Whamcloud - gitweb
099eb628dd8b7a483a0f179dba6bf899bebefc1d
[fs/lustre-release.git] / lustre / lov / lov_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/lov/lov_obd.c
37  *
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>
42  */
43
44 #define DEBUG_SUBSYSTEM S_LOV
45 #ifdef __KERNEL__
46 #include <libcfs/libcfs.h>
47 #else
48 #include <liblustre.h>
49 #endif
50
51 #include <obd_support.h>
52 #include <lustre_ioctl.h>
53 #include <lustre_lib.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_dlm.h>
57 #include <lustre_mds.h>
58 #include <obd_class.h>
59 #include <lprocfs_status.h>
60 #include <lustre_param.h>
61 #include <cl_object.h>
62 #include <lclient.h>
63 #include <lustre/ll_fiemap.h>
64 #include <lustre_fid.h>
65
66 #include "lov_internal.h"
67
68 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
69    Any function that expects lov_tgts to remain stationary must take a ref. */
70 static void lov_getref(struct obd_device *obd)
71 {
72         struct lov_obd *lov = &obd->u.lov;
73
74         /* nobody gets through here until lov_putref is done */
75         mutex_lock(&lov->lov_lock);
76         atomic_inc(&lov->lov_refcount);
77         mutex_unlock(&lov->lov_lock);
78         return;
79 }
80
81 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
82
83 static void lov_putref(struct obd_device *obd)
84 {
85         struct lov_obd *lov = &obd->u.lov;
86
87         mutex_lock(&lov->lov_lock);
88         /* ok to dec to 0 more than once -- ltd_exp's will be null */
89         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
90                 struct list_head kill = LIST_HEAD_INIT(kill);
91                 struct lov_tgt_desc *tgt, *n;
92                 int i;
93
94                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
95                        lov->lov_death_row);
96                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
97                         tgt = lov->lov_tgts[i];
98
99                         if (!tgt || !tgt->ltd_reap)
100                                 continue;
101                         list_add(&tgt->ltd_kill, &kill);
102                         /* XXX - right now there is a dependency on ld_tgt_count
103                          * being the maximum tgt index for computing the
104                          * mds_max_easize. So we can't shrink it. */
105                         lov_ost_pool_remove(&lov->lov_packed, i);
106                         lov->lov_tgts[i] = NULL;
107                         lov->lov_death_row--;
108                 }
109                 mutex_unlock(&lov->lov_lock);
110
111                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
112                         list_del(&tgt->ltd_kill);
113                         /* Disconnect */
114                         __lov_del_obd(obd, tgt);
115                 }
116         } else {
117                 mutex_unlock(&lov->lov_lock);
118         }
119 }
120
121 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
122                               enum obd_notify_event ev);
123 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
124                       enum obd_notify_event ev, void *data);
125
126 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
127                     struct obd_connect_data *data)
128 {
129         struct lov_obd *lov = &obd->u.lov;
130         struct obd_uuid *tgt_uuid;
131         struct obd_device *tgt_obd;
132         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
133         struct obd_import *imp;
134         int rc;
135         ENTRY;
136
137         if (lov->lov_tgts[index] == NULL)
138                 RETURN(-EINVAL);
139
140         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
141         tgt_obd = lov->lov_tgts[index]->ltd_obd;
142
143         if (!tgt_obd->obd_set_up) {
144                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
145                 RETURN(-EINVAL);
146         }
147
148         /* override the sp_me from lov */
149         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
150
151         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
152                 data->ocd_index = index;
153
154         /*
155          * Divine LOV knows that OBDs under it are OSCs.
156          */
157         imp = tgt_obd->u.cli.cl_import;
158
159         if (activate) {
160                 tgt_obd->obd_no_recov = 0;
161                 /* FIXME this is probably supposed to be
162                    ptlrpc_set_import_active.  Horrible naming. */
163                 ptlrpc_activate_import(imp);
164         }
165
166         rc = obd_register_observer(tgt_obd, obd);
167         if (rc) {
168                 CERROR("Target %s register_observer error %d\n",
169                        obd_uuid2str(tgt_uuid), rc);
170                 RETURN(rc);
171         }
172
173
174         if (imp->imp_invalid) {
175                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
176                        "disabled\n", obd_uuid2str(tgt_uuid));
177                 RETURN(0);
178         }
179
180         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
181                          &lov_osc_uuid, data, NULL);
182         if (rc || !lov->lov_tgts[index]->ltd_exp) {
183                 CERROR("Target %s connect error %d\n",
184                        obd_uuid2str(tgt_uuid), rc);
185                 RETURN(-ENODEV);
186         }
187
188         lov->lov_tgts[index]->ltd_reap = 0;
189
190         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
191                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
192
193         if (obd->obd_type->typ_procsym != NULL) {
194                 struct proc_dir_entry *osc_symlink;
195                 struct obd_device *osc_obd;
196
197                 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
198
199                 LASSERT(osc_obd != NULL);
200                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
201                 LASSERT(osc_obd->obd_type->typ_name != NULL);
202
203                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
204                                                   obd->obd_type->typ_procsym,
205                                                   "../../../%s/%s",
206                                                   osc_obd->obd_type->typ_name,
207                                                   osc_obd->obd_name);
208                 if (osc_symlink == NULL) {
209                         CERROR("could not register LOV target "
210                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
211                                obd->obd_type->typ_name, obd->obd_name,
212                                osc_obd->obd_name);
213                         lprocfs_remove(&obd->obd_type->typ_procsym);
214                 }
215         }
216         RETURN(0);
217 }
218
219 static int lov_connect(const struct lu_env *env,
220                        struct obd_export **exp, struct obd_device *obd,
221                        struct obd_uuid *cluuid, struct obd_connect_data *data,
222                        void *localdata)
223 {
224         struct lov_obd *lov = &obd->u.lov;
225         struct lov_tgt_desc *tgt;
226         struct lustre_handle conn;
227         int i, rc;
228         ENTRY;
229
230         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
231
232         rc = class_connect(&conn, obd, cluuid);
233         if (rc)
234                 RETURN(rc);
235
236         *exp = class_conn2export(&conn);
237
238         /* Why should there ever be more than 1 connect? */
239         lov->lov_connects++;
240         LASSERT(lov->lov_connects == 1);
241
242         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
243         if (data)
244                 lov->lov_ocd = *data;
245
246         obd->obd_type->typ_procsym = lprocfs_seq_register("target_obds",
247                                                  obd->obd_proc_entry,
248                                                  NULL, NULL);
249         if (IS_ERR(obd->obd_type->typ_procsym)) {
250                 CERROR("%s: could not register /proc/fs/lustre/%s/%s/target_obds.",
251                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
252                 obd->obd_type->typ_procsym = NULL;
253         }
254
255         obd_getref(obd);
256         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
257                 tgt = lov->lov_tgts[i];
258                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
259                         continue;
260                 /* Flags will be lowest common denominator */
261                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
262                 if (rc) {
263                         CERROR("%s: lov connect tgt %d failed: %d\n",
264                                obd->obd_name, i, rc);
265                         continue;
266                 }
267                 /* connect to administrative disabled ost */
268                 if (!lov->lov_tgts[i]->ltd_exp)
269                         continue;
270
271                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
272                                 OBD_NOTIFY_CONNECT, (void *)&i);
273                 if (rc) {
274                         CERROR("%s error sending notify %d\n",
275                                obd->obd_name, rc);
276                 }
277         }
278         obd_putref(obd);
279
280         RETURN(0);
281 }
282
283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
284 {
285         struct lov_obd *lov = &obd->u.lov;
286         struct obd_device *osc_obd;
287         int rc;
288         ENTRY;
289
290         osc_obd = class_exp2obd(tgt->ltd_exp);
291         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
292                obd->obd_name, osc_obd->obd_name);
293
294         if (tgt->ltd_active) {
295                 tgt->ltd_active = 0;
296                 lov->desc.ld_active_tgt_count--;
297                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
298         }
299
300         if (osc_obd) {
301                 /* Pass it on to our clients.
302                  * XXX This should be an argument to disconnect,
303                  * XXX not a back-door flag on the OBD.  Ah well.
304                  */
305                 osc_obd->obd_force = obd->obd_force;
306                 osc_obd->obd_fail = obd->obd_fail;
307                 osc_obd->obd_no_recov = obd->obd_no_recov;
308
309                 if (obd->obd_type->typ_procsym)
310                         lprocfs_remove_proc_entry(osc_obd->obd_name,
311                                                   obd->obd_type->typ_procsym);
312         }
313
314         obd_register_observer(osc_obd, NULL);
315
316         rc = obd_disconnect(tgt->ltd_exp);
317         if (rc) {
318                 CERROR("Target %s disconnect error %d\n",
319                        tgt->ltd_uuid.uuid, rc);
320                 rc = 0;
321         }
322
323         tgt->ltd_exp = NULL;
324         RETURN(0);
325 }
326
327 static int lov_disconnect(struct obd_export *exp)
328 {
329         struct obd_device *obd = class_exp2obd(exp);
330         struct lov_obd *lov = &obd->u.lov;
331         int i, rc;
332         ENTRY;
333
334         if (!lov->lov_tgts)
335                 goto out;
336
337         /* Only disconnect the underlying layers on the final disconnect. */
338         lov->lov_connects--;
339         if (lov->lov_connects != 0) {
340                 /* why should there be more than 1 connect? */
341                 CERROR("disconnect #%d\n", lov->lov_connects);
342                 goto out;
343         }
344
345         /* Let's hold another reference so lov_del_obd doesn't spin through
346            putref every time */
347         obd_getref(obd);
348
349         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
350                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
351                         /* Disconnection is the last we know about an obd */
352                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
353                 }
354         }
355         obd_putref(obd);
356
357         if (obd->obd_type->typ_procsym)
358                 lprocfs_remove(&obd->obd_type->typ_procsym);
359
360 out:
361         rc = class_disconnect(exp); /* bz 9811 */
362         RETURN(rc);
363 }
364
365 /* Error codes:
366  *
367  *  -EINVAL  : UUID can't be found in the LOV's target list
368  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
369  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
370  *  any >= 0 : is log target index
371  */
372 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
373                               enum obd_notify_event ev)
374 {
375         struct lov_obd *lov = &obd->u.lov;
376         struct lov_tgt_desc *tgt;
377         int index, activate, active;
378         ENTRY;
379
380         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
381                lov, uuid->uuid, ev);
382
383         obd_getref(obd);
384         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
385                 tgt = lov->lov_tgts[index];
386                 if (!tgt)
387                         continue;
388                 /*
389                  * LU-642, initially inactive OSC could miss the obd_connect,
390                  * we make up for it here.
391                  */
392                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
393                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
394                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
395
396                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
397                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
398                 }
399                 if (!tgt->ltd_exp)
400                         continue;
401
402                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
403                        index, obd_uuid2str(&tgt->ltd_uuid),
404                        tgt->ltd_exp->exp_handle.h_cookie);
405                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
406                         break;
407         }
408
409         if (index == lov->desc.ld_tgt_count)
410                 GOTO(out, index = -EINVAL);
411
412         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
413                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
414
415                 if (lov->lov_tgts[index]->ltd_activate == activate) {
416                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
417                                uuid->uuid, activate ? "" : "de");
418                 } else {
419                         lov->lov_tgts[index]->ltd_activate = activate;
420                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
421                                activate ? "" : "de", obd_uuid2str(uuid));
422                 }
423
424         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
425                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
426
427                 if (lov->lov_tgts[index]->ltd_active == active) {
428                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
429                                uuid->uuid, active ? "" : "in");
430                         GOTO(out, index);
431                 } else {
432                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
433                                obd_uuid2str(uuid), active ? "" : "in");
434                 }
435
436                 lov->lov_tgts[index]->ltd_active = active;
437                 if (active) {
438                         lov->desc.ld_active_tgt_count++;
439                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
440                 } else {
441                         lov->desc.ld_active_tgt_count--;
442                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
443                 }
444         } else {
445                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
446         }
447
448  out:
449         obd_putref(obd);
450         RETURN(index);
451 }
452
453 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
454                       enum obd_notify_event ev, void *data)
455 {
456         int rc = 0;
457         struct lov_obd *lov = &obd->u.lov;
458         ENTRY;
459
460         down_read(&lov->lov_notify_lock);
461         if (!lov->lov_connects) {
462                 up_read(&lov->lov_notify_lock);
463                 RETURN(rc);
464         }
465
466         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
467             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
468                 struct obd_uuid *uuid;
469
470                 LASSERT(watched);
471
472                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
473                         up_read(&lov->lov_notify_lock);
474                         CERROR("unexpected notification of %s %s!\n",
475                                watched->obd_type->typ_name,
476                                watched->obd_name);
477                         RETURN(-EINVAL);
478                 }
479                 uuid = &watched->u.cli.cl_target_uuid;
480
481                 /* Set OSC as active before notifying the observer, so the
482                  * observer can use the OSC normally.
483                  */
484                 rc = lov_set_osc_active(obd, uuid, ev);
485                 if (rc < 0) {
486                         up_read(&lov->lov_notify_lock);
487                         CERROR("event(%d) of %s failed: %d\n", ev,
488                                obd_uuid2str(uuid), rc);
489                         RETURN(rc);
490                 }
491                 /* active event should be pass lov target index as data */
492                 data = &rc;
493         }
494
495         /* Pass the notification up the chain. */
496         if (watched) {
497                 rc = obd_notify_observer(obd, watched, ev, data);
498         } else {
499                 /* NULL watched means all osc's in the lov (only for syncs) */
500                 /* sync event should be send lov idx as data */
501                 struct lov_obd *lov = &obd->u.lov;
502                 int i, is_sync;
503
504                 data = &i;
505                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
506                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
507
508                 obd_getref(obd);
509                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
510                         if (!lov->lov_tgts[i])
511                                 continue;
512
513                         /* don't send sync event if target not
514                          * connected/activated */
515                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
516                                 continue;
517
518                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
519                                                  ev, data);
520                         if (rc) {
521                                 CERROR("%s: notify %s of %s failed %d\n",
522                                        obd->obd_name,
523                                        obd->obd_observer->obd_name,
524                                        lov->lov_tgts[i]->ltd_obd->obd_name,
525                                        rc);
526                         }
527                 }
528                 obd_putref(obd);
529         }
530
531         up_read(&lov->lov_notify_lock);
532         RETURN(rc);
533 }
534
535 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
536                           __u32 index, int gen, int active)
537 {
538         struct lov_obd *lov = &obd->u.lov;
539         struct lov_tgt_desc *tgt;
540         struct obd_device *tgt_obd;
541         int rc;
542         ENTRY;
543
544         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
545                uuidp->uuid, index, gen, active);
546
547         if (gen <= 0) {
548                 CERROR("request to add OBD %s with invalid generation: %d\n",
549                        uuidp->uuid, gen);
550                 RETURN(-EINVAL);
551         }
552
553         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
554                                         &obd->obd_uuid);
555         if (tgt_obd == NULL)
556                 RETURN(-EINVAL);
557
558         mutex_lock(&lov->lov_lock);
559
560         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
561                 tgt = lov->lov_tgts[index];
562                 CERROR("UUID %s already assigned at LOV target index %d\n",
563                        obd_uuid2str(&tgt->ltd_uuid), index);
564                 mutex_unlock(&lov->lov_lock);
565                 RETURN(-EEXIST);
566         }
567
568         if (index >= lov->lov_tgt_size) {
569                 /* We need to reallocate the lov target array. */
570                 struct lov_tgt_desc **newtgts, **old = NULL;
571                 __u32 newsize, oldsize = 0;
572
573                 newsize = max(lov->lov_tgt_size, (__u32)2);
574                 while (newsize < index + 1)
575                         newsize = newsize << 1;
576                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
577                 if (newtgts == NULL) {
578                         mutex_unlock(&lov->lov_lock);
579                         RETURN(-ENOMEM);
580                 }
581
582                 if (lov->lov_tgt_size) {
583                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
584                                lov->lov_tgt_size);
585                         old = lov->lov_tgts;
586                         oldsize = lov->lov_tgt_size;
587                 }
588
589                 lov->lov_tgts = newtgts;
590                 lov->lov_tgt_size = newsize;
591                 smp_rmb();
592                 if (old)
593                         OBD_FREE(old, sizeof(*old) * oldsize);
594
595                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
596                        lov->lov_tgts, lov->lov_tgt_size);
597         }
598
599         OBD_ALLOC_PTR(tgt);
600         if (!tgt) {
601                 mutex_unlock(&lov->lov_lock);
602                 RETURN(-ENOMEM);
603         }
604
605         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
606         if (rc) {
607                 mutex_unlock(&lov->lov_lock);
608                 OBD_FREE_PTR(tgt);
609                 RETURN(rc);
610         }
611
612         tgt->ltd_uuid = *uuidp;
613         tgt->ltd_obd = tgt_obd;
614         /* XXX - add a sanity check on the generation number. */
615         tgt->ltd_gen = gen;
616         tgt->ltd_index = index;
617         tgt->ltd_activate = active;
618         lov->lov_tgts[index] = tgt;
619         if (index >= lov->desc.ld_tgt_count)
620                 lov->desc.ld_tgt_count = index + 1;
621
622         mutex_unlock(&lov->lov_lock);
623
624         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
625                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
626
627         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
628
629         if (lov->lov_connects == 0) {
630                 /* lov_connect hasn't been called yet. We'll do the
631                    lov_connect_obd on this target when that fn first runs,
632                    because we don't know the connect flags yet. */
633                 RETURN(0);
634         }
635
636         obd_getref(obd);
637
638         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
639         if (rc)
640                 GOTO(out, rc);
641
642         /* connect to administrative disabled ost */
643         if (!tgt->ltd_exp)
644                 GOTO(out, rc = 0);
645
646         if (lov->lov_cache != NULL) {
647                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
648                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
649                                 sizeof(struct cl_client_cache), lov->lov_cache,
650                                 NULL);
651                 if (rc < 0)
652                         GOTO(out, rc);
653         }
654
655         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
656                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
657                         (void *)&index);
658
659 out:
660         if (rc) {
661                 CERROR("add failed (%d), deleting %s\n", rc,
662                        obd_uuid2str(&tgt->ltd_uuid));
663                 lov_del_target(obd, index, 0, 0);
664         }
665         obd_putref(obd);
666         RETURN(rc);
667 }
668
669 /* Schedule a target for deletion */
670 int lov_del_target(struct obd_device *obd, __u32 index,
671                    struct obd_uuid *uuidp, int gen)
672 {
673         struct lov_obd *lov = &obd->u.lov;
674         int count = lov->desc.ld_tgt_count;
675         int rc = 0;
676         ENTRY;
677
678         if (index >= count) {
679                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
680                        index, count);
681                 RETURN(-EINVAL);
682         }
683
684         /* to make sure there's no ongoing lov_notify() now */
685         down_write(&lov->lov_notify_lock);
686         obd_getref(obd);
687
688         if (!lov->lov_tgts[index]) {
689                 CERROR("LOV target at index %d is not setup.\n", index);
690                 GOTO(out, rc = -EINVAL);
691         }
692
693         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
694                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
695                        lov_uuid2str(lov, index), index,
696                        obd_uuid2str(uuidp));
697                 GOTO(out, rc = -EINVAL);
698         }
699
700         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
701                lov_uuid2str(lov, index), index,
702                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
703                lov->lov_tgts[index]->ltd_active);
704
705         lov->lov_tgts[index]->ltd_reap = 1;
706         lov->lov_death_row++;
707         /* we really delete it from obd_putref */
708 out:
709         obd_putref(obd);
710         up_write(&lov->lov_notify_lock);
711
712         RETURN(rc);
713 }
714
715 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
716 {
717         struct obd_device *osc_obd;
718
719         LASSERT(tgt);
720         LASSERT(tgt->ltd_reap);
721
722         osc_obd = class_exp2obd(tgt->ltd_exp);
723
724         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
725                tgt->ltd_uuid.uuid,
726                osc_obd ? osc_obd->obd_name : "<no obd>");
727
728         if (tgt->ltd_exp)
729                 lov_disconnect_obd(obd, tgt);
730
731         OBD_FREE_PTR(tgt);
732
733         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
734            do it ourselves. And we can't do it from lov_cleanup,
735            because we just lost our only reference to it. */
736         if (osc_obd)
737                 class_manual_cleanup(osc_obd);
738 }
739
740 void lov_fix_desc_stripe_size(__u64 *val)
741 {
742         if (*val < LOV_MIN_STRIPE_SIZE) {
743                 if (*val != 0)
744                         LCONSOLE_INFO("Increasing default stripe size to "
745                                       "minimum %u\n",
746                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
747                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
748         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
749                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
750                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
751                               "multiple of %u)\n",
752                               *val, LOV_MIN_STRIPE_SIZE);
753         }
754 }
755
756 void lov_fix_desc_stripe_count(__u32 *val)
757 {
758         if (*val == 0)
759                 *val = 1;
760 }
761
762 void lov_fix_desc_pattern(__u32 *val)
763 {
764         /* from lov_setstripe */
765         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
766                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
767                 *val = 0;
768         }
769 }
770
771 void lov_fix_desc_qos_maxage(__u32 *val)
772 {
773         if (*val == 0)
774                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
775 }
776
777 void lov_fix_desc(struct lov_desc *desc)
778 {
779         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
780         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
781         lov_fix_desc_pattern(&desc->ld_pattern);
782         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
783 }
784
785 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
786 {
787         struct lov_desc *desc;
788         struct lov_obd *lov = &obd->u.lov;
789 #ifdef LPROCFS
790         struct obd_type *type;
791 #endif
792         int rc;
793         ENTRY;
794
795         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
796                 CERROR("LOV setup requires a descriptor\n");
797                 RETURN(-EINVAL);
798         }
799
800         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
801
802         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
803                 CERROR("descriptor size wrong: %d > %d\n",
804                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
805                 RETURN(-EINVAL);
806         }
807
808         if (desc->ld_magic != LOV_DESC_MAGIC) {
809                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
810                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
811                                    obd->obd_name, desc);
812                             lustre_swab_lov_desc(desc);
813                 } else {
814                         CERROR("%s: Bad lov desc magic: %#x\n",
815                                obd->obd_name, desc->ld_magic);
816                         RETURN(-EINVAL);
817                 }
818         }
819
820         lov_fix_desc(desc);
821
822         desc->ld_active_tgt_count = 0;
823         lov->desc = *desc;
824         lov->lov_tgt_size = 0;
825
826         mutex_init(&lov->lov_lock);
827         atomic_set(&lov->lov_refcount, 0);
828         lov->lov_sp_me = LUSTRE_SP_CLI;
829
830         init_rwsem(&lov->lov_notify_lock);
831
832         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
833                                                    HASH_POOLS_MAX_BITS,
834                                                    HASH_POOLS_BKT_BITS, 0,
835                                                    CFS_HASH_MIN_THETA,
836                                                    CFS_HASH_MAX_THETA,
837                                                    &pool_hash_operations,
838                                                    CFS_HASH_DEFAULT);
839         INIT_LIST_HEAD(&lov->lov_pool_list);
840         lov->lov_pool_count = 0;
841         rc = lov_ost_pool_init(&lov->lov_packed, 0);
842         if (rc)
843                 GOTO(out, rc);
844
845 #ifdef LPROCFS
846         obd->obd_vars = lprocfs_lov_obd_vars;
847         /* If this is true then both client (lov) and server
848          * (lod) are on the same node. The lod layer if loaded
849          * first will register the lov proc directory. In that
850          * case obd->obd_type->typ_procroot will be not set.
851          * Instead we use type->typ_procsym as the parent. */
852         type = class_search_type(LUSTRE_LOD_NAME);
853         if (type != NULL && type->typ_procsym != NULL) {
854                 obd->obd_proc_entry = lprocfs_seq_register(obd->obd_name,
855                                                            type->typ_procsym,
856                                                            obd->obd_vars, obd);
857                 if (IS_ERR(obd->obd_proc_entry)) {
858                         rc = PTR_ERR(obd->obd_proc_entry);
859                         CERROR("error %d setting up lprocfs for %s\n", rc,
860                                obd->obd_name);
861                         obd->obd_proc_entry = NULL;
862                 }
863         } else {
864                 rc = lprocfs_seq_obd_setup(obd);
865         }
866
867         if (rc == 0) {
868                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
869                                         0444, &lov_proc_target_fops, obd);
870                 if (rc)
871                         CWARN("Error adding the target_obd file\n");
872
873                 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
874                                                         obd->obd_proc_entry,
875                                                         NULL, NULL);
876                 if (IS_ERR(lov->lov_pool_proc_entry)) {
877                         rc = PTR_ERR(lov->lov_pool_proc_entry);
878                         CERROR("error %d setting up lprocfs for pools\n", rc);
879                         lov->lov_pool_proc_entry = NULL;
880                 }
881         }
882 #endif
883         RETURN(0);
884
885 out:
886         return rc;
887 }
888
889 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
890 {
891         int rc = 0;
892         struct lov_obd *lov = &obd->u.lov;
893
894         ENTRY;
895
896         switch (stage) {
897         case OBD_CLEANUP_EARLY: {
898                 int i;
899                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
900                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
901                                 continue;
902                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
903                                        OBD_CLEANUP_EARLY);
904                 }
905                 break;
906         }
907         default:
908                 break;
909         }
910
911         RETURN(rc);
912 }
913
914 static int lov_cleanup(struct obd_device *obd)
915 {
916         struct lov_obd *lov = &obd->u.lov;
917         struct list_head *pos, *tmp;
918         struct pool_desc *pool;
919         ENTRY;
920
921         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
922                 pool = list_entry(pos, struct pool_desc, pool_list);
923                 /* free pool structs */
924                 CDEBUG(D_INFO, "delete pool %p\n", pool);
925                 /* In the function below, .hs_keycmp resolves to
926                  * pool_hashkey_keycmp() */
927                 /* coverity[overrun-buffer-val] */
928                 lov_pool_del(obd, pool->pool_name);
929         }
930         cfs_hash_putref(lov->lov_pools_hash_body);
931         lov_ost_pool_free(&lov->lov_packed);
932
933         lprocfs_obd_cleanup(obd);
934         if (lov->lov_tgts) {
935                 int i;
936                 obd_getref(obd);
937                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
938                         if (!lov->lov_tgts[i])
939                                 continue;
940
941                         /* Inactive targets may never have connected */
942                         if (lov->lov_tgts[i]->ltd_active ||
943                             atomic_read(&lov->lov_refcount))
944                                 /* We should never get here - these
945                                  * should have been removed in the
946                                  * disconnect. */
947                                 CERROR("%s: lov tgt %d not cleaned! "
948                                        "deathrow=%d, lovrc=%d\n",
949                                        obd->obd_name, i, lov->lov_death_row,
950                                        atomic_read(&lov->lov_refcount));
951                         lov_del_target(obd, i, 0, 0);
952                 }
953                 obd_putref(obd);
954                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
955                          lov->lov_tgt_size);
956                 lov->lov_tgt_size = 0;
957         }
958         RETURN(0);
959 }
960
961 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
962                             __u32 *indexp, int *genp)
963 {
964         struct obd_uuid obd_uuid;
965         int cmd;
966         int rc = 0;
967         ENTRY;
968
969         switch(cmd = lcfg->lcfg_command) {
970         case LCFG_LOV_ADD_OBD:
971         case LCFG_LOV_ADD_INA:
972         case LCFG_LOV_DEL_OBD: {
973                 __u32 index;
974                 int gen;
975                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
976                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
977                         GOTO(out, rc = -EINVAL);
978
979                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
980
981                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", indexp) != 1)
982                         GOTO(out, rc = -EINVAL);
983                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
984                         GOTO(out, rc = -EINVAL);
985                 index = *indexp;
986                 gen = *genp;
987                 if (cmd == LCFG_LOV_ADD_OBD)
988                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
989                 else if (cmd == LCFG_LOV_ADD_INA)
990                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
991                 else
992                         rc = lov_del_target(obd, index, &obd_uuid, gen);
993                 GOTO(out, rc);
994         }
995         case LCFG_PARAM: {
996                 struct lov_desc *desc = &(obd->u.lov.desc);
997
998                 if (!desc)
999                         GOTO(out, rc = -EINVAL);
1000
1001                 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
1002                                                   lcfg, obd);
1003                 if (rc > 0)
1004                         rc = 0;
1005                 GOTO(out, rc);
1006         }
1007         case LCFG_POOL_NEW:
1008         case LCFG_POOL_ADD:
1009         case LCFG_POOL_DEL:
1010         case LCFG_POOL_REM:
1011                 GOTO(out, rc);
1012
1013         default: {
1014                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1015                 GOTO(out, rc = -EINVAL);
1016
1017         }
1018         }
1019 out:
1020         RETURN(rc);
1021 }
1022
1023 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1024                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1025 {
1026         struct lov_stripe_md *obj_mdp, *lsm;
1027         struct lov_obd *lov = &exp->exp_obd->u.lov;
1028         unsigned ost_idx;
1029         int rc, i;
1030         ENTRY;
1031
1032         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1033                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1034
1035         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1036         if (obj_mdp == NULL)
1037                 RETURN(-ENOMEM);
1038
1039         ost_idx = src_oa->o_nlink;
1040         lsm = *ea;
1041         if (lsm == NULL)
1042                 GOTO(out, rc = -EINVAL);
1043         if (ost_idx >= lov->desc.ld_tgt_count ||
1044             !lov->lov_tgts[ost_idx])
1045                 GOTO(out, rc = -EINVAL);
1046
1047         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1048                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1049                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1050                                         ostid_id(&src_oa->o_oi))
1051                                 GOTO(out, rc = -EINVAL);
1052                         break;
1053                 }
1054         }
1055         if (i == lsm->lsm_stripe_count)
1056                 GOTO(out, rc = -EINVAL);
1057
1058         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1059                         src_oa, &obj_mdp, oti);
1060 out:
1061         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1062         RETURN(rc);
1063 }
1064
1065 /* the LOV expects oa->o_id to be set to the LOV object id */
1066 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1067                       struct obdo *src_oa, struct lov_stripe_md **ea,
1068                       struct obd_trans_info *oti)
1069 {
1070         struct lov_obd *lov;
1071         int rc = 0;
1072         ENTRY;
1073
1074         LASSERT(ea != NULL);
1075         if (exp == NULL)
1076                 RETURN(-EINVAL);
1077
1078         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1079             src_oa->o_flags == OBD_FL_DELORPHAN) {
1080                 /* should be used with LOV anymore */
1081                 LBUG();
1082         }
1083
1084         lov = &exp->exp_obd->u.lov;
1085         if (!lov->desc.ld_active_tgt_count)
1086                 RETURN(-EIO);
1087
1088         obd_getref(exp->exp_obd);
1089         /* Recreate a specific object id at the given OST index */
1090         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1091             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1092                  rc = lov_recreate(exp, src_oa, ea, oti);
1093         }
1094
1095         obd_putref(exp->exp_obd);
1096         RETURN(rc);
1097 }
1098
1099 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1100 do {                                                                            \
1101         LASSERT((lsmp) != NULL);                                                \
1102         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1103                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1104                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1105 } while (0)
1106
1107 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1108                        struct obdo *oa, struct lov_stripe_md *lsm,
1109                        struct obd_trans_info *oti, struct obd_export *md_exp,
1110                        void *capa)
1111 {
1112         struct lov_request_set *set;
1113         struct obd_info oinfo;
1114         struct lov_request *req;
1115         struct list_head *pos;
1116         struct lov_obd *lov;
1117         int rc = 0, err = 0;
1118         ENTRY;
1119
1120         ASSERT_LSM_MAGIC(lsm);
1121
1122         if (!exp || !exp->exp_obd)
1123                 RETURN(-ENODEV);
1124
1125         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1126                 LASSERT(oti);
1127                 LASSERT(oti->oti_logcookies);
1128         }
1129
1130         lov = &exp->exp_obd->u.lov;
1131         obd_getref(exp->exp_obd);
1132         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1133         if (rc)
1134                 GOTO(out, rc);
1135
1136         list_for_each(pos, &set->set_list) {
1137                 req = list_entry(pos, struct lov_request, rq_link);
1138
1139                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1140                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1141
1142                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1143                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1144                 err = lov_update_common_set(set, req, err);
1145                 if (err) {
1146                         CERROR("%s: destroying objid "DOSTID" subobj "
1147                                DOSTID" on OST idx %d: rc = %d\n",
1148                                exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1149                                POSTID(&req->rq_oi.oi_oa->o_oi),
1150                                req->rq_idx, err);
1151                         if (!rc)
1152                                 rc = err;
1153                 }
1154         }
1155
1156         if (rc == 0) {
1157                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1158                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1159         }
1160         err = lov_fini_destroy_set(set);
1161 out:
1162         obd_putref(exp->exp_obd);
1163         RETURN(rc ? rc : err);
1164 }
1165
1166 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1167                                  void *data, int rc)
1168 {
1169         struct lov_request_set *lovset = (struct lov_request_set *)data;
1170         int err;
1171         ENTRY;
1172
1173         /* don't do attribute merge if this aysnc op failed */
1174         if (rc)
1175                 atomic_set(&lovset->set_completes, 0);
1176         err = lov_fini_getattr_set(lovset);
1177         RETURN(rc ? rc : err);
1178 }
1179
1180 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1181                               struct ptlrpc_request_set *rqset)
1182 {
1183         struct lov_request_set *lovset;
1184         struct lov_obd *lov;
1185         struct list_head *pos;
1186         struct lov_request *req;
1187         int rc = 0, err;
1188         ENTRY;
1189
1190         LASSERT(oinfo);
1191         ASSERT_LSM_MAGIC(oinfo->oi_md);
1192
1193         if (!exp || !exp->exp_obd)
1194                 RETURN(-ENODEV);
1195
1196         lov = &exp->exp_obd->u.lov;
1197
1198         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1199         if (rc)
1200                 RETURN(rc);
1201
1202         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1203                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1204                oinfo->oi_md->lsm_stripe_size);
1205
1206         list_for_each(pos, &lovset->set_list) {
1207                 req = list_entry(pos, struct lov_request, rq_link);
1208
1209                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1210                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1211                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1212                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1213                                        &req->rq_oi, rqset);
1214                 if (rc) {
1215                         CERROR("%s: getattr objid "DOSTID" subobj"
1216                                DOSTID" on OST idx %d: rc = %d\n",
1217                                exp->exp_obd->obd_name,
1218                                POSTID(&oinfo->oi_oa->o_oi),
1219                                POSTID(&req->rq_oi.oi_oa->o_oi),
1220                                req->rq_idx, rc);
1221                         GOTO(out, rc);
1222                 }
1223         }
1224
1225         if (!list_empty(&rqset->set_requests)) {
1226                 LASSERT(rc == 0);
1227                 LASSERT (rqset->set_interpret == NULL);
1228                 rqset->set_interpret = lov_getattr_interpret;
1229                 rqset->set_arg = (void *)lovset;
1230                 RETURN(rc);
1231         }
1232 out:
1233         if (rc)
1234                 atomic_set(&lovset->set_completes, 0);
1235         err = lov_fini_getattr_set(lovset);
1236         RETURN(rc ? rc : err);
1237 }
1238
1239 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1240                                  void *data, int rc)
1241 {
1242         struct lov_request_set *lovset = (struct lov_request_set *)data;
1243         int err;
1244         ENTRY;
1245
1246         if (rc)
1247                 atomic_set(&lovset->set_completes, 0);
1248         err = lov_fini_setattr_set(lovset);
1249         RETURN(rc ? rc : err);
1250 }
1251
1252 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1253    needed. Otherwise, a client is waiting for responses. */
1254 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1255                              struct obd_trans_info *oti,
1256                              struct ptlrpc_request_set *rqset)
1257 {
1258         struct lov_request_set *set;
1259         struct lov_request *req;
1260         struct list_head *pos;
1261         struct lov_obd *lov;
1262         int rc = 0;
1263         ENTRY;
1264
1265         LASSERT(oinfo);
1266         ASSERT_LSM_MAGIC(oinfo->oi_md);
1267         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1268                 LASSERT(oti);
1269                 LASSERT(oti->oti_logcookies);
1270         }
1271
1272         if (!exp || !exp->exp_obd)
1273                 RETURN(-ENODEV);
1274
1275         lov = &exp->exp_obd->u.lov;
1276         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1277         if (rc)
1278                 RETURN(rc);
1279
1280         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1281                POSTID(&oinfo->oi_md->lsm_oi),
1282                oinfo->oi_md->lsm_stripe_count,
1283                oinfo->oi_md->lsm_stripe_size);
1284
1285         list_for_each(pos, &set->set_list) {
1286                 req = list_entry(pos, struct lov_request, rq_link);
1287
1288                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1289                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1290
1291                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1292                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1293                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1294
1295                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1296                                        &req->rq_oi, oti, rqset);
1297                 if (rc) {
1298                         CERROR("error: setattr objid "DOSTID" subobj"
1299                                DOSTID" on OST idx %d: rc = %d\n",
1300                                POSTID(&set->set_oi->oi_oa->o_oi),
1301                                POSTID(&req->rq_oi.oi_oa->o_oi),
1302                                req->rq_idx, rc);
1303                         break;
1304                 }
1305         }
1306
1307         /* If we are not waiting for responses on async requests, return. */
1308         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1309                 int err;
1310                 if (rc)
1311                         atomic_set(&set->set_completes, 0);
1312                 err = lov_fini_setattr_set(set);
1313                 RETURN(rc ? rc : err);
1314         }
1315
1316         LASSERT(rqset->set_interpret == NULL);
1317         rqset->set_interpret = lov_setattr_interpret;
1318         rqset->set_arg = (void *)set;
1319
1320         RETURN(0);
1321 }
1322
1323 static int lov_change_cbdata(struct obd_export *exp,
1324                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1325                              void *data)
1326 {
1327         struct lov_obd *lov;
1328         int rc = 0, i;
1329         ENTRY;
1330
1331         ASSERT_LSM_MAGIC(lsm);
1332
1333         if (!exp || !exp->exp_obd)
1334                 RETURN(-ENODEV);
1335
1336         lov = &exp->exp_obd->u.lov;
1337         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1338                 struct lov_stripe_md submd;
1339                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1340
1341                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1342                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1343                         continue;
1344                 }
1345
1346                 submd.lsm_oi = loi->loi_oi;
1347                 submd.lsm_stripe_count = 0;
1348                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1349                                        &submd, it, data);
1350         }
1351         RETURN(rc);
1352 }
1353
1354 /* find any ldlm lock of the inode in lov
1355  * return 0    not find
1356  *        1    find one
1357  *      < 0    error */
1358 static int lov_find_cbdata(struct obd_export *exp,
1359                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1360                            void *data)
1361 {
1362         struct lov_obd *lov;
1363         int rc = 0, i;
1364         ENTRY;
1365
1366         ASSERT_LSM_MAGIC(lsm);
1367
1368         if (!exp || !exp->exp_obd)
1369                 RETURN(-ENODEV);
1370
1371         lov = &exp->exp_obd->u.lov;
1372         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1373                 struct lov_stripe_md submd;
1374                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1375
1376                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1377                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1378                         continue;
1379                 }
1380                 submd.lsm_oi = loi->loi_oi;
1381                 submd.lsm_stripe_count = 0;
1382                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1383                                      &submd, it, data);
1384                 if (rc != 0)
1385                         RETURN(rc);
1386         }
1387         RETURN(rc);
1388 }
1389
1390 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1391 {
1392         struct lov_request_set *lovset = (struct lov_request_set *)data;
1393         int err;
1394         ENTRY;
1395
1396         if (rc)
1397                 atomic_set(&lovset->set_completes, 0);
1398
1399         err = lov_fini_statfs_set(lovset);
1400         RETURN(rc ? rc : err);
1401 }
1402
1403 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1404                             __u64 max_age, struct ptlrpc_request_set *rqset)
1405 {
1406         struct obd_device      *obd = class_exp2obd(exp);
1407         struct lov_request_set *set;
1408         struct lov_request *req;
1409         struct list_head *pos;
1410         struct lov_obd *lov;
1411         int rc = 0;
1412         ENTRY;
1413
1414         LASSERT(oinfo != NULL);
1415         LASSERT(oinfo->oi_osfs != NULL);
1416
1417         lov = &obd->u.lov;
1418         rc = lov_prep_statfs_set(obd, oinfo, &set);
1419         if (rc)
1420                 RETURN(rc);
1421
1422         list_for_each(pos, &set->set_list) {
1423                 req = list_entry(pos, struct lov_request, rq_link);
1424                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1425                                       &req->rq_oi, max_age, rqset);
1426                 if (rc)
1427                         break;
1428         }
1429
1430         if (rc || list_empty(&rqset->set_requests)) {
1431                 int err;
1432                 if (rc)
1433                         atomic_set(&set->set_completes, 0);
1434                 err = lov_fini_statfs_set(set);
1435                 RETURN(rc ? rc : err);
1436         }
1437
1438         LASSERT(rqset->set_interpret == NULL);
1439         rqset->set_interpret = lov_statfs_interpret;
1440         rqset->set_arg = (void *)set;
1441         RETURN(0);
1442 }
1443
1444 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1445                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1446 {
1447         struct ptlrpc_request_set *set = NULL;
1448         struct obd_info oinfo = { { { 0 } } };
1449         int rc = 0;
1450         ENTRY;
1451
1452
1453         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1454          * statfs requests */
1455         set = ptlrpc_prep_set();
1456         if (set == NULL)
1457                 RETURN(-ENOMEM);
1458
1459         oinfo.oi_osfs = osfs;
1460         oinfo.oi_flags = flags;
1461         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1462         if (rc == 0)
1463                 rc = ptlrpc_set_wait(set);
1464         ptlrpc_set_destroy(set);
1465
1466         RETURN(rc);
1467 }
1468
1469 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1470                          void *karg, void *uarg)
1471 {
1472         struct obd_device *obddev = class_exp2obd(exp);
1473         struct lov_obd *lov = &obddev->u.lov;
1474         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1475         struct obd_uuid *uuidp;
1476         ENTRY;
1477
1478         switch (cmd) {
1479         case IOC_OBD_STATFS: {
1480                 struct obd_ioctl_data *data = karg;
1481                 struct obd_device *osc_obd;
1482                 struct obd_statfs stat_buf = {0};
1483                 __u32 index;
1484                 __u32 flags;
1485
1486                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1487                 if ((index >= count))
1488                         RETURN(-ENODEV);
1489
1490                 if (!lov->lov_tgts[index])
1491                         /* Try again with the next index */
1492                         RETURN(-EAGAIN);
1493                 if (!lov->lov_tgts[index]->ltd_active)
1494                         RETURN(-ENODATA);
1495
1496                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1497                 if (!osc_obd)
1498                         RETURN(-EINVAL);
1499
1500                 /* copy UUID */
1501                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1502                                  min((int)data->ioc_plen2,
1503                                      (int)sizeof(struct obd_uuid))))
1504                         RETURN(-EFAULT);
1505
1506                 flags = uarg ? *(__u32*)uarg : 0;
1507                 /* got statfs data */
1508                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1509                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1510                                 flags);
1511                 if (rc)
1512                         RETURN(rc);
1513                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1514                                      min((int) data->ioc_plen1,
1515                                          (int) sizeof(stat_buf))))
1516                         RETURN(-EFAULT);
1517                 break;
1518         }
1519         case OBD_IOC_LOV_GET_CONFIG: {
1520                 struct obd_ioctl_data *data;
1521                 struct lov_desc *desc;
1522                 char *buf = NULL;
1523                 __u32 *genp;
1524
1525                 len = 0;
1526                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1527                         RETURN(-EINVAL);
1528
1529                 data = (struct obd_ioctl_data *)buf;
1530
1531                 if (sizeof(*desc) > data->ioc_inllen1) {
1532                         obd_ioctl_freedata(buf, len);
1533                         RETURN(-EINVAL);
1534                 }
1535
1536                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1537                         obd_ioctl_freedata(buf, len);
1538                         RETURN(-EINVAL);
1539                 }
1540
1541                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1542                         obd_ioctl_freedata(buf, len);
1543                         RETURN(-EINVAL);
1544                 }
1545
1546                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1547                 memcpy(desc, &(lov->desc), sizeof(*desc));
1548
1549                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1550                 genp = (__u32 *)data->ioc_inlbuf3;
1551                 /* the uuid will be empty for deleted OSTs */
1552                 for (i = 0; i < count; i++, uuidp++, genp++) {
1553                         if (!lov->lov_tgts[i])
1554                                 continue;
1555                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1556                         *genp = lov->lov_tgts[i]->ltd_gen;
1557                 }
1558
1559                 if (copy_to_user((void *)uarg, buf, len))
1560                         rc = -EFAULT;
1561                 obd_ioctl_freedata(buf, len);
1562                 break;
1563         }
1564         case LL_IOC_LOV_GETSTRIPE:
1565                 rc = lov_getstripe(exp, karg, uarg);
1566                 break;
1567         case OBD_IOC_QUOTACTL: {
1568                 struct if_quotactl *qctl = karg;
1569                 struct lov_tgt_desc *tgt = NULL;
1570                 struct obd_quotactl *oqctl;
1571
1572                 if (qctl->qc_valid == QC_OSTIDX) {
1573                         if (count <= qctl->qc_idx)
1574                                 RETURN(-EINVAL);
1575
1576                         tgt = lov->lov_tgts[qctl->qc_idx];
1577                         if (!tgt || !tgt->ltd_exp)
1578                                 RETURN(-EINVAL);
1579                 } else if (qctl->qc_valid == QC_UUID) {
1580                         for (i = 0; i < count; i++) {
1581                                 tgt = lov->lov_tgts[i];
1582                                 if (!tgt ||
1583                                     !obd_uuid_equals(&tgt->ltd_uuid,
1584                                                      &qctl->obd_uuid))
1585                                         continue;
1586
1587                                 if (tgt->ltd_exp == NULL)
1588                                         RETURN(-EINVAL);
1589
1590                                 break;
1591                         }
1592                 } else {
1593                         RETURN(-EINVAL);
1594                 }
1595
1596                 if (i >= count)
1597                         RETURN(-EAGAIN);
1598
1599                 LASSERT(tgt && tgt->ltd_exp);
1600                 OBD_ALLOC_PTR(oqctl);
1601                 if (!oqctl)
1602                         RETURN(-ENOMEM);
1603
1604                 QCTL_COPY(oqctl, qctl);
1605                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1606                 if (rc == 0) {
1607                         QCTL_COPY(qctl, oqctl);
1608                         qctl->qc_valid = QC_OSTIDX;
1609                         qctl->obd_uuid = tgt->ltd_uuid;
1610                 }
1611                 OBD_FREE_PTR(oqctl);
1612                 break;
1613         }
1614         default: {
1615                 int set = 0;
1616
1617                 if (count == 0)
1618                         RETURN(-ENOTTY);
1619
1620                 for (i = 0; i < count; i++) {
1621                         int err;
1622                         struct obd_device *osc_obd;
1623
1624                         /* OST was disconnected */
1625                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1626                                 continue;
1627
1628                         /* ll_umount_begin() sets force flag but for lov, not
1629                          * osc. Let's pass it through */
1630                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1631                         osc_obd->obd_force = obddev->obd_force;
1632                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1633                                             len, karg, uarg);
1634                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1635                                 RETURN(err);
1636                         } else if (err) {
1637                                 if (lov->lov_tgts[i]->ltd_active) {
1638                                         CDEBUG(err == -ENOTTY ?
1639                                                D_IOCTL : D_WARNING,
1640                                                "iocontrol OSC %s on OST "
1641                                                "idx %d cmd %x: err = %d\n",
1642                                                lov_uuid2str(lov, i),
1643                                                i, cmd, err);
1644                                         if (!rc)
1645                                                 rc = err;
1646                                 }
1647                         } else {
1648                                 set = 1;
1649                         }
1650                 }
1651                 if (!set && !rc)
1652                         rc = -EIO;
1653         }
1654         }
1655
1656         RETURN(rc);
1657 }
1658
1659 #define FIEMAP_BUFFER_SIZE 4096
1660
1661 /**
1662  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1663  * call. The local end offset and the device are sent in the first
1664  * fm_extent. This function calculates the stripe number from the index.
1665  * This function returns a stripe_no on which mapping is to be restarted.
1666  *
1667  * This function returns fm_end_offset which is the in-OST offset at which
1668  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1669  * will re-calculate proper offset in next stripe.
1670  * Note that the first extent is passed to lov_get_info via the value field.
1671  *
1672  * \param fiemap fiemap request header
1673  * \param lsm striping information for the file
1674  * \param fm_start logical start of mapping
1675  * \param fm_end logical end of mapping
1676  * \param start_stripe starting stripe will be returned in this
1677  */
1678 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1679                                    struct lov_stripe_md *lsm, obd_size fm_start,
1680                                    obd_size fm_end, int *start_stripe)
1681 {
1682         obd_size local_end = fiemap->fm_extents[0].fe_logical;
1683         obd_off lun_start, lun_end;
1684         obd_size fm_end_offset;
1685         int stripe_no = -1, i;
1686
1687         if (fiemap->fm_extent_count == 0 ||
1688             fiemap->fm_extents[0].fe_logical == 0)
1689                 return 0;
1690
1691         /* Find out stripe_no from ost_index saved in the fe_device */
1692         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1693                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
1694                                         fiemap->fm_extents[0].fe_device) {
1695                         stripe_no = i;
1696                         break;
1697                 }
1698         }
1699         if (stripe_no == -1)
1700                 return -EINVAL;
1701
1702         /* If we have finished mapping on previous device, shift logical
1703          * offset to start of next device */
1704         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1705                                    &lun_start, &lun_end)) != 0 &&
1706                                    local_end < lun_end) {
1707                 fm_end_offset = local_end;
1708                 *start_stripe = stripe_no;
1709         } else {
1710                 /* This is a special value to indicate that caller should
1711                  * calculate offset in next stripe. */
1712                 fm_end_offset = 0;
1713                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1714         }
1715
1716         return fm_end_offset;
1717 }
1718
1719 /**
1720  * We calculate on which OST the mapping will end. If the length of mapping
1721  * is greater than (stripe_size * stripe_count) then the last_stripe will
1722  * will be one just before start_stripe. Else we check if the mapping
1723  * intersects each OST and find last_stripe.
1724  * This function returns the last_stripe and also sets the stripe_count
1725  * over which the mapping is spread
1726  *
1727  * \param lsm striping information for the file
1728  * \param fm_start logical start of mapping
1729  * \param fm_end logical end of mapping
1730  * \param start_stripe starting stripe of the mapping
1731  * \param stripe_count the number of stripes across which to map is returned
1732  *
1733  * \retval last_stripe return the last stripe of the mapping
1734  */
1735 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1736                             obd_size fm_end, int start_stripe,
1737                             int *stripe_count)
1738 {
1739         int last_stripe;
1740         obd_off obd_start, obd_end;
1741         int i, j;
1742
1743         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1744                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1745                                                               start_stripe - 1);
1746                 *stripe_count = lsm->lsm_stripe_count;
1747         } else {
1748                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1749                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
1750                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1751                                                    &obd_start, &obd_end)) == 0)
1752                                 break;
1753                 }
1754                 *stripe_count = j;
1755                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1756         }
1757
1758         return last_stripe;
1759 }
1760
1761 /**
1762  * Set fe_device and copy extents from local buffer into main return buffer.
1763  *
1764  * \param fiemap fiemap request header
1765  * \param lcl_fm_ext array of local fiemap extents to be copied
1766  * \param ost_index OST index to be written into the fm_device field for each
1767                     extent
1768  * \param ext_count number of extents to be copied
1769  * \param current_extent where to start copying in main extent array
1770  */
1771 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1772                                   struct ll_fiemap_extent *lcl_fm_ext,
1773                                   int ost_index, unsigned int ext_count,
1774                                   int current_extent)
1775 {
1776         char *to;
1777         int ext;
1778
1779         for (ext = 0; ext < ext_count; ext++) {
1780                 lcl_fm_ext[ext].fe_device = ost_index;
1781                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1782         }
1783
1784         /* Copy fm_extent's from fm_local to return buffer */
1785         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1786         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1787 }
1788
1789 /**
1790  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1791  * This also handles the restarting of FIEMAP calls in case mapping overflows
1792  * the available number of extents in single call.
1793  */
1794 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1795                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1796 {
1797         struct ll_fiemap_info_key *fm_key = key;
1798         struct ll_user_fiemap *fiemap = val;
1799         struct ll_user_fiemap *fm_local = NULL;
1800         struct ll_fiemap_extent *lcl_fm_ext;
1801         int count_local;
1802         unsigned int get_num_extents = 0;
1803         int ost_index = 0, actual_start_stripe, start_stripe;
1804         obd_size fm_start, fm_end, fm_length, fm_end_offset;
1805         obd_size curr_loc;
1806         int current_extent = 0, rc = 0, i;
1807         int ost_eof = 0; /* EOF for object */
1808         int ost_done = 0; /* done with required mapping for this OST? */
1809         int last_stripe;
1810         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1811         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1812
1813         if (!lsm_has_objects(lsm)) {
1814                 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1815                     fm_key->oa.o_size)) {
1816                         /* released file, return a minimal FIEMAP if
1817                          * request fits in file-size.
1818                          */
1819                         fiemap->fm_mapped_extents = 1;
1820                         fiemap->fm_extents[0].fe_logical =
1821                                                 fm_key->fiemap.fm_start;
1822                         if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1823                             fm_key->oa.o_size)
1824                                 fiemap->fm_extents[0].fe_length =
1825                                                 fm_key->fiemap.fm_length;
1826                         else
1827                                 fiemap->fm_extents[0].fe_length =
1828                                                 fm_key->oa.o_size -
1829                                                 fm_key->fiemap.fm_start;
1830                         fiemap->fm_extents[0].fe_flags |=
1831                                                 (FIEMAP_EXTENT_UNKNOWN |
1832                                                  FIEMAP_EXTENT_LAST);
1833                 }
1834                 GOTO(out, rc = 0);
1835         }
1836
1837         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1838                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1839
1840         OBD_ALLOC_LARGE(fm_local, buffer_size);
1841         if (fm_local == NULL)
1842                 GOTO(out, rc = -ENOMEM);
1843         lcl_fm_ext = &fm_local->fm_extents[0];
1844
1845         count_local = fiemap_size_to_count(buffer_size);
1846
1847         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1848         fm_start = fiemap->fm_start;
1849         fm_length = fiemap->fm_length;
1850         /* Calculate start stripe, last stripe and length of mapping */
1851         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1852         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1853                                                 fm_start + fm_length - 1);
1854         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1855         if (fm_end > fm_key->oa.o_size)
1856                 fm_end = fm_key->oa.o_size;
1857
1858         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1859                                             actual_start_stripe, &stripe_count);
1860
1861         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1862                                                   fm_end, &start_stripe);
1863         if (fm_end_offset == -EINVAL)
1864                 GOTO(out, rc = -EINVAL);
1865
1866         if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1867                 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1868         if (fiemap->fm_extent_count == 0) {
1869                 get_num_extents = 1;
1870                 count_local = 0;
1871         }
1872         /* Check each stripe */
1873         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1874              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1875                 obd_size req_fm_len; /* Stores length of required mapping */
1876                 obd_size len_mapped_single_call;
1877                 obd_off lun_start, lun_end, obd_object_end;
1878                 unsigned int ext_count;
1879
1880                 cur_stripe_wrap = cur_stripe;
1881
1882                 /* Find out range of mapping on this stripe */
1883                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1884                                            &lun_start, &obd_object_end)) == 0)
1885                         continue;
1886
1887                 /* If this is a continuation FIEMAP call and we are on
1888                  * starting stripe then lun_start needs to be set to
1889                  * fm_end_offset */
1890                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1891                         lun_start = fm_end_offset;
1892
1893                 if (fm_length != ~0ULL) {
1894                         /* Handle fm_start + fm_length overflow */
1895                         if (fm_start + fm_length < fm_start)
1896                                 fm_length = ~0ULL - fm_start;
1897                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1898                                                      cur_stripe);
1899                 } else {
1900                         lun_end = ~0ULL;
1901                 }
1902
1903                 if (lun_start == lun_end)
1904                         continue;
1905
1906                 req_fm_len = obd_object_end - lun_start;
1907                 fm_local->fm_length = 0;
1908                 len_mapped_single_call = 0;
1909
1910                 /* If the output buffer is very large and the objects have many
1911                  * extents we may need to loop on a single OST repeatedly */
1912                 ost_eof = 0;
1913                 ost_done = 0;
1914                 do {
1915                         if (get_num_extents == 0) {
1916                                 /* Don't get too many extents. */
1917                                 if (current_extent + count_local >
1918                                     fiemap->fm_extent_count)
1919                                         count_local = fiemap->fm_extent_count -
1920                                                                  current_extent;
1921                         }
1922
1923                         lun_start += len_mapped_single_call;
1924                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
1925                         req_fm_len = fm_local->fm_length;
1926                         fm_local->fm_extent_count = count_local;
1927                         fm_local->fm_mapped_extents = 0;
1928                         fm_local->fm_flags = fiemap->fm_flags;
1929
1930                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1931                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1932
1933                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1934                                 GOTO(out, rc = -EINVAL);
1935
1936                         /* If OST is inactive, return extent with UNKNOWN flag */
1937                         if (!lov->lov_tgts[ost_index]->ltd_active) {
1938                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1939                                 fm_local->fm_mapped_extents = 1;
1940
1941                                 lcl_fm_ext[0].fe_logical = lun_start;
1942                                 lcl_fm_ext[0].fe_length = obd_object_end -
1943                                                                       lun_start;
1944                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1945
1946                                 goto inactive_tgt;
1947                         }
1948
1949                         fm_local->fm_start = lun_start;
1950                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1951                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1952                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1953                         rc = obd_get_info(NULL,
1954                                           lov->lov_tgts[ost_index]->ltd_exp,
1955                                           keylen, key, vallen, fm_local, lsm);
1956                         if (rc != 0)
1957                                 GOTO(out, rc);
1958
1959 inactive_tgt:
1960                         ext_count = fm_local->fm_mapped_extents;
1961                         if (ext_count == 0) {
1962                                 ost_done = 1;
1963                                 /* If last stripe has hole at the end,
1964                                  * then we need to return */
1965                                 if (cur_stripe_wrap == last_stripe) {
1966                                         fiemap->fm_mapped_extents = 0;
1967                                         goto finish;
1968                                 }
1969                                 break;
1970                         }
1971
1972                         /* If we just need num of extents then go to next device */
1973                         if (get_num_extents) {
1974                                 current_extent += ext_count;
1975                                 break;
1976                         }
1977
1978                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1979                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1980
1981                         /* Have we finished mapping on this device? */
1982                         if (req_fm_len <= len_mapped_single_call)
1983                                 ost_done = 1;
1984
1985                         /* Clear the EXTENT_LAST flag which can be present on
1986                          * last extent */
1987                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1988                                 lcl_fm_ext[ext_count - 1].fe_flags &=
1989                                                             ~FIEMAP_EXTENT_LAST;
1990
1991                         curr_loc = lov_stripe_size(lsm,
1992                                            lcl_fm_ext[ext_count - 1].fe_logical+
1993                                            lcl_fm_ext[ext_count - 1].fe_length,
1994                                            cur_stripe);
1995                         if (curr_loc >= fm_key->oa.o_size)
1996                                 ost_eof = 1;
1997
1998                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1999                                                      ost_index, ext_count,
2000                                                      current_extent);
2001
2002                         current_extent += ext_count;
2003
2004                         /* Ran out of available extents? */
2005                         if (current_extent >= fiemap->fm_extent_count)
2006                                 goto finish;
2007                 } while (ost_done == 0 && ost_eof == 0);
2008
2009                 if (cur_stripe_wrap == last_stripe)
2010                         goto finish;
2011         }
2012
2013 finish:
2014         /* Indicate that we are returning device offsets unless file just has
2015          * single stripe */
2016         if (lsm->lsm_stripe_count > 1)
2017                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2018
2019         if (get_num_extents)
2020                 goto skip_last_device_calc;
2021
2022         /* Check if we have reached the last stripe and whether mapping for that
2023          * stripe is done. */
2024         if (cur_stripe_wrap == last_stripe) {
2025                 if (ost_done || ost_eof)
2026                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2027                                                              FIEMAP_EXTENT_LAST;
2028         }
2029
2030 skip_last_device_calc:
2031         fiemap->fm_mapped_extents = current_extent;
2032
2033 out:
2034         if (fm_local)
2035                 OBD_FREE_LARGE(fm_local, buffer_size);
2036         return rc;
2037 }
2038
2039 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2040                         __u32 keylen, void *key, __u32 *vallen, void *val,
2041                         struct lov_stripe_md *lsm)
2042 {
2043         struct obd_device *obddev = class_exp2obd(exp);
2044         struct lov_obd *lov = &obddev->u.lov;
2045         int i, rc;
2046         ENTRY;
2047
2048         if (!vallen || !val)
2049                 RETURN(-EFAULT);
2050
2051         obd_getref(obddev);
2052
2053         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2054                 struct {
2055                         char name[16];
2056                         struct ldlm_lock *lock;
2057                 } *data = key;
2058                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2059                 struct lov_oinfo *loi;
2060                 __u32 *stripe = val;
2061
2062                 if (*vallen < sizeof(*stripe))
2063                         GOTO(out, rc = -EFAULT);
2064                 *vallen = sizeof(*stripe);
2065
2066                 /* XXX This is another one of those bits that will need to
2067                  * change if we ever actually support nested LOVs.  It uses
2068                  * the lock's export to find out which stripe it is. */
2069                 /* XXX - it's assumed all the locks for deleted OSTs have
2070                  * been cancelled. Also, the export for deleted OSTs will
2071                  * be NULL and won't match the lock's export. */
2072                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2073                         loi = lsm->lsm_oinfo[i];
2074                         if (!lov->lov_tgts[loi->loi_ost_idx])
2075                                 continue;
2076                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2077                             data->lock->l_conn_export &&
2078                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2079                                 *stripe = i;
2080                                 GOTO(out, rc = 0);
2081                         }
2082                 }
2083                 LDLM_ERROR(data->lock, "lock on inode without such object");
2084                 dump_lsm(D_ERROR, lsm);
2085                 GOTO(out, rc = -ENXIO);
2086         } else if (KEY_IS(KEY_LAST_ID)) {
2087                 struct obd_id_info *info = val;
2088                 __u32 size = sizeof(obd_id);
2089                 struct lov_tgt_desc *tgt;
2090
2091                 LASSERT(*vallen == sizeof(struct obd_id_info));
2092                 tgt = lov->lov_tgts[info->idx];
2093
2094                 if (!tgt || !tgt->ltd_active)
2095                         GOTO(out, rc = -ESRCH);
2096
2097                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2098                                   &size, info->data, NULL);
2099                 GOTO(out, rc = 0);
2100         } else if (KEY_IS(KEY_LOVDESC)) {
2101                 struct lov_desc *desc_ret = val;
2102                 *desc_ret = lov->desc;
2103
2104                 GOTO(out, rc = 0);
2105         } else if (KEY_IS(KEY_FIEMAP)) {
2106                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2107                 GOTO(out, rc);
2108         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2109                 struct lov_tgt_desc *tgt;
2110                 __u64 ost_idx = *((__u64*)val);
2111
2112                 LASSERT(*vallen == sizeof(__u64));
2113                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2114                 tgt = lov->lov_tgts[ost_idx];
2115
2116                 if (!tgt || !tgt->ltd_exp)
2117                         GOTO(out, rc = -ESRCH);
2118
2119                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2120                 GOTO(out, rc = 0);
2121         } else if (KEY_IS(KEY_TGT_COUNT)) {
2122                 *((int *)val) = lov->desc.ld_tgt_count;
2123                 GOTO(out, rc = 0);
2124         }
2125
2126         rc = -EINVAL;
2127
2128 out:
2129         obd_putref(obddev);
2130         RETURN(rc);
2131 }
2132
2133 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2134                               obd_count keylen, void *key, obd_count vallen,
2135                               void *val, struct ptlrpc_request_set *set)
2136 {
2137         struct obd_device *obddev = class_exp2obd(exp);
2138         struct lov_obd *lov = &obddev->u.lov;
2139         obd_count count;
2140         int i, rc = 0, err;
2141         struct lov_tgt_desc *tgt;
2142         unsigned incr, check_uuid,
2143                  do_inactive, no_set;
2144         unsigned next_id = 0,  mds_con = 0, capa = 0;
2145         ENTRY;
2146
2147         incr = check_uuid = do_inactive = no_set = 0;
2148         if (set == NULL) {
2149                 no_set = 1;
2150                 set = ptlrpc_prep_set();
2151                 if (!set)
2152                         RETURN(-ENOMEM);
2153         }
2154
2155         obd_getref(obddev);
2156         count = lov->desc.ld_tgt_count;
2157
2158         if (KEY_IS(KEY_NEXT_ID)) {
2159                 count = vallen / sizeof(struct obd_id_info);
2160                 vallen = sizeof(obd_id);
2161                 incr = sizeof(struct obd_id_info);
2162                 do_inactive = 1;
2163                 next_id = 1;
2164         } else if (KEY_IS(KEY_CHECKSUM)) {
2165                 do_inactive = 1;
2166         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2167                 /* use defaults:  do_inactive = incr = 0; */
2168         } else if (KEY_IS(KEY_MDS_CONN)) {
2169                 mds_con = 1;
2170         } else if (KEY_IS(KEY_CAPA_KEY)) {
2171                 capa = 1;
2172         } else if (KEY_IS(KEY_CACHE_SET)) {
2173                 LASSERT(lov->lov_cache == NULL);
2174                 lov->lov_cache = val;
2175                 do_inactive = 1;
2176         }
2177
2178         for (i = 0; i < count; i++, val = (char *)val + incr) {
2179                 if (next_id) {
2180                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2181                 } else {
2182                         tgt = lov->lov_tgts[i];
2183                 }
2184                 /* OST was disconnected */
2185                 if (!tgt || !tgt->ltd_exp)
2186                         continue;
2187
2188                 /* OST is inactive and we don't want inactive OSCs */
2189                 if (!tgt->ltd_active && !do_inactive)
2190                         continue;
2191
2192                 if (mds_con) {
2193                         struct mds_group_info *mgi;
2194
2195                         LASSERT(vallen == sizeof(*mgi));
2196                         mgi = (struct mds_group_info *)val;
2197
2198                         /* Only want a specific OSC */
2199                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2200                                                 &tgt->ltd_uuid))
2201                                 continue;
2202
2203                         err = obd_set_info_async(env, tgt->ltd_exp,
2204                                          keylen, key, sizeof(int),
2205                                          &mgi->group, set);
2206                 } else if (next_id) {
2207                         err = obd_set_info_async(env, tgt->ltd_exp,
2208                                          keylen, key, vallen,
2209                                          ((struct obd_id_info*)val)->data, set);
2210                 } else if (capa) {
2211                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2212
2213                         LASSERT(vallen == sizeof(*info));
2214
2215                          /* Only want a specific OSC */
2216                         if (info->uuid &&
2217                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2218                                 continue;
2219
2220                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2221                                                  key, sizeof(*info->capa),
2222                                                  info->capa, set);
2223                 } else {
2224                         /* Only want a specific OSC */
2225                         if (check_uuid &&
2226                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2227                                 continue;
2228
2229                         err = obd_set_info_async(env, tgt->ltd_exp,
2230                                          keylen, key, vallen, val, set);
2231                 }
2232
2233                 if (!rc)
2234                         rc = err;
2235         }
2236
2237         obd_putref(obddev);
2238         if (no_set) {
2239                 err = ptlrpc_set_wait(set);
2240                 if (!rc)
2241                         rc = err;
2242                 ptlrpc_set_destroy(set);
2243         }
2244         RETURN(rc);
2245 }
2246
2247 void lov_stripe_lock(struct lov_stripe_md *md)
2248 {
2249         LASSERT(md->lsm_lock_owner != current_pid());
2250         spin_lock(&md->lsm_lock);
2251         LASSERT(md->lsm_lock_owner == 0);
2252         md->lsm_lock_owner = current_pid();
2253 }
2254
2255 void lov_stripe_unlock(struct lov_stripe_md *md)
2256 {
2257         LASSERT(md->lsm_lock_owner == current_pid());
2258         md->lsm_lock_owner = 0;
2259         spin_unlock(&md->lsm_lock);
2260 }
2261
2262 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2263                         struct obd_quotactl *oqctl)
2264 {
2265         struct lov_obd      *lov = &obd->u.lov;
2266         struct lov_tgt_desc *tgt;
2267         __u64                curspace = 0;
2268         __u64                bhardlimit = 0;
2269         int                  i, rc = 0;
2270         ENTRY;
2271
2272         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2273             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2274             oqctl->qc_cmd != Q_GETOQUOTA &&
2275             oqctl->qc_cmd != Q_INITQUOTA &&
2276             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2277             oqctl->qc_cmd != Q_FINVALIDATE) {
2278                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2279                 RETURN(-EFAULT);
2280         }
2281
2282         /* for lov tgt */
2283         obd_getref(obd);
2284         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2285                 int err;
2286
2287                 tgt = lov->lov_tgts[i];
2288
2289                 if (!tgt)
2290                         continue;
2291
2292                 if (!tgt->ltd_active || tgt->ltd_reap) {
2293                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2294                             lov->lov_tgts[i]->ltd_activate) {
2295                                 rc = -ENETDOWN;
2296                                 CERROR("ost %d is inactive\n", i);
2297                         } else {
2298                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2299                         }
2300                         continue;
2301                 }
2302
2303                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2304                 if (err) {
2305                         if (tgt->ltd_active && !rc)
2306                                 rc = err;
2307                         continue;
2308                 }
2309
2310                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2311                         curspace += oqctl->qc_dqblk.dqb_curspace;
2312                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2313                 }
2314         }
2315         obd_putref(obd);
2316
2317         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2318                 oqctl->qc_dqblk.dqb_curspace = curspace;
2319                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2320         }
2321         RETURN(rc);
2322 }
2323
2324 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2325                           struct obd_quotactl *oqctl)
2326 {
2327         struct lov_obd *lov = &obd->u.lov;
2328         int             i, rc = 0;
2329         ENTRY;
2330
2331         obd_getref(obd);
2332
2333         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2334                 if (!lov->lov_tgts[i])
2335                         continue;
2336
2337                 /* Skip quota check on the administratively disabled OSTs. */
2338                 if (!lov->lov_tgts[i]->ltd_activate) {
2339                         CWARN("lov idx %d was administratively disabled, "
2340                               "skip quotacheck on it.\n", i);
2341                         continue;
2342                 }
2343
2344                 if (!lov->lov_tgts[i]->ltd_active) {
2345                         CERROR("lov idx %d inactive\n", i);
2346                         rc = -EIO;
2347                         goto out;
2348                 }
2349         }
2350
2351         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2352                 int err;
2353
2354                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2355                         continue;
2356
2357                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2358                 if (err && !rc)
2359                         rc = err;
2360         }
2361
2362 out:
2363         obd_putref(obd);
2364
2365         RETURN(rc);
2366 }
2367
2368 static struct obd_ops lov_obd_ops = {
2369         .o_owner                = THIS_MODULE,
2370         .o_setup                = lov_setup,
2371         .o_precleanup           = lov_precleanup,
2372         .o_cleanup              = lov_cleanup,
2373         .o_connect              = lov_connect,
2374         .o_disconnect           = lov_disconnect,
2375         .o_statfs               = lov_statfs,
2376         .o_statfs_async         = lov_statfs_async,
2377         .o_packmd               = lov_packmd,
2378         .o_unpackmd             = lov_unpackmd,
2379         .o_create               = lov_create,
2380         .o_destroy              = lov_destroy,
2381         .o_getattr_async        = lov_getattr_async,
2382         .o_setattr_async        = lov_setattr_async,
2383         .o_change_cbdata        = lov_change_cbdata,
2384         .o_find_cbdata          = lov_find_cbdata,
2385         .o_iocontrol            = lov_iocontrol,
2386         .o_get_info             = lov_get_info,
2387         .o_set_info_async       = lov_set_info_async,
2388         .o_notify               = lov_notify,
2389         .o_pool_new             = lov_pool_new,
2390         .o_pool_rem             = lov_pool_remove,
2391         .o_pool_add             = lov_pool_add,
2392         .o_pool_del             = lov_pool_del,
2393         .o_getref               = lov_getref,
2394         .o_putref               = lov_putref,
2395         .o_quotactl             = lov_quotactl,
2396         .o_quotacheck           = lov_quotacheck,
2397 };
2398
2399 struct kmem_cache *lov_oinfo_slab;
2400
2401 extern struct lu_kmem_descr lov_caches[];
2402
2403 int __init lov_init(void)
2404 {
2405         bool enable_proc = true;
2406         struct obd_type *type;
2407         int rc;
2408         ENTRY;
2409
2410         /* print an address of _any_ initialized kernel symbol from this
2411          * module, to allow debugging with gdb that doesn't support data
2412          * symbols from modules.*/
2413         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2414
2415         rc = lu_kmem_init(lov_caches);
2416         if (rc)
2417                 return rc;
2418
2419         lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2420                                            sizeof(struct lov_oinfo), 0,
2421                                            SLAB_HWCACHE_ALIGN, NULL);
2422         if (lov_oinfo_slab == NULL) {
2423                 lu_kmem_fini(lov_caches);
2424                 return -ENOMEM;
2425         }
2426
2427         type = class_search_type(LUSTRE_LOD_NAME);
2428         if (type != NULL && type->typ_procsym != NULL)
2429                 enable_proc = false;
2430
2431         rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
2432 #ifndef HAVE_ONLY_PROCFS_SEQ
2433                                  NULL,
2434 #endif
2435                                  LUSTRE_LOV_NAME, &lov_device_type);
2436
2437         if (rc) {
2438                 kmem_cache_destroy(lov_oinfo_slab);
2439                 lu_kmem_fini(lov_caches);
2440         }
2441
2442         RETURN(rc);
2443 }
2444
2445 #ifdef __KERNEL__
2446 static void /*__exit*/ lov_exit(void)
2447 {
2448         class_unregister_type(LUSTRE_LOV_NAME);
2449         kmem_cache_destroy(lov_oinfo_slab);
2450         lu_kmem_fini(lov_caches);
2451 }
2452
2453 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2454 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2455 MODULE_LICENSE("GPL");
2456
2457 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2458 #endif