1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_obd.c
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
45 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_LOV
49 #include <libcfs/libcfs.h>
51 #include <liblustre.h>
54 #include <obd_support.h>
55 #include <lustre_lib.h>
56 #include <lustre_net.h>
57 #include <lustre/lustre_idl.h>
58 #include <lustre_dlm.h>
59 #include <lustre_mds.h>
60 #include <lustre_debug.h>
61 #include <obd_class.h>
64 #include <lprocfs_status.h>
65 #include <lustre_param.h>
66 #include <lustre_cache.h>
67 #include <lustre/ll_fiemap.h>
69 #include "lov_internal.h"
72 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
73 Any function that expects lov_tgts to remain stationary must take a ref. */
74 void lov_getref(struct obd_device *obd)
76 struct lov_obd *lov = &obd->u.lov;
78 /* nobody gets through here until lov_putref is done */
79 mutex_down(&lov->lov_lock);
80 atomic_inc(&lov->lov_refcount);
81 mutex_up(&lov->lov_lock);
85 static void __lov_del_obd(struct obd_device *obd, __u32 index);
87 void lov_putref(struct obd_device *obd)
89 struct lov_obd *lov = &obd->u.lov;
90 mutex_down(&lov->lov_lock);
91 /* ok to dec to 0 more than once -- ltd_exp's will be null */
92 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
94 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
96 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
97 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap)
99 /* Disconnect and delete from list */
100 __lov_del_obd(obd, i);
101 lov->lov_death_row--;
104 mutex_up(&lov->lov_lock);
107 static int lov_register_page_removal_cb(struct obd_export *exp,
108 obd_page_removal_cb_t func,
109 obd_pin_extent_cb pin_cb)
111 struct lov_obd *lov = &exp->exp_obd->u.lov;
114 if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
117 if (lov->lov_page_pin_cb && lov->lov_page_pin_cb != pin_cb)
120 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
121 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
123 rc |= obd_register_page_removal_cb(lov->lov_tgts[i]->ltd_exp,
127 lov->lov_page_removal_cb = func;
128 lov->lov_page_pin_cb = pin_cb;
133 static int lov_unregister_page_removal_cb(struct obd_export *exp,
134 obd_page_removal_cb_t func)
136 struct lov_obd *lov = &exp->exp_obd->u.lov;
139 if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
142 lov->lov_page_removal_cb = NULL;
143 lov->lov_page_pin_cb = NULL;
145 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
146 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
148 rc |= obd_unregister_page_removal_cb(lov->lov_tgts[i]->ltd_exp,
155 static int lov_register_lock_cancel_cb(struct obd_export *exp,
156 obd_lock_cancel_cb func)
158 struct lov_obd *lov = &exp->exp_obd->u.lov;
161 if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
164 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
165 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
167 rc |= obd_register_lock_cancel_cb(lov->lov_tgts[i]->ltd_exp,
171 lov->lov_lock_cancel_cb = func;
176 static int lov_unregister_lock_cancel_cb(struct obd_export *exp,
177 obd_lock_cancel_cb func)
179 struct lov_obd *lov = &exp->exp_obd->u.lov;
182 if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
185 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
186 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
188 rc |= obd_unregister_lock_cancel_cb(lov->lov_tgts[i]->ltd_exp,
191 lov->lov_lock_cancel_cb = NULL;
195 #define MAX_STRING_SIZE 128
196 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
197 struct obd_connect_data *data)
199 struct lov_obd *lov = &obd->u.lov;
200 struct obd_uuid tgt_uuid;
201 struct obd_device *tgt_obd;
202 struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
203 struct lustre_handle conn = {0, };
204 struct obd_import *imp;
207 cfs_proc_dir_entry_t *lov_proc_dir;
212 if (!lov->lov_tgts[index])
215 tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
217 tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
221 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
224 if (!tgt_obd->obd_set_up) {
225 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
229 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
230 data->ocd_index = index;
233 * Divine LOV knows that OBDs under it are OSCs.
235 imp = tgt_obd->u.cli.cl_import;
238 tgt_obd->obd_no_recov = 0;
239 /* FIXME this is probably supposed to be
240 ptlrpc_set_import_active. Horrible naming. */
241 ptlrpc_activate_import(imp);
244 if (imp->imp_invalid) {
245 CERROR("not connecting OSC %s; administratively "
246 "disabled\n", obd_uuid2str(&tgt_uuid));
247 rc = obd_register_observer(tgt_obd, obd);
249 CERROR("Target %s register_observer error %d; "
250 "will not be able to reactivate\n",
251 obd_uuid2str(&tgt_uuid), rc);
256 rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data, NULL);
258 CERROR("Target %s connect error %d\n",
259 obd_uuid2str(&tgt_uuid), rc);
262 lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
263 if (!lov->lov_tgts[index]->ltd_exp) {
264 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
268 rc = obd_register_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
269 lov->lov_page_removal_cb,
270 lov->lov_page_pin_cb);
272 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
273 lov->lov_tgts[index]->ltd_exp = NULL;
277 rc = obd_register_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
278 lov->lov_lock_cancel_cb);
280 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
281 lov->lov_page_removal_cb);
282 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
283 lov->lov_tgts[index]->ltd_exp = NULL;
287 rc = obd_register_observer(tgt_obd, obd);
289 CERROR("Target %s register_observer error %d\n",
290 obd_uuid2str(&tgt_uuid), rc);
291 obd_unregister_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
292 lov->lov_lock_cancel_cb);
293 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
294 lov->lov_page_removal_cb);
295 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
296 lov->lov_tgts[index]->ltd_exp = NULL;
300 lov->lov_tgts[index]->ltd_reap = 0;
302 lov->lov_tgts[index]->ltd_active = 1;
303 lov->desc.ld_active_tgt_count++;
304 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
306 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
307 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
310 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
312 struct obd_device *osc_obd = class_conn2obd(&conn);
313 cfs_proc_dir_entry_t *osc_symlink;
314 char name[MAX_STRING_SIZE];
316 LASSERT(osc_obd != NULL);
317 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
318 LASSERT(osc_obd->obd_type->typ_name != NULL);
319 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
320 osc_obd->obd_type->typ_name,
322 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
324 if (osc_symlink == NULL) {
325 CERROR("could not register LOV target "
326 "/proc/fs/lustre/%s/%s/target_obds/%s.",
327 obd->obd_type->typ_name, obd->obd_name,
329 lprocfs_remove(&lov_proc_dir);
334 rc = qos_add_tgt(obd, index);
336 CERROR("qos_add_tgt failed %d\n", rc);
341 static int lov_connect(const struct lu_env *env,
342 struct lustre_handle *conn, struct obd_device *obd,
343 struct obd_uuid *cluuid, struct obd_connect_data *data,
346 struct lov_obd *lov = &obd->u.lov;
347 struct lov_tgt_desc *tgt;
351 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
353 rc = class_connect(conn, obd, cluuid);
357 /* Why should there ever be more than 1 connect? */
359 LASSERT(lov->lov_connects == 1);
361 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
363 lov->lov_ocd = *data;
366 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
367 tgt = lov->lov_tgts[i];
368 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
370 /* Flags will be lowest common denominator */
371 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
373 CERROR("%s: lov connect tgt %d failed: %d\n",
374 obd->obd_name, i, rc);
383 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
385 cfs_proc_dir_entry_t *lov_proc_dir;
386 struct lov_obd *lov = &obd->u.lov;
387 struct obd_device *osc_obd;
392 if (lov->lov_tgts[index] == NULL)
395 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
396 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
397 obd->obd_name, osc_obd->obd_name);
399 obd_unregister_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
400 lov->lov_lock_cancel_cb);
401 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
402 lov->lov_page_removal_cb);
404 if (lov->lov_tgts[index]->ltd_active) {
405 lov->lov_tgts[index]->ltd_active = 0;
406 lov->desc.ld_active_tgt_count--;
407 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
410 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
412 cfs_proc_dir_entry_t *osc_symlink;
414 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
416 lprocfs_remove(&osc_symlink);
418 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
419 obd->obd_type->typ_name, obd->obd_name,
425 /* Pass it on to our clients.
426 * XXX This should be an argument to disconnect,
427 * XXX not a back-door flag on the OBD. Ah well.
429 osc_obd->obd_force = obd->obd_force;
430 osc_obd->obd_fail = obd->obd_fail;
431 osc_obd->obd_no_recov = obd->obd_no_recov;
434 obd_register_observer(osc_obd, NULL);
436 rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
438 CERROR("Target %s disconnect error %d\n",
439 lov_uuid2str(lov, index), rc);
443 qos_del_tgt(obd, index);
445 lov->lov_tgts[index]->ltd_exp = NULL;
449 static int lov_del_target(struct obd_device *obd, __u32 index,
450 struct obd_uuid *uuidp, int gen);
452 static int lov_disconnect(struct obd_export *exp)
454 struct obd_device *obd = class_exp2obd(exp);
455 struct lov_obd *lov = &obd->u.lov;
462 /* Only disconnect the underlying layers on the final disconnect. */
464 if (lov->lov_connects != 0) {
465 /* why should there be more than 1 connect? */
466 CERROR("disconnect #%d\n", lov->lov_connects);
470 /* Let's hold another reference so lov_del_obd doesn't spin through
473 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
474 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
475 /* Disconnection is the last we know about an obd */
476 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
482 rc = class_disconnect(exp); /* bz 9811 */
488 * -EINVAL : UUID can't be found in the LOV's target list
489 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
490 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
492 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
495 struct lov_obd *lov = &obd->u.lov;
496 struct lov_tgt_desc *tgt;
500 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
501 lov, uuid->uuid, activate);
504 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
505 tgt = lov->lov_tgts[i];
506 if (!tgt || !tgt->ltd_exp)
509 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
510 i, obd_uuid2str(&tgt->ltd_uuid),
511 tgt->ltd_exp->exp_handle.h_cookie);
512 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
516 if (i == lov->desc.ld_tgt_count)
517 GOTO(out, rc = -EINVAL);
519 if (lov->lov_tgts[i]->ltd_active == activate) {
520 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
521 activate ? "" : "in");
525 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
526 activate ? "" : "in");
528 lov->lov_tgts[i]->ltd_active = activate;
531 lov->desc.ld_active_tgt_count++;
532 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
534 lov->desc.ld_active_tgt_count--;
535 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
537 /* remove any old qos penalty */
538 lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
545 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
546 enum obd_notify_event ev, void *data)
551 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
552 struct obd_uuid *uuid;
556 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
557 CERROR("unexpected notification of %s %s!\n",
558 watched->obd_type->typ_name,
562 uuid = &watched->u.cli.cl_target_uuid;
564 /* Set OSC as active before notifying the observer, so the
565 * observer can use the OSC normally.
567 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
569 CERROR("%sactivation of %s failed: %d\n",
570 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
571 obd_uuid2str(uuid), rc);
576 /* Pass the notification up the chain. */
578 rc = obd_notify_observer(obd, watched, ev, data);
580 /* NULL watched means all osc's in the lov (only for syncs) */
581 struct lov_obd *lov = &obd->u.lov;
582 struct obd_device *tgt_obd;
585 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
586 if (!lov->lov_tgts[i])
588 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
589 rc = obd_notify_observer(obd, tgt_obd, ev, data);
591 CERROR("%s: notify %s of %s failed %d\n",
593 obd->obd_observer->obd_name,
594 tgt_obd->obd_name, rc);
604 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
605 __u32 index, int gen, int active)
607 struct lov_obd *lov = &obd->u.lov;
608 struct lov_tgt_desc *tgt;
612 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
613 uuidp->uuid, index, gen, active);
616 CERROR("request to add OBD %s with invalid generation: %d\n",
621 mutex_down(&lov->lov_lock);
623 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
624 tgt = lov->lov_tgts[index];
625 CERROR("UUID %s already assigned at LOV target index %d\n",
626 obd_uuid2str(&tgt->ltd_uuid), index);
627 mutex_up(&lov->lov_lock);
631 if (index >= lov->lov_tgt_size) {
632 /* We need to reallocate the lov target array. */
633 struct lov_tgt_desc **newtgts, **old = NULL;
634 __u32 newsize, oldsize = 0;
636 newsize = max(lov->lov_tgt_size, (__u32)2);
637 while (newsize < index + 1)
638 newsize = newsize << 1;
639 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
640 if (newtgts == NULL) {
641 mutex_up(&lov->lov_lock);
645 if (lov->lov_tgt_size) {
646 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
649 oldsize = lov->lov_tgt_size;
652 lov->lov_tgts = newtgts;
653 lov->lov_tgt_size = newsize;
658 OBD_FREE(old, sizeof(*old) * oldsize);
660 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
661 lov->lov_tgts, lov->lov_tgt_size);
667 mutex_up(&lov->lov_lock);
671 memset(tgt, 0, sizeof(*tgt));
672 tgt->ltd_uuid = *uuidp;
673 /* XXX - add a sanity check on the generation number. */
675 tgt->ltd_index = index;
676 tgt->ltd_activate = active;
677 lov->lov_tgts[index] = tgt;
678 if (index >= lov->desc.ld_tgt_count)
679 lov->desc.ld_tgt_count = index + 1;
680 mutex_up(&lov->lov_lock);
682 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
683 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
685 if (lov->lov_connects == 0) {
686 /* lov_connect hasn't been called yet. We'll do the
687 lov_connect_obd on this target when that fn first runs,
688 because we don't know the connect flags yet. */
694 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
698 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
699 active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
704 CERROR("add failed (%d), deleting %s\n", rc,
705 obd_uuid2str(&tgt->ltd_uuid));
706 lov_del_target(obd, index, 0, 0);
712 /* Schedule a target for deletion */
713 static int lov_del_target(struct obd_device *obd, __u32 index,
714 struct obd_uuid *uuidp, int gen)
716 struct lov_obd *lov = &obd->u.lov;
717 int count = lov->desc.ld_tgt_count;
721 if (index >= count) {
722 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
729 if (!lov->lov_tgts[index]) {
730 CERROR("LOV target at index %d is not setup.\n", index);
731 GOTO(out, rc = -EINVAL);
734 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
735 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
736 lov_uuid2str(lov, index), index,
737 obd_uuid2str(uuidp));
738 GOTO(out, rc = -EINVAL);
741 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
742 lov_uuid2str(lov, index), index,
743 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
744 lov->lov_tgts[index]->ltd_active);
746 lov->lov_tgts[index]->ltd_reap = 1;
747 lov->lov_death_row++;
748 /* we really delete it from lov_putref */
755 /* We are holding lov_lock */
756 static void __lov_del_obd(struct obd_device *obd, __u32 index)
758 struct lov_obd *lov = &obd->u.lov;
759 struct obd_device *osc_obd;
760 struct lov_tgt_desc *tgt = lov->lov_tgts[index];
763 LASSERT(tgt->ltd_reap);
765 osc_obd = class_exp2obd(tgt->ltd_exp);
767 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
768 lov_uuid2str(lov, index),
769 osc_obd ? osc_obd->obd_name : "<no obd>");
772 lov_disconnect_obd(obd, index);
774 /* XXX - right now there is a dependency on ld_tgt_count being the
775 * maximum tgt index for computing the mds_max_easize. So we can't
778 lov->lov_tgts[index] = NULL;
781 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
782 do it ourselves. And we can't do it from lov_cleanup,
783 because we just lost our only reference to it. */
785 class_manual_cleanup(osc_obd);
788 void lov_fix_desc_stripe_size(__u64 *val)
790 if (*val < PTLRPC_MAX_BRW_SIZE) {
791 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
792 PTLRPC_MAX_BRW_SIZE);
793 *val = PTLRPC_MAX_BRW_SIZE;
794 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
795 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
796 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
798 *val, LOV_MIN_STRIPE_SIZE);
802 void lov_fix_desc_stripe_count(__u32 *val)
808 void lov_fix_desc_pattern(__u32 *val)
810 /* from lov_setstripe */
811 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
812 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
817 void lov_fix_desc_qos_maxage(__u32 *val)
821 *val = QOS_DEFAULT_MAXAGE;
824 void lov_fix_desc(struct lov_desc *desc)
826 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
827 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
828 lov_fix_desc_pattern(&desc->ld_pattern);
829 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
832 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
834 struct lprocfs_static_vars lvars = { 0 };
835 struct lov_desc *desc;
836 struct lov_obd *lov = &obd->u.lov;
840 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
841 CERROR("LOV setup requires a descriptor\n");
845 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
847 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
848 CERROR("descriptor size wrong: %d > %d\n",
849 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
853 if (desc->ld_magic != LOV_DESC_MAGIC) {
854 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
855 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
856 obd->obd_name, desc);
857 lustre_swab_lov_desc(desc);
859 CERROR("%s: Bad lov desc magic: %#x\n",
860 obd->obd_name, desc->ld_magic);
867 /* Because of 64-bit divide/mod operations only work with a 32-bit
868 * divisor in a 32-bit kernel, we cannot support a stripe width
869 * of 4GB or larger on 32-bit CPUs. */
870 count = desc->ld_default_stripe_count;
871 if ((count > 0 ? count : desc->ld_tgt_count) *
872 desc->ld_default_stripe_size > 0xffffffff) {
873 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
874 desc->ld_default_stripe_size, count);
878 desc->ld_active_tgt_count = 0;
880 lov->lov_tgt_size = 0;
881 sema_init(&lov->lov_lock, 1);
882 atomic_set(&lov->lov_refcount, 0);
883 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
884 init_rwsem(&lov->lov_qos.lq_rw_sem);
885 lov->lov_qos.lq_dirty = 1;
886 lov->lov_qos.lq_dirty_rr = 1;
887 lov->lov_qos.lq_reset = 1;
888 /* Default priority is toward free space balance */
889 lov->lov_qos.lq_prio_free = 232;
891 lprocfs_lov_init_vars(&lvars);
892 lprocfs_obd_setup(obd, lvars.obd_vars);
897 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
898 0444, &lov_proc_target_fops, obd);
900 CWARN("Error adding the target_obd file\n");
907 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
913 case OBD_CLEANUP_EARLY: {
914 struct lov_obd *lov = &obd->u.lov;
916 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
917 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
919 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
924 case OBD_CLEANUP_EXPORTS:
925 rc = obd_llog_finish(obd, 0);
927 CERROR("failed to cleanup llogging subsystems\n");
933 static int lov_cleanup(struct obd_device *obd)
935 struct lov_obd *lov = &obd->u.lov;
937 lprocfs_obd_cleanup(obd);
940 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
941 if (!lov->lov_tgts[i])
944 /* Inactive targets may never have connected */
945 if (lov->lov_tgts[i]->ltd_active ||
946 atomic_read(&lov->lov_refcount))
947 /* We should never get here - these
948 should have been removed in the
950 CERROR("lov tgt %d not cleaned!"
951 " deathrow=%d, lovrc=%d\n",
952 i, lov->lov_death_row,
953 atomic_read(&lov->lov_refcount));
954 lov_del_target(obd, i, 0, 0);
956 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
958 lov->lov_tgt_size = 0;
961 if (lov->lov_qos.lq_rr_size)
962 OBD_FREE(lov->lov_qos.lq_rr_array, lov->lov_qos.lq_rr_size);
967 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
969 struct lustre_cfg *lcfg = buf;
970 struct obd_uuid obd_uuid;
975 switch(cmd = lcfg->lcfg_command) {
976 case LCFG_LOV_ADD_OBD:
977 case LCFG_LOV_ADD_INA:
978 case LCFG_LOV_DEL_OBD: {
981 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
982 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
983 GOTO(out, rc = -EINVAL);
985 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
987 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
988 GOTO(out, rc = -EINVAL);
989 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
990 GOTO(out, rc = -EINVAL);
991 if (cmd == LCFG_LOV_ADD_OBD)
992 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
993 else if (cmd == LCFG_LOV_ADD_INA)
994 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
996 rc = lov_del_target(obd, index, &obd_uuid, gen);
1000 struct lprocfs_static_vars lvars = { 0 };
1001 struct lov_desc *desc = &(obd->u.lov.desc);
1004 GOTO(out, rc = -EINVAL);
1006 lprocfs_lov_init_vars(&lvars);
1008 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1013 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1014 GOTO(out, rc = -EINVAL);
1023 #define log2(n) ffz(~(n))
1026 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1027 struct lov_stripe_md **ea,
1028 struct obd_trans_info *oti)
1030 struct lov_obd *lov;
1031 struct obdo *tmp_oa;
1032 struct obd_uuid *ost_uuid = NULL;
1036 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1037 src_oa->o_flags == OBD_FL_DELORPHAN);
1039 lov = &export->exp_obd->u.lov;
1045 if (src_oa->o_valid & OBD_MD_FLINLINE) {
1046 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
1047 CDEBUG(D_HA, "clearing orphans only for %s\n",
1051 lov_getref(export->exp_obd);
1052 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1053 struct lov_stripe_md obj_md;
1054 struct lov_stripe_md *obj_mdp = &obj_md;
1055 struct lov_tgt_desc *tgt;
1058 tgt = lov->lov_tgts[i];
1062 /* if called for a specific target, we don't
1063 care if it is not active. */
1064 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1065 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1069 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1072 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1073 obd_uuid2str(ost_uuid));
1075 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1077 LASSERT(lov->lov_tgts[i]->ltd_exp);
1078 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1079 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1080 tmp_oa, &obj_mdp, oti);
1082 /* This export will be disabled until it is recovered,
1083 and then orphan recovery will be completed. */
1084 CERROR("error in orphan recovery on OST idx %d/%d: "
1085 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1090 lov_putref(export->exp_obd);
1096 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1097 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1099 struct lov_stripe_md *obj_mdp, *lsm;
1100 struct lov_obd *lov = &exp->exp_obd->u.lov;
1105 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1106 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1108 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1109 if (obj_mdp == NULL)
1112 ost_idx = src_oa->o_nlink;
1115 GOTO(out, rc = -EINVAL);
1116 if (ost_idx >= lov->desc.ld_tgt_count ||
1117 !lov->lov_tgts[ost_idx])
1118 GOTO(out, rc = -EINVAL);
1120 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1121 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1122 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1123 GOTO(out, rc = -EINVAL);
1127 if (i == lsm->lsm_stripe_count)
1128 GOTO(out, rc = -EINVAL);
1130 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1132 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1136 /* the LOV expects oa->o_id to be set to the LOV object id */
1137 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1138 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1140 struct lov_obd *lov;
1141 struct obd_info oinfo;
1142 struct lov_request_set *set = NULL;
1143 struct lov_request *req;
1144 struct obd_statfs osfs;
1149 LASSERT(ea != NULL);
1153 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1154 src_oa->o_flags == OBD_FL_DELORPHAN) {
1155 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1159 lov = &exp->exp_obd->u.lov;
1160 if (!lov->desc.ld_active_tgt_count)
1163 /* Recreate a specific object id at the given OST index */
1164 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1165 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1166 rc = lov_recreate(exp, src_oa, ea, oti);
1170 maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1171 obd_statfs_rqset(exp->exp_obd, &osfs, maxage, OBD_STATFS_NODELAY);
1173 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1177 list_for_each_entry(req, &set->set_list, rq_link) {
1178 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1179 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1180 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1181 lov_update_create_set(set, req, rc);
1183 rc = lov_fini_create_set(set, ea);
1187 #define ASSERT_LSM_MAGIC(lsmp) \
1189 LASSERT((lsmp) != NULL); \
1190 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC || \
1191 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1192 (lsmp), (lsmp)->lsm_magic); \
1195 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1196 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1197 struct obd_export *md_exp)
1199 struct lov_request_set *set;
1200 struct obd_info oinfo;
1201 struct lov_request *req;
1202 struct list_head *pos;
1203 struct lov_obd *lov;
1207 ASSERT_LSM_MAGIC(lsm);
1209 if (!exp || !exp->exp_obd)
1212 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1214 LASSERT(oti->oti_logcookies);
1217 lov = &exp->exp_obd->u.lov;
1218 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1222 list_for_each (pos, &set->set_list) {
1224 req = list_entry(pos, struct lov_request, rq_link);
1226 if (oa->o_valid & OBD_MD_FLCOOKIE)
1227 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1229 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1230 req->rq_oi.oi_oa, NULL, oti, NULL);
1231 err = lov_update_common_set(set, req, err);
1233 CERROR("error: destroying objid "LPX64" subobj "
1234 LPX64" on OST idx %d: rc = %d\n",
1235 oa->o_id, req->rq_oi.oi_oa->o_id,
1243 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1244 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1246 err = lov_fini_destroy_set(set);
1247 RETURN(rc ? rc : err);
1250 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1252 struct lov_request_set *set;
1253 struct lov_request *req;
1254 struct list_head *pos;
1255 struct lov_obd *lov;
1256 int err = 0, rc = 0;
1260 ASSERT_LSM_MAGIC(oinfo->oi_md);
1262 if (!exp || !exp->exp_obd)
1265 lov = &exp->exp_obd->u.lov;
1267 rc = lov_prep_getattr_set(exp, oinfo, &set);
1271 list_for_each (pos, &set->set_list) {
1272 req = list_entry(pos, struct lov_request, rq_link);
1274 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1275 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1276 req->rq_oi.oi_oa->o_id, req->rq_idx);
1278 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1280 err = lov_update_common_set(set, req, rc);
1282 CERROR("error: getattr objid "LPX64" subobj "
1283 LPX64" on OST idx %d: rc = %d\n",
1284 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1290 rc = lov_fini_getattr_set(set);
1296 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1299 struct lov_request_set *lovset = (struct lov_request_set *)data;
1303 /* don't do attribute merge if this aysnc op failed */
1305 lovset->set_completes = 0;
1306 err = lov_fini_getattr_set(lovset);
1307 RETURN(rc ? rc : err);
1310 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1311 struct ptlrpc_request_set *rqset)
1313 struct lov_request_set *lovset;
1314 struct lov_obd *lov;
1315 struct list_head *pos;
1316 struct lov_request *req;
1321 ASSERT_LSM_MAGIC(oinfo->oi_md);
1323 if (!exp || !exp->exp_obd)
1326 lov = &exp->exp_obd->u.lov;
1328 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1332 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1333 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1334 oinfo->oi_md->lsm_stripe_size);
1336 list_for_each (pos, &lovset->set_list) {
1337 req = list_entry(pos, struct lov_request, rq_link);
1339 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1340 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1341 req->rq_oi.oi_oa->o_id, req->rq_idx);
1342 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1343 &req->rq_oi, rqset);
1345 CERROR("error: getattr objid "LPX64" subobj "
1346 LPX64" on OST idx %d: rc = %d\n",
1347 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1353 if (!list_empty(&rqset->set_requests)) {
1355 LASSERT (rqset->set_interpret == NULL);
1356 rqset->set_interpret = lov_getattr_interpret;
1357 rqset->set_arg = (void *)lovset;
1362 lovset->set_completes = 0;
1363 err = lov_fini_getattr_set(lovset);
1364 RETURN(rc ? rc : err);
1367 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1368 struct obd_trans_info *oti)
1370 struct lov_request_set *set;
1371 struct lov_obd *lov;
1372 struct list_head *pos;
1373 struct lov_request *req;
1374 int err = 0, rc = 0;
1378 ASSERT_LSM_MAGIC(oinfo->oi_md);
1380 if (!exp || !exp->exp_obd)
1383 /* for now, we only expect the following updates here */
1384 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1385 OBD_MD_FLMODE | OBD_MD_FLATIME |
1386 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1387 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1388 OBD_MD_FLGROUP | OBD_MD_FLUID |
1389 OBD_MD_FLGID | OBD_MD_FLINLINE |
1390 OBD_MD_FLFID | OBD_MD_FLGENER)));
1391 lov = &exp->exp_obd->u.lov;
1392 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1396 list_for_each (pos, &set->set_list) {
1397 req = list_entry(pos, struct lov_request, rq_link);
1399 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1401 err = lov_update_setattr_set(set, req, rc);
1403 CERROR("error: setattr objid "LPX64" subobj "
1404 LPX64" on OST idx %d: rc = %d\n",
1405 set->set_oi->oi_oa->o_id,
1406 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1411 err = lov_fini_setattr_set(set);
1417 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1420 struct lov_request_set *lovset = (struct lov_request_set *)data;
1425 lovset->set_completes = 0;
1426 err = lov_fini_setattr_set(lovset);
1427 RETURN(rc ? rc : err);
1430 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1431 needed. Otherwise, a client is waiting for responses. */
1432 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1433 struct obd_trans_info *oti,
1434 struct ptlrpc_request_set *rqset)
1436 struct lov_request_set *set;
1437 struct lov_request *req;
1438 struct list_head *pos;
1439 struct lov_obd *lov;
1444 ASSERT_LSM_MAGIC(oinfo->oi_md);
1445 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1447 LASSERT(oti->oti_logcookies);
1450 if (!exp || !exp->exp_obd)
1453 lov = &exp->exp_obd->u.lov;
1454 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1458 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1459 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1460 oinfo->oi_md->lsm_stripe_size);
1462 list_for_each (pos, &set->set_list) {
1463 req = list_entry(pos, struct lov_request, rq_link);
1465 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1466 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1468 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1469 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1470 req->rq_oi.oi_oa->o_id, req->rq_idx);
1472 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1473 &req->rq_oi, oti, rqset);
1475 CERROR("error: setattr objid "LPX64" subobj "
1476 LPX64" on OST idx %d: rc = %d\n",
1477 set->set_oi->oi_oa->o_id,
1478 req->rq_oi.oi_oa->o_id,
1484 /* If we are not waiting for responses on async requests, return. */
1485 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1488 set->set_completes = 0;
1489 err = lov_fini_setattr_set(set);
1490 RETURN(rc ? rc : err);
1493 LASSERT(rqset->set_interpret == NULL);
1494 rqset->set_interpret = lov_setattr_interpret;
1495 rqset->set_arg = (void *)set;
1500 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1503 struct lov_request_set *lovset = (struct lov_request_set *)data;
1508 lovset->set_completes = 0;
1509 err = lov_fini_punch_set(lovset);
1510 RETURN(rc ? rc : err);
1513 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1514 * we can send this 'punch' to just the authoritative node and the nodes
1515 * that the punch will affect. */
1516 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1517 struct obd_trans_info *oti,
1518 struct ptlrpc_request_set *rqset)
1520 struct lov_request_set *set;
1521 struct lov_obd *lov;
1522 struct list_head *pos;
1523 struct lov_request *req;
1528 ASSERT_LSM_MAGIC(oinfo->oi_md);
1530 if (!exp || !exp->exp_obd)
1533 lov = &exp->exp_obd->u.lov;
1534 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1538 list_for_each (pos, &set->set_list) {
1539 req = list_entry(pos, struct lov_request, rq_link);
1541 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1542 &req->rq_oi, NULL, rqset);
1544 CERROR("error: punch objid "LPX64" subobj "LPX64
1545 " on OST idx %d: rc = %d\n",
1546 set->set_oi->oi_oa->o_id,
1547 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1552 if (rc || list_empty(&rqset->set_requests)) {
1554 err = lov_fini_punch_set(set);
1555 RETURN(rc ? rc : err);
1558 LASSERT(rqset->set_interpret == NULL);
1559 rqset->set_interpret = lov_punch_interpret;
1560 rqset->set_arg = (void *)set;
1565 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1566 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1569 struct lov_request_set *set;
1570 struct obd_info oinfo;
1571 struct lov_obd *lov;
1572 struct list_head *pos;
1573 struct lov_request *req;
1574 int err = 0, rc = 0;
1577 ASSERT_LSM_MAGIC(lsm);
1582 lov = &exp->exp_obd->u.lov;
1583 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1587 list_for_each (pos, &set->set_list) {
1588 req = list_entry(pos, struct lov_request, rq_link);
1590 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1591 req->rq_oi.oi_oa, NULL,
1592 req->rq_oi.oi_policy.l_extent.start,
1593 req->rq_oi.oi_policy.l_extent.end, capa);
1594 err = lov_update_common_set(set, req, rc);
1596 CERROR("error: fsync objid "LPX64" subobj "LPX64
1597 " on OST idx %d: rc = %d\n",
1598 set->set_oi->oi_oa->o_id,
1599 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1604 err = lov_fini_sync_set(set);
1610 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1611 obd_count oa_bufs, struct brw_page *pga)
1613 struct obd_info oinfo = { { { 0 } } };
1616 oinfo.oi_oa = lov_oinfo->oi_oa;
1618 /* The caller just wants to know if there's a chance that this
1619 * I/O can succeed */
1620 for (i = 0; i < oa_bufs; i++) {
1621 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1622 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1625 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1626 pga[i].off + pga[i].count,
1630 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1631 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1635 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1643 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1644 obd_count oa_bufs, struct brw_page *pga,
1645 struct obd_trans_info *oti)
1647 struct lov_request_set *set;
1648 struct lov_request *req;
1649 struct list_head *pos;
1650 struct lov_obd *lov = &exp->exp_obd->u.lov;
1654 ASSERT_LSM_MAGIC(oinfo->oi_md);
1656 if (cmd == OBD_BRW_CHECK) {
1657 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1661 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1665 list_for_each (pos, &set->set_list) {
1666 struct obd_export *sub_exp;
1667 struct brw_page *sub_pga;
1668 req = list_entry(pos, struct lov_request, rq_link);
1670 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1671 sub_pga = set->set_pga + req->rq_pgaidx;
1672 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1676 lov_update_common_set(set, req, rc);
1679 err = lov_fini_brw_set(set);
1685 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1688 struct lov_request_set *lovset = (struct lov_request_set *)data;
1692 lovset->set_completes = 0;
1693 lov_fini_brw_set(lovset);
1695 rc = lov_fini_brw_set(lovset);
1701 static int lov_brw_async(int cmd, struct obd_export *exp,
1702 struct obd_info *oinfo, obd_count oa_bufs,
1703 struct brw_page *pga, struct obd_trans_info *oti,
1704 struct ptlrpc_request_set *set)
1706 struct lov_request_set *lovset;
1707 struct lov_request *req;
1708 struct list_head *pos;
1709 struct lov_obd *lov = &exp->exp_obd->u.lov;
1714 ASSERT_LSM_MAGIC(oinfo->oi_md);
1716 if (cmd == OBD_BRW_CHECK) {
1717 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1721 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1725 list_for_each (pos, &lovset->set_list) {
1726 struct obd_export *sub_exp;
1727 struct brw_page *sub_pga;
1728 req = list_entry(pos, struct lov_request, rq_link);
1730 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1731 sub_pga = lovset->set_pga + req->rq_pgaidx;
1732 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1736 lov_update_common_set(lovset, req, rc);
1739 LASSERT(set->set_interpret == NULL);
1740 LASSERT(set->set_arg == NULL);
1741 rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1747 lov_fini_brw_set(lovset);
1751 static int lov_ap_make_ready(void *data, int cmd)
1753 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1755 return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1758 static int lov_ap_refresh_count(void *data, int cmd)
1760 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1762 return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1766 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1768 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1770 lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1771 /* XXX woah, shouldn't we be altering more here? size? */
1772 oa->o_id = lap->lap_loi_id;
1773 oa->o_gr = lap->lap_loi_gr;
1774 oa->o_valid |= OBD_MD_FLGROUP;
1775 oa->o_stripe_idx = lap->lap_stripe;
1778 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1781 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1783 lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1786 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1788 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1790 /* in a raid1 regime this would down a count of many ios
1791 * in flight, onl calling the caller_ops completion when all
1792 * the raid1 ios are complete */
1793 rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1797 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1799 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1800 return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1803 static struct obd_async_page_ops lov_async_page_ops = {
1804 .ap_make_ready = lov_ap_make_ready,
1805 .ap_refresh_count = lov_ap_refresh_count,
1806 .ap_fill_obdo = lov_ap_fill_obdo,
1807 .ap_update_obdo = lov_ap_update_obdo,
1808 .ap_completion = lov_ap_completion,
1809 .ap_lookup_capa = lov_ap_lookup_capa,
1812 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1813 struct lov_oinfo *loi, cfs_page_t *page,
1814 obd_off offset, struct obd_async_page_ops *ops,
1815 void *data, void **res, int nocache,
1816 struct lustre_handle *lockh)
1818 struct lov_obd *lov = &exp->exp_obd->u.lov;
1819 struct lov_async_page *lap;
1820 struct lov_lock_handles *lov_lockh = NULL;
1826 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1827 Only because of this layering limitation will a client
1828 mount with no osts fail */
1829 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1830 !lov->lov_tgts[i]->ltd_exp) {
1832 if (i >= lov->desc.ld_tgt_count)
1835 rc = size_round(sizeof(*lap)) +
1836 obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1837 NULL, NULL, 0, NULL, NULL, NULL, 0,
1841 ASSERT_LSM_MAGIC(lsm);
1842 LASSERT(loi == NULL);
1845 lap->lap_magic = LOV_AP_MAGIC;
1846 lap->lap_caller_ops = ops;
1847 lap->lap_caller_data = data;
1849 /* for now only raid 0 which passes through */
1850 lap->lap_stripe = lov_stripe_number(lsm, offset);
1851 lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1852 loi = lsm->lsm_oinfo[lap->lap_stripe];
1854 /* so the callback doesn't need the lsm */
1855 lap->lap_loi_id = loi->loi_id;
1856 lap->lap_loi_gr = lsm->lsm_object_gr;
1857 LASSERT(lsm->lsm_object_gr > 0);
1859 lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1862 lov_lockh = lov_handle2llh(lockh);
1864 lockh = lov_lockh->llh_handles + lap->lap_stripe;
1868 rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1869 lsm, loi, page, lap->lap_sub_offset,
1870 &lov_async_page_ops, lap,
1871 &lap->lap_sub_cookie, nocache, lockh);
1873 lov_llh_put(lov_lockh);
1876 CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1877 lap->lap_sub_cookie, offset);
1881 static int lov_queue_async_io(struct obd_export *exp,
1882 struct lov_stripe_md *lsm,
1883 struct lov_oinfo *loi, void *cookie,
1884 int cmd, obd_off off, int count,
1885 obd_flag brw_flags, obd_flag async_flags)
1887 struct lov_obd *lov = &exp->exp_obd->u.lov;
1888 struct lov_async_page *lap;
1891 LASSERT(loi == NULL);
1893 ASSERT_LSM_MAGIC(lsm);
1895 lap = LAP_FROM_COOKIE(cookie);
1897 loi = lsm->lsm_oinfo[lap->lap_stripe];
1899 rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1900 loi, lap->lap_sub_cookie, cmd, off, count,
1901 brw_flags, async_flags);
1905 static int lov_set_async_flags(struct obd_export *exp,
1906 struct lov_stripe_md *lsm,
1907 struct lov_oinfo *loi, void *cookie,
1908 obd_flag async_flags)
1910 struct lov_obd *lov = &exp->exp_obd->u.lov;
1911 struct lov_async_page *lap;
1914 LASSERT(loi == NULL);
1916 ASSERT_LSM_MAGIC(lsm);
1918 lap = LAP_FROM_COOKIE(cookie);
1920 loi = lsm->lsm_oinfo[lap->lap_stripe];
1922 rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1923 lsm, loi, lap->lap_sub_cookie, async_flags);
1927 static int lov_queue_group_io(struct obd_export *exp,
1928 struct lov_stripe_md *lsm,
1929 struct lov_oinfo *loi,
1930 struct obd_io_group *oig, void *cookie,
1931 int cmd, obd_off off, int count,
1932 obd_flag brw_flags, obd_flag async_flags)
1934 struct lov_obd *lov = &exp->exp_obd->u.lov;
1935 struct lov_async_page *lap;
1938 LASSERT(loi == NULL);
1940 ASSERT_LSM_MAGIC(lsm);
1942 lap = LAP_FROM_COOKIE(cookie);
1944 loi = lsm->lsm_oinfo[lap->lap_stripe];
1946 rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1947 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1948 brw_flags, async_flags);
1952 /* this isn't exactly optimal. we may have queued sync io in oscs on
1953 * all stripes, but we don't record that fact at queue time. so we
1954 * trigger sync io on all stripes. */
1955 static int lov_trigger_group_io(struct obd_export *exp,
1956 struct lov_stripe_md *lsm,
1957 struct lov_oinfo *loi,
1958 struct obd_io_group *oig)
1960 struct lov_obd *lov = &exp->exp_obd->u.lov;
1963 LASSERT(loi == NULL);
1965 ASSERT_LSM_MAGIC(lsm);
1967 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1968 loi = lsm->lsm_oinfo[i];
1969 if (!lov->lov_tgts[loi->loi_ost_idx] ||
1970 !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1971 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1975 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1977 if (rc == 0 && err != 0)
1983 static int lov_teardown_async_page(struct obd_export *exp,
1984 struct lov_stripe_md *lsm,
1985 struct lov_oinfo *loi, void *cookie)
1987 struct lov_obd *lov = &exp->exp_obd->u.lov;
1988 struct lov_async_page *lap;
1991 LASSERT(loi == NULL);
1993 ASSERT_LSM_MAGIC(lsm);
1995 lap = LAP_FROM_COOKIE(cookie);
1997 loi = lsm->lsm_oinfo[lap->lap_stripe];
1999 rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2000 lsm, loi, lap->lap_sub_cookie);
2002 CERROR("unable to teardown sub cookie %p: %d\n",
2003 lap->lap_sub_cookie, rc);
2009 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2012 struct lov_request_set *lovset = (struct lov_request_set *)data;
2014 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2018 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2019 struct ldlm_enqueue_info *einfo,
2020 struct ptlrpc_request_set *rqset)
2022 ldlm_mode_t mode = einfo->ei_mode;
2023 struct lov_request_set *set;
2024 struct lov_request *req;
2025 struct list_head *pos;
2026 struct lov_obd *lov;
2031 ASSERT_LSM_MAGIC(oinfo->oi_md);
2032 LASSERT(mode == (mode & -mode));
2034 /* we should never be asked to replay a lock this way. */
2035 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2037 if (!exp || !exp->exp_obd)
2040 lov = &exp->exp_obd->u.lov;
2041 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2045 list_for_each (pos, &set->set_list) {
2046 req = list_entry(pos, struct lov_request, rq_link);
2048 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2049 &req->rq_oi, einfo, rqset);
2054 if (rqset && !list_empty(&rqset->set_requests)) {
2056 LASSERT(rqset->set_interpret == NULL);
2057 rqset->set_interpret = lov_enqueue_interpret;
2058 rqset->set_arg = (void *)set;
2062 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2066 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2067 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2068 int *flags, void *data, struct lustre_handle *lockh)
2070 struct lov_request_set *set;
2071 struct obd_info oinfo;
2072 struct lov_request *req;
2073 struct list_head *pos;
2074 struct lov_obd *lov = &exp->exp_obd->u.lov;
2075 struct lustre_handle *lov_lockhp;
2076 int lov_flags, rc = 0;
2079 ASSERT_LSM_MAGIC(lsm);
2080 LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2082 if (!exp || !exp->exp_obd)
2085 lov = &exp->exp_obd->u.lov;
2086 rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2090 list_for_each (pos, &set->set_list) {
2091 ldlm_policy_data_t sub_policy;
2092 req = list_entry(pos, struct lov_request, rq_link);
2093 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2094 LASSERT(lov_lockhp);
2097 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2099 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2100 req->rq_oi.oi_md, type, &sub_policy,
2101 mode, &lov_flags, data, lov_lockhp);
2102 rc = lov_update_match_set(set, req, rc);
2106 lov_fini_match_set(set, mode, *flags);
2110 static int lov_change_cbdata(struct obd_export *exp,
2111 struct lov_stripe_md *lsm, ldlm_iterator_t it,
2114 struct lov_obd *lov;
2118 ASSERT_LSM_MAGIC(lsm);
2120 if (!exp || !exp->exp_obd)
2123 LASSERT(lsm->lsm_object_gr > 0);
2125 lov = &exp->exp_obd->u.lov;
2126 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2127 struct lov_stripe_md submd;
2128 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2130 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2131 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2135 submd.lsm_object_id = loi->loi_id;
2136 submd.lsm_object_gr = lsm->lsm_object_gr;
2137 submd.lsm_stripe_count = 0;
2138 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2144 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2145 __u32 mode, struct lustre_handle *lockh)
2147 struct lov_request_set *set;
2148 struct obd_info oinfo;
2149 struct lov_request *req;
2150 struct list_head *pos;
2151 struct lov_obd *lov = &exp->exp_obd->u.lov;
2152 struct lustre_handle *lov_lockhp;
2153 int err = 0, rc = 0;
2156 ASSERT_LSM_MAGIC(lsm);
2158 if (!exp || !exp->exp_obd)
2161 LASSERT(lsm->lsm_object_gr > 0);
2163 lov = &exp->exp_obd->u.lov;
2164 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2168 list_for_each (pos, &set->set_list) {
2169 req = list_entry(pos, struct lov_request, rq_link);
2170 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2172 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2173 req->rq_oi.oi_md, mode, lov_lockhp);
2174 rc = lov_update_common_set(set, req, rc);
2176 CERROR("error: cancel objid "LPX64" subobj "
2177 LPX64" on OST idx %d: rc = %d\n",
2179 req->rq_oi.oi_md->lsm_object_id,
2185 lov_fini_cancel_set(set);
2189 static int lov_cancel_unused(struct obd_export *exp,
2190 struct lov_stripe_md *lsm,
2191 int flags, void *opaque)
2193 struct lov_obd *lov;
2197 if (!exp || !exp->exp_obd)
2200 lov = &exp->exp_obd->u.lov;
2202 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2204 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2207 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2215 ASSERT_LSM_MAGIC(lsm);
2217 LASSERT(lsm->lsm_object_gr > 0);
2218 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2219 struct lov_stripe_md submd;
2220 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2223 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2224 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2228 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2229 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2231 submd.lsm_object_id = loi->loi_id;
2232 submd.lsm_object_gr = lsm->lsm_object_gr;
2233 submd.lsm_stripe_count = 0;
2234 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2235 &submd, flags, opaque);
2236 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2237 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2238 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2239 loi->loi_id, loi->loi_ost_idx, err);
2247 static int lov_join_lru(struct obd_export *exp,
2248 struct lov_stripe_md *lsm, int join)
2250 struct lov_obd *lov;
2254 ASSERT_LSM_MAGIC(lsm);
2255 if (!exp || !exp->exp_obd)
2258 lov = &exp->exp_obd->u.lov;
2259 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2260 struct lov_stripe_md submd;
2261 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2264 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2265 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2269 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2270 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2272 submd.lsm_object_id = loi->loi_id;
2273 submd.lsm_object_gr = lsm->lsm_object_gr;
2274 submd.lsm_stripe_count = 0;
2275 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2278 CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2279 " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2280 loi->loi_id, loi->loi_ost_idx, rc);
2289 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2292 struct lov_request_set *lovset = (struct lov_request_set *)data;
2297 lovset->set_completes = 0;
2299 err = lov_fini_statfs_set(lovset);
2300 RETURN(rc ? rc : err);
2303 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2304 __u64 max_age, struct ptlrpc_request_set *rqset)
2306 struct lov_request_set *set;
2307 struct lov_request *req;
2308 struct list_head *pos;
2309 struct lov_obd *lov;
2313 LASSERT(oinfo != NULL);
2314 LASSERT(oinfo->oi_osfs != NULL);
2317 rc = lov_prep_statfs_set(obd, oinfo, &set);
2321 list_for_each (pos, &set->set_list) {
2322 struct obd_device *osc_obd;
2324 req = list_entry(pos, struct lov_request, rq_link);
2326 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2327 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2332 if (rc || list_empty(&rqset->set_requests)) {
2335 set->set_completes = 0;
2336 err = lov_fini_statfs_set(set);
2337 RETURN(rc ? rc : err);
2340 LASSERT(rqset->set_interpret == NULL);
2341 rqset->set_interpret = lov_statfs_interpret;
2342 rqset->set_arg = (void *)set;
2346 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2347 __u64 max_age, __u32 flags)
2349 struct ptlrpc_request_set *set = NULL;
2350 struct obd_info oinfo = { { { 0 } } };
2355 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2356 * statfs requests */
2357 set = ptlrpc_prep_set();
2361 oinfo.oi_osfs = osfs;
2362 oinfo.oi_flags = flags;
2363 rc = lov_statfs_async(obd, &oinfo, max_age, set);
2365 rc = ptlrpc_set_wait(set);
2366 ptlrpc_set_destroy(set);
2371 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2372 void *karg, void *uarg)
2374 struct obd_device *obddev = class_exp2obd(exp);
2375 struct lov_obd *lov = &obddev->u.lov;
2376 int i, rc, count = lov->desc.ld_tgt_count;
2377 struct obd_uuid *uuidp;
2381 case IOC_OBD_STATFS: {
2382 struct obd_ioctl_data *data = karg;
2383 struct obd_device *osc_obd;
2384 struct obd_statfs stat_buf = {0};
2387 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2388 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
2390 if ((index >= count))
2393 if (!lov->lov_tgts[index])
2394 /* Try again with the next index */
2396 if (!lov->lov_tgts[index]->ltd_active)
2399 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2403 /* got statfs data */
2404 rc = obd_statfs(osc_obd, &stat_buf,
2405 cfs_time_current_64() - HZ, 0);
2408 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2411 rc = copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2415 case OBD_IOC_LOV_GET_CONFIG: {
2416 struct obd_ioctl_data *data;
2417 struct lov_desc *desc;
2422 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2425 data = (struct obd_ioctl_data *)buf;
2427 if (sizeof(*desc) > data->ioc_inllen1) {
2428 obd_ioctl_freedata(buf, len);
2432 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2433 obd_ioctl_freedata(buf, len);
2437 if (sizeof(__u32) * count > data->ioc_inllen3) {
2438 obd_ioctl_freedata(buf, len);
2442 desc = (struct lov_desc *)data->ioc_inlbuf1;
2443 memcpy(desc, &(lov->desc), sizeof(*desc));
2445 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2446 genp = (__u32 *)data->ioc_inlbuf3;
2447 /* the uuid will be empty for deleted OSTs */
2448 for (i = 0; i < count; i++, uuidp++, genp++) {
2449 if (!lov->lov_tgts[i])
2451 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2452 *genp = lov->lov_tgts[i]->ltd_gen;
2455 rc = copy_to_user((void *)uarg, buf, len);
2458 obd_ioctl_freedata(buf, len);
2461 case LL_IOC_LOV_SETSTRIPE:
2462 rc = lov_setstripe(exp, karg, uarg);
2464 case LL_IOC_LOV_GETSTRIPE:
2465 rc = lov_getstripe(exp, karg, uarg);
2467 case LL_IOC_LOV_SETEA:
2468 rc = lov_setea(exp, karg, uarg);
2477 for (i = 0; i < count; i++) {
2480 /* OST was disconnected */
2481 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2484 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2486 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2489 if (lov->lov_tgts[i]->ltd_active) {
2490 CDEBUG(err == -ENOTTY ?
2491 D_IOCTL : D_WARNING,
2492 "iocontrol OSC %s on OST "
2493 "idx %d cmd %x: err = %d\n",
2494 lov_uuid2str(lov, i),
2511 #define FIEMAP_BUFFER_SIZE 4096
2514 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2515 * call. The local end offset and the device are sent in the first
2516 * fm_extent. This function calculates the stripe number from the index.
2517 * This function returns a stripe_no on which mapping is to be restarted.
2519 * This function returns fm_end_offset which is the in-OST offset at which
2520 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2521 * will re-calculate proper offset in next stripe.
2522 * Note that the first extent is passed to lov_get_info via the value field.
2524 * \param fiemap fiemap request header
2525 * \param lsm striping information for the file
2526 * \param fm_start logical start of mapping
2527 * \param fm_end logical end of mapping
2528 * \param start_stripe starting stripe will be returned in this
2530 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2531 struct lov_stripe_md *lsm, obd_size fm_start,
2532 obd_size fm_end, int *start_stripe)
2534 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2535 obd_off lun_start, lun_end;
2536 obd_size fm_end_offset;
2537 int stripe_no = -1, i;
2539 if (fiemap->fm_extent_count == 0 ||
2540 fiemap->fm_extents[0].fe_logical == 0)
2543 /* Find out stripe_no from ost_index saved in the fe_device */
2544 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2545 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2546 fiemap->fm_extents[0].fe_device) {
2552 /* If we have finished mapping on previous device, shift logical
2553 * offset to start of next device */
2554 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2555 &lun_start, &lun_end)) != 0 &&
2556 local_end < lun_end) {
2557 fm_end_offset = local_end;
2558 *start_stripe = stripe_no;
2560 /* This is a special value to indicate that caller should
2561 * calculate offset in next stripe. */
2563 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2566 return fm_end_offset;
2570 * We calculate on which OST the mapping will end. If the length of mapping
2571 * is greater than (stripe_size * stripe_count) then the last_stripe will
2572 * will be one just before start_stripe. Else we check if the mapping
2573 * intersects each OST and find last_stripe.
2574 * This function returns the last_stripe and also sets the stripe_count
2575 * over which the mapping is spread
2577 * \param lsm striping information for the file
2578 * \param fm_start logical start of mapping
2579 * \param fm_end logical end of mapping
2580 * \param start_stripe starting stripe of the mapping
2581 * \param stripe_count the number of stripes across which to map is returned
2583 * \retval last_stripe return the last stripe of the mapping
2585 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2586 obd_size fm_end, int start_stripe,
2590 obd_off obd_start, obd_end;
2593 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2594 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2596 *stripe_count = lsm->lsm_stripe_count;
2598 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2599 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2600 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2601 &obd_start, &obd_end)) == 0)
2605 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2612 * Set fe_device and copy extents from local buffer into main return buffer.
2614 * \param fiemap fiemap request header
2615 * \param lcl_fm_ext array of local fiemap extents to be copied
2616 * \param ost_index OST index to be written into the fm_device field for each
2618 * \param ext_count number of extents to be copied
2619 * \param current_extent where to start copying in main extent array
2621 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2622 struct ll_fiemap_extent *lcl_fm_ext,
2623 int ost_index, unsigned int ext_count,
2629 for (ext = 0; ext < ext_count; ext++) {
2630 lcl_fm_ext[ext].fe_device = ost_index;
2631 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2634 /* Copy fm_extent's from fm_local to return buffer */
2635 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2636 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2640 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2641 * This also handles the restarting of FIEMAP calls in case mapping overflows
2642 * the available number of extents in single call.
2644 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2645 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2647 struct ll_fiemap_info_key *fm_key = key;
2648 struct ll_user_fiemap *fiemap = val;
2649 struct ll_user_fiemap *fm_local = NULL;
2650 struct ll_fiemap_extent *lcl_fm_ext;
2652 unsigned int get_num_extents = 0;
2653 int ost_index = 0, actual_start_stripe, start_stripe;
2654 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2656 int current_extent = 0, rc = 0, i;
2657 int ost_eof = 0; /* EOF for object */
2658 int ost_done = 0; /* done with required mapping for this OST? */
2660 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2661 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2666 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2667 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2669 OBD_ALLOC(fm_local, buffer_size);
2670 if (fm_local == NULL)
2671 GOTO(out, rc = -ENOMEM);
2672 lcl_fm_ext = &fm_local->fm_extents[0];
2674 count_local = fiemap_size_to_count(buffer_size);
2676 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2677 fm_start = fiemap->fm_start;
2678 fm_length = fiemap->fm_length;
2679 /* Calculate start stripe, last stripe and length of mapping */
2680 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2681 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2682 fm_start + fm_length - 1);
2683 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2684 if (fm_end > fm_key->oa.o_size)
2685 fm_end = fm_key->oa.o_size;
2687 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2688 actual_start_stripe, &stripe_count);
2690 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2693 if (fiemap->fm_extent_count == 0) {
2694 get_num_extents = 1;
2698 /* Check each stripe */
2699 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2700 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2701 obd_size req_fm_len; /* Stores length of required mapping */
2702 obd_size len_mapped_single_call;
2703 obd_off lun_start, lun_end, obd_object_end;
2704 unsigned int ext_count;
2706 cur_stripe_wrap = cur_stripe;
2708 /* Find out range of mapping on this stripe */
2709 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2710 &lun_start, &obd_object_end)) == 0)
2713 /* If this is a continuation FIEMAP call and we are on
2714 * starting stripe then lun_start needs to be set to
2716 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2717 lun_start = fm_end_offset;
2719 if (fm_length != ~0ULL) {
2720 /* Handle fm_start + fm_length overflow */
2721 if (fm_start + fm_length < fm_start)
2722 fm_length = ~0ULL - fm_start;
2723 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2729 if (lun_start == lun_end)
2732 req_fm_len = obd_object_end - lun_start;
2733 fm_local->fm_length = 0;
2734 len_mapped_single_call = 0;
2736 /* If the output buffer is very large and the objects have many
2737 * extents we may need to loop on a single OST repeatedly */
2741 if (get_num_extents == 0) {
2742 /* Don't get too many extents. */
2743 if (current_extent + count_local >
2744 fiemap->fm_extent_count)
2745 count_local = fiemap->fm_extent_count -
2749 lun_start += len_mapped_single_call;
2750 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2751 req_fm_len = fm_local->fm_length;
2752 fm_local->fm_extent_count = count_local;
2753 fm_local->fm_mapped_extents = 0;
2754 fm_local->fm_flags = fiemap->fm_flags;
2756 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2757 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2759 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2760 GOTO(out, rc = -EINVAL);
2762 /* If OST is inactive, return extent with UNKNOWN flag */
2763 if (lov && !lov->lov_tgts[ost_index]->ltd_active) {
2764 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2765 fm_local->fm_mapped_extents = 1;
2767 lcl_fm_ext[0].fe_logical = lun_start;
2768 lcl_fm_ext[0].fe_length = obd_object_end -
2770 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2775 fm_local->fm_start = lun_start;
2776 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2777 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2778 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2779 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2780 keylen, key, vallen, fm_local, lsm);
2785 ext_count = fm_local->fm_mapped_extents;
2786 if (ext_count == 0) {
2788 /* If last stripe has hole at the end,
2789 * then we need to return */
2790 if (cur_stripe_wrap == last_stripe) {
2791 fiemap->fm_mapped_extents = 0;
2797 /* If we just need num of extents then go to next device */
2798 if (get_num_extents) {
2799 current_extent += ext_count;
2803 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2804 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2806 /* Have we finished mapping on this device? */
2807 if (req_fm_len <= len_mapped_single_call)
2810 /* Clear the EXTENT_LAST flag which can be present on
2812 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2813 lcl_fm_ext[ext_count - 1].fe_flags &=
2814 ~FIEMAP_EXTENT_LAST;
2816 curr_loc = lov_stripe_size(lsm,
2817 lcl_fm_ext[ext_count - 1].fe_logical+
2818 lcl_fm_ext[ext_count - 1].fe_length,
2820 if (curr_loc >= fm_key->oa.o_size)
2823 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2824 ost_index, ext_count,
2827 current_extent += ext_count;
2829 /* Ran out of available extents? */
2830 if (current_extent >= fiemap->fm_extent_count)
2832 } while (ost_done == 0 && ost_eof == 0);
2834 if (cur_stripe_wrap == last_stripe)
2839 /* Indicate that we are returning device offsets unless file just has
2841 if (lsm->lsm_stripe_count > 1)
2842 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2844 if (get_num_extents)
2845 goto skip_last_device_calc;
2847 /* Check if we have reached the last stripe and whether mapping for that
2848 * stripe is done. */
2849 if (cur_stripe_wrap == last_stripe) {
2850 if (ost_done || ost_eof)
2851 fiemap->fm_extents[current_extent - 1].fe_flags |=
2855 skip_last_device_calc:
2856 fiemap->fm_mapped_extents = current_extent;
2859 OBD_FREE(fm_local, buffer_size);
2863 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2864 void *key, __u32 *vallen, void *val,
2865 struct lov_stripe_md *lsm)
2867 struct obd_device *obddev = class_exp2obd(exp);
2868 struct lov_obd *lov = &obddev->u.lov;
2872 if (!vallen || !val)
2877 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2880 struct ldlm_lock *lock;
2882 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2883 struct lov_oinfo *loi;
2884 __u32 *stripe = val;
2886 if (*vallen < sizeof(*stripe))
2887 GOTO(out, rc = -EFAULT);
2888 *vallen = sizeof(*stripe);
2890 /* XXX This is another one of those bits that will need to
2891 * change if we ever actually support nested LOVs. It uses
2892 * the lock's export to find out which stripe it is. */
2893 /* XXX - it's assumed all the locks for deleted OSTs have
2894 * been cancelled. Also, the export for deleted OSTs will
2895 * be NULL and won't match the lock's export. */
2896 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2897 loi = lsm->lsm_oinfo[i];
2898 if (!lov->lov_tgts[loi->loi_ost_idx])
2900 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2901 data->lock->l_conn_export &&
2902 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2907 LDLM_ERROR(data->lock, "lock on inode without such object");
2908 dump_lsm(D_ERROR, lsm);
2909 GOTO(out, rc = -ENXIO);
2910 } else if (KEY_IS(KEY_LAST_ID)) {
2911 struct obd_id_info *info = val;
2912 __u32 size = sizeof(obd_id);
2913 struct lov_tgt_desc *tgt;
2915 LASSERT(*vallen == sizeof(struct obd_id_info));
2916 tgt = lov->lov_tgts[info->idx];
2918 if (!tgt || !tgt->ltd_active)
2919 GOTO(out, rc = -ESRCH);
2921 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2923 } else if (KEY_IS(KEY_LOVDESC)) {
2924 struct lov_desc *desc_ret = val;
2925 *desc_ret = lov->desc;
2928 } else if (KEY_IS(KEY_LOV_IDX)) {
2929 struct lov_tgt_desc *tgt;
2931 for(i = 0; i < lov->desc.ld_tgt_count; i++) {
2932 tgt = lov->lov_tgts[i];
2933 if (tgt && obd_uuid_equals(val, &tgt->ltd_uuid))
2936 } else if (KEY_IS(KEY_FIEMAP)) {
2937 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2948 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2949 void *key, obd_count vallen, void *val,
2950 struct ptlrpc_request_set *set)
2952 struct obd_device *obddev = class_exp2obd(exp);
2953 struct lov_obd *lov = &obddev->u.lov;
2956 struct lov_tgt_desc *tgt;
2957 unsigned incr, check_uuid,
2958 do_inactive, no_set;
2959 unsigned next_id = 0, mds_con = 0;
2962 incr = check_uuid = do_inactive = no_set = 0;
2965 set = ptlrpc_prep_set();
2971 count = lov->desc.ld_tgt_count;
2973 if (KEY_IS(KEY_NEXT_ID)) {
2974 count = vallen / sizeof(struct obd_id_info);
2975 vallen = sizeof(obd_id);
2976 incr = sizeof(struct obd_id_info);
2979 } else if (KEY_IS(KEY_CHECKSUM)) {
2981 } else if (KEY_IS(KEY_UNLINKED)) {
2982 check_uuid = val ? 1 : 0;
2983 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2984 /* use defaults: do_inactive = incr = 0; */
2985 } else if (KEY_IS(KEY_MDS_CONN)) {
2989 for (i = 0; i < count; i++, val = (char *)val + incr) {
2991 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2993 tgt = lov->lov_tgts[i];
2995 /* OST was disconnected */
2996 if (!tgt || !tgt->ltd_exp)
2999 /* OST is inactive and we don't want inactive OSCs */
3000 if (!tgt->ltd_active && !do_inactive)
3004 struct mds_group_info *mgi;
3006 LASSERT(vallen == sizeof(*mgi));
3007 mgi = (struct mds_group_info *)val;
3009 /* Only want a specific OSC */
3010 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
3014 err = obd_set_info_async(tgt->ltd_exp,
3015 keylen, key, sizeof(int),
3017 } else if (next_id) {
3018 err = obd_set_info_async(tgt->ltd_exp,
3019 keylen, key, vallen,
3020 ((struct obd_id_info*)val)->data, set);
3022 /* Only want a specific OSC */
3024 !obd_uuid_equals(val, &tgt->ltd_uuid))
3027 err = obd_set_info_async(tgt->ltd_exp,
3028 keylen, key, vallen, val, set);
3037 err = ptlrpc_set_wait(set);
3040 ptlrpc_set_destroy(set);
3045 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3046 struct lov_stripe_md *lsm)
3054 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3055 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3060 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3065 for (i = 0; i < lsm->lsm_stripe_count; i++) {
3066 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
3067 if (loi->loi_ar.ar_rc && !rc)
3068 rc = loi->loi_ar.ar_rc;
3069 loi->loi_ar.ar_rc = 0;
3073 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3076 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3077 int cmd, __u64 *offset)
3079 __u32 ssize = lsm->lsm_stripe_size;
3083 do_div(start, ssize);
3084 start = start * ssize;
3086 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3087 ", end "LPU64"\n", *offset, ssize, start,
3089 if (cmd == OBD_CALC_STRIPE_END) {
3090 *offset = start + ssize - 1;
3091 } else if (cmd == OBD_CALC_STRIPE_START) {
3102 struct lov_multi_wait {
3103 struct ldlm_lock *lock;
3109 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3110 struct lustre_handle *lockh)
3112 struct lov_lock_handles *lov_lockh = NULL;
3113 struct lustre_handle *lov_lockhp;
3114 struct lov_obd *lov;
3115 struct lov_oinfo *loi;
3116 struct lov_multi_wait *queues;
3120 ASSERT_LSM_MAGIC(lsm);
3122 if (!exp || !exp->exp_obd)
3125 LASSERT(lockh != NULL);
3126 if (lsm->lsm_stripe_count > 1) {
3127 lov_lockh = lov_handle2llh(lockh);
3128 if (lov_lockh == NULL) {
3129 CERROR("LOV: invalid lov lock handle %p\n", lockh);
3133 lov_lockhp = lov_lockh->llh_handles;
3138 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3140 GOTO(out, rc = -ENOMEM);
3142 lov = &exp->exp_obd->u.lov;
3143 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3144 i++, loi++, lov_lockhp++) {
3145 struct ldlm_lock *lock;
3146 struct obd_device *obd;
3148 lock = ldlm_handle2lock(lov_lockhp);
3150 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3151 loi->loi_ost_idx, loi->loi_id);
3152 queues[i].completed = 1;
3156 queues[i].lock = lock;
3157 init_waitqueue_entry(&(queues[i].wait), current);
3158 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3160 obd = class_exp2obd(lock->l_conn_export);
3162 imp = obd->u.cli.cl_import;
3164 spin_lock(&imp->imp_lock);
3165 queues[i].generation = imp->imp_generation;
3166 spin_unlock(&imp->imp_lock);
3170 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3171 interrupted_completion_wait, &lwd);
3172 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3174 for (i = 0; i < lsm->lsm_stripe_count; i++)
3175 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3177 if (rc == -EINTR || rc == -ETIMEDOUT) {
3183 if (lov_lockh != NULL)
3184 lov_llh_put(lov_lockh);
3189 void lov_stripe_lock(struct lov_stripe_md *md)
3191 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
3192 spin_lock(&md->lsm_lock);
3193 LASSERT(md->lsm_lock_owner == 0);
3194 md->lsm_lock_owner = cfs_curproc_pid();
3196 EXPORT_SYMBOL(lov_stripe_lock);
3198 void lov_stripe_unlock(struct lov_stripe_md *md)
3200 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
3201 md->lsm_lock_owner = 0;
3202 spin_unlock(&md->lsm_lock);
3204 EXPORT_SYMBOL(lov_stripe_unlock);
3207 * Checks if requested extent lock is compatible with a lock under the page.
3209 * Checks if the lock under \a page is compatible with a read or write lock
3210 * (specified by \a rw) for an extent [\a start , \a end].
3212 * \param exp lov export
3213 * \param lsm striping information for the file
3214 * \param res lov_async_page placeholder
3215 * \param rw OBD_BRW_READ if requested for reading,
3216 * OBD_BRW_WRITE if requested for writing
3217 * \param start start of the requested extent
3218 * \param end end of the requested extent
3219 * \param cookie transparent parameter for passing locking context
3221 * \post result == 1, *cookie == context, appropriate lock is referenced or
3224 * \retval 1 owned lock is reused for the request
3225 * \retval 0 no lock reused for the request
3227 * \see lov_release_short_lock
3229 static int lov_reget_short_lock(struct obd_export *exp,
3230 struct lov_stripe_md *lsm,
3232 obd_off start, obd_off end,
3235 struct lov_async_page *l = *res;
3236 obd_off stripe_start, stripe_end = start;
3240 /* ensure we don't cross stripe boundaries */
3241 lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3242 if (stripe_end <= end)
3245 /* map the region limits to the object limits */
3246 lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3247 lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3249 RETURN(obd_reget_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3250 lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3251 ltd_exp, NULL, &l->lap_sub_cookie,
3252 rw, stripe_start, stripe_end, cookie));
3256 * Releases a reference to a lock taken in a "fast" way.
3258 * Releases a read or a write (specified by \a rw) lock
3259 * referenced by \a cookie.
3261 * \param exp lov export
3262 * \param lsm striping information for the file
3263 * \param end end of the locked extent
3264 * \param rw OBD_BRW_READ if requested for reading,
3265 * OBD_BRW_WRITE if requested for writing
3266 * \param cookie transparent parameter for passing locking context
3268 * \post appropriate lock is dereferenced
3270 * \see lov_reget_short_lock
3272 static int lov_release_short_lock(struct obd_export *exp,
3273 struct lov_stripe_md *lsm, obd_off end,
3274 void *cookie, int rw)
3280 stripe = lov_stripe_number(lsm, end);
3282 RETURN(obd_release_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3283 lsm_oinfo[stripe]->loi_ost_idx]->
3284 ltd_exp, NULL, end, cookie, rw));
3287 struct obd_ops lov_obd_ops = {
3288 .o_owner = THIS_MODULE,
3289 .o_setup = lov_setup,
3290 .o_precleanup = lov_precleanup,
3291 .o_cleanup = lov_cleanup,
3292 .o_process_config = lov_process_config,
3293 .o_connect = lov_connect,
3294 .o_disconnect = lov_disconnect,
3295 .o_statfs = lov_statfs,
3296 .o_statfs_async = lov_statfs_async,
3297 .o_packmd = lov_packmd,
3298 .o_unpackmd = lov_unpackmd,
3299 .o_checkmd = lov_checkmd,
3300 .o_create = lov_create,
3301 .o_destroy = lov_destroy,
3302 .o_getattr = lov_getattr,
3303 .o_getattr_async = lov_getattr_async,
3304 .o_setattr = lov_setattr,
3305 .o_setattr_async = lov_setattr_async,
3307 .o_brw_async = lov_brw_async,
3308 .o_prep_async_page = lov_prep_async_page,
3309 .o_reget_short_lock = lov_reget_short_lock,
3310 .o_release_short_lock = lov_release_short_lock,
3311 .o_queue_async_io = lov_queue_async_io,
3312 .o_set_async_flags = lov_set_async_flags,
3313 .o_queue_group_io = lov_queue_group_io,
3314 .o_trigger_group_io = lov_trigger_group_io,
3315 .o_teardown_async_page = lov_teardown_async_page,
3316 .o_merge_lvb = lov_merge_lvb,
3317 .o_adjust_kms = lov_adjust_kms,
3318 .o_punch = lov_punch,
3320 .o_enqueue = lov_enqueue,
3321 .o_match = lov_match,
3322 .o_change_cbdata = lov_change_cbdata,
3323 .o_cancel = lov_cancel,
3324 .o_cancel_unused = lov_cancel_unused,
3325 .o_join_lru = lov_join_lru,
3326 .o_iocontrol = lov_iocontrol,
3327 .o_get_info = lov_get_info,
3328 .o_set_info_async = lov_set_info_async,
3329 .o_extent_calc = lov_extent_calc,
3330 .o_llog_init = lov_llog_init,
3331 .o_llog_finish = lov_llog_finish,
3332 .o_notify = lov_notify,
3333 .o_register_page_removal_cb = lov_register_page_removal_cb,
3334 .o_unregister_page_removal_cb = lov_unregister_page_removal_cb,
3335 .o_register_lock_cancel_cb = lov_register_lock_cancel_cb,
3336 .o_unregister_lock_cancel_cb = lov_unregister_lock_cancel_cb,
3339 static quota_interface_t *quota_interface;
3340 extern quota_interface_t lov_quota_interface;
3342 cfs_mem_cache_t *lov_oinfo_slab;
3344 int __init lov_init(void)
3346 struct lprocfs_static_vars lvars = { 0 };
3350 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3351 sizeof(struct lov_oinfo),
3352 0, SLAB_HWCACHE_ALIGN);
3353 if (lov_oinfo_slab == NULL)
3355 lprocfs_lov_init_vars(&lvars);
3357 request_module("lquota");
3358 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3359 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3361 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3362 LUSTRE_LOV_NAME, NULL);
3364 if (quota_interface)
3365 PORTAL_SYMBOL_PUT(lov_quota_interface);
3366 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3374 static void /*__exit*/ lov_exit(void)
3378 if (quota_interface)
3379 PORTAL_SYMBOL_PUT(lov_quota_interface);
3381 class_unregister_type(LUSTRE_LOV_NAME);
3382 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3386 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3387 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3388 MODULE_LICENSE("GPL");
3390 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);