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>
68 #include "lov_internal.h"
71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
72 Any function that expects lov_tgts to remain stationary must take a ref. */
73 void lov_getref(struct obd_device *obd)
75 struct lov_obd *lov = &obd->u.lov;
77 /* nobody gets through here until lov_putref is done */
78 mutex_down(&lov->lov_lock);
79 atomic_inc(&lov->lov_refcount);
80 mutex_up(&lov->lov_lock);
84 static void __lov_del_obd(struct obd_device *obd, __u32 index);
86 void lov_putref(struct obd_device *obd)
88 struct lov_obd *lov = &obd->u.lov;
89 mutex_down(&lov->lov_lock);
90 /* ok to dec to 0 more than once -- ltd_exp's will be null */
91 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
93 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
95 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
96 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap)
98 /* Disconnect and delete from list */
99 __lov_del_obd(obd, i);
100 lov->lov_death_row--;
103 mutex_up(&lov->lov_lock);
106 static int lov_register_page_removal_cb(struct obd_export *exp,
107 obd_page_removal_cb_t func,
108 obd_pin_extent_cb pin_cb)
110 struct lov_obd *lov = &exp->exp_obd->u.lov;
113 if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
116 if (lov->lov_page_pin_cb && lov->lov_page_pin_cb != pin_cb)
119 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
120 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
122 rc |= obd_register_page_removal_cb(lov->lov_tgts[i]->ltd_exp,
126 lov->lov_page_removal_cb = func;
127 lov->lov_page_pin_cb = pin_cb;
132 static int lov_unregister_page_removal_cb(struct obd_export *exp,
133 obd_page_removal_cb_t func)
135 struct lov_obd *lov = &exp->exp_obd->u.lov;
138 if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
141 lov->lov_page_removal_cb = NULL;
142 lov->lov_page_pin_cb = NULL;
144 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
145 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
147 rc |= obd_unregister_page_removal_cb(lov->lov_tgts[i]->ltd_exp,
154 static int lov_register_lock_cancel_cb(struct obd_export *exp,
155 obd_lock_cancel_cb func)
157 struct lov_obd *lov = &exp->exp_obd->u.lov;
160 if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
163 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
164 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
166 rc |= obd_register_lock_cancel_cb(lov->lov_tgts[i]->ltd_exp,
170 lov->lov_lock_cancel_cb = func;
175 static int lov_unregister_lock_cancel_cb(struct obd_export *exp,
176 obd_lock_cancel_cb func)
178 struct lov_obd *lov = &exp->exp_obd->u.lov;
181 if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
184 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
185 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
187 rc |= obd_unregister_lock_cancel_cb(lov->lov_tgts[i]->ltd_exp,
190 lov->lov_lock_cancel_cb = NULL;
194 #define MAX_STRING_SIZE 128
195 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
196 struct obd_connect_data *data)
198 struct lov_obd *lov = &obd->u.lov;
199 struct obd_uuid tgt_uuid;
200 struct obd_device *tgt_obd;
201 struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
202 struct lustre_handle conn = {0, };
203 struct obd_import *imp;
206 cfs_proc_dir_entry_t *lov_proc_dir;
211 if (!lov->lov_tgts[index])
214 tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
216 tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
220 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
223 if (!tgt_obd->obd_set_up) {
224 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
228 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
229 data->ocd_index = index;
232 * Divine LOV knows that OBDs under it are OSCs.
234 imp = tgt_obd->u.cli.cl_import;
237 tgt_obd->obd_no_recov = 0;
238 /* FIXME this is probably supposed to be
239 ptlrpc_set_import_active. Horrible naming. */
240 ptlrpc_activate_import(imp);
243 if (imp->imp_invalid) {
244 CERROR("not connecting OSC %s; administratively "
245 "disabled\n", obd_uuid2str(&tgt_uuid));
246 rc = obd_register_observer(tgt_obd, obd);
248 CERROR("Target %s register_observer error %d; "
249 "will not be able to reactivate\n",
250 obd_uuid2str(&tgt_uuid), rc);
255 rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data, NULL);
257 CERROR("Target %s connect error %d\n",
258 obd_uuid2str(&tgt_uuid), rc);
261 lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
262 if (!lov->lov_tgts[index]->ltd_exp) {
263 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
267 rc = obd_register_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
268 lov->lov_page_removal_cb,
269 lov->lov_page_pin_cb);
271 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
272 lov->lov_tgts[index]->ltd_exp = NULL;
276 rc = obd_register_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
277 lov->lov_lock_cancel_cb);
279 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
280 lov->lov_page_removal_cb);
281 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
282 lov->lov_tgts[index]->ltd_exp = NULL;
286 rc = obd_register_observer(tgt_obd, obd);
288 CERROR("Target %s register_observer error %d\n",
289 obd_uuid2str(&tgt_uuid), rc);
290 obd_unregister_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
291 lov->lov_lock_cancel_cb);
292 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
293 lov->lov_page_removal_cb);
294 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
295 lov->lov_tgts[index]->ltd_exp = NULL;
299 lov->lov_tgts[index]->ltd_reap = 0;
301 lov->lov_tgts[index]->ltd_active = 1;
302 lov->desc.ld_active_tgt_count++;
303 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
305 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
306 obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
309 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
311 struct obd_device *osc_obd = class_conn2obd(&conn);
312 cfs_proc_dir_entry_t *osc_symlink;
313 char name[MAX_STRING_SIZE];
315 LASSERT(osc_obd != NULL);
316 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
317 LASSERT(osc_obd->obd_type->typ_name != NULL);
318 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
319 osc_obd->obd_type->typ_name,
321 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
323 if (osc_symlink == NULL) {
324 CERROR("could not register LOV target "
325 "/proc/fs/lustre/%s/%s/target_obds/%s.",
326 obd->obd_type->typ_name, obd->obd_name,
328 lprocfs_remove(&lov_proc_dir);
333 rc = qos_add_tgt(obd, index);
335 CERROR("qos_add_tgt failed %d\n", rc);
340 static int lov_connect(const struct lu_env *env,
341 struct lustre_handle *conn, struct obd_device *obd,
342 struct obd_uuid *cluuid, struct obd_connect_data *data,
345 struct lov_obd *lov = &obd->u.lov;
346 struct lov_tgt_desc *tgt;
350 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
352 rc = class_connect(conn, obd, cluuid);
356 /* Why should there ever be more than 1 connect? */
358 LASSERT(lov->lov_connects == 1);
360 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
362 lov->lov_ocd = *data;
365 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
366 tgt = lov->lov_tgts[i];
367 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
369 /* Flags will be lowest common denominator */
370 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
372 CERROR("%s: lov connect tgt %d failed: %d\n",
373 obd->obd_name, i, rc);
382 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
384 cfs_proc_dir_entry_t *lov_proc_dir;
385 struct lov_obd *lov = &obd->u.lov;
386 struct obd_device *osc_obd;
391 if (lov->lov_tgts[index] == NULL)
394 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
395 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
396 obd->obd_name, osc_obd->obd_name);
398 obd_unregister_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
399 lov->lov_lock_cancel_cb);
400 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
401 lov->lov_page_removal_cb);
403 if (lov->lov_tgts[index]->ltd_active) {
404 lov->lov_tgts[index]->ltd_active = 0;
405 lov->desc.ld_active_tgt_count--;
406 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
409 lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
411 cfs_proc_dir_entry_t *osc_symlink;
413 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
415 lprocfs_remove(&osc_symlink);
417 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
418 obd->obd_type->typ_name, obd->obd_name,
424 /* Pass it on to our clients.
425 * XXX This should be an argument to disconnect,
426 * XXX not a back-door flag on the OBD. Ah well.
428 osc_obd->obd_force = obd->obd_force;
429 osc_obd->obd_fail = obd->obd_fail;
430 osc_obd->obd_no_recov = obd->obd_no_recov;
433 obd_register_observer(osc_obd, NULL);
435 rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
437 CERROR("Target %s disconnect error %d\n",
438 lov_uuid2str(lov, index), rc);
442 qos_del_tgt(obd, index);
444 lov->lov_tgts[index]->ltd_exp = NULL;
448 static int lov_del_target(struct obd_device *obd, __u32 index,
449 struct obd_uuid *uuidp, int gen);
451 static int lov_disconnect(struct obd_export *exp)
453 struct obd_device *obd = class_exp2obd(exp);
454 struct lov_obd *lov = &obd->u.lov;
461 /* Only disconnect the underlying layers on the final disconnect. */
463 if (lov->lov_connects != 0) {
464 /* why should there be more than 1 connect? */
465 CERROR("disconnect #%d\n", lov->lov_connects);
469 /* Let's hold another reference so lov_del_obd doesn't spin through
472 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
473 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
474 /* Disconnection is the last we know about an obd */
475 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
481 rc = class_disconnect(exp); /* bz 9811 */
487 * -EINVAL : UUID can't be found in the LOV's target list
488 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
489 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
491 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
494 struct lov_obd *lov = &obd->u.lov;
495 struct lov_tgt_desc *tgt;
499 CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
500 lov, uuid->uuid, activate);
503 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
504 tgt = lov->lov_tgts[i];
505 if (!tgt || !tgt->ltd_exp)
508 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
509 i, obd_uuid2str(&tgt->ltd_uuid),
510 tgt->ltd_exp->exp_handle.h_cookie);
511 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
515 if (i == lov->desc.ld_tgt_count)
516 GOTO(out, rc = -EINVAL);
518 if (lov->lov_tgts[i]->ltd_active == activate) {
519 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
520 activate ? "" : "in");
524 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
525 activate ? "" : "in");
527 lov->lov_tgts[i]->ltd_active = activate;
530 lov->desc.ld_active_tgt_count++;
531 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
533 lov->desc.ld_active_tgt_count--;
534 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
536 /* remove any old qos penalty */
537 lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
544 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
545 enum obd_notify_event ev, void *data)
550 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
551 struct obd_uuid *uuid;
555 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
556 CERROR("unexpected notification of %s %s!\n",
557 watched->obd_type->typ_name,
561 uuid = &watched->u.cli.cl_target_uuid;
563 /* Set OSC as active before notifying the observer, so the
564 * observer can use the OSC normally.
566 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
568 CERROR("%sactivation of %s failed: %d\n",
569 (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
570 obd_uuid2str(uuid), rc);
575 /* Pass the notification up the chain. */
577 rc = obd_notify_observer(obd, watched, ev, data);
579 /* NULL watched means all osc's in the lov (only for syncs) */
580 struct lov_obd *lov = &obd->u.lov;
581 struct obd_device *tgt_obd;
584 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
585 if (!lov->lov_tgts[i])
587 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
588 rc = obd_notify_observer(obd, tgt_obd, ev, data);
590 CERROR("%s: notify %s of %s failed %d\n",
592 obd->obd_observer->obd_name,
593 tgt_obd->obd_name, rc);
603 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
604 __u32 index, int gen, int active)
606 struct lov_obd *lov = &obd->u.lov;
607 struct lov_tgt_desc *tgt;
611 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
612 uuidp->uuid, index, gen, active);
615 CERROR("request to add OBD %s with invalid generation: %d\n",
620 mutex_down(&lov->lov_lock);
622 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
623 tgt = lov->lov_tgts[index];
624 CERROR("UUID %s already assigned at LOV target index %d\n",
625 obd_uuid2str(&tgt->ltd_uuid), index);
626 mutex_up(&lov->lov_lock);
630 if (index >= lov->lov_tgt_size) {
631 /* We need to reallocate the lov target array. */
632 struct lov_tgt_desc **newtgts, **old = NULL;
633 __u32 newsize, oldsize = 0;
635 newsize = max(lov->lov_tgt_size, (__u32)2);
636 while (newsize < index + 1)
637 newsize = newsize << 1;
638 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
639 if (newtgts == NULL) {
640 mutex_up(&lov->lov_lock);
644 if (lov->lov_tgt_size) {
645 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
648 oldsize = lov->lov_tgt_size;
651 lov->lov_tgts = newtgts;
652 lov->lov_tgt_size = newsize;
657 OBD_FREE(old, sizeof(*old) * oldsize);
659 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
660 lov->lov_tgts, lov->lov_tgt_size);
666 mutex_up(&lov->lov_lock);
670 memset(tgt, 0, sizeof(*tgt));
671 tgt->ltd_uuid = *uuidp;
672 /* XXX - add a sanity check on the generation number. */
674 tgt->ltd_index = index;
675 tgt->ltd_activate = active;
676 lov->lov_tgts[index] = tgt;
677 if (index >= lov->desc.ld_tgt_count)
678 lov->desc.ld_tgt_count = index + 1;
679 mutex_up(&lov->lov_lock);
681 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
682 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
684 if (lov->lov_connects == 0) {
685 /* lov_connect hasn't been called yet. We'll do the
686 lov_connect_obd on this target when that fn first runs,
687 because we don't know the connect flags yet. */
693 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
697 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
698 active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
703 CERROR("add failed (%d), deleting %s\n", rc,
704 obd_uuid2str(&tgt->ltd_uuid));
705 lov_del_target(obd, index, 0, 0);
711 /* Schedule a target for deletion */
712 static int lov_del_target(struct obd_device *obd, __u32 index,
713 struct obd_uuid *uuidp, int gen)
715 struct lov_obd *lov = &obd->u.lov;
716 int count = lov->desc.ld_tgt_count;
720 if (index >= count) {
721 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
728 if (!lov->lov_tgts[index]) {
729 CERROR("LOV target at index %d is not setup.\n", index);
730 GOTO(out, rc = -EINVAL);
733 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
734 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
735 lov_uuid2str(lov, index), index,
736 obd_uuid2str(uuidp));
737 GOTO(out, rc = -EINVAL);
740 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
741 lov_uuid2str(lov, index), index,
742 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
743 lov->lov_tgts[index]->ltd_active);
745 lov->lov_tgts[index]->ltd_reap = 1;
746 lov->lov_death_row++;
747 /* we really delete it from lov_putref */
754 /* We are holding lov_lock */
755 static void __lov_del_obd(struct obd_device *obd, __u32 index)
757 struct lov_obd *lov = &obd->u.lov;
758 struct obd_device *osc_obd;
759 struct lov_tgt_desc *tgt = lov->lov_tgts[index];
762 LASSERT(tgt->ltd_reap);
764 osc_obd = class_exp2obd(tgt->ltd_exp);
766 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
767 lov_uuid2str(lov, index),
768 osc_obd ? osc_obd->obd_name : "<no obd>");
771 lov_disconnect_obd(obd, index);
773 /* XXX - right now there is a dependency on ld_tgt_count being the
774 * maximum tgt index for computing the mds_max_easize. So we can't
777 lov->lov_tgts[index] = NULL;
780 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
781 do it ourselves. And we can't do it from lov_cleanup,
782 because we just lost our only reference to it. */
784 class_manual_cleanup(osc_obd);
787 void lov_fix_desc_stripe_size(__u64 *val)
789 if (*val < PTLRPC_MAX_BRW_SIZE) {
790 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
791 PTLRPC_MAX_BRW_SIZE);
792 *val = PTLRPC_MAX_BRW_SIZE;
793 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
794 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
795 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
797 *val, LOV_MIN_STRIPE_SIZE);
801 void lov_fix_desc_stripe_count(__u32 *val)
807 void lov_fix_desc_pattern(__u32 *val)
809 /* from lov_setstripe */
810 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
811 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
816 void lov_fix_desc_qos_maxage(__u32 *val)
820 *val = QOS_DEFAULT_MAXAGE;
823 void lov_fix_desc(struct lov_desc *desc)
825 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
826 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
827 lov_fix_desc_pattern(&desc->ld_pattern);
828 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
831 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
833 struct lprocfs_static_vars lvars = { 0 };
834 struct lov_desc *desc;
835 struct lov_obd *lov = &obd->u.lov;
839 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
840 CERROR("LOV setup requires a descriptor\n");
844 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
846 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
847 CERROR("descriptor size wrong: %d > %d\n",
848 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
852 if (desc->ld_magic != LOV_DESC_MAGIC) {
853 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
854 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
855 obd->obd_name, desc);
856 lustre_swab_lov_desc(desc);
858 CERROR("%s: Bad lov desc magic: %#x\n",
859 obd->obd_name, desc->ld_magic);
866 /* Because of 64-bit divide/mod operations only work with a 32-bit
867 * divisor in a 32-bit kernel, we cannot support a stripe width
868 * of 4GB or larger on 32-bit CPUs. */
869 count = desc->ld_default_stripe_count;
870 if ((count > 0 ? count : desc->ld_tgt_count) *
871 desc->ld_default_stripe_size > 0xffffffff) {
872 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
873 desc->ld_default_stripe_size, count);
877 desc->ld_active_tgt_count = 0;
879 lov->lov_tgt_size = 0;
880 sema_init(&lov->lov_lock, 1);
881 atomic_set(&lov->lov_refcount, 0);
882 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
883 init_rwsem(&lov->lov_qos.lq_rw_sem);
884 lov->lov_qos.lq_dirty = 1;
885 lov->lov_qos.lq_dirty_rr = 1;
886 lov->lov_qos.lq_reset = 1;
887 /* Default priority is toward free space balance */
888 lov->lov_qos.lq_prio_free = 232;
890 lprocfs_lov_init_vars(&lvars);
891 lprocfs_obd_setup(obd, lvars.obd_vars);
896 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
897 0444, &lov_proc_target_fops, obd);
899 CWARN("Error adding the target_obd file\n");
906 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
912 case OBD_CLEANUP_EARLY: {
913 struct lov_obd *lov = &obd->u.lov;
915 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
916 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
918 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
923 case OBD_CLEANUP_EXPORTS:
924 rc = obd_llog_finish(obd, 0);
926 CERROR("failed to cleanup llogging subsystems\n");
932 static int lov_cleanup(struct obd_device *obd)
934 struct lov_obd *lov = &obd->u.lov;
936 lprocfs_obd_cleanup(obd);
939 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
940 if (!lov->lov_tgts[i])
943 /* Inactive targets may never have connected */
944 if (lov->lov_tgts[i]->ltd_active ||
945 atomic_read(&lov->lov_refcount))
946 /* We should never get here - these
947 should have been removed in the
949 CERROR("lov tgt %d not cleaned!"
950 " deathrow=%d, lovrc=%d\n",
951 i, lov->lov_death_row,
952 atomic_read(&lov->lov_refcount));
953 lov_del_target(obd, i, 0, 0);
955 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
957 lov->lov_tgt_size = 0;
960 if (lov->lov_qos.lq_rr_size)
961 OBD_FREE(lov->lov_qos.lq_rr_array, lov->lov_qos.lq_rr_size);
966 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
968 struct lustre_cfg *lcfg = buf;
969 struct obd_uuid obd_uuid;
974 switch(cmd = lcfg->lcfg_command) {
975 case LCFG_LOV_ADD_OBD:
976 case LCFG_LOV_ADD_INA:
977 case LCFG_LOV_DEL_OBD: {
980 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
981 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
982 GOTO(out, rc = -EINVAL);
984 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
986 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
987 GOTO(out, rc = -EINVAL);
988 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
989 GOTO(out, rc = -EINVAL);
990 if (cmd == LCFG_LOV_ADD_OBD)
991 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
992 else if (cmd == LCFG_LOV_ADD_INA)
993 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
995 rc = lov_del_target(obd, index, &obd_uuid, gen);
999 struct lprocfs_static_vars lvars = { 0 };
1000 struct lov_desc *desc = &(obd->u.lov.desc);
1003 GOTO(out, rc = -EINVAL);
1005 lprocfs_lov_init_vars(&lvars);
1007 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1012 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1013 GOTO(out, rc = -EINVAL);
1022 #define log2(n) ffz(~(n))
1025 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1026 struct lov_stripe_md **ea,
1027 struct obd_trans_info *oti)
1029 struct lov_obd *lov;
1030 struct obdo *tmp_oa;
1031 struct obd_uuid *ost_uuid = NULL;
1035 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1036 src_oa->o_flags == OBD_FL_DELORPHAN);
1038 lov = &export->exp_obd->u.lov;
1044 if (src_oa->o_valid & OBD_MD_FLINLINE) {
1045 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
1046 CDEBUG(D_HA, "clearing orphans only for %s\n",
1050 lov_getref(export->exp_obd);
1051 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1052 struct lov_stripe_md obj_md;
1053 struct lov_stripe_md *obj_mdp = &obj_md;
1054 struct lov_tgt_desc *tgt;
1057 tgt = lov->lov_tgts[i];
1061 /* if called for a specific target, we don't
1062 care if it is not active. */
1063 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1064 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1068 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1071 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1072 obd_uuid2str(ost_uuid));
1074 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1076 LASSERT(lov->lov_tgts[i]->ltd_exp);
1077 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1078 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1079 tmp_oa, &obj_mdp, oti);
1081 /* This export will be disabled until it is recovered,
1082 and then orphan recovery will be completed. */
1083 CERROR("error in orphan recovery on OST idx %d/%d: "
1084 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1089 lov_putref(export->exp_obd);
1095 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1096 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1098 struct lov_stripe_md *obj_mdp, *lsm;
1099 struct lov_obd *lov = &exp->exp_obd->u.lov;
1104 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1105 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1107 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1108 if (obj_mdp == NULL)
1111 ost_idx = src_oa->o_nlink;
1114 GOTO(out, rc = -EINVAL);
1115 if (ost_idx >= lov->desc.ld_tgt_count ||
1116 !lov->lov_tgts[ost_idx])
1117 GOTO(out, rc = -EINVAL);
1119 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1120 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1121 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1122 GOTO(out, rc = -EINVAL);
1126 if (i == lsm->lsm_stripe_count)
1127 GOTO(out, rc = -EINVAL);
1129 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1131 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1135 /* the LOV expects oa->o_id to be set to the LOV object id */
1136 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1137 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1139 struct lov_obd *lov;
1140 struct obd_info oinfo;
1141 struct lov_request_set *set = NULL;
1142 struct lov_request *req;
1143 struct obd_statfs osfs;
1148 LASSERT(ea != NULL);
1152 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1153 src_oa->o_flags == OBD_FL_DELORPHAN) {
1154 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1158 lov = &exp->exp_obd->u.lov;
1159 if (!lov->desc.ld_active_tgt_count)
1162 /* Recreate a specific object id at the given OST index */
1163 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1164 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1165 rc = lov_recreate(exp, src_oa, ea, oti);
1169 maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1170 obd_statfs_rqset(exp->exp_obd, &osfs, maxage, OBD_STATFS_NODELAY);
1172 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1176 list_for_each_entry(req, &set->set_list, rq_link) {
1177 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1178 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1179 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1180 lov_update_create_set(set, req, rc);
1182 rc = lov_fini_create_set(set, ea);
1186 #define ASSERT_LSM_MAGIC(lsmp) \
1188 LASSERT((lsmp) != NULL); \
1189 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC || \
1190 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1191 (lsmp), (lsmp)->lsm_magic); \
1194 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1195 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1196 struct obd_export *md_exp)
1198 struct lov_request_set *set;
1199 struct obd_info oinfo;
1200 struct lov_request *req;
1201 struct list_head *pos;
1202 struct lov_obd *lov;
1206 ASSERT_LSM_MAGIC(lsm);
1208 if (!exp || !exp->exp_obd)
1211 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1213 LASSERT(oti->oti_logcookies);
1216 lov = &exp->exp_obd->u.lov;
1217 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1221 list_for_each (pos, &set->set_list) {
1223 req = list_entry(pos, struct lov_request, rq_link);
1225 if (oa->o_valid & OBD_MD_FLCOOKIE)
1226 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1228 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1229 req->rq_oi.oi_oa, NULL, oti, NULL);
1230 err = lov_update_common_set(set, req, err);
1232 CERROR("error: destroying objid "LPX64" subobj "
1233 LPX64" on OST idx %d: rc = %d\n",
1234 oa->o_id, req->rq_oi.oi_oa->o_id,
1242 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1243 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1245 err = lov_fini_destroy_set(set);
1246 RETURN(rc ? rc : err);
1249 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1251 struct lov_request_set *set;
1252 struct lov_request *req;
1253 struct list_head *pos;
1254 struct lov_obd *lov;
1255 int err = 0, rc = 0;
1259 ASSERT_LSM_MAGIC(oinfo->oi_md);
1261 if (!exp || !exp->exp_obd)
1264 lov = &exp->exp_obd->u.lov;
1266 rc = lov_prep_getattr_set(exp, oinfo, &set);
1270 list_for_each (pos, &set->set_list) {
1271 req = list_entry(pos, struct lov_request, rq_link);
1273 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1274 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1275 req->rq_oi.oi_oa->o_id, req->rq_idx);
1277 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1279 err = lov_update_common_set(set, req, rc);
1281 CERROR("error: getattr objid "LPX64" subobj "
1282 LPX64" on OST idx %d: rc = %d\n",
1283 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1289 rc = lov_fini_getattr_set(set);
1295 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1298 struct lov_request_set *lovset = (struct lov_request_set *)data;
1302 /* don't do attribute merge if this aysnc op failed */
1304 lovset->set_completes = 0;
1305 err = lov_fini_getattr_set(lovset);
1306 RETURN(rc ? rc : err);
1309 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1310 struct ptlrpc_request_set *rqset)
1312 struct lov_request_set *lovset;
1313 struct lov_obd *lov;
1314 struct list_head *pos;
1315 struct lov_request *req;
1320 ASSERT_LSM_MAGIC(oinfo->oi_md);
1322 if (!exp || !exp->exp_obd)
1325 lov = &exp->exp_obd->u.lov;
1327 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1331 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1332 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1333 oinfo->oi_md->lsm_stripe_size);
1335 list_for_each (pos, &lovset->set_list) {
1336 req = list_entry(pos, struct lov_request, rq_link);
1338 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1339 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1340 req->rq_oi.oi_oa->o_id, req->rq_idx);
1341 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1342 &req->rq_oi, rqset);
1344 CERROR("error: getattr objid "LPX64" subobj "
1345 LPX64" on OST idx %d: rc = %d\n",
1346 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1352 if (!list_empty(&rqset->set_requests)) {
1354 LASSERT (rqset->set_interpret == NULL);
1355 rqset->set_interpret = lov_getattr_interpret;
1356 rqset->set_arg = (void *)lovset;
1361 lovset->set_completes = 0;
1362 err = lov_fini_getattr_set(lovset);
1363 RETURN(rc ? rc : err);
1366 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1367 struct obd_trans_info *oti)
1369 struct lov_request_set *set;
1370 struct lov_obd *lov;
1371 struct list_head *pos;
1372 struct lov_request *req;
1373 int err = 0, rc = 0;
1377 ASSERT_LSM_MAGIC(oinfo->oi_md);
1379 if (!exp || !exp->exp_obd)
1382 /* for now, we only expect the following updates here */
1383 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1384 OBD_MD_FLMODE | OBD_MD_FLATIME |
1385 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1386 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1387 OBD_MD_FLGROUP | OBD_MD_FLUID |
1388 OBD_MD_FLGID | OBD_MD_FLINLINE |
1389 OBD_MD_FLFID | OBD_MD_FLGENER)));
1390 lov = &exp->exp_obd->u.lov;
1391 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1395 list_for_each (pos, &set->set_list) {
1396 req = list_entry(pos, struct lov_request, rq_link);
1398 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1400 err = lov_update_setattr_set(set, req, rc);
1402 CERROR("error: setattr objid "LPX64" subobj "
1403 LPX64" on OST idx %d: rc = %d\n",
1404 set->set_oi->oi_oa->o_id,
1405 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1410 err = lov_fini_setattr_set(set);
1416 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1419 struct lov_request_set *lovset = (struct lov_request_set *)data;
1424 lovset->set_completes = 0;
1425 err = lov_fini_setattr_set(lovset);
1426 RETURN(rc ? rc : err);
1429 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1430 needed. Otherwise, a client is waiting for responses. */
1431 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1432 struct obd_trans_info *oti,
1433 struct ptlrpc_request_set *rqset)
1435 struct lov_request_set *set;
1436 struct lov_request *req;
1437 struct list_head *pos;
1438 struct lov_obd *lov;
1443 ASSERT_LSM_MAGIC(oinfo->oi_md);
1444 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1446 LASSERT(oti->oti_logcookies);
1449 if (!exp || !exp->exp_obd)
1452 lov = &exp->exp_obd->u.lov;
1453 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1457 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1458 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1459 oinfo->oi_md->lsm_stripe_size);
1461 list_for_each (pos, &set->set_list) {
1462 req = list_entry(pos, struct lov_request, rq_link);
1464 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1465 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1467 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1468 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1469 req->rq_oi.oi_oa->o_id, req->rq_idx);
1471 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1472 &req->rq_oi, oti, rqset);
1474 CERROR("error: setattr objid "LPX64" subobj "
1475 LPX64" on OST idx %d: rc = %d\n",
1476 set->set_oi->oi_oa->o_id,
1477 req->rq_oi.oi_oa->o_id,
1483 /* If we are not waiting for responses on async requests, return. */
1484 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1487 set->set_completes = 0;
1488 err = lov_fini_setattr_set(set);
1489 RETURN(rc ? rc : err);
1492 LASSERT(rqset->set_interpret == NULL);
1493 rqset->set_interpret = lov_setattr_interpret;
1494 rqset->set_arg = (void *)set;
1499 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1502 struct lov_request_set *lovset = (struct lov_request_set *)data;
1507 lovset->set_completes = 0;
1508 err = lov_fini_punch_set(lovset);
1509 RETURN(rc ? rc : err);
1512 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1513 * we can send this 'punch' to just the authoritative node and the nodes
1514 * that the punch will affect. */
1515 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1516 struct obd_trans_info *oti,
1517 struct ptlrpc_request_set *rqset)
1519 struct lov_request_set *set;
1520 struct lov_obd *lov;
1521 struct list_head *pos;
1522 struct lov_request *req;
1527 ASSERT_LSM_MAGIC(oinfo->oi_md);
1529 if (!exp || !exp->exp_obd)
1532 lov = &exp->exp_obd->u.lov;
1533 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1537 list_for_each (pos, &set->set_list) {
1538 req = list_entry(pos, struct lov_request, rq_link);
1540 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1541 &req->rq_oi, NULL, rqset);
1543 CERROR("error: punch objid "LPX64" subobj "LPX64
1544 " on OST idx %d: rc = %d\n",
1545 set->set_oi->oi_oa->o_id,
1546 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1551 if (rc || list_empty(&rqset->set_requests)) {
1553 err = lov_fini_punch_set(set);
1554 RETURN(rc ? rc : err);
1557 LASSERT(rqset->set_interpret == NULL);
1558 rqset->set_interpret = lov_punch_interpret;
1559 rqset->set_arg = (void *)set;
1564 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1565 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1568 struct lov_request_set *set;
1569 struct obd_info oinfo;
1570 struct lov_obd *lov;
1571 struct list_head *pos;
1572 struct lov_request *req;
1573 int err = 0, rc = 0;
1576 ASSERT_LSM_MAGIC(lsm);
1581 lov = &exp->exp_obd->u.lov;
1582 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1586 list_for_each (pos, &set->set_list) {
1587 req = list_entry(pos, struct lov_request, rq_link);
1589 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1590 req->rq_oi.oi_oa, NULL,
1591 req->rq_oi.oi_policy.l_extent.start,
1592 req->rq_oi.oi_policy.l_extent.end, capa);
1593 err = lov_update_common_set(set, req, rc);
1595 CERROR("error: fsync objid "LPX64" subobj "LPX64
1596 " on OST idx %d: rc = %d\n",
1597 set->set_oi->oi_oa->o_id,
1598 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1603 err = lov_fini_sync_set(set);
1609 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1610 obd_count oa_bufs, struct brw_page *pga)
1612 struct obd_info oinfo = { { { 0 } } };
1615 oinfo.oi_oa = lov_oinfo->oi_oa;
1617 /* The caller just wants to know if there's a chance that this
1618 * I/O can succeed */
1619 for (i = 0; i < oa_bufs; i++) {
1620 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1621 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1624 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1625 pga[i].off + pga[i].count,
1629 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1630 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1634 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1642 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1643 obd_count oa_bufs, struct brw_page *pga,
1644 struct obd_trans_info *oti)
1646 struct lov_request_set *set;
1647 struct lov_request *req;
1648 struct list_head *pos;
1649 struct lov_obd *lov = &exp->exp_obd->u.lov;
1653 ASSERT_LSM_MAGIC(oinfo->oi_md);
1655 if (cmd == OBD_BRW_CHECK) {
1656 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1660 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1664 list_for_each (pos, &set->set_list) {
1665 struct obd_export *sub_exp;
1666 struct brw_page *sub_pga;
1667 req = list_entry(pos, struct lov_request, rq_link);
1669 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1670 sub_pga = set->set_pga + req->rq_pgaidx;
1671 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1675 lov_update_common_set(set, req, rc);
1678 err = lov_fini_brw_set(set);
1684 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1687 struct lov_request_set *lovset = (struct lov_request_set *)data;
1691 lovset->set_completes = 0;
1692 lov_fini_brw_set(lovset);
1694 rc = lov_fini_brw_set(lovset);
1700 static int lov_brw_async(int cmd, struct obd_export *exp,
1701 struct obd_info *oinfo, obd_count oa_bufs,
1702 struct brw_page *pga, struct obd_trans_info *oti,
1703 struct ptlrpc_request_set *set)
1705 struct lov_request_set *lovset;
1706 struct lov_request *req;
1707 struct list_head *pos;
1708 struct lov_obd *lov = &exp->exp_obd->u.lov;
1713 ASSERT_LSM_MAGIC(oinfo->oi_md);
1715 if (cmd == OBD_BRW_CHECK) {
1716 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1720 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1724 list_for_each (pos, &lovset->set_list) {
1725 struct obd_export *sub_exp;
1726 struct brw_page *sub_pga;
1727 req = list_entry(pos, struct lov_request, rq_link);
1729 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1730 sub_pga = lovset->set_pga + req->rq_pgaidx;
1731 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1735 lov_update_common_set(lovset, req, rc);
1738 LASSERT(set->set_interpret == NULL);
1739 LASSERT(set->set_arg == NULL);
1740 rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1746 lov_fini_brw_set(lovset);
1750 static int lov_ap_make_ready(void *data, int cmd)
1752 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1754 return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1757 static int lov_ap_refresh_count(void *data, int cmd)
1759 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1761 return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1765 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1767 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1769 lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1770 /* XXX woah, shouldn't we be altering more here? size? */
1771 oa->o_id = lap->lap_loi_id;
1772 oa->o_gr = lap->lap_loi_gr;
1773 oa->o_valid |= OBD_MD_FLGROUP;
1774 oa->o_stripe_idx = lap->lap_stripe;
1777 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1780 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1782 lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1785 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1787 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1789 /* in a raid1 regime this would down a count of many ios
1790 * in flight, onl calling the caller_ops completion when all
1791 * the raid1 ios are complete */
1792 rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1796 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1798 struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1799 return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1802 static struct obd_async_page_ops lov_async_page_ops = {
1803 .ap_make_ready = lov_ap_make_ready,
1804 .ap_refresh_count = lov_ap_refresh_count,
1805 .ap_fill_obdo = lov_ap_fill_obdo,
1806 .ap_update_obdo = lov_ap_update_obdo,
1807 .ap_completion = lov_ap_completion,
1808 .ap_lookup_capa = lov_ap_lookup_capa,
1811 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1812 struct lov_oinfo *loi, cfs_page_t *page,
1813 obd_off offset, struct obd_async_page_ops *ops,
1814 void *data, void **res, int nocache,
1815 struct lustre_handle *lockh)
1817 struct lov_obd *lov = &exp->exp_obd->u.lov;
1818 struct lov_async_page *lap;
1819 struct lov_lock_handles *lov_lockh = NULL;
1825 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1826 Only because of this layering limitation will a client
1827 mount with no osts fail */
1828 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1829 !lov->lov_tgts[i]->ltd_exp) {
1831 if (i >= lov->desc.ld_tgt_count)
1834 rc = size_round(sizeof(*lap)) +
1835 obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1836 NULL, NULL, 0, NULL, NULL, NULL, 0,
1840 ASSERT_LSM_MAGIC(lsm);
1841 LASSERT(loi == NULL);
1844 lap->lap_magic = LOV_AP_MAGIC;
1845 lap->lap_caller_ops = ops;
1846 lap->lap_caller_data = data;
1848 /* for now only raid 0 which passes through */
1849 lap->lap_stripe = lov_stripe_number(lsm, offset);
1850 lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1851 loi = lsm->lsm_oinfo[lap->lap_stripe];
1853 /* so the callback doesn't need the lsm */
1854 lap->lap_loi_id = loi->loi_id;
1855 lap->lap_loi_gr = lsm->lsm_object_gr;
1856 LASSERT(lsm->lsm_object_gr > 0);
1858 lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1861 lov_lockh = lov_handle2llh(lockh);
1863 lockh = lov_lockh->llh_handles + lap->lap_stripe;
1867 rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1868 lsm, loi, page, lap->lap_sub_offset,
1869 &lov_async_page_ops, lap,
1870 &lap->lap_sub_cookie, nocache, lockh);
1872 lov_llh_put(lov_lockh);
1875 CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1876 lap->lap_sub_cookie, offset);
1880 static int lov_queue_async_io(struct obd_export *exp,
1881 struct lov_stripe_md *lsm,
1882 struct lov_oinfo *loi, void *cookie,
1883 int cmd, obd_off off, int count,
1884 obd_flag brw_flags, obd_flag async_flags)
1886 struct lov_obd *lov = &exp->exp_obd->u.lov;
1887 struct lov_async_page *lap;
1890 LASSERT(loi == NULL);
1892 ASSERT_LSM_MAGIC(lsm);
1894 lap = LAP_FROM_COOKIE(cookie);
1896 loi = lsm->lsm_oinfo[lap->lap_stripe];
1898 rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1899 loi, lap->lap_sub_cookie, cmd, off, count,
1900 brw_flags, async_flags);
1904 static int lov_set_async_flags(struct obd_export *exp,
1905 struct lov_stripe_md *lsm,
1906 struct lov_oinfo *loi, void *cookie,
1907 obd_flag async_flags)
1909 struct lov_obd *lov = &exp->exp_obd->u.lov;
1910 struct lov_async_page *lap;
1913 LASSERT(loi == NULL);
1915 ASSERT_LSM_MAGIC(lsm);
1917 lap = LAP_FROM_COOKIE(cookie);
1919 loi = lsm->lsm_oinfo[lap->lap_stripe];
1921 rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1922 lsm, loi, lap->lap_sub_cookie, async_flags);
1926 static int lov_queue_group_io(struct obd_export *exp,
1927 struct lov_stripe_md *lsm,
1928 struct lov_oinfo *loi,
1929 struct obd_io_group *oig, void *cookie,
1930 int cmd, obd_off off, int count,
1931 obd_flag brw_flags, obd_flag async_flags)
1933 struct lov_obd *lov = &exp->exp_obd->u.lov;
1934 struct lov_async_page *lap;
1937 LASSERT(loi == NULL);
1939 ASSERT_LSM_MAGIC(lsm);
1941 lap = LAP_FROM_COOKIE(cookie);
1943 loi = lsm->lsm_oinfo[lap->lap_stripe];
1945 rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1946 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1947 brw_flags, async_flags);
1951 /* this isn't exactly optimal. we may have queued sync io in oscs on
1952 * all stripes, but we don't record that fact at queue time. so we
1953 * trigger sync io on all stripes. */
1954 static int lov_trigger_group_io(struct obd_export *exp,
1955 struct lov_stripe_md *lsm,
1956 struct lov_oinfo *loi,
1957 struct obd_io_group *oig)
1959 struct lov_obd *lov = &exp->exp_obd->u.lov;
1962 LASSERT(loi == NULL);
1964 ASSERT_LSM_MAGIC(lsm);
1966 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1967 loi = lsm->lsm_oinfo[i];
1968 if (!lov->lov_tgts[loi->loi_ost_idx] ||
1969 !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1970 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1974 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1976 if (rc == 0 && err != 0)
1982 static int lov_teardown_async_page(struct obd_export *exp,
1983 struct lov_stripe_md *lsm,
1984 struct lov_oinfo *loi, void *cookie)
1986 struct lov_obd *lov = &exp->exp_obd->u.lov;
1987 struct lov_async_page *lap;
1990 LASSERT(loi == NULL);
1992 ASSERT_LSM_MAGIC(lsm);
1994 lap = LAP_FROM_COOKIE(cookie);
1996 loi = lsm->lsm_oinfo[lap->lap_stripe];
1998 rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1999 lsm, loi, lap->lap_sub_cookie);
2001 CERROR("unable to teardown sub cookie %p: %d\n",
2002 lap->lap_sub_cookie, rc);
2008 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2011 struct lov_request_set *lovset = (struct lov_request_set *)data;
2013 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2017 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2018 struct ldlm_enqueue_info *einfo,
2019 struct ptlrpc_request_set *rqset)
2021 ldlm_mode_t mode = einfo->ei_mode;
2022 struct lov_request_set *set;
2023 struct lov_request *req;
2024 struct list_head *pos;
2025 struct lov_obd *lov;
2030 ASSERT_LSM_MAGIC(oinfo->oi_md);
2031 LASSERT(mode == (mode & -mode));
2033 /* we should never be asked to replay a lock this way. */
2034 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2036 if (!exp || !exp->exp_obd)
2039 lov = &exp->exp_obd->u.lov;
2040 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2044 list_for_each (pos, &set->set_list) {
2045 req = list_entry(pos, struct lov_request, rq_link);
2047 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2048 &req->rq_oi, einfo, rqset);
2053 if (rqset && !list_empty(&rqset->set_requests)) {
2055 LASSERT(rqset->set_interpret == NULL);
2056 rqset->set_interpret = lov_enqueue_interpret;
2057 rqset->set_arg = (void *)set;
2061 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2065 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2066 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2067 int *flags, void *data, struct lustre_handle *lockh)
2069 struct lov_request_set *set;
2070 struct obd_info oinfo;
2071 struct lov_request *req;
2072 struct list_head *pos;
2073 struct lov_obd *lov = &exp->exp_obd->u.lov;
2074 struct lustre_handle *lov_lockhp;
2075 int lov_flags, rc = 0;
2078 ASSERT_LSM_MAGIC(lsm);
2079 LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2081 if (!exp || !exp->exp_obd)
2084 lov = &exp->exp_obd->u.lov;
2085 rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2089 list_for_each (pos, &set->set_list) {
2090 ldlm_policy_data_t sub_policy;
2091 req = list_entry(pos, struct lov_request, rq_link);
2092 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2093 LASSERT(lov_lockhp);
2096 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2098 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2099 req->rq_oi.oi_md, type, &sub_policy,
2100 mode, &lov_flags, data, lov_lockhp);
2101 rc = lov_update_match_set(set, req, rc);
2105 lov_fini_match_set(set, mode, *flags);
2109 static int lov_change_cbdata(struct obd_export *exp,
2110 struct lov_stripe_md *lsm, ldlm_iterator_t it,
2113 struct lov_obd *lov;
2117 ASSERT_LSM_MAGIC(lsm);
2119 if (!exp || !exp->exp_obd)
2122 LASSERT(lsm->lsm_object_gr > 0);
2124 lov = &exp->exp_obd->u.lov;
2125 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2126 struct lov_stripe_md submd;
2127 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2129 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2130 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2134 submd.lsm_object_id = loi->loi_id;
2135 submd.lsm_object_gr = lsm->lsm_object_gr;
2136 submd.lsm_stripe_count = 0;
2137 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2143 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2144 __u32 mode, struct lustre_handle *lockh)
2146 struct lov_request_set *set;
2147 struct obd_info oinfo;
2148 struct lov_request *req;
2149 struct list_head *pos;
2150 struct lov_obd *lov = &exp->exp_obd->u.lov;
2151 struct lustre_handle *lov_lockhp;
2152 int err = 0, rc = 0;
2155 ASSERT_LSM_MAGIC(lsm);
2157 if (!exp || !exp->exp_obd)
2160 LASSERT(lsm->lsm_object_gr > 0);
2162 lov = &exp->exp_obd->u.lov;
2163 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2167 list_for_each (pos, &set->set_list) {
2168 req = list_entry(pos, struct lov_request, rq_link);
2169 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2171 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2172 req->rq_oi.oi_md, mode, lov_lockhp);
2173 rc = lov_update_common_set(set, req, rc);
2175 CERROR("error: cancel objid "LPX64" subobj "
2176 LPX64" on OST idx %d: rc = %d\n",
2178 req->rq_oi.oi_md->lsm_object_id,
2184 lov_fini_cancel_set(set);
2188 static int lov_cancel_unused(struct obd_export *exp,
2189 struct lov_stripe_md *lsm,
2190 int flags, void *opaque)
2192 struct lov_obd *lov;
2196 if (!exp || !exp->exp_obd)
2199 lov = &exp->exp_obd->u.lov;
2201 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2203 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2206 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2214 ASSERT_LSM_MAGIC(lsm);
2216 LASSERT(lsm->lsm_object_gr > 0);
2217 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2218 struct lov_stripe_md submd;
2219 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2222 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2223 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2227 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2228 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2230 submd.lsm_object_id = loi->loi_id;
2231 submd.lsm_object_gr = lsm->lsm_object_gr;
2232 submd.lsm_stripe_count = 0;
2233 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2234 &submd, flags, opaque);
2235 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2236 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2237 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2238 loi->loi_id, loi->loi_ost_idx, err);
2246 static int lov_join_lru(struct obd_export *exp,
2247 struct lov_stripe_md *lsm, int join)
2249 struct lov_obd *lov;
2253 ASSERT_LSM_MAGIC(lsm);
2254 if (!exp || !exp->exp_obd)
2257 lov = &exp->exp_obd->u.lov;
2258 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2259 struct lov_stripe_md submd;
2260 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2263 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2264 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2268 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2269 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2271 submd.lsm_object_id = loi->loi_id;
2272 submd.lsm_object_gr = lsm->lsm_object_gr;
2273 submd.lsm_stripe_count = 0;
2274 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2277 CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2278 " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2279 loi->loi_id, loi->loi_ost_idx, rc);
2288 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2291 struct lov_request_set *lovset = (struct lov_request_set *)data;
2296 lovset->set_completes = 0;
2298 err = lov_fini_statfs_set(lovset);
2299 RETURN(rc ? rc : err);
2302 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2303 __u64 max_age, struct ptlrpc_request_set *rqset)
2305 struct lov_request_set *set;
2306 struct lov_request *req;
2307 struct list_head *pos;
2308 struct lov_obd *lov;
2312 LASSERT(oinfo != NULL);
2313 LASSERT(oinfo->oi_osfs != NULL);
2316 rc = lov_prep_statfs_set(obd, oinfo, &set);
2320 list_for_each (pos, &set->set_list) {
2321 struct obd_device *osc_obd;
2323 req = list_entry(pos, struct lov_request, rq_link);
2325 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2326 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2331 if (rc || list_empty(&rqset->set_requests)) {
2334 set->set_completes = 0;
2335 err = lov_fini_statfs_set(set);
2336 RETURN(rc ? rc : err);
2339 LASSERT(rqset->set_interpret == NULL);
2340 rqset->set_interpret = lov_statfs_interpret;
2341 rqset->set_arg = (void *)set;
2345 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2346 __u64 max_age, __u32 flags)
2348 struct ptlrpc_request_set *set = NULL;
2349 struct obd_info oinfo = { { { 0 } } };
2354 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2355 * statfs requests */
2356 set = ptlrpc_prep_set();
2360 oinfo.oi_osfs = osfs;
2361 oinfo.oi_flags = flags;
2362 rc = lov_statfs_async(obd, &oinfo, max_age, set);
2364 rc = ptlrpc_set_wait(set);
2365 ptlrpc_set_destroy(set);
2370 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2371 void *karg, void *uarg)
2373 struct obd_device *obddev = class_exp2obd(exp);
2374 struct lov_obd *lov = &obddev->u.lov;
2375 int i, rc, count = lov->desc.ld_tgt_count;
2376 struct obd_uuid *uuidp;
2380 case IOC_OBD_STATFS: {
2381 struct obd_ioctl_data *data = karg;
2382 struct obd_device *osc_obd;
2383 struct obd_statfs stat_buf = {0};
2386 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2387 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
2389 if ((index >= count))
2392 if (!lov->lov_tgts[index])
2393 /* Try again with the next index */
2395 if (!lov->lov_tgts[index]->ltd_active)
2398 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2402 /* got statfs data */
2403 rc = obd_statfs(osc_obd, &stat_buf,
2404 cfs_time_current_64() - HZ, 0);
2407 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2410 rc = copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2414 case OBD_IOC_LOV_GET_CONFIG: {
2415 struct obd_ioctl_data *data;
2416 struct lov_desc *desc;
2421 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2424 data = (struct obd_ioctl_data *)buf;
2426 if (sizeof(*desc) > data->ioc_inllen1) {
2427 obd_ioctl_freedata(buf, len);
2431 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2432 obd_ioctl_freedata(buf, len);
2436 if (sizeof(__u32) * count > data->ioc_inllen3) {
2437 obd_ioctl_freedata(buf, len);
2441 desc = (struct lov_desc *)data->ioc_inlbuf1;
2442 memcpy(desc, &(lov->desc), sizeof(*desc));
2444 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2445 genp = (__u32 *)data->ioc_inlbuf3;
2446 /* the uuid will be empty for deleted OSTs */
2447 for (i = 0; i < count; i++, uuidp++, genp++) {
2448 if (!lov->lov_tgts[i])
2450 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2451 *genp = lov->lov_tgts[i]->ltd_gen;
2454 rc = copy_to_user((void *)uarg, buf, len);
2457 obd_ioctl_freedata(buf, len);
2460 case LL_IOC_LOV_SETSTRIPE:
2461 rc = lov_setstripe(exp, karg, uarg);
2463 case LL_IOC_LOV_GETSTRIPE:
2464 rc = lov_getstripe(exp, karg, uarg);
2466 case LL_IOC_LOV_SETEA:
2467 rc = lov_setea(exp, karg, uarg);
2476 for (i = 0; i < count; i++) {
2479 /* OST was disconnected */
2480 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2483 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2485 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2488 if (lov->lov_tgts[i]->ltd_active) {
2489 CDEBUG(err == -ENOTTY ?
2490 D_IOCTL : D_WARNING,
2491 "iocontrol OSC %s on OST "
2492 "idx %d cmd %x: err = %d\n",
2493 lov_uuid2str(lov, i),
2510 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2511 void *key, __u32 *vallen, void *val,
2512 struct lov_stripe_md *lsm)
2514 struct obd_device *obddev = class_exp2obd(exp);
2515 struct lov_obd *lov = &obddev->u.lov;
2519 if (!vallen || !val)
2524 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2527 struct ldlm_lock *lock;
2529 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2530 struct lov_oinfo *loi;
2531 __u32 *stripe = val;
2533 if (*vallen < sizeof(*stripe))
2534 GOTO(out, rc = -EFAULT);
2535 *vallen = sizeof(*stripe);
2537 /* XXX This is another one of those bits that will need to
2538 * change if we ever actually support nested LOVs. It uses
2539 * the lock's export to find out which stripe it is. */
2540 /* XXX - it's assumed all the locks for deleted OSTs have
2541 * been cancelled. Also, the export for deleted OSTs will
2542 * be NULL and won't match the lock's export. */
2543 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2544 loi = lsm->lsm_oinfo[i];
2545 if (!lov->lov_tgts[loi->loi_ost_idx])
2547 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2548 data->lock->l_conn_export &&
2549 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2554 LDLM_ERROR(data->lock, "lock on inode without such object");
2555 dump_lsm(D_ERROR, lsm);
2556 GOTO(out, rc = -ENXIO);
2557 } else if (KEY_IS(KEY_LAST_ID)) {
2558 struct obd_id_info *info = val;
2559 __u32 size = sizeof(obd_id);
2560 struct lov_tgt_desc *tgt;
2562 LASSERT(*vallen == sizeof(struct obd_id_info));
2563 tgt = lov->lov_tgts[info->idx];
2565 if (!tgt || !tgt->ltd_active)
2566 GOTO(out, rc = -ESRCH);
2568 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2570 } else if (KEY_IS(KEY_LOVDESC)) {
2571 struct lov_desc *desc_ret = val;
2572 *desc_ret = lov->desc;
2575 } else if (KEY_IS(KEY_LOV_IDX)) {
2576 struct lov_tgt_desc *tgt;
2578 for(i = 0; i < lov->desc.ld_tgt_count; i++) {
2579 tgt = lov->lov_tgts[i];
2580 if (tgt && obd_uuid_equals(val, &tgt->ltd_uuid))
2591 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2592 void *key, obd_count vallen, void *val,
2593 struct ptlrpc_request_set *set)
2595 struct obd_device *obddev = class_exp2obd(exp);
2596 struct lov_obd *lov = &obddev->u.lov;
2599 struct lov_tgt_desc *tgt;
2600 unsigned incr, check_uuid,
2601 do_inactive, no_set;
2602 unsigned next_id = 0, mds_con = 0;
2605 incr = check_uuid = do_inactive = no_set = 0;
2608 set = ptlrpc_prep_set();
2614 count = lov->desc.ld_tgt_count;
2616 if (KEY_IS(KEY_NEXT_ID)) {
2617 count = vallen / sizeof(struct obd_id_info);
2618 vallen = sizeof(obd_id);
2619 incr = sizeof(struct obd_id_info);
2622 } else if (KEY_IS(KEY_CHECKSUM)) {
2624 } else if (KEY_IS(KEY_UNLINKED)) {
2625 check_uuid = val ? 1 : 0;
2626 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2627 /* use defaults: do_inactive = incr = 0; */
2628 } else if (KEY_IS(KEY_MDS_CONN)) {
2632 for (i = 0; i < count; i++, val = (char *)val + incr) {
2634 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2636 tgt = lov->lov_tgts[i];
2638 /* OST was disconnected */
2639 if (!tgt || !tgt->ltd_exp)
2642 /* OST is inactive and we don't want inactive OSCs */
2643 if (!tgt->ltd_active && !do_inactive)
2647 struct mds_group_info *mgi;
2649 LASSERT(vallen == sizeof(*mgi));
2650 mgi = (struct mds_group_info *)val;
2652 /* Only want a specific OSC */
2653 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2657 err = obd_set_info_async(tgt->ltd_exp,
2658 keylen, key, sizeof(int),
2660 } else if (next_id) {
2661 err = obd_set_info_async(tgt->ltd_exp,
2662 keylen, key, vallen,
2663 ((struct obd_id_info*)val)->data, set);
2665 /* Only want a specific OSC */
2667 !obd_uuid_equals(val, &tgt->ltd_uuid))
2670 err = obd_set_info_async(tgt->ltd_exp,
2671 keylen, key, vallen, val, set);
2680 err = ptlrpc_set_wait(set);
2683 ptlrpc_set_destroy(set);
2688 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2689 struct lov_stripe_md *lsm)
2697 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2698 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2703 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2708 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2709 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2710 if (loi->loi_ar.ar_rc && !rc)
2711 rc = loi->loi_ar.ar_rc;
2712 loi->loi_ar.ar_rc = 0;
2716 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2719 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2720 int cmd, __u64 *offset)
2722 __u32 ssize = lsm->lsm_stripe_size;
2726 do_div(start, ssize);
2727 start = start * ssize;
2729 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2730 ", end "LPU64"\n", *offset, ssize, start,
2732 if (cmd == OBD_CALC_STRIPE_END) {
2733 *offset = start + ssize - 1;
2734 } else if (cmd == OBD_CALC_STRIPE_START) {
2745 struct lov_multi_wait {
2746 struct ldlm_lock *lock;
2752 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2753 struct lustre_handle *lockh)
2755 struct lov_lock_handles *lov_lockh = NULL;
2756 struct lustre_handle *lov_lockhp;
2757 struct lov_obd *lov;
2758 struct lov_oinfo *loi;
2759 struct lov_multi_wait *queues;
2763 ASSERT_LSM_MAGIC(lsm);
2765 if (!exp || !exp->exp_obd)
2768 LASSERT(lockh != NULL);
2769 if (lsm->lsm_stripe_count > 1) {
2770 lov_lockh = lov_handle2llh(lockh);
2771 if (lov_lockh == NULL) {
2772 CERROR("LOV: invalid lov lock handle %p\n", lockh);
2776 lov_lockhp = lov_lockh->llh_handles;
2781 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2783 GOTO(out, rc = -ENOMEM);
2785 lov = &exp->exp_obd->u.lov;
2786 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2787 i++, loi++, lov_lockhp++) {
2788 struct ldlm_lock *lock;
2789 struct obd_device *obd;
2791 lock = ldlm_handle2lock(lov_lockhp);
2793 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2794 loi->loi_ost_idx, loi->loi_id);
2795 queues[i].completed = 1;
2799 queues[i].lock = lock;
2800 init_waitqueue_entry(&(queues[i].wait), current);
2801 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2803 obd = class_exp2obd(lock->l_conn_export);
2805 imp = obd->u.cli.cl_import;
2807 spin_lock(&imp->imp_lock);
2808 queues[i].generation = imp->imp_generation;
2809 spin_unlock(&imp->imp_lock);
2813 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2814 interrupted_completion_wait, &lwd);
2815 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2817 for (i = 0; i < lsm->lsm_stripe_count; i++)
2818 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2820 if (rc == -EINTR || rc == -ETIMEDOUT) {
2826 if (lov_lockh != NULL)
2827 lov_llh_put(lov_lockh);
2832 void lov_stripe_lock(struct lov_stripe_md *md)
2834 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2835 spin_lock(&md->lsm_lock);
2836 LASSERT(md->lsm_lock_owner == 0);
2837 md->lsm_lock_owner = cfs_curproc_pid();
2839 EXPORT_SYMBOL(lov_stripe_lock);
2841 void lov_stripe_unlock(struct lov_stripe_md *md)
2843 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2844 md->lsm_lock_owner = 0;
2845 spin_unlock(&md->lsm_lock);
2847 EXPORT_SYMBOL(lov_stripe_unlock);
2850 * Checks if requested extent lock is compatible with a lock under the page.
2852 * Checks if the lock under \a page is compatible with a read or write lock
2853 * (specified by \a rw) for an extent [\a start , \a end].
2855 * \param exp lov export
2856 * \param lsm striping information for the file
2857 * \param res lov_async_page placeholder
2858 * \param rw OBD_BRW_READ if requested for reading,
2859 * OBD_BRW_WRITE if requested for writing
2860 * \param start start of the requested extent
2861 * \param end end of the requested extent
2862 * \param cookie transparent parameter for passing locking context
2864 * \post result == 1, *cookie == context, appropriate lock is referenced or
2867 * \retval 1 owned lock is reused for the request
2868 * \retval 0 no lock reused for the request
2870 * \see lov_release_short_lock
2872 static int lov_reget_short_lock(struct obd_export *exp,
2873 struct lov_stripe_md *lsm,
2875 obd_off start, obd_off end,
2878 struct lov_async_page *l = *res;
2879 obd_off stripe_start, stripe_end = start;
2883 /* ensure we don't cross stripe boundaries */
2884 lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
2885 if (stripe_end <= end)
2888 /* map the region limits to the object limits */
2889 lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
2890 lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
2892 RETURN(obd_reget_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
2893 lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
2894 ltd_exp, NULL, &l->lap_sub_cookie,
2895 rw, stripe_start, stripe_end, cookie));
2899 * Releases a reference to a lock taken in a "fast" way.
2901 * Releases a read or a write (specified by \a rw) lock
2902 * referenced by \a cookie.
2904 * \param exp lov export
2905 * \param lsm striping information for the file
2906 * \param end end of the locked extent
2907 * \param rw OBD_BRW_READ if requested for reading,
2908 * OBD_BRW_WRITE if requested for writing
2909 * \param cookie transparent parameter for passing locking context
2911 * \post appropriate lock is dereferenced
2913 * \see lov_reget_short_lock
2915 static int lov_release_short_lock(struct obd_export *exp,
2916 struct lov_stripe_md *lsm, obd_off end,
2917 void *cookie, int rw)
2923 stripe = lov_stripe_number(lsm, end);
2925 RETURN(obd_release_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
2926 lsm_oinfo[stripe]->loi_ost_idx]->
2927 ltd_exp, NULL, end, cookie, rw));
2930 struct obd_ops lov_obd_ops = {
2931 .o_owner = THIS_MODULE,
2932 .o_setup = lov_setup,
2933 .o_precleanup = lov_precleanup,
2934 .o_cleanup = lov_cleanup,
2935 .o_process_config = lov_process_config,
2936 .o_connect = lov_connect,
2937 .o_disconnect = lov_disconnect,
2938 .o_statfs = lov_statfs,
2939 .o_statfs_async = lov_statfs_async,
2940 .o_packmd = lov_packmd,
2941 .o_unpackmd = lov_unpackmd,
2942 .o_checkmd = lov_checkmd,
2943 .o_create = lov_create,
2944 .o_destroy = lov_destroy,
2945 .o_getattr = lov_getattr,
2946 .o_getattr_async = lov_getattr_async,
2947 .o_setattr = lov_setattr,
2948 .o_setattr_async = lov_setattr_async,
2950 .o_brw_async = lov_brw_async,
2951 .o_prep_async_page = lov_prep_async_page,
2952 .o_reget_short_lock = lov_reget_short_lock,
2953 .o_release_short_lock = lov_release_short_lock,
2954 .o_queue_async_io = lov_queue_async_io,
2955 .o_set_async_flags = lov_set_async_flags,
2956 .o_queue_group_io = lov_queue_group_io,
2957 .o_trigger_group_io = lov_trigger_group_io,
2958 .o_teardown_async_page = lov_teardown_async_page,
2959 .o_merge_lvb = lov_merge_lvb,
2960 .o_adjust_kms = lov_adjust_kms,
2961 .o_punch = lov_punch,
2963 .o_enqueue = lov_enqueue,
2964 .o_match = lov_match,
2965 .o_change_cbdata = lov_change_cbdata,
2966 .o_cancel = lov_cancel,
2967 .o_cancel_unused = lov_cancel_unused,
2968 .o_join_lru = lov_join_lru,
2969 .o_iocontrol = lov_iocontrol,
2970 .o_get_info = lov_get_info,
2971 .o_set_info_async = lov_set_info_async,
2972 .o_extent_calc = lov_extent_calc,
2973 .o_llog_init = lov_llog_init,
2974 .o_llog_finish = lov_llog_finish,
2975 .o_notify = lov_notify,
2976 .o_register_page_removal_cb = lov_register_page_removal_cb,
2977 .o_unregister_page_removal_cb = lov_unregister_page_removal_cb,
2978 .o_register_lock_cancel_cb = lov_register_lock_cancel_cb,
2979 .o_unregister_lock_cancel_cb = lov_unregister_lock_cancel_cb,
2982 static quota_interface_t *quota_interface;
2983 extern quota_interface_t lov_quota_interface;
2985 cfs_mem_cache_t *lov_oinfo_slab;
2987 int __init lov_init(void)
2989 struct lprocfs_static_vars lvars = { 0 };
2993 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2994 sizeof(struct lov_oinfo),
2995 0, SLAB_HWCACHE_ALIGN);
2996 if (lov_oinfo_slab == NULL)
2998 lprocfs_lov_init_vars(&lvars);
3000 request_module("lquota");
3001 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3002 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3004 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3005 LUSTRE_LOV_NAME, NULL);
3007 if (quota_interface)
3008 PORTAL_SYMBOL_PUT(lov_quota_interface);
3009 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3017 static void /*__exit*/ lov_exit(void)
3021 if (quota_interface)
3022 PORTAL_SYMBOL_PUT(lov_quota_interface);
3024 class_unregister_type(LUSTRE_LOV_NAME);
3025 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3029 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3030 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3031 MODULE_LICENSE("GPL");
3033 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);