Whamcloud - gitweb
LU-6173 llite: allocate and free client cache asynchronously
[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, 2014, 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 #include <libcfs/libcfs.h>
46
47 #include <obd_support.h>
48 #include <lustre_ioctl.h>
49 #include <lustre_lib.h>
50 #include <lustre_net.h>
51 #include <lustre/lustre_idl.h>
52 #include <lustre_dlm.h>
53 #include <lustre_mds.h>
54 #include <obd_class.h>
55 #include <lprocfs_status.h>
56 #include <lustre_param.h>
57 #include <cl_object.h>
58 #include <lustre/ll_fiemap.h>
59 #include <lustre_fid.h>
60
61 #include "lov_internal.h"
62
63 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
64    Any function that expects lov_tgts to remain stationary must take a ref. */
65 static void lov_getref(struct obd_device *obd)
66 {
67         struct lov_obd *lov = &obd->u.lov;
68
69         /* nobody gets through here until lov_putref is done */
70         mutex_lock(&lov->lov_lock);
71         atomic_inc(&lov->lov_refcount);
72         mutex_unlock(&lov->lov_lock);
73         return;
74 }
75
76 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
77
78 static void lov_putref(struct obd_device *obd)
79 {
80         struct lov_obd *lov = &obd->u.lov;
81
82         mutex_lock(&lov->lov_lock);
83         /* ok to dec to 0 more than once -- ltd_exp's will be null */
84         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
85                 struct list_head kill = LIST_HEAD_INIT(kill);
86                 struct lov_tgt_desc *tgt, *n;
87                 int i;
88
89                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
90                        lov->lov_death_row);
91                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
92                         tgt = lov->lov_tgts[i];
93
94                         if (!tgt || !tgt->ltd_reap)
95                                 continue;
96                         list_add(&tgt->ltd_kill, &kill);
97                         /* XXX - right now there is a dependency on ld_tgt_count
98                          * being the maximum tgt index for computing the
99                          * mds_max_easize. So we can't shrink it. */
100                         lov_ost_pool_remove(&lov->lov_packed, i);
101                         lov->lov_tgts[i] = NULL;
102                         lov->lov_death_row--;
103                 }
104                 mutex_unlock(&lov->lov_lock);
105
106                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
107                         list_del(&tgt->ltd_kill);
108                         /* Disconnect */
109                         __lov_del_obd(obd, tgt);
110                 }
111         } else {
112                 mutex_unlock(&lov->lov_lock);
113         }
114 }
115
116 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
117                               enum obd_notify_event ev);
118 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
119                       enum obd_notify_event ev, void *data);
120
121 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
122                     struct obd_connect_data *data)
123 {
124         struct lov_obd *lov = &obd->u.lov;
125         struct obd_uuid *tgt_uuid;
126         struct obd_device *tgt_obd;
127         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
128         struct obd_import *imp;
129         int rc;
130         ENTRY;
131
132         if (lov->lov_tgts[index] == NULL)
133                 RETURN(-EINVAL);
134
135         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
136         tgt_obd = lov->lov_tgts[index]->ltd_obd;
137
138         if (!tgt_obd->obd_set_up) {
139                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
140                 RETURN(-EINVAL);
141         }
142
143         /* override the sp_me from lov */
144         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
145
146         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
147                 data->ocd_index = index;
148
149         /*
150          * Divine LOV knows that OBDs under it are OSCs.
151          */
152         imp = tgt_obd->u.cli.cl_import;
153
154         if (activate) {
155                 tgt_obd->obd_no_recov = 0;
156                 /* FIXME this is probably supposed to be
157                    ptlrpc_set_import_active.  Horrible naming. */
158                 ptlrpc_activate_import(imp);
159         }
160
161         rc = obd_register_observer(tgt_obd, obd);
162         if (rc) {
163                 CERROR("Target %s register_observer error %d\n",
164                        obd_uuid2str(tgt_uuid), rc);
165                 RETURN(rc);
166         }
167
168
169         if (imp->imp_invalid) {
170                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
171                        "disabled\n", obd_uuid2str(tgt_uuid));
172                 RETURN(0);
173         }
174
175         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
176                          &lov_osc_uuid, data, NULL);
177         if (rc || !lov->lov_tgts[index]->ltd_exp) {
178                 CERROR("Target %s connect error %d\n",
179                        obd_uuid2str(tgt_uuid), rc);
180                 RETURN(-ENODEV);
181         }
182
183         lov->lov_tgts[index]->ltd_reap = 0;
184
185         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
186                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
187
188         if (lov->targets_proc_entry != NULL) {
189                 struct proc_dir_entry *osc_symlink;
190                 struct obd_device *osc_obd;
191
192                 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
193
194                 LASSERT(osc_obd != NULL);
195                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
196                 LASSERT(osc_obd->obd_type->typ_name != NULL);
197
198                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
199                                                   lov->targets_proc_entry,
200                                                   "../../../%s/%s",
201                                                   osc_obd->obd_type->typ_name,
202                                                   osc_obd->obd_name);
203                 if (osc_symlink == NULL) {
204                         CERROR("cannot register LOV target "
205                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
206                                obd->obd_type->typ_name, obd->obd_name,
207                                osc_obd->obd_name);
208                 }
209         }
210         RETURN(0);
211 }
212
213 static int lov_connect(const struct lu_env *env,
214                        struct obd_export **exp, struct obd_device *obd,
215                        struct obd_uuid *cluuid, struct obd_connect_data *data,
216                        void *localdata)
217 {
218         struct lov_obd *lov = &obd->u.lov;
219         struct lov_tgt_desc *tgt;
220         struct lustre_handle conn;
221         int i, rc;
222         ENTRY;
223
224         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
225
226         rc = class_connect(&conn, obd, cluuid);
227         if (rc)
228                 RETURN(rc);
229
230         *exp = class_conn2export(&conn);
231
232         /* Why should there ever be more than 1 connect? */
233         lov->lov_connects++;
234         LASSERT(lov->lov_connects == 1);
235
236         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
237         if (data)
238                 lov->lov_ocd = *data;
239
240         lov->targets_proc_entry = lprocfs_register("target_obds",
241                                                    obd->obd_proc_entry,
242                                                    NULL, NULL);
243         if (IS_ERR(lov->targets_proc_entry)) {
244                 CERROR("%s: cannot register "
245                        "/proc/fs/lustre/%s/%s/target_obds\n",
246                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
247                 lov->targets_proc_entry = NULL;
248         }
249
250         obd_getref(obd);
251         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
252                 tgt = lov->lov_tgts[i];
253                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
254                         continue;
255                 /* Flags will be lowest common denominator */
256                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
257                 if (rc) {
258                         CERROR("%s: lov connect tgt %d failed: %d\n",
259                                obd->obd_name, i, rc);
260                         continue;
261                 }
262                 /* connect to administrative disabled ost */
263                 if (!lov->lov_tgts[i]->ltd_exp)
264                         continue;
265
266                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
267                                 OBD_NOTIFY_CONNECT, (void *)&i);
268                 if (rc) {
269                         CERROR("%s error sending notify %d\n",
270                                obd->obd_name, rc);
271                 }
272         }
273         obd_putref(obd);
274
275         RETURN(0);
276 }
277
278 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
279 {
280         struct lov_obd *lov = &obd->u.lov;
281         struct obd_device *osc_obd;
282         int rc;
283         ENTRY;
284
285         osc_obd = class_exp2obd(tgt->ltd_exp);
286         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
287                obd->obd_name, osc_obd->obd_name);
288
289         if (tgt->ltd_active) {
290                 tgt->ltd_active = 0;
291                 lov->desc.ld_active_tgt_count--;
292                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
293         }
294
295         if (osc_obd) {
296                 /* Pass it on to our clients.
297                  * XXX This should be an argument to disconnect,
298                  * XXX not a back-door flag on the OBD.  Ah well.
299                  */
300                 osc_obd->obd_force = obd->obd_force;
301                 osc_obd->obd_fail = obd->obd_fail;
302                 osc_obd->obd_no_recov = obd->obd_no_recov;
303
304                 if (lov->targets_proc_entry != NULL)
305                         lprocfs_remove_proc_entry(osc_obd->obd_name,
306                                                   lov->targets_proc_entry);
307         }
308
309         obd_register_observer(osc_obd, NULL);
310
311         rc = obd_disconnect(tgt->ltd_exp);
312         if (rc) {
313                 CERROR("Target %s disconnect error %d\n",
314                        tgt->ltd_uuid.uuid, rc);
315                 rc = 0;
316         }
317
318         tgt->ltd_exp = NULL;
319         RETURN(0);
320 }
321
322 static int lov_disconnect(struct obd_export *exp)
323 {
324         struct obd_device *obd = class_exp2obd(exp);
325         struct lov_obd *lov = &obd->u.lov;
326         int i, rc;
327         ENTRY;
328
329         if (!lov->lov_tgts)
330                 goto out;
331
332         /* Only disconnect the underlying layers on the final disconnect. */
333         lov->lov_connects--;
334         if (lov->lov_connects != 0) {
335                 /* why should there be more than 1 connect? */
336                 CERROR("disconnect #%d\n", lov->lov_connects);
337                 goto out;
338         }
339
340         /* Let's hold another reference so lov_del_obd doesn't spin through
341            putref every time */
342         obd_getref(obd);
343
344         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
345                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
346                         /* Disconnection is the last we know about an obd */
347                         lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
348                 }
349         }
350         obd_putref(obd);
351
352         if (lov->targets_proc_entry != NULL)
353                 lprocfs_remove(&lov->targets_proc_entry);
354
355 out:
356         rc = class_disconnect(exp); /* bz 9811 */
357         RETURN(rc);
358 }
359
360 /* Error codes:
361  *
362  *  -EINVAL  : UUID can't be found in the LOV's target list
363  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
364  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
365  *  any >= 0 : is log target index
366  */
367 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
368                               enum obd_notify_event ev)
369 {
370         struct lov_obd *lov = &obd->u.lov;
371         struct lov_tgt_desc *tgt;
372         int index, activate, active;
373         ENTRY;
374
375         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
376                lov, uuid->uuid, ev);
377
378         obd_getref(obd);
379         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
380                 tgt = lov->lov_tgts[index];
381                 if (!tgt)
382                         continue;
383                 /*
384                  * LU-642, initially inactive OSC could miss the obd_connect,
385                  * we make up for it here.
386                  */
387                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
388                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
389                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
390
391                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
392                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
393                 }
394                 if (!tgt->ltd_exp)
395                         continue;
396
397                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
398                        index, obd_uuid2str(&tgt->ltd_uuid),
399                        tgt->ltd_exp->exp_handle.h_cookie);
400                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
401                         break;
402         }
403
404         if (index == lov->desc.ld_tgt_count)
405                 GOTO(out, index = -EINVAL);
406
407         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
408                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
409
410                 if (lov->lov_tgts[index]->ltd_activate == activate) {
411                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
412                                uuid->uuid, activate ? "" : "de");
413                 } else {
414                         lov->lov_tgts[index]->ltd_activate = activate;
415                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
416                                activate ? "" : "de", obd_uuid2str(uuid));
417                 }
418
419         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
420                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
421
422                 if (lov->lov_tgts[index]->ltd_active == active) {
423                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
424                                uuid->uuid, active ? "" : "in");
425                         GOTO(out, index);
426                 } else {
427                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
428                                obd_uuid2str(uuid), active ? "" : "in");
429                 }
430
431                 lov->lov_tgts[index]->ltd_active = active;
432                 if (active) {
433                         lov->desc.ld_active_tgt_count++;
434                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
435                 } else {
436                         lov->desc.ld_active_tgt_count--;
437                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
438                 }
439         } else {
440                 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
441                        ev, uuid->uuid);
442         }
443
444  out:
445         obd_putref(obd);
446         RETURN(index);
447 }
448
449 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
450                       enum obd_notify_event ev, void *data)
451 {
452         int rc = 0;
453         struct lov_obd *lov = &obd->u.lov;
454         ENTRY;
455
456         down_read(&lov->lov_notify_lock);
457         if (!lov->lov_connects) {
458                 up_read(&lov->lov_notify_lock);
459                 RETURN(rc);
460         }
461
462         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
463             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
464                 struct obd_uuid *uuid;
465
466                 LASSERT(watched);
467
468                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
469                         up_read(&lov->lov_notify_lock);
470                         CERROR("unexpected notification of %s %s!\n",
471                                watched->obd_type->typ_name,
472                                watched->obd_name);
473                         RETURN(-EINVAL);
474                 }
475                 uuid = &watched->u.cli.cl_target_uuid;
476
477                 /* Set OSC as active before notifying the observer, so the
478                  * observer can use the OSC normally.
479                  */
480                 rc = lov_set_osc_active(obd, uuid, ev);
481                 if (rc < 0) {
482                         up_read(&lov->lov_notify_lock);
483                         CERROR("event(%d) of %s failed: %d\n", ev,
484                                obd_uuid2str(uuid), rc);
485                         RETURN(rc);
486                 }
487                 /* active event should be pass lov target index as data */
488                 data = &rc;
489         }
490
491         /* Pass the notification up the chain. */
492         if (watched) {
493                 rc = obd_notify_observer(obd, watched, ev, data);
494         } else {
495                 /* NULL watched means all osc's in the lov (only for syncs) */
496                 /* sync event should be send lov idx as data */
497                 struct lov_obd *lov = &obd->u.lov;
498                 int i, is_sync;
499
500                 data = &i;
501                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
502                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
503
504                 obd_getref(obd);
505                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
506                         if (!lov->lov_tgts[i])
507                                 continue;
508
509                         /* don't send sync event if target not
510                          * connected/activated */
511                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
512                                 continue;
513
514                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
515                                                  ev, data);
516                         if (rc) {
517                                 CERROR("%s: notify %s of %s failed %d\n",
518                                        obd->obd_name,
519                                        obd->obd_observer->obd_name,
520                                        lov->lov_tgts[i]->ltd_obd->obd_name,
521                                        rc);
522                         }
523                 }
524                 obd_putref(obd);
525         }
526
527         up_read(&lov->lov_notify_lock);
528         RETURN(rc);
529 }
530
531 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
532                           __u32 index, int gen, int active)
533 {
534         struct lov_obd *lov = &obd->u.lov;
535         struct lov_tgt_desc *tgt;
536         struct obd_device *tgt_obd;
537         int rc;
538         ENTRY;
539
540         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
541                uuidp->uuid, index, gen, active);
542
543         if (gen <= 0) {
544                 CERROR("request to add OBD %s with invalid generation: %d\n",
545                        uuidp->uuid, gen);
546                 RETURN(-EINVAL);
547         }
548
549         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
550                                         &obd->obd_uuid);
551         if (tgt_obd == NULL)
552                 RETURN(-EINVAL);
553
554         mutex_lock(&lov->lov_lock);
555
556         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
557                 tgt = lov->lov_tgts[index];
558                 CERROR("UUID %s already assigned at LOV target index %d\n",
559                        obd_uuid2str(&tgt->ltd_uuid), index);
560                 mutex_unlock(&lov->lov_lock);
561                 RETURN(-EEXIST);
562         }
563
564         if (index >= lov->lov_tgt_size) {
565                 /* We need to reallocate the lov target array. */
566                 struct lov_tgt_desc **newtgts, **old = NULL;
567                 __u32 newsize, oldsize = 0;
568
569                 newsize = max(lov->lov_tgt_size, (__u32)2);
570                 while (newsize < index + 1)
571                         newsize = newsize << 1;
572                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
573                 if (newtgts == NULL) {
574                         mutex_unlock(&lov->lov_lock);
575                         RETURN(-ENOMEM);
576                 }
577
578                 if (lov->lov_tgt_size) {
579                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
580                                lov->lov_tgt_size);
581                         old = lov->lov_tgts;
582                         oldsize = lov->lov_tgt_size;
583                 }
584
585                 lov->lov_tgts = newtgts;
586                 lov->lov_tgt_size = newsize;
587                 smp_rmb();
588                 if (old)
589                         OBD_FREE(old, sizeof(*old) * oldsize);
590
591                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
592                        lov->lov_tgts, lov->lov_tgt_size);
593         }
594
595         OBD_ALLOC_PTR(tgt);
596         if (!tgt) {
597                 mutex_unlock(&lov->lov_lock);
598                 RETURN(-ENOMEM);
599         }
600
601         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
602         if (rc) {
603                 mutex_unlock(&lov->lov_lock);
604                 OBD_FREE_PTR(tgt);
605                 RETURN(rc);
606         }
607
608         tgt->ltd_uuid = *uuidp;
609         tgt->ltd_obd = tgt_obd;
610         /* XXX - add a sanity check on the generation number. */
611         tgt->ltd_gen = gen;
612         tgt->ltd_index = index;
613         tgt->ltd_activate = active;
614         lov->lov_tgts[index] = tgt;
615         if (index >= lov->desc.ld_tgt_count)
616                 lov->desc.ld_tgt_count = index + 1;
617
618         mutex_unlock(&lov->lov_lock);
619
620         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
621                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
622
623         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
624
625         if (lov->lov_connects == 0) {
626                 /* lov_connect hasn't been called yet. We'll do the
627                    lov_connect_obd on this target when that fn first runs,
628                    because we don't know the connect flags yet. */
629                 RETURN(0);
630         }
631
632         obd_getref(obd);
633
634         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
635         if (rc)
636                 GOTO(out, rc);
637
638         /* connect to administrative disabled ost */
639         if (!tgt->ltd_exp)
640                 GOTO(out, rc = 0);
641
642         if (lov->lov_cache != NULL) {
643                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
644                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
645                                 sizeof(struct cl_client_cache), lov->lov_cache,
646                                 NULL);
647                 if (rc < 0)
648                         GOTO(out, rc);
649         }
650
651         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
652                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
653                         (void *)&index);
654
655 out:
656         if (rc) {
657                 CERROR("add failed (%d), deleting %s\n", rc,
658                        obd_uuid2str(&tgt->ltd_uuid));
659                 lov_del_target(obd, index, NULL, 0);
660         }
661         obd_putref(obd);
662         RETURN(rc);
663 }
664
665 /* Schedule a target for deletion */
666 int lov_del_target(struct obd_device *obd, __u32 index,
667                    struct obd_uuid *uuidp, int gen)
668 {
669         struct lov_obd *lov = &obd->u.lov;
670         int count = lov->desc.ld_tgt_count;
671         int rc = 0;
672         ENTRY;
673
674         if (index >= count) {
675                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
676                        index, count);
677                 RETURN(-EINVAL);
678         }
679
680         /* to make sure there's no ongoing lov_notify() now */
681         down_write(&lov->lov_notify_lock);
682         obd_getref(obd);
683
684         if (!lov->lov_tgts[index]) {
685                 CERROR("LOV target at index %d is not setup.\n", index);
686                 GOTO(out, rc = -EINVAL);
687         }
688
689         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
690                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
691                        lov_uuid2str(lov, index), index,
692                        obd_uuid2str(uuidp));
693                 GOTO(out, rc = -EINVAL);
694         }
695
696         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
697                lov_uuid2str(lov, index), index,
698                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
699                lov->lov_tgts[index]->ltd_active);
700
701         lov->lov_tgts[index]->ltd_reap = 1;
702         lov->lov_death_row++;
703         /* we really delete it from obd_putref */
704 out:
705         obd_putref(obd);
706         up_write(&lov->lov_notify_lock);
707
708         RETURN(rc);
709 }
710
711 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
712 {
713         struct obd_device *osc_obd;
714
715         LASSERT(tgt);
716         LASSERT(tgt->ltd_reap);
717
718         osc_obd = class_exp2obd(tgt->ltd_exp);
719
720         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
721                tgt->ltd_uuid.uuid,
722                osc_obd ? osc_obd->obd_name : "<no obd>");
723
724         if (tgt->ltd_exp)
725                 lov_disconnect_obd(obd, tgt);
726
727         OBD_FREE_PTR(tgt);
728
729         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
730            do it ourselves. And we can't do it from lov_cleanup,
731            because we just lost our only reference to it. */
732         if (osc_obd)
733                 class_manual_cleanup(osc_obd);
734 }
735
736 void lov_fix_desc_stripe_size(__u64 *val)
737 {
738         if (*val < LOV_MIN_STRIPE_SIZE) {
739                 if (*val != 0)
740                         LCONSOLE_INFO("Increasing default stripe size to "
741                                       "minimum %u\n",
742                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
743                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
744         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
745                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
746                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
747                               "multiple of %u)\n",
748                               *val, LOV_MIN_STRIPE_SIZE);
749         }
750 }
751
752 void lov_fix_desc_stripe_count(__u32 *val)
753 {
754         if (*val == 0)
755                 *val = 1;
756 }
757
758 void lov_fix_desc_pattern(__u32 *val)
759 {
760         /* from lov_setstripe */
761         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
762                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
763                 *val = 0;
764         }
765 }
766
767 void lov_fix_desc_qos_maxage(__u32 *val)
768 {
769         if (*val == 0)
770                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
771 }
772
773 void lov_fix_desc(struct lov_desc *desc)
774 {
775         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
776         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
777         lov_fix_desc_pattern(&desc->ld_pattern);
778         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
779 }
780
781 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
782 {
783         struct lov_desc *desc;
784         struct lov_obd *lov = &obd->u.lov;
785 #ifdef CONFIG_PROC_FS
786         struct obd_type *type;
787 #endif
788         int rc;
789         ENTRY;
790
791         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
792                 CERROR("LOV setup requires a descriptor\n");
793                 RETURN(-EINVAL);
794         }
795
796         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
797
798         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
799                 CERROR("descriptor size wrong: %d > %d\n",
800                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
801                 RETURN(-EINVAL);
802         }
803
804         if (desc->ld_magic != LOV_DESC_MAGIC) {
805                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
806                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
807                                    obd->obd_name, desc);
808                             lustre_swab_lov_desc(desc);
809                 } else {
810                         CERROR("%s: Bad lov desc magic: %#x\n",
811                                obd->obd_name, desc->ld_magic);
812                         RETURN(-EINVAL);
813                 }
814         }
815
816         lov_fix_desc(desc);
817
818         desc->ld_active_tgt_count = 0;
819         lov->desc = *desc;
820         lov->lov_tgt_size = 0;
821
822         mutex_init(&lov->lov_lock);
823         atomic_set(&lov->lov_refcount, 0);
824         lov->lov_sp_me = LUSTRE_SP_CLI;
825
826         init_rwsem(&lov->lov_notify_lock);
827
828         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
829                                                    HASH_POOLS_MAX_BITS,
830                                                    HASH_POOLS_BKT_BITS, 0,
831                                                    CFS_HASH_MIN_THETA,
832                                                    CFS_HASH_MAX_THETA,
833                                                    &pool_hash_operations,
834                                                    CFS_HASH_DEFAULT);
835         INIT_LIST_HEAD(&lov->lov_pool_list);
836         lov->lov_pool_count = 0;
837         rc = lov_ost_pool_init(&lov->lov_packed, 0);
838         if (rc)
839                 GOTO(out, rc);
840
841 #ifdef CONFIG_PROC_FS
842         obd->obd_vars = lprocfs_lov_obd_vars;
843         /* If this is true then both client (lov) and server
844          * (lod) are on the same node. The lod layer if loaded
845          * first will register the lov proc directory. In that
846          * case obd->obd_type->typ_procroot will be not set.
847          * Instead we use type->typ_procsym as the parent. */
848         type = class_search_type(LUSTRE_LOD_NAME);
849         if (type != NULL && type->typ_procsym != NULL) {
850                 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
851                                                        type->typ_procsym,
852                                                        obd->obd_vars, obd);
853                 if (IS_ERR(obd->obd_proc_entry)) {
854                         rc = PTR_ERR(obd->obd_proc_entry);
855                         CERROR("error %d setting up lprocfs for %s\n", rc,
856                                obd->obd_name);
857                         obd->obd_proc_entry = NULL;
858                 }
859         } else {
860                 rc = lprocfs_obd_setup(obd);
861         }
862
863         if (rc == 0) {
864                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
865                                         0444, &lov_proc_target_fops, obd);
866                 if (rc)
867                         CWARN("Error adding the target_obd file\n");
868
869                 lov->lov_pool_proc_entry = lprocfs_register("pools",
870                                                             obd->obd_proc_entry,
871                                                             NULL, NULL);
872                 if (IS_ERR(lov->lov_pool_proc_entry)) {
873                         rc = PTR_ERR(lov->lov_pool_proc_entry);
874                         CERROR("error %d setting up lprocfs for pools\n", rc);
875                         lov->lov_pool_proc_entry = NULL;
876                 }
877         }
878 #endif
879         RETURN(0);
880
881 out:
882         return rc;
883 }
884
885 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
886 {
887         int rc = 0;
888         struct lov_obd *lov = &obd->u.lov;
889
890         ENTRY;
891
892         switch (stage) {
893         case OBD_CLEANUP_EARLY: {
894                 int i;
895                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
896                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
897                                 continue;
898                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
899                                        OBD_CLEANUP_EARLY);
900                 }
901                 break;
902         }
903         default:
904                 break;
905         }
906
907         RETURN(rc);
908 }
909
910 static int lov_cleanup(struct obd_device *obd)
911 {
912         struct lov_obd *lov = &obd->u.lov;
913         struct list_head *pos, *tmp;
914         struct pool_desc *pool;
915         ENTRY;
916
917         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
918                 pool = list_entry(pos, struct pool_desc, pool_list);
919                 /* free pool structs */
920                 CDEBUG(D_INFO, "delete pool %p\n", pool);
921                 /* In the function below, .hs_keycmp resolves to
922                  * pool_hashkey_keycmp() */
923                 /* coverity[overrun-buffer-val] */
924                 lov_pool_del(obd, pool->pool_name);
925         }
926         cfs_hash_putref(lov->lov_pools_hash_body);
927         lov_ost_pool_free(&lov->lov_packed);
928
929         lprocfs_obd_cleanup(obd);
930         if (lov->lov_tgts) {
931                 int i;
932                 obd_getref(obd);
933                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
934                         if (!lov->lov_tgts[i])
935                                 continue;
936
937                         /* Inactive targets may never have connected */
938                         if (lov->lov_tgts[i]->ltd_active ||
939                             atomic_read(&lov->lov_refcount))
940                                 /* We should never get here - these
941                                  * should have been removed in the
942                                  * disconnect. */
943                                 CERROR("%s: lov tgt %d not cleaned! "
944                                        "deathrow=%d, lovrc=%d\n",
945                                        obd->obd_name, i, lov->lov_death_row,
946                                        atomic_read(&lov->lov_refcount));
947                         lov_del_target(obd, i, NULL, 0);
948                 }
949                 obd_putref(obd);
950                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
951                          lov->lov_tgt_size);
952                 lov->lov_tgt_size = 0;
953         }
954
955         if (lov->lov_cache != NULL) {
956                 cl_cache_decref(lov->lov_cache);
957                 lov->lov_cache = NULL;
958         }
959
960         RETURN(0);
961 }
962
963 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
964                             __u32 *indexp, int *genp)
965 {
966         struct obd_uuid obd_uuid;
967         int cmd;
968         int rc = 0;
969         ENTRY;
970
971         switch(cmd = lcfg->lcfg_command) {
972         case LCFG_LOV_ADD_OBD:
973         case LCFG_LOV_ADD_INA:
974         case LCFG_LOV_DEL_OBD: {
975                 __u32 index;
976                 int gen;
977                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
978                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
979                         GOTO(out, rc = -EINVAL);
980
981                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
982
983                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", indexp) != 1)
984                         GOTO(out, rc = -EINVAL);
985                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
986                         GOTO(out, rc = -EINVAL);
987                 index = *indexp;
988                 gen = *genp;
989                 if (cmd == LCFG_LOV_ADD_OBD)
990                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
991                 else if (cmd == LCFG_LOV_ADD_INA)
992                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
993                 else
994                         rc = lov_del_target(obd, index, &obd_uuid, gen);
995                 GOTO(out, rc);
996         }
997         case LCFG_PARAM: {
998                 struct lov_desc *desc = &(obd->u.lov.desc);
999
1000                 if (!desc)
1001                         GOTO(out, rc = -EINVAL);
1002
1003                 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
1004                                               lcfg, obd);
1005                 if (rc > 0)
1006                         rc = 0;
1007                 GOTO(out, rc);
1008         }
1009         case LCFG_POOL_NEW:
1010         case LCFG_POOL_ADD:
1011         case LCFG_POOL_DEL:
1012         case LCFG_POOL_REM:
1013                 GOTO(out, rc);
1014
1015         default: {
1016                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1017                 GOTO(out, rc = -EINVAL);
1018
1019         }
1020         }
1021 out:
1022         RETURN(rc);
1023 }
1024
1025 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1026 do {                                                                            \
1027         LASSERT((lsmp) != NULL);                                                \
1028         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1029                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1030                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1031 } while (0)
1032
1033 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1034                                  void *data, int rc)
1035 {
1036         struct lov_request_set *lovset = (struct lov_request_set *)data;
1037         int err;
1038         ENTRY;
1039
1040         /* don't do attribute merge if this aysnc op failed */
1041         if (rc)
1042                 atomic_set(&lovset->set_completes, 0);
1043         err = lov_fini_getattr_set(lovset);
1044         RETURN(rc ? rc : err);
1045 }
1046
1047 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1048                               struct ptlrpc_request_set *rqset)
1049 {
1050         struct lov_request_set *lovset;
1051         struct lov_obd *lov;
1052         struct list_head *pos;
1053         struct lov_request *req;
1054         int rc = 0, err;
1055         ENTRY;
1056
1057         LASSERT(oinfo);
1058         ASSERT_LSM_MAGIC(oinfo->oi_md);
1059
1060         if (!exp || !exp->exp_obd)
1061                 RETURN(-ENODEV);
1062
1063         lov = &exp->exp_obd->u.lov;
1064
1065         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1066         if (rc)
1067                 RETURN(rc);
1068
1069         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1070                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1071                oinfo->oi_md->lsm_stripe_size);
1072
1073         list_for_each(pos, &lovset->set_list) {
1074                 req = list_entry(pos, struct lov_request, rq_link);
1075
1076                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1077                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1078                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1079                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1080                                        &req->rq_oi, rqset);
1081                 if (rc) {
1082                         CERROR("%s: getattr objid "DOSTID" subobj"
1083                                DOSTID" on OST idx %d: rc = %d\n",
1084                                exp->exp_obd->obd_name,
1085                                POSTID(&oinfo->oi_oa->o_oi),
1086                                POSTID(&req->rq_oi.oi_oa->o_oi),
1087                                req->rq_idx, rc);
1088                         GOTO(out, rc);
1089                 }
1090         }
1091
1092         if (!list_empty(&rqset->set_requests)) {
1093                 LASSERT(rc == 0);
1094                 LASSERT (rqset->set_interpret == NULL);
1095                 rqset->set_interpret = lov_getattr_interpret;
1096                 rqset->set_arg = (void *)lovset;
1097                 RETURN(rc);
1098         }
1099 out:
1100         if (rc)
1101                 atomic_set(&lovset->set_completes, 0);
1102         err = lov_fini_getattr_set(lovset);
1103         RETURN(rc ? rc : err);
1104 }
1105
1106 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1107                                  void *data, int rc)
1108 {
1109         struct lov_request_set *lovset = (struct lov_request_set *)data;
1110         int err;
1111         ENTRY;
1112
1113         if (rc)
1114                 atomic_set(&lovset->set_completes, 0);
1115         err = lov_fini_setattr_set(lovset);
1116         RETURN(rc ? rc : err);
1117 }
1118
1119 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1120    needed. Otherwise, a client is waiting for responses. */
1121 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1122                              struct obd_trans_info *oti,
1123                              struct ptlrpc_request_set *rqset)
1124 {
1125         struct lov_request_set *set;
1126         struct lov_request *req;
1127         struct list_head *pos;
1128         struct lov_obd *lov;
1129         int rc = 0;
1130         ENTRY;
1131
1132         LASSERT(oinfo);
1133         ASSERT_LSM_MAGIC(oinfo->oi_md);
1134         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1135                 LASSERT(oti);
1136                 LASSERT(oti->oti_logcookies);
1137         }
1138
1139         if (!exp || !exp->exp_obd)
1140                 RETURN(-ENODEV);
1141
1142         lov = &exp->exp_obd->u.lov;
1143         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1144         if (rc)
1145                 RETURN(rc);
1146
1147         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1148                POSTID(&oinfo->oi_md->lsm_oi),
1149                oinfo->oi_md->lsm_stripe_count,
1150                oinfo->oi_md->lsm_stripe_size);
1151
1152         list_for_each(pos, &set->set_list) {
1153                 req = list_entry(pos, struct lov_request, rq_link);
1154
1155                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1156                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1157
1158                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1159                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1160                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1161
1162                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1163                                        &req->rq_oi, oti, rqset);
1164                 if (rc) {
1165                         CERROR("error: setattr objid "DOSTID" subobj"
1166                                DOSTID" on OST idx %d: rc = %d\n",
1167                                POSTID(&set->set_oi->oi_oa->o_oi),
1168                                POSTID(&req->rq_oi.oi_oa->o_oi),
1169                                req->rq_idx, rc);
1170                         break;
1171                 }
1172         }
1173
1174         /* If we are not waiting for responses on async requests, return. */
1175         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1176                 int err;
1177                 if (rc)
1178                         atomic_set(&set->set_completes, 0);
1179                 err = lov_fini_setattr_set(set);
1180                 RETURN(rc ? rc : err);
1181         }
1182
1183         LASSERT(rqset->set_interpret == NULL);
1184         rqset->set_interpret = lov_setattr_interpret;
1185         rqset->set_arg = (void *)set;
1186
1187         RETURN(0);
1188 }
1189
1190 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1191 {
1192         struct lov_request_set *lovset = (struct lov_request_set *)data;
1193         int err;
1194         ENTRY;
1195
1196         if (rc)
1197                 atomic_set(&lovset->set_completes, 0);
1198
1199         err = lov_fini_statfs_set(lovset);
1200         RETURN(rc ? rc : err);
1201 }
1202
1203 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1204                             __u64 max_age, struct ptlrpc_request_set *rqset)
1205 {
1206         struct obd_device      *obd = class_exp2obd(exp);
1207         struct lov_request_set *set;
1208         struct lov_request *req;
1209         struct list_head *pos;
1210         struct lov_obd *lov;
1211         int rc = 0;
1212         ENTRY;
1213
1214         LASSERT(oinfo != NULL);
1215         LASSERT(oinfo->oi_osfs != NULL);
1216
1217         lov = &obd->u.lov;
1218         rc = lov_prep_statfs_set(obd, oinfo, &set);
1219         if (rc)
1220                 RETURN(rc);
1221
1222         list_for_each(pos, &set->set_list) {
1223                 req = list_entry(pos, struct lov_request, rq_link);
1224                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1225                                       &req->rq_oi, max_age, rqset);
1226                 if (rc)
1227                         break;
1228         }
1229
1230         if (rc || list_empty(&rqset->set_requests)) {
1231                 int err;
1232                 if (rc)
1233                         atomic_set(&set->set_completes, 0);
1234                 err = lov_fini_statfs_set(set);
1235                 RETURN(rc ? rc : err);
1236         }
1237
1238         LASSERT(rqset->set_interpret == NULL);
1239         rqset->set_interpret = lov_statfs_interpret;
1240         rqset->set_arg = (void *)set;
1241         RETURN(0);
1242 }
1243
1244 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1245                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1246 {
1247         struct ptlrpc_request_set *set = NULL;
1248         struct obd_info oinfo = { { { 0 } } };
1249         int rc = 0;
1250         ENTRY;
1251
1252
1253         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1254          * statfs requests */
1255         set = ptlrpc_prep_set();
1256         if (set == NULL)
1257                 RETURN(-ENOMEM);
1258
1259         oinfo.oi_osfs = osfs;
1260         oinfo.oi_flags = flags;
1261         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1262         if (rc == 0)
1263                 rc = ptlrpc_set_wait(set);
1264         ptlrpc_set_destroy(set);
1265
1266         RETURN(rc);
1267 }
1268
1269 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1270                          void *karg, void __user *uarg)
1271 {
1272         struct obd_device *obddev = class_exp2obd(exp);
1273         struct lov_obd *lov = &obddev->u.lov;
1274         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1275         struct obd_uuid *uuidp;
1276         ENTRY;
1277
1278         switch (cmd) {
1279         case IOC_OBD_STATFS: {
1280                 struct obd_ioctl_data *data = karg;
1281                 struct obd_device *osc_obd;
1282                 struct obd_statfs stat_buf = {0};
1283                 __u32 index;
1284                 __u32 flags;
1285
1286                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1287                 if ((index >= count))
1288                         RETURN(-ENODEV);
1289
1290                 if (!lov->lov_tgts[index])
1291                         /* Try again with the next index */
1292                         RETURN(-EAGAIN);
1293                 if (!lov->lov_tgts[index]->ltd_active)
1294                         RETURN(-ENODATA);
1295
1296                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1297                 if (!osc_obd)
1298                         RETURN(-EINVAL);
1299
1300                 /* copy UUID */
1301                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1302                                  min((int)data->ioc_plen2,
1303                                      (int)sizeof(struct obd_uuid))))
1304                         RETURN(-EFAULT);
1305
1306                 flags = uarg ? *(__u32 __user *)uarg : 0;
1307                 /* got statfs data */
1308                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1309                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1310                                 flags);
1311                 if (rc)
1312                         RETURN(rc);
1313                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1314                                      min((int) data->ioc_plen1,
1315                                          (int) sizeof(stat_buf))))
1316                         RETURN(-EFAULT);
1317                 break;
1318         }
1319         case OBD_IOC_LOV_GET_CONFIG: {
1320                 struct obd_ioctl_data *data;
1321                 struct lov_desc *desc;
1322                 char *buf = NULL;
1323                 __u32 *genp;
1324
1325                 len = 0;
1326                 if (obd_ioctl_getdata(&buf, &len, uarg))
1327                         RETURN(-EINVAL);
1328
1329                 data = (struct obd_ioctl_data *)buf;
1330
1331                 if (sizeof(*desc) > data->ioc_inllen1) {
1332                         obd_ioctl_freedata(buf, len);
1333                         RETURN(-EINVAL);
1334                 }
1335
1336                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1337                         obd_ioctl_freedata(buf, len);
1338                         RETURN(-EINVAL);
1339                 }
1340
1341                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1342                         obd_ioctl_freedata(buf, len);
1343                         RETURN(-EINVAL);
1344                 }
1345
1346                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1347                 memcpy(desc, &(lov->desc), sizeof(*desc));
1348
1349                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1350                 genp = (__u32 *)data->ioc_inlbuf3;
1351                 /* the uuid will be empty for deleted OSTs */
1352                 for (i = 0; i < count; i++, uuidp++, genp++) {
1353                         if (!lov->lov_tgts[i])
1354                                 continue;
1355                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1356                         *genp = lov->lov_tgts[i]->ltd_gen;
1357                 }
1358
1359                 if (copy_to_user(uarg, buf, len))
1360                         rc = -EFAULT;
1361                 obd_ioctl_freedata(buf, len);
1362                 break;
1363         }
1364         case OBD_IOC_QUOTACTL: {
1365                 struct if_quotactl *qctl = karg;
1366                 struct lov_tgt_desc *tgt = NULL;
1367                 struct obd_quotactl *oqctl;
1368
1369                 if (qctl->qc_valid == QC_OSTIDX) {
1370                         if (count <= qctl->qc_idx)
1371                                 RETURN(-EINVAL);
1372
1373                         tgt = lov->lov_tgts[qctl->qc_idx];
1374                         if (!tgt || !tgt->ltd_exp)
1375                                 RETURN(-EINVAL);
1376                 } else if (qctl->qc_valid == QC_UUID) {
1377                         for (i = 0; i < count; i++) {
1378                                 tgt = lov->lov_tgts[i];
1379                                 if (!tgt ||
1380                                     !obd_uuid_equals(&tgt->ltd_uuid,
1381                                                      &qctl->obd_uuid))
1382                                         continue;
1383
1384                                 if (tgt->ltd_exp == NULL)
1385                                         RETURN(-EINVAL);
1386
1387                                 break;
1388                         }
1389                 } else {
1390                         RETURN(-EINVAL);
1391                 }
1392
1393                 if (i >= count)
1394                         RETURN(-EAGAIN);
1395
1396                 LASSERT(tgt && tgt->ltd_exp);
1397                 OBD_ALLOC_PTR(oqctl);
1398                 if (!oqctl)
1399                         RETURN(-ENOMEM);
1400
1401                 QCTL_COPY(oqctl, qctl);
1402                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1403                 if (rc == 0) {
1404                         QCTL_COPY(qctl, oqctl);
1405                         qctl->qc_valid = QC_OSTIDX;
1406                         qctl->obd_uuid = tgt->ltd_uuid;
1407                 }
1408                 OBD_FREE_PTR(oqctl);
1409                 break;
1410         }
1411         default: {
1412                 int set = 0;
1413
1414                 if (count == 0)
1415                         RETURN(-ENOTTY);
1416
1417                 for (i = 0; i < count; i++) {
1418                         int err;
1419                         struct obd_device *osc_obd;
1420
1421                         /* OST was disconnected */
1422                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1423                                 continue;
1424
1425                         /* ll_umount_begin() sets force flag but for lov, not
1426                          * osc. Let's pass it through */
1427                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1428                         osc_obd->obd_force = obddev->obd_force;
1429                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1430                                             len, karg, uarg);
1431                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1432                                 RETURN(err);
1433                         } else if (err) {
1434                                 if (lov->lov_tgts[i]->ltd_active) {
1435                                         CDEBUG(err == -ENOTTY ?
1436                                                D_IOCTL : D_WARNING,
1437                                                "iocontrol OSC %s on OST "
1438                                                "idx %d cmd %x: err = %d\n",
1439                                                lov_uuid2str(lov, i),
1440                                                i, cmd, err);
1441                                         if (!rc)
1442                                                 rc = err;
1443                                 }
1444                         } else {
1445                                 set = 1;
1446                         }
1447                 }
1448                 if (!set && !rc)
1449                         rc = -EIO;
1450         }
1451         }
1452
1453         RETURN(rc);
1454 }
1455
1456 #define FIEMAP_BUFFER_SIZE 4096
1457
1458 /**
1459  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1460  * call. The local end offset and the device are sent in the first
1461  * fm_extent. This function calculates the stripe number from the index.
1462  * This function returns a stripe_no on which mapping is to be restarted.
1463  *
1464  * This function returns fm_end_offset which is the in-OST offset at which
1465  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1466  * will re-calculate proper offset in next stripe.
1467  * Note that the first extent is passed to lov_get_info via the value field.
1468  *
1469  * \param fiemap fiemap request header
1470  * \param lsm striping information for the file
1471  * \param fm_start logical start of mapping
1472  * \param fm_end logical end of mapping
1473  * \param start_stripe starting stripe will be returned in this
1474  */
1475 static loff_t fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1476                                      struct lov_stripe_md *lsm,
1477                                      loff_t fm_start,
1478                                      loff_t fm_end, int *start_stripe)
1479 {
1480         loff_t local_end = fiemap->fm_extents[0].fe_logical;
1481         loff_t lun_start;
1482         loff_t lun_end;
1483         loff_t fm_end_offset;
1484         int stripe_no = -1;
1485         int i;
1486
1487         if (fiemap->fm_extent_count == 0 ||
1488             fiemap->fm_extents[0].fe_logical == 0)
1489                 return 0;
1490
1491         /* Find out stripe_no from ost_index saved in the fe_device */
1492         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1493                 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1494
1495                 if (lov_oinfo_is_dummy(oinfo))
1496                         continue;
1497
1498                 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1499                         stripe_no = i;
1500                         break;
1501                 }
1502         }
1503
1504         if (stripe_no == -1)
1505                 return -EINVAL;
1506
1507         /* If we have finished mapping on previous device, shift logical
1508          * offset to start of next device */
1509         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1510                                    &lun_start, &lun_end)) != 0 &&
1511                                    local_end < lun_end) {
1512                 fm_end_offset = local_end;
1513                 *start_stripe = stripe_no;
1514         } else {
1515                 /* This is a special value to indicate that caller should
1516                  * calculate offset in next stripe. */
1517                 fm_end_offset = 0;
1518                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1519         }
1520
1521         return fm_end_offset;
1522 }
1523
1524 /**
1525  * We calculate on which OST the mapping will end. If the length of mapping
1526  * is greater than (stripe_size * stripe_count) then the last_stripe will
1527  * will be one just before start_stripe. Else we check if the mapping
1528  * intersects each OST and find last_stripe.
1529  * This function returns the last_stripe and also sets the stripe_count
1530  * over which the mapping is spread
1531  *
1532  * \param lsm striping information for the file
1533  * \param fm_start logical start of mapping
1534  * \param fm_end logical end of mapping
1535  * \param start_stripe starting stripe of the mapping
1536  * \param stripe_count the number of stripes across which to map is returned
1537  *
1538  * \retval last_stripe return the last stripe of the mapping
1539  */
1540 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm,
1541                                    loff_t fm_start, loff_t fm_end,
1542                                    int start_stripe, int *stripe_count)
1543 {
1544         int last_stripe;
1545         loff_t obd_start;
1546         loff_t obd_end;
1547         int i, j;
1548
1549         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1550                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1551                                                               start_stripe - 1);
1552                 *stripe_count = lsm->lsm_stripe_count;
1553         } else {
1554                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1555                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
1556                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1557                                                    &obd_start, &obd_end)) == 0)
1558                                 break;
1559                 }
1560                 *stripe_count = j;
1561                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1562         }
1563
1564         return last_stripe;
1565 }
1566
1567 /**
1568  * Set fe_device and copy extents from local buffer into main return buffer.
1569  *
1570  * \param fiemap fiemap request header
1571  * \param lcl_fm_ext array of local fiemap extents to be copied
1572  * \param ost_index OST index to be written into the fm_device field for each
1573                     extent
1574  * \param ext_count number of extents to be copied
1575  * \param current_extent where to start copying in main extent array
1576  */
1577 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1578                                          struct ll_fiemap_extent *lcl_fm_ext,
1579                                          int ost_index, unsigned int ext_count,
1580                                          int current_extent)
1581 {
1582         char *to;
1583         int ext;
1584
1585         for (ext = 0; ext < ext_count; ext++) {
1586                 lcl_fm_ext[ext].fe_device = ost_index;
1587                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1588         }
1589
1590         /* Copy fm_extent's from fm_local to return buffer */
1591         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1592         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1593 }
1594
1595 /**
1596  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1597  * This also handles the restarting of FIEMAP calls in case mapping overflows
1598  * the available number of extents in single call.
1599  */
1600 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1601                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1602 {
1603         struct ll_fiemap_info_key *fm_key = key;
1604         struct ll_user_fiemap *fiemap = val;
1605         struct ll_user_fiemap *fm_local = NULL;
1606         struct ll_fiemap_extent *lcl_fm_ext;
1607         int count_local;
1608         unsigned int get_num_extents = 0;
1609         int ost_index = 0, actual_start_stripe, start_stripe;
1610         loff_t fm_start;
1611         loff_t fm_end;
1612         loff_t fm_length;
1613         loff_t fm_end_offset;
1614         u64 curr_loc;
1615         int current_extent = 0, rc = 0, i;
1616         /* Whether have we collected enough extents */
1617         bool enough = false;
1618         int ost_eof = 0; /* EOF for object */
1619         int ost_done = 0; /* done with required mapping for this OST? */
1620         int last_stripe;
1621         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1622         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1623
1624         if (!lsm_has_objects(lsm)) {
1625                 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1626                     fm_key->oa.o_size)) {
1627                         /* released file, return a minimal FIEMAP if
1628                          * request fits in file-size.
1629                          */
1630                         fiemap->fm_mapped_extents = 1;
1631                         fiemap->fm_extents[0].fe_logical =
1632                                                 fm_key->fiemap.fm_start;
1633                         if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1634                             fm_key->oa.o_size)
1635                                 fiemap->fm_extents[0].fe_length =
1636                                                 fm_key->fiemap.fm_length;
1637                         else
1638                                 fiemap->fm_extents[0].fe_length =
1639                                                 fm_key->oa.o_size -
1640                                                 fm_key->fiemap.fm_start;
1641                         fiemap->fm_extents[0].fe_flags |=
1642                                                 (FIEMAP_EXTENT_UNKNOWN |
1643                                                  FIEMAP_EXTENT_LAST);
1644                 }
1645                 GOTO(out, rc = 0);
1646         }
1647
1648         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1649                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1650
1651         OBD_ALLOC_LARGE(fm_local, buffer_size);
1652         if (fm_local == NULL)
1653                 GOTO(out, rc = -ENOMEM);
1654         lcl_fm_ext = &fm_local->fm_extents[0];
1655
1656         count_local = fiemap_size_to_count(buffer_size);
1657
1658         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1659         fm_start = fiemap->fm_start;
1660         fm_length = fiemap->fm_length;
1661         /* Calculate start stripe, last stripe and length of mapping */
1662         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1663         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1664                                                 fm_start + fm_length - 1);
1665         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1666         if (fm_end > fm_key->oa.o_size)
1667                 fm_end = fm_key->oa.o_size;
1668
1669         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1670                                             actual_start_stripe, &stripe_count);
1671
1672         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1673                                                   fm_end, &start_stripe);
1674         if (fm_end_offset == -EINVAL)
1675                 GOTO(out, rc = -EINVAL);
1676
1677         if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1678                 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1679         if (fiemap->fm_extent_count == 0) {
1680                 get_num_extents = 1;
1681                 count_local = 0;
1682         }
1683         /* Check each stripe */
1684         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1685              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1686                 loff_t req_fm_len; /* Stores length of required mapping */
1687                 loff_t len_mapped_single_call;
1688                 loff_t lun_start;
1689                 loff_t lun_end;
1690                 loff_t obd_object_end;
1691                 unsigned int ext_count;
1692
1693                 cur_stripe_wrap = cur_stripe;
1694
1695                 /* Find out range of mapping on this stripe */
1696                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1697                                            &lun_start, &obd_object_end)) == 0)
1698                         continue;
1699
1700                 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe]))
1701                         GOTO(out, rc = -EIO);
1702
1703                 /* If this is a continuation FIEMAP call and we are on
1704                  * starting stripe then lun_start needs to be set to
1705                  * fm_end_offset */
1706                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1707                         lun_start = fm_end_offset;
1708
1709                 if (fm_length != ~0ULL) {
1710                         /* Handle fm_start + fm_length overflow */
1711                         if (fm_start + fm_length < fm_start)
1712                                 fm_length = ~0ULL - fm_start;
1713                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1714                                                      cur_stripe);
1715                 } else {
1716                         lun_end = ~0ULL;
1717                 }
1718
1719                 if (lun_start == lun_end)
1720                         continue;
1721
1722                 req_fm_len = obd_object_end - lun_start;
1723                 fm_local->fm_length = 0;
1724                 len_mapped_single_call = 0;
1725
1726                 /* If the output buffer is very large and the objects have many
1727                  * extents we may need to loop on a single OST repeatedly */
1728                 ost_eof = 0;
1729                 ost_done = 0;
1730                 do {
1731                         if (get_num_extents == 0) {
1732                                 /* Don't get too many extents. */
1733                                 if (current_extent + count_local >
1734                                     fiemap->fm_extent_count)
1735                                         count_local = fiemap->fm_extent_count -
1736                                                                  current_extent;
1737                         }
1738
1739                         lun_start += len_mapped_single_call;
1740                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
1741                         req_fm_len = fm_local->fm_length;
1742                         fm_local->fm_extent_count = enough ? 1 : count_local;
1743                         fm_local->fm_mapped_extents = 0;
1744                         fm_local->fm_flags = fiemap->fm_flags;
1745
1746                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1747                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1748
1749                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1750                                 GOTO(out, rc = -EINVAL);
1751
1752                         /* If OST is inactive, return extent with UNKNOWN flag */
1753                         if (!lov->lov_tgts[ost_index]->ltd_active) {
1754                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1755                                 fm_local->fm_mapped_extents = 1;
1756
1757                                 lcl_fm_ext[0].fe_logical = lun_start;
1758                                 lcl_fm_ext[0].fe_length = obd_object_end -
1759                                                                       lun_start;
1760                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1761
1762                                 goto inactive_tgt;
1763                         }
1764
1765                         fm_local->fm_start = lun_start;
1766                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1767                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1768                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1769                         rc = obd_get_info(NULL,
1770                                           lov->lov_tgts[ost_index]->ltd_exp,
1771                                           keylen, key, vallen, fm_local, lsm);
1772                         if (rc != 0)
1773                                 GOTO(out, rc);
1774
1775 inactive_tgt:
1776                         ext_count = fm_local->fm_mapped_extents;
1777                         if (ext_count == 0) {
1778                                 ost_done = 1;
1779                                 /* If last stripe has hole at the end,
1780                                  * then we need to return */
1781                                 if (cur_stripe_wrap == last_stripe) {
1782                                         fiemap->fm_mapped_extents = 0;
1783                                         goto finish;
1784                                 }
1785                                 break;
1786                         } else if (enough) {
1787                                 /*
1788                                  * We've collected enough extents and there are
1789                                  * more extents after it.
1790                                  */
1791                                 goto finish;
1792                         }
1793
1794                         /* If we just need num of extents then go to next device */
1795                         if (get_num_extents) {
1796                                 current_extent += ext_count;
1797                                 break;
1798                         }
1799
1800                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1801                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1802
1803                         /* Have we finished mapping on this device? */
1804                         if (req_fm_len <= len_mapped_single_call)
1805                                 ost_done = 1;
1806
1807                         /* Clear the EXTENT_LAST flag which can be present on
1808                          * last extent */
1809                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1810                                 lcl_fm_ext[ext_count - 1].fe_flags &=
1811                                                             ~FIEMAP_EXTENT_LAST;
1812
1813                         curr_loc = lov_stripe_size(lsm,
1814                                            lcl_fm_ext[ext_count - 1].fe_logical+
1815                                            lcl_fm_ext[ext_count - 1].fe_length,
1816                                            cur_stripe);
1817                         if (curr_loc >= fm_key->oa.o_size)
1818                                 ost_eof = 1;
1819
1820                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1821                                                      ost_index, ext_count,
1822                                                      current_extent);
1823
1824                         current_extent += ext_count;
1825
1826                         /* Ran out of available extents? */
1827                         if (current_extent >= fiemap->fm_extent_count)
1828                                 enough = true;
1829                 } while (ost_done == 0 && ost_eof == 0);
1830
1831                 if (cur_stripe_wrap == last_stripe)
1832                         goto finish;
1833         }
1834
1835 finish:
1836         /* Indicate that we are returning device offsets unless file just has
1837          * single stripe */
1838         if (lsm->lsm_stripe_count > 1)
1839                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1840
1841         if (get_num_extents)
1842                 goto skip_last_device_calc;
1843
1844         /* Check if we have reached the last stripe and whether mapping for that
1845          * stripe is done. */
1846         if (cur_stripe_wrap == last_stripe) {
1847                 if (ost_done || ost_eof)
1848                         fiemap->fm_extents[current_extent - 1].fe_flags |=
1849                                                              FIEMAP_EXTENT_LAST;
1850         }
1851
1852 skip_last_device_calc:
1853         fiemap->fm_mapped_extents = current_extent;
1854
1855 out:
1856         if (fm_local)
1857                 OBD_FREE_LARGE(fm_local, buffer_size);
1858         return rc;
1859 }
1860
1861 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1862                         __u32 keylen, void *key,
1863                         __u32 *vallen, void *val,
1864                         struct lov_stripe_md *lsm)
1865 {
1866         struct obd_device *obddev = class_exp2obd(exp);
1867         struct lov_obd *lov = &obddev->u.lov;
1868         int rc;
1869         ENTRY;
1870
1871         if (!vallen || !val)
1872                 RETURN(-EFAULT);
1873
1874         obd_getref(obddev);
1875
1876         if (KEY_IS(KEY_LOVDESC)) {
1877                 struct lov_desc *desc_ret = val;
1878                 *desc_ret = lov->desc;
1879
1880                 GOTO(out, rc = 0);
1881         } else if (KEY_IS(KEY_FIEMAP)) {
1882                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
1883                 GOTO(out, rc);
1884         } else if (KEY_IS(KEY_TGT_COUNT)) {
1885                 *((int *)val) = lov->desc.ld_tgt_count;
1886                 GOTO(out, rc = 0);
1887         }
1888
1889         rc = -EINVAL;
1890
1891 out:
1892         obd_putref(obddev);
1893         RETURN(rc);
1894 }
1895
1896 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1897                               __u32 keylen, void *key,
1898                               __u32 vallen, void *val,
1899                               struct ptlrpc_request_set *set)
1900 {
1901         struct obd_device *obddev = class_exp2obd(exp);
1902         struct lov_obd *lov = &obddev->u.lov;
1903         struct lov_tgt_desc *tgt;
1904         int do_inactive = 0;
1905         int no_set = 0;
1906         u32 count;
1907         u32 i;
1908         int rc = 0;
1909         int err;
1910         ENTRY;
1911
1912         if (set == NULL) {
1913                 no_set = 1;
1914                 set = ptlrpc_prep_set();
1915                 if (!set)
1916                         RETURN(-ENOMEM);
1917         }
1918
1919         obd_getref(obddev);
1920         count = lov->desc.ld_tgt_count;
1921
1922         if (KEY_IS(KEY_CHECKSUM)) {
1923                 do_inactive = 1;
1924         } else if (KEY_IS(KEY_CACHE_SET)) {
1925                 LASSERT(lov->lov_cache == NULL);
1926                 lov->lov_cache = val;
1927                 do_inactive = 1;
1928                 cl_cache_incref(lov->lov_cache);
1929         }
1930
1931         for (i = 0; i < count; i++) {
1932                 tgt = lov->lov_tgts[i];
1933
1934                 /* OST was disconnected */
1935                 if (!tgt || !tgt->ltd_exp)
1936                         continue;
1937
1938                 /* OST is inactive and we don't want inactive OSCs */
1939                 if (!tgt->ltd_active && !do_inactive)
1940                         continue;
1941
1942                 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1943                                          vallen, val, set);
1944                 if (!rc)
1945                         rc = err;
1946         }
1947
1948         obd_putref(obddev);
1949         if (no_set) {
1950                 err = ptlrpc_set_wait(set);
1951                 if (!rc)
1952                         rc = err;
1953                 ptlrpc_set_destroy(set);
1954         }
1955         RETURN(rc);
1956 }
1957
1958 void lov_stripe_lock(struct lov_stripe_md *md)
1959 __acquires(&md->lsm_lock)
1960 {
1961         LASSERT(md->lsm_lock_owner != current_pid());
1962         spin_lock(&md->lsm_lock);
1963         LASSERT(md->lsm_lock_owner == 0);
1964         md->lsm_lock_owner = current_pid();
1965 }
1966
1967 void lov_stripe_unlock(struct lov_stripe_md *md)
1968 __releases(&md->lsm_lock)
1969 {
1970         LASSERT(md->lsm_lock_owner == current_pid());
1971         md->lsm_lock_owner = 0;
1972         spin_unlock(&md->lsm_lock);
1973 }
1974
1975 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1976                         struct obd_quotactl *oqctl)
1977 {
1978         struct lov_obd      *lov = &obd->u.lov;
1979         struct lov_tgt_desc *tgt;
1980         __u64                curspace = 0;
1981         __u64                bhardlimit = 0;
1982         int                  i, rc = 0;
1983         ENTRY;
1984
1985         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
1986             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
1987             oqctl->qc_cmd != Q_GETOQUOTA &&
1988             oqctl->qc_cmd != Q_INITQUOTA &&
1989             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
1990             oqctl->qc_cmd != Q_FINVALIDATE) {
1991                 CERROR("%s: bad quota opc %x for lov obd\n",
1992                        obd->obd_name, oqctl->qc_cmd);
1993                 RETURN(-EFAULT);
1994         }
1995
1996         /* for lov tgt */
1997         obd_getref(obd);
1998         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1999                 int err;
2000
2001                 tgt = lov->lov_tgts[i];
2002
2003                 if (!tgt)
2004                         continue;
2005
2006                 if (!tgt->ltd_active || tgt->ltd_reap) {
2007                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2008                             lov->lov_tgts[i]->ltd_activate) {
2009                                 rc = -ENETDOWN;
2010                                 CERROR("ost %d is inactive\n", i);
2011                         } else {
2012                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2013                         }
2014                         continue;
2015                 }
2016
2017                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2018                 if (err) {
2019                         if (tgt->ltd_active && !rc)
2020                                 rc = err;
2021                         continue;
2022                 }
2023
2024                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2025                         curspace += oqctl->qc_dqblk.dqb_curspace;
2026                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2027                 }
2028         }
2029         obd_putref(obd);
2030
2031         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2032                 oqctl->qc_dqblk.dqb_curspace = curspace;
2033                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2034         }
2035         RETURN(rc);
2036 }
2037
2038 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2039                           struct obd_quotactl *oqctl)
2040 {
2041         struct lov_obd *lov = &obd->u.lov;
2042         int             i, rc = 0;
2043         ENTRY;
2044
2045         obd_getref(obd);
2046
2047         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2048                 if (!lov->lov_tgts[i])
2049                         continue;
2050
2051                 /* Skip quota check on the administratively disabled OSTs. */
2052                 if (!lov->lov_tgts[i]->ltd_activate) {
2053                         CWARN("lov idx %d was administratively disabled, "
2054                               "skip quotacheck on it.\n", i);
2055                         continue;
2056                 }
2057
2058                 if (!lov->lov_tgts[i]->ltd_active) {
2059                         CERROR("lov idx %d inactive\n", i);
2060                         rc = -EIO;
2061                         goto out;
2062                 }
2063         }
2064
2065         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2066                 int err;
2067
2068                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2069                         continue;
2070
2071                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2072                 if (err && !rc)
2073                         rc = err;
2074         }
2075
2076 out:
2077         obd_putref(obd);
2078
2079         RETURN(rc);
2080 }
2081
2082 static struct obd_ops lov_obd_ops = {
2083         .o_owner                = THIS_MODULE,
2084         .o_setup                = lov_setup,
2085         .o_precleanup           = lov_precleanup,
2086         .o_cleanup              = lov_cleanup,
2087         .o_connect              = lov_connect,
2088         .o_disconnect           = lov_disconnect,
2089         .o_statfs               = lov_statfs,
2090         .o_statfs_async         = lov_statfs_async,
2091         .o_packmd               = lov_packmd,
2092         .o_unpackmd             = lov_unpackmd,
2093         .o_getattr_async        = lov_getattr_async,
2094         .o_setattr_async        = lov_setattr_async,
2095         .o_iocontrol            = lov_iocontrol,
2096         .o_get_info             = lov_get_info,
2097         .o_set_info_async       = lov_set_info_async,
2098         .o_notify               = lov_notify,
2099         .o_pool_new             = lov_pool_new,
2100         .o_pool_rem             = lov_pool_remove,
2101         .o_pool_add             = lov_pool_add,
2102         .o_pool_del             = lov_pool_del,
2103         .o_getref               = lov_getref,
2104         .o_putref               = lov_putref,
2105         .o_quotactl             = lov_quotactl,
2106         .o_quotacheck           = lov_quotacheck,
2107 };
2108
2109 struct kmem_cache *lov_oinfo_slab;
2110
2111 static int __init lov_init(void)
2112 {
2113         bool enable_proc = true;
2114         struct obd_type *type;
2115         int rc;
2116         ENTRY;
2117
2118         /* print an address of _any_ initialized kernel symbol from this
2119          * module, to allow debugging with gdb that doesn't support data
2120          * symbols from modules.*/
2121         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2122
2123         rc = lu_kmem_init(lov_caches);
2124         if (rc)
2125                 return rc;
2126
2127         lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2128                                            sizeof(struct lov_oinfo), 0,
2129                                            SLAB_HWCACHE_ALIGN, NULL);
2130         if (lov_oinfo_slab == NULL) {
2131                 lu_kmem_fini(lov_caches);
2132                 return -ENOMEM;
2133         }
2134
2135         type = class_search_type(LUSTRE_LOD_NAME);
2136         if (type != NULL && type->typ_procsym != NULL)
2137                 enable_proc = false;
2138
2139         rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
2140                                  LUSTRE_LOV_NAME, &lov_device_type);
2141
2142         if (rc) {
2143                 kmem_cache_destroy(lov_oinfo_slab);
2144                 lu_kmem_fini(lov_caches);
2145         }
2146
2147         RETURN(rc);
2148 }
2149
2150 static void /*__exit*/ lov_exit(void)
2151 {
2152         class_unregister_type(LUSTRE_LOV_NAME);
2153         kmem_cache_destroy(lov_oinfo_slab);
2154         lu_kmem_fini(lov_caches);
2155 }
2156
2157 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2158 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2159 MODULE_LICENSE("GPL");
2160
2161 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);