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, __u32 index);
87 void lov_putref(struct obd_device *obd)
89 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) {
94 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
96 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
97 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap)
99 /* Disconnect and delete from list */
100 __lov_del_obd(obd, i);
101 lov->lov_death_row--;
104 mutex_up(&lov->lov_lock);
107 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
109 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
110 enum obd_notify_event ev, void *data);
113 #define MAX_STRING_SIZE 128
114 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
115 struct obd_connect_data *data)
117 struct lov_obd *lov = &obd->u.lov;
118 struct obd_uuid tgt_uuid;
119 struct obd_device *tgt_obd;
120 struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
121 struct lustre_handle conn = {0, };
122 struct obd_import *imp;
125 cfs_proc_dir_entry_t *lov_proc_dir;
130 if (!lov->lov_tgts[index])
133 tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
135 tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
139 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
142 if (!tgt_obd->obd_set_up) {
143 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
147 /* override the sp_me from lov */
148 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
150 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
151 data->ocd_index = index;
154 * Divine LOV knows that OBDs under it are OSCs.
156 imp = tgt_obd->u.cli.cl_import;
159 tgt_obd->obd_no_recov = 0;
160 /* FIXME this is probably supposed to be
161 ptlrpc_set_import_active. Horrible naming. */
162 ptlrpc_activate_import(imp);
165 if (imp->imp_invalid) {
166 CERROR("not connecting OSC %s; administratively "
167 "disabled\n", obd_uuid2str(&tgt_uuid));
168 rc = obd_register_observer(tgt_obd, obd);
170 CERROR("Target %s register_observer error %d; "
171 "will not be able to reactivate\n",
172 obd_uuid2str(&tgt_uuid), rc);
177 rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data, NULL);
179 CERROR("Target %s connect error %d\n",
180 obd_uuid2str(&tgt_uuid), rc);
183 lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
184 if (!lov->lov_tgts[index]->ltd_exp) {
185 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
189 rc = obd_register_observer(tgt_obd, obd);
191 CERROR("Target %s register_observer error %d\n",
192 obd_uuid2str(&tgt_uuid), rc);
193 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
194 lov->lov_tgts[index]->ltd_exp = NULL;
198 lov->lov_tgts[index]->ltd_reap = 0;
200 lov->lov_tgts[index]->ltd_active = 1;
201 lov->desc.ld_active_tgt_count++;
202 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
204 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
205 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
208 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
210 struct obd_device *osc_obd = class_conn2obd(&conn);
211 cfs_proc_dir_entry_t *osc_symlink;
212 char name[MAX_STRING_SIZE];
214 LASSERT(osc_obd != NULL);
215 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
216 LASSERT(osc_obd->obd_type->typ_name != NULL);
217 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
218 osc_obd->obd_type->typ_name,
220 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
222 if (osc_symlink == NULL) {
223 CERROR("could not register LOV target "
224 "/proc/fs/lustre/%s/%s/target_obds/%s.",
225 obd->obd_type->typ_name, obd->obd_name,
227 lprocfs_remove(&lov_proc_dir);
232 rc = qos_add_tgt(obd, index);
234 CERROR("qos_add_tgt failed %d\n", rc);
239 static int lov_connect(const struct lu_env *env,
240 struct lustre_handle *conn, struct obd_device *obd,
241 struct obd_uuid *cluuid, struct obd_connect_data *data,
244 struct lov_obd *lov = &obd->u.lov;
245 struct lov_tgt_desc *tgt;
249 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
251 rc = class_connect(conn, obd, cluuid);
255 /* Why should there ever be more than 1 connect? */
257 LASSERT(lov->lov_connects == 1);
259 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
261 lov->lov_ocd = *data;
264 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
265 tgt = lov->lov_tgts[i];
266 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
268 /* Flags will be lowest common denominator */
269 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
271 CERROR("%s: lov connect tgt %d failed: %d\n",
272 obd->obd_name, i, rc);
275 /* connect to administrative disabled ost */
276 if (!lov->lov_tgts[i]->ltd_exp)
279 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
280 OBD_NOTIFY_ACTIVE, (void *)&i);
282 CERROR("%s error sending notify %d\n",
291 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
293 cfs_proc_dir_entry_t *lov_proc_dir;
294 struct lov_obd *lov = &obd->u.lov;
295 struct obd_device *osc_obd;
300 if (lov->lov_tgts[index] == NULL)
303 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
304 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
305 obd->obd_name, osc_obd->obd_name);
307 if (lov->lov_tgts[index]->ltd_active) {
308 lov->lov_tgts[index]->ltd_active = 0;
309 lov->desc.ld_active_tgt_count--;
310 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
313 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
315 cfs_proc_dir_entry_t *osc_symlink;
317 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
319 lprocfs_remove(&osc_symlink);
321 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
322 obd->obd_type->typ_name, obd->obd_name,
328 /* Pass it on to our clients.
329 * XXX This should be an argument to disconnect,
330 * XXX not a back-door flag on the OBD. Ah well.
332 osc_obd->obd_force = obd->obd_force;
333 osc_obd->obd_fail = obd->obd_fail;
334 osc_obd->obd_no_recov = obd->obd_no_recov;
337 obd_register_observer(osc_obd, NULL);
339 rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
341 CERROR("Target %s disconnect error %d\n",
342 lov_uuid2str(lov, index), rc);
346 qos_del_tgt(obd, index);
348 lov->lov_tgts[index]->ltd_exp = NULL;
352 static int lov_disconnect(struct obd_export *exp)
354 struct obd_device *obd = class_exp2obd(exp);
355 struct lov_obd *lov = &obd->u.lov;
362 /* Only disconnect the underlying layers on the final disconnect. */
364 if (lov->lov_connects != 0) {
365 /* why should there be more than 1 connect? */
366 CERROR("disconnect #%d\n", lov->lov_connects);
370 /* Let's hold another reference so lov_del_obd doesn't spin through
374 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
375 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
376 /* Disconnection is the last we know about an obd */
377 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
383 rc = class_disconnect(exp); /* bz 9811 */
389 * -EINVAL : UUID can't be found in the LOV's target list
390 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
391 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
392 * any >= 0 : is log target index
394 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
397 struct lov_obd *lov = &obd->u.lov;
398 struct lov_tgt_desc *tgt;
402 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
403 lov, uuid->uuid, activate);
406 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
407 tgt = lov->lov_tgts[index];
408 if (!tgt || !tgt->ltd_exp)
411 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
412 index, obd_uuid2str(&tgt->ltd_uuid),
413 tgt->ltd_exp->exp_handle.h_cookie);
414 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
418 if (index == lov->desc.ld_tgt_count)
419 GOTO(out, index = -EINVAL);
421 if (lov->lov_tgts[index]->ltd_active == activate) {
422 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
423 activate ? "" : "in");
427 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
428 activate ? "" : "in");
430 lov->lov_tgts[index]->ltd_active = activate;
433 lov->desc.ld_active_tgt_count++;
434 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
436 lov->desc.ld_active_tgt_count--;
437 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
439 /* remove any old qos penalty */
440 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
447 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
448 enum obd_notify_event ev, void *data)
453 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
454 struct obd_uuid *uuid;
458 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
459 CERROR("unexpected notification of %s %s!\n",
460 watched->obd_type->typ_name,
464 uuid = &watched->u.cli.cl_target_uuid;
466 /* Set OSC as active before notifying the observer, so the
467 * observer can use the OSC normally.
469 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
471 CERROR("%sactivation of %s failed: %d\n",
472 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
473 obd_uuid2str(uuid), rc);
476 /* active event should be pass lov target index as data */
480 /* Pass the notification up the chain. */
482 rc = obd_notify_observer(obd, watched, ev, data);
484 /* NULL watched means all osc's in the lov (only for syncs) */
485 /* sync event should be send lov idx as data */
486 struct lov_obd *lov = &obd->u.lov;
487 struct obd_device *tgt_obd;
490 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
491 if (!lov->lov_tgts[i])
494 if ((ev == OBD_NOTIFY_SYNC) ||
495 (ev == OBD_NOTIFY_SYNC_NONBLOCK))
498 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
500 rc = obd_notify_observer(obd, tgt_obd, ev, data);
502 CERROR("%s: notify %s of %s failed %d\n",
504 obd->obd_observer->obd_name,
505 tgt_obd->obd_name, rc);
515 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
516 __u32 index, int gen, int active)
518 struct lov_obd *lov = &obd->u.lov;
519 struct lov_tgt_desc *tgt;
523 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
524 uuidp->uuid, index, gen, active);
527 CERROR("request to add OBD %s with invalid generation: %d\n",
532 mutex_down(&lov->lov_lock);
534 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
535 tgt = lov->lov_tgts[index];
536 CERROR("UUID %s already assigned at LOV target index %d\n",
537 obd_uuid2str(&tgt->ltd_uuid), index);
538 mutex_up(&lov->lov_lock);
542 if (index >= lov->lov_tgt_size) {
543 /* We need to reallocate the lov target array. */
544 struct lov_tgt_desc **newtgts, **old = NULL;
545 __u32 newsize, oldsize = 0;
547 newsize = max(lov->lov_tgt_size, (__u32)2);
548 while (newsize < index + 1)
549 newsize = newsize << 1;
550 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
551 if (newtgts == NULL) {
552 mutex_up(&lov->lov_lock);
556 if (lov->lov_tgt_size) {
557 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
560 oldsize = lov->lov_tgt_size;
563 lov->lov_tgts = newtgts;
564 lov->lov_tgt_size = newsize;
569 OBD_FREE(old, sizeof(*old) * oldsize);
571 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
572 lov->lov_tgts, lov->lov_tgt_size);
577 mutex_up(&lov->lov_lock);
581 memset(tgt, 0, sizeof(*tgt));
582 tgt->ltd_uuid = *uuidp;
583 /* XXX - add a sanity check on the generation number. */
585 tgt->ltd_index = index;
586 tgt->ltd_activate = active;
587 lov->lov_tgts[index] = tgt;
588 if (index >= lov->desc.ld_tgt_count)
589 lov->desc.ld_tgt_count = index + 1;
591 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
595 mutex_up(&lov->lov_lock);
597 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
598 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
600 if (lov->lov_connects == 0) {
601 /* lov_connect hasn't been called yet. We'll do the
602 lov_connect_obd on this target when that fn first runs,
603 because we don't know the connect flags yet. */
609 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
613 /* connect to administrative disabled ost */
617 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
618 active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
623 CERROR("add failed (%d), deleting %s\n", rc,
624 obd_uuid2str(&tgt->ltd_uuid));
625 lov_del_target(obd, index, 0, 0);
631 /* Schedule a target for deletion */
632 int lov_del_target(struct obd_device *obd, __u32 index,
633 struct obd_uuid *uuidp, int gen)
635 struct lov_obd *lov = &obd->u.lov;
636 int count = lov->desc.ld_tgt_count;
640 if (index >= count) {
641 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
648 if (!lov->lov_tgts[index]) {
649 CERROR("LOV target at index %d is not setup.\n", index);
650 GOTO(out, rc = -EINVAL);
653 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
654 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
655 lov_uuid2str(lov, index), index,
656 obd_uuid2str(uuidp));
657 GOTO(out, rc = -EINVAL);
660 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
661 lov_uuid2str(lov, index), index,
662 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
663 lov->lov_tgts[index]->ltd_active);
665 lov->lov_tgts[index]->ltd_reap = 1;
666 lov->lov_death_row++;
667 /* we really delete it from lov_putref */
674 /* We are holding lov_lock */
675 static void __lov_del_obd(struct obd_device *obd, __u32 index)
677 struct lov_obd *lov = &obd->u.lov;
678 struct obd_device *osc_obd;
679 struct lov_tgt_desc *tgt = lov->lov_tgts[index];
682 LASSERT(tgt->ltd_reap);
684 osc_obd = class_exp2obd(tgt->ltd_exp);
686 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
687 lov_uuid2str(lov, index),
688 osc_obd ? osc_obd->obd_name : "<no obd>");
691 lov_disconnect_obd(obd, index);
693 /* XXX - right now there is a dependency on ld_tgt_count being the
694 * maximum tgt index for computing the mds_max_easize. So we can't
697 lov_ost_pool_remove(&lov->lov_packed, index);
698 lov->lov_tgts[index] = NULL;
701 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
702 do it ourselves. And we can't do it from lov_cleanup,
703 because we just lost our only reference to it. */
705 class_manual_cleanup(osc_obd);
708 void lov_fix_desc_stripe_size(__u64 *val)
710 if (*val < PTLRPC_MAX_BRW_SIZE) {
711 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
712 PTLRPC_MAX_BRW_SIZE);
713 *val = PTLRPC_MAX_BRW_SIZE;
714 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
715 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
716 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
718 *val, LOV_MIN_STRIPE_SIZE);
722 void lov_fix_desc_stripe_count(__u32 *val)
728 void lov_fix_desc_pattern(__u32 *val)
730 /* from lov_setstripe */
731 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
732 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
737 void lov_fix_desc_qos_maxage(__u32 *val)
741 *val = QOS_DEFAULT_MAXAGE;
744 void lov_fix_desc(struct lov_desc *desc)
746 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
747 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
748 lov_fix_desc_pattern(&desc->ld_pattern);
749 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
752 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
754 struct lprocfs_static_vars lvars = { 0 };
755 struct lov_desc *desc;
756 struct lov_obd *lov = &obd->u.lov;
761 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
762 CERROR("LOV setup requires a descriptor\n");
766 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
768 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
769 CERROR("descriptor size wrong: %d > %d\n",
770 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
774 if (desc->ld_magic != LOV_DESC_MAGIC) {
775 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
776 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
777 obd->obd_name, desc);
778 lustre_swab_lov_desc(desc);
780 CERROR("%s: Bad lov desc magic: %#x\n",
781 obd->obd_name, desc->ld_magic);
788 /* Because of 64-bit divide/mod operations only work with a 32-bit
789 * divisor in a 32-bit kernel, we cannot support a stripe width
790 * of 4GB or larger on 32-bit CPUs. */
791 count = desc->ld_default_stripe_count;
792 if ((count > 0 ? count : desc->ld_tgt_count) *
793 desc->ld_default_stripe_size > 0xffffffff) {
794 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
795 desc->ld_default_stripe_size, count);
799 desc->ld_active_tgt_count = 0;
801 lov->lov_tgt_size = 0;
803 sema_init(&lov->lov_lock, 1);
804 atomic_set(&lov->lov_refcount, 0);
805 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
806 init_rwsem(&lov->lov_qos.lq_rw_sem);
807 lov->lov_sp_me = LUSTRE_SP_CLI;
808 lov->lov_qos.lq_dirty = 1;
809 lov->lov_qos.lq_rr.lqr_dirty = 1;
810 lov->lov_qos.lq_reset = 1;
811 /* Default priority is toward free space balance */
812 lov->lov_qos.lq_prio_free = 232;
814 lov->lov_pools_hash_body = lustre_hash_init("POOLS", 7, 7,
815 &pool_hash_operations, 0);
816 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
817 lov->lov_pool_count = 0;
818 rc = lov_ost_pool_init(&lov->lov_packed, 0);
821 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
823 lov_ost_pool_free(&lov->lov_packed);
827 lprocfs_lov_init_vars(&lvars);
828 lprocfs_obd_setup(obd, lvars.obd_vars);
833 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
834 0444, &lov_proc_target_fops, obd);
836 CWARN("Error adding the target_obd file\n");
839 lov->lov_pool_proc_entry = lprocfs_register("pools",
846 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
852 case OBD_CLEANUP_EARLY: {
853 struct lov_obd *lov = &obd->u.lov;
855 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
856 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
858 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
863 case OBD_CLEANUP_EXPORTS:
864 rc = obd_llog_finish(obd, 0);
866 CERROR("failed to cleanup llogging subsystems\n");
872 static int lov_cleanup(struct obd_device *obd)
874 struct lov_obd *lov = &obd->u.lov;
875 struct list_head *pos, *tmp;
876 struct pool_desc *pool;
878 lprocfs_obd_cleanup(obd);
880 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
881 pool = list_entry(pos, struct pool_desc, pool_list);
882 /* free pool structs */
883 lov_pool_del(obd, pool->pool_name);
885 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
886 lov_ost_pool_free(&lov->lov_packed);
887 lustre_hash_exit(lov->lov_pools_hash_body);
891 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
892 if (!lov->lov_tgts[i])
895 /* Inactive targets may never have connected */
896 if (lov->lov_tgts[i]->ltd_active ||
897 atomic_read(&lov->lov_refcount))
898 /* We should never get here - these
899 should have been removed in the
901 CERROR("lov tgt %d not cleaned!"
902 " deathrow=%d, lovrc=%d\n",
903 i, lov->lov_death_row,
904 atomic_read(&lov->lov_refcount));
905 lov_del_target(obd, i, 0, 0);
907 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
909 lov->lov_tgt_size = 0;
915 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
916 __u32 *indexp, int *genp)
918 struct obd_uuid obd_uuid;
923 switch(cmd = lcfg->lcfg_command) {
924 case LCFG_LOV_ADD_OBD:
925 case LCFG_LOV_ADD_INA:
926 case LCFG_LOV_DEL_OBD: {
929 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
930 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
931 GOTO(out, rc = -EINVAL);
933 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
935 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
936 GOTO(out, rc = -EINVAL);
937 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
938 GOTO(out, rc = -EINVAL);
941 if (cmd == LCFG_LOV_ADD_OBD)
942 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
943 else if (cmd == LCFG_LOV_ADD_INA)
944 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
946 rc = lov_del_target(obd, index, &obd_uuid, gen);
950 struct lprocfs_static_vars lvars = { 0 };
951 struct lov_desc *desc = &(obd->u.lov.desc);
954 GOTO(out, rc = -EINVAL);
956 lprocfs_lov_init_vars(&lvars);
958 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
971 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
972 GOTO(out, rc = -EINVAL);
981 #define log2(n) ffz(~(n))
984 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
985 struct lov_stripe_md **ea,
986 struct obd_trans_info *oti)
990 struct obd_uuid *ost_uuid = NULL;
994 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
995 src_oa->o_flags == OBD_FL_DELORPHAN);
997 lov = &export->exp_obd->u.lov;
1003 if (oti->oti_ost_uuid) {
1004 ost_uuid = oti->oti_ost_uuid;
1005 CDEBUG(D_HA, "clearing orphans only for %s\n",
1009 lov_getref(export->exp_obd);
1010 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1011 struct lov_stripe_md obj_md;
1012 struct lov_stripe_md *obj_mdp = &obj_md;
1013 struct lov_tgt_desc *tgt;
1016 tgt = lov->lov_tgts[i];
1020 /* if called for a specific target, we don't
1021 care if it is not active. */
1022 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1023 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1027 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1030 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1031 obd_uuid2str(ost_uuid));
1033 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1035 LASSERT(lov->lov_tgts[i]->ltd_exp);
1036 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1037 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1038 tmp_oa, &obj_mdp, oti);
1040 /* This export will be disabled until it is recovered,
1041 and then orphan recovery will be completed. */
1042 CERROR("error in orphan recovery on OST idx %d/%d: "
1043 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1050 lov_putref(export->exp_obd);
1056 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1057 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1059 struct lov_stripe_md *obj_mdp, *lsm;
1060 struct lov_obd *lov = &exp->exp_obd->u.lov;
1065 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1066 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1068 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1069 if (obj_mdp == NULL)
1072 ost_idx = src_oa->o_nlink;
1075 GOTO(out, rc = -EINVAL);
1076 if (ost_idx >= lov->desc.ld_tgt_count ||
1077 !lov->lov_tgts[ost_idx])
1078 GOTO(out, rc = -EINVAL);
1080 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1081 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1082 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1083 GOTO(out, rc = -EINVAL);
1087 if (i == lsm->lsm_stripe_count)
1088 GOTO(out, rc = -EINVAL);
1090 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1092 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1096 /* the LOV expects oa->o_id to be set to the LOV object id */
1097 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1098 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1100 struct lov_obd *lov;
1101 struct obd_info oinfo;
1102 struct lov_request_set *set = NULL;
1103 struct lov_request *req;
1104 struct obd_statfs osfs;
1109 LASSERT(ea != NULL);
1113 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1114 src_oa->o_flags == OBD_FL_DELORPHAN) {
1115 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1119 lov = &exp->exp_obd->u.lov;
1120 if (!lov->desc.ld_active_tgt_count)
1123 lov_getref(exp->exp_obd);
1124 /* Recreate a specific object id at the given OST index */
1125 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1126 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1127 rc = lov_recreate(exp, src_oa, ea, oti);
1131 maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1132 obd_statfs_rqset(exp->exp_obd, &osfs, maxage, OBD_STATFS_NODELAY);
1134 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1138 list_for_each_entry(req, &set->set_list, rq_link) {
1139 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1140 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1141 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1142 lov_update_create_set(set, req, rc);
1144 rc = lov_fini_create_set(set, ea);
1146 lov_putref(exp->exp_obd);
1150 #define ASSERT_LSM_MAGIC(lsmp) \
1152 LASSERT((lsmp) != NULL); \
1153 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1154 (lsmp)->lsm_magic == LOV_MAGIC_V3 || \
1155 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1156 (lsmp), (lsmp)->lsm_magic); \
1159 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1160 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1161 struct obd_export *md_exp, void *capa)
1163 struct lov_request_set *set;
1164 struct obd_info oinfo;
1165 struct lov_request *req;
1166 struct list_head *pos;
1167 struct lov_obd *lov;
1168 int rc = 0, err = 0;
1171 ASSERT_LSM_MAGIC(lsm);
1173 if (!exp || !exp->exp_obd)
1176 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1178 LASSERT(oti->oti_logcookies);
1181 lov = &exp->exp_obd->u.lov;
1182 lov_getref(exp->exp_obd);
1183 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1187 list_for_each (pos, &set->set_list) {
1188 req = list_entry(pos, struct lov_request, rq_link);
1190 if (oa->o_valid & OBD_MD_FLCOOKIE)
1191 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1193 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1194 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1195 err = lov_update_common_set(set, req, err);
1197 CERROR("error: destroying objid "LPX64" subobj "
1198 LPX64" on OST idx %d: rc = %d\n",
1199 oa->o_id, req->rq_oi.oi_oa->o_id,
1207 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1208 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1210 err = lov_fini_destroy_set(set);
1212 lov_putref(exp->exp_obd);
1213 RETURN(rc ? rc : err);
1216 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1218 struct lov_request_set *set;
1219 struct lov_request *req;
1220 struct list_head *pos;
1221 struct lov_obd *lov;
1222 int err = 0, rc = 0;
1226 ASSERT_LSM_MAGIC(oinfo->oi_md);
1228 if (!exp || !exp->exp_obd)
1231 lov = &exp->exp_obd->u.lov;
1233 rc = lov_prep_getattr_set(exp, oinfo, &set);
1237 list_for_each (pos, &set->set_list) {
1238 req = list_entry(pos, struct lov_request, rq_link);
1240 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1241 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1242 req->rq_oi.oi_oa->o_id, req->rq_idx);
1244 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1246 err = lov_update_common_set(set, req, rc);
1248 CERROR("error: getattr objid "LPX64" subobj "
1249 LPX64" on OST idx %d: rc = %d\n",
1250 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1256 rc = lov_fini_getattr_set(set);
1262 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1265 struct lov_request_set *lovset = (struct lov_request_set *)data;
1269 /* don't do attribute merge if this aysnc op failed */
1271 lovset->set_completes = 0;
1272 err = lov_fini_getattr_set(lovset);
1273 RETURN(rc ? rc : err);
1276 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1277 struct ptlrpc_request_set *rqset)
1279 struct lov_request_set *lovset;
1280 struct lov_obd *lov;
1281 struct list_head *pos;
1282 struct lov_request *req;
1287 ASSERT_LSM_MAGIC(oinfo->oi_md);
1289 if (!exp || !exp->exp_obd)
1292 lov = &exp->exp_obd->u.lov;
1294 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1298 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1299 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1300 oinfo->oi_md->lsm_stripe_size);
1302 list_for_each (pos, &lovset->set_list) {
1303 req = list_entry(pos, struct lov_request, rq_link);
1305 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1306 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1307 req->rq_oi.oi_oa->o_id, req->rq_idx);
1308 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1309 &req->rq_oi, rqset);
1311 CERROR("error: getattr objid "LPX64" subobj "
1312 LPX64" on OST idx %d: rc = %d\n",
1313 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1319 if (!list_empty(&rqset->set_requests)) {
1321 LASSERT (rqset->set_interpret == NULL);
1322 rqset->set_interpret = lov_getattr_interpret;
1323 rqset->set_arg = (void *)lovset;
1328 lovset->set_completes = 0;
1329 err = lov_fini_getattr_set(lovset);
1330 RETURN(rc ? rc : err);
1333 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1334 struct obd_trans_info *oti)
1336 struct lov_request_set *set;
1337 struct lov_obd *lov;
1338 struct list_head *pos;
1339 struct lov_request *req;
1340 int err = 0, rc = 0;
1344 ASSERT_LSM_MAGIC(oinfo->oi_md);
1346 if (!exp || !exp->exp_obd)
1349 /* for now, we only expect the following updates here */
1350 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1351 OBD_MD_FLMODE | OBD_MD_FLATIME |
1352 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1353 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1354 OBD_MD_FLGROUP | OBD_MD_FLUID |
1355 OBD_MD_FLGID | OBD_MD_FLFID |
1357 lov = &exp->exp_obd->u.lov;
1358 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1362 list_for_each (pos, &set->set_list) {
1363 req = list_entry(pos, struct lov_request, rq_link);
1365 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1367 err = lov_update_setattr_set(set, req, rc);
1369 CERROR("error: setattr objid "LPX64" subobj "
1370 LPX64" on OST idx %d: rc = %d\n",
1371 set->set_oi->oi_oa->o_id,
1372 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1377 err = lov_fini_setattr_set(set);
1383 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1386 struct lov_request_set *lovset = (struct lov_request_set *)data;
1391 lovset->set_completes = 0;
1392 err = lov_fini_setattr_set(lovset);
1393 RETURN(rc ? rc : err);
1396 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1397 needed. Otherwise, a client is waiting for responses. */
1398 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1399 struct obd_trans_info *oti,
1400 struct ptlrpc_request_set *rqset)
1402 struct lov_request_set *set;
1403 struct lov_request *req;
1404 struct list_head *pos;
1405 struct lov_obd *lov;
1410 ASSERT_LSM_MAGIC(oinfo->oi_md);
1411 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1413 LASSERT(oti->oti_logcookies);
1416 if (!exp || !exp->exp_obd)
1419 lov = &exp->exp_obd->u.lov;
1420 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1424 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1425 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1426 oinfo->oi_md->lsm_stripe_size);
1428 list_for_each (pos, &set->set_list) {
1429 req = list_entry(pos, struct lov_request, rq_link);
1431 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1432 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1434 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1435 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1436 req->rq_oi.oi_oa->o_id, req->rq_idx);
1438 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1439 &req->rq_oi, oti, rqset);
1441 CERROR("error: setattr objid "LPX64" subobj "
1442 LPX64" on OST idx %d: rc = %d\n",
1443 set->set_oi->oi_oa->o_id,
1444 req->rq_oi.oi_oa->o_id,
1450 /* If we are not waiting for responses on async requests, return. */
1451 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1454 set->set_completes = 0;
1455 err = lov_fini_setattr_set(set);
1456 RETURN(rc ? rc : err);
1459 LASSERT(rqset->set_interpret == NULL);
1460 rqset->set_interpret = lov_setattr_interpret;
1461 rqset->set_arg = (void *)set;
1466 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1469 struct lov_request_set *lovset = (struct lov_request_set *)data;
1474 lovset->set_completes = 0;
1475 err = lov_fini_punch_set(lovset);
1476 RETURN(rc ? rc : err);
1479 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1480 * we can send this 'punch' to just the authoritative node and the nodes
1481 * that the punch will affect. */
1482 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1483 struct obd_trans_info *oti,
1484 struct ptlrpc_request_set *rqset)
1486 struct lov_request_set *set;
1487 struct lov_obd *lov;
1488 struct list_head *pos;
1489 struct lov_request *req;
1494 ASSERT_LSM_MAGIC(oinfo->oi_md);
1496 if (!exp || !exp->exp_obd)
1499 lov = &exp->exp_obd->u.lov;
1500 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1504 list_for_each (pos, &set->set_list) {
1505 req = list_entry(pos, struct lov_request, rq_link);
1507 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1508 &req->rq_oi, NULL, rqset);
1510 CERROR("error: punch objid "LPX64" subobj "LPX64
1511 " on OST idx %d: rc = %d\n",
1512 set->set_oi->oi_oa->o_id,
1513 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1518 if (rc || list_empty(&rqset->set_requests)) {
1520 err = lov_fini_punch_set(set);
1521 RETURN(rc ? rc : err);
1524 LASSERT(rqset->set_interpret == NULL);
1525 rqset->set_interpret = lov_punch_interpret;
1526 rqset->set_arg = (void *)set;
1531 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1532 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1535 struct lov_request_set *set;
1536 struct obd_info oinfo;
1537 struct lov_obd *lov;
1538 struct list_head *pos;
1539 struct lov_request *req;
1540 int err = 0, rc = 0;
1543 ASSERT_LSM_MAGIC(lsm);
1548 lov = &exp->exp_obd->u.lov;
1549 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1553 list_for_each (pos, &set->set_list) {
1554 req = list_entry(pos, struct lov_request, rq_link);
1556 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1557 req->rq_oi.oi_oa, NULL,
1558 req->rq_oi.oi_policy.l_extent.start,
1559 req->rq_oi.oi_policy.l_extent.end, capa);
1560 err = lov_update_common_set(set, req, rc);
1562 CERROR("error: fsync objid "LPX64" subobj "LPX64
1563 " on OST idx %d: rc = %d\n",
1564 set->set_oi->oi_oa->o_id,
1565 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1570 err = lov_fini_sync_set(set);
1576 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1577 obd_count oa_bufs, struct brw_page *pga)
1579 struct obd_info oinfo = { { { 0 } } };
1582 oinfo.oi_oa = lov_oinfo->oi_oa;
1584 /* The caller just wants to know if there's a chance that this
1585 * I/O can succeed */
1586 for (i = 0; i < oa_bufs; i++) {
1587 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1588 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1591 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1592 pga[i].off + pga[i].count - 1,
1596 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1597 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1601 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1609 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1610 obd_count oa_bufs, struct brw_page *pga,
1611 struct obd_trans_info *oti)
1613 struct lov_request_set *set;
1614 struct lov_request *req;
1615 struct list_head *pos;
1616 struct lov_obd *lov = &exp->exp_obd->u.lov;
1620 ASSERT_LSM_MAGIC(oinfo->oi_md);
1622 if (cmd == OBD_BRW_CHECK) {
1623 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1627 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1631 list_for_each (pos, &set->set_list) {
1632 struct obd_export *sub_exp;
1633 struct brw_page *sub_pga;
1634 req = list_entry(pos, struct lov_request, rq_link);
1636 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1637 sub_pga = set->set_pga + req->rq_pgaidx;
1638 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1642 lov_update_common_set(set, req, rc);
1645 err = lov_fini_brw_set(set);
1651 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1654 struct lov_request_set *lovset = (struct lov_request_set *)data;
1656 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1660 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1661 struct ldlm_enqueue_info *einfo,
1662 struct ptlrpc_request_set *rqset)
1664 ldlm_mode_t mode = einfo->ei_mode;
1665 struct lov_request_set *set;
1666 struct lov_request *req;
1667 struct list_head *pos;
1668 struct lov_obd *lov;
1673 ASSERT_LSM_MAGIC(oinfo->oi_md);
1674 LASSERT(mode == (mode & -mode));
1676 /* we should never be asked to replay a lock this way. */
1677 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1679 if (!exp || !exp->exp_obd)
1682 lov = &exp->exp_obd->u.lov;
1683 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1687 list_for_each (pos, &set->set_list) {
1688 req = list_entry(pos, struct lov_request, rq_link);
1690 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1691 &req->rq_oi, einfo, rqset);
1696 if (rqset && !list_empty(&rqset->set_requests)) {
1698 LASSERT(rqset->set_interpret == NULL);
1699 rqset->set_interpret = lov_enqueue_interpret;
1700 rqset->set_arg = (void *)set;
1704 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1708 static int lov_change_cbdata(struct obd_export *exp,
1709 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1712 struct lov_obd *lov;
1716 ASSERT_LSM_MAGIC(lsm);
1718 if (!exp || !exp->exp_obd)
1721 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1723 lov = &exp->exp_obd->u.lov;
1724 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1725 struct lov_stripe_md submd;
1726 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1728 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1729 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1733 submd.lsm_object_id = loi->loi_id;
1734 submd.lsm_object_gr = lsm->lsm_object_gr;
1735 submd.lsm_stripe_count = 0;
1736 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1742 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1743 __u32 mode, struct lustre_handle *lockh)
1745 struct lov_request_set *set;
1746 struct obd_info oinfo;
1747 struct lov_request *req;
1748 struct list_head *pos;
1749 struct lov_obd *lov = &exp->exp_obd->u.lov;
1750 struct lustre_handle *lov_lockhp;
1751 int err = 0, rc = 0;
1754 ASSERT_LSM_MAGIC(lsm);
1756 if (!exp || !exp->exp_obd)
1759 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1761 lov = &exp->exp_obd->u.lov;
1762 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1766 list_for_each (pos, &set->set_list) {
1767 req = list_entry(pos, struct lov_request, rq_link);
1768 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1770 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1771 req->rq_oi.oi_md, mode, lov_lockhp);
1772 rc = lov_update_common_set(set, req, rc);
1774 CERROR("error: cancel objid "LPX64" subobj "
1775 LPX64" on OST idx %d: rc = %d\n",
1777 req->rq_oi.oi_md->lsm_object_id,
1783 lov_fini_cancel_set(set);
1787 static int lov_cancel_unused(struct obd_export *exp,
1788 struct lov_stripe_md *lsm,
1789 int flags, void *opaque)
1791 struct lov_obd *lov;
1795 if (!exp || !exp->exp_obd)
1798 lov = &exp->exp_obd->u.lov;
1800 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1802 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1805 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1813 ASSERT_LSM_MAGIC(lsm);
1815 LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1816 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1817 struct lov_stripe_md submd;
1818 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1821 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1822 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1826 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1827 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1829 submd.lsm_object_id = loi->loi_id;
1830 submd.lsm_object_gr = lsm->lsm_object_gr;
1831 submd.lsm_stripe_count = 0;
1832 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1833 &submd, flags, opaque);
1834 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1835 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1836 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1837 loi->loi_id, loi->loi_ost_idx, err);
1845 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
1848 struct lov_request_set *lovset = (struct lov_request_set *)data;
1853 lovset->set_completes = 0;
1855 err = lov_fini_statfs_set(lovset);
1856 RETURN(rc ? rc : err);
1859 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1860 __u64 max_age, struct ptlrpc_request_set *rqset)
1862 struct lov_request_set *set;
1863 struct lov_request *req;
1864 struct list_head *pos;
1865 struct lov_obd *lov;
1869 LASSERT(oinfo != NULL);
1870 LASSERT(oinfo->oi_osfs != NULL);
1873 rc = lov_prep_statfs_set(obd, oinfo, &set);
1877 list_for_each (pos, &set->set_list) {
1878 struct obd_device *osc_obd;
1880 req = list_entry(pos, struct lov_request, rq_link);
1882 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1883 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1888 if (rc || list_empty(&rqset->set_requests)) {
1891 set->set_completes = 0;
1892 err = lov_fini_statfs_set(set);
1893 RETURN(rc ? rc : err);
1896 LASSERT(rqset->set_interpret == NULL);
1897 rqset->set_interpret = lov_statfs_interpret;
1898 rqset->set_arg = (void *)set;
1902 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1903 __u64 max_age, __u32 flags)
1905 struct ptlrpc_request_set *set = NULL;
1906 struct obd_info oinfo = { { { 0 } } };
1911 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1912 * statfs requests */
1913 set = ptlrpc_prep_set();
1917 oinfo.oi_osfs = osfs;
1918 oinfo.oi_flags = flags;
1919 rc = lov_statfs_async(obd, &oinfo, max_age, set);
1921 rc = ptlrpc_set_wait(set);
1922 ptlrpc_set_destroy(set);
1927 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1928 void *karg, void *uarg)
1930 struct obd_device *obddev = class_exp2obd(exp);
1931 struct lov_obd *lov = &obddev->u.lov;
1932 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1933 struct obd_uuid *uuidp;
1937 case IOC_OBD_STATFS: {
1938 struct obd_ioctl_data *data = karg;
1939 struct obd_device *osc_obd;
1940 struct obd_statfs stat_buf = {0};
1943 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1944 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1946 if ((index >= count))
1949 if (!lov->lov_tgts[index])
1950 /* Try again with the next index */
1952 if (!lov->lov_tgts[index]->ltd_active)
1955 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1959 /* got statfs data */
1960 rc = obd_statfs(osc_obd, &stat_buf,
1961 cfs_time_current_64() - HZ, 0);
1964 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
1967 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1972 case OBD_IOC_LOV_GET_CONFIG: {
1973 struct obd_ioctl_data *data;
1974 struct lov_desc *desc;
1979 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1982 data = (struct obd_ioctl_data *)buf;
1984 if (sizeof(*desc) > data->ioc_inllen1) {
1985 obd_ioctl_freedata(buf, len);
1989 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1990 obd_ioctl_freedata(buf, len);
1994 if (sizeof(__u32) * count > data->ioc_inllen3) {
1995 obd_ioctl_freedata(buf, len);
1999 desc = (struct lov_desc *)data->ioc_inlbuf1;
2000 memcpy(desc, &(lov->desc), sizeof(*desc));
2002 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2003 genp = (__u32 *)data->ioc_inlbuf3;
2004 /* the uuid will be empty for deleted OSTs */
2005 for (i = 0; i < count; i++, uuidp++, genp++) {
2006 if (!lov->lov_tgts[i])
2008 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2009 *genp = lov->lov_tgts[i]->ltd_gen;
2012 if (copy_to_user((void *)uarg, buf, len))
2014 obd_ioctl_freedata(buf, len);
2017 case LL_IOC_LOV_SETSTRIPE:
2018 rc = lov_setstripe(exp, karg, uarg);
2020 case LL_IOC_LOV_GETSTRIPE:
2021 rc = lov_getstripe(exp, karg, uarg);
2023 case LL_IOC_LOV_SETEA:
2024 rc = lov_setea(exp, karg, uarg);
2026 case OBD_IOC_QUOTACTL: {
2027 struct if_quotactl *qctl = karg;
2028 struct lov_tgt_desc *tgt = NULL;
2029 struct obd_quotactl *oqctl;
2031 if (qctl->qc_valid == QC_OSTIDX) {
2032 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2035 tgt = lov->lov_tgts[qctl->qc_idx];
2036 if (!tgt || !tgt->ltd_exp)
2038 } else if (qctl->qc_valid == QC_UUID) {
2039 for (i = 0; i < count; i++) {
2040 tgt = lov->lov_tgts[i];
2042 !obd_uuid_equals(&tgt->ltd_uuid,
2046 if (tgt->ltd_exp == NULL)
2058 LASSERT(tgt && tgt->ltd_exp);
2059 OBD_ALLOC_PTR(oqctl);
2063 QCTL_COPY(oqctl, qctl);
2064 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2066 QCTL_COPY(qctl, oqctl);
2067 qctl->qc_valid = QC_OSTIDX;
2068 qctl->obd_uuid = tgt->ltd_uuid;
2070 OBD_FREE_PTR(oqctl);
2079 for (i = 0; i < count; i++) {
2082 /* OST was disconnected */
2083 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2086 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2088 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2091 if (lov->lov_tgts[i]->ltd_active) {
2092 CDEBUG(err == -ENOTTY ?
2093 D_IOCTL : D_WARNING,
2094 "iocontrol OSC %s on OST "
2095 "idx %d cmd %x: err = %d\n",
2096 lov_uuid2str(lov, i),
2113 #define FIEMAP_BUFFER_SIZE 4096
2116 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2117 * call. The local end offset and the device are sent in the first
2118 * fm_extent. This function calculates the stripe number from the index.
2119 * This function returns a stripe_no on which mapping is to be restarted.
2121 * This function returns fm_end_offset which is the in-OST offset at which
2122 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2123 * will re-calculate proper offset in next stripe.
2124 * Note that the first extent is passed to lov_get_info via the value field.
2126 * \param fiemap fiemap request header
2127 * \param lsm striping information for the file
2128 * \param fm_start logical start of mapping
2129 * \param fm_end logical end of mapping
2130 * \param start_stripe starting stripe will be returned in this
2132 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2133 struct lov_stripe_md *lsm, obd_size fm_start,
2134 obd_size fm_end, int *start_stripe)
2136 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2137 obd_off lun_start, lun_end;
2138 obd_size fm_end_offset;
2139 int stripe_no = -1, i;
2141 if (fiemap->fm_extent_count == 0 ||
2142 fiemap->fm_extents[0].fe_logical == 0)
2145 /* Find out stripe_no from ost_index saved in the fe_device */
2146 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2147 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2148 fiemap->fm_extents[0].fe_device) {
2154 /* If we have finished mapping on previous device, shift logical
2155 * offset to start of next device */
2156 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2157 &lun_start, &lun_end)) != 0 &&
2158 local_end < lun_end) {
2159 fm_end_offset = local_end;
2160 *start_stripe = stripe_no;
2162 /* This is a special value to indicate that caller should
2163 * calculate offset in next stripe. */
2165 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2168 return fm_end_offset;
2172 * We calculate on which OST the mapping will end. If the length of mapping
2173 * is greater than (stripe_size * stripe_count) then the last_stripe will
2174 * will be one just before start_stripe. Else we check if the mapping
2175 * intersects each OST and find last_stripe.
2176 * This function returns the last_stripe and also sets the stripe_count
2177 * over which the mapping is spread
2179 * \param lsm striping information for the file
2180 * \param fm_start logical start of mapping
2181 * \param fm_end logical end of mapping
2182 * \param start_stripe starting stripe of the mapping
2183 * \param stripe_count the number of stripes across which to map is returned
2185 * \retval last_stripe return the last stripe of the mapping
2187 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2188 obd_size fm_end, int start_stripe,
2192 obd_off obd_start, obd_end;
2195 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2196 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2198 *stripe_count = lsm->lsm_stripe_count;
2200 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2201 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2202 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2203 &obd_start, &obd_end)) == 0)
2207 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2214 * Set fe_device and copy extents from local buffer into main return buffer.
2216 * \param fiemap fiemap request header
2217 * \param lcl_fm_ext array of local fiemap extents to be copied
2218 * \param ost_index OST index to be written into the fm_device field for each
2220 * \param ext_count number of extents to be copied
2221 * \param current_extent where to start copying in main extent array
2223 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2224 struct ll_fiemap_extent *lcl_fm_ext,
2225 int ost_index, unsigned int ext_count,
2231 for (ext = 0; ext < ext_count; ext++) {
2232 lcl_fm_ext[ext].fe_device = ost_index;
2233 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2236 /* Copy fm_extent's from fm_local to return buffer */
2237 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2238 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2242 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2243 * This also handles the restarting of FIEMAP calls in case mapping overflows
2244 * the available number of extents in single call.
2246 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2247 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2249 struct ll_fiemap_info_key *fm_key = key;
2250 struct ll_user_fiemap *fiemap = val;
2251 struct ll_user_fiemap *fm_local = NULL;
2252 struct ll_fiemap_extent *lcl_fm_ext;
2254 unsigned int get_num_extents = 0;
2255 int ost_index = 0, actual_start_stripe, start_stripe;
2256 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2258 int current_extent = 0, rc = 0, i;
2259 int ost_eof = 0; /* EOF for object */
2260 int ost_done = 0; /* done with required mapping for this OST? */
2262 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2263 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2268 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2269 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2271 OBD_ALLOC(fm_local, buffer_size);
2272 if (fm_local == NULL)
2273 GOTO(out, rc = -ENOMEM);
2274 lcl_fm_ext = &fm_local->fm_extents[0];
2276 count_local = fiemap_size_to_count(buffer_size);
2278 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2279 fm_start = fiemap->fm_start;
2280 fm_length = fiemap->fm_length;
2281 /* Calculate start stripe, last stripe and length of mapping */
2282 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2283 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2284 fm_start + fm_length - 1);
2285 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2286 if (fm_end > fm_key->oa.o_size)
2287 fm_end = fm_key->oa.o_size;
2289 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2290 actual_start_stripe, &stripe_count);
2292 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2295 if (fiemap->fm_extent_count == 0) {
2296 get_num_extents = 1;
2300 /* Check each stripe */
2301 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2302 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2303 obd_size req_fm_len; /* Stores length of required mapping */
2304 obd_size len_mapped_single_call;
2305 obd_off lun_start, lun_end, obd_object_end;
2306 unsigned int ext_count;
2308 cur_stripe_wrap = cur_stripe;
2310 /* Find out range of mapping on this stripe */
2311 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2312 &lun_start, &obd_object_end)) == 0)
2315 /* If this is a continuation FIEMAP call and we are on
2316 * starting stripe then lun_start needs to be set to
2318 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2319 lun_start = fm_end_offset;
2321 if (fm_length != ~0ULL) {
2322 /* Handle fm_start + fm_length overflow */
2323 if (fm_start + fm_length < fm_start)
2324 fm_length = ~0ULL - fm_start;
2325 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2331 if (lun_start == lun_end)
2334 req_fm_len = obd_object_end - lun_start;
2335 fm_local->fm_length = 0;
2336 len_mapped_single_call = 0;
2338 /* If the output buffer is very large and the objects have many
2339 * extents we may need to loop on a single OST repeatedly */
2343 if (get_num_extents == 0) {
2344 /* Don't get too many extents. */
2345 if (current_extent + count_local >
2346 fiemap->fm_extent_count)
2347 count_local = fiemap->fm_extent_count -
2351 lun_start += len_mapped_single_call;
2352 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2353 req_fm_len = fm_local->fm_length;
2354 fm_local->fm_extent_count = count_local;
2355 fm_local->fm_mapped_extents = 0;
2356 fm_local->fm_flags = fiemap->fm_flags;
2358 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2359 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2361 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2362 GOTO(out, rc = -EINVAL);
2364 /* If OST is inactive, return extent with UNKNOWN flag */
2365 if (lov && !lov->lov_tgts[ost_index]->ltd_active) {
2366 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2367 fm_local->fm_mapped_extents = 1;
2369 lcl_fm_ext[0].fe_logical = lun_start;
2370 lcl_fm_ext[0].fe_length = obd_object_end -
2372 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2377 fm_local->fm_start = lun_start;
2378 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2379 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2380 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2381 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2382 keylen, key, vallen, fm_local, lsm);
2387 ext_count = fm_local->fm_mapped_extents;
2388 if (ext_count == 0) {
2390 /* If last stripe has hole at the end,
2391 * then we need to return */
2392 if (cur_stripe_wrap == last_stripe) {
2393 fiemap->fm_mapped_extents = 0;
2399 /* If we just need num of extents then go to next device */
2400 if (get_num_extents) {
2401 current_extent += ext_count;
2405 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2406 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2408 /* Have we finished mapping on this device? */
2409 if (req_fm_len <= len_mapped_single_call)
2412 /* Clear the EXTENT_LAST flag which can be present on
2414 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2415 lcl_fm_ext[ext_count - 1].fe_flags &=
2416 ~FIEMAP_EXTENT_LAST;
2418 curr_loc = lov_stripe_size(lsm,
2419 lcl_fm_ext[ext_count - 1].fe_logical+
2420 lcl_fm_ext[ext_count - 1].fe_length,
2422 if (curr_loc >= fm_key->oa.o_size)
2425 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2426 ost_index, ext_count,
2429 current_extent += ext_count;
2431 /* Ran out of available extents? */
2432 if (current_extent >= fiemap->fm_extent_count)
2434 } while (ost_done == 0 && ost_eof == 0);
2436 if (cur_stripe_wrap == last_stripe)
2441 /* Indicate that we are returning device offsets unless file just has
2443 if (lsm->lsm_stripe_count > 1)
2444 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2446 if (get_num_extents)
2447 goto skip_last_device_calc;
2449 /* Check if we have reached the last stripe and whether mapping for that
2450 * stripe is done. */
2451 if (cur_stripe_wrap == last_stripe) {
2452 if (ost_done || ost_eof)
2453 fiemap->fm_extents[current_extent - 1].fe_flags |=
2457 skip_last_device_calc:
2458 fiemap->fm_mapped_extents = current_extent;
2461 OBD_FREE(fm_local, buffer_size);
2465 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2466 void *key, __u32 *vallen, void *val,
2467 struct lov_stripe_md *lsm)
2469 struct obd_device *obddev = class_exp2obd(exp);
2470 struct lov_obd *lov = &obddev->u.lov;
2474 if (!vallen || !val)
2479 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2482 struct ldlm_lock *lock;
2484 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2485 struct lov_oinfo *loi;
2486 __u32 *stripe = val;
2488 if (*vallen < sizeof(*stripe))
2489 GOTO(out, rc = -EFAULT);
2490 *vallen = sizeof(*stripe);
2492 /* XXX This is another one of those bits that will need to
2493 * change if we ever actually support nested LOVs. It uses
2494 * the lock's export to find out which stripe it is. */
2495 /* XXX - it's assumed all the locks for deleted OSTs have
2496 * been cancelled. Also, the export for deleted OSTs will
2497 * be NULL and won't match the lock's export. */
2498 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2499 loi = lsm->lsm_oinfo[i];
2500 if (!lov->lov_tgts[loi->loi_ost_idx])
2502 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2503 data->lock->l_conn_export &&
2504 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2509 LDLM_ERROR(data->lock, "lock on inode without such object");
2510 dump_lsm(D_ERROR, lsm);
2511 GOTO(out, rc = -ENXIO);
2512 } else if (KEY_IS(KEY_LAST_ID)) {
2513 struct obd_id_info *info = val;
2514 __u32 size = sizeof(obd_id);
2515 struct lov_tgt_desc *tgt;
2517 LASSERT(*vallen == sizeof(struct obd_id_info));
2518 tgt = lov->lov_tgts[info->idx];
2520 if (!tgt || !tgt->ltd_active)
2521 GOTO(out, rc = -ESRCH);
2523 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2525 } else if (KEY_IS(KEY_LOVDESC)) {
2526 struct lov_desc *desc_ret = val;
2527 *desc_ret = lov->desc;
2530 } else if (KEY_IS(KEY_FIEMAP)) {
2531 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2542 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2543 void *key, obd_count vallen, void *val,
2544 struct ptlrpc_request_set *set)
2546 struct obd_device *obddev = class_exp2obd(exp);
2547 struct lov_obd *lov = &obddev->u.lov;
2550 struct lov_tgt_desc *tgt;
2551 unsigned incr, check_uuid,
2552 do_inactive, no_set;
2553 unsigned next_id = 0, mds_con = 0;
2556 incr = check_uuid = do_inactive = no_set = 0;
2559 set = ptlrpc_prep_set();
2565 count = lov->desc.ld_tgt_count;
2567 if (KEY_IS(KEY_NEXT_ID)) {
2568 count = vallen / sizeof(struct obd_id_info);
2569 vallen = sizeof(obd_id);
2570 incr = sizeof(struct obd_id_info);
2573 } else if (KEY_IS(KEY_CHECKSUM)) {
2575 } else if (KEY_IS(KEY_UNLINKED)) {
2576 check_uuid = val ? 1 : 0;
2577 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2578 /* use defaults: do_inactive = incr = 0; */
2579 } else if (KEY_IS(KEY_MDS_CONN)) {
2583 for (i = 0; i < count; i++, val = (char *)val + incr) {
2585 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2587 tgt = lov->lov_tgts[i];
2589 /* OST was disconnected */
2590 if (!tgt || !tgt->ltd_exp)
2593 /* OST is inactive and we don't want inactive OSCs */
2594 if (!tgt->ltd_active && !do_inactive)
2598 struct mds_group_info *mgi;
2600 LASSERT(vallen == sizeof(*mgi));
2601 mgi = (struct mds_group_info *)val;
2603 /* Only want a specific OSC */
2604 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2608 err = obd_set_info_async(tgt->ltd_exp,
2609 keylen, key, sizeof(int),
2611 } else if (next_id) {
2612 err = obd_set_info_async(tgt->ltd_exp,
2613 keylen, key, vallen,
2614 ((struct obd_id_info*)val)->data, set);
2616 /* Only want a specific OSC */
2618 !obd_uuid_equals(val, &tgt->ltd_uuid))
2621 err = obd_set_info_async(tgt->ltd_exp,
2622 keylen, key, vallen, val, set);
2631 err = ptlrpc_set_wait(set);
2634 ptlrpc_set_destroy(set);
2639 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2640 struct lov_stripe_md *lsm)
2648 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2649 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2654 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2659 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2660 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2661 if (loi->loi_ar.ar_rc && !rc)
2662 rc = loi->loi_ar.ar_rc;
2663 loi->loi_ar.ar_rc = 0;
2667 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2670 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2671 int cmd, __u64 *offset)
2673 __u32 ssize = lsm->lsm_stripe_size;
2677 do_div(start, ssize);
2678 start = start * ssize;
2680 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2681 ", end "LPU64"\n", *offset, ssize, start,
2683 if (cmd == OBD_CALC_STRIPE_END) {
2684 *offset = start + ssize - 1;
2685 } else if (cmd == OBD_CALC_STRIPE_START) {
2696 struct lov_multi_wait {
2697 struct ldlm_lock *lock;
2703 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2704 struct lustre_handle *lockh)
2706 struct lov_lock_handles *lov_lockh = NULL;
2707 struct lustre_handle *lov_lockhp;
2708 struct lov_obd *lov;
2709 struct lov_oinfo *loi;
2710 struct lov_multi_wait *queues;
2714 ASSERT_LSM_MAGIC(lsm);
2716 if (!exp || !exp->exp_obd)
2719 LASSERT(lockh != NULL);
2720 if (lsm->lsm_stripe_count > 1) {
2721 lov_lockh = lov_handle2llh(lockh);
2722 if (lov_lockh == NULL) {
2723 CERROR("LOV: invalid lov lock handle %p\n", lockh);
2727 lov_lockhp = lov_lockh->llh_handles;
2732 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2734 GOTO(out, rc = -ENOMEM);
2736 lov = &exp->exp_obd->u.lov;
2737 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2738 i++, loi++, lov_lockhp++) {
2739 struct ldlm_lock *lock;
2740 struct obd_device *obd;
2742 lock = ldlm_handle2lock_long(lov_lockhp, 0);
2744 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2745 loi->loi_ost_idx, loi->loi_id);
2746 queues[i].completed = 1;
2750 queues[i].lock = lock;
2751 init_waitqueue_entry(&(queues[i].wait), current);
2752 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2754 obd = class_exp2obd(lock->l_conn_export);
2756 imp = obd->u.cli.cl_import;
2758 spin_lock(&imp->imp_lock);
2759 queues[i].generation = imp->imp_generation;
2760 spin_unlock(&imp->imp_lock);
2764 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2765 interrupted_completion_wait, &lwd);
2766 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2768 for (i = 0; i < lsm->lsm_stripe_count; i++)
2769 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2771 if (rc == -EINTR || rc == -ETIMEDOUT) {
2777 if (lov_lockh != NULL)
2778 lov_llh_put(lov_lockh);
2783 void lov_stripe_lock(struct lov_stripe_md *md)
2785 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2786 spin_lock(&md->lsm_lock);
2787 LASSERT(md->lsm_lock_owner == 0);
2788 md->lsm_lock_owner = cfs_curproc_pid();
2790 EXPORT_SYMBOL(lov_stripe_lock);
2792 void lov_stripe_unlock(struct lov_stripe_md *md)
2794 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2795 md->lsm_lock_owner = 0;
2796 spin_unlock(&md->lsm_lock);
2798 EXPORT_SYMBOL(lov_stripe_unlock);
2801 struct obd_ops lov_obd_ops = {
2802 .o_owner = THIS_MODULE,
2803 .o_setup = lov_setup,
2804 .o_precleanup = lov_precleanup,
2805 .o_cleanup = lov_cleanup,
2806 //.o_process_config = lov_process_config,
2807 .o_connect = lov_connect,
2808 .o_disconnect = lov_disconnect,
2809 .o_statfs = lov_statfs,
2810 .o_statfs_async = lov_statfs_async,
2811 .o_packmd = lov_packmd,
2812 .o_unpackmd = lov_unpackmd,
2813 .o_checkmd = lov_checkmd,
2814 .o_create = lov_create,
2815 .o_destroy = lov_destroy,
2816 .o_getattr = lov_getattr,
2817 .o_getattr_async = lov_getattr_async,
2818 .o_setattr = lov_setattr,
2819 .o_setattr_async = lov_setattr_async,
2821 .o_merge_lvb = lov_merge_lvb,
2822 .o_adjust_kms = lov_adjust_kms,
2823 .o_punch = lov_punch,
2825 .o_enqueue = lov_enqueue,
2826 .o_change_cbdata = lov_change_cbdata,
2827 .o_cancel = lov_cancel,
2828 .o_cancel_unused = lov_cancel_unused,
2829 .o_iocontrol = lov_iocontrol,
2830 .o_get_info = lov_get_info,
2831 .o_set_info_async = lov_set_info_async,
2832 .o_extent_calc = lov_extent_calc,
2833 .o_llog_init = lov_llog_init,
2834 .o_llog_finish = lov_llog_finish,
2835 .o_notify = lov_notify,
2836 .o_pool_new = lov_pool_new,
2837 .o_pool_rem = lov_pool_remove,
2838 .o_pool_add = lov_pool_add,
2839 .o_pool_del = lov_pool_del,
2842 static quota_interface_t *quota_interface;
2843 extern quota_interface_t lov_quota_interface;
2845 cfs_mem_cache_t *lov_oinfo_slab;
2847 extern struct lu_kmem_descr lov_caches[];
2849 int __init lov_init(void)
2851 struct lprocfs_static_vars lvars = { 0 };
2855 /* print an address of _any_ initialized kernel symbol from this
2856 * module, to allow debugging with gdb that doesn't support data
2857 * symbols from modules.*/
2858 CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2860 rc = lu_kmem_init(lov_caches);
2864 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2865 sizeof(struct lov_oinfo),
2866 0, SLAB_HWCACHE_ALIGN);
2867 if (lov_oinfo_slab == NULL) {
2868 lu_kmem_fini(lov_caches);
2871 lprocfs_lov_init_vars(&lvars);
2873 request_module("lquota");
2874 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2875 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2877 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2878 LUSTRE_LOV_NAME, &lov_device_type);
2881 if (quota_interface)
2882 PORTAL_SYMBOL_PUT(lov_quota_interface);
2883 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2885 lu_kmem_fini(lov_caches);
2892 static void /*__exit*/ lov_exit(void)
2896 lu_device_type_fini(&lov_device_type);
2897 lu_kmem_fini(lov_caches);
2899 if (quota_interface)
2900 PORTAL_SYMBOL_PUT(lov_quota_interface);
2902 class_unregister_type(LUSTRE_LOV_NAME);
2903 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2907 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2908 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2909 MODULE_LICENSE("GPL");
2911 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);