Whamcloud - gitweb
81c01f2c3b46725b23d14c479c256fbc568866a9
[fs/lustre-release.git] / lustre / lov / lov_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/lov/lov_obd.c
37  *
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Peter Braam <braam@clusterfs.com>
40  * Author: Mike Shaver <shaver@clusterfs.com>
41  * Author: Nathan Rutman <nathan@clusterfs.com>
42  */
43
44 #define DEBUG_SUBSYSTEM S_LOV
45 #ifdef __KERNEL__
46 #include <libcfs/libcfs.h>
47 #else
48 #include <liblustre.h>
49 #endif
50
51 #include <obd_support.h>
52 #include <lustre_lib.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_dlm.h>
56 #include <lustre_mds.h>
57 #include <lustre_debug.h>
58 #include <obd_class.h>
59 #include <obd_lov.h>
60 #include <obd_ost.h>
61 #include <lprocfs_status.h>
62 #include <lustre_param.h>
63 #include <cl_object.h>
64 #include <lclient.h> /* for cl_client_lru */
65 #include <lustre/ll_fiemap.h>
66 #include <lustre_log.h>
67 #include <lustre_fid.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         mutex_lock(&lov->lov_lock);
79         cfs_atomic_inc(&lov->lov_refcount);
80         mutex_unlock(&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         mutex_lock(&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                 mutex_unlock(&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                 mutex_unlock(&lov->lov_lock);
120         }
121 }
122
123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
124                               enum obd_notify_event ev);
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 #ifdef __KERNEL__
139         cfs_proc_dir_entry_t *lov_proc_dir;
140 #endif
141         int rc;
142         ENTRY;
143
144         if (!lov->lov_tgts[index])
145                 RETURN(-EINVAL);
146
147         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
148         tgt_obd = lov->lov_tgts[index]->ltd_obd;
149
150         if (!tgt_obd->obd_set_up) {
151                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
152                 RETURN(-EINVAL);
153         }
154
155         /* override the sp_me from lov */
156         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
157
158         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
159                 data->ocd_index = index;
160
161         /*
162          * Divine LOV knows that OBDs under it are OSCs.
163          */
164         imp = tgt_obd->u.cli.cl_import;
165
166         if (activate) {
167                 tgt_obd->obd_no_recov = 0;
168                 /* FIXME this is probably supposed to be
169                    ptlrpc_set_import_active.  Horrible naming. */
170                 ptlrpc_activate_import(imp);
171         }
172
173         rc = obd_register_observer(tgt_obd, obd);
174         if (rc) {
175                 CERROR("Target %s register_observer error %d\n",
176                        obd_uuid2str(tgt_uuid), rc);
177                 RETURN(rc);
178         }
179
180
181         if (imp->imp_invalid) {
182                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
183                        "disabled\n", obd_uuid2str(tgt_uuid));
184                 RETURN(0);
185         }
186
187         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
188                          &lov_osc_uuid, data, NULL);
189         if (rc || !lov->lov_tgts[index]->ltd_exp) {
190                 CERROR("Target %s connect error %d\n",
191                        obd_uuid2str(tgt_uuid), rc);
192                 RETURN(-ENODEV);
193         }
194
195         lov->lov_tgts[index]->ltd_reap = 0;
196
197         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
198                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
199
200 #ifdef __KERNEL__
201         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
202         if (lov_proc_dir) {
203                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
204                 cfs_proc_dir_entry_t *osc_symlink;
205
206                 LASSERT(osc_obd != NULL);
207                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
208                 LASSERT(osc_obd->obd_type->typ_name != NULL);
209
210                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
211                                                   lov_proc_dir,
212                                                   "../../../%s/%s",
213                                                   osc_obd->obd_type->typ_name,
214                                                   osc_obd->obd_name);
215                 if (osc_symlink == NULL) {
216                         CERROR("could not register LOV target "
217                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
218                                 obd->obd_type->typ_name, obd->obd_name,
219                                 osc_obd->obd_name);
220                         lprocfs_remove(&lov_proc_dir);
221                 }
222         }
223 #endif
224
225         RETURN(0);
226 }
227
228 static int lov_connect(const struct lu_env *env,
229                        struct obd_export **exp, struct obd_device *obd,
230                        struct obd_uuid *cluuid, struct obd_connect_data *data,
231                        void *localdata)
232 {
233         struct lov_obd *lov = &obd->u.lov;
234         struct lov_tgt_desc *tgt;
235         struct lustre_handle conn;
236         int i, rc;
237         ENTRY;
238
239         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
240
241         rc = class_connect(&conn, obd, cluuid);
242         if (rc)
243                 RETURN(rc);
244
245         *exp = class_conn2export(&conn);
246
247         /* Why should there ever be more than 1 connect? */
248         lov->lov_connects++;
249         LASSERT(lov->lov_connects == 1);
250
251         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
252         if (data)
253                 lov->lov_ocd = *data;
254
255         obd_getref(obd);
256         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
257                 tgt = lov->lov_tgts[i];
258                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
259                         continue;
260                 /* Flags will be lowest common denominator */
261                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
262                 if (rc) {
263                         CERROR("%s: lov connect tgt %d failed: %d\n",
264                                obd->obd_name, i, rc);
265                         continue;
266                 }
267                 /* connect to administrative disabled ost */
268                 if (!lov->lov_tgts[i]->ltd_exp)
269                         continue;
270
271                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
272                                 OBD_NOTIFY_CONNECT, (void *)&i);
273                 if (rc) {
274                         CERROR("%s error sending notify %d\n",
275                                obd->obd_name, rc);
276                 }
277         }
278         obd_putref(obd);
279
280         RETURN(0);
281 }
282
283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
284 {
285         cfs_proc_dir_entry_t *lov_proc_dir;
286         struct lov_obd *lov = &obd->u.lov;
287         struct obd_device *osc_obd;
288         int rc;
289         ENTRY;
290
291         osc_obd = class_exp2obd(tgt->ltd_exp);
292         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
293                obd->obd_name, osc_obd->obd_name);
294
295         if (tgt->ltd_active) {
296                 tgt->ltd_active = 0;
297                 lov->desc.ld_active_tgt_count--;
298                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
299         }
300
301         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
302         if (lov_proc_dir) {
303                 cfs_proc_dir_entry_t *osc_symlink;
304
305                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
306                 if (osc_symlink) {
307                         lprocfs_remove(&osc_symlink);
308                 } else {
309                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
310                                obd->obd_type->typ_name, obd->obd_name,
311                                osc_obd->obd_name);
312                 }
313         }
314
315         if (osc_obd) {
316                 /* Pass it on to our clients.
317                  * XXX This should be an argument to disconnect,
318                  * XXX not a back-door flag on the OBD.  Ah well.
319                  */
320                 osc_obd->obd_force = obd->obd_force;
321                 osc_obd->obd_fail = obd->obd_fail;
322                 osc_obd->obd_no_recov = obd->obd_no_recov;
323         }
324
325         obd_register_observer(osc_obd, NULL);
326
327         rc = obd_disconnect(tgt->ltd_exp);
328         if (rc) {
329                 CERROR("Target %s disconnect error %d\n",
330                        tgt->ltd_uuid.uuid, rc);
331                 rc = 0;
332         }
333
334         tgt->ltd_exp = NULL;
335         RETURN(0);
336 }
337
338 static int lov_disconnect(struct obd_export *exp)
339 {
340         struct obd_device *obd = class_exp2obd(exp);
341         struct lov_obd *lov = &obd->u.lov;
342         int i, rc;
343         ENTRY;
344
345         if (!lov->lov_tgts)
346                 goto out;
347
348         /* Only disconnect the underlying layers on the final disconnect. */
349         lov->lov_connects--;
350         if (lov->lov_connects != 0) {
351                 /* why should there be more than 1 connect? */
352                 CERROR("disconnect #%d\n", lov->lov_connects);
353                 goto out;
354         }
355
356         /* Let's hold another reference so lov_del_obd doesn't spin through
357            putref every time */
358         obd_getref(obd);
359
360         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
361                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
362                         /* Disconnection is the last we know about an obd */
363                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
364                 }
365         }
366         obd_putref(obd);
367
368 out:
369         rc = class_disconnect(exp); /* bz 9811 */
370         RETURN(rc);
371 }
372
373 /* Error codes:
374  *
375  *  -EINVAL  : UUID can't be found in the LOV's target list
376  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
377  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
378  *  any >= 0 : is log target index
379  */
380 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
381                               enum obd_notify_event ev)
382 {
383         struct lov_obd *lov = &obd->u.lov;
384         struct lov_tgt_desc *tgt;
385         int index, activate, active;
386         ENTRY;
387
388         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
389                lov, uuid->uuid, ev);
390
391         obd_getref(obd);
392         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
393                 tgt = lov->lov_tgts[index];
394                 if (!tgt)
395                         continue;
396                 /*
397                  * LU-642, initially inactive OSC could miss the obd_connect,
398                  * we make up for it here.
399                  */
400                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
401                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
402                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
403
404                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
405                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
406                 }
407                 if (!tgt->ltd_exp)
408                         continue;
409
410                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
411                        index, obd_uuid2str(&tgt->ltd_uuid),
412                        tgt->ltd_exp->exp_handle.h_cookie);
413                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
414                         break;
415         }
416
417         if (index == lov->desc.ld_tgt_count)
418                 GOTO(out, index = -EINVAL);
419
420         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
421                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
422
423                 if (lov->lov_tgts[index]->ltd_activate == activate) {
424                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
425                                uuid->uuid, activate ? "" : "de");
426                 } else {
427                         lov->lov_tgts[index]->ltd_activate = activate;
428                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
429                                activate ? "" : "de", obd_uuid2str(uuid));
430                 }
431
432         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
433                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
434
435                 if (lov->lov_tgts[index]->ltd_active == active) {
436                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
437                                uuid->uuid, active ? "" : "in");
438                         GOTO(out, index);
439                 } else {
440                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
441                                obd_uuid2str(uuid), active ? "" : "in");
442                 }
443
444                 lov->lov_tgts[index]->ltd_active = active;
445                 if (active) {
446                         lov->desc.ld_active_tgt_count++;
447                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
448                 } else {
449                         lov->desc.ld_active_tgt_count--;
450                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
451                 }
452         } else {
453                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
454         }
455
456  out:
457         obd_putref(obd);
458         RETURN(index);
459 }
460
461 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
462                       enum obd_notify_event ev, void *data)
463 {
464         int rc = 0;
465         struct lov_obd *lov = &obd->u.lov;
466         ENTRY;
467
468         down_read(&lov->lov_notify_lock);
469         if (!lov->lov_connects) {
470                 up_read(&lov->lov_notify_lock);
471                 RETURN(rc);
472         }
473
474         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
475             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
476                 struct obd_uuid *uuid;
477
478                 LASSERT(watched);
479
480                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
481                         up_read(&lov->lov_notify_lock);
482                         CERROR("unexpected notification of %s %s!\n",
483                                watched->obd_type->typ_name,
484                                watched->obd_name);
485                         RETURN(-EINVAL);
486                 }
487                 uuid = &watched->u.cli.cl_target_uuid;
488
489                 /* Set OSC as active before notifying the observer, so the
490                  * observer can use the OSC normally.
491                  */
492                 rc = lov_set_osc_active(obd, uuid, ev);
493                 if (rc < 0) {
494                         up_read(&lov->lov_notify_lock);
495                         CERROR("event(%d) of %s failed: %d\n", ev,
496                                obd_uuid2str(uuid), rc);
497                         RETURN(rc);
498                 }
499                 /* active event should be pass lov target index as data */
500                 data = &rc;
501         }
502
503         /* Pass the notification up the chain. */
504         if (watched) {
505                 rc = obd_notify_observer(obd, watched, ev, data);
506         } else {
507                 /* NULL watched means all osc's in the lov (only for syncs) */
508                 /* sync event should be send lov idx as data */
509                 struct lov_obd *lov = &obd->u.lov;
510                 int i, is_sync;
511
512                 data = &i;
513                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
514                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
515
516                 obd_getref(obd);
517                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
518                         if (!lov->lov_tgts[i])
519                                 continue;
520
521                         /* don't send sync event if target not
522                          * connected/activated */
523                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
524                                 continue;
525
526                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
527                                                  ev, data);
528                         if (rc) {
529                                 CERROR("%s: notify %s of %s failed %d\n",
530                                        obd->obd_name,
531                                        obd->obd_observer->obd_name,
532                                        lov->lov_tgts[i]->ltd_obd->obd_name,
533                                        rc);
534                         }
535                 }
536                 obd_putref(obd);
537         }
538
539         up_read(&lov->lov_notify_lock);
540         RETURN(rc);
541 }
542
543 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
544                           __u32 index, int gen, int active)
545 {
546         struct lov_obd *lov = &obd->u.lov;
547         struct lov_tgt_desc *tgt;
548         struct obd_device *tgt_obd;
549         int rc;
550         ENTRY;
551
552         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
553                uuidp->uuid, index, gen, active);
554
555         if (gen <= 0) {
556                 CERROR("request to add OBD %s with invalid generation: %d\n",
557                        uuidp->uuid, gen);
558                 RETURN(-EINVAL);
559         }
560
561         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
562                                         &obd->obd_uuid);
563         if (tgt_obd == NULL)
564                 RETURN(-EINVAL);
565
566         mutex_lock(&lov->lov_lock);
567
568         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
569                 tgt = lov->lov_tgts[index];
570                 CERROR("UUID %s already assigned at LOV target index %d\n",
571                        obd_uuid2str(&tgt->ltd_uuid), index);
572                 mutex_unlock(&lov->lov_lock);
573                 RETURN(-EEXIST);
574         }
575
576         if (index >= lov->lov_tgt_size) {
577                 /* We need to reallocate the lov target array. */
578                 struct lov_tgt_desc **newtgts, **old = NULL;
579                 __u32 newsize, oldsize = 0;
580
581                 newsize = max(lov->lov_tgt_size, (__u32)2);
582                 while (newsize < index + 1)
583                         newsize = newsize << 1;
584                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
585                 if (newtgts == NULL) {
586                         mutex_unlock(&lov->lov_lock);
587                         RETURN(-ENOMEM);
588                 }
589
590                 if (lov->lov_tgt_size) {
591                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
592                                lov->lov_tgt_size);
593                         old = lov->lov_tgts;
594                         oldsize = lov->lov_tgt_size;
595                 }
596
597                 lov->lov_tgts = newtgts;
598                 lov->lov_tgt_size = newsize;
599                 smp_rmb();
600                 if (old)
601                         OBD_FREE(old, sizeof(*old) * oldsize);
602
603                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
604                        lov->lov_tgts, lov->lov_tgt_size);
605         }
606
607         OBD_ALLOC_PTR(tgt);
608         if (!tgt) {
609                 mutex_unlock(&lov->lov_lock);
610                 RETURN(-ENOMEM);
611         }
612
613         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
614         if (rc) {
615                 mutex_unlock(&lov->lov_lock);
616                 OBD_FREE_PTR(tgt);
617                 RETURN(rc);
618         }
619
620         tgt->ltd_uuid = *uuidp;
621         tgt->ltd_obd = tgt_obd;
622         /* XXX - add a sanity check on the generation number. */
623         tgt->ltd_gen = gen;
624         tgt->ltd_index = index;
625         tgt->ltd_activate = active;
626         lov->lov_tgts[index] = tgt;
627         if (index >= lov->desc.ld_tgt_count)
628                 lov->desc.ld_tgt_count = index + 1;
629
630         mutex_unlock(&lov->lov_lock);
631
632         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
633                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
634
635         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
636
637         if (lov->lov_connects == 0) {
638                 /* lov_connect hasn't been called yet. We'll do the
639                    lov_connect_obd on this target when that fn first runs,
640                    because we don't know the connect flags yet. */
641                 RETURN(0);
642         }
643
644         obd_getref(obd);
645
646         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
647         if (rc)
648                 GOTO(out, rc);
649
650         /* connect to administrative disabled ost */
651         if (!tgt->ltd_exp)
652                 GOTO(out, rc = 0);
653
654         if (lov->lov_cache != NULL) {
655                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
656                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
657                                 sizeof(struct cl_client_cache), lov->lov_cache,
658                                 NULL);
659                 if (rc < 0)
660                         GOTO(out, rc);
661         }
662
663         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
664                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
665                         (void *)&index);
666
667 out:
668         if (rc) {
669                 CERROR("add failed (%d), deleting %s\n", rc,
670                        obd_uuid2str(&tgt->ltd_uuid));
671                 lov_del_target(obd, index, 0, 0);
672         }
673         obd_putref(obd);
674         RETURN(rc);
675 }
676
677 /* Schedule a target for deletion */
678 int lov_del_target(struct obd_device *obd, __u32 index,
679                    struct obd_uuid *uuidp, int gen)
680 {
681         struct lov_obd *lov = &obd->u.lov;
682         int count = lov->desc.ld_tgt_count;
683         int rc = 0;
684         ENTRY;
685
686         if (index >= count) {
687                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
688                        index, count);
689                 RETURN(-EINVAL);
690         }
691
692         /* to make sure there's no ongoing lov_notify() now */
693         down_write(&lov->lov_notify_lock);
694         obd_getref(obd);
695
696         if (!lov->lov_tgts[index]) {
697                 CERROR("LOV target at index %d is not setup.\n", index);
698                 GOTO(out, rc = -EINVAL);
699         }
700
701         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
702                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
703                        lov_uuid2str(lov, index), index,
704                        obd_uuid2str(uuidp));
705                 GOTO(out, rc = -EINVAL);
706         }
707
708         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
709                lov_uuid2str(lov, index), index,
710                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
711                lov->lov_tgts[index]->ltd_active);
712
713         lov->lov_tgts[index]->ltd_reap = 1;
714         lov->lov_death_row++;
715         /* we really delete it from obd_putref */
716 out:
717         obd_putref(obd);
718         up_write(&lov->lov_notify_lock);
719
720         RETURN(rc);
721 }
722
723 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
724 {
725         struct obd_device *osc_obd;
726
727         LASSERT(tgt);
728         LASSERT(tgt->ltd_reap);
729
730         osc_obd = class_exp2obd(tgt->ltd_exp);
731
732         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
733                tgt->ltd_uuid.uuid,
734                osc_obd ? osc_obd->obd_name : "<no obd>");
735
736         if (tgt->ltd_exp)
737                 lov_disconnect_obd(obd, tgt);
738
739         OBD_FREE_PTR(tgt);
740
741         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
742            do it ourselves. And we can't do it from lov_cleanup,
743            because we just lost our only reference to it. */
744         if (osc_obd)
745                 class_manual_cleanup(osc_obd);
746 }
747
748 void lov_fix_desc_stripe_size(__u64 *val)
749 {
750         if (*val < LOV_DEFAULT_STRIPE_SIZE) {
751                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
752                               LOV_DEFAULT_STRIPE_SIZE);
753                 *val = LOV_DEFAULT_STRIPE_SIZE;
754         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
755                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
756                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
757                               "multiple of %u)\n",
758                               *val, LOV_MIN_STRIPE_SIZE);
759         }
760 }
761
762 void lov_fix_desc_stripe_count(__u32 *val)
763 {
764         if (*val == 0)
765                 *val = 1;
766 }
767
768 void lov_fix_desc_pattern(__u32 *val)
769 {
770         /* from lov_setstripe */
771         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
772                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
773                 *val = 0;
774         }
775 }
776
777 void lov_fix_desc_qos_maxage(__u32 *val)
778 {
779         /* fix qos_maxage */
780         if (*val == 0)
781                 *val = QOS_DEFAULT_MAXAGE;
782 }
783
784 void lov_fix_desc(struct lov_desc *desc)
785 {
786         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
787         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
788         lov_fix_desc_pattern(&desc->ld_pattern);
789         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
790 }
791
792 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
793 {
794         struct lprocfs_static_vars lvars = { 0 };
795         struct lov_desc *desc;
796         struct lov_obd *lov = &obd->u.lov;
797         int rc;
798         ENTRY;
799
800         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
801                 CERROR("LOV setup requires a descriptor\n");
802                 RETURN(-EINVAL);
803         }
804
805         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
806
807         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
808                 CERROR("descriptor size wrong: %d > %d\n",
809                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
810                 RETURN(-EINVAL);
811         }
812
813         if (desc->ld_magic != LOV_DESC_MAGIC) {
814                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
815                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
816                                    obd->obd_name, desc);
817                             lustre_swab_lov_desc(desc);
818                 } else {
819                         CERROR("%s: Bad lov desc magic: %#x\n",
820                                obd->obd_name, desc->ld_magic);
821                         RETURN(-EINVAL);
822                 }
823         }
824
825         lov_fix_desc(desc);
826
827         desc->ld_active_tgt_count = 0;
828         lov->desc = *desc;
829         lov->lov_tgt_size = 0;
830
831         mutex_init(&lov->lov_lock);
832         cfs_atomic_set(&lov->lov_refcount, 0);
833         lov->lov_sp_me = LUSTRE_SP_CLI;
834
835         init_rwsem(&lov->lov_notify_lock);
836
837         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
838                                                    HASH_POOLS_MAX_BITS,
839                                                    HASH_POOLS_BKT_BITS, 0,
840                                                    CFS_HASH_MIN_THETA,
841                                                    CFS_HASH_MAX_THETA,
842                                                    &pool_hash_operations,
843                                                    CFS_HASH_DEFAULT);
844         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
845         lov->lov_pool_count = 0;
846         rc = lov_ost_pool_init(&lov->lov_packed, 0);
847         if (rc)
848                 GOTO(out, rc);
849
850         lprocfs_lov_init_vars(&lvars);
851         lprocfs_obd_setup(obd, lvars.obd_vars);
852 #ifdef LPROCFS
853         {
854                 int rc;
855
856                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
857                                         0444, &lov_proc_target_fops, obd);
858                 if (rc)
859                         CWARN("Error adding the target_obd file\n");
860         }
861 #endif
862         lov->lov_pool_proc_entry = lprocfs_register("pools",
863                                                     obd->obd_proc_entry,
864                                                     NULL, NULL);
865
866         RETURN(0);
867
868 out:
869         return rc;
870 }
871
872 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
873 {
874         int rc = 0;
875         struct lov_obd *lov = &obd->u.lov;
876
877         ENTRY;
878
879         switch (stage) {
880         case OBD_CLEANUP_EARLY: {
881                 int i;
882                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
883                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
884                                 continue;
885                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
886                                        OBD_CLEANUP_EARLY);
887                 }
888                 break;
889         }
890         case OBD_CLEANUP_EXPORTS:
891                 rc = obd_llog_finish(obd, 0);
892                 if (rc != 0)
893                         CERROR("failed to cleanup llogging subsystems\n");
894                 break;
895         }
896         RETURN(rc);
897 }
898
899 static int lov_cleanup(struct obd_device *obd)
900 {
901         struct lov_obd *lov = &obd->u.lov;
902         cfs_list_t *pos, *tmp;
903         struct pool_desc *pool;
904         ENTRY;
905
906         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
907                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
908                 /* free pool structs */
909                 CDEBUG(D_INFO, "delete pool %p\n", pool);
910                 /* In the function below, .hs_keycmp resolves to
911                  * pool_hashkey_keycmp() */
912                 /* coverity[overrun-buffer-val] */
913                 lov_pool_del(obd, pool->pool_name);
914         }
915         cfs_hash_putref(lov->lov_pools_hash_body);
916         lov_ost_pool_free(&lov->lov_packed);
917
918         lprocfs_obd_cleanup(obd);
919         if (lov->lov_tgts) {
920                 int i;
921                 obd_getref(obd);
922                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
923                         if (!lov->lov_tgts[i])
924                                 continue;
925
926                         /* Inactive targets may never have connected */
927                         if (lov->lov_tgts[i]->ltd_active ||
928                             cfs_atomic_read(&lov->lov_refcount))
929                             /* We should never get here - these
930                                should have been removed in the
931                              disconnect. */
932                                 CERROR("lov tgt %d not cleaned!"
933                                        " deathrow=%d, lovrc=%d\n",
934                                        i, lov->lov_death_row,
935                                        cfs_atomic_read(&lov->lov_refcount));
936                         lov_del_target(obd, i, 0, 0);
937                 }
938                 obd_putref(obd);
939                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
940                          lov->lov_tgt_size);
941                 lov->lov_tgt_size = 0;
942         }
943         RETURN(0);
944 }
945
946 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
947                             __u32 *indexp, int *genp)
948 {
949         struct obd_uuid obd_uuid;
950         int cmd;
951         int rc = 0;
952         ENTRY;
953
954         switch(cmd = lcfg->lcfg_command) {
955         case LCFG_LOV_ADD_OBD:
956         case LCFG_LOV_ADD_INA:
957         case LCFG_LOV_DEL_OBD: {
958                 __u32 index;
959                 int gen;
960                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
961                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
962                         GOTO(out, rc = -EINVAL);
963
964                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
965
966                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
967                         GOTO(out, rc = -EINVAL);
968                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
969                         GOTO(out, rc = -EINVAL);
970                 index = *indexp;
971                 gen = *genp;
972                 if (cmd == LCFG_LOV_ADD_OBD)
973                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
974                 else if (cmd == LCFG_LOV_ADD_INA)
975                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
976                 else
977                         rc = lov_del_target(obd, index, &obd_uuid, gen);
978                 GOTO(out, rc);
979         }
980         case LCFG_PARAM: {
981                 struct lprocfs_static_vars lvars = { 0 };
982                 struct lov_desc *desc = &(obd->u.lov.desc);
983
984                 if (!desc)
985                         GOTO(out, rc = -EINVAL);
986
987                 lprocfs_lov_init_vars(&lvars);
988
989                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
990                                               lcfg, obd);
991                 if (rc > 0)
992                         rc = 0;
993                 GOTO(out, rc);
994         }
995         case LCFG_POOL_NEW:
996         case LCFG_POOL_ADD:
997         case LCFG_POOL_DEL:
998         case LCFG_POOL_REM:
999                 GOTO(out, rc);
1000
1001         default: {
1002                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1003                 GOTO(out, rc = -EINVAL);
1004
1005         }
1006         }
1007 out:
1008         RETURN(rc);
1009 }
1010
1011 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1012                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1013 {
1014         struct lov_stripe_md *obj_mdp, *lsm;
1015         struct lov_obd *lov = &exp->exp_obd->u.lov;
1016         unsigned ost_idx;
1017         int rc, i;
1018         ENTRY;
1019
1020         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1021                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1022
1023         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1024         if (obj_mdp == NULL)
1025                 RETURN(-ENOMEM);
1026
1027         ost_idx = src_oa->o_nlink;
1028         lsm = *ea;
1029         if (lsm == NULL)
1030                 GOTO(out, rc = -EINVAL);
1031         if (ost_idx >= lov->desc.ld_tgt_count ||
1032             !lov->lov_tgts[ost_idx])
1033                 GOTO(out, rc = -EINVAL);
1034
1035         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1036                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1037                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1038                                         ostid_id(&src_oa->o_oi))
1039                                 GOTO(out, rc = -EINVAL);
1040                         break;
1041                 }
1042         }
1043         if (i == lsm->lsm_stripe_count)
1044                 GOTO(out, rc = -EINVAL);
1045
1046         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1047                         src_oa, &obj_mdp, oti);
1048 out:
1049         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1050         RETURN(rc);
1051 }
1052
1053 /* the LOV expects oa->o_id to be set to the LOV object id */
1054 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1055                       struct obdo *src_oa, struct lov_stripe_md **ea,
1056                       struct obd_trans_info *oti)
1057 {
1058         struct lov_obd *lov;
1059         int rc = 0;
1060         ENTRY;
1061
1062         LASSERT(ea != NULL);
1063         if (exp == NULL)
1064                 RETURN(-EINVAL);
1065
1066         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1067             src_oa->o_flags == OBD_FL_DELORPHAN) {
1068                 /* should be used with LOV anymore */
1069                 LBUG();
1070         }
1071
1072         lov = &exp->exp_obd->u.lov;
1073         if (!lov->desc.ld_active_tgt_count)
1074                 RETURN(-EIO);
1075
1076         obd_getref(exp->exp_obd);
1077         /* Recreate a specific object id at the given OST index */
1078         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1079             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1080                  rc = lov_recreate(exp, src_oa, ea, oti);
1081         }
1082
1083         obd_putref(exp->exp_obd);
1084         RETURN(rc);
1085 }
1086
1087 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1088 do {                                                                            \
1089         LASSERT((lsmp) != NULL);                                                \
1090         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1091                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1092                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1093 } while (0)
1094
1095 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1096                        struct obdo *oa, struct lov_stripe_md *lsm,
1097                        struct obd_trans_info *oti, struct obd_export *md_exp,
1098                        void *capa)
1099 {
1100         struct lov_request_set *set;
1101         struct obd_info oinfo;
1102         struct lov_request *req;
1103         cfs_list_t *pos;
1104         struct lov_obd *lov;
1105         int rc = 0, err = 0;
1106         ENTRY;
1107
1108         ASSERT_LSM_MAGIC(lsm);
1109
1110         if (!exp || !exp->exp_obd)
1111                 RETURN(-ENODEV);
1112
1113         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1114                 LASSERT(oti);
1115                 LASSERT(oti->oti_logcookies);
1116         }
1117
1118         lov = &exp->exp_obd->u.lov;
1119         obd_getref(exp->exp_obd);
1120         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1121         if (rc)
1122                 GOTO(out, rc);
1123
1124         cfs_list_for_each (pos, &set->set_list) {
1125                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1126
1127                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1128                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1129
1130                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1131                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1132                 err = lov_update_common_set(set, req, err);
1133                 if (err) {
1134                         CERROR("%s: destroying objid "DOSTID" subobj "
1135                                DOSTID" on OST idx %d: rc = %d\n",
1136                                exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1137                                POSTID(&req->rq_oi.oi_oa->o_oi),
1138                                req->rq_idx, err);
1139                         if (!rc)
1140                                 rc = err;
1141                 }
1142         }
1143
1144         if (rc == 0) {
1145                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1146                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1147         }
1148         err = lov_fini_destroy_set(set);
1149 out:
1150         obd_putref(exp->exp_obd);
1151         RETURN(rc ? rc : err);
1152 }
1153
1154 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1155                        struct obd_info *oinfo)
1156 {
1157         struct lov_request_set *set;
1158         struct lov_request *req;
1159         cfs_list_t *pos;
1160         struct lov_obd *lov;
1161         int err = 0, rc = 0;
1162         ENTRY;
1163
1164         LASSERT(oinfo);
1165         ASSERT_LSM_MAGIC(oinfo->oi_md);
1166
1167         if (!exp || !exp->exp_obd)
1168                 RETURN(-ENODEV);
1169
1170         lov = &exp->exp_obd->u.lov;
1171
1172         rc = lov_prep_getattr_set(exp, oinfo, &set);
1173         if (rc)
1174                 RETURN(rc);
1175
1176         cfs_list_for_each (pos, &set->set_list) {
1177                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1178
1179                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1180                        " %u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1181                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1182
1183                 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1184                                  &req->rq_oi);
1185                 err = lov_update_common_set(set, req, rc);
1186                 if (err) {
1187                         CERROR("%s: getattr objid "DOSTID" subobj "
1188                                DOSTID" on OST idx %d: rc = %d\n",
1189                                exp->exp_obd->obd_name,
1190                                POSTID(&oinfo->oi_oa->o_oi),
1191                                POSTID(&req->rq_oi.oi_oa->o_oi),
1192                                req->rq_idx, err);
1193                         break;
1194                 }
1195         }
1196
1197         rc = lov_fini_getattr_set(set);
1198         if (err)
1199                 rc = err;
1200         RETURN(rc);
1201 }
1202
1203 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1204                                  void *data, int rc)
1205 {
1206         struct lov_request_set *lovset = (struct lov_request_set *)data;
1207         int err;
1208         ENTRY;
1209
1210         /* don't do attribute merge if this aysnc op failed */
1211         if (rc)
1212                 cfs_atomic_set(&lovset->set_completes, 0);
1213         err = lov_fini_getattr_set(lovset);
1214         RETURN(rc ? rc : err);
1215 }
1216
1217 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1218                               struct ptlrpc_request_set *rqset)
1219 {
1220         struct lov_request_set *lovset;
1221         struct lov_obd *lov;
1222         cfs_list_t *pos;
1223         struct lov_request *req;
1224         int rc = 0, err;
1225         ENTRY;
1226
1227         LASSERT(oinfo);
1228         ASSERT_LSM_MAGIC(oinfo->oi_md);
1229
1230         if (!exp || !exp->exp_obd)
1231                 RETURN(-ENODEV);
1232
1233         lov = &exp->exp_obd->u.lov;
1234
1235         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1236         if (rc)
1237                 RETURN(rc);
1238
1239         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1240                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1241                oinfo->oi_md->lsm_stripe_size);
1242
1243         cfs_list_for_each(pos, &lovset->set_list) {
1244                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1245
1246                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1247                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1248                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1249                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1250                                        &req->rq_oi, rqset);
1251                 if (rc) {
1252                         CERROR("%s: getattr objid "DOSTID" subobj"
1253                                DOSTID" on OST idx %d: rc = %d\n",
1254                                exp->exp_obd->obd_name,
1255                                POSTID(&oinfo->oi_oa->o_oi),
1256                                POSTID(&req->rq_oi.oi_oa->o_oi),
1257                                req->rq_idx, rc);
1258                         GOTO(out, rc);
1259                 }
1260         }
1261
1262         if (!cfs_list_empty(&rqset->set_requests)) {
1263                 LASSERT(rc == 0);
1264                 LASSERT (rqset->set_interpret == NULL);
1265                 rqset->set_interpret = lov_getattr_interpret;
1266                 rqset->set_arg = (void *)lovset;
1267                 RETURN(rc);
1268         }
1269 out:
1270         if (rc)
1271                 cfs_atomic_set(&lovset->set_completes, 0);
1272         err = lov_fini_getattr_set(lovset);
1273         RETURN(rc ? rc : err);
1274 }
1275
1276 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1277                        struct obd_info *oinfo, struct obd_trans_info *oti)
1278 {
1279         struct lov_request_set *set;
1280         struct lov_obd *lov;
1281         cfs_list_t *pos;
1282         struct lov_request *req;
1283         int err = 0, rc = 0;
1284         ENTRY;
1285
1286         LASSERT(oinfo);
1287         ASSERT_LSM_MAGIC(oinfo->oi_md);
1288
1289         if (!exp || !exp->exp_obd)
1290                 RETURN(-ENODEV);
1291
1292         /* for now, we only expect the following updates here */
1293         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1294                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1295                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1296                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1297                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1298                                             OBD_MD_FLGID | OBD_MD_FLFID |
1299                                             OBD_MD_FLGENER)));
1300         lov = &exp->exp_obd->u.lov;
1301         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1302         if (rc)
1303                 RETURN(rc);
1304
1305         cfs_list_for_each (pos, &set->set_list) {
1306                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1307
1308                 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1309                                  &req->rq_oi, NULL);
1310                 err = lov_update_setattr_set(set, req, rc);
1311                 if (err) {
1312                         CERROR("%s: setattr objid "DOSTID" subobj "
1313                                DOSTID" on OST idx %d: rc = %d\n",
1314                                exp->exp_obd->obd_name,
1315                                POSTID(&set->set_oi->oi_oa->o_oi),
1316                                POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1317                                err);
1318                         if (!rc)
1319                                 rc = err;
1320                 }
1321         }
1322         err = lov_fini_setattr_set(set);
1323         if (!rc)
1324                 rc = err;
1325         RETURN(rc);
1326 }
1327
1328 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1329                                  void *data, int rc)
1330 {
1331         struct lov_request_set *lovset = (struct lov_request_set *)data;
1332         int err;
1333         ENTRY;
1334
1335         if (rc)
1336                 cfs_atomic_set(&lovset->set_completes, 0);
1337         err = lov_fini_setattr_set(lovset);
1338         RETURN(rc ? rc : err);
1339 }
1340
1341 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1342    needed. Otherwise, a client is waiting for responses. */
1343 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1344                              struct obd_trans_info *oti,
1345                              struct ptlrpc_request_set *rqset)
1346 {
1347         struct lov_request_set *set;
1348         struct lov_request *req;
1349         cfs_list_t *pos;
1350         struct lov_obd *lov;
1351         int rc = 0;
1352         ENTRY;
1353
1354         LASSERT(oinfo);
1355         ASSERT_LSM_MAGIC(oinfo->oi_md);
1356         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1357                 LASSERT(oti);
1358                 LASSERT(oti->oti_logcookies);
1359         }
1360
1361         if (!exp || !exp->exp_obd)
1362                 RETURN(-ENODEV);
1363
1364         lov = &exp->exp_obd->u.lov;
1365         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1366         if (rc)
1367                 RETURN(rc);
1368
1369         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1370                POSTID(&oinfo->oi_md->lsm_oi),
1371                oinfo->oi_md->lsm_stripe_count,
1372                oinfo->oi_md->lsm_stripe_size);
1373
1374         cfs_list_for_each(pos, &set->set_list) {
1375                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1376
1377                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1378                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1379
1380                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1381                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1382                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1383
1384                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1385                                        &req->rq_oi, oti, rqset);
1386                 if (rc) {
1387                         CERROR("error: setattr objid "DOSTID" subobj"
1388                                DOSTID" on OST idx %d: rc = %d\n",
1389                                POSTID(&set->set_oi->oi_oa->o_oi),
1390                                POSTID(&req->rq_oi.oi_oa->o_oi),
1391                                req->rq_idx, rc);
1392                         break;
1393                 }
1394         }
1395
1396         /* If we are not waiting for responses on async requests, return. */
1397         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1398                 int err;
1399                 if (rc)
1400                         cfs_atomic_set(&set->set_completes, 0);
1401                 err = lov_fini_setattr_set(set);
1402                 RETURN(rc ? rc : err);
1403         }
1404
1405         LASSERT(rqset->set_interpret == NULL);
1406         rqset->set_interpret = lov_setattr_interpret;
1407         rqset->set_arg = (void *)set;
1408
1409         RETURN(0);
1410 }
1411
1412 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1413                                void *data, int rc)
1414 {
1415         struct lov_request_set *lovset = (struct lov_request_set *)data;
1416         int err;
1417         ENTRY;
1418
1419         if (rc)
1420                 cfs_atomic_set(&lovset->set_completes, 0);
1421         err = lov_fini_punch_set(lovset);
1422         RETURN(rc ? rc : err);
1423 }
1424
1425 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1426  * we can send this 'punch' to just the authoritative node and the nodes
1427  * that the punch will affect. */
1428 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1429                      struct obd_info *oinfo, struct obd_trans_info *oti,
1430                      struct ptlrpc_request_set *rqset)
1431 {
1432         struct lov_request_set *set;
1433         struct lov_obd *lov;
1434         cfs_list_t *pos;
1435         struct lov_request *req;
1436         int rc = 0;
1437         ENTRY;
1438
1439         LASSERT(oinfo);
1440         ASSERT_LSM_MAGIC(oinfo->oi_md);
1441
1442         if (!exp || !exp->exp_obd)
1443                 RETURN(-ENODEV);
1444
1445         lov = &exp->exp_obd->u.lov;
1446         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1447         if (rc)
1448                 RETURN(rc);
1449
1450         cfs_list_for_each (pos, &set->set_list) {
1451                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1452
1453                 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1454                                &req->rq_oi, NULL, rqset);
1455                 if (rc) {
1456                         CERROR("%s: punch objid "DOSTID" subobj "DOSTID
1457                                " on OST idx %d: rc = %d\n",
1458                                exp->exp_obd->obd_name,
1459                                POSTID(&set->set_oi->oi_oa->o_oi),
1460                                POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, rc);
1461                         break;
1462                 }
1463         }
1464
1465         if (rc || cfs_list_empty(&rqset->set_requests)) {
1466                 int err;
1467                 err = lov_fini_punch_set(set);
1468                 RETURN(rc ? rc : err);
1469         }
1470
1471         LASSERT(rqset->set_interpret == NULL);
1472         rqset->set_interpret = lov_punch_interpret;
1473         rqset->set_arg = (void *)set;
1474
1475         RETURN(0);
1476 }
1477
1478 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1479                               void *data, int rc)
1480 {
1481         struct lov_request_set *lovset = data;
1482         int err;
1483         ENTRY;
1484
1485         if (rc)
1486                 cfs_atomic_set(&lovset->set_completes, 0);
1487         err = lov_fini_sync_set(lovset);
1488         RETURN(rc ?: err);
1489 }
1490
1491 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1492                     struct obd_info *oinfo, obd_off start, obd_off end,
1493                     struct ptlrpc_request_set *rqset)
1494 {
1495         struct lov_request_set *set = NULL;
1496         struct lov_obd *lov;
1497         cfs_list_t *pos;
1498         struct lov_request *req;
1499         int rc = 0;
1500         ENTRY;
1501
1502         ASSERT_LSM_MAGIC(oinfo->oi_md);
1503         LASSERT(rqset != NULL);
1504
1505         if (!exp->exp_obd)
1506                 RETURN(-ENODEV);
1507
1508         lov = &exp->exp_obd->u.lov;
1509         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1510         if (rc)
1511                 RETURN(rc);
1512
1513         CDEBUG(D_INFO, "fsync objid "DOSTID" ["LPX64", "LPX64"]\n",
1514                POSTID(&set->set_oi->oi_oa->o_oi), start, end);
1515
1516         cfs_list_for_each (pos, &set->set_list) {
1517                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1518
1519                 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1520                               &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1521                               req->rq_oi.oi_policy.l_extent.end, rqset);
1522                 if (rc) {
1523                         CERROR("%s: fsync objid "DOSTID" subobj "DOSTID
1524                                " on OST idx %d: rc = %d\n",
1525                                exp->exp_obd->obd_name,
1526                                POSTID(&set->set_oi->oi_oa->o_oi),
1527                                POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1528                                rc);
1529                         break;
1530                 }
1531         }
1532
1533         /* If we are not waiting for responses on async requests, return. */
1534         if (rc || cfs_list_empty(&rqset->set_requests)) {
1535                 int err = lov_fini_sync_set(set);
1536
1537                 RETURN(rc ?: err);
1538         }
1539
1540         LASSERT(rqset->set_interpret == NULL);
1541         rqset->set_interpret = lov_sync_interpret;
1542         rqset->set_arg = (void *)set;
1543
1544         RETURN(0);
1545 }
1546
1547 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1548                          obd_count oa_bufs, struct brw_page *pga)
1549 {
1550         struct obd_info oinfo = { { { 0 } } };
1551         int i, rc = 0;
1552
1553         oinfo.oi_oa = lov_oinfo->oi_oa;
1554
1555         /* The caller just wants to know if there's a chance that this
1556          * I/O can succeed */
1557         for (i = 0; i < oa_bufs; i++) {
1558                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1559                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1560                 obd_off start, end;
1561
1562                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1563                                            pga[i].off + pga[i].count - 1,
1564                                            &start, &end))
1565                         continue;
1566
1567                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1568                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1569                         return -EIO;
1570                 }
1571
1572                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1573                              1, &pga[i], NULL);
1574                 if (rc)
1575                         break;
1576         }
1577         return rc;
1578 }
1579
1580 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1581                    obd_count oa_bufs, struct brw_page *pga,
1582                    struct obd_trans_info *oti)
1583 {
1584         struct lov_request_set *set;
1585         struct lov_request *req;
1586         cfs_list_t *pos;
1587         struct lov_obd *lov = &exp->exp_obd->u.lov;
1588         int err, rc = 0;
1589         ENTRY;
1590
1591         ASSERT_LSM_MAGIC(oinfo->oi_md);
1592
1593         if (cmd == OBD_BRW_CHECK) {
1594                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1595                 RETURN(rc);
1596         }
1597
1598         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1599         if (rc)
1600                 RETURN(rc);
1601
1602         cfs_list_for_each (pos, &set->set_list) {
1603                 struct obd_export *sub_exp;
1604                 struct brw_page *sub_pga;
1605                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1606
1607                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1608                 sub_pga = set->set_pga + req->rq_pgaidx;
1609                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1610                              sub_pga, oti);
1611                 if (rc)
1612                         break;
1613                 lov_update_common_set(set, req, rc);
1614         }
1615
1616         err = lov_fini_brw_set(set);
1617         if (!rc)
1618                 rc = err;
1619         RETURN(rc);
1620 }
1621
1622 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1623                                  void *data, int rc)
1624 {
1625         struct lov_request_set *lovset = (struct lov_request_set *)data;
1626         ENTRY;
1627         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1628         RETURN(rc);
1629 }
1630
1631 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1632                        struct ldlm_enqueue_info *einfo,
1633                        struct ptlrpc_request_set *rqset)
1634 {
1635         ldlm_mode_t mode = einfo->ei_mode;
1636         struct lov_request_set *set;
1637         struct lov_request *req;
1638         cfs_list_t *pos;
1639         struct lov_obd *lov;
1640         ldlm_error_t rc;
1641         ENTRY;
1642
1643         LASSERT(oinfo);
1644         ASSERT_LSM_MAGIC(oinfo->oi_md);
1645         LASSERT(mode == (mode & -mode));
1646
1647         /* we should never be asked to replay a lock this way. */
1648         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1649
1650         if (!exp || !exp->exp_obd)
1651                 RETURN(-ENODEV);
1652
1653         lov = &exp->exp_obd->u.lov;
1654         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1655         if (rc)
1656                 RETURN(rc);
1657
1658         cfs_list_for_each (pos, &set->set_list) {
1659                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1660
1661                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1662                                  &req->rq_oi, einfo, rqset);
1663                 if (rc != ELDLM_OK)
1664                         GOTO(out, rc);
1665         }
1666
1667         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1668                 LASSERT(rc == 0);
1669                 LASSERT(rqset->set_interpret == NULL);
1670                 rqset->set_interpret = lov_enqueue_interpret;
1671                 rqset->set_arg = (void *)set;
1672                 RETURN(rc);
1673         }
1674 out:
1675         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1676         RETURN(rc);
1677 }
1678
1679 static int lov_change_cbdata(struct obd_export *exp,
1680                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1681                              void *data)
1682 {
1683         struct lov_obd *lov;
1684         int rc = 0, i;
1685         ENTRY;
1686
1687         ASSERT_LSM_MAGIC(lsm);
1688
1689         if (!exp || !exp->exp_obd)
1690                 RETURN(-ENODEV);
1691
1692         lov = &exp->exp_obd->u.lov;
1693         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1694                 struct lov_stripe_md submd;
1695                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1696
1697                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1698                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1699                         continue;
1700                 }
1701
1702                 submd.lsm_oi = loi->loi_oi;
1703                 submd.lsm_stripe_count = 0;
1704                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1705                                        &submd, it, data);
1706         }
1707         RETURN(rc);
1708 }
1709
1710 /* find any ldlm lock of the inode in lov
1711  * return 0    not find
1712  *        1    find one
1713  *      < 0    error */
1714 static int lov_find_cbdata(struct obd_export *exp,
1715                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1716                            void *data)
1717 {
1718         struct lov_obd *lov;
1719         int rc = 0, i;
1720         ENTRY;
1721
1722         ASSERT_LSM_MAGIC(lsm);
1723
1724         if (!exp || !exp->exp_obd)
1725                 RETURN(-ENODEV);
1726
1727         lov = &exp->exp_obd->u.lov;
1728         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1729                 struct lov_stripe_md submd;
1730                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1731
1732                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1733                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1734                         continue;
1735                 }
1736                 submd.lsm_oi = loi->loi_oi;
1737                 submd.lsm_stripe_count = 0;
1738                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1739                                      &submd, it, data);
1740                 if (rc != 0)
1741                         RETURN(rc);
1742         }
1743         RETURN(rc);
1744 }
1745
1746 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1747                       __u32 mode, struct lustre_handle *lockh)
1748 {
1749         struct lov_request_set *set;
1750         struct obd_info oinfo;
1751         struct lov_request *req;
1752         cfs_list_t *pos;
1753         struct lov_obd *lov;
1754         struct lustre_handle *lov_lockhp;
1755         int err = 0, rc = 0;
1756         ENTRY;
1757
1758         ASSERT_LSM_MAGIC(lsm);
1759
1760         if (!exp || !exp->exp_obd)
1761                 RETURN(-ENODEV);
1762
1763         LASSERT(lockh);
1764         lov = &exp->exp_obd->u.lov;
1765         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1766         if (rc)
1767                 RETURN(rc);
1768
1769         cfs_list_for_each(pos, &set->set_list) {
1770                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1771                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1772
1773                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1774                                 req->rq_oi.oi_md, mode, lov_lockhp);
1775                 rc = lov_update_common_set(set, req, rc);
1776                 if (rc) {
1777                         CERROR("%s: cancel objid "DOSTID" subobj "
1778                                DOSTID" on OST idx %d: rc = %d\n",
1779                                exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1780                                POSTID(&req->rq_oi.oi_md->lsm_oi),
1781                                req->rq_idx, rc);
1782                         err = rc;
1783                 }
1784
1785         }
1786         lov_fini_cancel_set(set);
1787         RETURN(err);
1788 }
1789
1790 static int lov_cancel_unused(struct obd_export *exp,
1791                              struct lov_stripe_md *lsm,
1792                              ldlm_cancel_flags_t flags, void *opaque)
1793 {
1794         struct lov_obd *lov;
1795         int rc = 0, i;
1796         ENTRY;
1797
1798         if (!exp || !exp->exp_obd)
1799                 RETURN(-ENODEV);
1800
1801         lov = &exp->exp_obd->u.lov;
1802         if (lsm == NULL) {
1803                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1804                         int err;
1805                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1806                                 continue;
1807
1808                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1809                                                 flags, opaque);
1810                         if (!rc)
1811                                 rc = err;
1812                 }
1813                 RETURN(rc);
1814         }
1815
1816         ASSERT_LSM_MAGIC(lsm);
1817
1818         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1819                 struct lov_stripe_md submd;
1820                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1821                 int idx = loi->loi_ost_idx;
1822                 int err;
1823
1824                 if (!lov->lov_tgts[idx]) {
1825                         CDEBUG(D_HA, "lov idx %d NULL\n", idx);
1826                         continue;
1827                 }
1828
1829                 if (!lov->lov_tgts[idx]->ltd_active)
1830                         CDEBUG(D_HA, "lov idx %d inactive\n", idx);
1831
1832                 submd.lsm_oi = loi->loi_oi;
1833                 submd.lsm_stripe_count = 0;
1834                 err = obd_cancel_unused(lov->lov_tgts[idx]->ltd_exp,
1835                                         &submd, flags, opaque);
1836                 if (err && lov->lov_tgts[idx]->ltd_active) {
1837                         CERROR("%s: cancel unused objid "DOSTID
1838                                " subobj "DOSTID" on OST idx %d: rc = %d\n",
1839                                exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1840                                POSTID(&loi->loi_oi), idx, err);
1841                         if (!rc)
1842                                 rc = err;
1843                 }
1844         }
1845         RETURN(rc);
1846 }
1847
1848 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1849 {
1850         struct lov_request_set *lovset = (struct lov_request_set *)data;
1851         int err;
1852         ENTRY;
1853
1854         if (rc)
1855                 cfs_atomic_set(&lovset->set_completes, 0);
1856
1857         err = lov_fini_statfs_set(lovset);
1858         RETURN(rc ? rc : err);
1859 }
1860
1861 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1862                             __u64 max_age, struct ptlrpc_request_set *rqset)
1863 {
1864         struct obd_device      *obd = class_exp2obd(exp);
1865         struct lov_request_set *set;
1866         struct lov_request *req;
1867         cfs_list_t *pos;
1868         struct lov_obd *lov;
1869         int rc = 0;
1870         ENTRY;
1871
1872         LASSERT(oinfo != NULL);
1873         LASSERT(oinfo->oi_osfs != NULL);
1874
1875         lov = &obd->u.lov;
1876         rc = lov_prep_statfs_set(obd, oinfo, &set);
1877         if (rc)
1878                 RETURN(rc);
1879
1880         cfs_list_for_each (pos, &set->set_list) {
1881                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1882                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1883                                       &req->rq_oi, max_age, rqset);
1884                 if (rc)
1885                         break;
1886         }
1887
1888         if (rc || cfs_list_empty(&rqset->set_requests)) {
1889                 int err;
1890                 if (rc)
1891                         cfs_atomic_set(&set->set_completes, 0);
1892                 err = lov_fini_statfs_set(set);
1893                 RETURN(rc ? rc : err);
1894         }
1895
1896         LASSERT(rqset->set_interpret == NULL);
1897         rqset->set_interpret = lov_statfs_interpret;
1898         rqset->set_arg = (void *)set;
1899         RETURN(0);
1900 }
1901
1902 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1903                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1904 {
1905         struct ptlrpc_request_set *set = NULL;
1906         struct obd_info oinfo = { { { 0 } } };
1907         int rc = 0;
1908         ENTRY;
1909
1910
1911         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1912          * statfs requests */
1913         set = ptlrpc_prep_set();
1914         if (set == NULL)
1915                 RETURN(-ENOMEM);
1916
1917         oinfo.oi_osfs = osfs;
1918         oinfo.oi_flags = flags;
1919         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1920         if (rc == 0)
1921                 rc = ptlrpc_set_wait(set);
1922         ptlrpc_set_destroy(set);
1923
1924         RETURN(rc);
1925 }
1926
1927 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1928                          void *karg, void *uarg)
1929 {
1930         struct obd_device *obddev = class_exp2obd(exp);
1931         struct lov_obd *lov = &obddev->u.lov;
1932         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1933         struct obd_uuid *uuidp;
1934         ENTRY;
1935
1936         switch (cmd) {
1937         case IOC_OBD_STATFS: {
1938                 struct obd_ioctl_data *data = karg;
1939                 struct obd_device *osc_obd;
1940                 struct obd_statfs stat_buf = {0};
1941                 __u32 index;
1942                 __u32 flags;
1943
1944                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1945                 if ((index >= count))
1946                         RETURN(-ENODEV);
1947
1948                 if (!lov->lov_tgts[index])
1949                         /* Try again with the next index */
1950                         RETURN(-EAGAIN);
1951                 if (!lov->lov_tgts[index]->ltd_active)
1952                         RETURN(-ENODATA);
1953
1954                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1955                 if (!osc_obd)
1956                         RETURN(-EINVAL);
1957
1958                 /* copy UUID */
1959                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1960                                      min((int) data->ioc_plen2,
1961                                          (int) sizeof(struct obd_uuid))))
1962                         RETURN(-EFAULT);
1963
1964                 flags = uarg ? *(__u32*)uarg : 0;
1965                 /* got statfs data */
1966                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1967                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1968                                 flags);
1969                 if (rc)
1970                         RETURN(rc);
1971                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1972                                      min((int) data->ioc_plen1,
1973                                          (int) sizeof(stat_buf))))
1974                         RETURN(-EFAULT);
1975                 break;
1976         }
1977         case OBD_IOC_LOV_GET_CONFIG: {
1978                 struct obd_ioctl_data *data;
1979                 struct lov_desc *desc;
1980                 char *buf = NULL;
1981                 __u32 *genp;
1982
1983                 len = 0;
1984                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1985                         RETURN(-EINVAL);
1986
1987                 data = (struct obd_ioctl_data *)buf;
1988
1989                 if (sizeof(*desc) > data->ioc_inllen1) {
1990                         obd_ioctl_freedata(buf, len);
1991                         RETURN(-EINVAL);
1992                 }
1993
1994                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1995                         obd_ioctl_freedata(buf, len);
1996                         RETURN(-EINVAL);
1997                 }
1998
1999                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2000                         obd_ioctl_freedata(buf, len);
2001                         RETURN(-EINVAL);
2002                 }
2003
2004                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2005                 memcpy(desc, &(lov->desc), sizeof(*desc));
2006
2007                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2008                 genp = (__u32 *)data->ioc_inlbuf3;
2009                 /* the uuid will be empty for deleted OSTs */
2010                 for (i = 0; i < count; i++, uuidp++, genp++) {
2011                         if (!lov->lov_tgts[i])
2012                                 continue;
2013                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2014                         *genp = lov->lov_tgts[i]->ltd_gen;
2015                 }
2016
2017                 if (cfs_copy_to_user((void *)uarg, buf, len))
2018                         rc = -EFAULT;
2019                 obd_ioctl_freedata(buf, len);
2020                 break;
2021         }
2022         case LL_IOC_LOV_SETSTRIPE:
2023                 rc = lov_setstripe(exp, len, karg, uarg);
2024                 break;
2025         case LL_IOC_LOV_GETSTRIPE:
2026                 rc = lov_getstripe(exp, karg, uarg);
2027                 break;
2028         case LL_IOC_LOV_SETEA:
2029                 rc = lov_setea(exp, karg, uarg);
2030                 break;
2031         case OBD_IOC_QUOTACTL: {
2032                 struct if_quotactl *qctl = karg;
2033                 struct lov_tgt_desc *tgt = NULL;
2034                 struct obd_quotactl *oqctl;
2035
2036                 if (qctl->qc_valid == QC_OSTIDX) {
2037                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2038                                 RETURN(-EINVAL);
2039
2040                         tgt = lov->lov_tgts[qctl->qc_idx];
2041                         if (!tgt || !tgt->ltd_exp)
2042                                 RETURN(-EINVAL);
2043                 } else if (qctl->qc_valid == QC_UUID) {
2044                         for (i = 0; i < count; i++) {
2045                                 tgt = lov->lov_tgts[i];
2046                                 if (!tgt ||
2047                                     !obd_uuid_equals(&tgt->ltd_uuid,
2048                                                      &qctl->obd_uuid))
2049                                         continue;
2050
2051                                 if (tgt->ltd_exp == NULL)
2052                                         RETURN(-EINVAL);
2053
2054                                 break;
2055                         }
2056                 } else {
2057                         RETURN(-EINVAL);
2058                 }
2059
2060                 if (i >= count)
2061                         RETURN(-EAGAIN);
2062
2063                 LASSERT(tgt && tgt->ltd_exp);
2064                 OBD_ALLOC_PTR(oqctl);
2065                 if (!oqctl)
2066                         RETURN(-ENOMEM);
2067
2068                 QCTL_COPY(oqctl, qctl);
2069                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2070                 if (rc == 0) {
2071                         QCTL_COPY(qctl, oqctl);
2072                         qctl->qc_valid = QC_OSTIDX;
2073                         qctl->obd_uuid = tgt->ltd_uuid;
2074                 }
2075                 OBD_FREE_PTR(oqctl);
2076                 break;
2077         }
2078         default: {
2079                 int set = 0;
2080
2081                 if (count == 0)
2082                         RETURN(-ENOTTY);
2083
2084                 for (i = 0; i < count; i++) {
2085                         int err;
2086                         struct obd_device *osc_obd;
2087
2088                         /* OST was disconnected */
2089                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2090                                 continue;
2091
2092                         /* ll_umount_begin() sets force flag but for lov, not
2093                          * osc. Let's pass it through */
2094                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2095                         osc_obd->obd_force = obddev->obd_force;
2096                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2097                                             len, karg, uarg);
2098                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2099                                 RETURN(err);
2100                         } else if (err) {
2101                                 if (lov->lov_tgts[i]->ltd_active) {
2102                                         CDEBUG(err == -ENOTTY ?
2103                                                D_IOCTL : D_WARNING,
2104                                                "iocontrol OSC %s on OST "
2105                                                "idx %d cmd %x: err = %d\n",
2106                                                lov_uuid2str(lov, i),
2107                                                i, cmd, err);
2108                                         if (!rc)
2109                                                 rc = err;
2110                                 }
2111                         } else {
2112                                 set = 1;
2113                         }
2114                 }
2115                 if (!set && !rc)
2116                         rc = -EIO;
2117         }
2118         }
2119
2120         RETURN(rc);
2121 }
2122
2123 #define FIEMAP_BUFFER_SIZE 4096
2124
2125 /**
2126  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2127  * call. The local end offset and the device are sent in the first
2128  * fm_extent. This function calculates the stripe number from the index.
2129  * This function returns a stripe_no on which mapping is to be restarted.
2130  *
2131  * This function returns fm_end_offset which is the in-OST offset at which
2132  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2133  * will re-calculate proper offset in next stripe.
2134  * Note that the first extent is passed to lov_get_info via the value field.
2135  *
2136  * \param fiemap fiemap request header
2137  * \param lsm striping information for the file
2138  * \param fm_start logical start of mapping
2139  * \param fm_end logical end of mapping
2140  * \param start_stripe starting stripe will be returned in this
2141  */
2142 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2143                                    struct lov_stripe_md *lsm, obd_size fm_start,
2144                                    obd_size fm_end, int *start_stripe)
2145 {
2146         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2147         obd_off lun_start, lun_end;
2148         obd_size fm_end_offset;
2149         int stripe_no = -1, i;
2150
2151         if (fiemap->fm_extent_count == 0 ||
2152             fiemap->fm_extents[0].fe_logical == 0)
2153                 return 0;
2154
2155         /* Find out stripe_no from ost_index saved in the fe_device */
2156         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2157                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2158                                         fiemap->fm_extents[0].fe_device) {
2159                         stripe_no = i;
2160                         break;
2161                 }
2162         }
2163         if (stripe_no == -1)
2164                 return -EINVAL;
2165
2166         /* If we have finished mapping on previous device, shift logical
2167          * offset to start of next device */
2168         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2169                                    &lun_start, &lun_end)) != 0 &&
2170                                    local_end < lun_end) {
2171                 fm_end_offset = local_end;
2172                 *start_stripe = stripe_no;
2173         } else {
2174                 /* This is a special value to indicate that caller should
2175                  * calculate offset in next stripe. */
2176                 fm_end_offset = 0;
2177                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2178         }
2179
2180         return fm_end_offset;
2181 }
2182
2183 /**
2184  * We calculate on which OST the mapping will end. If the length of mapping
2185  * is greater than (stripe_size * stripe_count) then the last_stripe will
2186  * will be one just before start_stripe. Else we check if the mapping
2187  * intersects each OST and find last_stripe.
2188  * This function returns the last_stripe and also sets the stripe_count
2189  * over which the mapping is spread
2190  *
2191  * \param lsm striping information for the file
2192  * \param fm_start logical start of mapping
2193  * \param fm_end logical end of mapping
2194  * \param start_stripe starting stripe of the mapping
2195  * \param stripe_count the number of stripes across which to map is returned
2196  *
2197  * \retval last_stripe return the last stripe of the mapping
2198  */
2199 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2200                             obd_size fm_end, int start_stripe,
2201                             int *stripe_count)
2202 {
2203         int last_stripe;
2204         obd_off obd_start, obd_end;
2205         int i, j;
2206
2207         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2208                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2209                                                               start_stripe - 1);
2210                 *stripe_count = lsm->lsm_stripe_count;
2211         } else {
2212                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2213                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2214                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2215                                                    &obd_start, &obd_end)) == 0)
2216                                 break;
2217                 }
2218                 *stripe_count = j;
2219                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2220         }
2221
2222         return last_stripe;
2223 }
2224
2225 /**
2226  * Set fe_device and copy extents from local buffer into main return buffer.
2227  *
2228  * \param fiemap fiemap request header
2229  * \param lcl_fm_ext array of local fiemap extents to be copied
2230  * \param ost_index OST index to be written into the fm_device field for each
2231                     extent
2232  * \param ext_count number of extents to be copied
2233  * \param current_extent where to start copying in main extent array
2234  */
2235 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2236                                   struct ll_fiemap_extent *lcl_fm_ext,
2237                                   int ost_index, unsigned int ext_count,
2238                                   int current_extent)
2239 {
2240         char *to;
2241         int ext;
2242
2243         for (ext = 0; ext < ext_count; ext++) {
2244                 lcl_fm_ext[ext].fe_device = ost_index;
2245                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2246         }
2247
2248         /* Copy fm_extent's from fm_local to return buffer */
2249         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2250         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2251 }
2252
2253 /**
2254  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2255  * This also handles the restarting of FIEMAP calls in case mapping overflows
2256  * the available number of extents in single call.
2257  */
2258 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2259                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2260 {
2261         struct ll_fiemap_info_key *fm_key = key;
2262         struct ll_user_fiemap *fiemap = val;
2263         struct ll_user_fiemap *fm_local = NULL;
2264         struct ll_fiemap_extent *lcl_fm_ext;
2265         int count_local;
2266         unsigned int get_num_extents = 0;
2267         int ost_index = 0, actual_start_stripe, start_stripe;
2268         obd_size fm_start, fm_end, fm_length, fm_end_offset;
2269         obd_size curr_loc;
2270         int current_extent = 0, rc = 0, i;
2271         int ost_eof = 0; /* EOF for object */
2272         int ost_done = 0; /* done with required mapping for this OST? */
2273         int last_stripe;
2274         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2275         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2276
2277         if (lsm == NULL)
2278                 GOTO(out, rc = 0);
2279
2280         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2281                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2282
2283         OBD_ALLOC_LARGE(fm_local, buffer_size);
2284         if (fm_local == NULL)
2285                 GOTO(out, rc = -ENOMEM);
2286         lcl_fm_ext = &fm_local->fm_extents[0];
2287
2288         count_local = fiemap_size_to_count(buffer_size);
2289
2290         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2291         fm_start = fiemap->fm_start;
2292         fm_length = fiemap->fm_length;
2293         /* Calculate start stripe, last stripe and length of mapping */
2294         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2295         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2296                                                 fm_start + fm_length - 1);
2297         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2298         if (fm_end > fm_key->oa.o_size)
2299                 fm_end = fm_key->oa.o_size;
2300
2301         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2302                                             actual_start_stripe, &stripe_count);
2303
2304         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2305                                                   fm_end, &start_stripe);
2306         if (fm_end_offset == -EINVAL)
2307                 GOTO(out, rc = -EINVAL);
2308
2309         if (fiemap->fm_extent_count == 0) {
2310                 get_num_extents = 1;
2311                 count_local = 0;
2312         }
2313
2314         /* Check each stripe */
2315         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2316              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2317                 obd_size req_fm_len; /* Stores length of required mapping */
2318                 obd_size len_mapped_single_call;
2319                 obd_off lun_start, lun_end, obd_object_end;
2320                 unsigned int ext_count;
2321
2322                 cur_stripe_wrap = cur_stripe;
2323
2324                 /* Find out range of mapping on this stripe */
2325                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2326                                            &lun_start, &obd_object_end)) == 0)
2327                         continue;
2328
2329                 /* If this is a continuation FIEMAP call and we are on
2330                  * starting stripe then lun_start needs to be set to
2331                  * fm_end_offset */
2332                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2333                         lun_start = fm_end_offset;
2334
2335                 if (fm_length != ~0ULL) {
2336                         /* Handle fm_start + fm_length overflow */
2337                         if (fm_start + fm_length < fm_start)
2338                                 fm_length = ~0ULL - fm_start;
2339                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2340                                                      cur_stripe);
2341                 } else {
2342                         lun_end = ~0ULL;
2343                 }
2344
2345                 if (lun_start == lun_end)
2346                         continue;
2347
2348                 req_fm_len = obd_object_end - lun_start;
2349                 fm_local->fm_length = 0;
2350                 len_mapped_single_call = 0;
2351
2352                 /* If the output buffer is very large and the objects have many
2353                  * extents we may need to loop on a single OST repeatedly */
2354                 ost_eof = 0;
2355                 ost_done = 0;
2356                 do {
2357                         if (get_num_extents == 0) {
2358                                 /* Don't get too many extents. */
2359                                 if (current_extent + count_local >
2360                                     fiemap->fm_extent_count)
2361                                         count_local = fiemap->fm_extent_count -
2362                                                                  current_extent;
2363                         }
2364
2365                         lun_start += len_mapped_single_call;
2366                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2367                         req_fm_len = fm_local->fm_length;
2368                         fm_local->fm_extent_count = count_local;
2369                         fm_local->fm_mapped_extents = 0;
2370                         fm_local->fm_flags = fiemap->fm_flags;
2371
2372                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
2373                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2374
2375                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2376                                 GOTO(out, rc = -EINVAL);
2377
2378                         /* If OST is inactive, return extent with UNKNOWN flag */
2379                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2380                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2381                                 fm_local->fm_mapped_extents = 1;
2382
2383                                 lcl_fm_ext[0].fe_logical = lun_start;
2384                                 lcl_fm_ext[0].fe_length = obd_object_end -
2385                                                                       lun_start;
2386                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2387
2388                                 goto inactive_tgt;
2389                         }
2390
2391                         fm_local->fm_start = lun_start;
2392                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2393                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2394                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2395                         rc = obd_get_info(NULL,
2396                                           lov->lov_tgts[ost_index]->ltd_exp,
2397                                           keylen, key, vallen, fm_local, lsm);
2398                         if (rc != 0)
2399                                 GOTO(out, rc);
2400
2401 inactive_tgt:
2402                         ext_count = fm_local->fm_mapped_extents;
2403                         if (ext_count == 0) {
2404                                 ost_done = 1;
2405                                 /* If last stripe has hole at the end,
2406                                  * then we need to return */
2407                                 if (cur_stripe_wrap == last_stripe) {
2408                                         fiemap->fm_mapped_extents = 0;
2409                                         goto finish;
2410                                 }
2411                                 break;
2412                         }
2413
2414                         /* If we just need num of extents then go to next device */
2415                         if (get_num_extents) {
2416                                 current_extent += ext_count;
2417                                 break;
2418                         }
2419
2420                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2421                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2422
2423                         /* Have we finished mapping on this device? */
2424                         if (req_fm_len <= len_mapped_single_call)
2425                                 ost_done = 1;
2426
2427                         /* Clear the EXTENT_LAST flag which can be present on
2428                          * last extent */
2429                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2430                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2431                                                             ~FIEMAP_EXTENT_LAST;
2432
2433                         curr_loc = lov_stripe_size(lsm,
2434                                            lcl_fm_ext[ext_count - 1].fe_logical+
2435                                            lcl_fm_ext[ext_count - 1].fe_length,
2436                                            cur_stripe);
2437                         if (curr_loc >= fm_key->oa.o_size)
2438                                 ost_eof = 1;
2439
2440                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2441                                                      ost_index, ext_count,
2442                                                      current_extent);
2443
2444                         current_extent += ext_count;
2445
2446                         /* Ran out of available extents? */
2447                         if (current_extent >= fiemap->fm_extent_count)
2448                                 goto finish;
2449                 } while (ost_done == 0 && ost_eof == 0);
2450
2451                 if (cur_stripe_wrap == last_stripe)
2452                         goto finish;
2453         }
2454
2455 finish:
2456         /* Indicate that we are returning device offsets unless file just has
2457          * single stripe */
2458         if (lsm->lsm_stripe_count > 1)
2459                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2460
2461         if (get_num_extents)
2462                 goto skip_last_device_calc;
2463
2464         /* Check if we have reached the last stripe and whether mapping for that
2465          * stripe is done. */
2466         if (cur_stripe_wrap == last_stripe) {
2467                 if (ost_done || ost_eof)
2468                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2469                                                              FIEMAP_EXTENT_LAST;
2470         }
2471
2472 skip_last_device_calc:
2473         fiemap->fm_mapped_extents = current_extent;
2474
2475 out:
2476         OBD_FREE_LARGE(fm_local, buffer_size);
2477         return rc;
2478 }
2479
2480 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2481                         __u32 keylen, void *key, __u32 *vallen, void *val,
2482                         struct lov_stripe_md *lsm)
2483 {
2484         struct obd_device *obddev = class_exp2obd(exp);
2485         struct lov_obd *lov = &obddev->u.lov;
2486         int i, rc;
2487         ENTRY;
2488
2489         if (!vallen || !val)
2490                 RETURN(-EFAULT);
2491
2492         obd_getref(obddev);
2493
2494         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2495                 struct {
2496                         char name[16];
2497                         struct ldlm_lock *lock;
2498                 } *data = key;
2499                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2500                 struct lov_oinfo *loi;
2501                 __u32 *stripe = val;
2502
2503                 if (*vallen < sizeof(*stripe))
2504                         GOTO(out, rc = -EFAULT);
2505                 *vallen = sizeof(*stripe);
2506
2507                 /* XXX This is another one of those bits that will need to
2508                  * change if we ever actually support nested LOVs.  It uses
2509                  * the lock's export to find out which stripe it is. */
2510                 /* XXX - it's assumed all the locks for deleted OSTs have
2511                  * been cancelled. Also, the export for deleted OSTs will
2512                  * be NULL and won't match the lock's export. */
2513                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2514                         loi = lsm->lsm_oinfo[i];
2515                         if (!lov->lov_tgts[loi->loi_ost_idx])
2516                                 continue;
2517                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2518                             data->lock->l_conn_export &&
2519                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2520                                 *stripe = i;
2521                                 GOTO(out, rc = 0);
2522                         }
2523                 }
2524                 LDLM_ERROR(data->lock, "lock on inode without such object");
2525                 dump_lsm(D_ERROR, lsm);
2526                 GOTO(out, rc = -ENXIO);
2527         } else if (KEY_IS(KEY_LAST_ID)) {
2528                 struct obd_id_info *info = val;
2529                 __u32 size = sizeof(obd_id);
2530                 struct lov_tgt_desc *tgt;
2531
2532                 LASSERT(*vallen == sizeof(struct obd_id_info));
2533                 tgt = lov->lov_tgts[info->idx];
2534
2535                 if (!tgt || !tgt->ltd_active)
2536                         GOTO(out, rc = -ESRCH);
2537
2538                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2539                                   &size, info->data, NULL);
2540                 GOTO(out, rc = 0);
2541         } else if (KEY_IS(KEY_LOVDESC)) {
2542                 struct lov_desc *desc_ret = val;
2543                 *desc_ret = lov->desc;
2544
2545                 GOTO(out, rc = 0);
2546         } else if (KEY_IS(KEY_FIEMAP)) {
2547                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2548                 GOTO(out, rc);
2549         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2550                 struct lov_tgt_desc *tgt;
2551                 __u64 ost_idx = *((__u64*)val);
2552
2553                 LASSERT(*vallen == sizeof(__u64));
2554                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2555                 tgt = lov->lov_tgts[ost_idx];
2556
2557                 if (!tgt || !tgt->ltd_exp)
2558                         GOTO(out, rc = -ESRCH);
2559
2560                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2561                 GOTO(out, rc = 0);
2562         } else if (KEY_IS(KEY_TGT_COUNT)) {
2563                 *((int *)val) = lov->desc.ld_tgt_count;
2564                 GOTO(out, rc = 0);
2565         }
2566
2567         rc = -EINVAL;
2568
2569 out:
2570         obd_putref(obddev);
2571         RETURN(rc);
2572 }
2573
2574 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2575                               obd_count keylen, void *key, obd_count vallen,
2576                               void *val, struct ptlrpc_request_set *set)
2577 {
2578         struct obd_device *obddev = class_exp2obd(exp);
2579         struct lov_obd *lov = &obddev->u.lov;
2580         obd_count count;
2581         int i, rc = 0, err;
2582         struct lov_tgt_desc *tgt;
2583         unsigned incr, check_uuid,
2584                  do_inactive, no_set;
2585         unsigned next_id = 0,  mds_con = 0, capa = 0;
2586         ENTRY;
2587
2588         incr = check_uuid = do_inactive = no_set = 0;
2589         if (set == NULL) {
2590                 no_set = 1;
2591                 set = ptlrpc_prep_set();
2592                 if (!set)
2593                         RETURN(-ENOMEM);
2594         }
2595
2596         obd_getref(obddev);
2597         count = lov->desc.ld_tgt_count;
2598
2599         if (KEY_IS(KEY_NEXT_ID)) {
2600                 count = vallen / sizeof(struct obd_id_info);
2601                 vallen = sizeof(obd_id);
2602                 incr = sizeof(struct obd_id_info);
2603                 do_inactive = 1;
2604                 next_id = 1;
2605         } else if (KEY_IS(KEY_CHECKSUM)) {
2606                 do_inactive = 1;
2607         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2608                 /* use defaults:  do_inactive = incr = 0; */
2609         } else if (KEY_IS(KEY_MDS_CONN)) {
2610                 mds_con = 1;
2611         } else if (KEY_IS(KEY_CAPA_KEY)) {
2612                 capa = 1;
2613         } else if (KEY_IS(KEY_CACHE_SET)) {
2614                 LASSERT(lov->lov_cache == NULL);
2615                 lov->lov_cache = val;
2616                 do_inactive = 1;
2617         }
2618
2619         for (i = 0; i < count; i++, val = (char *)val + incr) {
2620                 if (next_id) {
2621                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2622                 } else {
2623                         tgt = lov->lov_tgts[i];
2624                 }
2625                 /* OST was disconnected */
2626                 if (!tgt || !tgt->ltd_exp)
2627                         continue;
2628
2629                 /* OST is inactive and we don't want inactive OSCs */
2630                 if (!tgt->ltd_active && !do_inactive)
2631                         continue;
2632
2633                 if (mds_con) {
2634                         struct mds_group_info *mgi;
2635
2636                         LASSERT(vallen == sizeof(*mgi));
2637                         mgi = (struct mds_group_info *)val;
2638
2639                         /* Only want a specific OSC */
2640                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2641                                                 &tgt->ltd_uuid))
2642                                 continue;
2643
2644                         err = obd_set_info_async(env, tgt->ltd_exp,
2645                                          keylen, key, sizeof(int),
2646                                          &mgi->group, set);
2647                 } else if (next_id) {
2648                         err = obd_set_info_async(env, tgt->ltd_exp,
2649                                          keylen, key, vallen,
2650                                          ((struct obd_id_info*)val)->data, set);
2651                 } else if (capa) {
2652                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2653
2654                         LASSERT(vallen == sizeof(*info));
2655
2656                          /* Only want a specific OSC */
2657                         if (info->uuid &&
2658                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2659                                 continue;
2660
2661                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2662                                                  key, sizeof(*info->capa),
2663                                                  info->capa, set);
2664                 } else {
2665                         /* Only want a specific OSC */
2666                         if (check_uuid &&
2667                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2668                                 continue;
2669
2670                         err = obd_set_info_async(env, tgt->ltd_exp,
2671                                          keylen, key, vallen, val, set);
2672                 }
2673
2674                 if (!rc)
2675                         rc = err;
2676         }
2677
2678         obd_putref(obddev);
2679         if (no_set) {
2680                 err = ptlrpc_set_wait(set);
2681                 if (!rc)
2682                         rc = err;
2683                 ptlrpc_set_destroy(set);
2684         }
2685         RETURN(rc);
2686 }
2687
2688 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2689                            int cmd, __u64 *offset)
2690 {
2691         __u32 ssize = lsm->lsm_stripe_size;
2692         __u64 start;
2693
2694         start = *offset;
2695         lov_do_div64(start, ssize);
2696         start = start * ssize;
2697
2698         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2699                            ", end "LPU64"\n", *offset, ssize, start,
2700                            start + ssize - 1);
2701         if (cmd == OBD_CALC_STRIPE_END) {
2702                 *offset = start + ssize - 1;
2703         } else if (cmd == OBD_CALC_STRIPE_START) {
2704                 *offset = start;
2705         } else {
2706                 LBUG();
2707         }
2708
2709         RETURN(0);
2710 }
2711
2712 void lov_stripe_lock(struct lov_stripe_md *md)
2713 {
2714         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2715         spin_lock(&md->lsm_lock);
2716         LASSERT(md->lsm_lock_owner == 0);
2717         md->lsm_lock_owner = cfs_curproc_pid();
2718 }
2719 EXPORT_SYMBOL(lov_stripe_lock);
2720
2721 void lov_stripe_unlock(struct lov_stripe_md *md)
2722 {
2723         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2724         md->lsm_lock_owner = 0;
2725         spin_unlock(&md->lsm_lock);
2726 }
2727 EXPORT_SYMBOL(lov_stripe_unlock);
2728
2729 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2730                         struct obd_quotactl *oqctl)
2731 {
2732         struct lov_obd      *lov = &obd->u.lov;
2733         struct lov_tgt_desc *tgt;
2734         __u64                curspace = 0;
2735         __u64                bhardlimit = 0;
2736         int                  i, rc = 0;
2737         ENTRY;
2738
2739         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2740             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2741             oqctl->qc_cmd != Q_GETOQUOTA &&
2742             oqctl->qc_cmd != Q_INITQUOTA &&
2743             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2744             oqctl->qc_cmd != Q_FINVALIDATE) {
2745                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2746                 RETURN(-EFAULT);
2747         }
2748
2749         /* for lov tgt */
2750         obd_getref(obd);
2751         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2752                 int err;
2753
2754                 tgt = lov->lov_tgts[i];
2755
2756                 if (!tgt)
2757                         continue;
2758
2759                 if (!tgt->ltd_active || tgt->ltd_reap) {
2760                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2761                             lov->lov_tgts[i]->ltd_activate) {
2762                                 rc = -EREMOTEIO;
2763                                 CERROR("ost %d is inactive\n", i);
2764                         } else {
2765                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2766                         }
2767                         continue;
2768                 }
2769
2770                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2771                 if (err) {
2772                         if (tgt->ltd_active && !rc)
2773                                 rc = err;
2774                         continue;
2775                 }
2776
2777                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2778                         curspace += oqctl->qc_dqblk.dqb_curspace;
2779                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2780                 }
2781         }
2782         obd_putref(obd);
2783
2784         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2785                 oqctl->qc_dqblk.dqb_curspace = curspace;
2786                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2787         }
2788         RETURN(rc);
2789 }
2790
2791 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2792                           struct obd_quotactl *oqctl)
2793 {
2794         struct lov_obd *lov = &obd->u.lov;
2795         int             i, rc = 0;
2796         ENTRY;
2797
2798         obd_getref(obd);
2799
2800         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2801                 if (!lov->lov_tgts[i])
2802                         continue;
2803
2804                 /* Skip quota check on the administratively disabled OSTs. */
2805                 if (!lov->lov_tgts[i]->ltd_activate) {
2806                         CWARN("lov idx %d was administratively disabled, "
2807                               "skip quotacheck on it.\n", i);
2808                         continue;
2809                 }
2810
2811                 if (!lov->lov_tgts[i]->ltd_active) {
2812                         CERROR("lov idx %d inactive\n", i);
2813                         rc = -EIO;
2814                         goto out;
2815                 }
2816         }
2817
2818         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2819                 int err;
2820
2821                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2822                         continue;
2823
2824                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2825                 if (err && !rc)
2826                         rc = err;
2827         }
2828
2829 out:
2830         obd_putref(obd);
2831
2832         RETURN(rc);
2833 }
2834
2835 struct obd_ops lov_obd_ops = {
2836         .o_owner               = THIS_MODULE,
2837         .o_setup               = lov_setup,
2838         .o_precleanup          = lov_precleanup,
2839         .o_cleanup             = lov_cleanup,
2840         //.o_process_config      = lov_process_config,
2841         .o_connect             = lov_connect,
2842         .o_disconnect          = lov_disconnect,
2843         .o_statfs              = lov_statfs,
2844         .o_statfs_async        = lov_statfs_async,
2845         .o_packmd              = lov_packmd,
2846         .o_unpackmd            = lov_unpackmd,
2847         .o_create              = lov_create,
2848         .o_destroy             = lov_destroy,
2849         .o_getattr             = lov_getattr,
2850         .o_getattr_async       = lov_getattr_async,
2851         .o_setattr             = lov_setattr,
2852         .o_setattr_async       = lov_setattr_async,
2853         .o_brw                 = lov_brw,
2854         .o_merge_lvb           = lov_merge_lvb,
2855         .o_adjust_kms          = lov_adjust_kms,
2856         .o_punch               = lov_punch,
2857         .o_sync                = lov_sync,
2858         .o_enqueue             = lov_enqueue,
2859         .o_change_cbdata       = lov_change_cbdata,
2860         .o_find_cbdata         = lov_find_cbdata,
2861         .o_cancel              = lov_cancel,
2862         .o_cancel_unused       = lov_cancel_unused,
2863         .o_iocontrol           = lov_iocontrol,
2864         .o_get_info            = lov_get_info,
2865         .o_set_info_async      = lov_set_info_async,
2866         .o_extent_calc         = lov_extent_calc,
2867         .o_llog_init           = lov_llog_init,
2868         .o_llog_finish         = lov_llog_finish,
2869         .o_notify              = lov_notify,
2870         .o_pool_new            = lov_pool_new,
2871         .o_pool_rem            = lov_pool_remove,
2872         .o_pool_add            = lov_pool_add,
2873         .o_pool_del            = lov_pool_del,
2874         .o_getref              = lov_getref,
2875         .o_putref              = lov_putref,
2876         .o_quotactl            = lov_quotactl,
2877         .o_quotacheck          = lov_quotacheck,
2878 };
2879
2880 cfs_mem_cache_t *lov_oinfo_slab;
2881
2882 extern struct lu_kmem_descr lov_caches[];
2883
2884 int __init lov_init(void)
2885 {
2886         struct lprocfs_static_vars lvars = { 0 };
2887         int rc, rc2;
2888         ENTRY;
2889
2890         /* print an address of _any_ initialized kernel symbol from this
2891          * module, to allow debugging with gdb that doesn't support data
2892          * symbols from modules.*/
2893         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2894
2895         rc = lu_kmem_init(lov_caches);
2896         if (rc)
2897                 return rc;
2898
2899         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2900                                               sizeof(struct lov_oinfo),
2901                                               0, CFS_SLAB_HWCACHE_ALIGN);
2902         if (lov_oinfo_slab == NULL) {
2903                 lu_kmem_fini(lov_caches);
2904                 return -ENOMEM;
2905         }
2906         lprocfs_lov_init_vars(&lvars);
2907
2908         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2909                                  LUSTRE_LOV_NAME, &lov_device_type);
2910
2911         if (rc) {
2912                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2913                 LASSERT(rc2 == 0);
2914                 lu_kmem_fini(lov_caches);
2915         }
2916
2917         RETURN(rc);
2918 }
2919
2920 #ifdef __KERNEL__
2921 static void /*__exit*/ lov_exit(void)
2922 {
2923         int rc;
2924
2925         class_unregister_type(LUSTRE_LOV_NAME);
2926         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2927         LASSERT(rc == 0);
2928
2929         lu_kmem_fini(lov_caches);
2930 }
2931
2932 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2933 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2934 MODULE_LICENSE("GPL");
2935
2936 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2937 #endif