109 lov->lov_tgts[i] = NULL;
110 lov->lov_death_row--;
112 mutex_up(&lov->lov_lock);
114 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
115 list_del(&tgt->ltd_kill);
117 __lov_del_obd(obd, tgt);
120 mutex_up(&lov->lov_lock);
124 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
126 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
127 enum obd_notify_event ev, void *data);
130 #define MAX_STRING_SIZE 128
131 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
132 struct obd_connect_data *data)
134 struct lov_obd *lov = &obd->u.lov;
135 struct obd_uuid tgt_uuid;
136 struct obd_device *tgt_obd;
137 struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
138 struct obd_import *imp;
141 cfs_proc_dir_entry_t *lov_proc_dir;
146 if (!lov->lov_tgts[index])
149 tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
151 tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
155 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
158 if (!tgt_obd->obd_set_up) {
159 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
163 /* override the sp_me from lov */
164 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
166 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
167 data->ocd_index = index;
170 * Divine LOV knows that OBDs under it are OSCs.
172 imp = tgt_obd->u.cli.cl_import;
175 tgt_obd->obd_no_recov = 0;
176 /* FIXME this is probably supposed to be
177 ptlrpc_set_import_active. Horrible naming. */
178 ptlrpc_activate_import(imp);
181 rc = obd_register_observer(tgt_obd, obd);
183 CERROR("Target %s register_observer error %d\n",
184 obd_uuid2str(&tgt_uuid), rc);
189 if (imp->imp_invalid) {
190 CERROR("not connecting OSC %s; administratively "
191 "disabled\n", obd_uuid2str(&tgt_uuid));
195 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
196 &lov_osc_uuid, data, NULL);
197 if (rc || !lov->lov_tgts[index]->ltd_exp) {
198 CERROR("Target %s connect error %d\n",
199 obd_uuid2str(&tgt_uuid), rc);
203 lov->lov_tgts[index]->ltd_reap = 0;
205 lov->lov_tgts[index]->ltd_active = 1;
206 lov->desc.ld_active_tgt_count++;
207 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
209 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
210 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
213 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
215 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
216 cfs_proc_dir_entry_t *osc_symlink;
217 char name[MAX_STRING_SIZE];
219 LASSERT(osc_obd != NULL);
220 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
221 LASSERT(osc_obd->obd_type->typ_name != NULL);
222 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
223 osc_obd->obd_type->typ_name,
225 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
227 if (osc_symlink == NULL) {
228 CERROR("could not register LOV target "
229 "/proc/fs/lustre/%s/%s/target_obds/%s.",
230 obd->obd_type->typ_name, obd->obd_name,
232 lprocfs_remove(&lov_proc_dir);
237 rc = qos_add_tgt(obd, index);
239 CERROR("qos_add_tgt failed %d\n", rc);
244 static int lov_connect(const struct lu_env *env,
245 struct obd_export **exp, struct obd_device *obd,
246 struct obd_uuid *cluuid, struct obd_connect_data *data,
249 struct lov_obd *lov = &obd->u.lov;
250 struct lov_tgt_desc *tgt;
251 struct lustre_handle conn;
255 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
257 rc = class_connect(&conn, obd, cluuid);
261 *exp = class_conn2export(&conn);
263 /* Why should there ever be more than 1 connect? */
265 LASSERT(lov->lov_connects == 1);
267 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
269 lov->lov_ocd = *data;
272 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
273 tgt = lov->lov_tgts[i];
274 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
276 /* Flags will be lowest common denominator */
277 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
279 CERROR("%s: lov connect tgt %d failed: %d\n",
280 obd->obd_name, i, rc);
283 /* connect to administrative disabled ost */
284 if (!lov->lov_tgts[i]->ltd_exp)
287 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
288 OBD_NOTIFY_CONNECT, (void *)&i);
290 CERROR("%s error sending notify %d\n",
299 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
301 cfs_proc_dir_entry_t *lov_proc_dir;
302 struct lov_obd *lov = &obd->u.lov;
303 struct obd_device *osc_obd;
307 osc_obd = class_exp2obd(tgt->ltd_exp);
308 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
309 obd->obd_name, osc_obd->obd_name);
311 if (tgt->ltd_active) {
313 lov->desc.ld_active_tgt_count--;
314 tgt->ltd_exp->exp_obd->obd_inactive = 1;
317 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
319 cfs_proc_dir_entry_t *osc_symlink;
321 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
323 lprocfs_remove(&osc_symlink);
325 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
326 obd->obd_type->typ_name, obd->obd_name,
332 /* Pass it on to our clients.
333 * XXX This should be an argument to disconnect,
334 * XXX not a back-door flag on the OBD. Ah well.
336 osc_obd->obd_force = obd->obd_force;
337 osc_obd->obd_fail = obd->obd_fail;
338 osc_obd->obd_no_recov = obd->obd_no_recov;
341 obd_register_observer(osc_obd, NULL);
343 rc = obd_disconnect(tgt->ltd_exp);
345 CERROR("Target %s disconnect error %d\n",
346 tgt->ltd_uuid.uuid, rc);
350 qos_del_tgt(obd, tgt);
356 static int lov_disconnect(struct obd_export *exp)
358 struct obd_device *obd = class_exp2obd(exp);
359 struct lov_obd *lov = &obd->u.lov;
366 /* Only disconnect the underlying layers on the final disconnect. */
368 if (lov->lov_connects != 0) {
369 /* why should there be more than 1 connect? */
370 CERROR("disconnect #%d\n", lov->lov_connects);
374 /* Let's hold another reference so lov_del_obd doesn't spin through
378 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
379 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
380 /* Disconnection is the last we know about an obd */
381 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
387 rc = class_disconnect(exp); /* bz 9811 */
393 * -EINVAL : UUID can't be found in the LOV's target list
394 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
395 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
396 * any >= 0 : is log target index
398 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
401 struct lov_obd *lov = &obd->u.lov;
402 struct lov_tgt_desc *tgt;
406 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
407 lov, uuid->uuid, activate);
410 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
411 tgt = lov->lov_tgts[index];
412 if (!tgt || !tgt->ltd_exp)
415 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
416 index, obd_uuid2str(&tgt->ltd_uuid),
417 tgt->ltd_exp->exp_handle.h_cookie);
418 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
422 if (index == lov->desc.ld_tgt_count)
423 GOTO(out, index = -EINVAL);
425 if (lov->lov_tgts[index]->ltd_active == activate) {
426 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
427 activate ? "" : "in");
431 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
432 activate ? "" : "in");
434 lov->lov_tgts[index]->ltd_active = activate;
437 lov->desc.ld_active_tgt_count++;
438 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
440 lov->desc.ld_active_tgt_count--;
441 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
443 /* remove any old qos penalty */
444 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
451 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
452 enum obd_notify_event ev, void *data)
457 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
458 struct obd_uuid *uuid;
462 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
463 CERROR("unexpected notification of %s %s!\n",
464 watched->obd_type->typ_name,
468 uuid = &watched->u.cli.cl_target_uuid;
470 /* Set OSC as active before notifying the observer, so the
471 * observer can use the OSC normally.
473 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
475 CERROR("%sactivation of %s failed: %d\n",
476 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
477 obd_uuid2str(uuid), rc);
480 /* active event should be pass lov target index as data */
484 /* Pass the notification up the chain. */
486 rc = obd_notify_observer(obd, watched, ev, data);
488 /* NULL watched means all osc's in the lov (only for syncs) */
489 /* sync event should be send lov idx as data */
490 struct lov_obd *lov = &obd->u.lov;
491 struct obd_device *tgt_obd;
494 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
495 /* don't send sync event if target not
496 * connected/activated */
497 if (!lov->lov_tgts[i] ||
498 !lov->lov_tgts[i]->ltd_active)
501 if ((ev == OBD_NOTIFY_SYNC) ||
502 (ev == OBD_NOTIFY_SYNC_NONBLOCK))
505 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
507 rc = obd_notify_observer(obd, tgt_obd, ev, data);
509 CERROR("%s: notify %s of %s failed %d\n",
511 obd->obd_observer->obd_name,
512 tgt_obd->obd_name, rc);
522 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
523 __u32 index, int gen, int active)
525 struct lov_obd *lov = &obd->u.lov;
526 struct lov_tgt_desc *tgt;
530 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
531 uuidp->uuid, index, gen, active);
534 CERROR("request to add OBD %s with invalid generation: %d\n",
539 mutex_down(&lov->lov_lock);
541 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
542 tgt = lov->lov_tgts[index];
543 CERROR("UUID %s already assigned at LOV target index %d\n",
544 obd_uuid2str(&tgt->ltd_uuid), index);
545 mutex_up(&lov->lov_lock);
549 if (index >= lov->lov_tgt_size) {
550 /* We need to reallocate the lov target array. */
551 struct lov_tgt_desc **newtgts, **old = NULL;
552 __u32 newsize, oldsize = 0;
554 newsize = max(lov->lov_tgt_size, (__u32)2);
555 while (newsize < index + 1)
556 newsize = newsize << 1;
557 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
558 if (newtgts == NULL) {
559 mutex_up(&lov->lov_lock);
563 if (lov->lov_tgt_size) {
564 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
567 oldsize = lov->lov_tgt_size;
570 lov->lov_tgts = newtgts;
571 lov->lov_tgt_size = newsize;
576 OBD_FREE(old, sizeof(*old) * oldsize);
578 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
579 lov->lov_tgts, lov->lov_tgt_size);
584 mutex_up(&lov->lov_lock);
588 memset(tgt, 0, sizeof(*tgt));
589 tgt->ltd_uuid = *uuidp;
590 /* XXX - add a sanity check on the generation number. */
592 tgt->ltd_index = index;
593 tgt->ltd_activate = active;
594 lov->lov_tgts[index] = tgt;
595 if (index >= lov->desc.ld_tgt_count)
596 lov->desc.ld_tgt_count = index + 1;
598 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
602 mutex_up(&lov->lov_lock);
604 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
605 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
607 if (lov->lov_connects == 0) {
608 /* lov_connect hasn't been called yet. We'll do the
609 lov_connect_obd on this target when that fn first runs,
610 because we don't know the connect flags yet. */
616 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
620 /* connect to administrative disabled ost */
624 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
625 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
630 CERROR("add failed (%d), deleting %s\n", rc,
631 obd_uuid2str(&tgt->ltd_uuid));
632 lov_del_target(obd, index, 0, 0);
638 /* Schedule a target for deletion */
639 int lov_del_target(struct obd_device *obd, __u32 index,
640 struct obd_uuid *uuidp, int gen)
642 struct lov_obd *lov = &obd->u.lov;
643 int count = lov->desc.ld_tgt_count;
647 if (index >= count) {
648 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
655 if (!lov->lov_tgts[index]) {
656 CERROR("LOV target at index %d is not setup.\n", index);
657 GOTO(out, rc = -EINVAL);
660 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
661 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
662 lov_uuid2str(lov, index), index,
663 obd_uuid2str(uuidp));
664 GOTO(out, rc = -EINVAL);
667 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
668 lov_uuid2str(lov, index), index,
669 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
670 lov->lov_tgts[index]->ltd_active);
672 lov->lov_tgts[index]->ltd_reap = 1;
673 lov->lov_death_row++;
674 /* we really delete it from lov_putref */
681 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
683 struct obd_device *osc_obd;
686 LASSERT(tgt->ltd_reap);
688 osc_obd = class_exp2obd(tgt->ltd_exp);
690 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
692 osc_obd ? osc_obd->obd_name : "<no obd>");
695 lov_disconnect_obd(obd, tgt);
699 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
700 do it ourselves. And we can't do it from lov_cleanup,
701 because we just lost our only reference to it. */
703 class_manual_cleanup(osc_obd);
706 void lov_fix_desc_stripe_size(__u64 *val)
708 if (*val < PTLRPC_MAX_BRW_SIZE) {
709 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
710 PTLRPC_MAX_BRW_SIZE);
711 *val = PTLRPC_MAX_BRW_SIZE;
712 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
713 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
714 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
716 *val, LOV_MIN_STRIPE_SIZE);
720 void lov_fix_desc_stripe_count(__u32 *val)
726 void lov_fix_desc_pattern(__u32 *val)
728 /* from lov_setstripe */
729 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
730 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
735 void lov_fix_desc_qos_maxage(__u32 *val)
739 *val = QOS_DEFAULT_MAXAGE;
742 void lov_fix_desc(struct lov_desc *desc)
744 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
745 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
746 lov_fix_desc_pattern(&desc->ld_pattern);
747 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
750 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
752 struct lprocfs_static_vars lvars = { 0 };
753 struct lov_desc *desc;
754 struct lov_obd *lov = &obd->u.lov;
759 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
760 CERROR("LOV setup requires a descriptor\n");
764 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
766 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
767 CERROR("descriptor size wrong: %d > %d\n",
768 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
772 if (desc->ld_magic != LOV_DESC_MAGIC) {
773 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
774 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
775 obd->obd_name, desc);
776 lustre_swab_lov_desc(desc);
778 CERROR("%s: Bad lov desc magic: %#x\n",
779 obd->obd_name, desc->ld_magic);
786 /* Because of 64-bit divide/mod operations only work with a 32-bit
787 * divisor in a 32-bit kernel, we cannot support a stripe width
788 * of 4GB or larger on 32-bit CPUs. */
789 count = desc->ld_default_stripe_count;
790 if ((count > 0 ? count : desc->ld_tgt_count) *
791 desc->ld_default_stripe_size > 0xffffffff) {
792 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
793 desc->ld_default_stripe_size, count);
797 desc->ld_active_tgt_count = 0;
799 lov->lov_tgt_size = 0;
801 sema_init(&lov->lov_lock, 1);
802 atomic_set(&lov->lov_refcount, 0);
803 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
804 init_rwsem(&lov->lov_qos.lq_rw_sem);
805 lov->lov_sp_me = LUSTRE_SP_CLI;
806 lov->lov_qos.lq_dirty = 1;
807 lov->lov_qos.lq_rr.lqr_dirty = 1;
808 lov->lov_qos.lq_reset = 1;
809 /* Default priority is toward free space balance */
810 lov->lov_qos.lq_prio_free = 232;
811 /* Default threshold for rr (roughly 17%) */
812 lov->lov_qos.lq_threshold_rr = 43;
813 /* Init statfs fields */
814 OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
815 if (NULL == lov->lov_qos.lq_statfs_data)
817 cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
819 lov->lov_pools_hash_body = lustre_hash_init("POOLS", 7, 7,
820 &pool_hash_operations, 0);
821 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
822 lov->lov_pool_count = 0;
823 rc = lov_ost_pool_init(&lov->lov_packed, 0);
826 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
828 lov_ost_pool_free(&lov->lov_packed);
832 lprocfs_lov_init_vars(&lvars);
833 lprocfs_obd_setup(obd, lvars.obd_vars);
838 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
839 0444, &lov_proc_target_fops, obd);
841 CWARN("Error adding the target_obd file\n");
844 lov->lov_pool_proc_entry = lprocfs_register("pools",
851 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
854 struct lov_obd *lov = &obd->u.lov;
859 case OBD_CLEANUP_EARLY: {
861 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
862 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
864 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
869 case OBD_CLEANUP_EXPORTS:
870 rc = obd_llog_finish(obd, 0);
872 CERROR("failed to cleanup llogging subsystems\n");
878 static int lov_cleanup(struct obd_device *obd)
880 struct lov_obd *lov = &obd->u.lov;
881 struct list_head *pos, *tmp;
882 struct pool_desc *pool;
884 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
885 pool = list_entry(pos, struct pool_desc, pool_list);
886 /* free pool structs */
887 CDEBUG(D_INFO, "delete pool %p\n", pool);
888 lov_pool_del(obd, pool->pool_name);
890 lustre_hash_exit(lov->lov_pools_hash_body);
891 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
892 lov_ost_pool_free(&lov->lov_packed);
897 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
898 if (!lov->lov_tgts[i])
901 /* Inactive targets may never have connected */
902 if (lov->lov_tgts[i]->ltd_active ||
903 atomic_read(&lov->lov_refcount))
904 /* We should never get here - these
905 should have been removed in the
907 CERROR("lov tgt %d not cleaned!"
908 " deathrow=%d, lovrc=%d\n",
909 i, lov->lov_death_row,
910 atomic_read(&lov->lov_refcount));
911 lov_del_target(obd, i, 0, 0);
914 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
916 lov->lov_tgt_size = 0;
919 /* clear pools parent proc entry only after all pools is killed */
920 lprocfs_obd_cleanup(obd);
922 OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
926 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
927 __u32 *indexp, int *genp)
929 struct obd_uuid obd_uuid;
934 switch(cmd = lcfg->lcfg_command) {
935 case LCFG_LOV_ADD_OBD:
936 case LCFG_LOV_ADD_INA:
937 case LCFG_LOV_DEL_OBD: {
940 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
941 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
942 GOTO(out, rc = -EINVAL);
944 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
946 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
947 GOTO(out, rc = -EINVAL);
948 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
949 GOTO(out, rc = -EINVAL);
952 if (cmd == LCFG_LOV_ADD_OBD)
953 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
954 else if (cmd == LCFG_LOV_ADD_INA)
955 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
957 rc = lov_del_target(obd, index, &obd_uuid, gen);
961 struct lprocfs_static_vars lvars = { 0 };
962 struct lov_desc *desc = &(obd->u.lov.desc);
965 GOTO(out, rc = -EINVAL);
967 lprocfs_lov_init_vars(&lvars);
969 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
982 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
983 GOTO(out, rc = -EINVAL);
992 #define log2(n) ffz(~(n))
995 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
996 struct lov_stripe_md **ea,
997 struct obd_trans_info *oti)
1000 struct obdo *tmp_oa;
1001 struct obd_uuid *ost_uuid = NULL;
1005 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1006 src_oa->o_flags == OBD_FL_DELORPHAN);
1008 lov = &export->exp_obd->u.lov;
1014 if (oti->oti_ost_uuid) {
1015 ost_uuid = oti->oti_ost_uuid;
1016 CDEBUG(D_HA, "clearing orphans only for %s\n",
1020 lov_getref(export->exp_obd);
1021 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1022 struct lov_stripe_md obj_md;
1023 struct lov_stripe_md *obj_mdp = &obj_md;
1024 struct lov_tgt_desc *tgt;
1027 tgt = lov->lov_tgts[i];
1031 /* if called for a specific target, we don't
1032 care if it is not active. */
1033 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1034 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1038 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1041 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1042 obd_uuid2str(ost_uuid));
1044 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1046 LASSERT(lov->lov_tgts[i]->ltd_exp);
1047 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1048 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1049 tmp_oa, &obj_mdp, oti);
1051 /* This export will be disabled until it is recovered,
1052 and then orphan recovery will be completed. */
1053 CERROR("error in orphan recovery on OST idx %d/%d: "
1054 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1061 lov_putref(export->exp_obd);
1067 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1068 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1070 struct lov_stripe_md *obj_mdp, *lsm;
1071 struct lov_obd *lov = &exp->exp_obd->u.lov;
1076 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1077 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1079 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1080 if (obj_mdp == NULL)
1083 ost_idx = src_oa->o_nlink;
1086 GOTO(out, rc = -EINVAL);
1087 if (ost_idx >= lov->desc.ld_tgt_count ||
1088 !lov->lov_tgts[ost_idx])
1089 GOTO(out, rc = -EINVAL);
1091 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1092 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1093 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1094 GOTO(out, rc = -EINVAL);
1098 if (i == lsm->lsm_stripe_count)
1099 GOTO(out, rc = -EINVAL);
1101 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1103 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1107 /* the LOV expects oa->o_id to be set to the LOV object id */
1108 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1109 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1111 struct lov_obd *lov;
1112 struct obd_info oinfo;
1113 struct lov_request_set *set = NULL;
1114 struct lov_request *req;
1118 LASSERT(ea != NULL);
1122 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1123 src_oa->o_flags == OBD_FL_DELORPHAN) {
1124 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1128 lov = &exp->exp_obd->u.lov;
1129 if (!lov->desc.ld_active_tgt_count)
1132 lov_getref(exp->exp_obd);
1133 /* Recreate a specific object id at the given OST index */
1134 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1135 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1136 rc = lov_recreate(exp, src_oa, ea, oti);
1140 /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1141 * later in alloc_qos(), we will wait for those rpcs to complete if
1142 * the osfs age is older than 2 * qos_maxage */
1143 qos_statfs_update(exp->exp_obd,
1144 cfs_time_shift_64(-lov->desc.ld_qos_maxage) + HZ, 0);
1146 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1150 list_for_each_entry(req, &set->set_list, rq_link) {
1151 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1152 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1153 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1154 lov_update_create_set(set, req, rc);
1156 rc = lov_fini_create_set(set, ea);
1158 lov_putref(exp->exp_obd);
1162 #define ASSERT_LSM_MAGIC(lsmp) \
1164 LASSERT((lsmp) != NULL); \
1165 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1166 (lsmp)->lsm_magic == LOV_MAGIC_V3 || \
1167 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1168 (lsmp), (lsmp)->lsm_magic); \
1171 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1172 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1173 struct obd_export *md_exp, void *capa)
1175 struct lov_request_set *set;
1176 struct obd_info oinfo;
1177 struct lov_request *req;
1178 struct list_head *pos;
1179 struct lov_obd *lov;
1180 int rc = 0, err = 0;
1183 ASSERT_LSM_MAGIC(lsm);
1185 if (!exp || !exp->exp_obd)
1188 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1190 LASSERT(oti->oti_logcookies);
1193 lov = &exp->exp_obd->u.lov;
1194 lov_getref(exp->exp_obd);
1195 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1199 list_for_each (pos, &set->set_list) {
1200 req = list_entry(pos, struct lov_request, rq_link);
1202 if (oa->o_valid & OBD_MD_FLCOOKIE)
1203 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1205 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1206 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1207 err = lov_update_common_set(set, req, err);
1209 CERROR("error: destroying objid "LPX64" subobj "
1210 LPX64" on OST idx %d: rc = %d\n",
1211 oa->o_id, req->rq_oi.oi_oa->o_id,
1219 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1220 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1222 err = lov_fini_destroy_set(set);
1224 lov_putref(exp->exp_obd);
1225 RETURN(rc ? rc : err);
1228 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1230 struct lov_request_set *set;
1231 struct lov_request *req;
1232 struct list_head *pos;
1233 struct lov_obd *lov;
1234 int err = 0, rc = 0;
1238 ASSERT_LSM_MAGIC(oinfo->oi_md);
1240 if (!exp || !exp->exp_obd)
1243 lov = &exp->exp_obd->u.lov;
1245 rc = lov_prep_getattr_set(exp, oinfo, &set);
1249 list_for_each (pos, &set->set_list) {
1250 req = list_entry(pos, struct lov_request, rq_link);
1252 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1253 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1254 req->rq_oi.oi_oa->o_id, req->rq_idx);
1256 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1258 err = lov_update_common_set(set, req, rc);
1260 CERROR("error: getattr objid "LPX64" subobj "
1261 LPX64" on OST idx %d: rc = %d\n",
1262 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1268 rc = lov_fini_getattr_set(set);
1274 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1277 struct lov_request_set *lovset = (struct lov_request_set *)data;
1281 /* don't do attribute merge if this aysnc op failed */
1283 lovset->set_completes = 0;
1284 err = lov_fini_getattr_set(lovset);
1285 RETURN(rc ? rc : err);
1288 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1289 struct ptlrpc_request_set *rqset)
1291 struct lov_request_set *lovset;
1292 struct lov_obd *lov;
1293 struct list_head *pos;
1294 struct lov_request *req;
1299 ASSERT_LSM_MAGIC(oinfo->oi_md);
1301 if (!exp || !exp->exp_obd)
1304 lov = &exp->exp_obd->u.lov;
1306 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1310 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1311 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1312 oinfo->oi_md->lsm_stripe_size);
1314 list_for_each (pos, &lovset->set_list) {
1315 req = list_entry(pos, struct lov_request, rq_link);
1317 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1318 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1319 req->rq_oi.oi_oa->o_id, req->rq_idx);
1320 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1321 &req->rq_oi, rqset);
1323 CERROR("error: getattr objid "LPX64" subobj "
1324 LPX64" on OST idx %d: rc = %d\n",
1325 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1331 if (!list_empty(&rqset->set_requests)) {
1333 LASSERT (rqset->set_interpret == NULL);
1334 rqset->set_interpret = lov_getattr_interpret;
1335 rqset->set_arg = (void *)lovset;
1340 lovset->set_completes = 0;
1341 err = lov_fini_getattr_set(lovset);
1342 RETURN(rc ? rc : err);
1345 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1346 struct obd_trans_info *oti)
1348 struct lov_request_set *set;
1349 struct lov_obd *lov;
1350 struct list_head *pos;
1351 struct lov_request *req;
1352 int err = 0, rc = 0;
1356 ASSERT_LSM_MAGIC(oinfo->oi_md);
1358 if (!exp || !exp->exp_obd)
1361 /* for now, we only expect the following updates here */
1362 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1363 OBD_MD_FLMODE | OBD_MD_FLATIME |
1364 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1365 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1366 OBD_MD_FLGROUP | OBD_MD_FLUID |
1367 OBD_MD_FLGID | OBD_MD_FLFID |
1369 lov = &exp->exp_obd->u.lov;
1370 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1374 list_for_each (pos, &set->set_list) {
1375 req = list_entry(pos, struct lov_request, rq_link);
1377 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1379 err = lov_update_setattr_set(set, req, rc);
1381 CERROR("error: setattr objid "LPX64" subobj "
1382 LPX64" on OST idx %d: rc = %d\n",
1383 set->set_oi->oi_oa->o_id,
1384 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1389 err = lov_fini_setattr_set(set);
1395 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1398 struct lov_request_set *lovset = (struct lov_request_set *)data;
1403 lovset->set_completes = 0;
1404 err = lov_fini_setattr_set(lovset);
1405 RETURN(rc ? rc : err);
1408 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1409 needed. Otherwise, a client is waiting for responses. */
1410 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1411 struct obd_trans_info *oti,
1412 struct ptlrpc_request_set *rqset)
1414 struct lov_request_set *set;
1415 struct lov_request *req;
1416 struct list_head *pos;
1417 struct lov_obd *lov;
1422 ASSERT_LSM_MAGIC(oinfo->oi_md);
1423 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1425 LASSERT(oti->oti_logcookies);
1428 if (!exp || !exp->exp_obd)
1431 lov = &exp->exp_obd->u.lov;
1432 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1436 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1437 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1438 oinfo->oi_md->lsm_stripe_size);
1440 list_for_each (pos, &set->set_list) {
1441 req = list_entry(pos, struct lov_request, rq_link);
1443 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1444 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1446 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1447 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1448 req->rq_oi.oi_oa->o_id, req->rq_idx);
1450 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1451 &req->rq_oi, oti, rqset);
1453 CERROR("error: setattr objid "LPX64" subobj "
1454 LPX64" on OST idx %d: rc = %d\n",
1455 set->set_oi->oi_oa->o_id,
1456 req->rq_oi.oi_oa->o_id,
1462 /* If we are not waiting for responses on async requests, return. */
1463 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1466 set->set_completes = 0;
1467 err = lov_fini_setattr_set(set);
1468 RETURN(rc ? rc : err);
1471 LASSERT(rqset->set_interpret == NULL);
1472 rqset->set_interpret = lov_setattr_interpret;
1473 rqset->set_arg = (void *)set;
1478 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1481 struct lov_request_set *lovset = (struct lov_request_set *)data;
1486 lovset->set_completes = 0;
1487 err = lov_fini_punch_set(lovset);
1488 RETURN(rc ? rc : err);
1491 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1492 * we can send this 'punch' to just the authoritative node and the nodes
1493 * that the punch will affect. */
1494 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1495 struct obd_trans_info *oti,
1496 struct ptlrpc_request_set *rqset)
1498 struct lov_request_set *set;
1499 struct lov_obd *lov;
1500 struct list_head *pos;
1501 struct lov_request *req;
1506 ASSERT_LSM_MAGIC(oinfo->oi_md);
1508 if (!exp || !exp->exp_obd)
1511 lov = &exp->exp_obd->u.lov;
1512 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1516 list_for_each (pos, &set->set_list) {
1517 req = list_entry(pos, struct lov_request, rq_link);
1519 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1520 &req->rq_oi, NULL, rqset);
1522 CERROR("error: punch objid "LPX64" subobj "LPX64
1523 " on OST idx %d: rc = %d\n",
1524 set->set_oi->oi_oa->o_id,
1525 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1530 if (rc || list_empty(&rqset->set_requests)) {
1532 err = lov_fini_punch_set(set);
1533 RETURN(rc ? rc : err);
1536 LASSERT(rqset->set_interpret == NULL);
1537 rqset->set_interpret = lov_punch_interpret;
1538 rqset->set_arg = (void *)set;
1543 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1544 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1547 struct lov_request_set *set;
1548 struct obd_info oinfo;
1549 struct lov_obd *lov;
1550 struct list_head *pos;
1551 struct lov_request *req;
1552 int err = 0, rc = 0;
1555 ASSERT_LSM_MAGIC(lsm);
1560 lov = &exp->exp_obd->u.lov;
1561 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1565 list_for_each (pos, &set->set_list) {
1566 req = list_entry(pos, struct lov_request, rq_link);
1568 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1569 req->rq_oi.oi_oa, NULL,
1570 req->rq_oi.oi_policy.l_extent.start,
1571 req->rq_oi.oi_policy.l_extent.end, capa);
1572 err = lov_update_common_set(set, req, rc);
1574 CERROR("error: fsync objid "LPX64" subobj "LPX64
1575 " on OST idx %d: rc = %d\n",
1576 set->set_oi->oi_oa->o_id,
1577 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1582 err = lov_fini_sync_set(set);
1588 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1589 obd_count oa_bufs, struct brw_page *pga)
1591 struct obd_info oinfo = { { { 0 } } };
1594 oinfo.oi_oa = lov_oinfo->oi_oa;
1596 /* The caller just wants to know if there's a chance that this
1597 * I/O can succeed */
1598 for (i = 0; i < oa_bufs; i++) {
1599 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1600 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1603 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1604 pga[i].off + pga[i].count - 1,
1608 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1609 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1613 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1621 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1622 obd_count oa_bufs, struct brw_page *pga,
1623 struct obd_trans_info *oti)
1625 struct lov_request_set *set;
1626 struct lov_request *req;
1627 struct list_head *pos;
1628 struct lov_obd *lov = &exp->exp_obd->u.lov;
1632 ASSERT_LSM_MAGIC(oinfo->oi_md);
1634 if (cmd == OBD_BRW_CHECK) {
1635 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1639 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1643 list_for_each (pos, &set->set_list) {
1644 struct obd_export *sub_exp;
1645 struct brw_page *sub_pga;
1646 req = list_entry(pos, struct lov_request, rq_link);
1648 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1649 sub_pga = set->set_pga + req->rq_pgaidx;
1650 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1654 lov_update_common_set(set, req, rc);
1657 err = lov_fini_brw_set(set);
1663 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1666 struct lov_request_set *lovset = (struct lov_request_set *)data;
1668 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1672 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1673 struct ldlm_enqueue_info *einfo,
1674 struct ptlrpc_request_set *rqset)
1676 ldlm_mode_t mode = einfo->ei_mode;
1677 struct lov_request_set *set;
1678 struct lov_request *req;
1679 struct list_head *pos;
1680 struct lov_obd *lov;
1685 ASSERT_LSM_MAGIC(oinfo->oi_md);
1686 LASSERT(mode == (mode & -mode));
1688 /* we should never be asked to replay a lock this way. */
1689 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1691 if (!exp || !exp->exp_obd)
1694 lov = &exp->exp_obd->u.lov;
1695 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1699 list_for_each (pos, &set->set_list) {
1700 req = list_entry(pos, struct lov_request, rq_link);
1702 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1703 &req->rq_oi, einfo, rqset);
1708 if (rqset && !list_empty(&rqset->set_requests)) {
1710 LASSERT(rqset->set_interpret == NULL);
1711 rqset->set_interpret = lov_enqueue_interpret;
1712 rqset->set_arg = (void *)set;
1716 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1720 static int lov_change_cbdata(struct obd_export *exp,
1721 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1724 struct lov_obd *lov;
1728 ASSERT_LSM_MAGIC(lsm);
1730 if (!exp || !exp->exp_obd)
1733 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1735 lov = &exp->exp_obd->u.lov;
1736 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1737 struct lov_stripe_md submd;
1738 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1740 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1741 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1745 submd.lsm_object_id = loi->loi_id;
1746 submd.lsm_object_gr = lsm->lsm_object_gr;
1747 submd.lsm_stripe_count = 0;
1748 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1754 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1755 __u32 mode, struct lustre_handle *lockh)
1757 struct lov_request_set *set;
1758 struct obd_info oinfo;
1759 struct lov_request *req;
1760 struct list_head *pos;
1761 struct lov_obd *lov;
1762 struct lustre_handle *lov_lockhp;
1763 int err = 0, rc = 0;
1766 ASSERT_LSM_MAGIC(lsm);
1768 if (!exp || !exp->exp_obd)
1771 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1773 lov = &exp->exp_obd->u.lov;
1774 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1778 list_for_each (pos, &set->set_list) {
1779 req = list_entry(pos, struct lov_request, rq_link);
1780 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1782 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1783 req->rq_oi.oi_md, mode, lov_lockhp);
1784 rc = lov_update_common_set(set, req, rc);
1786 CERROR("error: cancel objid "LPX64" subobj "
1787 LPX64" on OST idx %d: rc = %d\n",
1789 req->rq_oi.oi_md->lsm_object_id,
1795 lov_fini_cancel_set(set);
1799 static int lov_cancel_unused(struct obd_export *exp,
1800 struct lov_stripe_md *lsm,
1801 int flags, void *opaque)
1803 struct lov_obd *lov;
1807 if (!exp || !exp->exp_obd)
1810 lov = &exp->exp_obd->u.lov;
1812 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1814 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1817 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1825 ASSERT_LSM_MAGIC(lsm);
1827 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1828 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1829 struct lov_stripe_md submd;
1830 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1833 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1834 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1838 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1839 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1841 submd.lsm_object_id = loi->loi_id;
1842 submd.lsm_object_gr = lsm->lsm_object_gr;
1843 submd.lsm_stripe_count = 0;
1844 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1845 &submd, flags, opaque);
1846 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1847 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1848 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1849 loi->loi_id, loi->loi_ost_idx, err);
1857 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1859 struct lov_request_set *lovset = (struct lov_request_set *)data;
1864 lovset->set_completes = 0;
1866 err = lov_fini_statfs_set(lovset);
1867 RETURN(rc ? rc : err);
1870 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1871 __u64 max_age, struct ptlrpc_request_set *rqset)
1873 struct lov_request_set *set;
1874 struct lov_request *req;
1875 struct list_head *pos;
1876 struct lov_obd *lov;
1880 LASSERT(oinfo != NULL);
1881 LASSERT(oinfo->oi_osfs != NULL);
1884 rc = lov_prep_statfs_set(obd, oinfo, &set);
1888 list_for_each (pos, &set->set_list) {
1889 struct obd_device *osc_obd;
1891 req = list_entry(pos, struct lov_request, rq_link);
1893 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1894 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1899 if (rc || list_empty(&rqset->set_requests)) {
1902 set->set_completes = 0;
1903 err = lov_fini_statfs_set(set);
1904 RETURN(rc ? rc : err);
1907 LASSERT(rqset->set_interpret == NULL);
1908 rqset->set_interpret = lov_statfs_interpret;
1909 rqset->set_arg = (void *)set;
1913 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1914 __u64 max_age, __u32 flags)
1916 struct ptlrpc_request_set *set = NULL;
1917 struct obd_info oinfo = { { { 0 } } };
1922 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1923 * statfs requests */
1924 set = ptlrpc_prep_set();
1928 oinfo.oi_osfs = osfs;
1929 oinfo.oi_flags = flags;
1930 rc = lov_statfs_async(obd, &oinfo, max_age, set);
1932 rc = ptlrpc_set_wait(set);
1933 ptlrpc_set_destroy(set);
1938 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1939 void *karg, void *uarg)
1941 struct obd_device *obddev = class_exp2obd(exp);
1942 struct lov_obd *lov = &obddev->u.lov;
1943 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1944 struct obd_uuid *uuidp;
1948 case IOC_OBD_STATFS: {
1949 struct obd_ioctl_data *data = karg;
1950 struct obd_device *osc_obd;
1951 struct obd_statfs stat_buf = {0};
1954 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1955 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1957 if ((index >= count))
1960 if (!lov->lov_tgts[index])
1961 /* Try again with the next index */
1963 if (!lov->lov_tgts[index]->ltd_active)
1966 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1970 /* got statfs data */
1971 rc = obd_statfs(osc_obd, &stat_buf,
1972 cfs_time_current_64() - HZ, 0);
1975 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1978 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1983 case OBD_IOC_LOV_GET_CONFIG: {
1984 struct obd_ioctl_data *data;
1985 struct lov_desc *desc;
1990 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1993 data = (struct obd_ioctl_data *)buf;
1995 if (sizeof(*desc) > data->ioc_inllen1) {
1996 obd_ioctl_freedata(buf, len);
2000 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2001 obd_ioctl_freedata(buf, len);
2005 if (sizeof(__u32) * count > data->ioc_inllen3) {
2006 obd_ioctl_freedata(buf, len);
2010 desc = (struct lov_desc *)data->ioc_inlbuf1;
2011 memcpy(desc, &(lov->desc), sizeof(*desc));
2013 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2014 genp = (__u32 *)data->ioc_inlbuf3;
2015 /* the uuid will be empty for deleted OSTs */
2016 for (i = 0; i < count; i++, uuidp++, genp++) {
2017 if (!lov->lov_tgts[i])
2019 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2020 *genp = lov->lov_tgts[i]->ltd_gen;
2023 if (copy_to_user((void *)uarg, buf, len))
2025 obd_ioctl_freedata(buf, len);
2028 case LL_IOC_LOV_SETSTRIPE:
2029 rc = lov_setstripe(exp, karg, uarg);
2031 case LL_IOC_LOV_GETSTRIPE:
2032 rc = lov_getstripe(exp, karg, uarg);
2034 case LL_IOC_LOV_SETEA:
2035 rc = lov_setea(exp, karg, uarg);
2037 case OBD_IOC_QUOTACTL: {
2038 struct if_quotactl *qctl = karg;
2039 struct lov_tgt_desc *tgt = NULL;
2040 struct obd_quotactl *oqctl;
2042 if (qctl->qc_valid == QC_OSTIDX) {
2043 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2046 tgt = lov->lov_tgts[qctl->qc_idx];
2047 if (!tgt || !tgt->ltd_exp)
2049 } else if (qctl->qc_valid == QC_UUID) {
2050 for (i = 0; i < count; i++) {
2051 tgt = lov->lov_tgts[i];
2053 !obd_uuid_equals(&tgt->ltd_uuid,
2057 if (tgt->ltd_exp == NULL)
2069 LASSERT(tgt && tgt->ltd_exp);
2070 OBD_ALLOC_PTR(oqctl);
2074 QCTL_COPY(oqctl, qctl);
2075 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2077 QCTL_COPY(qctl, oqctl);
2078 qctl->qc_valid = QC_OSTIDX;
2079 qctl->obd_uuid = tgt->ltd_uuid;
2081 OBD_FREE_PTR(oqctl);
2090 for (i = 0; i < count; i++) {
2093 /* OST was disconnected */
2094 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2097 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2099 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2102 if (lov->lov_tgts[i]->ltd_active) {
2103 CDEBUG(err == -ENOTTY ?
2104 D_IOCTL : D_WARNING,
2105 "iocontrol OSC %s on OST "
2106 "idx %d cmd %x: err = %d\n",
2107 lov_uuid2str(lov, i),
2124 #define FIEMAP_BUFFER_SIZE 4096
2127 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2128 * call. The local end offset and the device are sent in the first
2129 * fm_extent. This function calculates the stripe number from the index.
2130 * This function returns a stripe_no on which mapping is to be restarted.
2132 * This function returns fm_end_offset which is the in-OST offset at which
2133 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2134 * will re-calculate proper offset in next stripe.
2135 * Note that the first extent is passed to lov_get_info via the value field.
2137 * \param fiemap fiemap request header
2138 * \param lsm striping information for the file
2139 * \param fm_start logical start of mapping
2140 * \param fm_end logical end of mapping
2141 * \param start_stripe starting stripe will be returned in this
2143 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2144 struct lov_stripe_md *lsm, obd_size fm_start,
2145 obd_size fm_end, int *start_stripe)
2147 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2148 obd_off lun_start, lun_end;
2149 obd_size fm_end_offset;
2150 int stripe_no = -1, i;
2152 if (fiemap->fm_extent_count == 0 ||
2153 fiemap->fm_extents[0].fe_logical == 0)
2156 /* Find out stripe_no from ost_index saved in the fe_device */
2157 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2158 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2159 fiemap->fm_extents[0].fe_device) {
2165 /* If we have finished mapping on previous device, shift logical
2166 * offset to start of next device */
2167 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2168 &lun_start, &lun_end)) != 0 &&
2169 local_end < lun_end) {
2170 fm_end_offset = local_end;
2171 *start_stripe = stripe_no;
2173 /* This is a special value to indicate that caller should
2174 * calculate offset in next stripe. */
2176 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2179 return fm_end_offset;
2183 * We calculate on which OST the mapping will end. If the length of mapping
2184 * is greater than (stripe_size * stripe_count) then the last_stripe will
2185 * will be one just before start_stripe. Else we check if the mapping
2186 * intersects each OST and find last_stripe.
2187 * This function returns the last_stripe and also sets the stripe_count
2188 * over which the mapping is spread
2190 * \param lsm striping information for the file
2191 * \param fm_start logical start of mapping
2192 * \param fm_end logical end of mapping
2193 * \param start_stripe starting stripe of the mapping
2194 * \param stripe_count the number of stripes across which to map is returned
2196 * \retval last_stripe return the last stripe of the mapping
2198 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2199 obd_size fm_end, int start_stripe,
2203 obd_off obd_start, obd_end;
2206 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2207 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2209 *stripe_count = lsm->lsm_stripe_count;
2211 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2212 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2213 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2214 &obd_start, &obd_end)) == 0)
2218 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2225 * Set fe_device and copy extents from local buffer into main return buffer.
2227 * \param fiemap fiemap request header
2228 * \param lcl_fm_ext array of local fiemap extents to be copied
2229 * \param ost_index OST index to be written into the fm_device field for each
2231 * \param ext_count number of extents to be copied
2232 * \param current_extent where to start copying in main extent array
2234 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2235 struct ll_fiemap_extent *lcl_fm_ext,
2236 int ost_index, unsigned int ext_count,
2242 for (ext = 0; ext < ext_count; ext++) {
2243 lcl_fm_ext[ext].fe_device = ost_index;
2244 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2247 /* Copy fm_extent's from fm_local to return buffer */
2248 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2249 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2253 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2254 * This also handles the restarting of FIEMAP calls in case mapping overflows
2255 * the available number of extents in single call.
2257 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2258 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2260 struct ll_fiemap_info_key *fm_key = key;
2261 struct ll_user_fiemap *fiemap = val;
2262 struct ll_user_fiemap *fm_local = NULL;
2263 struct ll_fiemap_extent *lcl_fm_ext;
2265 unsigned int get_num_extents = 0;
2266 int ost_index = 0, actual_start_stripe, start_stripe;
2267 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2269 int current_extent = 0, rc = 0, i;
2270 int ost_eof = 0; /* EOF for object */
2271 int ost_done = 0; /* done with required mapping for this OST? */
2273 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2274 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2279 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2280 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2282 OBD_ALLOC(fm_local, buffer_size);
2283 if (fm_local == NULL)
2284 GOTO(out, rc = -ENOMEM);
2285 lcl_fm_ext = &fm_local->fm_extents[0];
2287 count_local = fiemap_size_to_count(buffer_size);
2289 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2290 fm_start = fiemap->fm_start;
2291 fm_length = fiemap->fm_length;
2292 /* Calculate start stripe, last stripe and length of mapping */
2293 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2294 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2295 fm_start + fm_length - 1);
2296 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2297 if (fm_end > fm_key->oa.o_size)
2298 fm_end = fm_key->oa.o_size;
2300 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2301 actual_start_stripe, &stripe_count);
2303 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2306 if (fiemap->fm_extent_count == 0) {
2307 get_num_extents = 1;
2311 /* Check each stripe */
2312 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2313 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2314 obd_size req_fm_len; /* Stores length of required mapping */
2315 obd_size len_mapped_single_call;
2316 obd_off lun_start, lun_end, obd_object_end;
2317 unsigned int ext_count;
2319 cur_stripe_wrap = cur_stripe;
2321 /* Find out range of mapping on this stripe */
2322 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2323 &lun_start, &obd_object_end)) == 0)
2326 /* If this is a continuation FIEMAP call and we are on
2327 * starting stripe then lun_start needs to be set to
2329 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2330 lun_start = fm_end_offset;
2332 if (fm_length != ~0ULL) {
2333 /* Handle fm_start + fm_length overflow */
2334 if (fm_start + fm_length < fm_start)
2335 fm_length = ~0ULL - fm_start;
2336 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2342 if (lun_start == lun_end)
2345 req_fm_len = obd_object_end - lun_start;
2346 fm_local->fm_length = 0;
2347 len_mapped_single_call = 0;
2349 /* If the output buffer is very large and the objects have many
2350 * extents we may need to loop on a single OST repeatedly */
2354 if (get_num_extents == 0) {
2355 /* Don't get too many extents. */
2356 if (current_extent + count_local >
2357 fiemap->fm_extent_count)
2358 count_local = fiemap->fm_extent_count -
2362 lun_start += len_mapped_single_call;
2363 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2364 req_fm_len = fm_local->fm_length;
2365 fm_local->fm_extent_count = count_local;
2366 fm_local->fm_mapped_extents = 0;
2367 fm_local->fm_flags = fiemap->fm_flags;
2369 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2370 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2372 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2373 GOTO(out, rc = -EINVAL);
2375 /* If OST is inactive, return extent with UNKNOWN flag */
2376 if (!lov->lov_tgts[ost_index]->ltd_active) {
2377 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2378 fm_local->fm_mapped_extents = 1;
2380 lcl_fm_ext[0].fe_logical = lun_start;
2381 lcl_fm_ext[0].fe_length = obd_object_end -
2383 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2388 fm_local->fm_start = lun_start;
2389 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2390 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2391 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2392 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2393 keylen, key, vallen, fm_local, lsm);
2398 ext_count = fm_local->fm_mapped_extents;
2399 if (ext_count == 0) {
2401 /* If last stripe has hole at the end,
2402 * then we need to return */
2403 if (cur_stripe_wrap == last_stripe) {
2404 fiemap->fm_mapped_extents = 0;
2410 /* If we just need num of extents then go to next device */
2411 if (get_num_extents) {
2412 current_extent += ext_count;
2416 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2417 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2419 /* Have we finished mapping on this device? */
2420 if (req_fm_len <= len_mapped_single_call)
2423 /* Clear the EXTENT_LAST flag which can be present on
2425 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2426 lcl_fm_ext[ext_count - 1].fe_flags &=
2427 ~FIEMAP_EXTENT_LAST;
2429 curr_loc = lov_stripe_size(lsm,
2430 lcl_fm_ext[ext_count - 1].fe_logical+
2431 lcl_fm_ext[ext_count - 1].fe_length,
2433 if (curr_loc >= fm_key->oa.o_size)
2436 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2437 ost_index, ext_count,
2440 current_extent += ext_count;
2442 /* Ran out of available extents? */
2443 if (current_extent >= fiemap->fm_extent_count)
2445 } while (ost_done == 0 && ost_eof == 0);
2447 if (cur_stripe_wrap == last_stripe)
2452 /* Indicate that we are returning device offsets unless file just has
2454 if (lsm->lsm_stripe_count > 1)
2455 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2457 if (get_num_extents)
2458 goto skip_last_device_calc;
2460 /* Check if we have reached the last stripe and whether mapping for that
2461 * stripe is done. */
2462 if (cur_stripe_wrap == last_stripe) {
2463 if (ost_done || ost_eof)
2464 fiemap->fm_extents[current_extent - 1].fe_flags |=
2468 skip_last_device_calc:
2469 fiemap->fm_mapped_extents = current_extent;
2472 OBD_FREE(fm_local, buffer_size);
2476 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2477 void *key, __u32 *vallen, void *val,
2478 struct lov_stripe_md *lsm)
2480 struct obd_device *obddev = class_exp2obd(exp);
2481 struct lov_obd *lov = &obddev->u.lov;
2485 if (!vallen || !val)
2490 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2493 struct ldlm_lock *lock;
2495 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2496 struct lov_oinfo *loi;
2497 __u32 *stripe = val;
2499 if (*vallen < sizeof(*stripe))
2500 GOTO(out, rc = -EFAULT);
2501 *vallen = sizeof(*stripe);
2503 /* XXX This is another one of those bits that will need to
2504 * change if we ever actually support nested LOVs. It uses
2505 * the lock's export to find out which stripe it is. */
2506 /* XXX - it's assumed all the locks for deleted OSTs have
2507 * been cancelled. Also, the export for deleted OSTs will
2508 * be NULL and won't match the lock's export. */
2509 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2510 loi = lsm->lsm_oinfo[i];
2511 if (!lov->lov_tgts[loi->loi_ost_idx])
2513 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2514 data->lock->l_conn_export &&
2515 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2520 LDLM_ERROR(data->lock, "lock on inode without such object");
2521 dump_lsm(D_ERROR, lsm);
2522 GOTO(out, rc = -ENXIO);
2523 } else if (KEY_IS(KEY_LAST_ID)) {
2524 struct obd_id_info *info = val;
2525 __u32 size = sizeof(obd_id);
2526 struct lov_tgt_desc *tgt;
2528 LASSERT(*vallen == sizeof(struct obd_id_info));
2529 tgt = lov->lov_tgts[info->idx];
2531 if (!tgt || !tgt->ltd_active)
2532 GOTO(out, rc = -ESRCH);
2534 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2536 } else if (KEY_IS(KEY_LOVDESC)) {
2537 struct lov_desc *desc_ret = val;
2538 *desc_ret = lov->desc;
2541 } else if (KEY_IS(KEY_FIEMAP)) {
2542 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2553 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2554 void *key, obd_count vallen, void *val,
2555 struct ptlrpc_request_set *set)
2557 struct obd_device *obddev = class_exp2obd(exp);
2558 struct lov_obd *lov = &obddev->u.lov;
2561 struct lov_tgt_desc *tgt;
2562 unsigned incr, check_uuid,
2563 do_inactive, no_set;
2564 unsigned next_id = 0, mds_con = 0;
2567 incr = check_uuid = do_inactive = no_set = 0;
2570 set = ptlrpc_prep_set();
2576 count = lov->desc.ld_tgt_count;
2578 if (KEY_IS(KEY_NEXT_ID)) {
2579 count = vallen / sizeof(struct obd_id_info);
2580 vallen = sizeof(obd_id);
2581 incr = sizeof(struct obd_id_info);
2584 } else if (KEY_IS(KEY_CHECKSUM)) {
2586 } else if (KEY_IS(KEY_UNLINKED)) {
2587 check_uuid = val ? 1 : 0;
2588 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2589 /* use defaults: do_inactive = incr = 0; */
2590 } else if (KEY_IS(KEY_MDS_CONN)) {
2594 for (i = 0; i < count; i++, val = (char *)val + incr) {
2596 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2598 tgt = lov->lov_tgts[i];
2600 /* OST was disconnected */
2601 if (!tgt || !tgt->ltd_exp)
2604 /* OST is inactive and we don't want inactive OSCs */
2605 if (!tgt->ltd_active && !do_inactive)
2609 struct mds_group_info *mgi;
2611 LASSERT(vallen == sizeof(*mgi));
2612 mgi = (struct mds_group_info *)val;
2614 /* Only want a specific OSC */
2615 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2619 err = obd_set_info_async(tgt->ltd_exp,
2620 keylen, key, sizeof(int),
2622 } else if (next_id) {
2623 err = obd_set_info_async(tgt->ltd_exp,
2624 keylen, key, vallen,
2625 ((struct obd_id_info*)val)->data, set);
2627 /* Only want a specific OSC */
2629 !obd_uuid_equals(val, &tgt->ltd_uuid))
2632 err = obd_set_info_async(tgt->ltd_exp,
2633 keylen, key, vallen, val, set);
2642 err = ptlrpc_set_wait(set);
2645 ptlrpc_set_destroy(set);
2650 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2651 struct lov_stripe_md *lsm)
2659 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2660 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2665 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2670 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2671 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2672 if (loi->loi_ar.ar_rc && !rc)
2673 rc = loi->loi_ar.ar_rc;
2674 loi->loi_ar.ar_rc = 0;
2678 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2681 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2682 int cmd, __u64 *offset)
2684 __u32 ssize = lsm->lsm_stripe_size;
2688 do_div(start, ssize);
2689 start = start * ssize;
2691 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2692 ", end "LPU64"\n", *offset, ssize, start,
2694 if (cmd == OBD_CALC_STRIPE_END) {
2695 *offset = start + ssize - 1;
2696 } else if (cmd == OBD_CALC_STRIPE_START) {
2705 void lov_stripe_lock(struct lov_stripe_md *md)
2707 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2708 spin_lock(&md->lsm_lock);
2709 LASSERT(md->lsm_lock_owner == 0);
2710 md->lsm_lock_owner = cfs_curproc_pid();
2712 EXPORT_SYMBOL(lov_stripe_lock);
2714 void lov_stripe_unlock(struct lov_stripe_md *md)
2716 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2717 md->lsm_lock_owner = 0;
2718 spin_unlock(&md->lsm_lock);
2720 EXPORT_SYMBOL(lov_stripe_unlock);
2723 struct obd_ops lov_obd_ops = {
2724 .o_owner = THIS_MODULE,
2725 .o_setup = lov_setup,
2726 .o_precleanup = lov_precleanup,
2727 .o_cleanup = lov_cleanup,
2728 //.o_process_config = lov_process_config,
2729 .o_connect = lov_connect,
2730 .o_disconnect = lov_disconnect,
2731 .o_statfs = lov_statfs,
2732 .o_statfs_async = lov_statfs_async,
2733 .o_packmd = lov_packmd,
2734 .o_unpackmd = lov_unpackmd,
2735 .o_checkmd = lov_checkmd,
2736 .o_create = lov_create,
2737 .o_destroy = lov_destroy,
2738 .o_getattr = lov_getattr,
2739 .o_getattr_async = lov_getattr_async,
2740 .o_setattr = lov_setattr,
2741 .o_setattr_async = lov_setattr_async,
2743 .o_merge_lvb = lov_merge_lvb,
2744 .o_adjust_kms = lov_adjust_kms,
2745 .o_punch = lov_punch,
2747 .o_enqueue = lov_enqueue,
2748 .o_change_cbdata = lov_change_cbdata,
2749 .o_cancel = lov_cancel,
2750 .o_cancel_unused = lov_cancel_unused,
2751 .o_iocontrol = lov_iocontrol,
2752 .o_get_info = lov_get_info,
2753 .o_set_info_async = lov_set_info_async,
2754 .o_extent_calc = lov_extent_calc,
2755 .o_llog_init = lov_llog_init,
2756 .o_llog_finish = lov_llog_finish,
2757 .o_notify = lov_notify,
2758 .o_pool_new = lov_pool_new,
2759 .o_pool_rem = lov_pool_remove,
2760 .o_pool_add = lov_pool_add,
2761 .o_pool_del = lov_pool_del,
2764 static quota_interface_t *quota_interface;
2765 extern quota_interface_t lov_quota_interface;
2767 cfs_mem_cache_t *lov_oinfo_slab;
2769 extern struct lu_kmem_descr lov_caches[];
2771 int __init lov_init(void)
2773 struct lprocfs_static_vars lvars = { 0 };
2777 /* print an address of _any_ initialized kernel symbol from this
2778 * module, to allow debugging with gdb that doesn't support data
2779 * symbols from modules.*/
2780 CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2782 rc = lu_kmem_init(lov_caches);
2786 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2787 sizeof(struct lov_oinfo),
2788 0, SLAB_HWCACHE_ALIGN);
2789 if (lov_oinfo_slab == NULL) {
2790 lu_kmem_fini(lov_caches);
2793 lprocfs_lov_init_vars(&lvars);
2795 request_module("lquota");
2796 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2797 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2799 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2800 LUSTRE_LOV_NAME, &lov_device_type);
2803 if (quota_interface)
2804 PORTAL_SYMBOL_PUT(lov_quota_interface);
2805 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2807 lu_kmem_fini(lov_caches);
2814 static void /*__exit*/ lov_exit(void)
2818 lu_device_type_fini(&lov_device_type);
2819 lu_kmem_fini(lov_caches);
2821 if (quota_interface)
2822 PORTAL_SYMBOL_PUT(lov_quota_interface);
2824 class_unregister_type(LUSTRE_LOV_NAME);
2825 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2829 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2830 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2831 MODULE_LICENSE("GPL");
2833 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);