Whamcloud - gitweb
LU-3319 procfs: fix symlink handling
[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_lib.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_dlm.h>
56 #include <lustre_mds.h>
57 #include <obd_class.h>
58 #include <lprocfs_status.h>
59 #include <lustre_param.h>
60 #include <cl_object.h>
61 #include <lclient.h>
62 #include <lustre/ll_fiemap.h>
63 #include <lustre_fid.h>
64
65 #include "lov_internal.h"
66
67 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
68    Any function that expects lov_tgts to remain stationary must take a ref. */
69 static void lov_getref(struct obd_device *obd)
70 {
71         struct lov_obd *lov = &obd->u.lov;
72
73         /* nobody gets through here until lov_putref is done */
74         mutex_lock(&lov->lov_lock);
75         atomic_inc(&lov->lov_refcount);
76         mutex_unlock(&lov->lov_lock);
77         return;
78 }
79
80 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
81
82 static void lov_putref(struct obd_device *obd)
83 {
84         struct lov_obd *lov = &obd->u.lov;
85
86         mutex_lock(&lov->lov_lock);
87         /* ok to dec to 0 more than once -- ltd_exp's will be null */
88         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
89                 struct list_head kill = LIST_HEAD_INIT(kill);
90                 struct lov_tgt_desc *tgt, *n;
91                 int i;
92
93                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
94                        lov->lov_death_row);
95                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
96                         tgt = lov->lov_tgts[i];
97
98                         if (!tgt || !tgt->ltd_reap)
99                                 continue;
100                         cfs_list_add(&tgt->ltd_kill, &kill);
101                         /* XXX - right now there is a dependency on ld_tgt_count
102                          * being the maximum tgt index for computing the
103                          * mds_max_easize. So we can't shrink it. */
104                         lov_ost_pool_remove(&lov->lov_packed, i);
105                         lov->lov_tgts[i] = NULL;
106                         lov->lov_death_row--;
107                 }
108                 mutex_unlock(&lov->lov_lock);
109
110                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
111                         cfs_list_del(&tgt->ltd_kill);
112                         /* Disconnect */
113                         __lov_del_obd(obd, tgt);
114                 }
115         } else {
116                 mutex_unlock(&lov->lov_lock);
117         }
118 }
119
120 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
121                               enum obd_notify_event ev);
122 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
123                       enum obd_notify_event ev, void *data);
124
125 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
126                     struct obd_connect_data *data)
127 {
128         struct lov_obd *lov = &obd->u.lov;
129         struct obd_uuid *tgt_uuid;
130         struct obd_device *tgt_obd;
131         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
132         struct obd_import *imp;
133         int rc;
134         ENTRY;
135
136         if (lov->lov_tgts[index] == NULL)
137                 RETURN(-EINVAL);
138
139         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
140         tgt_obd = lov->lov_tgts[index]->ltd_obd;
141
142         if (!tgt_obd->obd_set_up) {
143                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
144                 RETURN(-EINVAL);
145         }
146
147         /* override the sp_me from lov */
148         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
149
150         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
151                 data->ocd_index = index;
152
153         /*
154          * Divine LOV knows that OBDs under it are OSCs.
155          */
156         imp = tgt_obd->u.cli.cl_import;
157
158         if (activate) {
159                 tgt_obd->obd_no_recov = 0;
160                 /* FIXME this is probably supposed to be
161                    ptlrpc_set_import_active.  Horrible naming. */
162                 ptlrpc_activate_import(imp);
163         }
164
165         rc = obd_register_observer(tgt_obd, obd);
166         if (rc) {
167                 CERROR("Target %s register_observer error %d\n",
168                        obd_uuid2str(tgt_uuid), rc);
169                 RETURN(rc);
170         }
171
172
173         if (imp->imp_invalid) {
174                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
175                        "disabled\n", obd_uuid2str(tgt_uuid));
176                 RETURN(0);
177         }
178
179         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
180                          &lov_osc_uuid, data, NULL);
181         if (rc || !lov->lov_tgts[index]->ltd_exp) {
182                 CERROR("Target %s connect error %d\n",
183                        obd_uuid2str(tgt_uuid), rc);
184                 RETURN(-ENODEV);
185         }
186
187         lov->lov_tgts[index]->ltd_reap = 0;
188
189         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
190                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
191
192         if (obd->obd_type->typ_procsym != NULL) {
193                 struct proc_dir_entry *osc_symlink;
194                 struct obd_device *osc_obd;
195
196                 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
197
198                 LASSERT(osc_obd != NULL);
199                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
200                 LASSERT(osc_obd->obd_type->typ_name != NULL);
201
202                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
203                                                   obd->obd_type->typ_procsym,
204                                                   "../../../%s/%s",
205                                                   osc_obd->obd_type->typ_name,
206                                                   osc_obd->obd_name);
207                 if (osc_symlink == NULL) {
208                         CERROR("could not register LOV target "
209                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
210                                obd->obd_type->typ_name, obd->obd_name,
211                                osc_obd->obd_name);
212                         lprocfs_remove(&obd->obd_type->typ_procsym);
213                 }
214         }
215         RETURN(0);
216 }
217
218 static int lov_connect(const struct lu_env *env,
219                        struct obd_export **exp, struct obd_device *obd,
220                        struct obd_uuid *cluuid, struct obd_connect_data *data,
221                        void *localdata)
222 {
223         struct lov_obd *lov = &obd->u.lov;
224         struct lov_tgt_desc *tgt;
225         struct lustre_handle conn;
226         int i, rc;
227         ENTRY;
228
229         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
230
231         rc = class_connect(&conn, obd, cluuid);
232         if (rc)
233                 RETURN(rc);
234
235         *exp = class_conn2export(&conn);
236
237         /* Why should there ever be more than 1 connect? */
238         lov->lov_connects++;
239         LASSERT(lov->lov_connects == 1);
240
241         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
242         if (data)
243                 lov->lov_ocd = *data;
244
245         obd->obd_type->typ_procsym = lprocfs_seq_register("target_obds",
246                                                  obd->obd_proc_entry,
247                                                  NULL, NULL);
248         if (IS_ERR(obd->obd_type->typ_procsym)) {
249                 CERROR("%s: could not register /proc/fs/lustre/%s/%s/target_obds.",
250                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
251                 obd->obd_type->typ_procsym = NULL;
252         }
253
254         obd_getref(obd);
255         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
256                 tgt = lov->lov_tgts[i];
257                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
258                         continue;
259                 /* Flags will be lowest common denominator */
260                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
261                 if (rc) {
262                         CERROR("%s: lov connect tgt %d failed: %d\n",
263                                obd->obd_name, i, rc);
264                         continue;
265                 }
266                 /* connect to administrative disabled ost */
267                 if (!lov->lov_tgts[i]->ltd_exp)
268                         continue;
269
270                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
271                                 OBD_NOTIFY_CONNECT, (void *)&i);
272                 if (rc) {
273                         CERROR("%s error sending notify %d\n",
274                                obd->obd_name, rc);
275                 }
276         }
277         obd_putref(obd);
278
279         RETURN(0);
280 }
281
282 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
283 {
284         struct lov_obd *lov = &obd->u.lov;
285         struct obd_device *osc_obd;
286         int rc;
287         ENTRY;
288
289         osc_obd = class_exp2obd(tgt->ltd_exp);
290         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
291                obd->obd_name, osc_obd->obd_name);
292
293         if (tgt->ltd_active) {
294                 tgt->ltd_active = 0;
295                 lov->desc.ld_active_tgt_count--;
296                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
297         }
298
299         if (osc_obd) {
300                 /* Pass it on to our clients.
301                  * XXX This should be an argument to disconnect,
302                  * XXX not a back-door flag on the OBD.  Ah well.
303                  */
304                 osc_obd->obd_force = obd->obd_force;
305                 osc_obd->obd_fail = obd->obd_fail;
306                 osc_obd->obd_no_recov = obd->obd_no_recov;
307
308                 if (obd->obd_type->typ_procsym)
309                         lprocfs_remove_proc_entry(osc_obd->obd_name,
310                                                   obd->obd_type->typ_procsym);
311         }
312
313         obd_register_observer(osc_obd, NULL);
314
315         rc = obd_disconnect(tgt->ltd_exp);
316         if (rc) {
317                 CERROR("Target %s disconnect error %d\n",
318                        tgt->ltd_uuid.uuid, rc);
319                 rc = 0;
320         }
321
322         tgt->ltd_exp = NULL;
323         RETURN(0);
324 }
325
326 static int lov_disconnect(struct obd_export *exp)
327 {
328         struct obd_device *obd = class_exp2obd(exp);
329         struct lov_obd *lov = &obd->u.lov;
330         int i, rc;
331         ENTRY;
332
333         if (!lov->lov_tgts)
334                 goto out;
335
336         /* Only disconnect the underlying layers on the final disconnect. */
337         lov->lov_connects--;
338         if (lov->lov_connects != 0) {
339                 /* why should there be more than 1 connect? */
340                 CERROR("disconnect #%d\n", lov->lov_connects);
341                 goto out;
342         }
343
344         /* Let's hold another reference so lov_del_obd doesn't spin through
345            putref every time */
346         obd_getref(obd);
347
348         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
349                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
350                         /* Disconnection is the last we know about an obd */
351                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
352                 }
353         }
354         obd_putref(obd);
355
356         if (obd->obd_type->typ_procsym)
357                 lprocfs_remove(&obd->obd_type->typ_procsym);
358
359 out:
360         rc = class_disconnect(exp); /* bz 9811 */
361         RETURN(rc);
362 }
363
364 /* Error codes:
365  *
366  *  -EINVAL  : UUID can't be found in the LOV's target list
367  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
368  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
369  *  any >= 0 : is log target index
370  */
371 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
372                               enum obd_notify_event ev)
373 {
374         struct lov_obd *lov = &obd->u.lov;
375         struct lov_tgt_desc *tgt;
376         int index, activate, active;
377         ENTRY;
378
379         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
380                lov, uuid->uuid, ev);
381
382         obd_getref(obd);
383         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
384                 tgt = lov->lov_tgts[index];
385                 if (!tgt)
386                         continue;
387                 /*
388                  * LU-642, initially inactive OSC could miss the obd_connect,
389                  * we make up for it here.
390                  */
391                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
392                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
393                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
394
395                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
396                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
397                 }
398                 if (!tgt->ltd_exp)
399                         continue;
400
401                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
402                        index, obd_uuid2str(&tgt->ltd_uuid),
403                        tgt->ltd_exp->exp_handle.h_cookie);
404                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
405                         break;
406         }
407
408         if (index == lov->desc.ld_tgt_count)
409                 GOTO(out, index = -EINVAL);
410
411         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
412                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
413
414                 if (lov->lov_tgts[index]->ltd_activate == activate) {
415                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
416                                uuid->uuid, activate ? "" : "de");
417                 } else {
418                         lov->lov_tgts[index]->ltd_activate = activate;
419                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
420                                activate ? "" : "de", obd_uuid2str(uuid));
421                 }
422
423         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
424                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
425
426                 if (lov->lov_tgts[index]->ltd_active == active) {
427                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
428                                uuid->uuid, active ? "" : "in");
429                         GOTO(out, index);
430                 } else {
431                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
432                                obd_uuid2str(uuid), active ? "" : "in");
433                 }
434
435                 lov->lov_tgts[index]->ltd_active = active;
436                 if (active) {
437                         lov->desc.ld_active_tgt_count++;
438                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
439                 } else {
440                         lov->desc.ld_active_tgt_count--;
441                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
442                 }
443         } else {
444                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
445         }
446
447  out:
448         obd_putref(obd);
449         RETURN(index);
450 }
451
452 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
453                       enum obd_notify_event ev, void *data)
454 {
455         int rc = 0;
456         struct lov_obd *lov = &obd->u.lov;
457         ENTRY;
458
459         down_read(&lov->lov_notify_lock);
460         if (!lov->lov_connects) {
461                 up_read(&lov->lov_notify_lock);
462                 RETURN(rc);
463         }
464
465         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
466             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
467                 struct obd_uuid *uuid;
468
469                 LASSERT(watched);
470
471                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
472                         up_read(&lov->lov_notify_lock);
473                         CERROR("unexpected notification of %s %s!\n",
474                                watched->obd_type->typ_name,
475                                watched->obd_name);
476                         RETURN(-EINVAL);
477                 }
478                 uuid = &watched->u.cli.cl_target_uuid;
479
480                 /* Set OSC as active before notifying the observer, so the
481                  * observer can use the OSC normally.
482                  */
483                 rc = lov_set_osc_active(obd, uuid, ev);
484                 if (rc < 0) {
485                         up_read(&lov->lov_notify_lock);
486                         CERROR("event(%d) of %s failed: %d\n", ev,
487                                obd_uuid2str(uuid), rc);
488                         RETURN(rc);
489                 }
490                 /* active event should be pass lov target index as data */
491                 data = &rc;
492         }
493
494         /* Pass the notification up the chain. */
495         if (watched) {
496                 rc = obd_notify_observer(obd, watched, ev, data);
497         } else {
498                 /* NULL watched means all osc's in the lov (only for syncs) */
499                 /* sync event should be send lov idx as data */
500                 struct lov_obd *lov = &obd->u.lov;
501                 int i, is_sync;
502
503                 data = &i;
504                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
505                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
506
507                 obd_getref(obd);
508                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
509                         if (!lov->lov_tgts[i])
510                                 continue;
511
512                         /* don't send sync event if target not
513                          * connected/activated */
514                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
515                                 continue;
516
517                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
518                                                  ev, data);
519                         if (rc) {
520                                 CERROR("%s: notify %s of %s failed %d\n",
521                                        obd->obd_name,
522                                        obd->obd_observer->obd_name,
523                                        lov->lov_tgts[i]->ltd_obd->obd_name,
524                                        rc);
525                         }
526                 }
527                 obd_putref(obd);
528         }
529
530         up_read(&lov->lov_notify_lock);
531         RETURN(rc);
532 }
533
534 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
535                           __u32 index, int gen, int active)
536 {
537         struct lov_obd *lov = &obd->u.lov;
538         struct lov_tgt_desc *tgt;
539         struct obd_device *tgt_obd;
540         int rc;
541         ENTRY;
542
543         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
544                uuidp->uuid, index, gen, active);
545
546         if (gen <= 0) {
547                 CERROR("request to add OBD %s with invalid generation: %d\n",
548                        uuidp->uuid, gen);
549                 RETURN(-EINVAL);
550         }
551
552         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
553                                         &obd->obd_uuid);
554         if (tgt_obd == NULL)
555                 RETURN(-EINVAL);
556
557         mutex_lock(&lov->lov_lock);
558
559         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
560                 tgt = lov->lov_tgts[index];
561                 CERROR("UUID %s already assigned at LOV target index %d\n",
562                        obd_uuid2str(&tgt->ltd_uuid), index);
563                 mutex_unlock(&lov->lov_lock);
564                 RETURN(-EEXIST);
565         }
566
567         if (index >= lov->lov_tgt_size) {
568                 /* We need to reallocate the lov target array. */
569                 struct lov_tgt_desc **newtgts, **old = NULL;
570                 __u32 newsize, oldsize = 0;
571
572                 newsize = max(lov->lov_tgt_size, (__u32)2);
573                 while (newsize < index + 1)
574                         newsize = newsize << 1;
575                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
576                 if (newtgts == NULL) {
577                         mutex_unlock(&lov->lov_lock);
578                         RETURN(-ENOMEM);
579                 }
580
581                 if (lov->lov_tgt_size) {
582                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
583                                lov->lov_tgt_size);
584                         old = lov->lov_tgts;
585                         oldsize = lov->lov_tgt_size;
586                 }
587
588                 lov->lov_tgts = newtgts;
589                 lov->lov_tgt_size = newsize;
590                 smp_rmb();
591                 if (old)
592                         OBD_FREE(old, sizeof(*old) * oldsize);
593
594                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
595                        lov->lov_tgts, lov->lov_tgt_size);
596         }
597
598         OBD_ALLOC_PTR(tgt);
599         if (!tgt) {
600                 mutex_unlock(&lov->lov_lock);
601                 RETURN(-ENOMEM);
602         }
603
604         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
605         if (rc) {
606                 mutex_unlock(&lov->lov_lock);
607                 OBD_FREE_PTR(tgt);
608                 RETURN(rc);
609         }
610
611         tgt->ltd_uuid = *uuidp;
612         tgt->ltd_obd = tgt_obd;
613         /* XXX - add a sanity check on the generation number. */
614         tgt->ltd_gen = gen;
615         tgt->ltd_index = index;
616         tgt->ltd_activate = active;
617         lov->lov_tgts[index] = tgt;
618         if (index >= lov->desc.ld_tgt_count)
619                 lov->desc.ld_tgt_count = index + 1;
620
621         mutex_unlock(&lov->lov_lock);
622
623         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
624                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
625
626         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
627
628         if (lov->lov_connects == 0) {
629                 /* lov_connect hasn't been called yet. We'll do the
630                    lov_connect_obd on this target when that fn first runs,
631                    because we don't know the connect flags yet. */
632                 RETURN(0);
633         }
634
635         obd_getref(obd);
636
637         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
638         if (rc)
639                 GOTO(out, rc);
640
641         /* connect to administrative disabled ost */
642         if (!tgt->ltd_exp)
643                 GOTO(out, rc = 0);
644
645         if (lov->lov_cache != NULL) {
646                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
647                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
648                                 sizeof(struct cl_client_cache), lov->lov_cache,
649                                 NULL);
650                 if (rc < 0)
651                         GOTO(out, rc);
652         }
653
654         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
655                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
656                         (void *)&index);
657
658 out:
659         if (rc) {
660                 CERROR("add failed (%d), deleting %s\n", rc,
661                        obd_uuid2str(&tgt->ltd_uuid));
662                 lov_del_target(obd, index, 0, 0);
663         }
664         obd_putref(obd);
665         RETURN(rc);
666 }
667
668 /* Schedule a target for deletion */
669 int lov_del_target(struct obd_device *obd, __u32 index,
670                    struct obd_uuid *uuidp, int gen)
671 {
672         struct lov_obd *lov = &obd->u.lov;
673         int count = lov->desc.ld_tgt_count;
674         int rc = 0;
675         ENTRY;
676
677         if (index >= count) {
678                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
679                        index, count);
680                 RETURN(-EINVAL);
681         }
682
683         /* to make sure there's no ongoing lov_notify() now */
684         down_write(&lov->lov_notify_lock);
685         obd_getref(obd);
686
687         if (!lov->lov_tgts[index]) {
688                 CERROR("LOV target at index %d is not setup.\n", index);
689                 GOTO(out, rc = -EINVAL);
690         }
691
692         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
693                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
694                        lov_uuid2str(lov, index), index,
695                        obd_uuid2str(uuidp));
696                 GOTO(out, rc = -EINVAL);
697         }
698
699         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
700                lov_uuid2str(lov, index), index,
701                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
702                lov->lov_tgts[index]->ltd_active);
703
704         lov->lov_tgts[index]->ltd_reap = 1;
705         lov->lov_death_row++;
706         /* we really delete it from obd_putref */
707 out:
708         obd_putref(obd);
709         up_write(&lov->lov_notify_lock);
710
711         RETURN(rc);
712 }
713
714 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
715 {
716         struct obd_device *osc_obd;
717
718         LASSERT(tgt);
719         LASSERT(tgt->ltd_reap);
720
721         osc_obd = class_exp2obd(tgt->ltd_exp);
722
723         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
724                tgt->ltd_uuid.uuid,
725                osc_obd ? osc_obd->obd_name : "<no obd>");
726
727         if (tgt->ltd_exp)
728                 lov_disconnect_obd(obd, tgt);
729
730         OBD_FREE_PTR(tgt);
731
732         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
733            do it ourselves. And we can't do it from lov_cleanup,
734            because we just lost our only reference to it. */
735         if (osc_obd)
736                 class_manual_cleanup(osc_obd);
737 }
738
739 void lov_fix_desc_stripe_size(__u64 *val)
740 {
741         if (*val < LOV_MIN_STRIPE_SIZE) {
742                 if (*val != 0)
743                         LCONSOLE_INFO("Increasing default stripe size to "
744                                       "minimum %u\n",
745                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
746                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
747         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
748                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
749                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
750                               "multiple of %u)\n",
751                               *val, LOV_MIN_STRIPE_SIZE);
752         }
753 }
754
755 void lov_fix_desc_stripe_count(__u32 *val)
756 {
757         if (*val == 0)
758                 *val = 1;
759 }
760
761 void lov_fix_desc_pattern(__u32 *val)
762 {
763         /* from lov_setstripe */
764         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
765                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
766                 *val = 0;
767         }
768 }
769
770 void lov_fix_desc_qos_maxage(__u32 *val)
771 {
772         if (*val == 0)
773                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
774 }
775
776 void lov_fix_desc(struct lov_desc *desc)
777 {
778         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
779         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
780         lov_fix_desc_pattern(&desc->ld_pattern);
781         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
782 }
783
784 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
785 {
786         struct lov_desc *desc;
787         struct lov_obd *lov = &obd->u.lov;
788 #ifdef LPROCFS
789         struct obd_type *type;
790 #endif
791         int rc;
792         ENTRY;
793
794         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
795                 CERROR("LOV setup requires a descriptor\n");
796                 RETURN(-EINVAL);
797         }
798
799         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
800
801         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
802                 CERROR("descriptor size wrong: %d > %d\n",
803                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
804                 RETURN(-EINVAL);
805         }
806
807         if (desc->ld_magic != LOV_DESC_MAGIC) {
808                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
809                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
810                                    obd->obd_name, desc);
811                             lustre_swab_lov_desc(desc);
812                 } else {
813                         CERROR("%s: Bad lov desc magic: %#x\n",
814                                obd->obd_name, desc->ld_magic);
815                         RETURN(-EINVAL);
816                 }
817         }
818
819         lov_fix_desc(desc);
820
821         desc->ld_active_tgt_count = 0;
822         lov->desc = *desc;
823         lov->lov_tgt_size = 0;
824
825         mutex_init(&lov->lov_lock);
826         atomic_set(&lov->lov_refcount, 0);
827         lov->lov_sp_me = LUSTRE_SP_CLI;
828
829         init_rwsem(&lov->lov_notify_lock);
830
831         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
832                                                    HASH_POOLS_MAX_BITS,
833                                                    HASH_POOLS_BKT_BITS, 0,
834                                                    CFS_HASH_MIN_THETA,
835                                                    CFS_HASH_MAX_THETA,
836                                                    &pool_hash_operations,
837                                                    CFS_HASH_DEFAULT);
838         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
839         lov->lov_pool_count = 0;
840         rc = lov_ost_pool_init(&lov->lov_packed, 0);
841         if (rc)
842                 GOTO(out, rc);
843
844 #ifdef LPROCFS
845         obd->obd_vars = lprocfs_lov_obd_vars;
846         /* If this is true then both client (lov) and server
847          * (lod) are on the same node. The lod layer if loaded
848          * first will register the lov proc directory. In that
849          * case obd->obd_type->typ_procroot will be not set.
850          * Instead we use type->typ_procsym as the parent. */
851         type = class_search_type(LUSTRE_LOD_NAME);
852         if (type != NULL && type->typ_procsym != NULL) {
853                 obd->obd_proc_entry = lprocfs_seq_register(obd->obd_name,
854                                                            type->typ_procsym,
855                                                            obd->obd_vars, obd);
856                 if (IS_ERR(obd->obd_proc_entry)) {
857                         rc = PTR_ERR(obd->obd_proc_entry);
858                         CERROR("error %d setting up lprocfs for %s\n", rc,
859                                obd->obd_name);
860                         obd->obd_proc_entry = NULL;
861                 }
862         } else {
863                 rc = lprocfs_seq_obd_setup(obd);
864         }
865
866         if (rc == 0) {
867                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
868                                         0444, &lov_proc_target_fops, obd);
869                 if (rc)
870                         CWARN("Error adding the target_obd file\n");
871
872                 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
873                                                         obd->obd_proc_entry,
874                                                         NULL, NULL);
875                 if (IS_ERR(lov->lov_pool_proc_entry)) {
876                         rc = PTR_ERR(lov->lov_pool_proc_entry);
877                         CERROR("error %d setting up lprocfs for pools\n", rc);
878                         lov->lov_pool_proc_entry = NULL;
879                 }
880         }
881 #endif
882         RETURN(0);
883
884 out:
885         return rc;
886 }
887
888 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
889 {
890         int rc = 0;
891         struct lov_obd *lov = &obd->u.lov;
892
893         ENTRY;
894
895         switch (stage) {
896         case OBD_CLEANUP_EARLY: {
897                 int i;
898                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
899                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
900                                 continue;
901                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
902                                        OBD_CLEANUP_EARLY);
903                 }
904                 break;
905         }
906         default:
907                 break;
908         }
909
910         RETURN(rc);
911 }
912
913 static int lov_cleanup(struct obd_device *obd)
914 {
915         struct lov_obd *lov = &obd->u.lov;
916         cfs_list_t *pos, *tmp;
917         struct pool_desc *pool;
918         ENTRY;
919
920         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
921                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
922                 /* free pool structs */
923                 CDEBUG(D_INFO, "delete pool %p\n", pool);
924                 /* In the function below, .hs_keycmp resolves to
925                  * pool_hashkey_keycmp() */
926                 /* coverity[overrun-buffer-val] */
927                 lov_pool_del(obd, pool->pool_name);
928         }
929         cfs_hash_putref(lov->lov_pools_hash_body);
930         lov_ost_pool_free(&lov->lov_packed);
931
932         lprocfs_obd_cleanup(obd);
933         if (lov->lov_tgts) {
934                 int i;
935                 obd_getref(obd);
936                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
937                         if (!lov->lov_tgts[i])
938                                 continue;
939
940                         /* Inactive targets may never have connected */
941                         if (lov->lov_tgts[i]->ltd_active ||
942                             atomic_read(&lov->lov_refcount))
943                                 /* We should never get here - these
944                                  * should have been removed in the
945                                  * disconnect. */
946                                 CERROR("%s: lov tgt %d not cleaned! "
947                                        "deathrow=%d, lovrc=%d\n",
948                                        obd->obd_name, i, lov->lov_death_row,
949                                        atomic_read(&lov->lov_refcount));
950                         lov_del_target(obd, i, 0, 0);
951                 }
952                 obd_putref(obd);
953                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
954                          lov->lov_tgt_size);
955                 lov->lov_tgt_size = 0;
956         }
957         RETURN(0);
958 }
959
960 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
961                             __u32 *indexp, int *genp)
962 {
963         struct obd_uuid obd_uuid;
964         int cmd;
965         int rc = 0;
966         ENTRY;
967
968         switch(cmd = lcfg->lcfg_command) {
969         case LCFG_LOV_ADD_OBD:
970         case LCFG_LOV_ADD_INA:
971         case LCFG_LOV_DEL_OBD: {
972                 __u32 index;
973                 int gen;
974                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
975                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
976                         GOTO(out, rc = -EINVAL);
977
978                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
979
980                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
981                         GOTO(out, rc = -EINVAL);
982                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
983                         GOTO(out, rc = -EINVAL);
984                 index = *indexp;
985                 gen = *genp;
986                 if (cmd == LCFG_LOV_ADD_OBD)
987                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
988                 else if (cmd == LCFG_LOV_ADD_INA)
989                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
990                 else
991                         rc = lov_del_target(obd, index, &obd_uuid, gen);
992                 GOTO(out, rc);
993         }
994         case LCFG_PARAM: {
995                 struct lov_desc *desc = &(obd->u.lov.desc);
996
997                 if (!desc)
998                         GOTO(out, rc = -EINVAL);
999
1000                 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
1001                                                   lcfg, obd);
1002                 if (rc > 0)
1003                         rc = 0;
1004                 GOTO(out, rc);
1005         }
1006         case LCFG_POOL_NEW:
1007         case LCFG_POOL_ADD:
1008         case LCFG_POOL_DEL:
1009         case LCFG_POOL_REM:
1010                 GOTO(out, rc);
1011
1012         default: {
1013                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1014                 GOTO(out, rc = -EINVAL);
1015
1016         }
1017         }
1018 out:
1019         RETURN(rc);
1020 }
1021
1022 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1023                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1024 {
1025         struct lov_stripe_md *obj_mdp, *lsm;
1026         struct lov_obd *lov = &exp->exp_obd->u.lov;
1027         unsigned ost_idx;
1028         int rc, i;
1029         ENTRY;
1030
1031         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1032                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1033
1034         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1035         if (obj_mdp == NULL)
1036                 RETURN(-ENOMEM);
1037
1038         ost_idx = src_oa->o_nlink;
1039         lsm = *ea;
1040         if (lsm == NULL)
1041                 GOTO(out, rc = -EINVAL);
1042         if (ost_idx >= lov->desc.ld_tgt_count ||
1043             !lov->lov_tgts[ost_idx])
1044                 GOTO(out, rc = -EINVAL);
1045
1046         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1047                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1048                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1049                                         ostid_id(&src_oa->o_oi))
1050                                 GOTO(out, rc = -EINVAL);
1051                         break;
1052                 }
1053         }
1054         if (i == lsm->lsm_stripe_count)
1055                 GOTO(out, rc = -EINVAL);
1056
1057         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1058                         src_oa, &obj_mdp, oti);
1059 out:
1060         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1061         RETURN(rc);
1062 }
1063
1064 /* the LOV expects oa->o_id to be set to the LOV object id */
1065 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1066                       struct obdo *src_oa, struct lov_stripe_md **ea,
1067                       struct obd_trans_info *oti)
1068 {
1069         struct lov_obd *lov;
1070         int rc = 0;
1071         ENTRY;
1072
1073         LASSERT(ea != NULL);
1074         if (exp == NULL)
1075                 RETURN(-EINVAL);
1076
1077         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1078             src_oa->o_flags == OBD_FL_DELORPHAN) {
1079                 /* should be used with LOV anymore */
1080                 LBUG();
1081         }
1082
1083         lov = &exp->exp_obd->u.lov;
1084         if (!lov->desc.ld_active_tgt_count)
1085                 RETURN(-EIO);
1086
1087         obd_getref(exp->exp_obd);
1088         /* Recreate a specific object id at the given OST index */
1089         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1090             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1091                  rc = lov_recreate(exp, src_oa, ea, oti);
1092         }
1093
1094         obd_putref(exp->exp_obd);
1095         RETURN(rc);
1096 }
1097
1098 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1099 do {                                                                            \
1100         LASSERT((lsmp) != NULL);                                                \
1101         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1102                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1103                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1104 } while (0)
1105
1106 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1107                        struct obdo *oa, struct lov_stripe_md *lsm,
1108                        struct obd_trans_info *oti, struct obd_export *md_exp,
1109                        void *capa)
1110 {
1111         struct lov_request_set *set;
1112         struct obd_info oinfo;
1113         struct lov_request *req;
1114         cfs_list_t *pos;
1115         struct lov_obd *lov;
1116         int rc = 0, err = 0;
1117         ENTRY;
1118
1119         ASSERT_LSM_MAGIC(lsm);
1120
1121         if (!exp || !exp->exp_obd)
1122                 RETURN(-ENODEV);
1123
1124         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1125                 LASSERT(oti);
1126                 LASSERT(oti->oti_logcookies);
1127         }
1128
1129         lov = &exp->exp_obd->u.lov;
1130         obd_getref(exp->exp_obd);
1131         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1132         if (rc)
1133                 GOTO(out, rc);
1134
1135         cfs_list_for_each (pos, &set->set_list) {
1136                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1137
1138                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1139                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1140
1141                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1142                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1143                 err = lov_update_common_set(set, req, err);
1144                 if (err) {
1145                         CERROR("%s: destroying objid "DOSTID" subobj "
1146                                DOSTID" on OST idx %d: rc = %d\n",
1147                                exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1148                                POSTID(&req->rq_oi.oi_oa->o_oi),
1149                                req->rq_idx, err);
1150                         if (!rc)
1151                                 rc = err;
1152                 }
1153         }
1154
1155         if (rc == 0) {
1156                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1157                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1158         }
1159         err = lov_fini_destroy_set(set);
1160 out:
1161         obd_putref(exp->exp_obd);
1162         RETURN(rc ? rc : err);
1163 }
1164
1165 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1166                                  void *data, int rc)
1167 {
1168         struct lov_request_set *lovset = (struct lov_request_set *)data;
1169         int err;
1170         ENTRY;
1171
1172         /* don't do attribute merge if this aysnc op failed */
1173         if (rc)
1174                 atomic_set(&lovset->set_completes, 0);
1175         err = lov_fini_getattr_set(lovset);
1176         RETURN(rc ? rc : err);
1177 }
1178
1179 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1180                               struct ptlrpc_request_set *rqset)
1181 {
1182         struct lov_request_set *lovset;
1183         struct lov_obd *lov;
1184         cfs_list_t *pos;
1185         struct lov_request *req;
1186         int rc = 0, err;
1187         ENTRY;
1188
1189         LASSERT(oinfo);
1190         ASSERT_LSM_MAGIC(oinfo->oi_md);
1191
1192         if (!exp || !exp->exp_obd)
1193                 RETURN(-ENODEV);
1194
1195         lov = &exp->exp_obd->u.lov;
1196
1197         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1198         if (rc)
1199                 RETURN(rc);
1200
1201         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1202                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1203                oinfo->oi_md->lsm_stripe_size);
1204
1205         cfs_list_for_each(pos, &lovset->set_list) {
1206                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1207
1208                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1209                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1210                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1211                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1212                                        &req->rq_oi, rqset);
1213                 if (rc) {
1214                         CERROR("%s: getattr objid "DOSTID" subobj"
1215                                DOSTID" on OST idx %d: rc = %d\n",
1216                                exp->exp_obd->obd_name,
1217                                POSTID(&oinfo->oi_oa->o_oi),
1218                                POSTID(&req->rq_oi.oi_oa->o_oi),
1219                                req->rq_idx, rc);
1220                         GOTO(out, rc);
1221                 }
1222         }
1223
1224         if (!cfs_list_empty(&rqset->set_requests)) {
1225                 LASSERT(rc == 0);
1226                 LASSERT (rqset->set_interpret == NULL);
1227                 rqset->set_interpret = lov_getattr_interpret;
1228                 rqset->set_arg = (void *)lovset;
1229                 RETURN(rc);
1230         }
1231 out:
1232         if (rc)
1233                 atomic_set(&lovset->set_completes, 0);
1234         err = lov_fini_getattr_set(lovset);
1235         RETURN(rc ? rc : err);
1236 }
1237
1238 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1239                                  void *data, int rc)
1240 {
1241         struct lov_request_set *lovset = (struct lov_request_set *)data;
1242         int err;
1243         ENTRY;
1244
1245         if (rc)
1246                 atomic_set(&lovset->set_completes, 0);
1247         err = lov_fini_setattr_set(lovset);
1248         RETURN(rc ? rc : err);
1249 }
1250
1251 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1252    needed. Otherwise, a client is waiting for responses. */
1253 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1254                              struct obd_trans_info *oti,
1255                              struct ptlrpc_request_set *rqset)
1256 {
1257         struct lov_request_set *set;
1258         struct lov_request *req;
1259         cfs_list_t *pos;
1260         struct lov_obd *lov;
1261         int rc = 0;
1262         ENTRY;
1263
1264         LASSERT(oinfo);
1265         ASSERT_LSM_MAGIC(oinfo->oi_md);
1266         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1267                 LASSERT(oti);
1268                 LASSERT(oti->oti_logcookies);
1269         }
1270
1271         if (!exp || !exp->exp_obd)
1272                 RETURN(-ENODEV);
1273
1274         lov = &exp->exp_obd->u.lov;
1275         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1276         if (rc)
1277                 RETURN(rc);
1278
1279         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1280                POSTID(&oinfo->oi_md->lsm_oi),
1281                oinfo->oi_md->lsm_stripe_count,
1282                oinfo->oi_md->lsm_stripe_size);
1283
1284         cfs_list_for_each(pos, &set->set_list) {
1285                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1286
1287                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1288                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1289
1290                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1291                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1292                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1293
1294                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1295                                        &req->rq_oi, oti, rqset);
1296                 if (rc) {
1297                         CERROR("error: setattr objid "DOSTID" subobj"
1298                                DOSTID" on OST idx %d: rc = %d\n",
1299                                POSTID(&set->set_oi->oi_oa->o_oi),
1300                                POSTID(&req->rq_oi.oi_oa->o_oi),
1301                                req->rq_idx, rc);
1302                         break;
1303                 }
1304         }
1305
1306         /* If we are not waiting for responses on async requests, return. */
1307         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1308                 int err;
1309                 if (rc)
1310                         atomic_set(&set->set_completes, 0);
1311                 err = lov_fini_setattr_set(set);
1312                 RETURN(rc ? rc : err);
1313         }
1314
1315         LASSERT(rqset->set_interpret == NULL);
1316         rqset->set_interpret = lov_setattr_interpret;
1317         rqset->set_arg = (void *)set;
1318
1319         RETURN(0);
1320 }
1321
1322 static int lov_change_cbdata(struct obd_export *exp,
1323                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1324                              void *data)
1325 {
1326         struct lov_obd *lov;
1327         int rc = 0, i;
1328         ENTRY;
1329
1330         ASSERT_LSM_MAGIC(lsm);
1331
1332         if (!exp || !exp->exp_obd)
1333                 RETURN(-ENODEV);
1334
1335         lov = &exp->exp_obd->u.lov;
1336         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1337                 struct lov_stripe_md submd;
1338                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1339
1340                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1341                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1342                         continue;
1343                 }
1344
1345                 submd.lsm_oi = loi->loi_oi;
1346                 submd.lsm_stripe_count = 0;
1347                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1348                                        &submd, it, data);
1349         }
1350         RETURN(rc);
1351 }
1352
1353 /* find any ldlm lock of the inode in lov
1354  * return 0    not find
1355  *        1    find one
1356  *      < 0    error */
1357 static int lov_find_cbdata(struct obd_export *exp,
1358                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1359                            void *data)
1360 {
1361         struct lov_obd *lov;
1362         int rc = 0, i;
1363         ENTRY;
1364
1365         ASSERT_LSM_MAGIC(lsm);
1366
1367         if (!exp || !exp->exp_obd)
1368                 RETURN(-ENODEV);
1369
1370         lov = &exp->exp_obd->u.lov;
1371         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1372                 struct lov_stripe_md submd;
1373                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1374
1375                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1376                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1377                         continue;
1378                 }
1379                 submd.lsm_oi = loi->loi_oi;
1380                 submd.lsm_stripe_count = 0;
1381                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1382                                      &submd, it, data);
1383                 if (rc != 0)
1384                         RETURN(rc);
1385         }
1386         RETURN(rc);
1387 }
1388
1389 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1390 {
1391         struct lov_request_set *lovset = (struct lov_request_set *)data;
1392         int err;
1393         ENTRY;
1394
1395         if (rc)
1396                 atomic_set(&lovset->set_completes, 0);
1397
1398         err = lov_fini_statfs_set(lovset);
1399         RETURN(rc ? rc : err);
1400 }
1401
1402 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1403                             __u64 max_age, struct ptlrpc_request_set *rqset)
1404 {
1405         struct obd_device      *obd = class_exp2obd(exp);
1406         struct lov_request_set *set;
1407         struct lov_request *req;
1408         cfs_list_t *pos;
1409         struct lov_obd *lov;
1410         int rc = 0;
1411         ENTRY;
1412
1413         LASSERT(oinfo != NULL);
1414         LASSERT(oinfo->oi_osfs != NULL);
1415
1416         lov = &obd->u.lov;
1417         rc = lov_prep_statfs_set(obd, oinfo, &set);
1418         if (rc)
1419                 RETURN(rc);
1420
1421         cfs_list_for_each (pos, &set->set_list) {
1422                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1423                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1424                                       &req->rq_oi, max_age, rqset);
1425                 if (rc)
1426                         break;
1427         }
1428
1429         if (rc || cfs_list_empty(&rqset->set_requests)) {
1430                 int err;
1431                 if (rc)
1432                         atomic_set(&set->set_completes, 0);
1433                 err = lov_fini_statfs_set(set);
1434                 RETURN(rc ? rc : err);
1435         }
1436
1437         LASSERT(rqset->set_interpret == NULL);
1438         rqset->set_interpret = lov_statfs_interpret;
1439         rqset->set_arg = (void *)set;
1440         RETURN(0);
1441 }
1442
1443 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1444                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1445 {
1446         struct ptlrpc_request_set *set = NULL;
1447         struct obd_info oinfo = { { { 0 } } };
1448         int rc = 0;
1449         ENTRY;
1450
1451
1452         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1453          * statfs requests */
1454         set = ptlrpc_prep_set();
1455         if (set == NULL)
1456                 RETURN(-ENOMEM);
1457
1458         oinfo.oi_osfs = osfs;
1459         oinfo.oi_flags = flags;
1460         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1461         if (rc == 0)
1462                 rc = ptlrpc_set_wait(set);
1463         ptlrpc_set_destroy(set);
1464
1465         RETURN(rc);
1466 }
1467
1468 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1469                          void *karg, void *uarg)
1470 {
1471         struct obd_device *obddev = class_exp2obd(exp);
1472         struct lov_obd *lov = &obddev->u.lov;
1473         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1474         struct obd_uuid *uuidp;
1475         ENTRY;
1476
1477         switch (cmd) {
1478         case IOC_OBD_STATFS: {
1479                 struct obd_ioctl_data *data = karg;
1480                 struct obd_device *osc_obd;
1481                 struct obd_statfs stat_buf = {0};
1482                 __u32 index;
1483                 __u32 flags;
1484
1485                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1486                 if ((index >= count))
1487                         RETURN(-ENODEV);
1488
1489                 if (!lov->lov_tgts[index])
1490                         /* Try again with the next index */
1491                         RETURN(-EAGAIN);
1492                 if (!lov->lov_tgts[index]->ltd_active)
1493                         RETURN(-ENODATA);
1494
1495                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1496                 if (!osc_obd)
1497                         RETURN(-EINVAL);
1498
1499                 /* copy UUID */
1500                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1501                                  min((int)data->ioc_plen2,
1502                                      (int)sizeof(struct obd_uuid))))
1503                         RETURN(-EFAULT);
1504
1505                 flags = uarg ? *(__u32*)uarg : 0;
1506                 /* got statfs data */
1507                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1508                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1509                                 flags);
1510                 if (rc)
1511                         RETURN(rc);
1512                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1513                                      min((int) data->ioc_plen1,
1514                                          (int) sizeof(stat_buf))))
1515                         RETURN(-EFAULT);
1516                 break;
1517         }
1518         case OBD_IOC_LOV_GET_CONFIG: {
1519                 struct obd_ioctl_data *data;
1520                 struct lov_desc *desc;
1521                 char *buf = NULL;
1522                 __u32 *genp;
1523
1524                 len = 0;
1525                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1526                         RETURN(-EINVAL);
1527
1528                 data = (struct obd_ioctl_data *)buf;
1529
1530                 if (sizeof(*desc) > data->ioc_inllen1) {
1531                         obd_ioctl_freedata(buf, len);
1532                         RETURN(-EINVAL);
1533                 }
1534
1535                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1536                         obd_ioctl_freedata(buf, len);
1537                         RETURN(-EINVAL);
1538                 }
1539
1540                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1541                         obd_ioctl_freedata(buf, len);
1542                         RETURN(-EINVAL);
1543                 }
1544
1545                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1546                 memcpy(desc, &(lov->desc), sizeof(*desc));
1547
1548                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1549                 genp = (__u32 *)data->ioc_inlbuf3;
1550                 /* the uuid will be empty for deleted OSTs */
1551                 for (i = 0; i < count; i++, uuidp++, genp++) {
1552                         if (!lov->lov_tgts[i])
1553                                 continue;
1554                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1555                         *genp = lov->lov_tgts[i]->ltd_gen;
1556                 }
1557
1558                 if (copy_to_user((void *)uarg, buf, len))
1559                         rc = -EFAULT;
1560                 obd_ioctl_freedata(buf, len);
1561                 break;
1562         }
1563         case LL_IOC_LOV_GETSTRIPE:
1564                 rc = lov_getstripe(exp, karg, uarg);
1565                 break;
1566         case OBD_IOC_QUOTACTL: {
1567                 struct if_quotactl *qctl = karg;
1568                 struct lov_tgt_desc *tgt = NULL;
1569                 struct obd_quotactl *oqctl;
1570
1571                 if (qctl->qc_valid == QC_OSTIDX) {
1572                         if (count <= qctl->qc_idx)
1573                                 RETURN(-EINVAL);
1574
1575                         tgt = lov->lov_tgts[qctl->qc_idx];
1576                         if (!tgt || !tgt->ltd_exp)
1577                                 RETURN(-EINVAL);
1578                 } else if (qctl->qc_valid == QC_UUID) {
1579                         for (i = 0; i < count; i++) {
1580                                 tgt = lov->lov_tgts[i];
1581                                 if (!tgt ||
1582                                     !obd_uuid_equals(&tgt->ltd_uuid,
1583                                                      &qctl->obd_uuid))
1584                                         continue;
1585
1586                                 if (tgt->ltd_exp == NULL)
1587                                         RETURN(-EINVAL);
1588
1589                                 break;
1590                         }
1591                 } else {
1592                         RETURN(-EINVAL);
1593                 }
1594
1595                 if (i >= count)
1596                         RETURN(-EAGAIN);
1597
1598                 LASSERT(tgt && tgt->ltd_exp);
1599                 OBD_ALLOC_PTR(oqctl);
1600                 if (!oqctl)
1601                         RETURN(-ENOMEM);
1602
1603                 QCTL_COPY(oqctl, qctl);
1604                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1605                 if (rc == 0) {
1606                         QCTL_COPY(qctl, oqctl);
1607                         qctl->qc_valid = QC_OSTIDX;
1608                         qctl->obd_uuid = tgt->ltd_uuid;
1609                 }
1610                 OBD_FREE_PTR(oqctl);
1611                 break;
1612         }
1613         default: {
1614                 int set = 0;
1615
1616                 if (count == 0)
1617                         RETURN(-ENOTTY);
1618
1619                 for (i = 0; i < count; i++) {
1620                         int err;
1621                         struct obd_device *osc_obd;
1622
1623                         /* OST was disconnected */
1624                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1625                                 continue;
1626
1627                         /* ll_umount_begin() sets force flag but for lov, not
1628                          * osc. Let's pass it through */
1629                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1630                         osc_obd->obd_force = obddev->obd_force;
1631                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1632                                             len, karg, uarg);
1633                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1634                                 RETURN(err);
1635                         } else if (err) {
1636                                 if (lov->lov_tgts[i]->ltd_active) {
1637                                         CDEBUG(err == -ENOTTY ?
1638                                                D_IOCTL : D_WARNING,
1639                                                "iocontrol OSC %s on OST "
1640                                                "idx %d cmd %x: err = %d\n",
1641                                                lov_uuid2str(lov, i),
1642                                                i, cmd, err);
1643                                         if (!rc)
1644                                                 rc = err;
1645                                 }
1646                         } else {
1647                                 set = 1;
1648                         }
1649                 }
1650                 if (!set && !rc)
1651                         rc = -EIO;
1652         }
1653         }
1654
1655         RETURN(rc);
1656 }
1657
1658 #define FIEMAP_BUFFER_SIZE 4096
1659
1660 /**
1661  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1662  * call. The local end offset and the device are sent in the first
1663  * fm_extent. This function calculates the stripe number from the index.
1664  * This function returns a stripe_no on which mapping is to be restarted.
1665  *
1666  * This function returns fm_end_offset which is the in-OST offset at which
1667  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1668  * will re-calculate proper offset in next stripe.
1669  * Note that the first extent is passed to lov_get_info via the value field.
1670  *
1671  * \param fiemap fiemap request header
1672  * \param lsm striping information for the file
1673  * \param fm_start logical start of mapping
1674  * \param fm_end logical end of mapping
1675  * \param start_stripe starting stripe will be returned in this
1676  */
1677 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1678                                    struct lov_stripe_md *lsm, obd_size fm_start,
1679                                    obd_size fm_end, int *start_stripe)
1680 {
1681         obd_size local_end = fiemap->fm_extents[0].fe_logical;
1682         obd_off lun_start, lun_end;
1683         obd_size fm_end_offset;
1684         int stripe_no = -1, i;
1685
1686         if (fiemap->fm_extent_count == 0 ||
1687             fiemap->fm_extents[0].fe_logical == 0)
1688                 return 0;
1689
1690         /* Find out stripe_no from ost_index saved in the fe_device */
1691         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1692                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
1693                                         fiemap->fm_extents[0].fe_device) {
1694                         stripe_no = i;
1695                         break;
1696                 }
1697         }
1698         if (stripe_no == -1)
1699                 return -EINVAL;
1700
1701         /* If we have finished mapping on previous device, shift logical
1702          * offset to start of next device */
1703         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1704                                    &lun_start, &lun_end)) != 0 &&
1705                                    local_end < lun_end) {
1706                 fm_end_offset = local_end;
1707                 *start_stripe = stripe_no;
1708         } else {
1709                 /* This is a special value to indicate that caller should
1710                  * calculate offset in next stripe. */
1711                 fm_end_offset = 0;
1712                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1713         }
1714
1715         return fm_end_offset;
1716 }
1717
1718 /**
1719  * We calculate on which OST the mapping will end. If the length of mapping
1720  * is greater than (stripe_size * stripe_count) then the last_stripe will
1721  * will be one just before start_stripe. Else we check if the mapping
1722  * intersects each OST and find last_stripe.
1723  * This function returns the last_stripe and also sets the stripe_count
1724  * over which the mapping is spread
1725  *
1726  * \param lsm striping information for the file
1727  * \param fm_start logical start of mapping
1728  * \param fm_end logical end of mapping
1729  * \param start_stripe starting stripe of the mapping
1730  * \param stripe_count the number of stripes across which to map is returned
1731  *
1732  * \retval last_stripe return the last stripe of the mapping
1733  */
1734 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1735                             obd_size fm_end, int start_stripe,
1736                             int *stripe_count)
1737 {
1738         int last_stripe;
1739         obd_off obd_start, obd_end;
1740         int i, j;
1741
1742         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1743                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1744                                                               start_stripe - 1);
1745                 *stripe_count = lsm->lsm_stripe_count;
1746         } else {
1747                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1748                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
1749                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1750                                                    &obd_start, &obd_end)) == 0)
1751                                 break;
1752                 }
1753                 *stripe_count = j;
1754                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1755         }
1756
1757         return last_stripe;
1758 }
1759
1760 /**
1761  * Set fe_device and copy extents from local buffer into main return buffer.
1762  *
1763  * \param fiemap fiemap request header
1764  * \param lcl_fm_ext array of local fiemap extents to be copied
1765  * \param ost_index OST index to be written into the fm_device field for each
1766                     extent
1767  * \param ext_count number of extents to be copied
1768  * \param current_extent where to start copying in main extent array
1769  */
1770 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1771                                   struct ll_fiemap_extent *lcl_fm_ext,
1772                                   int ost_index, unsigned int ext_count,
1773                                   int current_extent)
1774 {
1775         char *to;
1776         int ext;
1777
1778         for (ext = 0; ext < ext_count; ext++) {
1779                 lcl_fm_ext[ext].fe_device = ost_index;
1780                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1781         }
1782
1783         /* Copy fm_extent's from fm_local to return buffer */
1784         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1785         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1786 }
1787
1788 /**
1789  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1790  * This also handles the restarting of FIEMAP calls in case mapping overflows
1791  * the available number of extents in single call.
1792  */
1793 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1794                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1795 {
1796         struct ll_fiemap_info_key *fm_key = key;
1797         struct ll_user_fiemap *fiemap = val;
1798         struct ll_user_fiemap *fm_local = NULL;
1799         struct ll_fiemap_extent *lcl_fm_ext;
1800         int count_local;
1801         unsigned int get_num_extents = 0;
1802         int ost_index = 0, actual_start_stripe, start_stripe;
1803         obd_size fm_start, fm_end, fm_length, fm_end_offset;
1804         obd_size curr_loc;
1805         int current_extent = 0, rc = 0, i;
1806         int ost_eof = 0; /* EOF for object */
1807         int ost_done = 0; /* done with required mapping for this OST? */
1808         int last_stripe;
1809         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1810         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1811
1812         if (!lsm_has_objects(lsm)) {
1813                 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1814                     fm_key->oa.o_size)) {
1815                         /* released file, return a minimal FIEMAP if
1816                          * request fits in file-size.
1817                          */
1818                         fiemap->fm_mapped_extents = 1;
1819                         fiemap->fm_extents[0].fe_logical =
1820                                                 fm_key->fiemap.fm_start;
1821                         if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1822                             fm_key->oa.o_size)
1823                                 fiemap->fm_extents[0].fe_length =
1824                                                 fm_key->fiemap.fm_length;
1825                         else
1826                                 fiemap->fm_extents[0].fe_length =
1827                                                 fm_key->oa.o_size -
1828                                                 fm_key->fiemap.fm_start;
1829                         fiemap->fm_extents[0].fe_flags |=
1830                                                 (FIEMAP_EXTENT_UNKNOWN |
1831                                                  FIEMAP_EXTENT_LAST);
1832                 }
1833                 GOTO(out, rc = 0);
1834         }
1835
1836         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1837                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1838
1839         OBD_ALLOC_LARGE(fm_local, buffer_size);
1840         if (fm_local == NULL)
1841                 GOTO(out, rc = -ENOMEM);
1842         lcl_fm_ext = &fm_local->fm_extents[0];
1843
1844         count_local = fiemap_size_to_count(buffer_size);
1845
1846         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1847         fm_start = fiemap->fm_start;
1848         fm_length = fiemap->fm_length;
1849         /* Calculate start stripe, last stripe and length of mapping */
1850         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1851         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1852                                                 fm_start + fm_length - 1);
1853         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1854         if (fm_end > fm_key->oa.o_size)
1855                 fm_end = fm_key->oa.o_size;
1856
1857         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1858                                             actual_start_stripe, &stripe_count);
1859
1860         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1861                                                   fm_end, &start_stripe);
1862         if (fm_end_offset == -EINVAL)
1863                 GOTO(out, rc = -EINVAL);
1864
1865         if (fiemap->fm_extent_count == 0) {
1866                 get_num_extents = 1;
1867                 count_local = 0;
1868         }
1869
1870         /* Check each stripe */
1871         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1872              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1873                 obd_size req_fm_len; /* Stores length of required mapping */
1874                 obd_size len_mapped_single_call;
1875                 obd_off lun_start, lun_end, obd_object_end;
1876                 unsigned int ext_count;
1877
1878                 cur_stripe_wrap = cur_stripe;
1879
1880                 /* Find out range of mapping on this stripe */
1881                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1882                                            &lun_start, &obd_object_end)) == 0)
1883                         continue;
1884
1885                 /* If this is a continuation FIEMAP call and we are on
1886                  * starting stripe then lun_start needs to be set to
1887                  * fm_end_offset */
1888                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1889                         lun_start = fm_end_offset;
1890
1891                 if (fm_length != ~0ULL) {
1892                         /* Handle fm_start + fm_length overflow */
1893                         if (fm_start + fm_length < fm_start)
1894                                 fm_length = ~0ULL - fm_start;
1895                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1896                                                      cur_stripe);
1897                 } else {
1898                         lun_end = ~0ULL;
1899                 }
1900
1901                 if (lun_start == lun_end)
1902                         continue;
1903
1904                 req_fm_len = obd_object_end - lun_start;
1905                 fm_local->fm_length = 0;
1906                 len_mapped_single_call = 0;
1907
1908                 /* If the output buffer is very large and the objects have many
1909                  * extents we may need to loop on a single OST repeatedly */
1910                 ost_eof = 0;
1911                 ost_done = 0;
1912                 do {
1913                         if (get_num_extents == 0) {
1914                                 /* Don't get too many extents. */
1915                                 if (current_extent + count_local >
1916                                     fiemap->fm_extent_count)
1917                                         count_local = fiemap->fm_extent_count -
1918                                                                  current_extent;
1919                         }
1920
1921                         lun_start += len_mapped_single_call;
1922                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
1923                         req_fm_len = fm_local->fm_length;
1924                         fm_local->fm_extent_count = count_local;
1925                         fm_local->fm_mapped_extents = 0;
1926                         fm_local->fm_flags = fiemap->fm_flags;
1927
1928                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1929                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1930
1931                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1932                                 GOTO(out, rc = -EINVAL);
1933
1934                         /* If OST is inactive, return extent with UNKNOWN flag */
1935                         if (!lov->lov_tgts[ost_index]->ltd_active) {
1936                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1937                                 fm_local->fm_mapped_extents = 1;
1938
1939                                 lcl_fm_ext[0].fe_logical = lun_start;
1940                                 lcl_fm_ext[0].fe_length = obd_object_end -
1941                                                                       lun_start;
1942                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1943
1944                                 goto inactive_tgt;
1945                         }
1946
1947                         fm_local->fm_start = lun_start;
1948                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1949                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1950                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1951                         rc = obd_get_info(NULL,
1952                                           lov->lov_tgts[ost_index]->ltd_exp,
1953                                           keylen, key, vallen, fm_local, lsm);
1954                         if (rc != 0)
1955                                 GOTO(out, rc);
1956
1957 inactive_tgt:
1958                         ext_count = fm_local->fm_mapped_extents;
1959                         if (ext_count == 0) {
1960                                 ost_done = 1;
1961                                 /* If last stripe has hole at the end,
1962                                  * then we need to return */
1963                                 if (cur_stripe_wrap == last_stripe) {
1964                                         fiemap->fm_mapped_extents = 0;
1965                                         goto finish;
1966                                 }
1967                                 break;
1968                         }
1969
1970                         /* If we just need num of extents then go to next device */
1971                         if (get_num_extents) {
1972                                 current_extent += ext_count;
1973                                 break;
1974                         }
1975
1976                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1977                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1978
1979                         /* Have we finished mapping on this device? */
1980                         if (req_fm_len <= len_mapped_single_call)
1981                                 ost_done = 1;
1982
1983                         /* Clear the EXTENT_LAST flag which can be present on
1984                          * last extent */
1985                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1986                                 lcl_fm_ext[ext_count - 1].fe_flags &=
1987                                                             ~FIEMAP_EXTENT_LAST;
1988
1989                         curr_loc = lov_stripe_size(lsm,
1990                                            lcl_fm_ext[ext_count - 1].fe_logical+
1991                                            lcl_fm_ext[ext_count - 1].fe_length,
1992                                            cur_stripe);
1993                         if (curr_loc >= fm_key->oa.o_size)
1994                                 ost_eof = 1;
1995
1996                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1997                                                      ost_index, ext_count,
1998                                                      current_extent);
1999
2000                         current_extent += ext_count;
2001
2002                         /* Ran out of available extents? */
2003                         if (current_extent >= fiemap->fm_extent_count)
2004                                 goto finish;
2005                 } while (ost_done == 0 && ost_eof == 0);
2006
2007                 if (cur_stripe_wrap == last_stripe)
2008                         goto finish;
2009         }
2010
2011 finish:
2012         /* Indicate that we are returning device offsets unless file just has
2013          * single stripe */
2014         if (lsm->lsm_stripe_count > 1)
2015                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2016
2017         if (get_num_extents)
2018                 goto skip_last_device_calc;
2019
2020         /* Check if we have reached the last stripe and whether mapping for that
2021          * stripe is done. */
2022         if (cur_stripe_wrap == last_stripe) {
2023                 if (ost_done || ost_eof)
2024                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2025                                                              FIEMAP_EXTENT_LAST;
2026         }
2027
2028 skip_last_device_calc:
2029         fiemap->fm_mapped_extents = current_extent;
2030
2031 out:
2032         if (fm_local)
2033                 OBD_FREE_LARGE(fm_local, buffer_size);
2034         return rc;
2035 }
2036
2037 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2038                         __u32 keylen, void *key, __u32 *vallen, void *val,
2039                         struct lov_stripe_md *lsm)
2040 {
2041         struct obd_device *obddev = class_exp2obd(exp);
2042         struct lov_obd *lov = &obddev->u.lov;
2043         int i, rc;
2044         ENTRY;
2045
2046         if (!vallen || !val)
2047                 RETURN(-EFAULT);
2048
2049         obd_getref(obddev);
2050
2051         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2052                 struct {
2053                         char name[16];
2054                         struct ldlm_lock *lock;
2055                 } *data = key;
2056                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2057                 struct lov_oinfo *loi;
2058                 __u32 *stripe = val;
2059
2060                 if (*vallen < sizeof(*stripe))
2061                         GOTO(out, rc = -EFAULT);
2062                 *vallen = sizeof(*stripe);
2063
2064                 /* XXX This is another one of those bits that will need to
2065                  * change if we ever actually support nested LOVs.  It uses
2066                  * the lock's export to find out which stripe it is. */
2067                 /* XXX - it's assumed all the locks for deleted OSTs have
2068                  * been cancelled. Also, the export for deleted OSTs will
2069                  * be NULL and won't match the lock's export. */
2070                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2071                         loi = lsm->lsm_oinfo[i];
2072                         if (!lov->lov_tgts[loi->loi_ost_idx])
2073                                 continue;
2074                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2075                             data->lock->l_conn_export &&
2076                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2077                                 *stripe = i;
2078                                 GOTO(out, rc = 0);
2079                         }
2080                 }
2081                 LDLM_ERROR(data->lock, "lock on inode without such object");
2082                 dump_lsm(D_ERROR, lsm);
2083                 GOTO(out, rc = -ENXIO);
2084         } else if (KEY_IS(KEY_LAST_ID)) {
2085                 struct obd_id_info *info = val;
2086                 __u32 size = sizeof(obd_id);
2087                 struct lov_tgt_desc *tgt;
2088
2089                 LASSERT(*vallen == sizeof(struct obd_id_info));
2090                 tgt = lov->lov_tgts[info->idx];
2091
2092                 if (!tgt || !tgt->ltd_active)
2093                         GOTO(out, rc = -ESRCH);
2094
2095                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2096                                   &size, info->data, NULL);
2097                 GOTO(out, rc = 0);
2098         } else if (KEY_IS(KEY_LOVDESC)) {
2099                 struct lov_desc *desc_ret = val;
2100                 *desc_ret = lov->desc;
2101
2102                 GOTO(out, rc = 0);
2103         } else if (KEY_IS(KEY_FIEMAP)) {
2104                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2105                 GOTO(out, rc);
2106         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2107                 struct lov_tgt_desc *tgt;
2108                 __u64 ost_idx = *((__u64*)val);
2109
2110                 LASSERT(*vallen == sizeof(__u64));
2111                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2112                 tgt = lov->lov_tgts[ost_idx];
2113
2114                 if (!tgt || !tgt->ltd_exp)
2115                         GOTO(out, rc = -ESRCH);
2116
2117                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2118                 GOTO(out, rc = 0);
2119         } else if (KEY_IS(KEY_TGT_COUNT)) {
2120                 *((int *)val) = lov->desc.ld_tgt_count;
2121                 GOTO(out, rc = 0);
2122         }
2123
2124         rc = -EINVAL;
2125
2126 out:
2127         obd_putref(obddev);
2128         RETURN(rc);
2129 }
2130
2131 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2132                               obd_count keylen, void *key, obd_count vallen,
2133                               void *val, struct ptlrpc_request_set *set)
2134 {
2135         struct obd_device *obddev = class_exp2obd(exp);
2136         struct lov_obd *lov = &obddev->u.lov;
2137         obd_count count;
2138         int i, rc = 0, err;
2139         struct lov_tgt_desc *tgt;
2140         unsigned incr, check_uuid,
2141                  do_inactive, no_set;
2142         unsigned next_id = 0,  mds_con = 0, capa = 0;
2143         ENTRY;
2144
2145         incr = check_uuid = do_inactive = no_set = 0;
2146         if (set == NULL) {
2147                 no_set = 1;
2148                 set = ptlrpc_prep_set();
2149                 if (!set)
2150                         RETURN(-ENOMEM);
2151         }
2152
2153         obd_getref(obddev);
2154         count = lov->desc.ld_tgt_count;
2155
2156         if (KEY_IS(KEY_NEXT_ID)) {
2157                 count = vallen / sizeof(struct obd_id_info);
2158                 vallen = sizeof(obd_id);
2159                 incr = sizeof(struct obd_id_info);
2160                 do_inactive = 1;
2161                 next_id = 1;
2162         } else if (KEY_IS(KEY_CHECKSUM)) {
2163                 do_inactive = 1;
2164         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2165                 /* use defaults:  do_inactive = incr = 0; */
2166         } else if (KEY_IS(KEY_MDS_CONN)) {
2167                 mds_con = 1;
2168         } else if (KEY_IS(KEY_CAPA_KEY)) {
2169                 capa = 1;
2170         } else if (KEY_IS(KEY_CACHE_SET)) {
2171                 LASSERT(lov->lov_cache == NULL);
2172                 lov->lov_cache = val;
2173                 do_inactive = 1;
2174         }
2175
2176         for (i = 0; i < count; i++, val = (char *)val + incr) {
2177                 if (next_id) {
2178                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2179                 } else {
2180                         tgt = lov->lov_tgts[i];
2181                 }
2182                 /* OST was disconnected */
2183                 if (!tgt || !tgt->ltd_exp)
2184                         continue;
2185
2186                 /* OST is inactive and we don't want inactive OSCs */
2187                 if (!tgt->ltd_active && !do_inactive)
2188                         continue;
2189
2190                 if (mds_con) {
2191                         struct mds_group_info *mgi;
2192
2193                         LASSERT(vallen == sizeof(*mgi));
2194                         mgi = (struct mds_group_info *)val;
2195
2196                         /* Only want a specific OSC */
2197                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2198                                                 &tgt->ltd_uuid))
2199                                 continue;
2200
2201                         err = obd_set_info_async(env, tgt->ltd_exp,
2202                                          keylen, key, sizeof(int),
2203                                          &mgi->group, set);
2204                 } else if (next_id) {
2205                         err = obd_set_info_async(env, tgt->ltd_exp,
2206                                          keylen, key, vallen,
2207                                          ((struct obd_id_info*)val)->data, set);
2208                 } else if (capa) {
2209                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2210
2211                         LASSERT(vallen == sizeof(*info));
2212
2213                          /* Only want a specific OSC */
2214                         if (info->uuid &&
2215                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2216                                 continue;
2217
2218                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2219                                                  key, sizeof(*info->capa),
2220                                                  info->capa, set);
2221                 } else {
2222                         /* Only want a specific OSC */
2223                         if (check_uuid &&
2224                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2225                                 continue;
2226
2227                         err = obd_set_info_async(env, tgt->ltd_exp,
2228                                          keylen, key, vallen, val, set);
2229                 }
2230
2231                 if (!rc)
2232                         rc = err;
2233         }
2234
2235         obd_putref(obddev);
2236         if (no_set) {
2237                 err = ptlrpc_set_wait(set);
2238                 if (!rc)
2239                         rc = err;
2240                 ptlrpc_set_destroy(set);
2241         }
2242         RETURN(rc);
2243 }
2244
2245 void lov_stripe_lock(struct lov_stripe_md *md)
2246 {
2247         LASSERT(md->lsm_lock_owner != current_pid());
2248         spin_lock(&md->lsm_lock);
2249         LASSERT(md->lsm_lock_owner == 0);
2250         md->lsm_lock_owner = current_pid();
2251 }
2252
2253 void lov_stripe_unlock(struct lov_stripe_md *md)
2254 {
2255         LASSERT(md->lsm_lock_owner == current_pid());
2256         md->lsm_lock_owner = 0;
2257         spin_unlock(&md->lsm_lock);
2258 }
2259
2260 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2261                         struct obd_quotactl *oqctl)
2262 {
2263         struct lov_obd      *lov = &obd->u.lov;
2264         struct lov_tgt_desc *tgt;
2265         __u64                curspace = 0;
2266         __u64                bhardlimit = 0;
2267         int                  i, rc = 0;
2268         ENTRY;
2269
2270         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2271             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2272             oqctl->qc_cmd != Q_GETOQUOTA &&
2273             oqctl->qc_cmd != Q_INITQUOTA &&
2274             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2275             oqctl->qc_cmd != Q_FINVALIDATE) {
2276                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2277                 RETURN(-EFAULT);
2278         }
2279
2280         /* for lov tgt */
2281         obd_getref(obd);
2282         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2283                 int err;
2284
2285                 tgt = lov->lov_tgts[i];
2286
2287                 if (!tgt)
2288                         continue;
2289
2290                 if (!tgt->ltd_active || tgt->ltd_reap) {
2291                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2292                             lov->lov_tgts[i]->ltd_activate) {
2293                                 rc = -ENETDOWN;
2294                                 CERROR("ost %d is inactive\n", i);
2295                         } else {
2296                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2297                         }
2298                         continue;
2299                 }
2300
2301                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2302                 if (err) {
2303                         if (tgt->ltd_active && !rc)
2304                                 rc = err;
2305                         continue;
2306                 }
2307
2308                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2309                         curspace += oqctl->qc_dqblk.dqb_curspace;
2310                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2311                 }
2312         }
2313         obd_putref(obd);
2314
2315         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2316                 oqctl->qc_dqblk.dqb_curspace = curspace;
2317                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2318         }
2319         RETURN(rc);
2320 }
2321
2322 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2323                           struct obd_quotactl *oqctl)
2324 {
2325         struct lov_obd *lov = &obd->u.lov;
2326         int             i, rc = 0;
2327         ENTRY;
2328
2329         obd_getref(obd);
2330
2331         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2332                 if (!lov->lov_tgts[i])
2333                         continue;
2334
2335                 /* Skip quota check on the administratively disabled OSTs. */
2336                 if (!lov->lov_tgts[i]->ltd_activate) {
2337                         CWARN("lov idx %d was administratively disabled, "
2338                               "skip quotacheck on it.\n", i);
2339                         continue;
2340                 }
2341
2342                 if (!lov->lov_tgts[i]->ltd_active) {
2343                         CERROR("lov idx %d inactive\n", i);
2344                         rc = -EIO;
2345                         goto out;
2346                 }
2347         }
2348
2349         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2350                 int err;
2351
2352                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2353                         continue;
2354
2355                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2356                 if (err && !rc)
2357                         rc = err;
2358         }
2359
2360 out:
2361         obd_putref(obd);
2362
2363         RETURN(rc);
2364 }
2365
2366 static struct obd_ops lov_obd_ops = {
2367         .o_owner                = THIS_MODULE,
2368         .o_setup                = lov_setup,
2369         .o_precleanup           = lov_precleanup,
2370         .o_cleanup              = lov_cleanup,
2371         .o_connect              = lov_connect,
2372         .o_disconnect           = lov_disconnect,
2373         .o_statfs               = lov_statfs,
2374         .o_statfs_async         = lov_statfs_async,
2375         .o_packmd               = lov_packmd,
2376         .o_unpackmd             = lov_unpackmd,
2377         .o_create               = lov_create,
2378         .o_destroy              = lov_destroy,
2379         .o_getattr_async        = lov_getattr_async,
2380         .o_setattr_async        = lov_setattr_async,
2381         .o_change_cbdata        = lov_change_cbdata,
2382         .o_find_cbdata          = lov_find_cbdata,
2383         .o_iocontrol            = lov_iocontrol,
2384         .o_get_info             = lov_get_info,
2385         .o_set_info_async       = lov_set_info_async,
2386         .o_notify               = lov_notify,
2387         .o_pool_new             = lov_pool_new,
2388         .o_pool_rem             = lov_pool_remove,
2389         .o_pool_add             = lov_pool_add,
2390         .o_pool_del             = lov_pool_del,
2391         .o_getref               = lov_getref,
2392         .o_putref               = lov_putref,
2393         .o_quotactl             = lov_quotactl,
2394         .o_quotacheck           = lov_quotacheck,
2395 };
2396
2397 struct kmem_cache *lov_oinfo_slab;
2398
2399 extern struct lu_kmem_descr lov_caches[];
2400
2401 int __init lov_init(void)
2402 {
2403         bool enable_proc = true;
2404         struct obd_type *type;
2405         int rc;
2406         ENTRY;
2407
2408         /* print an address of _any_ initialized kernel symbol from this
2409          * module, to allow debugging with gdb that doesn't support data
2410          * symbols from modules.*/
2411         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2412
2413         rc = lu_kmem_init(lov_caches);
2414         if (rc)
2415                 return rc;
2416
2417         lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2418                                            sizeof(struct lov_oinfo), 0,
2419                                            SLAB_HWCACHE_ALIGN, NULL);
2420         if (lov_oinfo_slab == NULL) {
2421                 lu_kmem_fini(lov_caches);
2422                 return -ENOMEM;
2423         }
2424
2425         type = class_search_type(LUSTRE_LOD_NAME);
2426         if (type != NULL && type->typ_procsym != NULL)
2427                 enable_proc = false;
2428
2429         rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
2430 #ifndef HAVE_ONLY_PROCFS_SEQ
2431                                  NULL,
2432 #endif
2433                                  LUSTRE_LOV_NAME, &lov_device_type);
2434
2435         if (rc) {
2436                 kmem_cache_destroy(lov_oinfo_slab);
2437                 lu_kmem_fini(lov_caches);
2438         }
2439
2440         RETURN(rc);
2441 }
2442
2443 #ifdef __KERNEL__
2444 static void /*__exit*/ lov_exit(void)
2445 {
2446         class_unregister_type(LUSTRE_LOV_NAME);
2447         kmem_cache_destroy(lov_oinfo_slab);
2448         lu_kmem_fini(lov_caches);
2449 }
2450
2451 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2452 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2453 MODULE_LICENSE("GPL");
2454
2455 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2456 #endif