1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002-2006 Cluster File Systems, Inc.
5 * Author: Phil Schwan <phil@clusterfs.com>
6 * Peter Braam <braam@clusterfs.com>
7 * Mike Shaver <shaver@clusterfs.com>
8 * Nathan Rutman <nathan@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
30 # define EXPORT_SYMTAB
32 #define DEBUG_SUBSYSTEM S_LOV
34 #include <libcfs/libcfs.h>
36 #include <liblustre.h>
39 #include <obd_support.h>
40 #include <lustre_lib.h>
41 #include <lustre_net.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_dlm.h>
44 #include <lustre_mds.h>
45 #include <lustre_debug.h>
46 #include <obd_class.h>
49 #include <lprocfs_status.h>
50 #include <lustre_param.h>
52 #include "lov_internal.h"
55 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
56 Any function that expects lov_tgts to remain stationary must take a ref. */
57 void lov_getref(struct obd_device *obd)
59 struct lov_obd *lov = &obd->u.lov;
61 /* nobody gets through here until lov_putref is done */
62 mutex_down(&lov->lov_lock);
63 atomic_inc(&lov->lov_refcount);
64 mutex_up(&lov->lov_lock);
68 static void __lov_del_obd(struct obd_device *obd, __u32 index);
70 void lov_putref(struct obd_device *obd)
72 struct lov_obd *lov = &obd->u.lov;
73 mutex_down(&lov->lov_lock);
74 /* ok to dec to 0 more than once -- ltd_exp's will be null */
75 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
77 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
79 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
80 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap)
82 /* Disconnect and delete from list */
83 __lov_del_obd(obd, i);
87 mutex_up(&lov->lov_lock);
90 #define MAX_STRING_SIZE 128
91 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
92 struct obd_connect_data *data)
94 struct lov_obd *lov = &obd->u.lov;
95 struct obd_uuid tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
96 struct obd_device *tgt_obd;
97 struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
98 struct lustre_handle conn = {0, };
99 struct obd_import *imp;
102 cfs_proc_dir_entry_t *lov_proc_dir;
107 if (!lov->lov_tgts[index])
110 tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
114 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
117 if (!tgt_obd->obd_set_up) {
118 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
122 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
123 data->ocd_index = index;
126 * Divine LOV knows that OBDs under it are OSCs.
128 imp = tgt_obd->u.cli.cl_import;
131 tgt_obd->obd_no_recov = 0;
132 /* FIXME this is probably supposed to be
133 ptlrpc_set_import_active. Horrible naming. */
134 ptlrpc_activate_import(imp);
137 if (imp->imp_invalid) {
138 CERROR("not connecting OSC %s; administratively "
139 "disabled\n", obd_uuid2str(&tgt_uuid));
140 rc = obd_register_observer(tgt_obd, obd);
142 CERROR("Target %s register_observer error %d; "
143 "will not be able to reactivate\n",
144 obd_uuid2str(&tgt_uuid), rc);
149 rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data);
151 CERROR("Target %s connect error %d\n",
152 obd_uuid2str(&tgt_uuid), rc);
155 lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
156 if (!lov->lov_tgts[index]->ltd_exp) {
157 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
161 rc = obd_register_observer(tgt_obd, obd);
163 CERROR("Target %s register_observer error %d\n",
164 obd_uuid2str(&tgt_uuid), rc);
165 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
166 lov->lov_tgts[index]->ltd_exp = NULL;
170 lov->lov_tgts[index]->ltd_reap = 0;
172 lov->lov_tgts[index]->ltd_active = 1;
173 lov->desc.ld_active_tgt_count++;
174 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
176 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
177 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
180 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
182 struct obd_device *osc_obd = class_conn2obd(&conn);
183 cfs_proc_dir_entry_t *osc_symlink;
184 char name[MAX_STRING_SIZE];
186 LASSERT(osc_obd != NULL);
187 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
188 LASSERT(osc_obd->obd_type->typ_name != NULL);
189 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
190 osc_obd->obd_type->typ_name,
192 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
194 if (osc_symlink == NULL) {
195 CERROR("could not register LOV target "
196 "/proc/fs/lustre/%s/%s/target_obds/%s.",
197 obd->obd_type->typ_name, obd->obd_name,
199 lprocfs_remove(&lov_proc_dir);
204 rc = qos_add_tgt(obd, index);
206 CERROR("qos_add_tgt failed %d\n", rc);
211 static int lov_connect(const struct lu_env *env,
212 struct lustre_handle *conn, struct obd_device *obd,
213 struct obd_uuid *cluuid, struct obd_connect_data *data)
215 struct lov_obd *lov = &obd->u.lov;
216 struct lov_tgt_desc *tgt;
220 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
222 rc = class_connect(conn, obd, cluuid);
226 /* Why should there ever be more than 1 connect? */
228 LASSERT(lov->lov_connects == 1);
230 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
232 lov->lov_ocd = *data;
235 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
236 tgt = lov->lov_tgts[i];
237 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
239 /* Flags will be lowest common denominator */
240 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
242 CERROR("%s: lov connect tgt %d failed: %d\n",
243 obd->obd_name, i, rc);
252 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
254 cfs_proc_dir_entry_t *lov_proc_dir;
255 struct lov_obd *lov = &obd->u.lov;
256 struct obd_device *osc_obd =
257 class_exp2obd(lov->lov_tgts[index]->ltd_exp);
262 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
263 obd->obd_name, osc_obd->obd_name);
265 if (lov->lov_tgts[index]->ltd_active) {
266 lov->lov_tgts[index]->ltd_active = 0;
267 lov->desc.ld_active_tgt_count--;
268 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
271 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
273 cfs_proc_dir_entry_t *osc_symlink;
275 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
277 lprocfs_remove(&osc_symlink);
279 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
280 obd->obd_type->typ_name, obd->obd_name,
286 /* Pass it on to our clients.
287 * XXX This should be an argument to disconnect,
288 * XXX not a back-door flag on the OBD. Ah well.
290 osc_obd->obd_force = obd->obd_force;
291 osc_obd->obd_fail = obd->obd_fail;
292 osc_obd->obd_no_recov = obd->obd_no_recov;
295 obd_register_observer(osc_obd, NULL);
297 rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
299 CERROR("Target %s disconnect error %d\n",
300 lov_uuid2str(lov, index), rc);
304 qos_del_tgt(obd, index);
306 lov->lov_tgts[index]->ltd_exp = NULL;
310 static int lov_del_target(struct obd_device *obd, __u32 index,
311 struct obd_uuid *uuidp, int gen);
313 static int lov_disconnect(struct obd_export *exp)
315 struct obd_device *obd = class_exp2obd(exp);
316 struct lov_obd *lov = &obd->u.lov;
323 /* Only disconnect the underlying layers on the final disconnect. */
325 if (lov->lov_connects != 0) {
326 /* why should there be more than 1 connect? */
327 CERROR("disconnect #%d\n", lov->lov_connects);
331 /* Let's hold another reference so lov_del_obd doesn't spin through
334 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
335 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
336 /* Disconnection is the last we know about an obd */
337 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
343 rc = class_disconnect(exp); /* bz 9811 */
349 * -EINVAL : UUID can't be found in the LOV's target list
350 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
351 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
353 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
356 struct lov_obd *lov = &obd->u.lov;
357 struct lov_tgt_desc *tgt;
361 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
362 lov, uuid->uuid, activate);
365 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
366 tgt = lov->lov_tgts[i];
367 if (!tgt || !tgt->ltd_exp)
370 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
371 i, obd_uuid2str(&tgt->ltd_uuid),
372 tgt->ltd_exp->exp_handle.h_cookie);
373 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
377 if (i == lov->desc.ld_tgt_count)
378 GOTO(out, rc = -EINVAL);
380 if (lov->lov_tgts[i]->ltd_active == activate) {
381 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
382 activate ? "" : "in");
386 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
387 activate ? "" : "in");
389 lov->lov_tgts[i]->ltd_active = activate;
392 lov->desc.ld_active_tgt_count++;
393 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
395 lov->desc.ld_active_tgt_count--;
396 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
398 /* remove any old qos penalty */
399 lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
406 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
407 enum obd_notify_event ev, void *data)
412 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
413 struct obd_uuid *uuid;
417 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
418 CERROR("unexpected notification of %s %s!\n",
419 watched->obd_type->typ_name,
423 uuid = &watched->u.cli.cl_target_uuid;
425 /* Set OSC as active before notifying the observer, so the
426 * observer can use the OSC normally.
428 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
430 CERROR("%sactivation of %s failed: %d\n",
431 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
432 obd_uuid2str(uuid), rc);
437 /* Pass the notification up the chain. */
439 rc = obd_notify_observer(obd, watched, ev, data);
441 /* NULL watched means all osc's in the lov (only for syncs) */
442 struct lov_obd *lov = &obd->u.lov;
443 struct obd_device *tgt_obd;
446 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
447 if (!lov->lov_tgts[i])
449 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
450 rc = obd_notify_observer(obd, tgt_obd, ev, data);
452 CERROR("%s: notify %s of %s failed %d\n",
454 obd->obd_observer->obd_name,
455 tgt_obd->obd_name, rc);
465 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
466 __u32 index, int gen, int active)
468 struct lov_obd *lov = &obd->u.lov;
469 struct lov_tgt_desc *tgt;
473 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
474 uuidp->uuid, index, gen, active);
477 CERROR("request to add OBD %s with invalid generation: %d\n",
482 mutex_down(&lov->lov_lock);
484 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
485 tgt = lov->lov_tgts[index];
486 CERROR("UUID %s already assigned at LOV target index %d\n",
487 obd_uuid2str(&tgt->ltd_uuid), index);
488 mutex_up(&lov->lov_lock);
492 if (index >= lov->lov_tgt_size) {
493 /* We need to reallocate the lov target array. */
494 struct lov_tgt_desc **newtgts, **old = NULL;
495 __u32 newsize, oldsize = 0;
497 newsize = max(lov->lov_tgt_size, (__u32)2);
498 while (newsize < index + 1)
499 newsize = newsize << 1;
500 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
501 if (newtgts == NULL) {
502 mutex_up(&lov->lov_lock);
506 if (lov->lov_tgt_size) {
507 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
510 oldsize = lov->lov_tgt_size;
513 lov->lov_tgts = newtgts;
514 lov->lov_tgt_size = newsize;
519 OBD_FREE(old, sizeof(*old) * oldsize);
521 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
522 lov->lov_tgts, lov->lov_tgt_size);
528 mutex_up(&lov->lov_lock);
532 memset(tgt, 0, sizeof(*tgt));
533 tgt->ltd_uuid = *uuidp;
534 /* XXX - add a sanity check on the generation number. */
536 tgt->ltd_index = index;
537 tgt->ltd_activate = active;
538 lov->lov_tgts[index] = tgt;
539 if (index >= lov->desc.ld_tgt_count)
540 lov->desc.ld_tgt_count = index + 1;
541 mutex_up(&lov->lov_lock);
543 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
544 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
546 if (lov->lov_connects == 0) {
547 /* lov_connect hasn't been called yet. We'll do the
548 lov_connect_obd on this target when that fn first runs,
549 because we don't know the connect flags yet. */
555 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
559 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
560 active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
565 CERROR("add failed (%d), deleting %s\n", rc,
566 obd_uuid2str(&tgt->ltd_uuid));
567 lov_del_target(obd, index, 0, 0);
573 /* Schedule a target for deletion */
574 static int lov_del_target(struct obd_device *obd, __u32 index,
575 struct obd_uuid *uuidp, int gen)
577 struct lov_obd *lov = &obd->u.lov;
578 int count = lov->desc.ld_tgt_count;
582 if (index >= count) {
583 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
590 if (!lov->lov_tgts[index]) {
591 CERROR("LOV target at index %d is not setup.\n", index);
592 GOTO(out, rc = -EINVAL);
595 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
596 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
597 lov_uuid2str(lov, index), index,
598 obd_uuid2str(uuidp));
599 GOTO(out, rc = -EINVAL);
602 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
603 lov_uuid2str(lov, index), index,
604 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
605 lov->lov_tgts[index]->ltd_active);
607 lov->lov_tgts[index]->ltd_reap = 1;
608 lov->lov_death_row++;
609 /* we really delete it from lov_putref */
616 /* We are holding lov_lock */
617 static void __lov_del_obd(struct obd_device *obd, __u32 index)
619 struct lov_obd *lov = &obd->u.lov;
620 struct obd_device *osc_obd;
621 struct lov_tgt_desc *tgt = lov->lov_tgts[index];
624 LASSERT(tgt->ltd_reap);
626 osc_obd = class_exp2obd(tgt->ltd_exp);
628 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
629 lov_uuid2str(lov, index),
630 osc_obd ? osc_obd->obd_name : "<no obd>");
633 lov_disconnect_obd(obd, index);
635 /* XXX - right now there is a dependency on ld_tgt_count being the
636 * maximum tgt index for computing the mds_max_easize. So we can't
639 lov->lov_tgts[index] = NULL;
642 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
643 do it ourselves. And we can't do it from lov_cleanup,
644 because we just lost our only reference to it. */
646 class_manual_cleanup(osc_obd);
649 void lov_fix_desc(struct lov_desc *desc)
651 if (desc->ld_default_stripe_size < PTLRPC_MAX_BRW_SIZE) {
652 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
653 PTLRPC_MAX_BRW_SIZE);
654 desc->ld_default_stripe_size = PTLRPC_MAX_BRW_SIZE;
655 } else if (desc->ld_default_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
656 desc->ld_default_stripe_size &= ~(LOV_MIN_STRIPE_SIZE - 1);
657 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
659 desc->ld_default_stripe_size,LOV_MIN_STRIPE_SIZE);
662 if (desc->ld_default_stripe_count == 0)
663 desc->ld_default_stripe_count = 1;
665 /* from lov_setstripe */
666 if ((desc->ld_pattern != 0) &&
667 (desc->ld_pattern != LOV_PATTERN_RAID0)) {
668 LCONSOLE_WARN("Unknown stripe pattern: %#x\n",desc->ld_pattern);
669 desc->ld_pattern = 0;
672 if (desc->ld_qos_maxage == 0)
673 desc->ld_qos_maxage = QOS_DEFAULT_MAXAGE;
676 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
678 struct lprocfs_static_vars lvars;
679 struct lov_desc *desc;
680 struct lov_obd *lov = &obd->u.lov;
684 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
685 CERROR("LOV setup requires a descriptor\n");
689 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
691 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
692 CERROR("descriptor size wrong: %d > %d\n",
693 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
697 if (desc->ld_magic != LOV_DESC_MAGIC) {
698 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
699 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
700 obd->obd_name, desc);
701 lustre_swab_lov_desc(desc);
703 CERROR("%s: Bad lov desc magic: %#x\n",
704 obd->obd_name, desc->ld_magic);
711 /* Because of 64-bit divide/mod operations only work with a 32-bit
712 * divisor in a 32-bit kernel, we cannot support a stripe width
713 * of 4GB or larger on 32-bit CPUs. */
714 count = desc->ld_default_stripe_count;
715 if ((count > 0 ? count : desc->ld_tgt_count) *
716 desc->ld_default_stripe_size > 0xffffffff) {
717 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
718 desc->ld_default_stripe_size, count);
722 desc->ld_active_tgt_count = 0;
724 lov->lov_tgt_size = 0;
725 sema_init(&lov->lov_lock, 1);
726 atomic_set(&lov->lov_refcount, 0);
727 INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
728 init_rwsem(&lov->lov_qos.lq_rw_sem);
729 lov->lov_qos.lq_dirty = 1;
730 lov->lov_qos.lq_dirty_rr = 1;
731 lov->lov_qos.lq_reset = 1;
732 /* Default priority is toward free space balance */
733 lov->lov_qos.lq_prio_free = 232;
735 lprocfs_init_vars(lov, &lvars);
736 lprocfs_obd_setup(obd, lvars.obd_vars);
739 cfs_proc_dir_entry_t *entry;
741 entry = create_proc_entry("target_obd", 0444,
742 obd->obd_proc_entry);
744 entry->proc_fops = &lov_proc_target_fops;
753 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
759 case OBD_CLEANUP_EARLY: {
760 struct lov_obd *lov = &obd->u.lov;
762 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
763 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
765 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
770 case OBD_CLEANUP_EXPORTS:
772 case OBD_CLEANUP_SELF_EXP:
773 rc = obd_llog_finish(obd, 0);
775 CERROR("failed to cleanup llogging subsystems\n");
777 case OBD_CLEANUP_OBD:
783 static int lov_cleanup(struct obd_device *obd)
785 struct lov_obd *lov = &obd->u.lov;
787 lprocfs_obd_cleanup(obd);
790 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
791 if (lov->lov_tgts[i]) {
792 /* Inactive targets may never have connected */
793 if (lov->lov_tgts[i]->ltd_active ||
794 atomic_read(&lov->lov_refcount))
795 /* We should never get here - these
796 should have been removed in the
798 CERROR("lov tgt %d not cleaned!"
799 " deathrow=%d, lovrc=%d\n",
800 i, lov->lov_death_row,
801 atomic_read(&lov->lov_refcount));
802 lov_del_target(obd, i, 0, 0);
805 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
807 lov->lov_tgt_size = 0;
810 if (lov->lov_qos.lq_rr_size)
811 OBD_FREE(lov->lov_qos.lq_rr_array, lov->lov_qos.lq_rr_size);
816 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
818 struct lustre_cfg *lcfg = buf;
819 struct obd_uuid obd_uuid;
824 switch(cmd = lcfg->lcfg_command) {
825 case LCFG_LOV_ADD_OBD:
826 case LCFG_LOV_ADD_INA:
827 case LCFG_LOV_DEL_OBD: {
830 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
831 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
832 GOTO(out, rc = -EINVAL);
834 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
836 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
837 GOTO(out, rc = -EINVAL);
838 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
839 GOTO(out, rc = -EINVAL);
840 if (cmd == LCFG_LOV_ADD_OBD)
841 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
842 else if (cmd == LCFG_LOV_ADD_INA)
843 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
845 rc = lov_del_target(obd, index, &obd_uuid, gen);
849 struct lprocfs_static_vars lvars;
850 struct lov_desc *desc = &(obd->u.lov.desc);
853 GOTO(out, rc = -EINVAL);
855 lprocfs_init_vars(lov, &lvars);
857 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
862 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
863 GOTO(out, rc = -EINVAL);
872 #define log2(n) ffz(~(n))
875 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
876 struct lov_stripe_md **ea,
877 struct obd_trans_info *oti)
881 struct obd_uuid *ost_uuid = NULL;
885 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
886 src_oa->o_flags == OBD_FL_DELORPHAN);
888 lov = &export->exp_obd->u.lov;
894 if (src_oa->o_valid & OBD_MD_FLINLINE) {
895 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
896 CDEBUG(D_HA, "clearing orphans only for %s\n",
900 lov_getref(export->exp_obd);
901 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
902 struct lov_stripe_md obj_md;
903 struct lov_stripe_md *obj_mdp = &obj_md;
904 struct lov_tgt_desc *tgt;
907 tgt = lov->lov_tgts[i];
911 /* if called for a specific target, we don't
912 care if it is not active. */
913 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
914 CDEBUG(D_HA, "lov idx %d inactive\n", i);
918 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
921 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
922 obd_uuid2str(ost_uuid));
924 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
926 LASSERT(lov->lov_tgts[i]->ltd_exp);
927 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
928 err = obd_create(lov->lov_tgts[i]->ltd_exp,
929 tmp_oa, &obj_mdp, oti);
931 /* This export will be disabled until it is recovered,
932 and then orphan recovery will be completed. */
933 CERROR("error in orphan recovery on OST idx %d/%d: "
934 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
939 lov_putref(export->exp_obd);
945 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
946 struct lov_stripe_md **ea, struct obd_trans_info *oti)
948 struct lov_stripe_md *obj_mdp, *lsm;
949 struct lov_obd *lov = &exp->exp_obd->u.lov;
954 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
955 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
957 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
961 ost_idx = src_oa->o_nlink;
964 GOTO(out, rc = -EINVAL);
965 if (ost_idx >= lov->desc.ld_tgt_count ||
966 !lov->lov_tgts[ost_idx])
967 GOTO(out, rc = -EINVAL);
969 for (i = 0; i < lsm->lsm_stripe_count; i++) {
970 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
971 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
972 GOTO(out, rc = -EINVAL);
976 if (i == lsm->lsm_stripe_count)
977 GOTO(out, rc = -EINVAL);
979 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
981 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
985 /* the LOV expects oa->o_id to be set to the LOV object id */
986 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
987 struct lov_stripe_md **ea, struct obd_trans_info *oti)
990 struct obd_info oinfo;
991 struct lov_request_set *set = NULL;
992 struct lov_request *req;
993 struct obd_statfs osfs;
1002 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1003 src_oa->o_flags == OBD_FL_DELORPHAN) {
1004 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1008 lov = &exp->exp_obd->u.lov;
1009 if (!lov->desc.ld_active_tgt_count)
1012 /* Recreate a specific object id at the given OST index */
1013 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1014 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1015 rc = lov_recreate(exp, src_oa, ea, oti);
1019 maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1020 obd_statfs_rqset(exp->exp_obd, &osfs, maxage);
1022 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1026 list_for_each_entry(req, &set->set_list, rq_link) {
1027 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1028 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1029 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1030 lov_update_create_set(set, req, rc);
1032 rc = lov_fini_create_set(set, ea);
1036 #define ASSERT_LSM_MAGIC(lsmp) \
1038 LASSERT((lsmp) != NULL); \
1039 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC || \
1040 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1041 (lsmp), (lsmp)->lsm_magic); \
1044 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1045 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1046 struct obd_export *md_exp)
1048 struct lov_request_set *set;
1049 struct obd_info oinfo;
1050 struct lov_request *req;
1051 struct list_head *pos;
1052 struct lov_obd *lov;
1056 ASSERT_LSM_MAGIC(lsm);
1058 if (!exp || !exp->exp_obd)
1061 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1063 LASSERT(oti->oti_logcookies);
1066 lov = &exp->exp_obd->u.lov;
1067 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1071 list_for_each (pos, &set->set_list) {
1073 req = list_entry(pos, struct lov_request, rq_link);
1075 if (oa->o_valid & OBD_MD_FLCOOKIE)
1076 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1078 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1079 req->rq_oi.oi_oa, NULL, oti, NULL);
1080 err = lov_update_common_set(set, req, err);
1082 CERROR("error: destroying objid "LPX64" subobj "
1083 LPX64" on OST idx %d: rc = %d\n",
1084 oa->o_id, req->rq_oi.oi_oa->o_id,
1092 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1093 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1095 err = lov_fini_destroy_set(set);
1096 RETURN(rc ? rc : err);
1099 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1101 struct lov_request_set *set;
1102 struct lov_request *req;
1103 struct list_head *pos;
1104 struct lov_obd *lov;
1105 int err = 0, rc = 0;
1109 ASSERT_LSM_MAGIC(oinfo->oi_md);
1111 if (!exp || !exp->exp_obd)
1114 lov = &exp->exp_obd->u.lov;
1116 rc = lov_prep_getattr_set(exp, oinfo, &set);
1120 list_for_each (pos, &set->set_list) {
1121 req = list_entry(pos, struct lov_request, rq_link);
1123 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1124 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1125 req->rq_oi.oi_oa->o_id, req->rq_idx);
1127 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1129 err = lov_update_common_set(set, req, rc);
1131 CERROR("error: getattr objid "LPX64" subobj "
1132 LPX64" on OST idx %d: rc = %d\n",
1133 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1139 rc = lov_fini_getattr_set(set);
1145 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1148 struct lov_request_set *lovset = (struct lov_request_set *)data;
1152 /* don't do attribute merge if this aysnc op failed */
1154 lovset->set_completes = 0;
1155 err = lov_fini_getattr_set(lovset);
1156 RETURN(rc ? rc : err);
1159 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1160 struct ptlrpc_request_set *rqset)
1162 struct lov_request_set *lovset;
1163 struct lov_obd *lov;
1164 struct list_head *pos;
1165 struct lov_request *req;
1170 ASSERT_LSM_MAGIC(oinfo->oi_md);
1172 if (!exp || !exp->exp_obd)
1175 lov = &exp->exp_obd->u.lov;
1177 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1181 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1182 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1183 oinfo->oi_md->lsm_stripe_size);
1185 list_for_each (pos, &lovset->set_list) {
1186 req = list_entry(pos, struct lov_request, rq_link);
1188 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1189 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1190 req->rq_oi.oi_oa->o_id, req->rq_idx);
1191 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1192 &req->rq_oi, rqset);
1194 CERROR("error: getattr objid "LPX64" subobj "
1195 LPX64" on OST idx %d: rc = %d\n",
1196 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1202 if (!list_empty(&rqset->set_requests)) {
1204 LASSERT (rqset->set_interpret == NULL);
1205 rqset->set_interpret = lov_getattr_interpret;
1206 rqset->set_arg = (void *)lovset;
1211 lovset->set_completes = 0;
1212 err = lov_fini_getattr_set(lovset);
1213 RETURN(rc ? rc : err);
1216 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1217 struct obd_trans_info *oti)
1219 struct lov_request_set *set;
1220 struct lov_obd *lov;
1221 struct list_head *pos;
1222 struct lov_request *req;
1223 int err = 0, rc = 0;
1227 ASSERT_LSM_MAGIC(oinfo->oi_md);
1229 if (!exp || !exp->exp_obd)
1232 /* for now, we only expect the following updates here */
1233 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1234 OBD_MD_FLMODE | OBD_MD_FLATIME |
1235 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1236 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1237 OBD_MD_FLGROUP | OBD_MD_FLUID |
1238 OBD_MD_FLGID | OBD_MD_FLINLINE |
1239 OBD_MD_FLFID | OBD_MD_FLGENER)));
1240 lov = &exp->exp_obd->u.lov;
1241 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1245 list_for_each (pos, &set->set_list) {
1246 req = list_entry(pos, struct lov_request, rq_link);
1248 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1250 err = lov_update_setattr_set(set, req, rc);
1252 CERROR("error: setattr objid "LPX64" subobj "
1253 LPX64" on OST idx %d: rc = %d\n",
1254 set->set_oi->oi_oa->o_id,
1255 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1260 err = lov_fini_setattr_set(set);
1266 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1269 struct lov_request_set *lovset = (struct lov_request_set *)data;
1274 lovset->set_completes = 0;
1275 err = lov_fini_setattr_set(lovset);
1276 RETURN(rc ? rc : err);
1279 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1280 needed. Otherwise, a client is waiting for responses. */
1281 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1282 struct obd_trans_info *oti,
1283 struct ptlrpc_request_set *rqset)
1285 struct lov_request_set *set;
1286 struct lov_request *req;
1287 struct list_head *pos;
1288 struct lov_obd *lov;
1293 ASSERT_LSM_MAGIC(oinfo->oi_md);
1294 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1296 LASSERT(oti->oti_logcookies);
1299 if (!exp || !exp->exp_obd)
1302 lov = &exp->exp_obd->u.lov;
1303 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1307 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1308 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1309 oinfo->oi_md->lsm_stripe_size);
1311 list_for_each (pos, &set->set_list) {
1312 req = list_entry(pos, struct lov_request, rq_link);
1314 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1315 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
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);
1321 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1322 &req->rq_oi, oti, rqset);
1324 CERROR("error: setattr objid "LPX64" subobj "
1325 LPX64" on OST idx %d: rc = %d\n",
1326 set->set_oi->oi_oa->o_id,
1327 req->rq_oi.oi_oa->o_id,
1333 /* If we are not waiting for responses on async requests, return. */
1334 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1337 set->set_completes = 0;
1338 err = lov_fini_setattr_set(set);
1339 RETURN(rc ? rc : err);
1342 LASSERT(rqset->set_interpret == NULL);
1343 rqset->set_interpret = lov_setattr_interpret;
1344 rqset->set_arg = (void *)set;
1349 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1352 struct lov_request_set *lovset = (struct lov_request_set *)data;
1357 lovset->set_completes = 0;
1358 err = lov_fini_punch_set(lovset);
1359 RETURN(rc ? rc : err);
1362 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1363 * we can send this 'punch' to just the authoritative node and the nodes
1364 * that the punch will affect. */
1365 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1366 struct obd_trans_info *oti,
1367 struct ptlrpc_request_set *rqset)
1369 struct lov_request_set *set;
1370 struct lov_obd *lov;
1371 struct list_head *pos;
1372 struct lov_request *req;
1377 ASSERT_LSM_MAGIC(oinfo->oi_md);
1379 if (!exp || !exp->exp_obd)
1382 lov = &exp->exp_obd->u.lov;
1383 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1387 list_for_each (pos, &set->set_list) {
1388 req = list_entry(pos, struct lov_request, rq_link);
1390 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1391 &req->rq_oi, NULL, rqset);
1393 CERROR("error: punch objid "LPX64" subobj "LPX64
1394 " on OST idx %d: rc = %d\n",
1395 set->set_oi->oi_oa->o_id,
1396 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1401 if (rc || list_empty(&rqset->set_requests)) {
1403 err = lov_fini_punch_set(set);
1404 RETURN(rc ? rc : err);
1407 LASSERT(rqset->set_interpret == NULL);
1408 rqset->set_interpret = lov_punch_interpret;
1409 rqset->set_arg = (void *)set;
1414 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1415 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1418 struct lov_request_set *set;
1419 struct obd_info oinfo;
1420 struct lov_obd *lov;
1421 struct list_head *pos;
1422 struct lov_request *req;
1423 int err = 0, rc = 0;
1426 ASSERT_LSM_MAGIC(lsm);
1431 lov = &exp->exp_obd->u.lov;
1432 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1436 list_for_each (pos, &set->set_list) {
1437 req = list_entry(pos, struct lov_request, rq_link);
1439 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1440 req->rq_oi.oi_oa, NULL,
1441 req->rq_oi.oi_policy.l_extent.start,
1442 req->rq_oi.oi_policy.l_extent.end, capa);
1443 err = lov_update_common_set(set, req, rc);
1445 CERROR("error: fsync objid "LPX64" subobj "LPX64
1446 " on OST idx %d: rc = %d\n",
1447 set->set_oi->oi_oa->o_id,
1448 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1453 err = lov_fini_sync_set(set);
1459 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1460 obd_count oa_bufs, struct brw_page *pga)
1462 struct obd_info oinfo = { { { 0 } } };
1465 oinfo.oi_oa = lov_oinfo->oi_oa;
1467 /* The caller just wants to know if there's a chance that this
1468 * I/O can succeed */
1469 for (i = 0; i < oa_bufs; i++) {
1470 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1471 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1474 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1475 pga[i].off + pga[i].count,
1479 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1480 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1484 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1492 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1493 obd_count oa_bufs, struct brw_page *pga,
1494 struct obd_trans_info *oti)
1496 struct lov_request_set *set;
1497 struct lov_request *req;
1498 struct list_head *pos;
1499 struct lov_obd *lov = &exp->exp_obd->u.lov;
1503 ASSERT_LSM_MAGIC(oinfo->oi_md);
1505 if (cmd == OBD_BRW_CHECK) {
1506 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1510 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1514 list_for_each (pos, &set->set_list) {
1515 struct obd_export *sub_exp;
1516 struct brw_page *sub_pga;
1517 req = list_entry(pos, struct lov_request, rq_link);
1519 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1520 sub_pga = set->set_pga + req->rq_pgaidx;
1521 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1525 lov_update_common_set(set, req, rc);
1528 err = lov_fini_brw_set(set);
1534 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1537 struct lov_request_set *lovset = (struct lov_request_set *)data;
1541 lovset->set_completes = 0;
1542 lov_fini_brw_set(lovset);
1544 rc = lov_fini_brw_set(lovset);
1550 static int lov_brw_async(int cmd, struct obd_export *exp,
1551 struct obd_info *oinfo, obd_count oa_bufs,
1552 struct brw_page *pga, struct obd_trans_info *oti,
1553 struct ptlrpc_request_set *set)
1555 struct lov_request_set *lovset;
1556 struct lov_request *req;
1557 struct list_head *pos;
1558 struct lov_obd *lov = &exp->exp_obd->u.lov;
1563 ASSERT_LSM_MAGIC(oinfo->oi_md);
1565 if (cmd == OBD_BRW_CHECK) {
1566 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1570 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1574 list_for_each (pos, &lovset->set_list) {
1575 struct obd_export *sub_exp;
1576 struct brw_page *sub_pga;
1577 req = list_entry(pos, struct lov_request, rq_link);
1579 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1580 sub_pga = lovset->set_pga + req->rq_pgaidx;
1581 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1585 lov_update_common_set(lovset, req, rc);
1588 LASSERT(set->set_interpret == NULL);
1589 set->set_interpret = (set_interpreter_func)lov_brw_interpret;
1590 set->set_arg = (void *)lovset;
1594 lov_fini_brw_set(lovset);
1598 static int lov_ap_make_ready(void *data, int cmd)
1600 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1602 return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1605 static int lov_ap_refresh_count(void *data, int cmd)
1607 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1609 return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1613 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1615 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1617 lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1618 /* XXX woah, shouldn't we be altering more here? size? */
1619 oa->o_id = lap->lap_loi_id;
1620 oa->o_gr = lap->lap_loi_gr;
1621 oa->o_valid |= OBD_MD_FLGROUP;
1622 oa->o_stripe_idx = lap->lap_stripe;
1625 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1628 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1630 lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1633 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1635 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1637 /* in a raid1 regime this would down a count of many ios
1638 * in flight, onl calling the caller_ops completion when all
1639 * the raid1 ios are complete */
1640 rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1644 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1646 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1647 return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1650 static struct obd_async_page_ops lov_async_page_ops = {
1651 .ap_make_ready = lov_ap_make_ready,
1652 .ap_refresh_count = lov_ap_refresh_count,
1653 .ap_fill_obdo = lov_ap_fill_obdo,
1654 .ap_update_obdo = lov_ap_update_obdo,
1655 .ap_completion = lov_ap_completion,
1656 .ap_lookup_capa = lov_ap_lookup_capa,
1659 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1660 struct lov_oinfo *loi, cfs_page_t *page,
1661 obd_off offset, struct obd_async_page_ops *ops,
1662 void *data, void **res)
1664 struct lov_obd *lov = &exp->exp_obd->u.lov;
1665 struct lov_async_page *lap;
1671 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1672 Only because of this layering limitation will a client
1673 mount with no osts fail */
1674 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1675 !lov->lov_tgts[i]->ltd_exp) {
1677 if (i >= lov->desc.ld_tgt_count)
1680 rc = size_round(sizeof(*lap)) +
1681 obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1682 NULL, NULL, 0, NULL, NULL, NULL);
1685 ASSERT_LSM_MAGIC(lsm);
1686 LASSERT(loi == NULL);
1689 lap->lap_magic = LOV_AP_MAGIC;
1690 lap->lap_caller_ops = ops;
1691 lap->lap_caller_data = data;
1693 /* for now only raid 0 which passes through */
1694 lap->lap_stripe = lov_stripe_number(lsm, offset);
1695 lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1696 loi = lsm->lsm_oinfo[lap->lap_stripe];
1698 /* so the callback doesn't need the lsm */
1699 lap->lap_loi_id = loi->loi_id;
1700 lap->lap_loi_gr = lsm->lsm_object_gr;
1701 LASSERT(lsm->lsm_object_gr > 0);
1703 lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1705 rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1706 lsm, loi, page, lap->lap_sub_offset,
1707 &lov_async_page_ops, lap,
1708 &lap->lap_sub_cookie);
1711 CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1712 lap->lap_sub_cookie, offset);
1716 static int lov_queue_async_io(struct obd_export *exp,
1717 struct lov_stripe_md *lsm,
1718 struct lov_oinfo *loi, void *cookie,
1719 int cmd, obd_off off, int count,
1720 obd_flag brw_flags, obd_flag async_flags)
1722 struct lov_obd *lov = &exp->exp_obd->u.lov;
1723 struct lov_async_page *lap;
1726 LASSERT(loi == NULL);
1728 ASSERT_LSM_MAGIC(lsm);
1730 lap = LAP_FROM_COOKIE(cookie);
1732 loi = lsm->lsm_oinfo[lap->lap_stripe];
1734 rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1735 loi, lap->lap_sub_cookie, cmd, off, count,
1736 brw_flags, async_flags);
1740 static int lov_set_async_flags(struct obd_export *exp,
1741 struct lov_stripe_md *lsm,
1742 struct lov_oinfo *loi, void *cookie,
1743 obd_flag async_flags)
1745 struct lov_obd *lov = &exp->exp_obd->u.lov;
1746 struct lov_async_page *lap;
1749 LASSERT(loi == NULL);
1751 ASSERT_LSM_MAGIC(lsm);
1753 lap = LAP_FROM_COOKIE(cookie);
1755 loi = lsm->lsm_oinfo[lap->lap_stripe];
1757 rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1758 lsm, loi, lap->lap_sub_cookie, async_flags);
1762 static int lov_queue_group_io(struct obd_export *exp,
1763 struct lov_stripe_md *lsm,
1764 struct lov_oinfo *loi,
1765 struct obd_io_group *oig, void *cookie,
1766 int cmd, obd_off off, int count,
1767 obd_flag brw_flags, obd_flag async_flags)
1769 struct lov_obd *lov = &exp->exp_obd->u.lov;
1770 struct lov_async_page *lap;
1773 LASSERT(loi == NULL);
1775 ASSERT_LSM_MAGIC(lsm);
1777 lap = LAP_FROM_COOKIE(cookie);
1779 loi = lsm->lsm_oinfo[lap->lap_stripe];
1781 rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1782 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1783 brw_flags, async_flags);
1787 /* this isn't exactly optimal. we may have queued sync io in oscs on
1788 * all stripes, but we don't record that fact at queue time. so we
1789 * trigger sync io on all stripes. */
1790 static int lov_trigger_group_io(struct obd_export *exp,
1791 struct lov_stripe_md *lsm,
1792 struct lov_oinfo *loi,
1793 struct obd_io_group *oig)
1795 struct lov_obd *lov = &exp->exp_obd->u.lov;
1798 LASSERT(loi == NULL);
1800 ASSERT_LSM_MAGIC(lsm);
1802 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1803 loi = lsm->lsm_oinfo[i];
1804 if (!lov->lov_tgts[loi->loi_ost_idx] ||
1805 !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1806 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1810 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1812 if (rc == 0 && err != 0)
1818 static int lov_teardown_async_page(struct obd_export *exp,
1819 struct lov_stripe_md *lsm,
1820 struct lov_oinfo *loi, void *cookie)
1822 struct lov_obd *lov = &exp->exp_obd->u.lov;
1823 struct lov_async_page *lap;
1826 LASSERT(loi == NULL);
1828 ASSERT_LSM_MAGIC(lsm);
1830 lap = LAP_FROM_COOKIE(cookie);
1832 loi = lsm->lsm_oinfo[lap->lap_stripe];
1834 rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1835 lsm, loi, lap->lap_sub_cookie);
1837 CERROR("unable to teardown sub cookie %p: %d\n",
1838 lap->lap_sub_cookie, rc);
1844 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1847 struct lov_request_set *lovset = (struct lov_request_set *)data;
1849 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1853 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1854 struct ldlm_enqueue_info *einfo,
1855 struct ptlrpc_request_set *rqset)
1857 ldlm_mode_t mode = einfo->ei_mode;
1858 struct lov_request_set *set;
1859 struct lov_request *req;
1860 struct list_head *pos;
1861 struct lov_obd *lov;
1866 ASSERT_LSM_MAGIC(oinfo->oi_md);
1867 LASSERT(mode == (mode & -mode));
1869 /* we should never be asked to replay a lock this way. */
1870 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1872 if (!exp || !exp->exp_obd)
1875 lov = &exp->exp_obd->u.lov;
1876 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1880 list_for_each (pos, &set->set_list) {
1881 req = list_entry(pos, struct lov_request, rq_link);
1883 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1884 &req->rq_oi, einfo, rqset);
1889 if (rqset && !list_empty(&rqset->set_requests)) {
1891 LASSERT(rqset->set_interpret == NULL);
1892 rqset->set_interpret = lov_enqueue_interpret;
1893 rqset->set_arg = (void *)set;
1897 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1901 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
1902 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1903 int *flags, void *data, struct lustre_handle *lockh)
1905 struct lov_request_set *set;
1906 struct obd_info oinfo;
1907 struct lov_request *req;
1908 struct list_head *pos;
1909 struct lov_obd *lov = &exp->exp_obd->u.lov;
1910 struct lustre_handle *lov_lockhp;
1911 int lov_flags, rc = 0;
1914 ASSERT_LSM_MAGIC(lsm);
1915 LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
1917 if (!exp || !exp->exp_obd)
1920 lov = &exp->exp_obd->u.lov;
1921 rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
1925 list_for_each (pos, &set->set_list) {
1926 ldlm_policy_data_t sub_policy;
1927 req = list_entry(pos, struct lov_request, rq_link);
1928 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1929 LASSERT(lov_lockhp);
1932 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
1934 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
1935 req->rq_oi.oi_md, type, &sub_policy,
1936 mode, &lov_flags, data, lov_lockhp);
1937 rc = lov_update_match_set(set, req, rc);
1941 lov_fini_match_set(set, mode, *flags);
1945 static int lov_change_cbdata(struct obd_export *exp,
1946 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1949 struct lov_obd *lov;
1953 ASSERT_LSM_MAGIC(lsm);
1955 if (!exp || !exp->exp_obd)
1958 LASSERT(lsm->lsm_object_gr > 0);
1960 lov = &exp->exp_obd->u.lov;
1961 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1962 struct lov_stripe_md submd;
1963 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1965 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1966 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1970 submd.lsm_object_id = loi->loi_id;
1971 submd.lsm_object_gr = lsm->lsm_object_gr;
1972 submd.lsm_stripe_count = 0;
1973 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1979 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1980 __u32 mode, struct lustre_handle *lockh)
1982 struct lov_request_set *set;
1983 struct obd_info oinfo;
1984 struct lov_request *req;
1985 struct list_head *pos;
1986 struct lov_obd *lov = &exp->exp_obd->u.lov;
1987 struct lustre_handle *lov_lockhp;
1988 int err = 0, rc = 0;
1991 ASSERT_LSM_MAGIC(lsm);
1993 if (!exp || !exp->exp_obd)
1996 LASSERT(lsm->lsm_object_gr > 0);
1998 lov = &exp->exp_obd->u.lov;
1999 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2003 list_for_each (pos, &set->set_list) {
2004 req = list_entry(pos, struct lov_request, rq_link);
2005 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2007 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2008 req->rq_oi.oi_md, mode, lov_lockhp);
2009 rc = lov_update_common_set(set, req, rc);
2011 CERROR("error: cancel objid "LPX64" subobj "
2012 LPX64" on OST idx %d: rc = %d\n",
2014 req->rq_oi.oi_md->lsm_object_id,
2020 lov_fini_cancel_set(set);
2024 static int lov_cancel_unused(struct obd_export *exp,
2025 struct lov_stripe_md *lsm,
2026 int flags, void *opaque)
2028 struct lov_obd *lov;
2032 if (!exp || !exp->exp_obd)
2035 lov = &exp->exp_obd->u.lov;
2037 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2039 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2042 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2050 ASSERT_LSM_MAGIC(lsm);
2052 LASSERT(lsm->lsm_object_gr > 0);
2053 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2054 struct lov_stripe_md submd;
2055 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2058 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2059 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2063 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2064 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2066 submd.lsm_object_id = loi->loi_id;
2067 submd.lsm_object_gr = lsm->lsm_object_gr;
2068 submd.lsm_stripe_count = 0;
2069 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2070 &submd, flags, opaque);
2071 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2072 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2073 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2074 loi->loi_id, loi->loi_ost_idx, err);
2082 static int lov_join_lru(struct obd_export *exp,
2083 struct lov_stripe_md *lsm, int join)
2085 struct lov_obd *lov;
2089 ASSERT_LSM_MAGIC(lsm);
2090 if (!exp || !exp->exp_obd)
2093 lov = &exp->exp_obd->u.lov;
2094 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2095 struct lov_stripe_md submd;
2096 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2099 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2100 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2104 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2105 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2107 submd.lsm_object_id = loi->loi_id;
2108 submd.lsm_object_gr = lsm->lsm_object_gr;
2109 submd.lsm_stripe_count = 0;
2110 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2113 CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2114 " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2115 loi->loi_id, loi->loi_ost_idx, rc);
2124 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2127 struct lov_request_set *lovset = (struct lov_request_set *)data;
2132 lovset->set_completes = 0;
2134 err = lov_fini_statfs_set(lovset);
2135 RETURN(rc ? rc : err);
2138 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2139 __u64 max_age, struct ptlrpc_request_set *rqset)
2141 struct lov_request_set *set;
2142 struct lov_request *req;
2143 struct list_head *pos;
2144 struct lov_obd *lov;
2148 LASSERT(oinfo != NULL);
2149 LASSERT(oinfo->oi_osfs != NULL);
2152 rc = lov_prep_statfs_set(obd, oinfo, &set);
2156 list_for_each (pos, &set->set_list) {
2157 struct obd_device *osc_obd;
2159 req = list_entry(pos, struct lov_request, rq_link);
2161 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2162 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2167 if (rc || list_empty(&rqset->set_requests)) {
2170 set->set_completes = 0;
2171 err = lov_fini_statfs_set(set);
2172 RETURN(rc ? rc : err);
2175 LASSERT(rqset->set_interpret == NULL);
2176 rqset->set_interpret = lov_statfs_interpret;
2177 rqset->set_arg = (void *)set;
2181 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2184 struct lov_obd *lov = &obd->u.lov;
2185 struct obd_statfs lov_sfs;
2191 /* We only get block data from the OBD */
2192 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2193 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
2194 CDEBUG(D_HA, "lov idx %d inactive\n", i);
2198 err = obd_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
2201 if (lov->lov_tgts[i]->ltd_active && !rc)
2205 lov_update_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
2206 osfs, &lov_sfs, set);
2210 err = lov_fini_statfs(obd, osfs, set);
2213 RETURN(rc ? rc : err);
2216 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2217 void *karg, void *uarg)
2219 struct obd_device *obddev = class_exp2obd(exp);
2220 struct lov_obd *lov = &obddev->u.lov;
2221 int i, rc, count = lov->desc.ld_tgt_count;
2222 struct obd_uuid *uuidp;
2226 case OBD_IOC_LOV_GET_CONFIG: {
2227 struct obd_ioctl_data *data;
2228 struct lov_desc *desc;
2233 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2236 data = (struct obd_ioctl_data *)buf;
2238 if (sizeof(*desc) > data->ioc_inllen1) {
2239 obd_ioctl_freedata(buf, len);
2243 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2244 obd_ioctl_freedata(buf, len);
2248 if (sizeof(__u32) * count > data->ioc_inllen3) {
2249 obd_ioctl_freedata(buf, len);
2253 desc = (struct lov_desc *)data->ioc_inlbuf1;
2254 memcpy(desc, &(lov->desc), sizeof(*desc));
2256 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2257 genp = (__u32 *)data->ioc_inlbuf3;
2258 /* the uuid will be empty for deleted OSTs */
2259 for (i = 0; i < count; i++, uuidp++, genp++) {
2260 if (!lov->lov_tgts[i])
2262 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2263 *genp = lov->lov_tgts[i]->ltd_gen;
2266 rc = copy_to_user((void *)uarg, buf, len);
2269 obd_ioctl_freedata(buf, len);
2272 case LL_IOC_LOV_SETSTRIPE:
2273 rc = lov_setstripe(exp, karg, uarg);
2275 case LL_IOC_LOV_GETSTRIPE:
2276 rc = lov_getstripe(exp, karg, uarg);
2278 case LL_IOC_LOV_SETEA:
2279 rc = lov_setea(exp, karg, uarg);
2288 for (i = 0; i < count; i++) {
2291 /* OST was disconnected */
2292 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2295 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2297 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2300 if (lov->lov_tgts[i]->ltd_active) {
2301 CDEBUG(err == -ENOTTY ?
2302 D_IOCTL : D_WARNING,
2303 "iocontrol OSC %s on OST "
2304 "idx %d cmd %x: err = %d\n",
2305 lov_uuid2str(lov, i),
2322 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2323 void *key, __u32 *vallen, void *val)
2325 struct obd_device *obddev = class_exp2obd(exp);
2326 struct lov_obd *lov = &obddev->u.lov;
2330 if (!vallen || !val)
2335 if (KEY_IS("lock_to_stripe")) {
2338 struct ldlm_lock *lock;
2339 struct lov_stripe_md *lsm;
2341 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2342 struct lov_oinfo *loi;
2343 __u32 *stripe = val;
2345 if (*vallen < sizeof(*stripe))
2346 GOTO(out, rc = -EFAULT);
2347 *vallen = sizeof(*stripe);
2349 /* XXX This is another one of those bits that will need to
2350 * change if we ever actually support nested LOVs. It uses
2351 * the lock's export to find out which stripe it is. */
2352 /* XXX - it's assumed all the locks for deleted OSTs have
2353 * been cancelled. Also, the export for deleted OSTs will
2354 * be NULL and won't match the lock's export. */
2355 for (i = 0; i < data->lsm->lsm_stripe_count; i++) {
2356 loi = data->lsm->lsm_oinfo[i];
2357 if (!lov->lov_tgts[loi->loi_ost_idx])
2359 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2360 data->lock->l_conn_export &&
2361 loi->loi_id == res_id->name[0] &&
2362 loi->loi_gr == res_id->name[2]) {
2367 LDLM_ERROR(data->lock, "lock on inode without such object");
2368 dump_lsm(D_ERROR, data->lsm);
2369 GOTO(out, rc = -ENXIO);
2370 } else if (KEY_IS(KEY_LAST_ID)) {
2371 struct obd_id_info *info = val;
2372 int size = sizeof(obd_id);
2373 struct lov_tgt_desc *tgt;
2375 LASSERT(*vallen == sizeof(struct obd_id_info));
2376 tgt = lov->lov_tgts[info->idx];
2378 if (!tgt || !tgt->ltd_active)
2379 GOTO(out, rc = -ESRCH);
2381 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data);
2383 } else if (KEY_IS(KEY_LOVDESC)) {
2384 struct lov_desc *desc_ret = val;
2385 *desc_ret = lov->desc;
2388 } else if (KEY_IS(KEY_LOV_IDX)) {
2389 struct lov_tgt_desc *tgt;
2391 for(i = 0; i < lov->desc.ld_tgt_count; i++) {
2392 tgt = lov->lov_tgts[i];
2393 if (obd_uuid_equals(val, &tgt->ltd_uuid))
2404 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2405 void *key, obd_count vallen, void *val,
2406 struct ptlrpc_request_set *set)
2408 struct obd_device *obddev = class_exp2obd(exp);
2409 struct lov_obd *lov = &obddev->u.lov;
2412 struct lov_tgt_desc *tgt;
2413 unsigned incr, check_uuid,
2414 do_inactive, no_set;
2415 unsigned next_id = 0, mds_con = 0;
2418 incr = check_uuid = do_inactive = no_set = 0;
2421 set = ptlrpc_prep_set();
2427 count = lov->desc.ld_tgt_count;
2429 if (KEY_IS(KEY_NEXT_ID)) {
2430 count = vallen / sizeof(struct obd_id_info);
2431 vallen = sizeof(obd_id);
2432 incr = sizeof(struct obd_id_info);
2435 } else if (KEY_IS("checksum")) {
2437 } else if (KEY_IS("unlinked")) {
2438 check_uuid = val ? 1 : 0;
2439 } else if (KEY_IS("evict_by_nid")) {
2440 /* use defaults: do_inactive = incr = 0; */
2441 } else if (KEY_IS(KEY_MDS_CONN)) {
2445 for (i = 0; i < count; i++, val = (char *)val + incr) {
2447 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2449 tgt = lov->lov_tgts[i];
2451 /* OST was disconnected */
2452 if (!tgt || !tgt->ltd_exp)
2455 /* OST is inactive and we don't want inactive OSCs */
2456 if (!tgt->ltd_active && !do_inactive)
2460 struct mds_group_info *mgi;
2462 LASSERT(vallen == sizeof(*mgi));
2463 mgi = (struct mds_group_info *)val;
2465 /* Only want a specific OSC */
2466 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2470 err = obd_set_info_async(tgt->ltd_exp,
2471 keylen, key, sizeof(int),
2473 } else if (next_id) {
2474 err = obd_set_info_async(tgt->ltd_exp,
2475 keylen, key, vallen,
2476 ((struct obd_id_info*)val)->data, set);
2478 /* Only want a specific OSC */
2480 !obd_uuid_equals(val, &tgt->ltd_uuid))
2483 err = obd_set_info_async(tgt->ltd_exp,
2484 keylen, key, vallen, val, set);
2493 err = ptlrpc_set_wait(set);
2496 ptlrpc_set_destroy(set);
2501 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2502 struct lov_stripe_md *lsm)
2510 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2511 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2516 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2521 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2522 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2523 if (loi->loi_ar.ar_rc && !rc)
2524 rc = loi->loi_ar.ar_rc;
2525 loi->loi_ar.ar_rc = 0;
2529 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2532 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2533 int cmd, __u64 *offset)
2535 __u32 ssize = lsm->lsm_stripe_size;
2539 do_div(start, ssize);
2540 start = start * ssize;
2542 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2543 ", end "LPU64"\n", *offset, ssize, start,
2545 if (cmd == OBD_CALC_STRIPE_END) {
2546 *offset = start + ssize - 1;
2547 } else if (cmd == OBD_CALC_STRIPE_START) {
2558 struct lov_multi_wait {
2559 struct ldlm_lock *lock;
2565 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2566 struct lustre_handle *lockh)
2568 struct lov_lock_handles *lov_lockh = NULL;
2569 struct lustre_handle *lov_lockhp;
2570 struct lov_obd *lov;
2571 struct lov_oinfo *loi;
2572 struct lov_multi_wait *queues;
2576 ASSERT_LSM_MAGIC(lsm);
2578 if (!exp || !exp->exp_obd)
2581 LASSERT(lockh != NULL);
2582 if (lsm->lsm_stripe_count > 1) {
2583 lov_lockh = lov_handle2llh(lockh);
2584 if (lov_lockh == NULL) {
2585 CERROR("LOV: invalid lov lock handle %p\n", lockh);
2589 lov_lockhp = lov_lockh->llh_handles;
2594 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2596 GOTO(out, rc = -ENOMEM);
2598 lov = &exp->exp_obd->u.lov;
2599 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2600 i++, loi++, lov_lockhp++) {
2601 struct ldlm_lock *lock;
2602 struct obd_device *obd;
2604 lock = ldlm_handle2lock(lov_lockhp);
2606 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2607 loi->loi_ost_idx, loi->loi_id);
2608 queues[i].completed = 1;
2612 queues[i].lock = lock;
2613 init_waitqueue_entry(&(queues[i].wait), current);
2614 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2616 obd = class_exp2obd(lock->l_conn_export);
2618 imp = obd->u.cli.cl_import;
2620 spin_lock(&imp->imp_lock);
2621 queues[i].generation = imp->imp_generation;
2622 spin_unlock(&imp->imp_lock);
2626 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2627 interrupted_completion_wait, &lwd);
2628 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2630 for (i = 0; i < lsm->lsm_stripe_count; i++)
2631 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2633 if (rc == -EINTR || rc == -ETIMEDOUT) {
2639 if (lov_lockh != NULL)
2640 lov_llh_put(lov_lockh);
2645 void lov_stripe_lock(struct lov_stripe_md *md)
2647 LASSERT(md->lsm_lock_owner != cfs_current());
2648 spin_lock(&md->lsm_lock);
2649 LASSERT(md->lsm_lock_owner == NULL);
2650 md->lsm_lock_owner = cfs_current();
2652 EXPORT_SYMBOL(lov_stripe_lock);
2654 void lov_stripe_unlock(struct lov_stripe_md *md)
2656 LASSERT(md->lsm_lock_owner == cfs_current());
2657 md->lsm_lock_owner = NULL;
2658 spin_unlock(&md->lsm_lock);
2660 EXPORT_SYMBOL(lov_stripe_unlock);
2662 struct obd_ops lov_obd_ops = {
2663 .o_owner = THIS_MODULE,
2664 .o_setup = lov_setup,
2665 .o_precleanup = lov_precleanup,
2666 .o_cleanup = lov_cleanup,
2667 .o_process_config = lov_process_config,
2668 .o_connect = lov_connect,
2669 .o_disconnect = lov_disconnect,
2670 .o_statfs = lov_statfs,
2671 .o_statfs_async = lov_statfs_async,
2672 .o_packmd = lov_packmd,
2673 .o_unpackmd = lov_unpackmd,
2674 .o_checkmd = lov_checkmd,
2675 .o_create = lov_create,
2676 .o_destroy = lov_destroy,
2677 .o_getattr = lov_getattr,
2678 .o_getattr_async = lov_getattr_async,
2679 .o_setattr = lov_setattr,
2680 .o_setattr_async = lov_setattr_async,
2682 .o_brw_async = lov_brw_async,
2683 .o_prep_async_page = lov_prep_async_page,
2684 .o_queue_async_io = lov_queue_async_io,
2685 .o_set_async_flags = lov_set_async_flags,
2686 .o_queue_group_io = lov_queue_group_io,
2687 .o_trigger_group_io = lov_trigger_group_io,
2688 .o_teardown_async_page = lov_teardown_async_page,
2689 .o_merge_lvb = lov_merge_lvb,
2690 .o_adjust_kms = lov_adjust_kms,
2691 .o_punch = lov_punch,
2693 .o_enqueue = lov_enqueue,
2694 .o_match = lov_match,
2695 .o_change_cbdata = lov_change_cbdata,
2696 .o_cancel = lov_cancel,
2697 .o_cancel_unused = lov_cancel_unused,
2698 .o_join_lru = lov_join_lru,
2699 .o_iocontrol = lov_iocontrol,
2700 .o_get_info = lov_get_info,
2701 .o_set_info_async = lov_set_info_async,
2702 .o_extent_calc = lov_extent_calc,
2703 .o_llog_init = lov_llog_init,
2704 .o_llog_finish = lov_llog_finish,
2705 .o_notify = lov_notify,
2708 static quota_interface_t *quota_interface;
2709 extern quota_interface_t lov_quota_interface;
2711 cfs_mem_cache_t *lov_oinfo_slab;
2713 int __init lov_init(void)
2715 struct lprocfs_static_vars lvars;
2719 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2720 sizeof(struct lov_oinfo),
2721 0, SLAB_HWCACHE_ALIGN);
2722 if (lov_oinfo_slab == NULL)
2724 lprocfs_init_vars(lov, &lvars);
2726 request_module("lquota");
2727 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2728 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2730 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2731 LUSTRE_LOV_NAME, NULL);
2733 if (quota_interface)
2734 PORTAL_SYMBOL_PUT(lov_quota_interface);
2735 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2743 static void /*__exit*/ lov_exit(void)
2747 if (quota_interface)
2748 PORTAL_SYMBOL_PUT(lov_quota_interface);
2750 class_unregister_type(LUSTRE_LOV_NAME);
2751 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2755 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2756 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2757 MODULE_LICENSE("GPL");
2759 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);