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;
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_uuid = lov->lov_tgts[index]->ltd_uuid;
112 tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
116 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
119 if (!tgt_obd->obd_set_up) {
120 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
124 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
125 data->ocd_index = index;
128 * Divine LOV knows that OBDs under it are OSCs.
130 imp = tgt_obd->u.cli.cl_import;
133 tgt_obd->obd_no_recov = 0;
134 /* FIXME this is probably supposed to be
135 ptlrpc_set_import_active. Horrible naming. */
136 ptlrpc_activate_import(imp);
139 if (imp->imp_invalid) {
140 CERROR("not connecting OSC %s; administratively "
141 "disabled\n", obd_uuid2str(&tgt_uuid));
142 rc = obd_register_observer(tgt_obd, obd);
144 CERROR("Target %s register_observer error %d; "
145 "will not be able to reactivate\n",
146 obd_uuid2str(&tgt_uuid), rc);
151 rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data);
153 CERROR("Target %s connect error %d\n",
154 obd_uuid2str(&tgt_uuid), rc);
157 lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
158 if (!lov->lov_tgts[index]->ltd_exp) {
159 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
163 rc = obd_register_observer(tgt_obd, obd);
165 CERROR("Target %s register_observer error %d\n",
166 obd_uuid2str(&tgt_uuid), rc);
167 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
168 lov->lov_tgts[index]->ltd_exp = NULL;
172 lov->lov_tgts[index]->ltd_reap = 0;
174 lov->lov_tgts[index]->ltd_active = 1;
175 lov->desc.ld_active_tgt_count++;
176 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
178 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
179 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
182 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
184 struct obd_device *osc_obd = class_conn2obd(&conn);
185 cfs_proc_dir_entry_t *osc_symlink;
186 char name[MAX_STRING_SIZE];
188 LASSERT(osc_obd != NULL);
189 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
190 LASSERT(osc_obd->obd_type->typ_name != NULL);
191 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
192 osc_obd->obd_type->typ_name,
194 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
196 if (osc_symlink == NULL) {
197 CERROR("could not register LOV target "
198 "/proc/fs/lustre/%s/%s/target_obds/%s.",
199 obd->obd_type->typ_name, obd->obd_name,
201 lprocfs_remove(&lov_proc_dir);
206 rc = qos_add_tgt(obd, index);
208 CERROR("qos_add_tgt failed %d\n", rc);
213 static int lov_connect(const struct lu_env *env,
214 struct lustre_handle *conn, struct obd_device *obd,
215 struct obd_uuid *cluuid, struct obd_connect_data *data)
217 struct lov_obd *lov = &obd->u.lov;
218 struct lov_tgt_desc *tgt;
222 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
224 rc = class_connect(conn, obd, cluuid);
228 /* Why should there ever be more than 1 connect? */
230 LASSERT(lov->lov_connects == 1);
232 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
234 lov->lov_ocd = *data;
237 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
238 tgt = lov->lov_tgts[i];
239 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
241 /* Flags will be lowest common denominator */
242 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
244 CERROR("%s: lov connect tgt %d failed: %d\n",
245 obd->obd_name, i, rc);
254 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
256 cfs_proc_dir_entry_t *lov_proc_dir;
257 struct lov_obd *lov = &obd->u.lov;
258 struct obd_device *osc_obd;
263 if (lov->lov_tgts[index] == NULL)
266 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
267 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
268 obd->obd_name, osc_obd->obd_name);
271 if (lov->lov_tgts[index]->ltd_active) {
272 lov->lov_tgts[index]->ltd_active = 0;
273 lov->desc.ld_active_tgt_count--;
274 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
277 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
279 cfs_proc_dir_entry_t *osc_symlink;
281 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
283 lprocfs_remove(&osc_symlink);
285 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
286 obd->obd_type->typ_name, obd->obd_name,
292 /* Pass it on to our clients.
293 * XXX This should be an argument to disconnect,
294 * XXX not a back-door flag on the OBD. Ah well.
296 osc_obd->obd_force = obd->obd_force;
297 osc_obd->obd_fail = obd->obd_fail;
298 osc_obd->obd_no_recov = obd->obd_no_recov;
301 obd_register_observer(osc_obd, NULL);
303 rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
305 CERROR("Target %s disconnect error %d\n",
306 lov_uuid2str(lov, index), rc);
310 qos_del_tgt(obd, index);
312 lov->lov_tgts[index]->ltd_exp = NULL;
316 static int lov_del_target(struct obd_device *obd, __u32 index,
317 struct obd_uuid *uuidp, int gen);
319 static int lov_disconnect(struct obd_export *exp)
321 struct obd_device *obd = class_exp2obd(exp);
322 struct lov_obd *lov = &obd->u.lov;
329 /* Only disconnect the underlying layers on the final disconnect. */
331 if (lov->lov_connects != 0) {
332 /* why should there be more than 1 connect? */
333 CERROR("disconnect #%d\n", lov->lov_connects);
337 /* Let's hold another reference so lov_del_obd doesn't spin through
340 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
341 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
342 /* Disconnection is the last we know about an obd */
343 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
349 rc = class_disconnect(exp); /* bz 9811 */
355 * -EINVAL : UUID can't be found in the LOV's target list
356 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
357 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
359 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
362 struct lov_obd *lov = &obd->u.lov;
363 struct lov_tgt_desc *tgt;
367 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
368 lov, uuid->uuid, activate);
371 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
372 tgt = lov->lov_tgts[i];
373 if (!tgt || !tgt->ltd_exp)
376 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
377 i, obd_uuid2str(&tgt->ltd_uuid),
378 tgt->ltd_exp->exp_handle.h_cookie);
379 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
383 if (i == lov->desc.ld_tgt_count)
384 GOTO(out, rc = -EINVAL);
386 if (lov->lov_tgts[i]->ltd_active == activate) {
387 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
388 activate ? "" : "in");
392 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
393 activate ? "" : "in");
395 lov->lov_tgts[i]->ltd_active = activate;
398 lov->desc.ld_active_tgt_count++;
399 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
401 lov->desc.ld_active_tgt_count--;
402 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
404 /* remove any old qos penalty */
405 lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
412 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
413 enum obd_notify_event ev, void *data)
418 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
419 struct obd_uuid *uuid;
423 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
424 CERROR("unexpected notification of %s %s!\n",
425 watched->obd_type->typ_name,
429 uuid = &watched->u.cli.cl_target_uuid;
431 /* Set OSC as active before notifying the observer, so the
432 * observer can use the OSC normally.
434 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
436 CERROR("%sactivation of %s failed: %d\n",
437 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
438 obd_uuid2str(uuid), rc);
443 /* Pass the notification up the chain. */
445 rc = obd_notify_observer(obd, watched, ev, data);
447 /* NULL watched means all osc's in the lov (only for syncs) */
448 struct lov_obd *lov = &obd->u.lov;
449 struct obd_device *tgt_obd;
452 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
453 if (!lov->lov_tgts[i])
455 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
456 rc = obd_notify_observer(obd, tgt_obd, ev, data);
458 CERROR("%s: notify %s of %s failed %d\n",
460 obd->obd_observer->obd_name,
461 tgt_obd->obd_name, rc);
471 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
472 __u32 index, int gen, int active)
474 struct lov_obd *lov = &obd->u.lov;
475 struct lov_tgt_desc *tgt;
479 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
480 uuidp->uuid, index, gen, active);
483 CERROR("request to add OBD %s with invalid generation: %d\n",
488 mutex_down(&lov->lov_lock);
490 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
491 tgt = lov->lov_tgts[index];
492 CERROR("UUID %s already assigned at LOV target index %d\n",
493 obd_uuid2str(&tgt->ltd_uuid), index);
494 mutex_up(&lov->lov_lock);
498 if (index >= lov->lov_tgt_size) {
499 /* We need to reallocate the lov target array. */
500 struct lov_tgt_desc **newtgts, **old = NULL;
501 __u32 newsize, oldsize = 0;
503 newsize = max(lov->lov_tgt_size, (__u32)2);
504 while (newsize < index + 1)
505 newsize = newsize << 1;
506 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
507 if (newtgts == NULL) {
508 mutex_up(&lov->lov_lock);
512 if (lov->lov_tgt_size) {
513 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
516 oldsize = lov->lov_tgt_size;
519 lov->lov_tgts = newtgts;
520 lov->lov_tgt_size = newsize;
525 OBD_FREE(old, sizeof(*old) * oldsize);
527 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
528 lov->lov_tgts, lov->lov_tgt_size);
534 mutex_up(&lov->lov_lock);
538 memset(tgt, 0, sizeof(*tgt));
539 tgt->ltd_uuid = *uuidp;
540 /* XXX - add a sanity check on the generation number. */
542 tgt->ltd_index = index;
543 tgt->ltd_activate = active;
544 lov->lov_tgts[index] = tgt;
545 if (index >= lov->desc.ld_tgt_count)
546 lov->desc.ld_tgt_count = index + 1;
547 mutex_up(&lov->lov_lock);
549 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
550 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
552 if (lov->lov_connects == 0) {
553 /* lov_connect hasn't been called yet. We'll do the
554 lov_connect_obd on this target when that fn first runs,
555 because we don't know the connect flags yet. */
561 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
565 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
566 active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
571 CERROR("add failed (%d), deleting %s\n", rc,
572 obd_uuid2str(&tgt->ltd_uuid));
573 lov_del_target(obd, index, 0, 0);
579 /* Schedule a target for deletion */
580 static int lov_del_target(struct obd_device *obd, __u32 index,
581 struct obd_uuid *uuidp, int gen)
583 struct lov_obd *lov = &obd->u.lov;
584 int count = lov->desc.ld_tgt_count;
588 if (index >= count) {
589 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
596 if (!lov->lov_tgts[index]) {
597 CERROR("LOV target at index %d is not setup.\n", index);
598 GOTO(out, rc = -EINVAL);
601 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
602 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
603 lov_uuid2str(lov, index), index,
604 obd_uuid2str(uuidp));
605 GOTO(out, rc = -EINVAL);
608 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
609 lov_uuid2str(lov, index), index,
610 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
611 lov->lov_tgts[index]->ltd_active);
613 lov->lov_tgts[index]->ltd_reap = 1;
614 lov->lov_death_row++;
615 /* we really delete it from lov_putref */
622 /* We are holding lov_lock */
623 static void __lov_del_obd(struct obd_device *obd, __u32 index)
625 struct lov_obd *lov = &obd->u.lov;
626 struct obd_device *osc_obd;
627 struct lov_tgt_desc *tgt = lov->lov_tgts[index];
630 LASSERT(tgt->ltd_reap);
632 osc_obd = class_exp2obd(tgt->ltd_exp);
634 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
635 lov_uuid2str(lov, index),
636 osc_obd ? osc_obd->obd_name : "<no obd>");
639 lov_disconnect_obd(obd, index);
641 /* XXX - right now there is a dependency on ld_tgt_count being the
642 * maximum tgt index for computing the mds_max_easize. So we can't
645 lov->lov_tgts[index] = NULL;
648 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
649 do it ourselves. And we can't do it from lov_cleanup,
650 because we just lost our only reference to it. */
652 class_manual_cleanup(osc_obd);
655 void lov_fix_desc(struct lov_desc *desc)
657 if (desc->ld_default_stripe_size < PTLRPC_MAX_BRW_SIZE) {
658 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
659 PTLRPC_MAX_BRW_SIZE);
660 desc->ld_default_stripe_size = PTLRPC_MAX_BRW_SIZE;
661 } else if (desc->ld_default_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
662 desc->ld_default_stripe_size &= ~(LOV_MIN_STRIPE_SIZE - 1);
663 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
665 desc->ld_default_stripe_size,LOV_MIN_STRIPE_SIZE);
668 if (desc->ld_default_stripe_count == 0)
669 desc->ld_default_stripe_count = 1;
671 /* from lov_setstripe */
672 if ((desc->ld_pattern != 0) &&
673 (desc->ld_pattern != LOV_PATTERN_RAID0)) {
674 LCONSOLE_WARN("Unknown stripe pattern: %#x\n",desc->ld_pattern);
675 desc->ld_pattern = 0;
678 if (desc->ld_qos_maxage == 0)
679 desc->ld_qos_maxage = QOS_DEFAULT_MAXAGE;
682 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
684 struct lprocfs_static_vars lvars = { 0 };
685 struct lov_desc *desc;
686 struct lov_obd *lov = &obd->u.lov;
690 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
691 CERROR("LOV setup requires a descriptor\n");
695 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
697 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
698 CERROR("descriptor size wrong: %d > %d\n",
699 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
703 if (desc->ld_magic != LOV_DESC_MAGIC) {
704 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
705 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
706 obd->obd_name, desc);
707 lustre_swab_lov_desc(desc);
709 CERROR("%s: Bad lov desc magic: %#x\n",
710 obd->obd_name, desc->ld_magic);
717 /* Because of 64-bit divide/mod operations only work with a 32-bit
718 * divisor in a 32-bit kernel, we cannot support a stripe width
719 * of 4GB or larger on 32-bit CPUs. */
720 count = desc->ld_default_stripe_count;
721 if ((count > 0 ? count : desc->ld_tgt_count) *
722 desc->ld_default_stripe_size > 0xffffffff) {
723 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
724 desc->ld_default_stripe_size, count);
728 desc->ld_active_tgt_count = 0;
730 lov->lov_tgt_size = 0;
731 sema_init(&lov->lov_lock, 1);
732 atomic_set(&lov->lov_refcount, 0);
733 INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
734 init_rwsem(&lov->lov_qos.lq_rw_sem);
735 lov->lov_qos.lq_dirty = 1;
736 lov->lov_qos.lq_dirty_rr = 1;
737 lov->lov_qos.lq_reset = 1;
738 /* Default priority is toward free space balance */
739 lov->lov_qos.lq_prio_free = 232;
741 lprocfs_lov_init_vars(&lvars);
742 lprocfs_obd_setup(obd, lvars.obd_vars);
745 cfs_proc_dir_entry_t *entry;
747 entry = create_proc_entry("target_obd", 0444,
748 obd->obd_proc_entry);
750 entry->proc_fops = &lov_proc_target_fops;
759 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
765 case OBD_CLEANUP_EARLY: {
766 struct lov_obd *lov = &obd->u.lov;
768 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
769 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
771 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
776 case OBD_CLEANUP_EXPORTS:
778 case OBD_CLEANUP_SELF_EXP:
779 rc = obd_llog_finish(obd, 0);
781 CERROR("failed to cleanup llogging subsystems\n");
783 case OBD_CLEANUP_OBD:
789 static int lov_cleanup(struct obd_device *obd)
791 struct lov_obd *lov = &obd->u.lov;
793 lprocfs_obd_cleanup(obd);
796 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
797 if (!lov->lov_tgts[i])
800 /* Inactive targets may never have connected */
801 if (lov->lov_tgts[i]->ltd_active ||
802 atomic_read(&lov->lov_refcount))
803 /* We should never get here - these
804 should have been removed in the
806 CERROR("lov tgt %d not cleaned!"
807 " deathrow=%d, lovrc=%d\n",
808 i, lov->lov_death_row,
809 atomic_read(&lov->lov_refcount));
810 lov_del_target(obd, i, 0, 0);
812 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
814 lov->lov_tgt_size = 0;
817 if (lov->lov_qos.lq_rr_size)
818 OBD_FREE(lov->lov_qos.lq_rr_array, lov->lov_qos.lq_rr_size);
823 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
825 struct lustre_cfg *lcfg = buf;
826 struct obd_uuid obd_uuid;
831 switch(cmd = lcfg->lcfg_command) {
832 case LCFG_LOV_ADD_OBD:
833 case LCFG_LOV_ADD_INA:
834 case LCFG_LOV_DEL_OBD: {
837 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
838 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
839 GOTO(out, rc = -EINVAL);
841 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
843 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
844 GOTO(out, rc = -EINVAL);
845 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
846 GOTO(out, rc = -EINVAL);
847 if (cmd == LCFG_LOV_ADD_OBD)
848 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
849 else if (cmd == LCFG_LOV_ADD_INA)
850 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
852 rc = lov_del_target(obd, index, &obd_uuid, gen);
856 struct lprocfs_static_vars lvars = { 0 };
857 struct lov_desc *desc = &(obd->u.lov.desc);
860 GOTO(out, rc = -EINVAL);
862 lprocfs_lov_init_vars(&lvars);
864 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
869 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
870 GOTO(out, rc = -EINVAL);
879 #define log2(n) ffz(~(n))
882 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
883 struct lov_stripe_md **ea,
884 struct obd_trans_info *oti)
888 struct obd_uuid *ost_uuid = NULL;
892 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
893 src_oa->o_flags == OBD_FL_DELORPHAN);
895 lov = &export->exp_obd->u.lov;
901 if (src_oa->o_valid & OBD_MD_FLINLINE) {
902 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
903 CDEBUG(D_HA, "clearing orphans only for %s\n",
907 lov_getref(export->exp_obd);
908 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
909 struct lov_stripe_md obj_md;
910 struct lov_stripe_md *obj_mdp = &obj_md;
911 struct lov_tgt_desc *tgt;
914 tgt = lov->lov_tgts[i];
918 /* if called for a specific target, we don't
919 care if it is not active. */
920 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
921 CDEBUG(D_HA, "lov idx %d inactive\n", i);
925 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
928 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
929 obd_uuid2str(ost_uuid));
931 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
933 LASSERT(lov->lov_tgts[i]->ltd_exp);
934 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
935 err = obd_create(lov->lov_tgts[i]->ltd_exp,
936 tmp_oa, &obj_mdp, oti);
938 /* This export will be disabled until it is recovered,
939 and then orphan recovery will be completed. */
940 CERROR("error in orphan recovery on OST idx %d/%d: "
941 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
946 lov_putref(export->exp_obd);
952 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
953 struct lov_stripe_md **ea, struct obd_trans_info *oti)
955 struct lov_stripe_md *obj_mdp, *lsm;
956 struct lov_obd *lov = &exp->exp_obd->u.lov;
961 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
962 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
964 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
968 ost_idx = src_oa->o_nlink;
971 GOTO(out, rc = -EINVAL);
972 if (ost_idx >= lov->desc.ld_tgt_count ||
973 !lov->lov_tgts[ost_idx])
974 GOTO(out, rc = -EINVAL);
976 for (i = 0; i < lsm->lsm_stripe_count; i++) {
977 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
978 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
979 GOTO(out, rc = -EINVAL);
983 if (i == lsm->lsm_stripe_count)
984 GOTO(out, rc = -EINVAL);
986 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
988 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
992 /* the LOV expects oa->o_id to be set to the LOV object id */
993 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
994 struct lov_stripe_md **ea, struct obd_trans_info *oti)
997 struct obd_info oinfo;
998 struct lov_request_set *set = NULL;
999 struct lov_request *req;
1000 struct obd_statfs osfs;
1005 LASSERT(ea != NULL);
1009 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1010 src_oa->o_flags == OBD_FL_DELORPHAN) {
1011 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1015 lov = &exp->exp_obd->u.lov;
1016 if (!lov->desc.ld_active_tgt_count)
1019 /* Recreate a specific object id at the given OST index */
1020 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1021 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1022 rc = lov_recreate(exp, src_oa, ea, oti);
1026 maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1027 obd_statfs_rqset(exp->exp_obd, &osfs, maxage, 1);
1029 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1033 list_for_each_entry(req, &set->set_list, rq_link) {
1034 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1035 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1036 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1037 lov_update_create_set(set, req, rc);
1039 rc = lov_fini_create_set(set, ea);
1043 #define ASSERT_LSM_MAGIC(lsmp) \
1045 LASSERT((lsmp) != NULL); \
1046 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC || \
1047 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1048 (lsmp), (lsmp)->lsm_magic); \
1051 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1052 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1053 struct obd_export *md_exp)
1055 struct lov_request_set *set;
1056 struct obd_info oinfo;
1057 struct lov_request *req;
1058 struct list_head *pos;
1059 struct lov_obd *lov;
1063 ASSERT_LSM_MAGIC(lsm);
1065 if (!exp || !exp->exp_obd)
1068 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1070 LASSERT(oti->oti_logcookies);
1073 lov = &exp->exp_obd->u.lov;
1074 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1078 list_for_each (pos, &set->set_list) {
1080 req = list_entry(pos, struct lov_request, rq_link);
1082 if (oa->o_valid & OBD_MD_FLCOOKIE)
1083 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1085 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1086 req->rq_oi.oi_oa, NULL, oti, NULL);
1087 err = lov_update_common_set(set, req, err);
1089 CERROR("error: destroying objid "LPX64" subobj "
1090 LPX64" on OST idx %d: rc = %d\n",
1091 oa->o_id, req->rq_oi.oi_oa->o_id,
1099 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1100 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1102 err = lov_fini_destroy_set(set);
1103 RETURN(rc ? rc : err);
1106 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1108 struct lov_request_set *set;
1109 struct lov_request *req;
1110 struct list_head *pos;
1111 struct lov_obd *lov;
1112 int err = 0, rc = 0;
1116 ASSERT_LSM_MAGIC(oinfo->oi_md);
1118 if (!exp || !exp->exp_obd)
1121 lov = &exp->exp_obd->u.lov;
1123 rc = lov_prep_getattr_set(exp, oinfo, &set);
1127 list_for_each (pos, &set->set_list) {
1128 req = list_entry(pos, struct lov_request, rq_link);
1130 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1131 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1132 req->rq_oi.oi_oa->o_id, req->rq_idx);
1134 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1136 err = lov_update_common_set(set, req, rc);
1138 CERROR("error: getattr objid "LPX64" subobj "
1139 LPX64" on OST idx %d: rc = %d\n",
1140 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1146 rc = lov_fini_getattr_set(set);
1152 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1155 struct lov_request_set *lovset = (struct lov_request_set *)data;
1159 /* don't do attribute merge if this aysnc op failed */
1161 lovset->set_completes = 0;
1162 err = lov_fini_getattr_set(lovset);
1163 RETURN(rc ? rc : err);
1166 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1167 struct ptlrpc_request_set *rqset)
1169 struct lov_request_set *lovset;
1170 struct lov_obd *lov;
1171 struct list_head *pos;
1172 struct lov_request *req;
1177 ASSERT_LSM_MAGIC(oinfo->oi_md);
1179 if (!exp || !exp->exp_obd)
1182 lov = &exp->exp_obd->u.lov;
1184 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1188 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1189 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1190 oinfo->oi_md->lsm_stripe_size);
1192 list_for_each (pos, &lovset->set_list) {
1193 req = list_entry(pos, struct lov_request, rq_link);
1195 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1196 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1197 req->rq_oi.oi_oa->o_id, req->rq_idx);
1198 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1199 &req->rq_oi, rqset);
1201 CERROR("error: getattr objid "LPX64" subobj "
1202 LPX64" on OST idx %d: rc = %d\n",
1203 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1209 if (!list_empty(&rqset->set_requests)) {
1211 LASSERT (rqset->set_interpret == NULL);
1212 rqset->set_interpret = lov_getattr_interpret;
1213 rqset->set_arg = (void *)lovset;
1218 lovset->set_completes = 0;
1219 err = lov_fini_getattr_set(lovset);
1220 RETURN(rc ? rc : err);
1223 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1224 struct obd_trans_info *oti)
1226 struct lov_request_set *set;
1227 struct lov_obd *lov;
1228 struct list_head *pos;
1229 struct lov_request *req;
1230 int err = 0, rc = 0;
1234 ASSERT_LSM_MAGIC(oinfo->oi_md);
1236 if (!exp || !exp->exp_obd)
1239 /* for now, we only expect the following updates here */
1240 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1241 OBD_MD_FLMODE | OBD_MD_FLATIME |
1242 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1243 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1244 OBD_MD_FLGROUP | OBD_MD_FLUID |
1245 OBD_MD_FLGID | OBD_MD_FLINLINE |
1246 OBD_MD_FLFID | OBD_MD_FLGENER)));
1247 lov = &exp->exp_obd->u.lov;
1248 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1252 list_for_each (pos, &set->set_list) {
1253 req = list_entry(pos, struct lov_request, rq_link);
1255 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1257 err = lov_update_setattr_set(set, req, rc);
1259 CERROR("error: setattr objid "LPX64" subobj "
1260 LPX64" on OST idx %d: rc = %d\n",
1261 set->set_oi->oi_oa->o_id,
1262 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1267 err = lov_fini_setattr_set(set);
1273 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1276 struct lov_request_set *lovset = (struct lov_request_set *)data;
1281 lovset->set_completes = 0;
1282 err = lov_fini_setattr_set(lovset);
1283 RETURN(rc ? rc : err);
1286 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1287 needed. Otherwise, a client is waiting for responses. */
1288 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1289 struct obd_trans_info *oti,
1290 struct ptlrpc_request_set *rqset)
1292 struct lov_request_set *set;
1293 struct lov_request *req;
1294 struct list_head *pos;
1295 struct lov_obd *lov;
1300 ASSERT_LSM_MAGIC(oinfo->oi_md);
1301 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1303 LASSERT(oti->oti_logcookies);
1306 if (!exp || !exp->exp_obd)
1309 lov = &exp->exp_obd->u.lov;
1310 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1314 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1315 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1316 oinfo->oi_md->lsm_stripe_size);
1318 list_for_each (pos, &set->set_list) {
1319 req = list_entry(pos, struct lov_request, rq_link);
1321 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1322 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1324 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1325 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1326 req->rq_oi.oi_oa->o_id, req->rq_idx);
1328 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1329 &req->rq_oi, oti, rqset);
1331 CERROR("error: setattr objid "LPX64" subobj "
1332 LPX64" on OST idx %d: rc = %d\n",
1333 set->set_oi->oi_oa->o_id,
1334 req->rq_oi.oi_oa->o_id,
1340 /* If we are not waiting for responses on async requests, return. */
1341 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1344 set->set_completes = 0;
1345 err = lov_fini_setattr_set(set);
1346 RETURN(rc ? rc : err);
1349 LASSERT(rqset->set_interpret == NULL);
1350 rqset->set_interpret = lov_setattr_interpret;
1351 rqset->set_arg = (void *)set;
1356 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1359 struct lov_request_set *lovset = (struct lov_request_set *)data;
1364 lovset->set_completes = 0;
1365 err = lov_fini_punch_set(lovset);
1366 RETURN(rc ? rc : err);
1369 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1370 * we can send this 'punch' to just the authoritative node and the nodes
1371 * that the punch will affect. */
1372 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1373 struct obd_trans_info *oti,
1374 struct ptlrpc_request_set *rqset)
1376 struct lov_request_set *set;
1377 struct lov_obd *lov;
1378 struct list_head *pos;
1379 struct lov_request *req;
1384 ASSERT_LSM_MAGIC(oinfo->oi_md);
1386 if (!exp || !exp->exp_obd)
1389 lov = &exp->exp_obd->u.lov;
1390 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1394 list_for_each (pos, &set->set_list) {
1395 req = list_entry(pos, struct lov_request, rq_link);
1397 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1398 &req->rq_oi, NULL, rqset);
1400 CERROR("error: punch objid "LPX64" subobj "LPX64
1401 " on OST idx %d: rc = %d\n",
1402 set->set_oi->oi_oa->o_id,
1403 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1408 if (rc || list_empty(&rqset->set_requests)) {
1410 err = lov_fini_punch_set(set);
1411 RETURN(rc ? rc : err);
1414 LASSERT(rqset->set_interpret == NULL);
1415 rqset->set_interpret = lov_punch_interpret;
1416 rqset->set_arg = (void *)set;
1421 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1422 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1425 struct lov_request_set *set;
1426 struct obd_info oinfo;
1427 struct lov_obd *lov;
1428 struct list_head *pos;
1429 struct lov_request *req;
1430 int err = 0, rc = 0;
1433 ASSERT_LSM_MAGIC(lsm);
1438 lov = &exp->exp_obd->u.lov;
1439 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1443 list_for_each (pos, &set->set_list) {
1444 req = list_entry(pos, struct lov_request, rq_link);
1446 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1447 req->rq_oi.oi_oa, NULL,
1448 req->rq_oi.oi_policy.l_extent.start,
1449 req->rq_oi.oi_policy.l_extent.end, capa);
1450 err = lov_update_common_set(set, req, rc);
1452 CERROR("error: fsync objid "LPX64" subobj "LPX64
1453 " on OST idx %d: rc = %d\n",
1454 set->set_oi->oi_oa->o_id,
1455 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1460 err = lov_fini_sync_set(set);
1466 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1467 obd_count oa_bufs, struct brw_page *pga)
1469 struct obd_info oinfo = { { { 0 } } };
1472 oinfo.oi_oa = lov_oinfo->oi_oa;
1474 /* The caller just wants to know if there's a chance that this
1475 * I/O can succeed */
1476 for (i = 0; i < oa_bufs; i++) {
1477 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1478 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1481 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1482 pga[i].off + pga[i].count,
1486 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1487 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1491 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1499 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1500 obd_count oa_bufs, struct brw_page *pga,
1501 struct obd_trans_info *oti)
1503 struct lov_request_set *set;
1504 struct lov_request *req;
1505 struct list_head *pos;
1506 struct lov_obd *lov = &exp->exp_obd->u.lov;
1510 ASSERT_LSM_MAGIC(oinfo->oi_md);
1512 if (cmd == OBD_BRW_CHECK) {
1513 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1517 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1521 list_for_each (pos, &set->set_list) {
1522 struct obd_export *sub_exp;
1523 struct brw_page *sub_pga;
1524 req = list_entry(pos, struct lov_request, rq_link);
1526 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1527 sub_pga = set->set_pga + req->rq_pgaidx;
1528 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1532 lov_update_common_set(set, req, rc);
1535 err = lov_fini_brw_set(set);
1541 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1544 struct lov_request_set *lovset = (struct lov_request_set *)data;
1548 lovset->set_completes = 0;
1549 lov_fini_brw_set(lovset);
1551 rc = lov_fini_brw_set(lovset);
1557 static int lov_brw_async(int cmd, struct obd_export *exp,
1558 struct obd_info *oinfo, obd_count oa_bufs,
1559 struct brw_page *pga, struct obd_trans_info *oti,
1560 struct ptlrpc_request_set *set)
1562 struct lov_request_set *lovset;
1563 struct lov_request *req;
1564 struct list_head *pos;
1565 struct lov_obd *lov = &exp->exp_obd->u.lov;
1570 ASSERT_LSM_MAGIC(oinfo->oi_md);
1572 if (cmd == OBD_BRW_CHECK) {
1573 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1577 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1581 list_for_each (pos, &lovset->set_list) {
1582 struct obd_export *sub_exp;
1583 struct brw_page *sub_pga;
1584 req = list_entry(pos, struct lov_request, rq_link);
1586 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1587 sub_pga = lovset->set_pga + req->rq_pgaidx;
1588 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1592 lov_update_common_set(lovset, req, rc);
1595 LASSERT(set->set_interpret == NULL);
1596 LASSERT(set->set_arg == NULL);
1597 rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1603 lov_fini_brw_set(lovset);
1607 static int lov_ap_make_ready(void *data, int cmd)
1609 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1611 return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1614 static int lov_ap_refresh_count(void *data, int cmd)
1616 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1618 return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1622 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1624 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1626 lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1627 /* XXX woah, shouldn't we be altering more here? size? */
1628 oa->o_id = lap->lap_loi_id;
1629 oa->o_gr = lap->lap_loi_gr;
1630 oa->o_valid |= OBD_MD_FLGROUP;
1631 oa->o_stripe_idx = lap->lap_stripe;
1634 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1637 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1639 lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1642 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1644 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1646 /* in a raid1 regime this would down a count of many ios
1647 * in flight, onl calling the caller_ops completion when all
1648 * the raid1 ios are complete */
1649 rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1653 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1655 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1656 return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1659 static struct obd_async_page_ops lov_async_page_ops = {
1660 .ap_make_ready = lov_ap_make_ready,
1661 .ap_refresh_count = lov_ap_refresh_count,
1662 .ap_fill_obdo = lov_ap_fill_obdo,
1663 .ap_update_obdo = lov_ap_update_obdo,
1664 .ap_completion = lov_ap_completion,
1665 .ap_lookup_capa = lov_ap_lookup_capa,
1668 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1669 struct lov_oinfo *loi, cfs_page_t *page,
1670 obd_off offset, struct obd_async_page_ops *ops,
1671 void *data, void **res)
1673 struct lov_obd *lov = &exp->exp_obd->u.lov;
1674 struct lov_async_page *lap;
1680 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1681 Only because of this layering limitation will a client
1682 mount with no osts fail */
1683 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1684 !lov->lov_tgts[i]->ltd_exp) {
1686 if (i >= lov->desc.ld_tgt_count)
1689 rc = size_round(sizeof(*lap)) +
1690 obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1691 NULL, NULL, 0, NULL, NULL, NULL);
1694 ASSERT_LSM_MAGIC(lsm);
1695 LASSERT(loi == NULL);
1698 lap->lap_magic = LOV_AP_MAGIC;
1699 lap->lap_caller_ops = ops;
1700 lap->lap_caller_data = data;
1702 /* for now only raid 0 which passes through */
1703 lap->lap_stripe = lov_stripe_number(lsm, offset);
1704 lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1705 loi = lsm->lsm_oinfo[lap->lap_stripe];
1707 /* so the callback doesn't need the lsm */
1708 lap->lap_loi_id = loi->loi_id;
1709 lap->lap_loi_gr = lsm->lsm_object_gr;
1710 LASSERT(lsm->lsm_object_gr > 0);
1712 lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1714 rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1715 lsm, loi, page, lap->lap_sub_offset,
1716 &lov_async_page_ops, lap,
1717 &lap->lap_sub_cookie);
1720 CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1721 lap->lap_sub_cookie, offset);
1725 static int lov_queue_async_io(struct obd_export *exp,
1726 struct lov_stripe_md *lsm,
1727 struct lov_oinfo *loi, void *cookie,
1728 int cmd, obd_off off, int count,
1729 obd_flag brw_flags, obd_flag async_flags)
1731 struct lov_obd *lov = &exp->exp_obd->u.lov;
1732 struct lov_async_page *lap;
1735 LASSERT(loi == NULL);
1737 ASSERT_LSM_MAGIC(lsm);
1739 lap = LAP_FROM_COOKIE(cookie);
1741 loi = lsm->lsm_oinfo[lap->lap_stripe];
1743 rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1744 loi, lap->lap_sub_cookie, cmd, off, count,
1745 brw_flags, async_flags);
1749 static int lov_set_async_flags(struct obd_export *exp,
1750 struct lov_stripe_md *lsm,
1751 struct lov_oinfo *loi, void *cookie,
1752 obd_flag async_flags)
1754 struct lov_obd *lov = &exp->exp_obd->u.lov;
1755 struct lov_async_page *lap;
1758 LASSERT(loi == NULL);
1760 ASSERT_LSM_MAGIC(lsm);
1762 lap = LAP_FROM_COOKIE(cookie);
1764 loi = lsm->lsm_oinfo[lap->lap_stripe];
1766 rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1767 lsm, loi, lap->lap_sub_cookie, async_flags);
1771 static int lov_queue_group_io(struct obd_export *exp,
1772 struct lov_stripe_md *lsm,
1773 struct lov_oinfo *loi,
1774 struct obd_io_group *oig, void *cookie,
1775 int cmd, obd_off off, int count,
1776 obd_flag brw_flags, obd_flag async_flags)
1778 struct lov_obd *lov = &exp->exp_obd->u.lov;
1779 struct lov_async_page *lap;
1782 LASSERT(loi == NULL);
1784 ASSERT_LSM_MAGIC(lsm);
1786 lap = LAP_FROM_COOKIE(cookie);
1788 loi = lsm->lsm_oinfo[lap->lap_stripe];
1790 rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1791 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1792 brw_flags, async_flags);
1796 /* this isn't exactly optimal. we may have queued sync io in oscs on
1797 * all stripes, but we don't record that fact at queue time. so we
1798 * trigger sync io on all stripes. */
1799 static int lov_trigger_group_io(struct obd_export *exp,
1800 struct lov_stripe_md *lsm,
1801 struct lov_oinfo *loi,
1802 struct obd_io_group *oig)
1804 struct lov_obd *lov = &exp->exp_obd->u.lov;
1807 LASSERT(loi == NULL);
1809 ASSERT_LSM_MAGIC(lsm);
1811 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1812 loi = lsm->lsm_oinfo[i];
1813 if (!lov->lov_tgts[loi->loi_ost_idx] ||
1814 !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1815 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1819 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1821 if (rc == 0 && err != 0)
1827 static int lov_teardown_async_page(struct obd_export *exp,
1828 struct lov_stripe_md *lsm,
1829 struct lov_oinfo *loi, void *cookie)
1831 struct lov_obd *lov = &exp->exp_obd->u.lov;
1832 struct lov_async_page *lap;
1835 LASSERT(loi == NULL);
1837 ASSERT_LSM_MAGIC(lsm);
1839 lap = LAP_FROM_COOKIE(cookie);
1841 loi = lsm->lsm_oinfo[lap->lap_stripe];
1843 rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1844 lsm, loi, lap->lap_sub_cookie);
1846 CERROR("unable to teardown sub cookie %p: %d\n",
1847 lap->lap_sub_cookie, rc);
1853 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1856 struct lov_request_set *lovset = (struct lov_request_set *)data;
1858 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1862 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1863 struct ldlm_enqueue_info *einfo,
1864 struct ptlrpc_request_set *rqset)
1866 ldlm_mode_t mode = einfo->ei_mode;
1867 struct lov_request_set *set;
1868 struct lov_request *req;
1869 struct list_head *pos;
1870 struct lov_obd *lov;
1875 ASSERT_LSM_MAGIC(oinfo->oi_md);
1876 LASSERT(mode == (mode & -mode));
1878 /* we should never be asked to replay a lock this way. */
1879 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1881 if (!exp || !exp->exp_obd)
1884 lov = &exp->exp_obd->u.lov;
1885 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1889 list_for_each (pos, &set->set_list) {
1890 req = list_entry(pos, struct lov_request, rq_link);
1892 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1893 &req->rq_oi, einfo, rqset);
1898 if (rqset && !list_empty(&rqset->set_requests)) {
1900 LASSERT(rqset->set_interpret == NULL);
1901 rqset->set_interpret = lov_enqueue_interpret;
1902 rqset->set_arg = (void *)set;
1906 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1910 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
1911 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1912 int *flags, void *data, struct lustre_handle *lockh)
1914 struct lov_request_set *set;
1915 struct obd_info oinfo;
1916 struct lov_request *req;
1917 struct list_head *pos;
1918 struct lov_obd *lov = &exp->exp_obd->u.lov;
1919 struct lustre_handle *lov_lockhp;
1920 int lov_flags, rc = 0;
1923 ASSERT_LSM_MAGIC(lsm);
1924 LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
1926 if (!exp || !exp->exp_obd)
1929 lov = &exp->exp_obd->u.lov;
1930 rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
1934 list_for_each (pos, &set->set_list) {
1935 ldlm_policy_data_t sub_policy;
1936 req = list_entry(pos, struct lov_request, rq_link);
1937 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1938 LASSERT(lov_lockhp);
1941 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
1943 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
1944 req->rq_oi.oi_md, type, &sub_policy,
1945 mode, &lov_flags, data, lov_lockhp);
1946 rc = lov_update_match_set(set, req, rc);
1950 lov_fini_match_set(set, mode, *flags);
1954 static int lov_change_cbdata(struct obd_export *exp,
1955 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1958 struct lov_obd *lov;
1962 ASSERT_LSM_MAGIC(lsm);
1964 if (!exp || !exp->exp_obd)
1967 LASSERT(lsm->lsm_object_gr > 0);
1969 lov = &exp->exp_obd->u.lov;
1970 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1971 struct lov_stripe_md submd;
1972 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1974 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1975 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1979 submd.lsm_object_id = loi->loi_id;
1980 submd.lsm_object_gr = lsm->lsm_object_gr;
1981 submd.lsm_stripe_count = 0;
1982 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1988 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1989 __u32 mode, struct lustre_handle *lockh)
1991 struct lov_request_set *set;
1992 struct obd_info oinfo;
1993 struct lov_request *req;
1994 struct list_head *pos;
1995 struct lov_obd *lov = &exp->exp_obd->u.lov;
1996 struct lustre_handle *lov_lockhp;
1997 int err = 0, rc = 0;
2000 ASSERT_LSM_MAGIC(lsm);
2002 if (!exp || !exp->exp_obd)
2005 LASSERT(lsm->lsm_object_gr > 0);
2007 lov = &exp->exp_obd->u.lov;
2008 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2012 list_for_each (pos, &set->set_list) {
2013 req = list_entry(pos, struct lov_request, rq_link);
2014 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2016 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2017 req->rq_oi.oi_md, mode, lov_lockhp);
2018 rc = lov_update_common_set(set, req, rc);
2020 CERROR("error: cancel objid "LPX64" subobj "
2021 LPX64" on OST idx %d: rc = %d\n",
2023 req->rq_oi.oi_md->lsm_object_id,
2029 lov_fini_cancel_set(set);
2033 static int lov_cancel_unused(struct obd_export *exp,
2034 struct lov_stripe_md *lsm,
2035 int flags, void *opaque)
2037 struct lov_obd *lov;
2041 if (!exp || !exp->exp_obd)
2044 lov = &exp->exp_obd->u.lov;
2046 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2048 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2051 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2059 ASSERT_LSM_MAGIC(lsm);
2061 LASSERT(lsm->lsm_object_gr > 0);
2062 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2063 struct lov_stripe_md submd;
2064 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2067 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2068 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2072 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2073 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2075 submd.lsm_object_id = loi->loi_id;
2076 submd.lsm_object_gr = lsm->lsm_object_gr;
2077 submd.lsm_stripe_count = 0;
2078 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2079 &submd, flags, opaque);
2080 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2081 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2082 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2083 loi->loi_id, loi->loi_ost_idx, err);
2091 static int lov_join_lru(struct obd_export *exp,
2092 struct lov_stripe_md *lsm, int join)
2094 struct lov_obd *lov;
2098 ASSERT_LSM_MAGIC(lsm);
2099 if (!exp || !exp->exp_obd)
2102 lov = &exp->exp_obd->u.lov;
2103 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2104 struct lov_stripe_md submd;
2105 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2108 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2109 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2113 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2114 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2116 submd.lsm_object_id = loi->loi_id;
2117 submd.lsm_object_gr = lsm->lsm_object_gr;
2118 submd.lsm_stripe_count = 0;
2119 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2122 CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2123 " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2124 loi->loi_id, loi->loi_ost_idx, rc);
2133 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2136 struct lov_request_set *lovset = (struct lov_request_set *)data;
2141 lovset->set_completes = 0;
2143 err = lov_fini_statfs_set(lovset);
2144 RETURN(rc ? rc : err);
2147 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2148 __u64 max_age, struct ptlrpc_request_set *rqset)
2150 struct lov_request_set *set;
2151 struct lov_request *req;
2152 struct list_head *pos;
2153 struct lov_obd *lov;
2157 LASSERT(oinfo != NULL);
2158 LASSERT(oinfo->oi_osfs != NULL);
2161 rc = lov_prep_statfs_set(obd, oinfo, &set);
2165 list_for_each (pos, &set->set_list) {
2166 struct obd_device *osc_obd;
2168 req = list_entry(pos, struct lov_request, rq_link);
2170 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2171 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2176 if (rc || list_empty(&rqset->set_requests)) {
2179 set->set_completes = 0;
2180 err = lov_fini_statfs_set(set);
2181 RETURN(rc ? rc : err);
2184 LASSERT(rqset->set_interpret == NULL);
2185 rqset->set_interpret = lov_statfs_interpret;
2186 rqset->set_arg = (void *)set;
2190 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2193 struct lov_obd *lov = &obd->u.lov;
2194 struct obd_statfs lov_sfs;
2200 /* We only get block data from the OBD */
2201 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2202 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
2203 CDEBUG(D_HA, "lov idx %d inactive\n", i);
2207 err = obd_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
2210 if (lov->lov_tgts[i]->ltd_active && !rc)
2214 lov_update_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
2215 osfs, &lov_sfs, set);
2219 err = lov_fini_statfs(obd, osfs, set);
2222 RETURN(rc ? rc : err);
2225 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2226 void *karg, void *uarg)
2228 struct obd_device *obddev = class_exp2obd(exp);
2229 struct lov_obd *lov = &obddev->u.lov;
2230 int i, rc, count = lov->desc.ld_tgt_count;
2231 struct obd_uuid *uuidp;
2235 case OBD_IOC_LOV_GET_CONFIG: {
2236 struct obd_ioctl_data *data;
2237 struct lov_desc *desc;
2242 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2245 data = (struct obd_ioctl_data *)buf;
2247 if (sizeof(*desc) > data->ioc_inllen1) {
2248 obd_ioctl_freedata(buf, len);
2252 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2253 obd_ioctl_freedata(buf, len);
2257 if (sizeof(__u32) * count > data->ioc_inllen3) {
2258 obd_ioctl_freedata(buf, len);
2262 desc = (struct lov_desc *)data->ioc_inlbuf1;
2263 memcpy(desc, &(lov->desc), sizeof(*desc));
2265 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2266 genp = (__u32 *)data->ioc_inlbuf3;
2267 /* the uuid will be empty for deleted OSTs */
2268 for (i = 0; i < count; i++, uuidp++, genp++) {
2269 if (!lov->lov_tgts[i])
2271 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2272 *genp = lov->lov_tgts[i]->ltd_gen;
2275 rc = copy_to_user((void *)uarg, buf, len);
2278 obd_ioctl_freedata(buf, len);
2281 case LL_IOC_LOV_SETSTRIPE:
2282 rc = lov_setstripe(exp, karg, uarg);
2284 case LL_IOC_LOV_GETSTRIPE:
2285 rc = lov_getstripe(exp, karg, uarg);
2287 case LL_IOC_LOV_SETEA:
2288 rc = lov_setea(exp, karg, uarg);
2297 for (i = 0; i < count; i++) {
2300 /* OST was disconnected */
2301 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2304 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2306 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2309 if (lov->lov_tgts[i]->ltd_active) {
2310 CDEBUG(err == -ENOTTY ?
2311 D_IOCTL : D_WARNING,
2312 "iocontrol OSC %s on OST "
2313 "idx %d cmd %x: err = %d\n",
2314 lov_uuid2str(lov, i),
2331 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2332 void *key, __u32 *vallen, void *val)
2334 struct obd_device *obddev = class_exp2obd(exp);
2335 struct lov_obd *lov = &obddev->u.lov;
2339 if (!vallen || !val)
2344 if (KEY_IS("lock_to_stripe")) {
2347 struct ldlm_lock *lock;
2348 struct lov_stripe_md *lsm;
2350 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2351 struct lov_oinfo *loi;
2352 __u32 *stripe = val;
2354 if (*vallen < sizeof(*stripe))
2355 GOTO(out, rc = -EFAULT);
2356 *vallen = sizeof(*stripe);
2358 /* XXX This is another one of those bits that will need to
2359 * change if we ever actually support nested LOVs. It uses
2360 * the lock's export to find out which stripe it is. */
2361 /* XXX - it's assumed all the locks for deleted OSTs have
2362 * been cancelled. Also, the export for deleted OSTs will
2363 * be NULL and won't match the lock's export. */
2364 for (i = 0; i < data->lsm->lsm_stripe_count; i++) {
2365 loi = data->lsm->lsm_oinfo[i];
2366 if (!lov->lov_tgts[loi->loi_ost_idx])
2368 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2369 data->lock->l_conn_export &&
2370 loi->loi_id == res_id->name[0] &&
2371 loi->loi_gr == res_id->name[2]) {
2376 LDLM_ERROR(data->lock, "lock on inode without such object");
2377 dump_lsm(D_ERROR, data->lsm);
2378 GOTO(out, rc = -ENXIO);
2379 } else if (KEY_IS(KEY_LAST_ID)) {
2380 struct obd_id_info *info = val;
2381 __u32 size = sizeof(obd_id);
2382 struct lov_tgt_desc *tgt;
2384 LASSERT(*vallen == sizeof(struct obd_id_info));
2385 tgt = lov->lov_tgts[info->idx];
2387 if (!tgt || !tgt->ltd_active)
2388 GOTO(out, rc = -ESRCH);
2390 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data);
2392 } else if (KEY_IS(KEY_LOVDESC)) {
2393 struct lov_desc *desc_ret = val;
2394 *desc_ret = lov->desc;
2397 } else if (KEY_IS(KEY_LOV_IDX)) {
2398 struct lov_tgt_desc *tgt;
2400 for(i = 0; i < lov->desc.ld_tgt_count; i++) {
2401 tgt = lov->lov_tgts[i];
2402 if (tgt && obd_uuid_equals(val, &tgt->ltd_uuid))
2413 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2414 void *key, obd_count vallen, void *val,
2415 struct ptlrpc_request_set *set)
2417 struct obd_device *obddev = class_exp2obd(exp);
2418 struct lov_obd *lov = &obddev->u.lov;
2421 struct lov_tgt_desc *tgt;
2422 unsigned incr, check_uuid,
2423 do_inactive, no_set;
2424 unsigned next_id = 0, mds_con = 0;
2427 incr = check_uuid = do_inactive = no_set = 0;
2430 set = ptlrpc_prep_set();
2436 count = lov->desc.ld_tgt_count;
2438 if (KEY_IS(KEY_NEXT_ID)) {
2439 count = vallen / sizeof(struct obd_id_info);
2440 vallen = sizeof(obd_id);
2441 incr = sizeof(struct obd_id_info);
2444 } else if (KEY_IS("checksum")) {
2446 } else if (KEY_IS("unlinked")) {
2447 check_uuid = val ? 1 : 0;
2448 } else if (KEY_IS("evict_by_nid")) {
2449 /* use defaults: do_inactive = incr = 0; */
2450 } else if (KEY_IS(KEY_MDS_CONN)) {
2454 for (i = 0; i < count; i++, val = (char *)val + incr) {
2456 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2458 tgt = lov->lov_tgts[i];
2460 /* OST was disconnected */
2461 if (!tgt || !tgt->ltd_exp)
2464 /* OST is inactive and we don't want inactive OSCs */
2465 if (!tgt->ltd_active && !do_inactive)
2469 struct mds_group_info *mgi;
2471 LASSERT(vallen == sizeof(*mgi));
2472 mgi = (struct mds_group_info *)val;
2474 /* Only want a specific OSC */
2475 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2479 err = obd_set_info_async(tgt->ltd_exp,
2480 keylen, key, sizeof(int),
2482 } else if (next_id) {
2483 err = obd_set_info_async(tgt->ltd_exp,
2484 keylen, key, vallen,
2485 ((struct obd_id_info*)val)->data, set);
2487 /* Only want a specific OSC */
2489 !obd_uuid_equals(val, &tgt->ltd_uuid))
2492 err = obd_set_info_async(tgt->ltd_exp,
2493 keylen, key, vallen, val, set);
2502 err = ptlrpc_set_wait(set);
2505 ptlrpc_set_destroy(set);
2510 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2511 struct lov_stripe_md *lsm)
2519 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2520 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2525 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2530 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2531 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2532 if (loi->loi_ar.ar_rc && !rc)
2533 rc = loi->loi_ar.ar_rc;
2534 loi->loi_ar.ar_rc = 0;
2538 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2541 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2542 int cmd, __u64 *offset)
2544 __u32 ssize = lsm->lsm_stripe_size;
2548 do_div(start, ssize);
2549 start = start * ssize;
2551 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2552 ", end "LPU64"\n", *offset, ssize, start,
2554 if (cmd == OBD_CALC_STRIPE_END) {
2555 *offset = start + ssize - 1;
2556 } else if (cmd == OBD_CALC_STRIPE_START) {
2567 struct lov_multi_wait {
2568 struct ldlm_lock *lock;
2574 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2575 struct lustre_handle *lockh)
2577 struct lov_lock_handles *lov_lockh = NULL;
2578 struct lustre_handle *lov_lockhp;
2579 struct lov_obd *lov;
2580 struct lov_oinfo *loi;
2581 struct lov_multi_wait *queues;
2585 ASSERT_LSM_MAGIC(lsm);
2587 if (!exp || !exp->exp_obd)
2590 LASSERT(lockh != NULL);
2591 if (lsm->lsm_stripe_count > 1) {
2592 lov_lockh = lov_handle2llh(lockh);
2593 if (lov_lockh == NULL) {
2594 CERROR("LOV: invalid lov lock handle %p\n", lockh);
2598 lov_lockhp = lov_lockh->llh_handles;
2603 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2605 GOTO(out, rc = -ENOMEM);
2607 lov = &exp->exp_obd->u.lov;
2608 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2609 i++, loi++, lov_lockhp++) {
2610 struct ldlm_lock *lock;
2611 struct obd_device *obd;
2613 lock = ldlm_handle2lock(lov_lockhp);
2615 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2616 loi->loi_ost_idx, loi->loi_id);
2617 queues[i].completed = 1;
2621 queues[i].lock = lock;
2622 init_waitqueue_entry(&(queues[i].wait), current);
2623 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2625 obd = class_exp2obd(lock->l_conn_export);
2627 imp = obd->u.cli.cl_import;
2629 spin_lock(&imp->imp_lock);
2630 queues[i].generation = imp->imp_generation;
2631 spin_unlock(&imp->imp_lock);
2635 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2636 interrupted_completion_wait, &lwd);
2637 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2639 for (i = 0; i < lsm->lsm_stripe_count; i++)
2640 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2642 if (rc == -EINTR || rc == -ETIMEDOUT) {
2648 if (lov_lockh != NULL)
2649 lov_llh_put(lov_lockh);
2654 void lov_stripe_lock(struct lov_stripe_md *md)
2656 LASSERT(md->lsm_lock_owner != cfs_current());
2657 spin_lock(&md->lsm_lock);
2658 LASSERT(md->lsm_lock_owner == NULL);
2659 md->lsm_lock_owner = cfs_current();
2661 EXPORT_SYMBOL(lov_stripe_lock);
2663 void lov_stripe_unlock(struct lov_stripe_md *md)
2665 LASSERT(md->lsm_lock_owner == cfs_current());
2666 md->lsm_lock_owner = NULL;
2667 spin_unlock(&md->lsm_lock);
2669 EXPORT_SYMBOL(lov_stripe_unlock);
2671 struct obd_ops lov_obd_ops = {
2672 .o_owner = THIS_MODULE,
2673 .o_setup = lov_setup,
2674 .o_precleanup = lov_precleanup,
2675 .o_cleanup = lov_cleanup,
2676 .o_process_config = lov_process_config,
2677 .o_connect = lov_connect,
2678 .o_disconnect = lov_disconnect,
2679 .o_statfs = lov_statfs,
2680 .o_statfs_async = lov_statfs_async,
2681 .o_packmd = lov_packmd,
2682 .o_unpackmd = lov_unpackmd,
2683 .o_checkmd = lov_checkmd,
2684 .o_create = lov_create,
2685 .o_destroy = lov_destroy,
2686 .o_getattr = lov_getattr,
2687 .o_getattr_async = lov_getattr_async,
2688 .o_setattr = lov_setattr,
2689 .o_setattr_async = lov_setattr_async,
2691 .o_brw_async = lov_brw_async,
2692 .o_prep_async_page = lov_prep_async_page,
2693 .o_queue_async_io = lov_queue_async_io,
2694 .o_set_async_flags = lov_set_async_flags,
2695 .o_queue_group_io = lov_queue_group_io,
2696 .o_trigger_group_io = lov_trigger_group_io,
2697 .o_teardown_async_page = lov_teardown_async_page,
2698 .o_merge_lvb = lov_merge_lvb,
2699 .o_adjust_kms = lov_adjust_kms,
2700 .o_punch = lov_punch,
2702 .o_enqueue = lov_enqueue,
2703 .o_match = lov_match,
2704 .o_change_cbdata = lov_change_cbdata,
2705 .o_cancel = lov_cancel,
2706 .o_cancel_unused = lov_cancel_unused,
2707 .o_join_lru = lov_join_lru,
2708 .o_iocontrol = lov_iocontrol,
2709 .o_get_info = lov_get_info,
2710 .o_set_info_async = lov_set_info_async,
2711 .o_extent_calc = lov_extent_calc,
2712 .o_llog_init = lov_llog_init,
2713 .o_llog_finish = lov_llog_finish,
2714 .o_notify = lov_notify,
2717 static quota_interface_t *quota_interface;
2718 extern quota_interface_t lov_quota_interface;
2720 cfs_mem_cache_t *lov_oinfo_slab;
2722 int __init lov_init(void)
2724 struct lprocfs_static_vars lvars = { 0 };
2728 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2729 sizeof(struct lov_oinfo),
2730 0, SLAB_HWCACHE_ALIGN);
2731 if (lov_oinfo_slab == NULL)
2733 lprocfs_lov_init_vars(&lvars);
2735 request_module("lquota");
2736 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2737 init_obd_quota_ops(quota_interface, &lov_obd_ops);
2739 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2740 LUSTRE_LOV_NAME, NULL);
2742 if (quota_interface)
2743 PORTAL_SYMBOL_PUT(lov_quota_interface);
2744 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2752 static void /*__exit*/ lov_exit(void)
2756 if (quota_interface)
2757 PORTAL_SYMBOL_PUT(lov_quota_interface);
2759 class_unregister_type(LUSTRE_LOV_NAME);
2760 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2764 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2765 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2766 MODULE_LICENSE("GPL");
2768 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);