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 * - any other is lov index
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 (index = 0; index < lov->desc.ld_tgt_count; index++) {
504 tgt = lov->lov_tgts[index];
505 if (!tgt || !tgt->ltd_exp)
508 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
509 index, obd_uuid2str(&tgt->ltd_uuid),
510 tgt->ltd_exp->exp_handle.h_cookie);
511 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
515 if (index == lov->desc.ld_tgt_count)
516 GOTO(out, index = -EINVAL);
518 if (lov->lov_tgts[index]->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[index]->ltd_active = activate;
530 lov->desc.ld_active_tgt_count++;
531 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
533 lov->desc.ld_active_tgt_count--;
534 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
536 /* remove any old qos penalty */
537 lov->lov_tgts[index]->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);
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 /* sync event should be send lov idx as data */
582 struct lov_obd *lov = &obd->u.lov;
583 struct obd_device *tgt_obd;
586 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
587 if (!lov->lov_tgts[i])
590 if ((ev == OBD_NOTIFY_SYNC) ||
591 (ev == OBD_NOTIFY_SYNC_NONBLOCK))
594 tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
595 rc = obd_notify_observer(obd, tgt_obd, ev, data);
597 CERROR("%s: notify %s of %s failed %d\n",
599 obd->obd_observer->obd_name,
600 tgt_obd->obd_name, rc);
610 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
611 __u32 index, int gen, int active)
613 struct lov_obd *lov = &obd->u.lov;
614 struct lov_tgt_desc *tgt;
618 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
619 uuidp->uuid, index, gen, active);
622 CERROR("request to add OBD %s with invalid generation: %d\n",
627 mutex_down(&lov->lov_lock);
629 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
630 tgt = lov->lov_tgts[index];
631 CERROR("UUID %s already assigned at LOV target index %d\n",
632 obd_uuid2str(&tgt->ltd_uuid), index);
633 mutex_up(&lov->lov_lock);
637 if (index >= lov->lov_tgt_size) {
638 /* We need to reallocate the lov target array. */
639 struct lov_tgt_desc **newtgts, **old = NULL;
640 __u32 newsize, oldsize = 0;
642 newsize = max(lov->lov_tgt_size, (__u32)2);
643 while (newsize < index + 1)
644 newsize = newsize << 1;
645 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
646 if (newtgts == NULL) {
647 mutex_up(&lov->lov_lock);
651 if (lov->lov_tgt_size) {
652 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
655 oldsize = lov->lov_tgt_size;
658 lov->lov_tgts = newtgts;
659 lov->lov_tgt_size = newsize;
664 OBD_FREE(old, sizeof(*old) * oldsize);
666 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
667 lov->lov_tgts, lov->lov_tgt_size);
672 mutex_up(&lov->lov_lock);
676 memset(tgt, 0, sizeof(*tgt));
677 tgt->ltd_uuid = *uuidp;
678 /* XXX - add a sanity check on the generation number. */
680 tgt->ltd_index = index;
681 tgt->ltd_activate = active;
682 lov->lov_tgts[index] = tgt;
683 if (index >= lov->desc.ld_tgt_count)
684 lov->desc.ld_tgt_count = index + 1;
686 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
690 mutex_up(&lov->lov_lock);
692 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
693 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
695 if (lov->lov_connects == 0) {
696 /* lov_connect hasn't been called yet. We'll do the
697 lov_connect_obd on this target when that fn first runs,
698 because we don't know the connect flags yet. */
704 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
708 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
709 active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
714 CERROR("add failed (%d), deleting %s\n", rc,
715 obd_uuid2str(&tgt->ltd_uuid));
716 lov_del_target(obd, index, 0, 0);
722 /* Schedule a target for deletion */
723 static int lov_del_target(struct obd_device *obd, __u32 index,
724 struct obd_uuid *uuidp, int gen)
726 struct lov_obd *lov = &obd->u.lov;
727 int count = lov->desc.ld_tgt_count;
731 if (index >= count) {
732 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
739 if (!lov->lov_tgts[index]) {
740 CERROR("LOV target at index %d is not setup.\n", index);
741 GOTO(out, rc = -EINVAL);
744 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
745 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
746 lov_uuid2str(lov, index), index,
747 obd_uuid2str(uuidp));
748 GOTO(out, rc = -EINVAL);
751 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
752 lov_uuid2str(lov, index), index,
753 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
754 lov->lov_tgts[index]->ltd_active);
756 lov->lov_tgts[index]->ltd_reap = 1;
757 lov->lov_death_row++;
758 /* we really delete it from lov_putref */
765 /* We are holding lov_lock */
766 static void __lov_del_obd(struct obd_device *obd, __u32 index)
768 struct lov_obd *lov = &obd->u.lov;
769 struct obd_device *osc_obd;
770 struct lov_tgt_desc *tgt = lov->lov_tgts[index];
773 LASSERT(tgt->ltd_reap);
775 osc_obd = class_exp2obd(tgt->ltd_exp);
777 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
778 lov_uuid2str(lov, index),
779 osc_obd ? osc_obd->obd_name : "<no obd>");
782 lov_disconnect_obd(obd, index);
784 /* XXX - right now there is a dependency on ld_tgt_count being the
785 * maximum tgt index for computing the mds_max_easize. So we can't
788 lov_ost_pool_remove(&lov->lov_packed, index);
789 lov->lov_tgts[index] = NULL;
792 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
793 do it ourselves. And we can't do it from lov_cleanup,
794 because we just lost our only reference to it. */
796 class_manual_cleanup(osc_obd);
799 void lov_fix_desc_stripe_size(__u64 *val)
801 if (*val < PTLRPC_MAX_BRW_SIZE) {
802 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
803 PTLRPC_MAX_BRW_SIZE);
804 *val = PTLRPC_MAX_BRW_SIZE;
805 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
806 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
807 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
809 *val, LOV_MIN_STRIPE_SIZE);
813 void lov_fix_desc_stripe_count(__u32 *val)
819 void lov_fix_desc_pattern(__u32 *val)
821 /* from lov_setstripe */
822 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
823 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
828 void lov_fix_desc_qos_maxage(__u32 *val)
832 *val = QOS_DEFAULT_MAXAGE;
835 void lov_fix_desc(struct lov_desc *desc)
837 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
838 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
839 lov_fix_desc_pattern(&desc->ld_pattern);
840 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
843 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
845 struct lprocfs_static_vars lvars = { 0 };
846 struct lov_desc *desc;
847 struct lov_obd *lov = &obd->u.lov;
852 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
853 CERROR("LOV setup requires a descriptor\n");
857 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
859 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
860 CERROR("descriptor size wrong: %d > %d\n",
861 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
865 if (desc->ld_magic != LOV_DESC_MAGIC) {
866 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
867 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
868 obd->obd_name, desc);
869 lustre_swab_lov_desc(desc);
871 CERROR("%s: Bad lov desc magic: %#x\n",
872 obd->obd_name, desc->ld_magic);
879 /* Because of 64-bit divide/mod operations only work with a 32-bit
880 * divisor in a 32-bit kernel, we cannot support a stripe width
881 * of 4GB or larger on 32-bit CPUs. */
882 count = desc->ld_default_stripe_count;
883 if ((count > 0 ? count : desc->ld_tgt_count) *
884 desc->ld_default_stripe_size > 0xffffffff) {
885 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
886 desc->ld_default_stripe_size, count);
890 desc->ld_active_tgt_count = 0;
892 lov->lov_tgt_size = 0;
894 sema_init(&lov->lov_lock, 1);
895 atomic_set(&lov->lov_refcount, 0);
896 CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
897 init_rwsem(&lov->lov_qos.lq_rw_sem);
898 lov->lov_qos.lq_dirty = 1;
899 lov->lov_qos.lq_rr.lqr_dirty = 1;
900 lov->lov_qos.lq_reset = 1;
901 /* Default priority is toward free space balance */
902 lov->lov_qos.lq_prio_free = 232;
904 lov->lov_pools_hash_body = lustre_hash_init("POOLS", 128, 128,
905 &pool_hash_operations, 0);
906 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
907 lov->lov_pool_count = 0;
908 rc = lov_ost_pool_init(&lov->lov_packed, 0);
911 rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
913 lov_ost_pool_free(&lov->lov_packed);
917 lprocfs_lov_init_vars(&lvars);
918 lprocfs_obd_setup(obd, lvars.obd_vars);
923 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
924 0444, &lov_proc_target_fops, obd);
926 CWARN("Error adding the target_obd file\n");
929 lov->lov_pool_proc_entry = lprocfs_register("pools",
936 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
942 case OBD_CLEANUP_EARLY: {
943 struct lov_obd *lov = &obd->u.lov;
945 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
946 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
948 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
953 case OBD_CLEANUP_EXPORTS:
954 rc = obd_llog_finish(obd, 0);
956 CERROR("failed to cleanup llogging subsystems\n");
962 static int lov_cleanup(struct obd_device *obd)
964 struct lov_obd *lov = &obd->u.lov;
965 struct list_head *pos, *tmp;
966 struct pool_desc *pool;
968 lprocfs_obd_cleanup(obd);
970 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
971 pool = list_entry(pos, struct pool_desc, pool_list);
972 list_del(&pool->pool_list);
973 lov_ost_pool_free(&(pool->pool_rr.lqr_pool));
974 lov_ost_pool_free(&(pool->pool_obds));
977 lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
978 lov_ost_pool_free(&lov->lov_packed);
979 lustre_hash_exit(lov->lov_pools_hash_body);
983 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
984 if (!lov->lov_tgts[i])
987 /* Inactive targets may never have connected */
988 if (lov->lov_tgts[i]->ltd_active ||
989 atomic_read(&lov->lov_refcount))
990 /* We should never get here - these
991 should have been removed in the
993 CERROR("lov tgt %d not cleaned!"
994 " deathrow=%d, lovrc=%d\n",
995 i, lov->lov_death_row,
996 atomic_read(&lov->lov_refcount));
997 lov_del_target(obd, i, 0, 0);
999 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
1001 lov->lov_tgt_size = 0;
1007 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
1009 struct lustre_cfg *lcfg = buf;
1010 struct obd_uuid obd_uuid;
1015 switch(cmd = lcfg->lcfg_command) {
1016 case LCFG_LOV_ADD_OBD:
1017 case LCFG_LOV_ADD_INA:
1018 case LCFG_LOV_DEL_OBD: {
1021 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
1022 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1023 GOTO(out, rc = -EINVAL);
1025 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1027 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1028 GOTO(out, rc = -EINVAL);
1029 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1030 GOTO(out, rc = -EINVAL);
1031 if (cmd == LCFG_LOV_ADD_OBD)
1032 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
1033 else if (cmd == LCFG_LOV_ADD_INA)
1034 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
1036 rc = lov_del_target(obd, index, &obd_uuid, gen);
1040 struct lprocfs_static_vars lvars = { 0 };
1041 struct lov_desc *desc = &(obd->u.lov.desc);
1044 GOTO(out, rc = -EINVAL);
1046 lprocfs_lov_init_vars(&lvars);
1048 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1059 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1060 GOTO(out, rc = -EINVAL);
1069 #define log2(n) ffz(~(n))
1072 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1073 struct lov_stripe_md **ea,
1074 struct obd_trans_info *oti)
1076 struct lov_obd *lov;
1077 struct obdo *tmp_oa;
1078 struct obd_uuid *ost_uuid = NULL;
1082 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1083 src_oa->o_flags == OBD_FL_DELORPHAN);
1085 lov = &export->exp_obd->u.lov;
1091 if (oti->oti_ost_uuid) {
1092 ost_uuid = oti->oti_ost_uuid;
1093 CDEBUG(D_HA, "clearing orphans only for %s\n",
1097 lov_getref(export->exp_obd);
1098 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1099 struct lov_stripe_md obj_md;
1100 struct lov_stripe_md *obj_mdp = &obj_md;
1101 struct lov_tgt_desc *tgt;
1104 tgt = lov->lov_tgts[i];
1108 /* if called for a specific target, we don't
1109 care if it is not active. */
1110 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1111 CDEBUG(D_HA, "lov idx %d inactive\n", i);
1115 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1118 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1119 obd_uuid2str(ost_uuid));
1121 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1123 LASSERT(lov->lov_tgts[i]->ltd_exp);
1124 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1125 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1126 tmp_oa, &obj_mdp, oti);
1128 /* This export will be disabled until it is recovered,
1129 and then orphan recovery will be completed. */
1130 CERROR("error in orphan recovery on OST idx %d/%d: "
1131 "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1138 lov_putref(export->exp_obd);
1144 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1145 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1147 struct lov_stripe_md *obj_mdp, *lsm;
1148 struct lov_obd *lov = &exp->exp_obd->u.lov;
1153 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1154 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1156 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1157 if (obj_mdp == NULL)
1160 ost_idx = src_oa->o_nlink;
1163 GOTO(out, rc = -EINVAL);
1164 if (ost_idx >= lov->desc.ld_tgt_count ||
1165 !lov->lov_tgts[ost_idx])
1166 GOTO(out, rc = -EINVAL);
1168 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1169 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1170 if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1171 GOTO(out, rc = -EINVAL);
1175 if (i == lsm->lsm_stripe_count)
1176 GOTO(out, rc = -EINVAL);
1178 rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1180 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1184 /* the LOV expects oa->o_id to be set to the LOV object id */
1185 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1186 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1188 struct lov_obd *lov;
1189 struct obd_info oinfo;
1190 struct lov_request_set *set = NULL;
1191 struct lov_request *req;
1192 struct obd_statfs osfs;
1197 LASSERT(ea != NULL);
1201 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1202 src_oa->o_flags == OBD_FL_DELORPHAN) {
1203 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1207 lov = &exp->exp_obd->u.lov;
1208 if (!lov->desc.ld_active_tgt_count)
1211 lov_getref(exp->exp_obd);
1212 /* Recreate a specific object id at the given OST index */
1213 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1214 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1215 rc = lov_recreate(exp, src_oa, ea, oti);
1219 maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1220 obd_statfs_rqset(exp->exp_obd, &osfs, maxage, OBD_STATFS_NODELAY);
1222 rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1226 list_for_each_entry(req, &set->set_list, rq_link) {
1227 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1228 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1229 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1230 lov_update_create_set(set, req, rc);
1232 rc = lov_fini_create_set(set, ea);
1234 lov_putref(exp->exp_obd);
1238 #define ASSERT_LSM_MAGIC(lsmp) \
1240 LASSERT((lsmp) != NULL); \
1241 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1242 (lsmp)->lsm_magic == LOV_MAGIC_V3 || \
1243 (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n", \
1244 (lsmp), (lsmp)->lsm_magic); \
1247 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1248 struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1249 struct obd_export *md_exp)
1251 struct lov_request_set *set;
1252 struct obd_info oinfo;
1253 struct lov_request *req;
1254 struct list_head *pos;
1255 struct lov_obd *lov;
1259 ASSERT_LSM_MAGIC(lsm);
1261 if (!exp || !exp->exp_obd)
1264 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1266 LASSERT(oti->oti_logcookies);
1269 lov = &exp->exp_obd->u.lov;
1270 lov_getref(exp->exp_obd);
1271 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1275 list_for_each (pos, &set->set_list) {
1277 req = list_entry(pos, struct lov_request, rq_link);
1279 if (oa->o_valid & OBD_MD_FLCOOKIE)
1280 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1282 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1283 req->rq_oi.oi_oa, NULL, oti, NULL);
1284 err = lov_update_common_set(set, req, err);
1286 CERROR("error: destroying objid "LPX64" subobj "
1287 LPX64" on OST idx %d: rc = %d\n",
1288 oa->o_id, req->rq_oi.oi_oa->o_id,
1296 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1297 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1299 err = lov_fini_destroy_set(set);
1301 lov_putref(exp->exp_obd);
1302 RETURN(rc ? rc : err);
1305 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1307 struct lov_request_set *set;
1308 struct lov_request *req;
1309 struct list_head *pos;
1310 struct lov_obd *lov;
1311 int err = 0, rc = 0;
1315 ASSERT_LSM_MAGIC(oinfo->oi_md);
1317 if (!exp || !exp->exp_obd)
1320 lov = &exp->exp_obd->u.lov;
1322 rc = lov_prep_getattr_set(exp, oinfo, &set);
1326 list_for_each (pos, &set->set_list) {
1327 req = list_entry(pos, struct lov_request, rq_link);
1329 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1330 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1331 req->rq_oi.oi_oa->o_id, req->rq_idx);
1333 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1335 err = lov_update_common_set(set, req, rc);
1337 CERROR("error: getattr objid "LPX64" subobj "
1338 LPX64" on OST idx %d: rc = %d\n",
1339 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1345 rc = lov_fini_getattr_set(set);
1351 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1354 struct lov_request_set *lovset = (struct lov_request_set *)data;
1358 /* don't do attribute merge if this aysnc op failed */
1360 lovset->set_completes = 0;
1361 err = lov_fini_getattr_set(lovset);
1362 RETURN(rc ? rc : err);
1365 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1366 struct ptlrpc_request_set *rqset)
1368 struct lov_request_set *lovset;
1369 struct lov_obd *lov;
1370 struct list_head *pos;
1371 struct lov_request *req;
1376 ASSERT_LSM_MAGIC(oinfo->oi_md);
1378 if (!exp || !exp->exp_obd)
1381 lov = &exp->exp_obd->u.lov;
1383 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1387 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1388 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1389 oinfo->oi_md->lsm_stripe_size);
1391 list_for_each (pos, &lovset->set_list) {
1392 req = list_entry(pos, struct lov_request, rq_link);
1394 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1395 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1396 req->rq_oi.oi_oa->o_id, req->rq_idx);
1397 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1398 &req->rq_oi, rqset);
1400 CERROR("error: getattr objid "LPX64" subobj "
1401 LPX64" on OST idx %d: rc = %d\n",
1402 oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1408 if (!list_empty(&rqset->set_requests)) {
1410 LASSERT (rqset->set_interpret == NULL);
1411 rqset->set_interpret = lov_getattr_interpret;
1412 rqset->set_arg = (void *)lovset;
1417 lovset->set_completes = 0;
1418 err = lov_fini_getattr_set(lovset);
1419 RETURN(rc ? rc : err);
1422 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1423 struct obd_trans_info *oti)
1425 struct lov_request_set *set;
1426 struct lov_obd *lov;
1427 struct list_head *pos;
1428 struct lov_request *req;
1429 int err = 0, rc = 0;
1433 ASSERT_LSM_MAGIC(oinfo->oi_md);
1435 if (!exp || !exp->exp_obd)
1438 /* for now, we only expect the following updates here */
1439 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1440 OBD_MD_FLMODE | OBD_MD_FLATIME |
1441 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1442 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1443 OBD_MD_FLGROUP | OBD_MD_FLUID |
1444 OBD_MD_FLGID | OBD_MD_FLFID |
1446 lov = &exp->exp_obd->u.lov;
1447 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1451 list_for_each (pos, &set->set_list) {
1452 req = list_entry(pos, struct lov_request, rq_link);
1454 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1456 err = lov_update_setattr_set(set, req, rc);
1458 CERROR("error: setattr objid "LPX64" subobj "
1459 LPX64" on OST idx %d: rc = %d\n",
1460 set->set_oi->oi_oa->o_id,
1461 req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1466 err = lov_fini_setattr_set(set);
1472 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1475 struct lov_request_set *lovset = (struct lov_request_set *)data;
1480 lovset->set_completes = 0;
1481 err = lov_fini_setattr_set(lovset);
1482 RETURN(rc ? rc : err);
1485 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1486 needed. Otherwise, a client is waiting for responses. */
1487 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1488 struct obd_trans_info *oti,
1489 struct ptlrpc_request_set *rqset)
1491 struct lov_request_set *set;
1492 struct lov_request *req;
1493 struct list_head *pos;
1494 struct lov_obd *lov;
1499 ASSERT_LSM_MAGIC(oinfo->oi_md);
1500 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1502 LASSERT(oti->oti_logcookies);
1505 if (!exp || !exp->exp_obd)
1508 lov = &exp->exp_obd->u.lov;
1509 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1513 CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1514 oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1515 oinfo->oi_md->lsm_stripe_size);
1517 list_for_each (pos, &set->set_list) {
1518 req = list_entry(pos, struct lov_request, rq_link);
1520 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1521 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1523 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1524 "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1525 req->rq_oi.oi_oa->o_id, req->rq_idx);
1527 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1528 &req->rq_oi, oti, rqset);
1530 CERROR("error: setattr objid "LPX64" subobj "
1531 LPX64" on OST idx %d: rc = %d\n",
1532 set->set_oi->oi_oa->o_id,
1533 req->rq_oi.oi_oa->o_id,
1539 /* If we are not waiting for responses on async requests, return. */
1540 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1543 set->set_completes = 0;
1544 err = lov_fini_setattr_set(set);
1545 RETURN(rc ? rc : err);
1548 LASSERT(rqset->set_interpret == NULL);
1549 rqset->set_interpret = lov_setattr_interpret;
1550 rqset->set_arg = (void *)set;
1555 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1558 struct lov_request_set *lovset = (struct lov_request_set *)data;
1563 lovset->set_completes = 0;
1564 err = lov_fini_punch_set(lovset);
1565 RETURN(rc ? rc : err);
1568 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1569 * we can send this 'punch' to just the authoritative node and the nodes
1570 * that the punch will affect. */
1571 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1572 struct obd_trans_info *oti,
1573 struct ptlrpc_request_set *rqset)
1575 struct lov_request_set *set;
1576 struct lov_obd *lov;
1577 struct list_head *pos;
1578 struct lov_request *req;
1583 ASSERT_LSM_MAGIC(oinfo->oi_md);
1585 if (!exp || !exp->exp_obd)
1588 lov = &exp->exp_obd->u.lov;
1589 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1593 list_for_each (pos, &set->set_list) {
1594 req = list_entry(pos, struct lov_request, rq_link);
1596 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1597 &req->rq_oi, NULL, rqset);
1599 CERROR("error: punch objid "LPX64" subobj "LPX64
1600 " on OST idx %d: rc = %d\n",
1601 set->set_oi->oi_oa->o_id,
1602 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1607 if (rc || list_empty(&rqset->set_requests)) {
1609 err = lov_fini_punch_set(set);
1610 RETURN(rc ? rc : err);
1613 LASSERT(rqset->set_interpret == NULL);
1614 rqset->set_interpret = lov_punch_interpret;
1615 rqset->set_arg = (void *)set;
1620 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1621 struct lov_stripe_md *lsm, obd_off start, obd_off end,
1624 struct lov_request_set *set;
1625 struct obd_info oinfo;
1626 struct lov_obd *lov;
1627 struct list_head *pos;
1628 struct lov_request *req;
1629 int err = 0, rc = 0;
1632 ASSERT_LSM_MAGIC(lsm);
1637 lov = &exp->exp_obd->u.lov;
1638 rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1642 list_for_each (pos, &set->set_list) {
1643 req = list_entry(pos, struct lov_request, rq_link);
1645 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1646 req->rq_oi.oi_oa, NULL,
1647 req->rq_oi.oi_policy.l_extent.start,
1648 req->rq_oi.oi_policy.l_extent.end, capa);
1649 err = lov_update_common_set(set, req, rc);
1651 CERROR("error: fsync objid "LPX64" subobj "LPX64
1652 " on OST idx %d: rc = %d\n",
1653 set->set_oi->oi_oa->o_id,
1654 req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1659 err = lov_fini_sync_set(set);
1665 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1666 obd_count oa_bufs, struct brw_page *pga)
1668 struct obd_info oinfo = { { { 0 } } };
1671 oinfo.oi_oa = lov_oinfo->oi_oa;
1673 /* The caller just wants to know if there's a chance that this
1674 * I/O can succeed */
1675 for (i = 0; i < oa_bufs; i++) {
1676 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1677 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1680 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1681 pga[i].off + pga[i].count,
1685 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1686 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1690 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1698 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1699 obd_count oa_bufs, struct brw_page *pga,
1700 struct obd_trans_info *oti)
1702 struct lov_request_set *set;
1703 struct lov_request *req;
1704 struct list_head *pos;
1705 struct lov_obd *lov = &exp->exp_obd->u.lov;
1709 ASSERT_LSM_MAGIC(oinfo->oi_md);
1711 if (cmd == OBD_BRW_CHECK) {
1712 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1716 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1720 list_for_each (pos, &set->set_list) {
1721 struct obd_export *sub_exp;
1722 struct brw_page *sub_pga;
1723 req = list_entry(pos, struct lov_request, rq_link);
1725 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1726 sub_pga = set->set_pga + req->rq_pgaidx;
1727 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1731 lov_update_common_set(set, req, rc);
1734 err = lov_fini_brw_set(set);
1740 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1743 struct lov_request_set *lovset = (struct lov_request_set *)data;
1747 lovset->set_completes = 0;
1748 lov_fini_brw_set(lovset);
1750 rc = lov_fini_brw_set(lovset);
1756 static int lov_brw_async(int cmd, struct obd_export *exp,
1757 struct obd_info *oinfo, obd_count oa_bufs,
1758 struct brw_page *pga, struct obd_trans_info *oti,
1759 struct ptlrpc_request_set *set)
1761 struct lov_request_set *lovset;
1762 struct lov_request *req;
1763 struct list_head *pos;
1764 struct lov_obd *lov = &exp->exp_obd->u.lov;
1769 ASSERT_LSM_MAGIC(oinfo->oi_md);
1771 if (cmd == OBD_BRW_CHECK) {
1772 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1776 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1780 list_for_each (pos, &lovset->set_list) {
1781 struct obd_export *sub_exp;
1782 struct brw_page *sub_pga;
1783 req = list_entry(pos, struct lov_request, rq_link);
1785 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1786 sub_pga = lovset->set_pga + req->rq_pgaidx;
1787 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1791 lov_update_common_set(lovset, req, rc);
1794 LASSERT(set->set_interpret == NULL);
1795 LASSERT(set->set_arg == NULL);
1796 rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1802 lov_fini_brw_set(lovset);
1806 static int lov_ap_make_ready(void *data, int cmd)
1808 struct lov_async_page *lap = lap_from_cookie(data);
1810 return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1813 static int lov_ap_refresh_count(void *data, int cmd)
1815 struct lov_async_page *lap = lap_from_cookie(data);
1817 return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1821 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1823 struct lov_async_page *lap = lap_from_cookie(data);
1825 lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1826 /* XXX woah, shouldn't we be altering more here? size? */
1827 oa->o_id = lap->lap_loi_id;
1828 oa->o_gr = lap->lap_loi_gr;
1829 oa->o_valid |= OBD_MD_FLGROUP;
1830 oa->o_stripe_idx = lap->lap_stripe;
1833 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1836 struct lov_async_page *lap = lap_from_cookie(data);
1838 lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1841 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1843 struct lov_async_page *lap = lap_from_cookie(data);
1845 /* in a raid1 regime this would down a count of many ios
1846 * in flight, onl calling the caller_ops completion when all
1847 * the raid1 ios are complete */
1848 rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1852 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1854 struct lov_async_page *lap = lap_from_cookie(data);
1855 return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1858 static struct obd_async_page_ops lov_async_page_ops = {
1859 .ap_make_ready = lov_ap_make_ready,
1860 .ap_refresh_count = lov_ap_refresh_count,
1861 .ap_fill_obdo = lov_ap_fill_obdo,
1862 .ap_update_obdo = lov_ap_update_obdo,
1863 .ap_completion = lov_ap_completion,
1864 .ap_lookup_capa = lov_ap_lookup_capa,
1867 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1868 struct lov_oinfo *loi, cfs_page_t *page,
1869 obd_off offset, struct obd_async_page_ops *ops,
1870 void *data, void **res, int nocache,
1871 struct lustre_handle *lockh)
1873 struct lov_obd *lov = &exp->exp_obd->u.lov;
1874 struct lov_async_page *lap;
1875 struct lov_lock_handles *lov_lockh = NULL;
1881 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1882 Only because of this layering limitation will a client
1883 mount with no osts fail */
1884 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1885 !lov->lov_tgts[i]->ltd_exp) {
1887 if (i >= lov->desc.ld_tgt_count)
1890 rc = size_round(sizeof(*lap)) +
1891 obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1892 NULL, NULL, 0, NULL, NULL, NULL, 0,
1896 ASSERT_LSM_MAGIC(lsm);
1897 LASSERT(loi == NULL);
1900 lap->lap_magic = LOV_AP_MAGIC;
1901 lap->lap_caller_ops = ops;
1902 lap->lap_caller_data = data;
1904 /* for now only raid 0 which passes through */
1905 lap->lap_stripe = lov_stripe_number(lsm, offset);
1906 lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1907 loi = lsm->lsm_oinfo[lap->lap_stripe];
1909 /* so the callback doesn't need the lsm */
1910 lap->lap_loi_id = loi->loi_id;
1911 lap->lap_loi_gr = lsm->lsm_object_gr;
1912 LASSERT(lsm->lsm_object_gr > 0);
1914 lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1917 lov_lockh = lov_handle2llh(lockh);
1919 lockh = lov_lockh->llh_handles + lap->lap_stripe;
1923 rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1924 lsm, loi, page, lap->lap_sub_offset,
1925 &lov_async_page_ops, lap,
1926 &lap->lap_sub_cookie, nocache, lockh);
1928 lov_llh_put(lov_lockh);
1931 CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1932 lap->lap_sub_cookie, offset);
1936 static int lov_queue_async_io(struct obd_export *exp,
1937 struct lov_stripe_md *lsm,
1938 struct lov_oinfo *loi, void *cookie,
1939 int cmd, obd_off off, int count,
1940 obd_flag brw_flags, obd_flag async_flags)
1942 struct lov_obd *lov = &exp->exp_obd->u.lov;
1943 struct lov_async_page *lap;
1946 LASSERT(loi == NULL);
1948 ASSERT_LSM_MAGIC(lsm);
1950 lap = lap_from_cookie(cookie);
1952 loi = lsm->lsm_oinfo[lap->lap_stripe];
1954 rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1955 loi, lap->lap_sub_cookie, cmd, off, count,
1956 brw_flags, async_flags);
1960 static int lov_set_async_flags(struct obd_export *exp,
1961 struct lov_stripe_md *lsm,
1962 struct lov_oinfo *loi, void *cookie,
1963 obd_flag async_flags)
1965 struct lov_obd *lov = &exp->exp_obd->u.lov;
1966 struct lov_async_page *lap;
1969 LASSERT(loi == NULL);
1971 ASSERT_LSM_MAGIC(lsm);
1973 lap = lap_from_cookie(cookie);
1975 loi = lsm->lsm_oinfo[lap->lap_stripe];
1977 rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1978 lsm, loi, lap->lap_sub_cookie, async_flags);
1982 static int lov_queue_group_io(struct obd_export *exp,
1983 struct lov_stripe_md *lsm,
1984 struct lov_oinfo *loi,
1985 struct obd_io_group *oig, void *cookie,
1986 int cmd, obd_off off, int count,
1987 obd_flag brw_flags, obd_flag async_flags)
1989 struct lov_obd *lov = &exp->exp_obd->u.lov;
1990 struct lov_async_page *lap;
1993 LASSERT(loi == NULL);
1995 ASSERT_LSM_MAGIC(lsm);
1997 lap = lap_from_cookie(cookie);
1999 loi = lsm->lsm_oinfo[lap->lap_stripe];
2001 rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
2002 loi, oig, lap->lap_sub_cookie, cmd, off, count,
2003 brw_flags, async_flags);
2007 /* this isn't exactly optimal. we may have queued sync io in oscs on
2008 * all stripes, but we don't record that fact at queue time. so we
2009 * trigger sync io on all stripes. */
2010 static int lov_trigger_group_io(struct obd_export *exp,
2011 struct lov_stripe_md *lsm,
2012 struct lov_oinfo *loi,
2013 struct obd_io_group *oig)
2015 struct lov_obd *lov = &exp->exp_obd->u.lov;
2018 LASSERT(loi == NULL);
2020 ASSERT_LSM_MAGIC(lsm);
2022 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2023 loi = lsm->lsm_oinfo[i];
2024 if (!lov->lov_tgts[loi->loi_ost_idx] ||
2025 !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2026 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2030 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2032 if (rc == 0 && err != 0)
2038 static int lov_teardown_async_page(struct obd_export *exp,
2039 struct lov_stripe_md *lsm,
2040 struct lov_oinfo *loi, void *cookie)
2042 struct lov_obd *lov = &exp->exp_obd->u.lov;
2043 struct lov_async_page *lap;
2046 LASSERT(loi == NULL);
2048 ASSERT_LSM_MAGIC(lsm);
2050 lap = lap_from_cookie(cookie);
2052 loi = lsm->lsm_oinfo[lap->lap_stripe];
2054 rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2055 lsm, loi, lap->lap_sub_cookie);
2057 CERROR("unable to teardown sub cookie %p: %d\n",
2058 lap->lap_sub_cookie, rc);
2064 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2067 struct lov_request_set *lovset = (struct lov_request_set *)data;
2069 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2073 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2074 struct ldlm_enqueue_info *einfo,
2075 struct ptlrpc_request_set *rqset)
2077 ldlm_mode_t mode = einfo->ei_mode;
2078 struct lov_request_set *set;
2079 struct lov_request *req;
2080 struct list_head *pos;
2081 struct lov_obd *lov;
2086 ASSERT_LSM_MAGIC(oinfo->oi_md);
2087 LASSERT(mode == (mode & -mode));
2089 /* we should never be asked to replay a lock this way. */
2090 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2092 if (!exp || !exp->exp_obd)
2095 lov = &exp->exp_obd->u.lov;
2096 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2100 list_for_each (pos, &set->set_list) {
2101 req = list_entry(pos, struct lov_request, rq_link);
2103 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2104 &req->rq_oi, einfo, rqset);
2109 if (rqset && !list_empty(&rqset->set_requests)) {
2111 LASSERT(rqset->set_interpret == NULL);
2112 rqset->set_interpret = lov_enqueue_interpret;
2113 rqset->set_arg = (void *)set;
2117 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2121 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2122 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2123 int *flags, void *data, struct lustre_handle *lockh)
2125 struct lov_request_set *set;
2126 struct obd_info oinfo;
2127 struct lov_request *req;
2128 struct list_head *pos;
2129 struct lov_obd *lov = &exp->exp_obd->u.lov;
2130 struct lustre_handle *lov_lockhp;
2131 int lov_flags, rc = 0;
2134 ASSERT_LSM_MAGIC(lsm);
2135 LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2137 if (!exp || !exp->exp_obd)
2140 lov = &exp->exp_obd->u.lov;
2141 rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2145 list_for_each (pos, &set->set_list) {
2146 ldlm_policy_data_t sub_policy;
2147 req = list_entry(pos, struct lov_request, rq_link);
2148 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2149 LASSERT(lov_lockhp);
2152 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2154 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2155 req->rq_oi.oi_md, type, &sub_policy,
2156 mode, &lov_flags, data, lov_lockhp);
2157 rc = lov_update_match_set(set, req, rc);
2161 lov_fini_match_set(set, mode, *flags);
2165 static int lov_change_cbdata(struct obd_export *exp,
2166 struct lov_stripe_md *lsm, ldlm_iterator_t it,
2169 struct lov_obd *lov;
2173 ASSERT_LSM_MAGIC(lsm);
2175 if (!exp || !exp->exp_obd)
2178 LASSERT(lsm->lsm_object_gr > 0);
2180 lov = &exp->exp_obd->u.lov;
2181 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2182 struct lov_stripe_md submd;
2183 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2185 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2186 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2190 submd.lsm_object_id = loi->loi_id;
2191 submd.lsm_object_gr = lsm->lsm_object_gr;
2192 submd.lsm_stripe_count = 0;
2193 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2199 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2200 __u32 mode, struct lustre_handle *lockh)
2202 struct lov_request_set *set;
2203 struct obd_info oinfo;
2204 struct lov_request *req;
2205 struct list_head *pos;
2206 struct lov_obd *lov = &exp->exp_obd->u.lov;
2207 struct lustre_handle *lov_lockhp;
2208 int err = 0, rc = 0;
2211 ASSERT_LSM_MAGIC(lsm);
2213 if (!exp || !exp->exp_obd)
2216 LASSERT(lsm->lsm_object_gr > 0);
2218 lov = &exp->exp_obd->u.lov;
2219 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2223 list_for_each (pos, &set->set_list) {
2224 req = list_entry(pos, struct lov_request, rq_link);
2225 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2227 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2228 req->rq_oi.oi_md, mode, lov_lockhp);
2229 rc = lov_update_common_set(set, req, rc);
2231 CERROR("error: cancel objid "LPX64" subobj "
2232 LPX64" on OST idx %d: rc = %d\n",
2234 req->rq_oi.oi_md->lsm_object_id,
2240 lov_fini_cancel_set(set);
2244 static int lov_cancel_unused(struct obd_export *exp,
2245 struct lov_stripe_md *lsm,
2246 int flags, void *opaque)
2248 struct lov_obd *lov;
2252 if (!exp || !exp->exp_obd)
2255 lov = &exp->exp_obd->u.lov;
2257 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2259 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2262 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2270 ASSERT_LSM_MAGIC(lsm);
2272 LASSERT(lsm->lsm_object_gr > 0);
2273 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2274 struct lov_stripe_md submd;
2275 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2278 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2279 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2283 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2284 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2286 submd.lsm_object_id = loi->loi_id;
2287 submd.lsm_object_gr = lsm->lsm_object_gr;
2288 submd.lsm_stripe_count = 0;
2289 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2290 &submd, flags, opaque);
2291 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2292 CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2293 " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2294 loi->loi_id, loi->loi_ost_idx, err);
2302 static int lov_join_lru(struct obd_export *exp,
2303 struct lov_stripe_md *lsm, int join)
2305 struct lov_obd *lov;
2309 ASSERT_LSM_MAGIC(lsm);
2310 if (!exp || !exp->exp_obd)
2313 lov = &exp->exp_obd->u.lov;
2314 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2315 struct lov_stripe_md submd;
2316 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2319 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2320 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2324 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2325 CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2327 submd.lsm_object_id = loi->loi_id;
2328 submd.lsm_object_gr = lsm->lsm_object_gr;
2329 submd.lsm_stripe_count = 0;
2330 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2333 CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2334 " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2335 loi->loi_id, loi->loi_ost_idx, rc);
2344 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2347 struct lov_request_set *lovset = (struct lov_request_set *)data;
2352 lovset->set_completes = 0;
2354 err = lov_fini_statfs_set(lovset);
2355 RETURN(rc ? rc : err);
2358 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2359 __u64 max_age, struct ptlrpc_request_set *rqset)
2361 struct lov_request_set *set;
2362 struct lov_request *req;
2363 struct list_head *pos;
2364 struct lov_obd *lov;
2368 LASSERT(oinfo != NULL);
2369 LASSERT(oinfo->oi_osfs != NULL);
2372 rc = lov_prep_statfs_set(obd, oinfo, &set);
2376 list_for_each (pos, &set->set_list) {
2377 struct obd_device *osc_obd;
2379 req = list_entry(pos, struct lov_request, rq_link);
2381 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2382 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2387 if (rc || list_empty(&rqset->set_requests)) {
2390 set->set_completes = 0;
2391 err = lov_fini_statfs_set(set);
2392 RETURN(rc ? rc : err);
2395 LASSERT(rqset->set_interpret == NULL);
2396 rqset->set_interpret = lov_statfs_interpret;
2397 rqset->set_arg = (void *)set;
2401 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2402 __u64 max_age, __u32 flags)
2404 struct ptlrpc_request_set *set = NULL;
2405 struct obd_info oinfo = { { { 0 } } };
2410 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2411 * statfs requests */
2412 set = ptlrpc_prep_set();
2416 oinfo.oi_osfs = osfs;
2417 oinfo.oi_flags = flags;
2418 rc = lov_statfs_async(obd, &oinfo, max_age, set);
2420 rc = ptlrpc_set_wait(set);
2421 ptlrpc_set_destroy(set);
2426 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2427 void *karg, void *uarg)
2429 struct obd_device *obddev = class_exp2obd(exp);
2430 struct lov_obd *lov = &obddev->u.lov;
2431 int i, rc, count = lov->desc.ld_tgt_count;
2432 struct obd_uuid *uuidp;
2436 case IOC_OBD_STATFS: {
2437 struct obd_ioctl_data *data = karg;
2438 struct obd_device *osc_obd;
2439 struct obd_statfs stat_buf = {0};
2442 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2443 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
2445 if ((index >= count))
2448 if (!lov->lov_tgts[index])
2449 /* Try again with the next index */
2451 if (!lov->lov_tgts[index]->ltd_active)
2454 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2458 /* got statfs data */
2459 rc = obd_statfs(osc_obd, &stat_buf,
2460 cfs_time_current_64() - HZ, 0);
2463 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2466 rc = copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2470 case OBD_IOC_LOV_GET_CONFIG: {
2471 struct obd_ioctl_data *data;
2472 struct lov_desc *desc;
2477 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2480 data = (struct obd_ioctl_data *)buf;
2482 if (sizeof(*desc) > data->ioc_inllen1) {
2483 obd_ioctl_freedata(buf, len);
2487 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2488 obd_ioctl_freedata(buf, len);
2492 if (sizeof(__u32) * count > data->ioc_inllen3) {
2493 obd_ioctl_freedata(buf, len);
2497 desc = (struct lov_desc *)data->ioc_inlbuf1;
2498 memcpy(desc, &(lov->desc), sizeof(*desc));
2500 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2501 genp = (__u32 *)data->ioc_inlbuf3;
2502 /* the uuid will be empty for deleted OSTs */
2503 for (i = 0; i < count; i++, uuidp++, genp++) {
2504 if (!lov->lov_tgts[i])
2506 *uuidp = lov->lov_tgts[i]->ltd_uuid;
2507 *genp = lov->lov_tgts[i]->ltd_gen;
2510 rc = copy_to_user((void *)uarg, buf, len);
2513 obd_ioctl_freedata(buf, len);
2516 case LL_IOC_LOV_SETSTRIPE:
2517 rc = lov_setstripe(exp, karg, uarg);
2519 case LL_IOC_LOV_GETSTRIPE:
2520 rc = lov_getstripe(exp, karg, uarg);
2522 case LL_IOC_LOV_SETEA:
2523 rc = lov_setea(exp, karg, uarg);
2532 for (i = 0; i < count; i++) {
2535 /* OST was disconnected */
2536 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2539 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2541 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2544 if (lov->lov_tgts[i]->ltd_active) {
2545 CDEBUG(err == -ENOTTY ?
2546 D_IOCTL : D_WARNING,
2547 "iocontrol OSC %s on OST "
2548 "idx %d cmd %x: err = %d\n",
2549 lov_uuid2str(lov, i),
2566 #define FIEMAP_BUFFER_SIZE 4096
2569 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2570 * call. The local end offset and the device are sent in the first
2571 * fm_extent. This function calculates the stripe number from the index.
2572 * This function returns a stripe_no on which mapping is to be restarted.
2574 * This function returns fm_end_offset which is the in-OST offset at which
2575 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2576 * will re-calculate proper offset in next stripe.
2577 * Note that the first extent is passed to lov_get_info via the value field.
2579 * \param fiemap fiemap request header
2580 * \param lsm striping information for the file
2581 * \param fm_start logical start of mapping
2582 * \param fm_end logical end of mapping
2583 * \param start_stripe starting stripe will be returned in this
2585 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2586 struct lov_stripe_md *lsm, obd_size fm_start,
2587 obd_size fm_end, int *start_stripe)
2589 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2590 obd_off lun_start, lun_end;
2591 obd_size fm_end_offset;
2592 int stripe_no = -1, i;
2594 if (fiemap->fm_extent_count == 0 ||
2595 fiemap->fm_extents[0].fe_logical == 0)
2598 /* Find out stripe_no from ost_index saved in the fe_device */
2599 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2600 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2601 fiemap->fm_extents[0].fe_device) {
2607 /* If we have finished mapping on previous device, shift logical
2608 * offset to start of next device */
2609 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2610 &lun_start, &lun_end)) != 0 &&
2611 local_end < lun_end) {
2612 fm_end_offset = local_end;
2613 *start_stripe = stripe_no;
2615 /* This is a special value to indicate that caller should
2616 * calculate offset in next stripe. */
2618 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2621 return fm_end_offset;
2625 * We calculate on which OST the mapping will end. If the length of mapping
2626 * is greater than (stripe_size * stripe_count) then the last_stripe will
2627 * will be one just before start_stripe. Else we check if the mapping
2628 * intersects each OST and find last_stripe.
2629 * This function returns the last_stripe and also sets the stripe_count
2630 * over which the mapping is spread
2632 * \param lsm striping information for the file
2633 * \param fm_start logical start of mapping
2634 * \param fm_end logical end of mapping
2635 * \param start_stripe starting stripe of the mapping
2636 * \param stripe_count the number of stripes across which to map is returned
2638 * \retval last_stripe return the last stripe of the mapping
2640 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2641 obd_size fm_end, int start_stripe,
2645 obd_off obd_start, obd_end;
2648 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2649 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2651 *stripe_count = lsm->lsm_stripe_count;
2653 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2654 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2655 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2656 &obd_start, &obd_end)) == 0)
2660 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2667 * Set fe_device and copy extents from local buffer into main return buffer.
2669 * \param fiemap fiemap request header
2670 * \param lcl_fm_ext array of local fiemap extents to be copied
2671 * \param ost_index OST index to be written into the fm_device field for each
2673 * \param ext_count number of extents to be copied
2674 * \param current_extent where to start copying in main extent array
2676 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2677 struct ll_fiemap_extent *lcl_fm_ext,
2678 int ost_index, unsigned int ext_count,
2684 for (ext = 0; ext < ext_count; ext++) {
2685 lcl_fm_ext[ext].fe_device = ost_index;
2686 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2689 /* Copy fm_extent's from fm_local to return buffer */
2690 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2691 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2695 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2696 * This also handles the restarting of FIEMAP calls in case mapping overflows
2697 * the available number of extents in single call.
2699 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2700 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2702 struct ll_fiemap_info_key *fm_key = key;
2703 struct ll_user_fiemap *fiemap = val;
2704 struct ll_user_fiemap *fm_local = NULL;
2705 struct ll_fiemap_extent *lcl_fm_ext;
2707 unsigned int get_num_extents = 0;
2708 int ost_index = 0, actual_start_stripe, start_stripe;
2709 obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2711 int current_extent = 0, rc = 0, i;
2712 int ost_eof = 0; /* EOF for object */
2713 int ost_done = 0; /* done with required mapping for this OST? */
2715 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2716 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2721 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2722 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2724 OBD_ALLOC(fm_local, buffer_size);
2725 if (fm_local == NULL)
2726 GOTO(out, rc = -ENOMEM);
2727 lcl_fm_ext = &fm_local->fm_extents[0];
2729 count_local = fiemap_size_to_count(buffer_size);
2731 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2732 fm_start = fiemap->fm_start;
2733 fm_length = fiemap->fm_length;
2734 /* Calculate start stripe, last stripe and length of mapping */
2735 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2736 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2737 fm_start + fm_length - 1);
2738 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2739 if (fm_end > fm_key->oa.o_size)
2740 fm_end = fm_key->oa.o_size;
2742 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2743 actual_start_stripe, &stripe_count);
2745 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2748 if (fiemap->fm_extent_count == 0) {
2749 get_num_extents = 1;
2753 /* Check each stripe */
2754 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2755 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2756 obd_size req_fm_len; /* Stores length of required mapping */
2757 obd_size len_mapped_single_call;
2758 obd_off lun_start, lun_end, obd_object_end;
2759 unsigned int ext_count;
2761 cur_stripe_wrap = cur_stripe;
2763 /* Find out range of mapping on this stripe */
2764 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2765 &lun_start, &obd_object_end)) == 0)
2768 /* If this is a continuation FIEMAP call and we are on
2769 * starting stripe then lun_start needs to be set to
2771 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2772 lun_start = fm_end_offset;
2774 if (fm_length != ~0ULL) {
2775 /* Handle fm_start + fm_length overflow */
2776 if (fm_start + fm_length < fm_start)
2777 fm_length = ~0ULL - fm_start;
2778 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2784 if (lun_start == lun_end)
2787 req_fm_len = obd_object_end - lun_start;
2788 fm_local->fm_length = 0;
2789 len_mapped_single_call = 0;
2791 /* If the output buffer is very large and the objects have many
2792 * extents we may need to loop on a single OST repeatedly */
2796 if (get_num_extents == 0) {
2797 /* Don't get too many extents. */
2798 if (current_extent + count_local >
2799 fiemap->fm_extent_count)
2800 count_local = fiemap->fm_extent_count -
2804 lun_start += len_mapped_single_call;
2805 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2806 req_fm_len = fm_local->fm_length;
2807 fm_local->fm_extent_count = count_local;
2808 fm_local->fm_mapped_extents = 0;
2809 fm_local->fm_flags = fiemap->fm_flags;
2811 fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2812 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2814 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2815 GOTO(out, rc = -EINVAL);
2817 /* If OST is inactive, return extent with UNKNOWN flag */
2818 if (lov && !lov->lov_tgts[ost_index]->ltd_active) {
2819 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2820 fm_local->fm_mapped_extents = 1;
2822 lcl_fm_ext[0].fe_logical = lun_start;
2823 lcl_fm_ext[0].fe_length = obd_object_end -
2825 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2830 fm_local->fm_start = lun_start;
2831 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2832 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2833 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2834 rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2835 keylen, key, vallen, fm_local, lsm);
2840 ext_count = fm_local->fm_mapped_extents;
2841 if (ext_count == 0) {
2843 /* If last stripe has hole at the end,
2844 * then we need to return */
2845 if (cur_stripe_wrap == last_stripe) {
2846 fiemap->fm_mapped_extents = 0;
2852 /* If we just need num of extents then go to next device */
2853 if (get_num_extents) {
2854 current_extent += ext_count;
2858 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2859 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2861 /* Have we finished mapping on this device? */
2862 if (req_fm_len <= len_mapped_single_call)
2865 /* Clear the EXTENT_LAST flag which can be present on
2867 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2868 lcl_fm_ext[ext_count - 1].fe_flags &=
2869 ~FIEMAP_EXTENT_LAST;
2871 curr_loc = lov_stripe_size(lsm,
2872 lcl_fm_ext[ext_count - 1].fe_logical+
2873 lcl_fm_ext[ext_count - 1].fe_length,
2875 if (curr_loc >= fm_key->oa.o_size)
2878 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2879 ost_index, ext_count,
2882 current_extent += ext_count;
2884 /* Ran out of available extents? */
2885 if (current_extent >= fiemap->fm_extent_count)
2887 } while (ost_done == 0 && ost_eof == 0);
2889 if (cur_stripe_wrap == last_stripe)
2894 /* Indicate that we are returning device offsets unless file just has
2896 if (lsm->lsm_stripe_count > 1)
2897 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2899 if (get_num_extents)
2900 goto skip_last_device_calc;
2902 /* Check if we have reached the last stripe and whether mapping for that
2903 * stripe is done. */
2904 if (cur_stripe_wrap == last_stripe) {
2905 if (ost_done || ost_eof)
2906 fiemap->fm_extents[current_extent - 1].fe_flags |=
2910 skip_last_device_calc:
2911 fiemap->fm_mapped_extents = current_extent;
2914 OBD_FREE(fm_local, buffer_size);
2918 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2919 void *key, __u32 *vallen, void *val,
2920 struct lov_stripe_md *lsm)
2922 struct obd_device *obddev = class_exp2obd(exp);
2923 struct lov_obd *lov = &obddev->u.lov;
2927 if (!vallen || !val)
2932 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2935 struct ldlm_lock *lock;
2937 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2938 struct lov_oinfo *loi;
2939 __u32 *stripe = val;
2941 if (*vallen < sizeof(*stripe))
2942 GOTO(out, rc = -EFAULT);
2943 *vallen = sizeof(*stripe);
2945 /* XXX This is another one of those bits that will need to
2946 * change if we ever actually support nested LOVs. It uses
2947 * the lock's export to find out which stripe it is. */
2948 /* XXX - it's assumed all the locks for deleted OSTs have
2949 * been cancelled. Also, the export for deleted OSTs will
2950 * be NULL and won't match the lock's export. */
2951 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2952 loi = lsm->lsm_oinfo[i];
2953 if (!lov->lov_tgts[loi->loi_ost_idx])
2955 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2956 data->lock->l_conn_export &&
2957 osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2962 LDLM_ERROR(data->lock, "lock on inode without such object");
2963 dump_lsm(D_ERROR, lsm);
2964 GOTO(out, rc = -ENXIO);
2965 } else if (KEY_IS(KEY_LAST_ID)) {
2966 struct obd_id_info *info = val;
2967 __u32 size = sizeof(obd_id);
2968 struct lov_tgt_desc *tgt;
2970 LASSERT(*vallen == sizeof(struct obd_id_info));
2971 tgt = lov->lov_tgts[info->idx];
2973 if (!tgt || !tgt->ltd_active)
2974 GOTO(out, rc = -ESRCH);
2976 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2978 } else if (KEY_IS(KEY_LOVDESC)) {
2979 struct lov_desc *desc_ret = val;
2980 *desc_ret = lov->desc;
2983 } else if (KEY_IS(KEY_FIEMAP)) {
2984 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2995 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2996 void *key, obd_count vallen, void *val,
2997 struct ptlrpc_request_set *set)
2999 struct obd_device *obddev = class_exp2obd(exp);
3000 struct lov_obd *lov = &obddev->u.lov;
3003 struct lov_tgt_desc *tgt;
3004 unsigned incr, check_uuid,
3005 do_inactive, no_set;
3006 unsigned next_id = 0, mds_con = 0;
3009 incr = check_uuid = do_inactive = no_set = 0;
3012 set = ptlrpc_prep_set();
3018 count = lov->desc.ld_tgt_count;
3020 if (KEY_IS(KEY_NEXT_ID)) {
3021 count = vallen / sizeof(struct obd_id_info);
3022 vallen = sizeof(obd_id);
3023 incr = sizeof(struct obd_id_info);
3026 } else if (KEY_IS(KEY_CHECKSUM)) {
3028 } else if (KEY_IS(KEY_UNLINKED)) {
3029 check_uuid = val ? 1 : 0;
3030 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
3031 /* use defaults: do_inactive = incr = 0; */
3032 } else if (KEY_IS(KEY_MDS_CONN)) {
3036 for (i = 0; i < count; i++, val = (char *)val + incr) {
3038 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
3040 tgt = lov->lov_tgts[i];
3042 /* OST was disconnected */
3043 if (!tgt || !tgt->ltd_exp)
3046 /* OST is inactive and we don't want inactive OSCs */
3047 if (!tgt->ltd_active && !do_inactive)
3051 struct mds_group_info *mgi;
3053 LASSERT(vallen == sizeof(*mgi));
3054 mgi = (struct mds_group_info *)val;
3056 /* Only want a specific OSC */
3057 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
3061 err = obd_set_info_async(tgt->ltd_exp,
3062 keylen, key, sizeof(int),
3064 } else if (next_id) {
3065 err = obd_set_info_async(tgt->ltd_exp,
3066 keylen, key, vallen,
3067 ((struct obd_id_info*)val)->data, set);
3069 /* Only want a specific OSC */
3071 !obd_uuid_equals(val, &tgt->ltd_uuid))
3074 err = obd_set_info_async(tgt->ltd_exp,
3075 keylen, key, vallen, val, set);
3084 err = ptlrpc_set_wait(set);
3087 ptlrpc_set_destroy(set);
3092 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3093 struct lov_stripe_md *lsm)
3101 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3102 rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3107 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3112 for (i = 0; i < lsm->lsm_stripe_count; i++) {
3113 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
3114 if (loi->loi_ar.ar_rc && !rc)
3115 rc = loi->loi_ar.ar_rc;
3116 loi->loi_ar.ar_rc = 0;
3120 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3123 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3124 int cmd, __u64 *offset)
3126 __u32 ssize = lsm->lsm_stripe_size;
3130 do_div(start, ssize);
3131 start = start * ssize;
3133 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3134 ", end "LPU64"\n", *offset, ssize, start,
3136 if (cmd == OBD_CALC_STRIPE_END) {
3137 *offset = start + ssize - 1;
3138 } else if (cmd == OBD_CALC_STRIPE_START) {
3149 struct lov_multi_wait {
3150 struct ldlm_lock *lock;
3156 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3157 struct lustre_handle *lockh)
3159 struct lov_lock_handles *lov_lockh = NULL;
3160 struct lustre_handle *lov_lockhp;
3161 struct lov_obd *lov;
3162 struct lov_oinfo *loi;
3163 struct lov_multi_wait *queues;
3167 ASSERT_LSM_MAGIC(lsm);
3169 if (!exp || !exp->exp_obd)
3172 LASSERT(lockh != NULL);
3173 if (lsm->lsm_stripe_count > 1) {
3174 lov_lockh = lov_handle2llh(lockh);
3175 if (lov_lockh == NULL) {
3176 CERROR("LOV: invalid lov lock handle %p\n", lockh);
3180 lov_lockhp = lov_lockh->llh_handles;
3185 OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3187 GOTO(out, rc = -ENOMEM);
3189 lov = &exp->exp_obd->u.lov;
3190 for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3191 i++, loi++, lov_lockhp++) {
3192 struct ldlm_lock *lock;
3193 struct obd_device *obd;
3195 lock = ldlm_handle2lock(lov_lockhp);
3197 CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3198 loi->loi_ost_idx, loi->loi_id);
3199 queues[i].completed = 1;
3203 queues[i].lock = lock;
3204 init_waitqueue_entry(&(queues[i].wait), current);
3205 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3207 obd = class_exp2obd(lock->l_conn_export);
3209 imp = obd->u.cli.cl_import;
3211 spin_lock(&imp->imp_lock);
3212 queues[i].generation = imp->imp_generation;
3213 spin_unlock(&imp->imp_lock);
3217 lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3218 interrupted_completion_wait, &lwd);
3219 rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3221 for (i = 0; i < lsm->lsm_stripe_count; i++)
3222 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3224 if (rc == -EINTR || rc == -ETIMEDOUT) {
3230 if (lov_lockh != NULL)
3231 lov_llh_put(lov_lockh);
3236 void lov_stripe_lock(struct lov_stripe_md *md)
3238 LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
3239 spin_lock(&md->lsm_lock);
3240 LASSERT(md->lsm_lock_owner == 0);
3241 md->lsm_lock_owner = cfs_curproc_pid();
3243 EXPORT_SYMBOL(lov_stripe_lock);
3245 void lov_stripe_unlock(struct lov_stripe_md *md)
3247 LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
3248 md->lsm_lock_owner = 0;
3249 spin_unlock(&md->lsm_lock);
3251 EXPORT_SYMBOL(lov_stripe_unlock);
3254 * Checks if requested extent lock is compatible with a lock under the page.
3256 * Checks if the lock under \a page is compatible with a read or write lock
3257 * (specified by \a rw) for an extent [\a start , \a end].
3259 * \param exp lov export
3260 * \param lsm striping information for the file
3261 * \param res lov_async_page placeholder
3262 * \param rw OBD_BRW_READ if requested for reading,
3263 * OBD_BRW_WRITE if requested for writing
3264 * \param start start of the requested extent
3265 * \param end end of the requested extent
3266 * \param cookie transparent parameter for passing locking context
3268 * \post result == 1, *cookie == context, appropriate lock is referenced or
3271 * \retval 1 owned lock is reused for the request
3272 * \retval 0 no lock reused for the request
3274 * \see lov_release_short_lock
3276 static int lov_reget_short_lock(struct obd_export *exp,
3277 struct lov_stripe_md *lsm,
3279 obd_off start, obd_off end,
3282 struct lov_async_page *l = *res;
3283 obd_off stripe_start, stripe_end = start;
3287 /* ensure we don't cross stripe boundaries */
3288 lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3289 if (stripe_end <= end)
3292 /* map the region limits to the object limits */
3293 lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3294 lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3296 RETURN(obd_reget_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3297 lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3298 ltd_exp, NULL, &l->lap_sub_cookie,
3299 rw, stripe_start, stripe_end, cookie));
3303 * Releases a reference to a lock taken in a "fast" way.
3305 * Releases a read or a write (specified by \a rw) lock
3306 * referenced by \a cookie.
3308 * \param exp lov export
3309 * \param lsm striping information for the file
3310 * \param end end of the locked extent
3311 * \param rw OBD_BRW_READ if requested for reading,
3312 * OBD_BRW_WRITE if requested for writing
3313 * \param cookie transparent parameter for passing locking context
3315 * \post appropriate lock is dereferenced
3317 * \see lov_reget_short_lock
3319 static int lov_release_short_lock(struct obd_export *exp,
3320 struct lov_stripe_md *lsm, obd_off end,
3321 void *cookie, int rw)
3327 stripe = lov_stripe_number(lsm, end);
3329 RETURN(obd_release_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3330 lsm_oinfo[stripe]->loi_ost_idx]->
3331 ltd_exp, NULL, end, cookie, rw));
3334 struct obd_ops lov_obd_ops = {
3335 .o_owner = THIS_MODULE,
3336 .o_setup = lov_setup,
3337 .o_precleanup = lov_precleanup,
3338 .o_cleanup = lov_cleanup,
3339 .o_process_config = lov_process_config,
3340 .o_connect = lov_connect,
3341 .o_disconnect = lov_disconnect,
3342 .o_statfs = lov_statfs,
3343 .o_statfs_async = lov_statfs_async,
3344 .o_packmd = lov_packmd,
3345 .o_unpackmd = lov_unpackmd,
3346 .o_checkmd = lov_checkmd,
3347 .o_create = lov_create,
3348 .o_destroy = lov_destroy,
3349 .o_getattr = lov_getattr,
3350 .o_getattr_async = lov_getattr_async,
3351 .o_setattr = lov_setattr,
3352 .o_setattr_async = lov_setattr_async,
3354 .o_brw_async = lov_brw_async,
3355 .o_prep_async_page = lov_prep_async_page,
3356 .o_reget_short_lock = lov_reget_short_lock,
3357 .o_release_short_lock = lov_release_short_lock,
3358 .o_queue_async_io = lov_queue_async_io,
3359 .o_set_async_flags = lov_set_async_flags,
3360 .o_queue_group_io = lov_queue_group_io,
3361 .o_trigger_group_io = lov_trigger_group_io,
3362 .o_teardown_async_page = lov_teardown_async_page,
3363 .o_merge_lvb = lov_merge_lvb,
3364 .o_adjust_kms = lov_adjust_kms,
3365 .o_punch = lov_punch,
3367 .o_enqueue = lov_enqueue,
3368 .o_match = lov_match,
3369 .o_change_cbdata = lov_change_cbdata,
3370 .o_cancel = lov_cancel,
3371 .o_cancel_unused = lov_cancel_unused,
3372 .o_join_lru = lov_join_lru,
3373 .o_iocontrol = lov_iocontrol,
3374 .o_get_info = lov_get_info,
3375 .o_set_info_async = lov_set_info_async,
3376 .o_extent_calc = lov_extent_calc,
3377 .o_llog_init = lov_llog_init,
3378 .o_llog_finish = lov_llog_finish,
3379 .o_notify = lov_notify,
3380 .o_register_page_removal_cb = lov_register_page_removal_cb,
3381 .o_unregister_page_removal_cb = lov_unregister_page_removal_cb,
3382 .o_register_lock_cancel_cb = lov_register_lock_cancel_cb,
3383 .o_unregister_lock_cancel_cb = lov_unregister_lock_cancel_cb,
3384 .o_pool_new = lov_pool_new,
3385 .o_pool_rem = lov_pool_remove,
3386 .o_pool_add = lov_pool_add,
3387 .o_pool_del = lov_pool_del,
3390 static quota_interface_t *quota_interface;
3391 extern quota_interface_t lov_quota_interface;
3393 cfs_mem_cache_t *lov_oinfo_slab;
3395 int __init lov_init(void)
3397 struct lprocfs_static_vars lvars = { 0 };
3401 lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3402 sizeof(struct lov_oinfo),
3403 0, SLAB_HWCACHE_ALIGN);
3404 if (lov_oinfo_slab == NULL)
3406 lprocfs_lov_init_vars(&lvars);
3408 request_module("lquota");
3409 quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3410 init_obd_quota_ops(quota_interface, &lov_obd_ops);
3412 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3413 LUSTRE_LOV_NAME, NULL);
3415 if (quota_interface)
3416 PORTAL_SYMBOL_PUT(lov_quota_interface);
3417 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3425 static void /*__exit*/ lov_exit(void)
3429 if (quota_interface)
3430 PORTAL_SYMBOL_PUT(lov_quota_interface);
3432 class_unregister_type(LUSTRE_LOV_NAME);
3433 rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3437 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3438 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3439 MODULE_LICENSE("GPL");
3441 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);