Whamcloud - gitweb
9f84ece07d75ea2367a9e1fbb48d0ed43b75dbf6
[fs/lustre-release.git] / lustre / lov / lov_obd.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 only,
10  * as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful, but
13  * WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * General Public License version 2 for more details (a copy is included
16  * in the LICENSE file that accompanied this code).
17  *
18  * You should have received a copy of the GNU General Public License
19  * version 2 along with this program; If not, see
20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
21  *
22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23  * CA 95054 USA or visit www.sun.com if you need additional information or
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
30  * Use is subject to license terms.
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 #ifndef EXPORT_SYMTAB
45 # define EXPORT_SYMTAB
46 #endif
47 #define DEBUG_SUBSYSTEM S_LOV
48 #ifdef __KERNEL__
49 #include <libcfs/libcfs.h>
50 #else
51 #include <liblustre.h>
52 #endif
53
54 #include <obd_support.h>
55 #include <lustre_lib.h>
56 #include <lustre_net.h>
57 #include <lustre/lustre_idl.h>
58 #include <lustre_dlm.h>
59 #include <lustre_mds.h>
60 #include <lustre_debug.h>
61 #include <obd_class.h>
62 #include <obd_lov.h>
63 #include <obd_ost.h>
64 #include <lprocfs_status.h>
65 #include <lustre_param.h>
66 #include <cl_object.h>
67 #include <lustre/ll_fiemap.h>
68
69 #include "lov_internal.h"
70
71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
72    Any function that expects lov_tgts to remain stationary must take a ref. */
73 static void lov_getref(struct obd_device *obd)
74 {
75         struct lov_obd *lov = &obd->u.lov;
76
77         /* nobody gets through here until lov_putref is done */
78         cfs_mutex_down(&lov->lov_lock);
79         cfs_atomic_inc(&lov->lov_refcount);
80         cfs_mutex_up(&lov->lov_lock);
81         return;
82 }
83
84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
85
86 static void lov_putref(struct obd_device *obd)
87 {
88         struct lov_obd *lov = &obd->u.lov;
89
90         cfs_mutex_down(&lov->lov_lock);
91         /* ok to dec to 0 more than once -- ltd_exp's will be null */
92         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
93                 CFS_LIST_HEAD(kill);
94                 int i;
95                 struct lov_tgt_desc *tgt, *n;
96                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
97                        lov->lov_death_row);
98                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
99                         tgt = lov->lov_tgts[i];
100
101                         if (!tgt || !tgt->ltd_reap)
102                                 continue;
103                         cfs_list_add(&tgt->ltd_kill, &kill);
104                         /* XXX - right now there is a dependency on ld_tgt_count
105                          * being the maximum tgt index for computing the
106                          * mds_max_easize. So we can't shrink it. */
107                         lov_ost_pool_remove(&lov->lov_packed, i);
108                         lov->lov_tgts[i] = NULL;
109                         lov->lov_death_row--;
110                 }
111                 cfs_mutex_up(&lov->lov_lock);
112
113                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
114                         cfs_list_del(&tgt->ltd_kill);
115                         /* Disconnect */
116                         __lov_del_obd(obd, tgt);
117                 }
118         } else {
119                 cfs_mutex_up(&lov->lov_lock);
120         }
121 }
122
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
124                               int activate);
125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
126                       enum obd_notify_event ev, void *data);
127
128
129 #define MAX_STRING_SIZE 128
130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
131                     struct obd_connect_data *data)
132 {
133         struct lov_obd *lov = &obd->u.lov;
134         struct obd_uuid tgt_uuid;
135         struct obd_device *tgt_obd;
136         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
137         struct obd_import *imp;
138
139 #ifdef __KERNEL__
140         cfs_proc_dir_entry_t *lov_proc_dir;
141 #endif
142         int rc;
143         ENTRY;
144
145         if (!lov->lov_tgts[index])
146                 RETURN(-EINVAL);
147
148         tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
149         tgt_obd = lov->lov_tgts[index]->ltd_obd;
150
151         if (!tgt_obd->obd_set_up) {
152                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
153                 RETURN(-EINVAL);
154         }
155
156         /* override the sp_me from lov */
157         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
158
159         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
160                 data->ocd_index = index;
161
162         /*
163          * Divine LOV knows that OBDs under it are OSCs.
164          */
165         imp = tgt_obd->u.cli.cl_import;
166
167         if (activate) {
168                 tgt_obd->obd_no_recov = 0;
169                 /* FIXME this is probably supposed to be
170                    ptlrpc_set_import_active.  Horrible naming. */
171                 ptlrpc_activate_import(imp);
172         }
173
174         rc = obd_register_observer(tgt_obd, obd);
175         if (rc) {
176                 CERROR("Target %s register_observer error %d\n",
177                        obd_uuid2str(&tgt_uuid), rc);
178                 RETURN(rc);
179         }
180
181
182         if (imp->imp_invalid) {
183                 CERROR("not connecting OSC %s; administratively "
184                        "disabled\n", obd_uuid2str(&tgt_uuid));
185                 RETURN(0);
186         }
187
188         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
189                          &lov_osc_uuid, data, NULL);
190         if (rc || !lov->lov_tgts[index]->ltd_exp) {
191                 CERROR("Target %s connect error %d\n",
192                        obd_uuid2str(&tgt_uuid), rc);
193                 RETURN(-ENODEV);
194         }
195
196         lov->lov_tgts[index]->ltd_reap = 0;
197
198         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
199                obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
200
201 #ifdef __KERNEL__
202         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
203         if (lov_proc_dir) {
204                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
205                 cfs_proc_dir_entry_t *osc_symlink;
206                 char name[MAX_STRING_SIZE];
207
208                 LASSERT(osc_obd != NULL);
209                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
210                 LASSERT(osc_obd->obd_type->typ_name != NULL);
211                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
212                          osc_obd->obd_type->typ_name,
213                          osc_obd->obd_name);
214                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
215                                                   name);
216                 if (osc_symlink == NULL) {
217                         CERROR("could not register LOV target "
218                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
219                                obd->obd_type->typ_name, obd->obd_name,
220                                osc_obd->obd_name);
221                         lprocfs_remove(&lov_proc_dir);
222                 }
223         }
224 #endif
225
226         rc = qos_add_tgt(obd, index);
227         if (rc)
228                 CERROR("qos_add_tgt failed %d\n", rc);
229
230         RETURN(0);
231 }
232
233 static int lov_connect(const struct lu_env *env,
234                        struct obd_export **exp, struct obd_device *obd,
235                        struct obd_uuid *cluuid, struct obd_connect_data *data,
236                        void *localdata)
237 {
238         struct lov_obd *lov = &obd->u.lov;
239         struct lov_tgt_desc *tgt;
240         struct lustre_handle conn;
241         int i, rc;
242         ENTRY;
243
244         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
245
246         rc = class_connect(&conn, obd, cluuid);
247         if (rc)
248                 RETURN(rc);
249
250         *exp = class_conn2export(&conn);
251
252         /* Why should there ever be more than 1 connect? */
253         lov->lov_connects++;
254         LASSERT(lov->lov_connects == 1);
255
256         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
257         if (data)
258                 lov->lov_ocd = *data;
259
260         obd_getref(obd);
261         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
262                 tgt = lov->lov_tgts[i];
263                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
264                         continue;
265                 /* Flags will be lowest common denominator */
266                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
267                 if (rc) {
268                         CERROR("%s: lov connect tgt %d failed: %d\n",
269                                obd->obd_name, i, rc);
270                         continue;
271                 }
272                 /* connect to administrative disabled ost */
273                 if (!lov->lov_tgts[i]->ltd_exp)
274                         continue;
275
276                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
277                                 OBD_NOTIFY_CONNECT, (void *)&i);
278                 if (rc) {
279                         CERROR("%s error sending notify %d\n",
280                                obd->obd_name, rc);
281                 }
282         }
283         obd_putref(obd);
284
285         RETURN(0);
286 }
287
288 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
289 {
290         cfs_proc_dir_entry_t *lov_proc_dir;
291         struct lov_obd *lov = &obd->u.lov;
292         struct obd_device *osc_obd;
293         int rc;
294         ENTRY;
295
296         osc_obd = class_exp2obd(tgt->ltd_exp);
297         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
298                obd->obd_name, osc_obd->obd_name);
299
300         if (tgt->ltd_active) {
301                 tgt->ltd_active = 0;
302                 lov->desc.ld_active_tgt_count--;
303                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
304         }
305
306         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
307         if (lov_proc_dir) {
308                 cfs_proc_dir_entry_t *osc_symlink;
309
310                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
311                 if (osc_symlink) {
312                         lprocfs_remove(&osc_symlink);
313                 } else {
314                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
315                                obd->obd_type->typ_name, obd->obd_name,
316                                osc_obd->obd_name);
317                 }
318         }
319
320         if (osc_obd) {
321                 /* Pass it on to our clients.
322                  * XXX This should be an argument to disconnect,
323                  * XXX not a back-door flag on the OBD.  Ah well.
324                  */
325                 osc_obd->obd_force = obd->obd_force;
326                 osc_obd->obd_fail = obd->obd_fail;
327                 osc_obd->obd_no_recov = obd->obd_no_recov;
328         }
329
330         obd_register_observer(osc_obd, NULL);
331
332         rc = obd_disconnect(tgt->ltd_exp);
333         if (rc) {
334                 CERROR("Target %s disconnect error %d\n",
335                        tgt->ltd_uuid.uuid, rc);
336                 rc = 0;
337         }
338
339         qos_del_tgt(obd, tgt);
340
341         tgt->ltd_exp = NULL;
342         RETURN(0);
343 }
344
345 static int lov_disconnect(struct obd_export *exp)
346 {
347         struct obd_device *obd = class_exp2obd(exp);
348         struct lov_obd *lov = &obd->u.lov;
349         int i, rc;
350         ENTRY;
351
352         if (!lov->lov_tgts)
353                 goto out;
354
355         /* Only disconnect the underlying layers on the final disconnect. */
356         lov->lov_connects--;
357         if (lov->lov_connects != 0) {
358                 /* why should there be more than 1 connect? */
359                 CERROR("disconnect #%d\n", lov->lov_connects);
360                 goto out;
361         }
362
363         /* Let's hold another reference so lov_del_obd doesn't spin through
364            putref every time */
365         obd_getref(obd);
366
367         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
368                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
369                         /* Disconnection is the last we know about an obd */
370                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
371                 }
372         }
373         obd_putref(obd);
374
375 out:
376         rc = class_disconnect(exp); /* bz 9811 */
377         RETURN(rc);
378 }
379
380 /* Error codes:
381  *
382  *  -EINVAL  : UUID can't be found in the LOV's target list
383  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
384  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
385  *  any >= 0 : is log target index
386  */
387 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
388                               int activate)
389 {
390         struct lov_obd *lov = &obd->u.lov;
391         struct lov_tgt_desc *tgt;
392         int index;
393         ENTRY;
394
395         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
396                lov, uuid->uuid, activate);
397
398         obd_getref(obd);
399         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
400                 tgt = lov->lov_tgts[index];
401                 if (!tgt || !tgt->ltd_exp)
402                         continue;
403
404                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
405                        index, obd_uuid2str(&tgt->ltd_uuid),
406                        tgt->ltd_exp->exp_handle.h_cookie);
407                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
408                         break;
409         }
410
411         if (index == lov->desc.ld_tgt_count)
412                 GOTO(out, index = -EINVAL);
413
414         if (lov->lov_tgts[index]->ltd_active == activate) {
415                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
416                        activate ? "" : "in");
417                 GOTO(out, index);
418         }
419
420         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
421                activate ? "" : "in");
422
423         lov->lov_tgts[index]->ltd_active = activate;
424
425         if (activate) {
426                 lov->desc.ld_active_tgt_count++;
427                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
428         } else {
429                 lov->desc.ld_active_tgt_count--;
430                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
431         }
432         /* remove any old qos penalty */
433         lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
434
435  out:
436         obd_putref(obd);
437         RETURN(index);
438 }
439
440 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
441                       enum obd_notify_event ev, void *data)
442 {
443         int rc = 0;
444         ENTRY;
445
446         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
447                 struct obd_uuid *uuid;
448
449                 LASSERT(watched);
450
451                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
452                         CERROR("unexpected notification of %s %s!\n",
453                                watched->obd_type->typ_name,
454                                watched->obd_name);
455                         RETURN(-EINVAL);
456                 }
457                 uuid = &watched->u.cli.cl_target_uuid;
458
459                 /* Set OSC as active before notifying the observer, so the
460                  * observer can use the OSC normally.
461                  */
462                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
463                 if (rc < 0) {
464                         CERROR("%sactivation of %s failed: %d\n",
465                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
466                                obd_uuid2str(uuid), rc);
467                         RETURN(rc);
468                 }
469                 /* active event should be pass lov target index as data */
470                 data = &rc;
471         }
472
473         /* Pass the notification up the chain. */
474         if (watched) {
475                 rc = obd_notify_observer(obd, watched, ev, data);
476         } else {
477                 /* NULL watched means all osc's in the lov (only for syncs) */
478                 /* sync event should be send lov idx as data */
479                 struct lov_obd *lov = &obd->u.lov;
480                 struct obd_device *tgt_obd;
481                 int i;
482                 obd_getref(obd);
483                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
484                         /* don't send sync event if target not
485                          * connected/activated */
486                         if (!lov->lov_tgts[i] ||
487                             !lov->lov_tgts[i]->ltd_active)
488                                  continue;
489
490                         if ((ev == OBD_NOTIFY_SYNC) ||
491                             (ev == OBD_NOTIFY_SYNC_NONBLOCK))
492                                 data = &i;
493
494                         tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
495
496                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
497                         if (rc) {
498                                 CERROR("%s: notify %s of %s failed %d\n",
499                                        obd->obd_name,
500                                        obd->obd_observer->obd_name,
501                                        tgt_obd->obd_name, rc);
502                                 break;
503                         }
504                 }
505                 obd_putref(obd);
506         }
507
508         RETURN(rc);
509 }
510
511 int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
512                    __u32 index, int gen, int active)
513 {
514         struct lov_obd *lov = &obd->u.lov;
515         struct lov_tgt_desc *tgt;
516         struct obd_device *tgt_obd;
517         int rc;
518         ENTRY;
519
520         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
521                uuidp->uuid, index, gen, active);
522
523         if (gen <= 0) {
524                 CERROR("request to add OBD %s with invalid generation: %d\n",
525                        uuidp->uuid, gen);
526                 RETURN(-EINVAL);
527         }
528
529         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
530                                         &obd->obd_uuid);
531         if (tgt_obd == NULL)
532                 RETURN(-EINVAL);
533
534         cfs_mutex_down(&lov->lov_lock);
535
536         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
537                 tgt = lov->lov_tgts[index];
538                 CERROR("UUID %s already assigned at LOV target index %d\n",
539                        obd_uuid2str(&tgt->ltd_uuid), index);
540                 cfs_mutex_up(&lov->lov_lock);
541                 RETURN(-EEXIST);
542         }
543
544         if (index >= lov->lov_tgt_size) {
545                 /* We need to reallocate the lov target array. */
546                 struct lov_tgt_desc **newtgts, **old = NULL;
547                 __u32 newsize, oldsize = 0;
548
549                 newsize = max(lov->lov_tgt_size, (__u32)2);
550                 while (newsize < index + 1)
551                         newsize = newsize << 1;
552                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
553                 if (newtgts == NULL) {
554                         cfs_mutex_up(&lov->lov_lock);
555                         RETURN(-ENOMEM);
556                 }
557
558                 if (lov->lov_tgt_size) {
559                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
560                                lov->lov_tgt_size);
561                         old = lov->lov_tgts;
562                         oldsize = lov->lov_tgt_size;
563                 }
564
565                 lov->lov_tgts = newtgts;
566                 lov->lov_tgt_size = newsize;
567 #ifdef __KERNEL__
568                 smp_rmb();
569 #endif
570                 if (old)
571                         OBD_FREE(old, sizeof(*old) * oldsize);
572
573                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
574                        lov->lov_tgts, lov->lov_tgt_size);
575         }
576
577         OBD_ALLOC_PTR(tgt);
578         if (!tgt) {
579                 cfs_mutex_up(&lov->lov_lock);
580                 RETURN(-ENOMEM);
581         }
582
583         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
584         if (rc) {
585                 cfs_mutex_up(&lov->lov_lock);
586                 OBD_FREE_PTR(tgt);
587                 RETURN(rc);
588         }
589
590         memset(tgt, 0, sizeof(*tgt));
591         tgt->ltd_uuid = *uuidp;
592         tgt->ltd_obd = tgt_obd;
593         /* XXX - add a sanity check on the generation number. */
594         tgt->ltd_gen = gen;
595         tgt->ltd_index = index;
596         tgt->ltd_activate = active;
597         lov->lov_tgts[index] = tgt;
598         if (index >= lov->desc.ld_tgt_count)
599                 lov->desc.ld_tgt_count = index + 1;
600
601         cfs_mutex_up(&lov->lov_lock);
602
603         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
604                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
605
606         if (lov->lov_connects == 0) {
607                 /* lov_connect hasn't been called yet. We'll do the
608                    lov_connect_obd on this target when that fn first runs,
609                    because we don't know the connect flags yet. */
610                 RETURN(0);
611         }
612
613         obd_getref(obd);
614
615         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
616         if (rc)
617                 GOTO(out, rc);
618
619         /* connect to administrative disabled ost */
620         if (!tgt->ltd_exp)
621                 GOTO(out, rc = 0);
622
623         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
624                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
625                         (void *)&index);
626
627 out:
628         if (rc) {
629                 CERROR("add failed (%d), deleting %s\n", rc,
630                        obd_uuid2str(&tgt->ltd_uuid));
631                 lov_del_target(obd, index, 0, 0);
632         }
633         obd_putref(obd);
634         RETURN(rc);
635 }
636
637 /* Schedule a target for deletion */
638 int lov_del_target(struct obd_device *obd, __u32 index,
639                    struct obd_uuid *uuidp, int gen)
640 {
641         struct lov_obd *lov = &obd->u.lov;
642         int count = lov->desc.ld_tgt_count;
643         int rc = 0;
644         ENTRY;
645
646         if (index >= count) {
647                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
648                        index, count);
649                 RETURN(-EINVAL);
650         }
651
652         obd_getref(obd);
653
654         if (!lov->lov_tgts[index]) {
655                 CERROR("LOV target at index %d is not setup.\n", index);
656                 GOTO(out, rc = -EINVAL);
657         }
658
659         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
660                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
661                        lov_uuid2str(lov, index), index,
662                        obd_uuid2str(uuidp));
663                 GOTO(out, rc = -EINVAL);
664         }
665
666         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
667                lov_uuid2str(lov, index), index,
668                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
669                lov->lov_tgts[index]->ltd_active);
670
671         lov->lov_tgts[index]->ltd_reap = 1;
672         lov->lov_death_row++;
673         /* we really delete it from obd_putref */
674 out:
675         obd_putref(obd);
676
677         RETURN(rc);
678 }
679
680 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
681 {
682         struct obd_device *osc_obd;
683
684         LASSERT(tgt);
685         LASSERT(tgt->ltd_reap);
686
687         osc_obd = class_exp2obd(tgt->ltd_exp);
688
689         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
690                tgt->ltd_uuid.uuid,
691                osc_obd ? osc_obd->obd_name : "<no obd>");
692
693         if (tgt->ltd_exp)
694                 lov_disconnect_obd(obd, tgt);
695
696         OBD_FREE_PTR(tgt);
697
698         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
699            do it ourselves. And we can't do it from lov_cleanup,
700            because we just lost our only reference to it. */
701         if (osc_obd)
702                 class_manual_cleanup(osc_obd);
703 }
704
705 void lov_fix_desc_stripe_size(__u64 *val)
706 {
707         if (*val < PTLRPC_MAX_BRW_SIZE) {
708                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
709                               PTLRPC_MAX_BRW_SIZE);
710                 *val = PTLRPC_MAX_BRW_SIZE;
711         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
712                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
713                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
714                               "multiple of %u)\n",
715                               *val, LOV_MIN_STRIPE_SIZE);
716         }
717 }
718
719 void lov_fix_desc_stripe_count(__u32 *val)
720 {
721         if (*val == 0)
722                 *val = 1;
723 }
724
725 void lov_fix_desc_pattern(__u32 *val)
726 {
727         /* from lov_setstripe */
728         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
729                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
730                 *val = 0;
731         }
732 }
733
734 void lov_fix_desc_qos_maxage(__u32 *val)
735 {
736         /* fix qos_maxage */
737         if (*val == 0)
738                 *val = QOS_DEFAULT_MAXAGE;
739 }
740
741 void lov_fix_desc(struct lov_desc *desc)
742 {
743         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
744         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
745         lov_fix_desc_pattern(&desc->ld_pattern);
746         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
747 }
748
749 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
750 {
751         struct lprocfs_static_vars lvars = { 0 };
752         struct lov_desc *desc;
753         struct lov_obd *lov = &obd->u.lov;
754         int rc;
755         ENTRY;
756
757         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
758                 CERROR("LOV setup requires a descriptor\n");
759                 RETURN(-EINVAL);
760         }
761
762         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
763
764         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
765                 CERROR("descriptor size wrong: %d > %d\n",
766                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
767                 RETURN(-EINVAL);
768         }
769
770         if (desc->ld_magic != LOV_DESC_MAGIC) {
771                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
772                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
773                                    obd->obd_name, desc);
774                             lustre_swab_lov_desc(desc);
775                 } else {
776                         CERROR("%s: Bad lov desc magic: %#x\n",
777                                obd->obd_name, desc->ld_magic);
778                         RETURN(-EINVAL);
779                 }
780         }
781
782         lov_fix_desc(desc);
783
784         desc->ld_active_tgt_count = 0;
785         lov->desc = *desc;
786         lov->lov_tgt_size = 0;
787
788         cfs_sema_init(&lov->lov_lock, 1);
789         cfs_atomic_set(&lov->lov_refcount, 0);
790         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
791         cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
792         lov->lov_sp_me = LUSTRE_SP_CLI;
793         lov->lov_qos.lq_dirty = 1;
794         lov->lov_qos.lq_rr.lqr_dirty = 1;
795         lov->lov_qos.lq_reset = 1;
796         /* Default priority is toward free space balance */
797         lov->lov_qos.lq_prio_free = 232;
798         /* Default threshold for rr (roughly 17%) */
799         lov->lov_qos.lq_threshold_rr = 43;
800         /* Init statfs fields */
801         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
802         if (NULL == lov->lov_qos.lq_statfs_data)
803                 RETURN(-ENOMEM);
804         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
805
806         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
807                                                    HASH_POOLS_CUR_BITS,
808                                                    &pool_hash_operations, 0);
809         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
810         lov->lov_pool_count = 0;
811         rc = lov_ost_pool_init(&lov->lov_packed, 0);
812         if (rc)
813                 RETURN(rc);
814         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
815         if (rc) {
816                 lov_ost_pool_free(&lov->lov_packed);
817                 RETURN(rc);
818         }
819
820         lprocfs_lov_init_vars(&lvars);
821         lprocfs_obd_setup(obd, lvars.obd_vars);
822 #ifdef LPROCFS
823         {
824                 int rc;
825
826                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
827                                         0444, &lov_proc_target_fops, obd);
828                 if (rc)
829                         CWARN("Error adding the target_obd file\n");
830         }
831 #endif
832         lov->lov_pool_proc_entry = lprocfs_register("pools",
833                                                     obd->obd_proc_entry,
834                                                     NULL, NULL);
835
836         RETURN(0);
837 }
838
839 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
840 {
841         int rc = 0;
842         struct lov_obd *lov = &obd->u.lov;
843
844         ENTRY;
845
846         switch (stage) {
847         case OBD_CLEANUP_EARLY: {
848                 int i;
849                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
850                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
851                                 continue;
852                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
853                                        OBD_CLEANUP_EARLY);
854                 }
855                 break;
856         }
857         case OBD_CLEANUP_EXPORTS:
858                 rc = obd_llog_finish(obd, 0);
859                 if (rc != 0)
860                         CERROR("failed to cleanup llogging subsystems\n");
861                 break;
862         }
863         RETURN(rc);
864 }
865
866 static int lov_cleanup(struct obd_device *obd)
867 {
868         struct lov_obd *lov = &obd->u.lov;
869         cfs_list_t *pos, *tmp;
870         struct pool_desc *pool;
871
872         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
873                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
874                 /* free pool structs */
875                 CDEBUG(D_INFO, "delete pool %p\n", pool);
876                 lov_pool_del(obd, pool->pool_name);
877         }
878         cfs_hash_destroy(lov->lov_pools_hash_body);
879         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
880         lov_ost_pool_free(&lov->lov_packed);
881
882         if (lov->lov_tgts) {
883                 int i;
884                 obd_getref(obd);
885                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
886                         if (!lov->lov_tgts[i])
887                                 continue;
888
889                         /* Inactive targets may never have connected */
890                         if (lov->lov_tgts[i]->ltd_active ||
891                             cfs_atomic_read(&lov->lov_refcount))
892                             /* We should never get here - these
893                                should have been removed in the
894                              disconnect. */
895                                 CERROR("lov tgt %d not cleaned!"
896                                        " deathrow=%d, lovrc=%d\n",
897                                        i, lov->lov_death_row,
898                                        cfs_atomic_read(&lov->lov_refcount));
899                         lov_del_target(obd, i, 0, 0);
900                 }
901                 obd_putref(obd);
902                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
903                          lov->lov_tgt_size);
904                 lov->lov_tgt_size = 0;
905         }
906
907         /* clear pools parent proc entry only after all pools is killed */
908         lprocfs_obd_cleanup(obd);
909
910         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
911         RETURN(0);
912 }
913
914 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
915                             __u32 *indexp, int *genp)
916 {
917         struct obd_uuid obd_uuid;
918         int cmd;
919         int rc = 0;
920         ENTRY;
921
922         switch(cmd = lcfg->lcfg_command) {
923         case LCFG_LOV_ADD_OBD:
924         case LCFG_LOV_ADD_INA:
925         case LCFG_LOV_DEL_OBD: {
926                 __u32 index;
927                 int gen;
928                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
929                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
930                         GOTO(out, rc = -EINVAL);
931
932                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
933
934                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
935                         GOTO(out, rc = -EINVAL);
936                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
937                         GOTO(out, rc = -EINVAL);
938                 index = *indexp;
939                 gen = *genp;
940                 if (cmd == LCFG_LOV_ADD_OBD)
941                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
942                 else if (cmd == LCFG_LOV_ADD_INA)
943                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
944                 else
945                         rc = lov_del_target(obd, index, &obd_uuid, gen);
946                 GOTO(out, rc);
947         }
948         case LCFG_PARAM: {
949                 struct lprocfs_static_vars lvars = { 0 };
950                 struct lov_desc *desc = &(obd->u.lov.desc);
951
952                 if (!desc)
953                         GOTO(out, rc = -EINVAL);
954
955                 lprocfs_lov_init_vars(&lvars);
956
957                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
958                                               lcfg, obd);
959                 if (rc > 0)
960                         rc = 0;
961                 GOTO(out, rc);
962         }
963         case LCFG_POOL_NEW:
964         case LCFG_POOL_ADD:
965         case LCFG_POOL_DEL:
966         case LCFG_POOL_REM:
967                 GOTO(out, rc);
968
969         default: {
970                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
971                 GOTO(out, rc = -EINVAL);
972
973         }
974         }
975 out:
976         RETURN(rc);
977 }
978
979 #ifndef log2
980 #define log2(n) cfs_ffz(~(n))
981 #endif
982
983 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
984                              struct lov_stripe_md **ea,
985                              struct obd_trans_info *oti)
986 {
987         struct lov_obd *lov;
988         struct obdo *tmp_oa;
989         struct obd_uuid *ost_uuid = NULL;
990         int rc = 0, i;
991         ENTRY;
992
993         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
994                 src_oa->o_flags == OBD_FL_DELORPHAN);
995
996         lov = &export->exp_obd->u.lov;
997
998         OBDO_ALLOC(tmp_oa);
999         if (tmp_oa == NULL)
1000                 RETURN(-ENOMEM);
1001
1002         if (oti->oti_ost_uuid) {
1003                 ost_uuid = oti->oti_ost_uuid;
1004                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1005                        ost_uuid->uuid);
1006         }
1007
1008         obd_getref(export->exp_obd);
1009         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1010                 struct lov_stripe_md obj_md;
1011                 struct lov_stripe_md *obj_mdp = &obj_md;
1012                 struct lov_tgt_desc *tgt;
1013                 int err;
1014
1015                 tgt = lov->lov_tgts[i];
1016                 if (!tgt)
1017                         continue;
1018
1019                 /* if called for a specific target, we don't
1020                    care if it is not active. */
1021                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1022                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1023                         continue;
1024                 }
1025
1026                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1027                         continue;
1028
1029                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1030                        obd_uuid2str(ost_uuid));
1031
1032                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1033
1034                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1035                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1036                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1037                                  tmp_oa, &obj_mdp, oti);
1038                 if (err) {
1039                         /* This export will be disabled until it is recovered,
1040                            and then orphan recovery will be completed. */
1041                         CERROR("error in orphan recovery on OST idx %d/%d: "
1042                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1043                         rc = err;
1044                 }
1045
1046                 if (ost_uuid)
1047                         break;
1048         }
1049         obd_putref(export->exp_obd);
1050
1051         OBDO_FREE(tmp_oa);
1052         RETURN(rc);
1053 }
1054
1055 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1056                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1057 {
1058         struct lov_stripe_md *obj_mdp, *lsm;
1059         struct lov_obd *lov = &exp->exp_obd->u.lov;
1060         unsigned ost_idx;
1061         int rc, i;
1062         ENTRY;
1063
1064         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1065                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1066
1067         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1068         if (obj_mdp == NULL)
1069                 RETURN(-ENOMEM);
1070
1071         ost_idx = src_oa->o_nlink;
1072         lsm = *ea;
1073         if (lsm == NULL)
1074                 GOTO(out, rc = -EINVAL);
1075         if (ost_idx >= lov->desc.ld_tgt_count ||
1076             !lov->lov_tgts[ost_idx])
1077                 GOTO(out, rc = -EINVAL);
1078
1079         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1080                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1081                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1082                                 GOTO(out, rc = -EINVAL);
1083                         break;
1084                 }
1085         }
1086         if (i == lsm->lsm_stripe_count)
1087                 GOTO(out, rc = -EINVAL);
1088
1089         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1090 out:
1091         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1092         RETURN(rc);
1093 }
1094
1095 /* the LOV expects oa->o_id to be set to the LOV object id */
1096 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1097                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1098 {
1099         struct lov_obd *lov;
1100         struct obd_info oinfo;
1101         struct lov_request_set *set = NULL;
1102         struct lov_request *req;
1103         struct l_wait_info  lwi = { 0 };
1104         int rc = 0;
1105         ENTRY;
1106
1107         LASSERT(ea != NULL);
1108         if (exp == NULL)
1109                 RETURN(-EINVAL);
1110
1111         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1112             src_oa->o_flags == OBD_FL_DELORPHAN) {
1113                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1114                 RETURN(rc);
1115         }
1116
1117         lov = &exp->exp_obd->u.lov;
1118         if (!lov->desc.ld_active_tgt_count)
1119                 RETURN(-EIO);
1120
1121         obd_getref(exp->exp_obd);
1122         /* Recreate a specific object id at the given OST index */
1123         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1124             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1125                  rc = lov_recreate(exp, src_oa, ea, oti);
1126                  GOTO(out, rc);
1127         }
1128
1129         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1130          * later in alloc_qos(), we will wait for those rpcs to complete if
1131          * the osfs age is older than 2 * qos_maxage */
1132         qos_statfs_update(exp->exp_obd,
1133                           cfs_time_shift_64(-lov->desc.ld_qos_maxage) + CFS_HZ,
1134                           0);
1135
1136         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1137         if (rc)
1138                 GOTO(out, rc);
1139
1140         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1141                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1142                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1143                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1144         }
1145
1146         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1147          * longer then this */
1148         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1149
1150         /* we not have ptlrpc set for assign set->interpret and should
1151          * be call interpret function himself. calling from cb_create_update
1152          * not permited because lov_fini_create_set can sleep for long time,
1153          * but we must avoid sleeping in ptlrpcd interpret function. */
1154         rc = lov_fini_create_set(set, ea);
1155 out:
1156         obd_putref(exp->exp_obd);
1157         RETURN(rc);
1158 }
1159
1160 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1161 do {                                                                            \
1162         LASSERT((lsmp) != NULL);                                                \
1163         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1164                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1165                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1166 } while (0)
1167
1168 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1169                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1170                        struct obd_export *md_exp, void *capa)
1171 {
1172         struct lov_request_set *set;
1173         struct obd_info oinfo;
1174         struct lov_request *req;
1175         cfs_list_t *pos;
1176         struct lov_obd *lov;
1177         int rc = 0, err = 0;
1178         ENTRY;
1179
1180         ASSERT_LSM_MAGIC(lsm);
1181
1182         if (!exp || !exp->exp_obd)
1183                 RETURN(-ENODEV);
1184
1185         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1186                 LASSERT(oti);
1187                 LASSERT(oti->oti_logcookies);
1188         }
1189
1190         lov = &exp->exp_obd->u.lov;
1191         obd_getref(exp->exp_obd);
1192         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1193         if (rc)
1194                 GOTO(out, rc);
1195
1196         cfs_list_for_each (pos, &set->set_list) {
1197                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1198
1199                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1200                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1201
1202                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1203                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1204                 err = lov_update_common_set(set, req, err);
1205                 if (err) {
1206                         CERROR("error: destroying objid "LPX64" subobj "
1207                                LPX64" on OST idx %d: rc = %d\n",
1208                                oa->o_id, req->rq_oi.oi_oa->o_id,
1209                                req->rq_idx, err);
1210                         if (!rc)
1211                                 rc = err;
1212                 }
1213         }
1214
1215         if (rc == 0) {
1216                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1217                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1218         }
1219         err = lov_fini_destroy_set(set);
1220 out:
1221         obd_putref(exp->exp_obd);
1222         RETURN(rc ? rc : err);
1223 }
1224
1225 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1226 {
1227         struct lov_request_set *set;
1228         struct lov_request *req;
1229         cfs_list_t *pos;
1230         struct lov_obd *lov;
1231         int err = 0, rc = 0;
1232         ENTRY;
1233
1234         LASSERT(oinfo);
1235         ASSERT_LSM_MAGIC(oinfo->oi_md);
1236
1237         if (!exp || !exp->exp_obd)
1238                 RETURN(-ENODEV);
1239
1240         lov = &exp->exp_obd->u.lov;
1241
1242         rc = lov_prep_getattr_set(exp, oinfo, &set);
1243         if (rc)
1244                 RETURN(rc);
1245
1246         cfs_list_for_each (pos, &set->set_list) {
1247                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1248
1249                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1250                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1251                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1252
1253                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1254                                  &req->rq_oi);
1255                 err = lov_update_common_set(set, req, rc);
1256                 if (err) {
1257                         CERROR("error: getattr objid "LPX64" subobj "
1258                                LPX64" on OST idx %d: rc = %d\n",
1259                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1260                                req->rq_idx, err);
1261                         break;
1262                 }
1263         }
1264
1265         rc = lov_fini_getattr_set(set);
1266         if (err)
1267                 rc = err;
1268         RETURN(rc);
1269 }
1270
1271 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1272                                  void *data, int rc)
1273 {
1274         struct lov_request_set *lovset = (struct lov_request_set *)data;
1275         int err;
1276         ENTRY;
1277
1278         /* don't do attribute merge if this aysnc op failed */
1279         if (rc)
1280                 lovset->set_completes = 0;
1281         err = lov_fini_getattr_set(lovset);
1282         RETURN(rc ? rc : err);
1283 }
1284
1285 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1286                               struct ptlrpc_request_set *rqset)
1287 {
1288         struct lov_request_set *lovset;
1289         struct lov_obd *lov;
1290         cfs_list_t *pos;
1291         struct lov_request *req;
1292         int rc = 0, err;
1293         ENTRY;
1294
1295         LASSERT(oinfo);
1296         ASSERT_LSM_MAGIC(oinfo->oi_md);
1297
1298         if (!exp || !exp->exp_obd)
1299                 RETURN(-ENODEV);
1300
1301         lov = &exp->exp_obd->u.lov;
1302
1303         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1304         if (rc)
1305                 RETURN(rc);
1306
1307         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1308                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1309                oinfo->oi_md->lsm_stripe_size);
1310
1311         cfs_list_for_each (pos, &lovset->set_list) {
1312                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1313
1314                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1315                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1316                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1317                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1318                                        &req->rq_oi, rqset);
1319                 if (rc) {
1320                         CERROR("error: getattr objid "LPX64" subobj "
1321                                LPX64" on OST idx %d: rc = %d\n",
1322                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1323                                req->rq_idx, rc);
1324                         GOTO(out, rc);
1325                 }
1326         }
1327
1328         if (!cfs_list_empty(&rqset->set_requests)) {
1329                 LASSERT(rc == 0);
1330                 LASSERT (rqset->set_interpret == NULL);
1331                 rqset->set_interpret = lov_getattr_interpret;
1332                 rqset->set_arg = (void *)lovset;
1333                 RETURN(rc);
1334         }
1335 out:
1336         if (rc)
1337                 lovset->set_completes = 0;
1338         err = lov_fini_getattr_set(lovset);
1339         RETURN(rc ? rc : err);
1340 }
1341
1342 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1343                        struct obd_trans_info *oti)
1344 {
1345         struct lov_request_set *set;
1346         struct lov_obd *lov;
1347         cfs_list_t *pos;
1348         struct lov_request *req;
1349         int err = 0, rc = 0;
1350         ENTRY;
1351
1352         LASSERT(oinfo);
1353         ASSERT_LSM_MAGIC(oinfo->oi_md);
1354
1355         if (!exp || !exp->exp_obd)
1356                 RETURN(-ENODEV);
1357
1358         /* for now, we only expect the following updates here */
1359         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1360                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1361                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1362                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1363                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1364                                             OBD_MD_FLGID | OBD_MD_FLFID |
1365                                             OBD_MD_FLGENER)));
1366         lov = &exp->exp_obd->u.lov;
1367         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1368         if (rc)
1369                 RETURN(rc);
1370
1371         cfs_list_for_each (pos, &set->set_list) {
1372                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1373
1374                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1375                                  &req->rq_oi, NULL);
1376                 err = lov_update_setattr_set(set, req, rc);
1377                 if (err) {
1378                         CERROR("error: setattr objid "LPX64" subobj "
1379                                LPX64" on OST idx %d: rc = %d\n",
1380                                set->set_oi->oi_oa->o_id,
1381                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1382                         if (!rc)
1383                                 rc = err;
1384                 }
1385         }
1386         err = lov_fini_setattr_set(set);
1387         if (!rc)
1388                 rc = err;
1389         RETURN(rc);
1390 }
1391
1392 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1393                                  void *data, int rc)
1394 {
1395         struct lov_request_set *lovset = (struct lov_request_set *)data;
1396         int err;
1397         ENTRY;
1398
1399         if (rc)
1400                 lovset->set_completes = 0;
1401         err = lov_fini_setattr_set(lovset);
1402         RETURN(rc ? rc : err);
1403 }
1404
1405 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1406    needed. Otherwise, a client is waiting for responses. */
1407 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1408                              struct obd_trans_info *oti,
1409                              struct ptlrpc_request_set *rqset)
1410 {
1411         struct lov_request_set *set;
1412         struct lov_request *req;
1413         cfs_list_t *pos;
1414         struct lov_obd *lov;
1415         int rc = 0;
1416         ENTRY;
1417
1418         LASSERT(oinfo);
1419         ASSERT_LSM_MAGIC(oinfo->oi_md);
1420         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1421                 LASSERT(oti);
1422                 LASSERT(oti->oti_logcookies);
1423         }
1424
1425         if (!exp || !exp->exp_obd)
1426                 RETURN(-ENODEV);
1427
1428         lov = &exp->exp_obd->u.lov;
1429         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1430         if (rc)
1431                 RETURN(rc);
1432
1433         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1434                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1435                oinfo->oi_md->lsm_stripe_size);
1436
1437         cfs_list_for_each (pos, &set->set_list) {
1438                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1439
1440                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1441                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1442
1443                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1444                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1445                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1446
1447                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1448                                        &req->rq_oi, oti, rqset);
1449                 if (rc) {
1450                         CERROR("error: setattr objid "LPX64" subobj "
1451                                LPX64" on OST idx %d: rc = %d\n",
1452                                set->set_oi->oi_oa->o_id,
1453                                req->rq_oi.oi_oa->o_id,
1454                                req->rq_idx, rc);
1455                         break;
1456                 }
1457         }
1458
1459         /* If we are not waiting for responses on async requests, return. */
1460         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1461                 int err;
1462                 if (rc)
1463                         set->set_completes = 0;
1464                 err = lov_fini_setattr_set(set);
1465                 RETURN(rc ? rc : err);
1466         }
1467
1468         LASSERT(rqset->set_interpret == NULL);
1469         rqset->set_interpret = lov_setattr_interpret;
1470         rqset->set_arg = (void *)set;
1471
1472         RETURN(0);
1473 }
1474
1475 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1476                                void *data, int rc)
1477 {
1478         struct lov_request_set *lovset = (struct lov_request_set *)data;
1479         int err;
1480         ENTRY;
1481
1482         if (rc)
1483                 lovset->set_completes = 0;
1484         err = lov_fini_punch_set(lovset);
1485         RETURN(rc ? rc : err);
1486 }
1487
1488 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1489  * we can send this 'punch' to just the authoritative node and the nodes
1490  * that the punch will affect. */
1491 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1492                      struct obd_trans_info *oti,
1493                      struct ptlrpc_request_set *rqset)
1494 {
1495         struct lov_request_set *set;
1496         struct lov_obd *lov;
1497         cfs_list_t *pos;
1498         struct lov_request *req;
1499         int rc = 0;
1500         ENTRY;
1501
1502         LASSERT(oinfo);
1503         ASSERT_LSM_MAGIC(oinfo->oi_md);
1504
1505         if (!exp || !exp->exp_obd)
1506                 RETURN(-ENODEV);
1507
1508         lov = &exp->exp_obd->u.lov;
1509         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1510         if (rc)
1511                 RETURN(rc);
1512
1513         cfs_list_for_each (pos, &set->set_list) {
1514                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1515
1516                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1517                                &req->rq_oi, NULL, rqset);
1518                 if (rc) {
1519                         CERROR("error: punch objid "LPX64" subobj "LPX64
1520                                " on OST idx %d: rc = %d\n",
1521                                set->set_oi->oi_oa->o_id,
1522                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1523                         break;
1524                 }
1525         }
1526
1527         if (rc || cfs_list_empty(&rqset->set_requests)) {
1528                 int err;
1529                 err = lov_fini_punch_set(set);
1530                 RETURN(rc ? rc : err);
1531         }
1532
1533         LASSERT(rqset->set_interpret == NULL);
1534         rqset->set_interpret = lov_punch_interpret;
1535         rqset->set_arg = (void *)set;
1536
1537         RETURN(0);
1538 }
1539
1540 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1541                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1542                     void *capa)
1543 {
1544         struct lov_request_set *set;
1545         struct obd_info oinfo;
1546         struct lov_obd *lov;
1547         cfs_list_t *pos;
1548         struct lov_request *req;
1549         int err = 0, rc = 0;
1550         ENTRY;
1551
1552         ASSERT_LSM_MAGIC(lsm);
1553
1554         if (!exp->exp_obd)
1555                 RETURN(-ENODEV);
1556
1557         lov = &exp->exp_obd->u.lov;
1558         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1559         if (rc)
1560                 RETURN(rc);
1561
1562         cfs_list_for_each (pos, &set->set_list) {
1563                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1564
1565                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1566                               req->rq_oi.oi_oa, NULL,
1567                               req->rq_oi.oi_policy.l_extent.start,
1568                               req->rq_oi.oi_policy.l_extent.end, capa);
1569                 err = lov_update_common_set(set, req, rc);
1570                 if (err) {
1571                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1572                                " on OST idx %d: rc = %d\n",
1573                                set->set_oi->oi_oa->o_id,
1574                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1575                         if (!rc)
1576                                 rc = err;
1577                 }
1578         }
1579         err = lov_fini_sync_set(set);
1580         if (!rc)
1581                 rc = err;
1582         RETURN(rc);
1583 }
1584
1585 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1586                          obd_count oa_bufs, struct brw_page *pga)
1587 {
1588         struct obd_info oinfo = { { { 0 } } };
1589         int i, rc = 0;
1590
1591         oinfo.oi_oa = lov_oinfo->oi_oa;
1592
1593         /* The caller just wants to know if there's a chance that this
1594          * I/O can succeed */
1595         for (i = 0; i < oa_bufs; i++) {
1596                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1597                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1598                 obd_off start, end;
1599
1600                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1601                                            pga[i].off + pga[i].count - 1,
1602                                            &start, &end))
1603                         continue;
1604
1605                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1606                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1607                         return -EIO;
1608                 }
1609
1610                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1611                              1, &pga[i], NULL);
1612                 if (rc)
1613                         break;
1614         }
1615         return rc;
1616 }
1617
1618 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1619                    obd_count oa_bufs, struct brw_page *pga,
1620                    struct obd_trans_info *oti)
1621 {
1622         struct lov_request_set *set;
1623         struct lov_request *req;
1624         cfs_list_t *pos;
1625         struct lov_obd *lov = &exp->exp_obd->u.lov;
1626         int err, rc = 0;
1627         ENTRY;
1628
1629         ASSERT_LSM_MAGIC(oinfo->oi_md);
1630
1631         if (cmd == OBD_BRW_CHECK) {
1632                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1633                 RETURN(rc);
1634         }
1635
1636         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1637         if (rc)
1638                 RETURN(rc);
1639
1640         cfs_list_for_each (pos, &set->set_list) {
1641                 struct obd_export *sub_exp;
1642                 struct brw_page *sub_pga;
1643                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1644
1645                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1646                 sub_pga = set->set_pga + req->rq_pgaidx;
1647                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1648                              sub_pga, oti);
1649                 if (rc)
1650                         break;
1651                 lov_update_common_set(set, req, rc);
1652         }
1653
1654         err = lov_fini_brw_set(set);
1655         if (!rc)
1656                 rc = err;
1657         RETURN(rc);
1658 }
1659
1660 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1661                                  void *data, int rc)
1662 {
1663         struct lov_request_set *lovset = (struct lov_request_set *)data;
1664         ENTRY;
1665         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1666         RETURN(rc);
1667 }
1668
1669 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1670                        struct ldlm_enqueue_info *einfo,
1671                        struct ptlrpc_request_set *rqset)
1672 {
1673         ldlm_mode_t mode = einfo->ei_mode;
1674         struct lov_request_set *set;
1675         struct lov_request *req;
1676         cfs_list_t *pos;
1677         struct lov_obd *lov;
1678         ldlm_error_t rc;
1679         ENTRY;
1680
1681         LASSERT(oinfo);
1682         ASSERT_LSM_MAGIC(oinfo->oi_md);
1683         LASSERT(mode == (mode & -mode));
1684
1685         /* we should never be asked to replay a lock this way. */
1686         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1687
1688         if (!exp || !exp->exp_obd)
1689                 RETURN(-ENODEV);
1690
1691         lov = &exp->exp_obd->u.lov;
1692         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1693         if (rc)
1694                 RETURN(rc);
1695
1696         cfs_list_for_each (pos, &set->set_list) {
1697                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1698
1699                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1700                                  &req->rq_oi, einfo, rqset);
1701                 if (rc != ELDLM_OK)
1702                         GOTO(out, rc);
1703         }
1704
1705         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1706                 LASSERT(rc == 0);
1707                 LASSERT(rqset->set_interpret == NULL);
1708                 rqset->set_interpret = lov_enqueue_interpret;
1709                 rqset->set_arg = (void *)set;
1710                 RETURN(rc);
1711         }
1712 out:
1713         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1714         RETURN(rc);
1715 }
1716
1717 static int lov_change_cbdata(struct obd_export *exp,
1718                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1719                              void *data)
1720 {
1721         struct lov_obd *lov;
1722         int rc = 0, i;
1723         ENTRY;
1724
1725         ASSERT_LSM_MAGIC(lsm);
1726
1727         if (!exp || !exp->exp_obd)
1728                 RETURN(-ENODEV);
1729
1730         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1731
1732         lov = &exp->exp_obd->u.lov;
1733         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1734                 struct lov_stripe_md submd;
1735                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1736
1737                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1738                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1739                         continue;
1740                 }
1741
1742                 submd.lsm_object_id = loi->loi_id;
1743                 submd.lsm_object_gr = lsm->lsm_object_gr;
1744                 submd.lsm_stripe_count = 0;
1745                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1746                                        &submd, it, data);
1747         }
1748         RETURN(rc);
1749 }
1750
1751 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1752                       __u32 mode, struct lustre_handle *lockh)
1753 {
1754         struct lov_request_set *set;
1755         struct obd_info oinfo;
1756         struct lov_request *req;
1757         cfs_list_t *pos;
1758         struct lov_obd *lov;
1759         struct lustre_handle *lov_lockhp;
1760         int err = 0, rc = 0;
1761         ENTRY;
1762
1763         ASSERT_LSM_MAGIC(lsm);
1764
1765         if (!exp || !exp->exp_obd)
1766                 RETURN(-ENODEV);
1767
1768         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1769         LASSERT(lockh);
1770         lov = &exp->exp_obd->u.lov;
1771         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1772         if (rc)
1773                 RETURN(rc);
1774
1775         cfs_list_for_each (pos, &set->set_list) {
1776                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1777                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1778
1779                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1780                                 req->rq_oi.oi_md, mode, lov_lockhp);
1781                 rc = lov_update_common_set(set, req, rc);
1782                 if (rc) {
1783                         CERROR("error: cancel objid "LPX64" subobj "
1784                                LPX64" on OST idx %d: rc = %d\n",
1785                                lsm->lsm_object_id,
1786                                req->rq_oi.oi_md->lsm_object_id,
1787                                req->rq_idx, rc);
1788                         err = rc;
1789                 }
1790
1791         }
1792         lov_fini_cancel_set(set);
1793         RETURN(err);
1794 }
1795
1796 static int lov_cancel_unused(struct obd_export *exp,
1797                              struct lov_stripe_md *lsm,
1798                              int flags, void *opaque)
1799 {
1800         struct lov_obd *lov;
1801         int rc = 0, i;
1802         ENTRY;
1803
1804         if (!exp || !exp->exp_obd)
1805                 RETURN(-ENODEV);
1806
1807         lov = &exp->exp_obd->u.lov;
1808         if (lsm == NULL) {
1809                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1810                         int err;
1811                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1812                                 continue;
1813
1814                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1815                                                 flags, opaque);
1816                         if (!rc)
1817                                 rc = err;
1818                 }
1819                 RETURN(rc);
1820         }
1821
1822         ASSERT_LSM_MAGIC(lsm);
1823
1824         LASSERT_MDS_GROUP(lsm->lsm_object_gr);
1825         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1826                 struct lov_stripe_md submd;
1827                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1828                 int err;
1829
1830                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1831                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1832                         continue;
1833                 }
1834
1835                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1836                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1837
1838                 submd.lsm_object_id = loi->loi_id;
1839                 submd.lsm_object_gr = lsm->lsm_object_gr;
1840                 submd.lsm_stripe_count = 0;
1841                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1842                                         &submd, flags, opaque);
1843                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1844                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1845                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1846                                loi->loi_id, loi->loi_ost_idx, err);
1847                         if (!rc)
1848                                 rc = err;
1849                 }
1850         }
1851         RETURN(rc);
1852 }
1853
1854 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1855 {
1856         struct lov_request_set *lovset = (struct lov_request_set *)data;
1857         int err;
1858         ENTRY;
1859
1860         if (rc)
1861                 lovset->set_completes = 0;
1862
1863         err = lov_fini_statfs_set(lovset);
1864         RETURN(rc ? rc : err);
1865 }
1866
1867 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
1868                             __u64 max_age, struct ptlrpc_request_set *rqset)
1869 {
1870         struct lov_request_set *set;
1871         struct lov_request *req;
1872         cfs_list_t *pos;
1873         struct lov_obd *lov;
1874         int rc = 0;
1875         ENTRY;
1876
1877         LASSERT(oinfo != NULL);
1878         LASSERT(oinfo->oi_osfs != NULL);
1879
1880         lov = &obd->u.lov;
1881         rc = lov_prep_statfs_set(obd, oinfo, &set);
1882         if (rc)
1883                 RETURN(rc);
1884
1885         cfs_list_for_each (pos, &set->set_list) {
1886                 struct obd_device *osc_obd;
1887
1888                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1889
1890                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
1891                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
1892                 if (rc)
1893                         break;
1894         }
1895
1896         if (rc || cfs_list_empty(&rqset->set_requests)) {
1897                 int err;
1898                 if (rc)
1899                         set->set_completes = 0;
1900                 err = lov_fini_statfs_set(set);
1901                 RETURN(rc ? rc : err);
1902         }
1903
1904         LASSERT(rqset->set_interpret == NULL);
1905         rqset->set_interpret = lov_statfs_interpret;
1906         rqset->set_arg = (void *)set;
1907         RETURN(0);
1908 }
1909
1910 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1911                       __u64 max_age, __u32 flags)
1912 {
1913         struct ptlrpc_request_set *set = NULL;
1914         struct obd_info oinfo = { { { 0 } } };
1915         int rc = 0;
1916         ENTRY;
1917
1918
1919         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1920          * statfs requests */
1921         set = ptlrpc_prep_set();
1922         if (set == NULL)
1923                 RETURN(-ENOMEM);
1924
1925         oinfo.oi_osfs = osfs;
1926         oinfo.oi_flags = flags;
1927         rc = lov_statfs_async(obd, &oinfo, max_age, set);
1928         if (rc == 0)
1929                 rc = ptlrpc_set_wait(set);
1930         ptlrpc_set_destroy(set);
1931
1932         RETURN(rc);
1933 }
1934
1935 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1936                          void *karg, void *uarg)
1937 {
1938         struct obd_device *obddev = class_exp2obd(exp);
1939         struct lov_obd *lov = &obddev->u.lov;
1940         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1941         struct obd_uuid *uuidp;
1942         ENTRY;
1943
1944         switch (cmd) {
1945         case IOC_OBD_STATFS: {
1946                 struct obd_ioctl_data *data = karg;
1947                 struct obd_device *osc_obd;
1948                 struct obd_statfs stat_buf = {0};
1949                 __u32 index;
1950
1951                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1952                 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
1953
1954                 if ((index >= count))
1955                         RETURN(-ENODEV);
1956
1957                 if (!lov->lov_tgts[index])
1958                         /* Try again with the next index */
1959                         RETURN(-EAGAIN);
1960                 if (!lov->lov_tgts[index]->ltd_active)
1961                         RETURN(-ENODATA);
1962
1963                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1964                 if (!osc_obd)
1965                         RETURN(-EINVAL);
1966
1967                 /* got statfs data */
1968                 rc = obd_statfs(osc_obd, &stat_buf,
1969                                 cfs_time_current_64() - CFS_HZ, 0);
1970                 if (rc)
1971                         RETURN(rc);
1972                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1973                                      data->ioc_plen1))
1974                         RETURN(-EFAULT);
1975                 /* copy UUID */
1976                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1977                                      data->ioc_plen2))
1978                         RETURN(-EFAULT);
1979                 break;
1980         }
1981         case OBD_IOC_LOV_GET_CONFIG: {
1982                 struct obd_ioctl_data *data;
1983                 struct lov_desc *desc;
1984                 char *buf = NULL;
1985                 __u32 *genp;
1986
1987                 len = 0;
1988                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1989                         RETURN(-EINVAL);
1990
1991                 data = (struct obd_ioctl_data *)buf;
1992
1993                 if (sizeof(*desc) > data->ioc_inllen1) {
1994                         obd_ioctl_freedata(buf, len);
1995                         RETURN(-EINVAL);
1996                 }
1997
1998                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1999                         obd_ioctl_freedata(buf, len);
2000                         RETURN(-EINVAL);
2001                 }
2002
2003                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2004                         obd_ioctl_freedata(buf, len);
2005                         RETURN(-EINVAL);
2006                 }
2007
2008                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2009                 memcpy(desc, &(lov->desc), sizeof(*desc));
2010
2011                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2012                 genp = (__u32 *)data->ioc_inlbuf3;
2013                 /* the uuid will be empty for deleted OSTs */
2014                 for (i = 0; i < count; i++, uuidp++, genp++) {
2015                         if (!lov->lov_tgts[i])
2016                                 continue;
2017                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2018                         *genp = lov->lov_tgts[i]->ltd_gen;
2019                 }
2020
2021                 if (cfs_copy_to_user((void *)uarg, buf, len))
2022                         rc = -EFAULT;
2023                 obd_ioctl_freedata(buf, len);
2024                 break;
2025         }
2026         case LL_IOC_LOV_SETSTRIPE:
2027                 rc = lov_setstripe(exp, len, karg, uarg);
2028                 break;
2029         case LL_IOC_LOV_GETSTRIPE:
2030                 rc = lov_getstripe(exp, karg, uarg);
2031                 break;
2032         case LL_IOC_LOV_SETEA:
2033                 rc = lov_setea(exp, karg, uarg);
2034                 break;
2035         case OBD_IOC_QUOTACTL: {
2036                 struct if_quotactl *qctl = karg;
2037                 struct lov_tgt_desc *tgt = NULL;
2038                 struct obd_quotactl *oqctl;
2039
2040                 if (qctl->qc_valid == QC_OSTIDX) {
2041                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2042                                 RETURN(-EINVAL);
2043
2044                         tgt = lov->lov_tgts[qctl->qc_idx];
2045                         if (!tgt || !tgt->ltd_exp)
2046                                 RETURN(-EINVAL);
2047                 } else if (qctl->qc_valid == QC_UUID) {
2048                         for (i = 0; i < count; i++) {
2049                                 tgt = lov->lov_tgts[i];
2050                                 if (!tgt ||
2051                                     !obd_uuid_equals(&tgt->ltd_uuid,
2052                                                      &qctl->obd_uuid))
2053                                         continue;
2054
2055                                 if (tgt->ltd_exp == NULL)
2056                                         RETURN(-EINVAL);
2057
2058                                 break;
2059                         }
2060                 } else {
2061                         RETURN(-EINVAL);
2062                 }
2063
2064                 if (i >= count)
2065                         RETURN(-EAGAIN);
2066
2067                 LASSERT(tgt && tgt->ltd_exp);
2068                 OBD_ALLOC_PTR(oqctl);
2069                 if (!oqctl)
2070                         RETURN(-ENOMEM);
2071
2072                 QCTL_COPY(oqctl, qctl);
2073                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2074                 if (rc == 0) {
2075                         QCTL_COPY(qctl, oqctl);
2076                         qctl->qc_valid = QC_OSTIDX;
2077                         qctl->obd_uuid = tgt->ltd_uuid;
2078                 }
2079                 OBD_FREE_PTR(oqctl);
2080                 break;
2081         }
2082         default: {
2083                 int set = 0;
2084
2085                 if (count == 0)
2086                         RETURN(-ENOTTY);
2087
2088                 for (i = 0; i < count; i++) {
2089                         int err;
2090
2091                         /* OST was disconnected */
2092                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2093                                 continue;
2094
2095                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2096                                             len, karg, uarg);
2097                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2098                                 RETURN(err);
2099                         } else if (err) {
2100                                 if (lov->lov_tgts[i]->ltd_active) {
2101                                         CDEBUG(err == -ENOTTY ?
2102                                                D_IOCTL : D_WARNING,
2103                                                "iocontrol OSC %s on OST "
2104                                                "idx %d cmd %x: err = %d\n",
2105                                                lov_uuid2str(lov, i),
2106                                                i, cmd, err);
2107                                         if (!rc)
2108                                                 rc = err;
2109                                 }
2110                         } else {
2111                                 set = 1;
2112                         }
2113                 }
2114                 if (!set && !rc)
2115                         rc = -EIO;
2116         }
2117         }
2118
2119         RETURN(rc);
2120 }
2121
2122 #define FIEMAP_BUFFER_SIZE 4096
2123
2124 /**
2125  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2126  * call. The local end offset and the device are sent in the first
2127  * fm_extent. This function calculates the stripe number from the index.
2128  * This function returns a stripe_no on which mapping is to be restarted.
2129  *
2130  * This function returns fm_end_offset which is the in-OST offset at which
2131  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2132  * will re-calculate proper offset in next stripe.
2133  * Note that the first extent is passed to lov_get_info via the value field.
2134  *
2135  * \param fiemap fiemap request header
2136  * \param lsm striping information for the file
2137  * \param fm_start logical start of mapping
2138  * \param fm_end logical end of mapping
2139  * \param start_stripe starting stripe will be returned in this
2140  */
2141 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2142                                    struct lov_stripe_md *lsm, obd_size fm_start,
2143                                    obd_size fm_end, int *start_stripe)
2144 {
2145         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2146         obd_off lun_start, lun_end;
2147         obd_size fm_end_offset;
2148         int stripe_no = -1, i;
2149
2150         if (fiemap->fm_extent_count == 0 ||
2151             fiemap->fm_extents[0].fe_logical == 0)
2152                 return 0;
2153
2154         /* Find out stripe_no from ost_index saved in the fe_device */
2155         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2156                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2157                                         fiemap->fm_extents[0].fe_device) {
2158                         stripe_no = i;
2159                         break;
2160                 }
2161         }
2162
2163         /* If we have finished mapping on previous device, shift logical
2164          * offset to start of next device */
2165         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2166                                    &lun_start, &lun_end)) != 0 &&
2167                                    local_end < lun_end) {
2168                 fm_end_offset = local_end;
2169                 *start_stripe = stripe_no;
2170         } else {
2171                 /* This is a special value to indicate that caller should
2172                  * calculate offset in next stripe. */
2173                 fm_end_offset = 0;
2174                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2175         }
2176
2177         return fm_end_offset;
2178 }
2179
2180 /**
2181  * We calculate on which OST the mapping will end. If the length of mapping
2182  * is greater than (stripe_size * stripe_count) then the last_stripe will
2183  * will be one just before start_stripe. Else we check if the mapping
2184  * intersects each OST and find last_stripe.
2185  * This function returns the last_stripe and also sets the stripe_count
2186  * over which the mapping is spread
2187  *
2188  * \param lsm striping information for the file
2189  * \param fm_start logical start of mapping
2190  * \param fm_end logical end of mapping
2191  * \param start_stripe starting stripe of the mapping
2192  * \param stripe_count the number of stripes across which to map is returned
2193  *
2194  * \retval last_stripe return the last stripe of the mapping
2195  */
2196 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2197                             obd_size fm_end, int start_stripe,
2198                             int *stripe_count)
2199 {
2200         int last_stripe;
2201         obd_off obd_start, obd_end;
2202         int i, j;
2203
2204         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2205                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2206                                                               start_stripe - 1);
2207                 *stripe_count = lsm->lsm_stripe_count;
2208         } else {
2209                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2210                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2211                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2212                                                    &obd_start, &obd_end)) == 0)
2213                                 break;
2214                 }
2215                 *stripe_count = j;
2216                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2217         }
2218
2219         return last_stripe;
2220 }
2221
2222 /**
2223  * Set fe_device and copy extents from local buffer into main return buffer.
2224  *
2225  * \param fiemap fiemap request header
2226  * \param lcl_fm_ext array of local fiemap extents to be copied
2227  * \param ost_index OST index to be written into the fm_device field for each
2228                     extent
2229  * \param ext_count number of extents to be copied
2230  * \param current_extent where to start copying in main extent array
2231  */
2232 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2233                                   struct ll_fiemap_extent *lcl_fm_ext,
2234                                   int ost_index, unsigned int ext_count,
2235                                   int current_extent)
2236 {
2237         char *to;
2238         int ext;
2239
2240         for (ext = 0; ext < ext_count; ext++) {
2241                 lcl_fm_ext[ext].fe_device = ost_index;
2242                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2243         }
2244
2245         /* Copy fm_extent's from fm_local to return buffer */
2246         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2247         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2248 }
2249
2250 /**
2251  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2252  * This also handles the restarting of FIEMAP calls in case mapping overflows
2253  * the available number of extents in single call.
2254  */
2255 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2256                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2257 {
2258         struct ll_fiemap_info_key *fm_key = key;
2259         struct ll_user_fiemap *fiemap = val;
2260         struct ll_user_fiemap *fm_local = NULL;
2261         struct ll_fiemap_extent *lcl_fm_ext;
2262         int count_local;
2263         unsigned int get_num_extents = 0;
2264         int ost_index = 0, actual_start_stripe, start_stripe;
2265         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2266         obd_size curr_loc;
2267         int current_extent = 0, rc = 0, i;
2268         int ost_eof = 0; /* EOF for object */
2269         int ost_done = 0; /* done with required mapping for this OST? */
2270         int last_stripe;
2271         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2272         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2273
2274         if (lsm == NULL)
2275                 GOTO(out, rc = 0);
2276
2277         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2278                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2279
2280         OBD_ALLOC(fm_local, buffer_size);
2281         if (fm_local == NULL)
2282                 GOTO(out, rc = -ENOMEM);
2283         lcl_fm_ext = &fm_local->fm_extents[0];
2284
2285         count_local = fiemap_size_to_count(buffer_size);
2286
2287         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2288         fm_start = fiemap->fm_start;
2289         fm_length = fiemap->fm_length;
2290         /* Calculate start stripe, last stripe and length of mapping */
2291         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2292         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2293                                                 fm_start + fm_length - 1);
2294         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2295         if (fm_end > fm_key->oa.o_size)
2296                 fm_end = fm_key->oa.o_size;
2297
2298         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2299                                             actual_start_stripe, &stripe_count);
2300
2301         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2302                                                   &start_stripe);
2303
2304         if (fiemap->fm_extent_count == 0) {
2305                 get_num_extents = 1;
2306                 count_local = 0;
2307         }
2308
2309         /* Check each stripe */
2310         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2311              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2312                 obd_size req_fm_len; /* Stores length of required mapping */
2313                 obd_size len_mapped_single_call;
2314                 obd_off lun_start, lun_end, obd_object_end;
2315                 unsigned int ext_count;
2316
2317                 cur_stripe_wrap = cur_stripe;
2318
2319                 /* Find out range of mapping on this stripe */
2320                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2321                                            &lun_start, &obd_object_end)) == 0)
2322                         continue;
2323
2324                 /* If this is a continuation FIEMAP call and we are on
2325                  * starting stripe then lun_start needs to be set to
2326                  * fm_end_offset */
2327                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2328                         lun_start = fm_end_offset;
2329
2330                 if (fm_length != ~0ULL) {
2331                         /* Handle fm_start + fm_length overflow */
2332                         if (fm_start + fm_length < fm_start)
2333                                 fm_length = ~0ULL - fm_start;
2334                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2335                                                      cur_stripe);
2336                 } else {
2337                         lun_end = ~0ULL;
2338                 }
2339
2340                 if (lun_start == lun_end)
2341                         continue;
2342
2343                 req_fm_len = obd_object_end - lun_start;
2344                 fm_local->fm_length = 0;
2345                 len_mapped_single_call = 0;
2346
2347                 /* If the output buffer is very large and the objects have many
2348                  * extents we may need to loop on a single OST repeatedly */
2349                 ost_eof = 0;
2350                 ost_done = 0;
2351                 do {
2352                         if (get_num_extents == 0) {
2353                                 /* Don't get too many extents. */
2354                                 if (current_extent + count_local >
2355                                     fiemap->fm_extent_count)
2356                                         count_local = fiemap->fm_extent_count -
2357                                                                  current_extent;
2358                         }
2359
2360                         lun_start += len_mapped_single_call;
2361                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2362                         req_fm_len = fm_local->fm_length;
2363                         fm_local->fm_extent_count = count_local;
2364                         fm_local->fm_mapped_extents = 0;
2365                         fm_local->fm_flags = fiemap->fm_flags;
2366
2367                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2368                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2369
2370                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2371                                 GOTO(out, rc = -EINVAL);
2372
2373                         /* If OST is inactive, return extent with UNKNOWN flag */
2374                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2375                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2376                                 fm_local->fm_mapped_extents = 1;
2377
2378                                 lcl_fm_ext[0].fe_logical = lun_start;
2379                                 lcl_fm_ext[0].fe_length = obd_object_end -
2380                                                                       lun_start;
2381                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2382
2383                                 goto inactive_tgt;
2384                         }
2385
2386                         fm_local->fm_start = lun_start;
2387                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2388                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2389                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2390                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2391                                           keylen, key, vallen, fm_local, lsm);
2392                         if (rc != 0)
2393                                 GOTO(out, rc);
2394
2395 inactive_tgt:
2396                         ext_count = fm_local->fm_mapped_extents;
2397                         if (ext_count == 0) {
2398                                 ost_done = 1;
2399                                 /* If last stripe has hole at the end,
2400                                  * then we need to return */
2401                                 if (cur_stripe_wrap == last_stripe) {
2402                                         fiemap->fm_mapped_extents = 0;
2403                                         goto finish;
2404                                 }
2405                                 break;
2406                         }
2407
2408                         /* If we just need num of extents then go to next device */
2409                         if (get_num_extents) {
2410                                 current_extent += ext_count;
2411                                 break;
2412                         }
2413
2414                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2415                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2416
2417                         /* Have we finished mapping on this device? */
2418                         if (req_fm_len <= len_mapped_single_call)
2419                                 ost_done = 1;
2420
2421                         /* Clear the EXTENT_LAST flag which can be present on
2422                          * last extent */
2423                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2424                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2425                                                             ~FIEMAP_EXTENT_LAST;
2426
2427                         curr_loc = lov_stripe_size(lsm,
2428                                            lcl_fm_ext[ext_count - 1].fe_logical+
2429                                            lcl_fm_ext[ext_count - 1].fe_length,
2430                                            cur_stripe);
2431                         if (curr_loc >= fm_key->oa.o_size)
2432                                 ost_eof = 1;
2433
2434                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2435                                                      ost_index, ext_count,
2436                                                      current_extent);
2437
2438                         current_extent += ext_count;
2439
2440                         /* Ran out of available extents? */
2441                         if (current_extent >= fiemap->fm_extent_count)
2442                                 goto finish;
2443                 } while (ost_done == 0 && ost_eof == 0);
2444
2445                 if (cur_stripe_wrap == last_stripe)
2446                         goto finish;
2447         }
2448
2449 finish:
2450         /* Indicate that we are returning device offsets unless file just has
2451          * single stripe */
2452         if (lsm->lsm_stripe_count > 1)
2453                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2454
2455         if (get_num_extents)
2456                 goto skip_last_device_calc;
2457
2458         /* Check if we have reached the last stripe and whether mapping for that
2459          * stripe is done. */
2460         if (cur_stripe_wrap == last_stripe) {
2461                 if (ost_done || ost_eof)
2462                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2463                                                              FIEMAP_EXTENT_LAST;
2464         }
2465
2466 skip_last_device_calc:
2467         fiemap->fm_mapped_extents = current_extent;
2468
2469 out:
2470         OBD_FREE(fm_local, buffer_size);
2471         return rc;
2472 }
2473
2474 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2475                         void *key, __u32 *vallen, void *val,
2476                         struct lov_stripe_md *lsm)
2477 {
2478         struct obd_device *obddev = class_exp2obd(exp);
2479         struct lov_obd *lov = &obddev->u.lov;
2480         int i, rc;
2481         ENTRY;
2482
2483         if (!vallen || !val)
2484                 RETURN(-EFAULT);
2485
2486         obd_getref(obddev);
2487
2488         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2489                 struct {
2490                         char name[16];
2491                         struct ldlm_lock *lock;
2492                 } *data = key;
2493                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2494                 struct lov_oinfo *loi;
2495                 __u32 *stripe = val;
2496
2497                 if (*vallen < sizeof(*stripe))
2498                         GOTO(out, rc = -EFAULT);
2499                 *vallen = sizeof(*stripe);
2500
2501                 /* XXX This is another one of those bits that will need to
2502                  * change if we ever actually support nested LOVs.  It uses
2503                  * the lock's export to find out which stripe it is. */
2504                 /* XXX - it's assumed all the locks for deleted OSTs have
2505                  * been cancelled. Also, the export for deleted OSTs will
2506                  * be NULL and won't match the lock's export. */
2507                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2508                         loi = lsm->lsm_oinfo[i];
2509                         if (!lov->lov_tgts[loi->loi_ost_idx])
2510                                 continue;
2511                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2512                             data->lock->l_conn_export &&
2513                             osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2514                                 *stripe = i;
2515                                 GOTO(out, rc = 0);
2516                         }
2517                 }
2518                 LDLM_ERROR(data->lock, "lock on inode without such object");
2519                 dump_lsm(D_ERROR, lsm);
2520                 GOTO(out, rc = -ENXIO);
2521         } else if (KEY_IS(KEY_LAST_ID)) {
2522                 struct obd_id_info *info = val;
2523                 __u32 size = sizeof(obd_id);
2524                 struct lov_tgt_desc *tgt;
2525
2526                 LASSERT(*vallen == sizeof(struct obd_id_info));
2527                 tgt = lov->lov_tgts[info->idx];
2528
2529                 if (!tgt || !tgt->ltd_active)
2530                         GOTO(out, rc = -ESRCH);
2531
2532                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2533                 GOTO(out, rc = 0);
2534         } else if (KEY_IS(KEY_LOVDESC)) {
2535                 struct lov_desc *desc_ret = val;
2536                 *desc_ret = lov->desc;
2537
2538                 GOTO(out, rc = 0);
2539         } else if (KEY_IS(KEY_FIEMAP)) {
2540                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2541                 GOTO(out, rc);
2542         }
2543
2544         rc = -EINVAL;
2545
2546 out:
2547         obd_putref(obddev);
2548         RETURN(rc);
2549 }
2550
2551 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2552                               void *key, obd_count vallen, void *val,
2553                               struct ptlrpc_request_set *set)
2554 {
2555         struct obd_device *obddev = class_exp2obd(exp);
2556         struct lov_obd *lov = &obddev->u.lov;
2557         obd_count count;
2558         int i, rc = 0, err;
2559         struct lov_tgt_desc *tgt;
2560         unsigned incr, check_uuid,
2561                  do_inactive, no_set;
2562         unsigned next_id = 0,  mds_con = 0, capa = 0;
2563         ENTRY;
2564
2565         incr = check_uuid = do_inactive = no_set = 0;
2566         if (set == NULL) {
2567                 no_set = 1;
2568                 set = ptlrpc_prep_set();
2569                 if (!set)
2570                         RETURN(-ENOMEM);
2571         }
2572
2573         obd_getref(obddev);
2574         count = lov->desc.ld_tgt_count;
2575
2576         if (KEY_IS(KEY_NEXT_ID)) {
2577                 count = vallen / sizeof(struct obd_id_info);
2578                 vallen = sizeof(obd_id);
2579                 incr = sizeof(struct obd_id_info);
2580                 do_inactive = 1;
2581                 next_id = 1;
2582         } else if (KEY_IS(KEY_CHECKSUM)) {
2583                 do_inactive = 1;
2584         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2585                 /* use defaults:  do_inactive = incr = 0; */
2586         } else if (KEY_IS(KEY_MDS_CONN)) {
2587                 mds_con = 1;
2588         } else if (KEY_IS(KEY_CAPA_KEY)) {
2589                 capa = 1;
2590         }
2591
2592         for (i = 0; i < count; i++, val = (char *)val + incr) {
2593                 if (next_id) {
2594                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2595                 } else {
2596                         tgt = lov->lov_tgts[i];
2597                 }
2598                 /* OST was disconnected */
2599                 if (!tgt || !tgt->ltd_exp)
2600                         continue;
2601
2602                 /* OST is inactive and we don't want inactive OSCs */
2603                 if (!tgt->ltd_active && !do_inactive)
2604                         continue;
2605
2606                 if (mds_con) {
2607                         struct mds_group_info *mgi;
2608
2609                         LASSERT(vallen == sizeof(*mgi));
2610                         mgi = (struct mds_group_info *)val;
2611
2612                         /* Only want a specific OSC */
2613                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2614                                                 &tgt->ltd_uuid))
2615                                 continue;
2616
2617                         err = obd_set_info_async(tgt->ltd_exp,
2618                                          keylen, key, sizeof(int),
2619                                          &mgi->group, set);
2620                 } else if (next_id) {
2621                         err = obd_set_info_async(tgt->ltd_exp,
2622                                          keylen, key, vallen,
2623                                          ((struct obd_id_info*)val)->data, set);
2624                 } else if (capa) {
2625                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2626
2627                         LASSERT(vallen == sizeof(*info));
2628
2629                          /* Only want a specific OSC */
2630                         if (info->uuid &&
2631                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2632                                 continue;
2633
2634                         err = obd_set_info_async(tgt->ltd_exp, keylen, key,
2635                                                  sizeof(*info->capa),
2636                                                  info->capa, set);
2637                        
2638                 } else {
2639                         /* Only want a specific OSC */
2640                         if (check_uuid &&
2641                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2642                                 continue;
2643
2644                         err = obd_set_info_async(tgt->ltd_exp,
2645                                          keylen, key, vallen, val, set);
2646                 }
2647
2648                 if (!rc)
2649                         rc = err;
2650         }
2651
2652         obd_putref(obddev);
2653         if (no_set) {
2654                 err = ptlrpc_set_wait(set);
2655                 if (!rc)
2656                         rc = err;
2657                 ptlrpc_set_destroy(set);
2658         }
2659         RETURN(rc);
2660 }
2661
2662 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2663 {
2664         int i, rc = 0;
2665         ENTRY;
2666
2667         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2668                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2669                 if (loi->loi_ar.ar_rc && !rc)
2670                         rc = loi->loi_ar.ar_rc;
2671                 loi->loi_ar.ar_rc = 0;
2672         }
2673         RETURN(rc);
2674 }
2675 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2676
2677
2678 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2679                            int cmd, __u64 *offset)
2680 {
2681         __u32 ssize = lsm->lsm_stripe_size;
2682         __u64 start;
2683
2684         start = *offset;
2685         do_div(start, ssize);
2686         start = start * ssize;
2687
2688         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2689                            ", end "LPU64"\n", *offset, ssize, start,
2690                            start + ssize - 1);
2691         if (cmd == OBD_CALC_STRIPE_END) {
2692                 *offset = start + ssize - 1;
2693         } else if (cmd == OBD_CALC_STRIPE_START) {
2694                 *offset = start;
2695         } else {
2696                 LBUG();
2697         }
2698
2699         RETURN(0);
2700 }
2701
2702 void lov_stripe_lock(struct lov_stripe_md *md)
2703 {
2704         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2705         cfs_spin_lock(&md->lsm_lock);
2706         LASSERT(md->lsm_lock_owner == 0);
2707         md->lsm_lock_owner = cfs_curproc_pid();
2708 }
2709 EXPORT_SYMBOL(lov_stripe_lock);
2710
2711 void lov_stripe_unlock(struct lov_stripe_md *md)
2712 {
2713         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2714         md->lsm_lock_owner = 0;
2715         cfs_spin_unlock(&md->lsm_lock);
2716 }
2717 EXPORT_SYMBOL(lov_stripe_unlock);
2718
2719
2720 struct obd_ops lov_obd_ops = {
2721         .o_owner               = THIS_MODULE,
2722         .o_setup               = lov_setup,
2723         .o_precleanup          = lov_precleanup,
2724         .o_cleanup             = lov_cleanup,
2725         //.o_process_config      = lov_process_config,
2726         .o_connect             = lov_connect,
2727         .o_disconnect          = lov_disconnect,
2728         .o_statfs              = lov_statfs,
2729         .o_statfs_async        = lov_statfs_async,
2730         .o_packmd              = lov_packmd,
2731         .o_unpackmd            = lov_unpackmd,
2732         .o_create              = lov_create,
2733         .o_destroy             = lov_destroy,
2734         .o_getattr             = lov_getattr,
2735         .o_getattr_async       = lov_getattr_async,
2736         .o_setattr             = lov_setattr,
2737         .o_setattr_async       = lov_setattr_async,
2738         .o_brw                 = lov_brw,
2739         .o_merge_lvb           = lov_merge_lvb,
2740         .o_adjust_kms          = lov_adjust_kms,
2741         .o_punch               = lov_punch,
2742         .o_sync                = lov_sync,
2743         .o_enqueue             = lov_enqueue,
2744         .o_change_cbdata       = lov_change_cbdata,
2745         .o_cancel              = lov_cancel,
2746         .o_cancel_unused       = lov_cancel_unused,
2747         .o_iocontrol           = lov_iocontrol,
2748         .o_get_info            = lov_get_info,
2749         .o_set_info_async      = lov_set_info_async,
2750         .o_extent_calc         = lov_extent_calc,
2751         .o_llog_init           = lov_llog_init,
2752         .o_llog_finish         = lov_llog_finish,
2753         .o_notify              = lov_notify,
2754         .o_pool_new            = lov_pool_new,
2755         .o_pool_rem            = lov_pool_remove,
2756         .o_pool_add            = lov_pool_add,
2757         .o_pool_del            = lov_pool_del,
2758         .o_getref              = lov_getref,
2759         .o_putref              = lov_putref,
2760 };
2761
2762 static quota_interface_t *quota_interface;
2763 extern quota_interface_t lov_quota_interface;
2764
2765 cfs_mem_cache_t *lov_oinfo_slab;
2766
2767 extern struct lu_kmem_descr lov_caches[];
2768
2769 int __init lov_init(void)
2770 {
2771         struct lprocfs_static_vars lvars = { 0 };
2772         int rc, rc2;
2773         ENTRY;
2774
2775         /* print an address of _any_ initialized kernel symbol from this
2776          * module, to allow debugging with gdb that doesn't support data
2777          * symbols from modules.*/
2778         CDEBUG(D_CONSOLE, "Lustre LOV module (%p).\n", &lov_caches);
2779
2780         rc = lu_kmem_init(lov_caches);
2781         if (rc)
2782                 return rc;
2783
2784         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2785                                               sizeof(struct lov_oinfo),
2786                                               0, CFS_SLAB_HWCACHE_ALIGN);
2787         if (lov_oinfo_slab == NULL) {
2788                 lu_kmem_fini(lov_caches);
2789                 return -ENOMEM;
2790         }
2791         lprocfs_lov_init_vars(&lvars);
2792
2793         cfs_request_module("lquota");
2794         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2795         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2796
2797         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2798                                  LUSTRE_LOV_NAME, &lov_device_type);
2799
2800         if (rc) {
2801                 if (quota_interface)
2802                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2803                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2804                 LASSERT(rc2 == 0);
2805                 lu_kmem_fini(lov_caches);
2806         }
2807
2808         RETURN(rc);
2809 }
2810
2811 #ifdef __KERNEL__
2812 static void /*__exit*/ lov_exit(void)
2813 {
2814         int rc;
2815
2816         lu_device_type_fini(&lov_device_type);
2817         lu_kmem_fini(lov_caches);
2818
2819         if (quota_interface)
2820                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2821
2822         class_unregister_type(LUSTRE_LOV_NAME);
2823         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2824         LASSERT(rc == 0);
2825 }
2826
2827 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2828 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2829 MODULE_LICENSE("GPL");
2830
2831 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2832 #endif