Whamcloud - gitweb
2ef90dbf29ecc97950f3ff1d14ed639f9fc7def0
[fs/lustre-release.git] / lustre / lov / lov_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/lov/lov_obd.c
37  *
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Peter Braam <braam@clusterfs.com>
40  * Author: Mike Shaver <shaver@clusterfs.com>
41  * Author: Nathan Rutman <nathan@clusterfs.com>
42  */
43
44 #define DEBUG_SUBSYSTEM S_LOV
45 #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 <lclient.h>
59 #include <lustre/ll_fiemap.h>
60 #include <lustre_fid.h>
61
62 #include "lov_internal.h"
63
64 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
65    Any function that expects lov_tgts to remain stationary must take a ref. */
66 static void lov_getref(struct obd_device *obd)
67 {
68         struct lov_obd *lov = &obd->u.lov;
69
70         /* nobody gets through here until lov_putref is done */
71         mutex_lock(&lov->lov_lock);
72         atomic_inc(&lov->lov_refcount);
73         mutex_unlock(&lov->lov_lock);
74         return;
75 }
76
77 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
78
79 static void lov_putref(struct obd_device *obd)
80 {
81         struct lov_obd *lov = &obd->u.lov;
82
83         mutex_lock(&lov->lov_lock);
84         /* ok to dec to 0 more than once -- ltd_exp's will be null */
85         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
86                 struct list_head kill = LIST_HEAD_INIT(kill);
87                 struct lov_tgt_desc *tgt, *n;
88                 int i;
89
90                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
91                        lov->lov_death_row);
92                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
93                         tgt = lov->lov_tgts[i];
94
95                         if (!tgt || !tgt->ltd_reap)
96                                 continue;
97                         list_add(&tgt->ltd_kill, &kill);
98                         /* XXX - right now there is a dependency on ld_tgt_count
99                          * being the maximum tgt index for computing the
100                          * mds_max_easize. So we can't shrink it. */
101                         lov_ost_pool_remove(&lov->lov_packed, i);
102                         lov->lov_tgts[i] = NULL;
103                         lov->lov_death_row--;
104                 }
105                 mutex_unlock(&lov->lov_lock);
106
107                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
108                         list_del(&tgt->ltd_kill);
109                         /* Disconnect */
110                         __lov_del_obd(obd, tgt);
111                 }
112         } else {
113                 mutex_unlock(&lov->lov_lock);
114         }
115 }
116
117 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
118                               enum obd_notify_event ev);
119 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
120                       enum obd_notify_event ev, void *data);
121
122 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
123                     struct obd_connect_data *data)
124 {
125         struct lov_obd *lov = &obd->u.lov;
126         struct obd_uuid *tgt_uuid;
127         struct obd_device *tgt_obd;
128         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
129         struct obd_import *imp;
130         int rc;
131         ENTRY;
132
133         if (lov->lov_tgts[index] == NULL)
134                 RETURN(-EINVAL);
135
136         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
137         tgt_obd = lov->lov_tgts[index]->ltd_obd;
138
139         if (!tgt_obd->obd_set_up) {
140                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
141                 RETURN(-EINVAL);
142         }
143
144         /* override the sp_me from lov */
145         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
146
147         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
148                 data->ocd_index = index;
149
150         /*
151          * Divine LOV knows that OBDs under it are OSCs.
152          */
153         imp = tgt_obd->u.cli.cl_import;
154
155         if (activate) {
156                 tgt_obd->obd_no_recov = 0;
157                 /* FIXME this is probably supposed to be
158                    ptlrpc_set_import_active.  Horrible naming. */
159                 ptlrpc_activate_import(imp);
160         }
161
162         rc = obd_register_observer(tgt_obd, obd);
163         if (rc) {
164                 CERROR("Target %s register_observer error %d\n",
165                        obd_uuid2str(tgt_uuid), rc);
166                 RETURN(rc);
167         }
168
169
170         if (imp->imp_invalid) {
171                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
172                        "disabled\n", obd_uuid2str(tgt_uuid));
173                 RETURN(0);
174         }
175
176         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
177                          &lov_osc_uuid, data, NULL);
178         if (rc || !lov->lov_tgts[index]->ltd_exp) {
179                 CERROR("Target %s connect error %d\n",
180                        obd_uuid2str(tgt_uuid), rc);
181                 RETURN(-ENODEV);
182         }
183
184         lov->lov_tgts[index]->ltd_reap = 0;
185
186         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
187                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
188
189         if (lov->targets_proc_entry != NULL) {
190                 struct proc_dir_entry *osc_symlink;
191                 struct obd_device *osc_obd;
192
193                 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
194
195                 LASSERT(osc_obd != NULL);
196                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
197                 LASSERT(osc_obd->obd_type->typ_name != NULL);
198
199                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
200                                                   lov->targets_proc_entry,
201                                                   "../../../%s/%s",
202                                                   osc_obd->obd_type->typ_name,
203                                                   osc_obd->obd_name);
204                 if (osc_symlink == NULL) {
205                         CERROR("cannot register LOV target "
206                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
207                                obd->obd_type->typ_name, obd->obd_name,
208                                osc_obd->obd_name);
209                 }
210         }
211         RETURN(0);
212 }
213
214 static int lov_connect(const struct lu_env *env,
215                        struct obd_export **exp, struct obd_device *obd,
216                        struct obd_uuid *cluuid, struct obd_connect_data *data,
217                        void *localdata)
218 {
219         struct lov_obd *lov = &obd->u.lov;
220         struct lov_tgt_desc *tgt;
221         struct lustre_handle conn;
222         int i, rc;
223         ENTRY;
224
225         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
226
227         rc = class_connect(&conn, obd, cluuid);
228         if (rc)
229                 RETURN(rc);
230
231         *exp = class_conn2export(&conn);
232
233         /* Why should there ever be more than 1 connect? */
234         lov->lov_connects++;
235         LASSERT(lov->lov_connects == 1);
236
237         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
238         if (data)
239                 lov->lov_ocd = *data;
240
241         lov->targets_proc_entry = lprocfs_seq_register("target_obds",
242                                                        obd->obd_proc_entry,
243                                                        NULL, NULL);
244         if (IS_ERR(lov->targets_proc_entry)) {
245                 CERROR("%s: cannot register "
246                        "/proc/fs/lustre/%s/%s/target_obds\n",
247                        obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
248                 lov->targets_proc_entry = NULL;
249         }
250
251         obd_getref(obd);
252         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
253                 tgt = lov->lov_tgts[i];
254                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
255                         continue;
256                 /* Flags will be lowest common denominator */
257                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
258                 if (rc) {
259                         CERROR("%s: lov connect tgt %d failed: %d\n",
260                                obd->obd_name, i, rc);
261                         continue;
262                 }
263                 /* connect to administrative disabled ost */
264                 if (!lov->lov_tgts[i]->ltd_exp)
265                         continue;
266
267                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
268                                 OBD_NOTIFY_CONNECT, (void *)&i);
269                 if (rc) {
270                         CERROR("%s error sending notify %d\n",
271                                obd->obd_name, rc);
272                 }
273         }
274         obd_putref(obd);
275
276         RETURN(0);
277 }
278
279 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
280 {
281         struct lov_obd *lov = &obd->u.lov;
282         struct obd_device *osc_obd;
283         int rc;
284         ENTRY;
285
286         osc_obd = class_exp2obd(tgt->ltd_exp);
287         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
288                obd->obd_name, osc_obd->obd_name);
289
290         if (tgt->ltd_active) {
291                 tgt->ltd_active = 0;
292                 lov->desc.ld_active_tgt_count--;
293                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
294         }
295
296         if (osc_obd) {
297                 /* Pass it on to our clients.
298                  * XXX This should be an argument to disconnect,
299                  * XXX not a back-door flag on the OBD.  Ah well.
300                  */
301                 osc_obd->obd_force = obd->obd_force;
302                 osc_obd->obd_fail = obd->obd_fail;
303                 osc_obd->obd_no_recov = obd->obd_no_recov;
304
305                 if (lov->targets_proc_entry != NULL)
306                         lprocfs_remove_proc_entry(osc_obd->obd_name,
307                                                   lov->targets_proc_entry);
308         }
309
310         obd_register_observer(osc_obd, NULL);
311
312         rc = obd_disconnect(tgt->ltd_exp);
313         if (rc) {
314                 CERROR("Target %s disconnect error %d\n",
315                        tgt->ltd_uuid.uuid, rc);
316                 rc = 0;
317         }
318
319         tgt->ltd_exp = NULL;
320         RETURN(0);
321 }
322
323 static int lov_disconnect(struct obd_export *exp)
324 {
325         struct obd_device *obd = class_exp2obd(exp);
326         struct lov_obd *lov = &obd->u.lov;
327         int i, rc;
328         ENTRY;
329
330         if (!lov->lov_tgts)
331                 goto out;
332
333         /* Only disconnect the underlying layers on the final disconnect. */
334         lov->lov_connects--;
335         if (lov->lov_connects != 0) {
336                 /* why should there be more than 1 connect? */
337                 CERROR("disconnect #%d\n", lov->lov_connects);
338                 goto out;
339         }
340
341         /* Let's hold another reference so lov_del_obd doesn't spin through
342            putref every time */
343         obd_getref(obd);
344
345         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
346                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
347                         /* Disconnection is the last we know about an obd */
348                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
349                 }
350         }
351         obd_putref(obd);
352
353         if (lov->targets_proc_entry != NULL)
354                 lprocfs_remove(&lov->targets_proc_entry);
355
356 out:
357         rc = class_disconnect(exp); /* bz 9811 */
358         RETURN(rc);
359 }
360
361 /* Error codes:
362  *
363  *  -EINVAL  : UUID can't be found in the LOV's target list
364  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
365  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
366  *  any >= 0 : is log target index
367  */
368 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
369                               enum obd_notify_event ev)
370 {
371         struct lov_obd *lov = &obd->u.lov;
372         struct lov_tgt_desc *tgt;
373         int index, activate, active;
374         ENTRY;
375
376         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
377                lov, uuid->uuid, ev);
378
379         obd_getref(obd);
380         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
381                 tgt = lov->lov_tgts[index];
382                 if (!tgt)
383                         continue;
384                 /*
385                  * LU-642, initially inactive OSC could miss the obd_connect,
386                  * we make up for it here.
387                  */
388                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
389                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
390                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
391
392                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
393                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
394                 }
395                 if (!tgt->ltd_exp)
396                         continue;
397
398                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
399                        index, obd_uuid2str(&tgt->ltd_uuid),
400                        tgt->ltd_exp->exp_handle.h_cookie);
401                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
402                         break;
403         }
404
405         if (index == lov->desc.ld_tgt_count)
406                 GOTO(out, index = -EINVAL);
407
408         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
409                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
410
411                 if (lov->lov_tgts[index]->ltd_activate == activate) {
412                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
413                                uuid->uuid, activate ? "" : "de");
414                 } else {
415                         lov->lov_tgts[index]->ltd_activate = activate;
416                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
417                                activate ? "" : "de", obd_uuid2str(uuid));
418                 }
419
420         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
421                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
422
423                 if (lov->lov_tgts[index]->ltd_active == active) {
424                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
425                                uuid->uuid, active ? "" : "in");
426                         GOTO(out, index);
427                 } else {
428                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
429                                obd_uuid2str(uuid), active ? "" : "in");
430                 }
431
432                 lov->lov_tgts[index]->ltd_active = active;
433                 if (active) {
434                         lov->desc.ld_active_tgt_count++;
435                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
436                 } else {
437                         lov->desc.ld_active_tgt_count--;
438                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
439                 }
440         } else {
441                 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
442                        ev, uuid->uuid);
443         }
444
445  out:
446         obd_putref(obd);
447         RETURN(index);
448 }
449
450 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
451                       enum obd_notify_event ev, void *data)
452 {
453         int rc = 0;
454         struct lov_obd *lov = &obd->u.lov;
455         ENTRY;
456
457         down_read(&lov->lov_notify_lock);
458         if (!lov->lov_connects) {
459                 up_read(&lov->lov_notify_lock);
460                 RETURN(rc);
461         }
462
463         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
464             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
465                 struct obd_uuid *uuid;
466
467                 LASSERT(watched);
468
469                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
470                         up_read(&lov->lov_notify_lock);
471                         CERROR("unexpected notification of %s %s!\n",
472                                watched->obd_type->typ_name,
473                                watched->obd_name);
474                         RETURN(-EINVAL);
475                 }
476                 uuid = &watched->u.cli.cl_target_uuid;
477
478                 /* Set OSC as active before notifying the observer, so the
479                  * observer can use the OSC normally.
480                  */
481                 rc = lov_set_osc_active(obd, uuid, ev);
482                 if (rc < 0) {
483                         up_read(&lov->lov_notify_lock);
484                         CERROR("event(%d) of %s failed: %d\n", ev,
485                                obd_uuid2str(uuid), rc);
486                         RETURN(rc);
487                 }
488                 /* active event should be pass lov target index as data */
489                 data = &rc;
490         }
491
492         /* Pass the notification up the chain. */
493         if (watched) {
494                 rc = obd_notify_observer(obd, watched, ev, data);
495         } else {
496                 /* NULL watched means all osc's in the lov (only for syncs) */
497                 /* sync event should be send lov idx as data */
498                 struct lov_obd *lov = &obd->u.lov;
499                 int i, is_sync;
500
501                 data = &i;
502                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
503                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
504
505                 obd_getref(obd);
506                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
507                         if (!lov->lov_tgts[i])
508                                 continue;
509
510                         /* don't send sync event if target not
511                          * connected/activated */
512                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
513                                 continue;
514
515                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
516                                                  ev, data);
517                         if (rc) {
518                                 CERROR("%s: notify %s of %s failed %d\n",
519                                        obd->obd_name,
520                                        obd->obd_observer->obd_name,
521                                        lov->lov_tgts[i]->ltd_obd->obd_name,
522                                        rc);
523                         }
524                 }
525                 obd_putref(obd);
526         }
527
528         up_read(&lov->lov_notify_lock);
529         RETURN(rc);
530 }
531
532 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
533                           __u32 index, int gen, int active)
534 {
535         struct lov_obd *lov = &obd->u.lov;
536         struct lov_tgt_desc *tgt;
537         struct obd_device *tgt_obd;
538         int rc;
539         ENTRY;
540
541         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
542                uuidp->uuid, index, gen, active);
543
544         if (gen <= 0) {
545                 CERROR("request to add OBD %s with invalid generation: %d\n",
546                        uuidp->uuid, gen);
547                 RETURN(-EINVAL);
548         }
549
550         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
551                                         &obd->obd_uuid);
552         if (tgt_obd == NULL)
553                 RETURN(-EINVAL);
554
555         mutex_lock(&lov->lov_lock);
556
557         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
558                 tgt = lov->lov_tgts[index];
559                 CERROR("UUID %s already assigned at LOV target index %d\n",
560                        obd_uuid2str(&tgt->ltd_uuid), index);
561                 mutex_unlock(&lov->lov_lock);
562                 RETURN(-EEXIST);
563         }
564
565         if (index >= lov->lov_tgt_size) {
566                 /* We need to reallocate the lov target array. */
567                 struct lov_tgt_desc **newtgts, **old = NULL;
568                 __u32 newsize, oldsize = 0;
569
570                 newsize = max(lov->lov_tgt_size, (__u32)2);
571                 while (newsize < index + 1)
572                         newsize = newsize << 1;
573                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
574                 if (newtgts == NULL) {
575                         mutex_unlock(&lov->lov_lock);
576                         RETURN(-ENOMEM);
577                 }
578
579                 if (lov->lov_tgt_size) {
580                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
581                                lov->lov_tgt_size);
582                         old = lov->lov_tgts;
583                         oldsize = lov->lov_tgt_size;
584                 }
585
586                 lov->lov_tgts = newtgts;
587                 lov->lov_tgt_size = newsize;
588                 smp_rmb();
589                 if (old)
590                         OBD_FREE(old, sizeof(*old) * oldsize);
591
592                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
593                        lov->lov_tgts, lov->lov_tgt_size);
594         }
595
596         OBD_ALLOC_PTR(tgt);
597         if (!tgt) {
598                 mutex_unlock(&lov->lov_lock);
599                 RETURN(-ENOMEM);
600         }
601
602         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
603         if (rc) {
604                 mutex_unlock(&lov->lov_lock);
605                 OBD_FREE_PTR(tgt);
606                 RETURN(rc);
607         }
608
609         tgt->ltd_uuid = *uuidp;
610         tgt->ltd_obd = tgt_obd;
611         /* XXX - add a sanity check on the generation number. */
612         tgt->ltd_gen = gen;
613         tgt->ltd_index = index;
614         tgt->ltd_activate = active;
615         lov->lov_tgts[index] = tgt;
616         if (index >= lov->desc.ld_tgt_count)
617                 lov->desc.ld_tgt_count = index + 1;
618
619         mutex_unlock(&lov->lov_lock);
620
621         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
622                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
623
624         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
625
626         if (lov->lov_connects == 0) {
627                 /* lov_connect hasn't been called yet. We'll do the
628                    lov_connect_obd on this target when that fn first runs,
629                    because we don't know the connect flags yet. */
630                 RETURN(0);
631         }
632
633         obd_getref(obd);
634
635         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
636         if (rc)
637                 GOTO(out, rc);
638
639         /* connect to administrative disabled ost */
640         if (!tgt->ltd_exp)
641                 GOTO(out, rc = 0);
642
643         if (lov->lov_cache != NULL) {
644                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
645                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
646                                 sizeof(struct cl_client_cache), lov->lov_cache,
647                                 NULL);
648                 if (rc < 0)
649                         GOTO(out, rc);
650         }
651
652         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
653                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
654                         (void *)&index);
655
656 out:
657         if (rc) {
658                 CERROR("add failed (%d), deleting %s\n", rc,
659                        obd_uuid2str(&tgt->ltd_uuid));
660                 lov_del_target(obd, index, 0, 0);
661         }
662         obd_putref(obd);
663         RETURN(rc);
664 }
665
666 /* Schedule a target for deletion */
667 int lov_del_target(struct obd_device *obd, __u32 index,
668                    struct obd_uuid *uuidp, int gen)
669 {
670         struct lov_obd *lov = &obd->u.lov;
671         int count = lov->desc.ld_tgt_count;
672         int rc = 0;
673         ENTRY;
674
675         if (index >= count) {
676                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
677                        index, count);
678                 RETURN(-EINVAL);
679         }
680
681         /* to make sure there's no ongoing lov_notify() now */
682         down_write(&lov->lov_notify_lock);
683         obd_getref(obd);
684
685         if (!lov->lov_tgts[index]) {
686                 CERROR("LOV target at index %d is not setup.\n", index);
687                 GOTO(out, rc = -EINVAL);
688         }
689
690         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
691                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
692                        lov_uuid2str(lov, index), index,
693                        obd_uuid2str(uuidp));
694                 GOTO(out, rc = -EINVAL);
695         }
696
697         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
698                lov_uuid2str(lov, index), index,
699                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
700                lov->lov_tgts[index]->ltd_active);
701
702         lov->lov_tgts[index]->ltd_reap = 1;
703         lov->lov_death_row++;
704         /* we really delete it from obd_putref */
705 out:
706         obd_putref(obd);
707         up_write(&lov->lov_notify_lock);
708
709         RETURN(rc);
710 }
711
712 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
713 {
714         struct obd_device *osc_obd;
715
716         LASSERT(tgt);
717         LASSERT(tgt->ltd_reap);
718
719         osc_obd = class_exp2obd(tgt->ltd_exp);
720
721         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
722                tgt->ltd_uuid.uuid,
723                osc_obd ? osc_obd->obd_name : "<no obd>");
724
725         if (tgt->ltd_exp)
726                 lov_disconnect_obd(obd, tgt);
727
728         OBD_FREE_PTR(tgt);
729
730         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
731            do it ourselves. And we can't do it from lov_cleanup,
732            because we just lost our only reference to it. */
733         if (osc_obd)
734                 class_manual_cleanup(osc_obd);
735 }
736
737 void lov_fix_desc_stripe_size(__u64 *val)
738 {
739         if (*val < LOV_MIN_STRIPE_SIZE) {
740                 if (*val != 0)
741                         LCONSOLE_INFO("Increasing default stripe size to "
742                                       "minimum %u\n",
743                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
744                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
745         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
746                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
747                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
748                               "multiple of %u)\n",
749                               *val, LOV_MIN_STRIPE_SIZE);
750         }
751 }
752
753 void lov_fix_desc_stripe_count(__u32 *val)
754 {
755         if (*val == 0)
756                 *val = 1;
757 }
758
759 void lov_fix_desc_pattern(__u32 *val)
760 {
761         /* from lov_setstripe */
762         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
763                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
764                 *val = 0;
765         }
766 }
767
768 void lov_fix_desc_qos_maxage(__u32 *val)
769 {
770         if (*val == 0)
771                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
772 }
773
774 void lov_fix_desc(struct lov_desc *desc)
775 {
776         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
777         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
778         lov_fix_desc_pattern(&desc->ld_pattern);
779         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
780 }
781
782 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
783 {
784         struct lov_desc *desc;
785         struct lov_obd *lov = &obd->u.lov;
786 #ifdef LPROCFS
787         struct obd_type *type;
788 #endif
789         int rc;
790         ENTRY;
791
792         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
793                 CERROR("LOV setup requires a descriptor\n");
794                 RETURN(-EINVAL);
795         }
796
797         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
798
799         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
800                 CERROR("descriptor size wrong: %d > %d\n",
801                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
802                 RETURN(-EINVAL);
803         }
804
805         if (desc->ld_magic != LOV_DESC_MAGIC) {
806                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
807                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
808                                    obd->obd_name, desc);
809                             lustre_swab_lov_desc(desc);
810                 } else {
811                         CERROR("%s: Bad lov desc magic: %#x\n",
812                                obd->obd_name, desc->ld_magic);
813                         RETURN(-EINVAL);
814                 }
815         }
816
817         lov_fix_desc(desc);
818
819         desc->ld_active_tgt_count = 0;
820         lov->desc = *desc;
821         lov->lov_tgt_size = 0;
822
823         mutex_init(&lov->lov_lock);
824         atomic_set(&lov->lov_refcount, 0);
825         lov->lov_sp_me = LUSTRE_SP_CLI;
826
827         init_rwsem(&lov->lov_notify_lock);
828
829         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
830                                                    HASH_POOLS_MAX_BITS,
831                                                    HASH_POOLS_BKT_BITS, 0,
832                                                    CFS_HASH_MIN_THETA,
833                                                    CFS_HASH_MAX_THETA,
834                                                    &pool_hash_operations,
835                                                    CFS_HASH_DEFAULT);
836         INIT_LIST_HEAD(&lov->lov_pool_list);
837         lov->lov_pool_count = 0;
838         rc = lov_ost_pool_init(&lov->lov_packed, 0);
839         if (rc)
840                 GOTO(out, rc);
841
842 #ifdef LPROCFS
843         obd->obd_vars = lprocfs_lov_obd_vars;
844         /* If this is true then both client (lov) and server
845          * (lod) are on the same node. The lod layer if loaded
846          * first will register the lov proc directory. In that
847          * case obd->obd_type->typ_procroot will be not set.
848          * Instead we use type->typ_procsym as the parent. */
849         type = class_search_type(LUSTRE_LOD_NAME);
850         if (type != NULL && type->typ_procsym != NULL) {
851                 obd->obd_proc_entry = lprocfs_seq_register(obd->obd_name,
852                                                            type->typ_procsym,
853                                                            obd->obd_vars, obd);
854                 if (IS_ERR(obd->obd_proc_entry)) {
855                         rc = PTR_ERR(obd->obd_proc_entry);
856                         CERROR("error %d setting up lprocfs for %s\n", rc,
857                                obd->obd_name);
858                         obd->obd_proc_entry = NULL;
859                 }
860         } else {
861                 rc = lprocfs_obd_setup(obd);
862         }
863
864         if (rc == 0) {
865                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
866                                         0444, &lov_proc_target_fops, obd);
867                 if (rc)
868                         CWARN("Error adding the target_obd file\n");
869
870                 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
871                                                         obd->obd_proc_entry,
872                                                         NULL, NULL);
873                 if (IS_ERR(lov->lov_pool_proc_entry)) {
874                         rc = PTR_ERR(lov->lov_pool_proc_entry);
875                         CERROR("error %d setting up lprocfs for pools\n", rc);
876                         lov->lov_pool_proc_entry = NULL;
877                 }
878         }
879 #endif
880         RETURN(0);
881
882 out:
883         return rc;
884 }
885
886 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
887 {
888         int rc = 0;
889         struct lov_obd *lov = &obd->u.lov;
890
891         ENTRY;
892
893         switch (stage) {
894         case OBD_CLEANUP_EARLY: {
895                 int i;
896                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
897                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
898                                 continue;
899                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
900                                        OBD_CLEANUP_EARLY);
901                 }
902                 break;
903         }
904         default:
905                 break;
906         }
907
908         RETURN(rc);
909 }
910
911 static int lov_cleanup(struct obd_device *obd)
912 {
913         struct lov_obd *lov = &obd->u.lov;
914         struct list_head *pos, *tmp;
915         struct pool_desc *pool;
916         ENTRY;
917
918         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
919                 pool = list_entry(pos, struct pool_desc, pool_list);
920                 /* free pool structs */
921                 CDEBUG(D_INFO, "delete pool %p\n", pool);
922                 /* In the function below, .hs_keycmp resolves to
923                  * pool_hashkey_keycmp() */
924                 /* coverity[overrun-buffer-val] */
925                 lov_pool_del(obd, pool->pool_name);
926         }
927         cfs_hash_putref(lov->lov_pools_hash_body);
928         lov_ost_pool_free(&lov->lov_packed);
929
930         lprocfs_obd_cleanup(obd);
931         if (lov->lov_tgts) {
932                 int i;
933                 obd_getref(obd);
934                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
935                         if (!lov->lov_tgts[i])
936                                 continue;
937
938                         /* Inactive targets may never have connected */
939                         if (lov->lov_tgts[i]->ltd_active ||
940                             atomic_read(&lov->lov_refcount))
941                                 /* We should never get here - these
942                                  * should have been removed in the
943                                  * disconnect. */
944                                 CERROR("%s: lov tgt %d not cleaned! "
945                                        "deathrow=%d, lovrc=%d\n",
946                                        obd->obd_name, i, lov->lov_death_row,
947                                        atomic_read(&lov->lov_refcount));
948                         lov_del_target(obd, i, 0, 0);
949                 }
950                 obd_putref(obd);
951                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
952                          lov->lov_tgt_size);
953                 lov->lov_tgt_size = 0;
954         }
955         RETURN(0);
956 }
957
958 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
959                             __u32 *indexp, int *genp)
960 {
961         struct obd_uuid obd_uuid;
962         int cmd;
963         int rc = 0;
964         ENTRY;
965
966         switch(cmd = lcfg->lcfg_command) {
967         case LCFG_LOV_ADD_OBD:
968         case LCFG_LOV_ADD_INA:
969         case LCFG_LOV_DEL_OBD: {
970                 __u32 index;
971                 int gen;
972                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
973                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
974                         GOTO(out, rc = -EINVAL);
975
976                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
977
978                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", indexp) != 1)
979                         GOTO(out, rc = -EINVAL);
980                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
981                         GOTO(out, rc = -EINVAL);
982                 index = *indexp;
983                 gen = *genp;
984                 if (cmd == LCFG_LOV_ADD_OBD)
985                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
986                 else if (cmd == LCFG_LOV_ADD_INA)
987                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
988                 else
989                         rc = lov_del_target(obd, index, &obd_uuid, gen);
990                 GOTO(out, rc);
991         }
992         case LCFG_PARAM: {
993                 struct lov_desc *desc = &(obd->u.lov.desc);
994
995                 if (!desc)
996                         GOTO(out, rc = -EINVAL);
997
998                 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
999                                               lcfg, obd);
1000                 if (rc > 0)
1001                         rc = 0;
1002                 GOTO(out, rc);
1003         }
1004         case LCFG_POOL_NEW:
1005         case LCFG_POOL_ADD:
1006         case LCFG_POOL_DEL:
1007         case LCFG_POOL_REM:
1008                 GOTO(out, rc);
1009
1010         default: {
1011                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1012                 GOTO(out, rc = -EINVAL);
1013
1014         }
1015         }
1016 out:
1017         RETURN(rc);
1018 }
1019
1020 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1021 do {                                                                            \
1022         LASSERT((lsmp) != NULL);                                                \
1023         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1024                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1025                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1026 } while (0)
1027
1028 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1029                                  void *data, int rc)
1030 {
1031         struct lov_request_set *lovset = (struct lov_request_set *)data;
1032         int err;
1033         ENTRY;
1034
1035         /* don't do attribute merge if this aysnc op failed */
1036         if (rc)
1037                 atomic_set(&lovset->set_completes, 0);
1038         err = lov_fini_getattr_set(lovset);
1039         RETURN(rc ? rc : err);
1040 }
1041
1042 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1043                               struct ptlrpc_request_set *rqset)
1044 {
1045         struct lov_request_set *lovset;
1046         struct lov_obd *lov;
1047         struct list_head *pos;
1048         struct lov_request *req;
1049         int rc = 0, err;
1050         ENTRY;
1051
1052         LASSERT(oinfo);
1053         ASSERT_LSM_MAGIC(oinfo->oi_md);
1054
1055         if (!exp || !exp->exp_obd)
1056                 RETURN(-ENODEV);
1057
1058         lov = &exp->exp_obd->u.lov;
1059
1060         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1061         if (rc)
1062                 RETURN(rc);
1063
1064         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1065                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1066                oinfo->oi_md->lsm_stripe_size);
1067
1068         list_for_each(pos, &lovset->set_list) {
1069                 req = list_entry(pos, struct lov_request, rq_link);
1070
1071                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1072                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1073                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1074                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1075                                        &req->rq_oi, rqset);
1076                 if (rc) {
1077                         CERROR("%s: getattr objid "DOSTID" subobj"
1078                                DOSTID" on OST idx %d: rc = %d\n",
1079                                exp->exp_obd->obd_name,
1080                                POSTID(&oinfo->oi_oa->o_oi),
1081                                POSTID(&req->rq_oi.oi_oa->o_oi),
1082                                req->rq_idx, rc);
1083                         GOTO(out, rc);
1084                 }
1085         }
1086
1087         if (!list_empty(&rqset->set_requests)) {
1088                 LASSERT(rc == 0);
1089                 LASSERT (rqset->set_interpret == NULL);
1090                 rqset->set_interpret = lov_getattr_interpret;
1091                 rqset->set_arg = (void *)lovset;
1092                 RETURN(rc);
1093         }
1094 out:
1095         if (rc)
1096                 atomic_set(&lovset->set_completes, 0);
1097         err = lov_fini_getattr_set(lovset);
1098         RETURN(rc ? rc : err);
1099 }
1100
1101 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1102                                  void *data, int rc)
1103 {
1104         struct lov_request_set *lovset = (struct lov_request_set *)data;
1105         int err;
1106         ENTRY;
1107
1108         if (rc)
1109                 atomic_set(&lovset->set_completes, 0);
1110         err = lov_fini_setattr_set(lovset);
1111         RETURN(rc ? rc : err);
1112 }
1113
1114 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1115    needed. Otherwise, a client is waiting for responses. */
1116 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1117                              struct obd_trans_info *oti,
1118                              struct ptlrpc_request_set *rqset)
1119 {
1120         struct lov_request_set *set;
1121         struct lov_request *req;
1122         struct list_head *pos;
1123         struct lov_obd *lov;
1124         int rc = 0;
1125         ENTRY;
1126
1127         LASSERT(oinfo);
1128         ASSERT_LSM_MAGIC(oinfo->oi_md);
1129         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1130                 LASSERT(oti);
1131                 LASSERT(oti->oti_logcookies);
1132         }
1133
1134         if (!exp || !exp->exp_obd)
1135                 RETURN(-ENODEV);
1136
1137         lov = &exp->exp_obd->u.lov;
1138         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1139         if (rc)
1140                 RETURN(rc);
1141
1142         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1143                POSTID(&oinfo->oi_md->lsm_oi),
1144                oinfo->oi_md->lsm_stripe_count,
1145                oinfo->oi_md->lsm_stripe_size);
1146
1147         list_for_each(pos, &set->set_list) {
1148                 req = list_entry(pos, struct lov_request, rq_link);
1149
1150                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1151                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1152
1153                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1154                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1155                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1156
1157                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1158                                        &req->rq_oi, oti, rqset);
1159                 if (rc) {
1160                         CERROR("error: setattr objid "DOSTID" subobj"
1161                                DOSTID" on OST idx %d: rc = %d\n",
1162                                POSTID(&set->set_oi->oi_oa->o_oi),
1163                                POSTID(&req->rq_oi.oi_oa->o_oi),
1164                                req->rq_idx, rc);
1165                         break;
1166                 }
1167         }
1168
1169         /* If we are not waiting for responses on async requests, return. */
1170         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1171                 int err;
1172                 if (rc)
1173                         atomic_set(&set->set_completes, 0);
1174                 err = lov_fini_setattr_set(set);
1175                 RETURN(rc ? rc : err);
1176         }
1177
1178         LASSERT(rqset->set_interpret == NULL);
1179         rqset->set_interpret = lov_setattr_interpret;
1180         rqset->set_arg = (void *)set;
1181
1182         RETURN(0);
1183 }
1184
1185 static int lov_change_cbdata(struct obd_export *exp,
1186                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1187                              void *data)
1188 {
1189         struct lov_obd *lov;
1190         int rc = 0, i;
1191         ENTRY;
1192
1193         ASSERT_LSM_MAGIC(lsm);
1194
1195         if (!exp || !exp->exp_obd)
1196                 RETURN(-ENODEV);
1197
1198         lov = &exp->exp_obd->u.lov;
1199         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1200                 struct lov_stripe_md submd;
1201                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1202
1203                 if (lov_oinfo_is_dummy(loi))
1204                         continue;
1205
1206                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1207                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1208                         continue;
1209                 }
1210
1211                 submd.lsm_oi = loi->loi_oi;
1212                 submd.lsm_stripe_count = 0;
1213                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1214                                        &submd, it, data);
1215         }
1216         RETURN(rc);
1217 }
1218
1219 /* find any ldlm lock of the inode in lov
1220  * return 0    not find
1221  *        1    find one
1222  *      < 0    error */
1223 static int lov_find_cbdata(struct obd_export *exp,
1224                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1225                            void *data)
1226 {
1227         struct lov_obd *lov;
1228         int rc = 0, i;
1229         ENTRY;
1230
1231         ASSERT_LSM_MAGIC(lsm);
1232
1233         if (!exp || !exp->exp_obd)
1234                 RETURN(-ENODEV);
1235
1236         lov = &exp->exp_obd->u.lov;
1237         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1238                 struct lov_stripe_md submd;
1239                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1240
1241                 if (lov_oinfo_is_dummy(loi))
1242                         continue;
1243
1244                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1245                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1246                         continue;
1247                 }
1248
1249                 submd.lsm_oi = loi->loi_oi;
1250                 submd.lsm_stripe_count = 0;
1251                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1252                                      &submd, it, data);
1253                 if (rc != 0)
1254                         RETURN(rc);
1255         }
1256         RETURN(rc);
1257 }
1258
1259 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1260 {
1261         struct lov_request_set *lovset = (struct lov_request_set *)data;
1262         int err;
1263         ENTRY;
1264
1265         if (rc)
1266                 atomic_set(&lovset->set_completes, 0);
1267
1268         err = lov_fini_statfs_set(lovset);
1269         RETURN(rc ? rc : err);
1270 }
1271
1272 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1273                             __u64 max_age, struct ptlrpc_request_set *rqset)
1274 {
1275         struct obd_device      *obd = class_exp2obd(exp);
1276         struct lov_request_set *set;
1277         struct lov_request *req;
1278         struct list_head *pos;
1279         struct lov_obd *lov;
1280         int rc = 0;
1281         ENTRY;
1282
1283         LASSERT(oinfo != NULL);
1284         LASSERT(oinfo->oi_osfs != NULL);
1285
1286         lov = &obd->u.lov;
1287         rc = lov_prep_statfs_set(obd, oinfo, &set);
1288         if (rc)
1289                 RETURN(rc);
1290
1291         list_for_each(pos, &set->set_list) {
1292                 req = list_entry(pos, struct lov_request, rq_link);
1293                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1294                                       &req->rq_oi, max_age, rqset);
1295                 if (rc)
1296                         break;
1297         }
1298
1299         if (rc || list_empty(&rqset->set_requests)) {
1300                 int err;
1301                 if (rc)
1302                         atomic_set(&set->set_completes, 0);
1303                 err = lov_fini_statfs_set(set);
1304                 RETURN(rc ? rc : err);
1305         }
1306
1307         LASSERT(rqset->set_interpret == NULL);
1308         rqset->set_interpret = lov_statfs_interpret;
1309         rqset->set_arg = (void *)set;
1310         RETURN(0);
1311 }
1312
1313 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1314                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1315 {
1316         struct ptlrpc_request_set *set = NULL;
1317         struct obd_info oinfo = { { { 0 } } };
1318         int rc = 0;
1319         ENTRY;
1320
1321
1322         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1323          * statfs requests */
1324         set = ptlrpc_prep_set();
1325         if (set == NULL)
1326                 RETURN(-ENOMEM);
1327
1328         oinfo.oi_osfs = osfs;
1329         oinfo.oi_flags = flags;
1330         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1331         if (rc == 0)
1332                 rc = ptlrpc_set_wait(set);
1333         ptlrpc_set_destroy(set);
1334
1335         RETURN(rc);
1336 }
1337
1338 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1339                          void *karg, void __user *uarg)
1340 {
1341         struct obd_device *obddev = class_exp2obd(exp);
1342         struct lov_obd *lov = &obddev->u.lov;
1343         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1344         struct obd_uuid *uuidp;
1345         ENTRY;
1346
1347         switch (cmd) {
1348         case IOC_OBD_STATFS: {
1349                 struct obd_ioctl_data *data = karg;
1350                 struct obd_device *osc_obd;
1351                 struct obd_statfs stat_buf = {0};
1352                 __u32 index;
1353                 __u32 flags;
1354
1355                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1356                 if ((index >= count))
1357                         RETURN(-ENODEV);
1358
1359                 if (!lov->lov_tgts[index])
1360                         /* Try again with the next index */
1361                         RETURN(-EAGAIN);
1362                 if (!lov->lov_tgts[index]->ltd_active)
1363                         RETURN(-ENODATA);
1364
1365                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1366                 if (!osc_obd)
1367                         RETURN(-EINVAL);
1368
1369                 /* copy UUID */
1370                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1371                                  min((int)data->ioc_plen2,
1372                                      (int)sizeof(struct obd_uuid))))
1373                         RETURN(-EFAULT);
1374
1375                 flags = uarg ? *(__u32 __user *)uarg : 0;
1376                 /* got statfs data */
1377                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1378                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1379                                 flags);
1380                 if (rc)
1381                         RETURN(rc);
1382                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1383                                      min((int) data->ioc_plen1,
1384                                          (int) sizeof(stat_buf))))
1385                         RETURN(-EFAULT);
1386                 break;
1387         }
1388         case OBD_IOC_LOV_GET_CONFIG: {
1389                 struct obd_ioctl_data *data;
1390                 struct lov_desc *desc;
1391                 char *buf = NULL;
1392                 __u32 *genp;
1393
1394                 len = 0;
1395                 if (obd_ioctl_getdata(&buf, &len, uarg))
1396                         RETURN(-EINVAL);
1397
1398                 data = (struct obd_ioctl_data *)buf;
1399
1400                 if (sizeof(*desc) > data->ioc_inllen1) {
1401                         obd_ioctl_freedata(buf, len);
1402                         RETURN(-EINVAL);
1403                 }
1404
1405                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1406                         obd_ioctl_freedata(buf, len);
1407                         RETURN(-EINVAL);
1408                 }
1409
1410                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1411                         obd_ioctl_freedata(buf, len);
1412                         RETURN(-EINVAL);
1413                 }
1414
1415                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1416                 memcpy(desc, &(lov->desc), sizeof(*desc));
1417
1418                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1419                 genp = (__u32 *)data->ioc_inlbuf3;
1420                 /* the uuid will be empty for deleted OSTs */
1421                 for (i = 0; i < count; i++, uuidp++, genp++) {
1422                         if (!lov->lov_tgts[i])
1423                                 continue;
1424                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1425                         *genp = lov->lov_tgts[i]->ltd_gen;
1426                 }
1427
1428                 if (copy_to_user(uarg, buf, len))
1429                         rc = -EFAULT;
1430                 obd_ioctl_freedata(buf, len);
1431                 break;
1432         }
1433         case LL_IOC_LOV_GETSTRIPE:
1434                 rc = lov_getstripe(exp, karg, uarg);
1435                 break;
1436         case OBD_IOC_QUOTACTL: {
1437                 struct if_quotactl *qctl = karg;
1438                 struct lov_tgt_desc *tgt = NULL;
1439                 struct obd_quotactl *oqctl;
1440
1441                 if (qctl->qc_valid == QC_OSTIDX) {
1442                         if (count <= qctl->qc_idx)
1443                                 RETURN(-EINVAL);
1444
1445                         tgt = lov->lov_tgts[qctl->qc_idx];
1446                         if (!tgt || !tgt->ltd_exp)
1447                                 RETURN(-EINVAL);
1448                 } else if (qctl->qc_valid == QC_UUID) {
1449                         for (i = 0; i < count; i++) {
1450                                 tgt = lov->lov_tgts[i];
1451                                 if (!tgt ||
1452                                     !obd_uuid_equals(&tgt->ltd_uuid,
1453                                                      &qctl->obd_uuid))
1454                                         continue;
1455
1456                                 if (tgt->ltd_exp == NULL)
1457                                         RETURN(-EINVAL);
1458
1459                                 break;
1460                         }
1461                 } else {
1462                         RETURN(-EINVAL);
1463                 }
1464
1465                 if (i >= count)
1466                         RETURN(-EAGAIN);
1467
1468                 LASSERT(tgt && tgt->ltd_exp);
1469                 OBD_ALLOC_PTR(oqctl);
1470                 if (!oqctl)
1471                         RETURN(-ENOMEM);
1472
1473                 QCTL_COPY(oqctl, qctl);
1474                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1475                 if (rc == 0) {
1476                         QCTL_COPY(qctl, oqctl);
1477                         qctl->qc_valid = QC_OSTIDX;
1478                         qctl->obd_uuid = tgt->ltd_uuid;
1479                 }
1480                 OBD_FREE_PTR(oqctl);
1481                 break;
1482         }
1483         default: {
1484                 int set = 0;
1485
1486                 if (count == 0)
1487                         RETURN(-ENOTTY);
1488
1489                 for (i = 0; i < count; i++) {
1490                         int err;
1491                         struct obd_device *osc_obd;
1492
1493                         /* OST was disconnected */
1494                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1495                                 continue;
1496
1497                         /* ll_umount_begin() sets force flag but for lov, not
1498                          * osc. Let's pass it through */
1499                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1500                         osc_obd->obd_force = obddev->obd_force;
1501                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1502                                             len, karg, uarg);
1503                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1504                                 RETURN(err);
1505                         } else if (err) {
1506                                 if (lov->lov_tgts[i]->ltd_active) {
1507                                         CDEBUG(err == -ENOTTY ?
1508                                                D_IOCTL : D_WARNING,
1509                                                "iocontrol OSC %s on OST "
1510                                                "idx %d cmd %x: err = %d\n",
1511                                                lov_uuid2str(lov, i),
1512                                                i, cmd, err);
1513                                         if (!rc)
1514                                                 rc = err;
1515                                 }
1516                         } else {
1517                                 set = 1;
1518                         }
1519                 }
1520                 if (!set && !rc)
1521                         rc = -EIO;
1522         }
1523         }
1524
1525         RETURN(rc);
1526 }
1527
1528 #define FIEMAP_BUFFER_SIZE 4096
1529
1530 /**
1531  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1532  * call. The local end offset and the device are sent in the first
1533  * fm_extent. This function calculates the stripe number from the index.
1534  * This function returns a stripe_no on which mapping is to be restarted.
1535  *
1536  * This function returns fm_end_offset which is the in-OST offset at which
1537  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1538  * will re-calculate proper offset in next stripe.
1539  * Note that the first extent is passed to lov_get_info via the value field.
1540  *
1541  * \param fiemap fiemap request header
1542  * \param lsm striping information for the file
1543  * \param fm_start logical start of mapping
1544  * \param fm_end logical end of mapping
1545  * \param start_stripe starting stripe will be returned in this
1546  */
1547 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1548                                    struct lov_stripe_md *lsm, obd_size fm_start,
1549                                    obd_size fm_end, int *start_stripe)
1550 {
1551         obd_size local_end = fiemap->fm_extents[0].fe_logical;
1552         obd_off lun_start, lun_end;
1553         obd_size fm_end_offset;
1554         int stripe_no = -1, i;
1555
1556         if (fiemap->fm_extent_count == 0 ||
1557             fiemap->fm_extents[0].fe_logical == 0)
1558                 return 0;
1559
1560         /* Find out stripe_no from ost_index saved in the fe_device */
1561         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1562                 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1563
1564                 if (lov_oinfo_is_dummy(oinfo))
1565                         continue;
1566
1567                 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1568                         stripe_no = i;
1569                         break;
1570                 }
1571         }
1572
1573         if (stripe_no == -1)
1574                 return -EINVAL;
1575
1576         /* If we have finished mapping on previous device, shift logical
1577          * offset to start of next device */
1578         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1579                                    &lun_start, &lun_end)) != 0 &&
1580                                    local_end < lun_end) {
1581                 fm_end_offset = local_end;
1582                 *start_stripe = stripe_no;
1583         } else {
1584                 /* This is a special value to indicate that caller should
1585                  * calculate offset in next stripe. */
1586                 fm_end_offset = 0;
1587                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1588         }
1589
1590         return fm_end_offset;
1591 }
1592
1593 /**
1594  * We calculate on which OST the mapping will end. If the length of mapping
1595  * is greater than (stripe_size * stripe_count) then the last_stripe will
1596  * will be one just before start_stripe. Else we check if the mapping
1597  * intersects each OST and find last_stripe.
1598  * This function returns the last_stripe and also sets the stripe_count
1599  * over which the mapping is spread
1600  *
1601  * \param lsm striping information for the file
1602  * \param fm_start logical start of mapping
1603  * \param fm_end logical end of mapping
1604  * \param start_stripe starting stripe of the mapping
1605  * \param stripe_count the number of stripes across which to map is returned
1606  *
1607  * \retval last_stripe return the last stripe of the mapping
1608  */
1609 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1610                             obd_size fm_end, int start_stripe,
1611                             int *stripe_count)
1612 {
1613         int last_stripe;
1614         obd_off obd_start, obd_end;
1615         int i, j;
1616
1617         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1618                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1619                                                               start_stripe - 1);
1620                 *stripe_count = lsm->lsm_stripe_count;
1621         } else {
1622                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1623                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
1624                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1625                                                    &obd_start, &obd_end)) == 0)
1626                                 break;
1627                 }
1628                 *stripe_count = j;
1629                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1630         }
1631
1632         return last_stripe;
1633 }
1634
1635 /**
1636  * Set fe_device and copy extents from local buffer into main return buffer.
1637  *
1638  * \param fiemap fiemap request header
1639  * \param lcl_fm_ext array of local fiemap extents to be copied
1640  * \param ost_index OST index to be written into the fm_device field for each
1641                     extent
1642  * \param ext_count number of extents to be copied
1643  * \param current_extent where to start copying in main extent array
1644  */
1645 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1646                                   struct ll_fiemap_extent *lcl_fm_ext,
1647                                   int ost_index, unsigned int ext_count,
1648                                   int current_extent)
1649 {
1650         char *to;
1651         int ext;
1652
1653         for (ext = 0; ext < ext_count; ext++) {
1654                 lcl_fm_ext[ext].fe_device = ost_index;
1655                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1656         }
1657
1658         /* Copy fm_extent's from fm_local to return buffer */
1659         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1660         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1661 }
1662
1663 /**
1664  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1665  * This also handles the restarting of FIEMAP calls in case mapping overflows
1666  * the available number of extents in single call.
1667  */
1668 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1669                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1670 {
1671         struct ll_fiemap_info_key *fm_key = key;
1672         struct ll_user_fiemap *fiemap = val;
1673         struct ll_user_fiemap *fm_local = NULL;
1674         struct ll_fiemap_extent *lcl_fm_ext;
1675         int count_local;
1676         unsigned int get_num_extents = 0;
1677         int ost_index = 0, actual_start_stripe, start_stripe;
1678         obd_size fm_start, fm_end, fm_length, fm_end_offset;
1679         obd_size curr_loc;
1680         int current_extent = 0, rc = 0, i;
1681         int ost_eof = 0; /* EOF for object */
1682         int ost_done = 0; /* done with required mapping for this OST? */
1683         int last_stripe;
1684         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1685         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1686
1687         if (!lsm_has_objects(lsm)) {
1688                 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1689                     fm_key->oa.o_size)) {
1690                         /* released file, return a minimal FIEMAP if
1691                          * request fits in file-size.
1692                          */
1693                         fiemap->fm_mapped_extents = 1;
1694                         fiemap->fm_extents[0].fe_logical =
1695                                                 fm_key->fiemap.fm_start;
1696                         if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1697                             fm_key->oa.o_size)
1698                                 fiemap->fm_extents[0].fe_length =
1699                                                 fm_key->fiemap.fm_length;
1700                         else
1701                                 fiemap->fm_extents[0].fe_length =
1702                                                 fm_key->oa.o_size -
1703                                                 fm_key->fiemap.fm_start;
1704                         fiemap->fm_extents[0].fe_flags |=
1705                                                 (FIEMAP_EXTENT_UNKNOWN |
1706                                                  FIEMAP_EXTENT_LAST);
1707                 }
1708                 GOTO(out, rc = 0);
1709         }
1710
1711         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1712                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1713
1714         OBD_ALLOC_LARGE(fm_local, buffer_size);
1715         if (fm_local == NULL)
1716                 GOTO(out, rc = -ENOMEM);
1717         lcl_fm_ext = &fm_local->fm_extents[0];
1718
1719         count_local = fiemap_size_to_count(buffer_size);
1720
1721         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1722         fm_start = fiemap->fm_start;
1723         fm_length = fiemap->fm_length;
1724         /* Calculate start stripe, last stripe and length of mapping */
1725         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1726         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1727                                                 fm_start + fm_length - 1);
1728         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1729         if (fm_end > fm_key->oa.o_size)
1730                 fm_end = fm_key->oa.o_size;
1731
1732         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1733                                             actual_start_stripe, &stripe_count);
1734
1735         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1736                                                   fm_end, &start_stripe);
1737         if (fm_end_offset == -EINVAL)
1738                 GOTO(out, rc = -EINVAL);
1739
1740         if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1741                 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1742         if (fiemap->fm_extent_count == 0) {
1743                 get_num_extents = 1;
1744                 count_local = 0;
1745         }
1746         /* Check each stripe */
1747         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1748              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1749                 obd_size req_fm_len; /* Stores length of required mapping */
1750                 obd_size len_mapped_single_call;
1751                 obd_off lun_start, lun_end, obd_object_end;
1752                 unsigned int ext_count;
1753
1754                 cur_stripe_wrap = cur_stripe;
1755
1756                 /* Find out range of mapping on this stripe */
1757                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1758                                            &lun_start, &obd_object_end)) == 0)
1759                         continue;
1760
1761                 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe]))
1762                         GOTO(out, rc = -EIO);
1763
1764                 /* If this is a continuation FIEMAP call and we are on
1765                  * starting stripe then lun_start needs to be set to
1766                  * fm_end_offset */
1767                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1768                         lun_start = fm_end_offset;
1769
1770                 if (fm_length != ~0ULL) {
1771                         /* Handle fm_start + fm_length overflow */
1772                         if (fm_start + fm_length < fm_start)
1773                                 fm_length = ~0ULL - fm_start;
1774                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1775                                                      cur_stripe);
1776                 } else {
1777                         lun_end = ~0ULL;
1778                 }
1779
1780                 if (lun_start == lun_end)
1781                         continue;
1782
1783                 req_fm_len = obd_object_end - lun_start;
1784                 fm_local->fm_length = 0;
1785                 len_mapped_single_call = 0;
1786
1787                 /* If the output buffer is very large and the objects have many
1788                  * extents we may need to loop on a single OST repeatedly */
1789                 ost_eof = 0;
1790                 ost_done = 0;
1791                 do {
1792                         if (get_num_extents == 0) {
1793                                 /* Don't get too many extents. */
1794                                 if (current_extent + count_local >
1795                                     fiemap->fm_extent_count)
1796                                         count_local = fiemap->fm_extent_count -
1797                                                                  current_extent;
1798                         }
1799
1800                         lun_start += len_mapped_single_call;
1801                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
1802                         req_fm_len = fm_local->fm_length;
1803                         fm_local->fm_extent_count = count_local;
1804                         fm_local->fm_mapped_extents = 0;
1805                         fm_local->fm_flags = fiemap->fm_flags;
1806
1807                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1808                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1809
1810                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1811                                 GOTO(out, rc = -EINVAL);
1812
1813                         /* If OST is inactive, return extent with UNKNOWN flag */
1814                         if (!lov->lov_tgts[ost_index]->ltd_active) {
1815                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1816                                 fm_local->fm_mapped_extents = 1;
1817
1818                                 lcl_fm_ext[0].fe_logical = lun_start;
1819                                 lcl_fm_ext[0].fe_length = obd_object_end -
1820                                                                       lun_start;
1821                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1822
1823                                 goto inactive_tgt;
1824                         }
1825
1826                         fm_local->fm_start = lun_start;
1827                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1828                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1829                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1830                         rc = obd_get_info(NULL,
1831                                           lov->lov_tgts[ost_index]->ltd_exp,
1832                                           keylen, key, vallen, fm_local, lsm);
1833                         if (rc != 0)
1834                                 GOTO(out, rc);
1835
1836 inactive_tgt:
1837                         ext_count = fm_local->fm_mapped_extents;
1838                         if (ext_count == 0) {
1839                                 ost_done = 1;
1840                                 /* If last stripe has hole at the end,
1841                                  * then we need to return */
1842                                 if (cur_stripe_wrap == last_stripe) {
1843                                         fiemap->fm_mapped_extents = 0;
1844                                         goto finish;
1845                                 }
1846                                 break;
1847                         }
1848
1849                         /* If we just need num of extents then go to next device */
1850                         if (get_num_extents) {
1851                                 current_extent += ext_count;
1852                                 break;
1853                         }
1854
1855                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1856                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1857
1858                         /* Have we finished mapping on this device? */
1859                         if (req_fm_len <= len_mapped_single_call)
1860                                 ost_done = 1;
1861
1862                         /* Clear the EXTENT_LAST flag which can be present on
1863                          * last extent */
1864                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1865                                 lcl_fm_ext[ext_count - 1].fe_flags &=
1866                                                             ~FIEMAP_EXTENT_LAST;
1867
1868                         curr_loc = lov_stripe_size(lsm,
1869                                            lcl_fm_ext[ext_count - 1].fe_logical+
1870                                            lcl_fm_ext[ext_count - 1].fe_length,
1871                                            cur_stripe);
1872                         if (curr_loc >= fm_key->oa.o_size)
1873                                 ost_eof = 1;
1874
1875                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1876                                                      ost_index, ext_count,
1877                                                      current_extent);
1878
1879                         current_extent += ext_count;
1880
1881                         /* Ran out of available extents? */
1882                         if (current_extent >= fiemap->fm_extent_count)
1883                                 goto finish;
1884                 } while (ost_done == 0 && ost_eof == 0);
1885
1886                 if (cur_stripe_wrap == last_stripe)
1887                         goto finish;
1888         }
1889
1890 finish:
1891         /* Indicate that we are returning device offsets unless file just has
1892          * single stripe */
1893         if (lsm->lsm_stripe_count > 1)
1894                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1895
1896         if (get_num_extents)
1897                 goto skip_last_device_calc;
1898
1899         /* Check if we have reached the last stripe and whether mapping for that
1900          * stripe is done. */
1901         if (cur_stripe_wrap == last_stripe) {
1902                 if (ost_done || ost_eof)
1903                         fiemap->fm_extents[current_extent - 1].fe_flags |=
1904                                                              FIEMAP_EXTENT_LAST;
1905         }
1906
1907 skip_last_device_calc:
1908         fiemap->fm_mapped_extents = current_extent;
1909
1910 out:
1911         if (fm_local)
1912                 OBD_FREE_LARGE(fm_local, buffer_size);
1913         return rc;
1914 }
1915
1916 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1917                         __u32 keylen, void *key, __u32 *vallen, void *val,
1918                         struct lov_stripe_md *lsm)
1919 {
1920         struct obd_device *obddev = class_exp2obd(exp);
1921         struct lov_obd *lov = &obddev->u.lov;
1922         int rc;
1923         ENTRY;
1924
1925         if (!vallen || !val)
1926                 RETURN(-EFAULT);
1927
1928         obd_getref(obddev);
1929
1930         if (KEY_IS(KEY_LOVDESC)) {
1931                 struct lov_desc *desc_ret = val;
1932                 *desc_ret = lov->desc;
1933
1934                 GOTO(out, rc = 0);
1935         } else if (KEY_IS(KEY_FIEMAP)) {
1936                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
1937                 GOTO(out, rc);
1938         } else if (KEY_IS(KEY_TGT_COUNT)) {
1939                 *((int *)val) = lov->desc.ld_tgt_count;
1940                 GOTO(out, rc = 0);
1941         }
1942
1943         rc = -EINVAL;
1944
1945 out:
1946         obd_putref(obddev);
1947         RETURN(rc);
1948 }
1949
1950 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1951                               obd_count keylen, void *key, obd_count vallen,
1952                               void *val, struct ptlrpc_request_set *set)
1953 {
1954         struct obd_device *obddev = class_exp2obd(exp);
1955         struct lov_obd *lov = &obddev->u.lov;
1956         obd_count count;
1957         int i, rc = 0, err;
1958         struct lov_tgt_desc *tgt;
1959         int do_inactive = 0;
1960         int no_set = 0;
1961         ENTRY;
1962
1963         if (set == NULL) {
1964                 no_set = 1;
1965                 set = ptlrpc_prep_set();
1966                 if (!set)
1967                         RETURN(-ENOMEM);
1968         }
1969
1970         obd_getref(obddev);
1971         count = lov->desc.ld_tgt_count;
1972
1973         if (KEY_IS(KEY_CHECKSUM)) {
1974                 do_inactive = 1;
1975         } else if (KEY_IS(KEY_CACHE_SET)) {
1976                 LASSERT(lov->lov_cache == NULL);
1977                 lov->lov_cache = val;
1978                 do_inactive = 1;
1979         }
1980
1981         for (i = 0; i < count; i++) {
1982                 tgt = lov->lov_tgts[i];
1983
1984                 /* OST was disconnected */
1985                 if (!tgt || !tgt->ltd_exp)
1986                         continue;
1987
1988                 /* OST is inactive and we don't want inactive OSCs */
1989                 if (!tgt->ltd_active && !do_inactive)
1990                         continue;
1991
1992                 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1993                                          vallen, val, set);
1994                 if (!rc)
1995                         rc = err;
1996         }
1997
1998         obd_putref(obddev);
1999         if (no_set) {
2000                 err = ptlrpc_set_wait(set);
2001                 if (!rc)
2002                         rc = err;
2003                 ptlrpc_set_destroy(set);
2004         }
2005         RETURN(rc);
2006 }
2007
2008 void lov_stripe_lock(struct lov_stripe_md *md)
2009 __acquires(&md->lsm_lock)
2010 {
2011         LASSERT(md->lsm_lock_owner != current_pid());
2012         spin_lock(&md->lsm_lock);
2013         LASSERT(md->lsm_lock_owner == 0);
2014         md->lsm_lock_owner = current_pid();
2015 }
2016
2017 void lov_stripe_unlock(struct lov_stripe_md *md)
2018 __releases(&md->lsm_lock)
2019 {
2020         LASSERT(md->lsm_lock_owner == current_pid());
2021         md->lsm_lock_owner = 0;
2022         spin_unlock(&md->lsm_lock);
2023 }
2024
2025 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2026                         struct obd_quotactl *oqctl)
2027 {
2028         struct lov_obd      *lov = &obd->u.lov;
2029         struct lov_tgt_desc *tgt;
2030         __u64                curspace = 0;
2031         __u64                bhardlimit = 0;
2032         int                  i, rc = 0;
2033         ENTRY;
2034
2035         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2036             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2037             oqctl->qc_cmd != Q_GETOQUOTA &&
2038             oqctl->qc_cmd != Q_INITQUOTA &&
2039             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2040             oqctl->qc_cmd != Q_FINVALIDATE) {
2041                 CERROR("%s: bad quota opc %x for lov obd\n",
2042                        obd->obd_name, oqctl->qc_cmd);
2043                 RETURN(-EFAULT);
2044         }
2045
2046         /* for lov tgt */
2047         obd_getref(obd);
2048         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2049                 int err;
2050
2051                 tgt = lov->lov_tgts[i];
2052
2053                 if (!tgt)
2054                         continue;
2055
2056                 if (!tgt->ltd_active || tgt->ltd_reap) {
2057                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2058                             lov->lov_tgts[i]->ltd_activate) {
2059                                 rc = -ENETDOWN;
2060                                 CERROR("ost %d is inactive\n", i);
2061                         } else {
2062                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2063                         }
2064                         continue;
2065                 }
2066
2067                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2068                 if (err) {
2069                         if (tgt->ltd_active && !rc)
2070                                 rc = err;
2071                         continue;
2072                 }
2073
2074                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2075                         curspace += oqctl->qc_dqblk.dqb_curspace;
2076                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2077                 }
2078         }
2079         obd_putref(obd);
2080
2081         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2082                 oqctl->qc_dqblk.dqb_curspace = curspace;
2083                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2084         }
2085         RETURN(rc);
2086 }
2087
2088 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2089                           struct obd_quotactl *oqctl)
2090 {
2091         struct lov_obd *lov = &obd->u.lov;
2092         int             i, rc = 0;
2093         ENTRY;
2094
2095         obd_getref(obd);
2096
2097         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2098                 if (!lov->lov_tgts[i])
2099                         continue;
2100
2101                 /* Skip quota check on the administratively disabled OSTs. */
2102                 if (!lov->lov_tgts[i]->ltd_activate) {
2103                         CWARN("lov idx %d was administratively disabled, "
2104                               "skip quotacheck on it.\n", i);
2105                         continue;
2106                 }
2107
2108                 if (!lov->lov_tgts[i]->ltd_active) {
2109                         CERROR("lov idx %d inactive\n", i);
2110                         rc = -EIO;
2111                         goto out;
2112                 }
2113         }
2114
2115         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2116                 int err;
2117
2118                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2119                         continue;
2120
2121                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2122                 if (err && !rc)
2123                         rc = err;
2124         }
2125
2126 out:
2127         obd_putref(obd);
2128
2129         RETURN(rc);
2130 }
2131
2132 static struct obd_ops lov_obd_ops = {
2133         .o_owner                = THIS_MODULE,
2134         .o_setup                = lov_setup,
2135         .o_precleanup           = lov_precleanup,
2136         .o_cleanup              = lov_cleanup,
2137         .o_connect              = lov_connect,
2138         .o_disconnect           = lov_disconnect,
2139         .o_statfs               = lov_statfs,
2140         .o_statfs_async         = lov_statfs_async,
2141         .o_packmd               = lov_packmd,
2142         .o_unpackmd             = lov_unpackmd,
2143         .o_getattr_async        = lov_getattr_async,
2144         .o_setattr_async        = lov_setattr_async,
2145         .o_change_cbdata        = lov_change_cbdata,
2146         .o_find_cbdata          = lov_find_cbdata,
2147         .o_iocontrol            = lov_iocontrol,
2148         .o_get_info             = lov_get_info,
2149         .o_set_info_async       = lov_set_info_async,
2150         .o_notify               = lov_notify,
2151         .o_pool_new             = lov_pool_new,
2152         .o_pool_rem             = lov_pool_remove,
2153         .o_pool_add             = lov_pool_add,
2154         .o_pool_del             = lov_pool_del,
2155         .o_getref               = lov_getref,
2156         .o_putref               = lov_putref,
2157         .o_quotactl             = lov_quotactl,
2158         .o_quotacheck           = lov_quotacheck,
2159 };
2160
2161 struct kmem_cache *lov_oinfo_slab;
2162
2163 extern struct lu_kmem_descr lov_caches[];
2164
2165 int __init lov_init(void)
2166 {
2167         bool enable_proc = true;
2168         struct obd_type *type;
2169         int rc;
2170         ENTRY;
2171
2172         /* print an address of _any_ initialized kernel symbol from this
2173          * module, to allow debugging with gdb that doesn't support data
2174          * symbols from modules.*/
2175         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2176
2177         rc = lu_kmem_init(lov_caches);
2178         if (rc)
2179                 return rc;
2180
2181         lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2182                                            sizeof(struct lov_oinfo), 0,
2183                                            SLAB_HWCACHE_ALIGN, NULL);
2184         if (lov_oinfo_slab == NULL) {
2185                 lu_kmem_fini(lov_caches);
2186                 return -ENOMEM;
2187         }
2188
2189         type = class_search_type(LUSTRE_LOD_NAME);
2190         if (type != NULL && type->typ_procsym != NULL)
2191                 enable_proc = false;
2192
2193         rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
2194                                  LUSTRE_LOV_NAME, &lov_device_type);
2195
2196         if (rc) {
2197                 kmem_cache_destroy(lov_oinfo_slab);
2198                 lu_kmem_fini(lov_caches);
2199         }
2200
2201         RETURN(rc);
2202 }
2203
2204 static void /*__exit*/ lov_exit(void)
2205 {
2206         class_unregister_type(LUSTRE_LOV_NAME);
2207         kmem_cache_destroy(lov_oinfo_slab);
2208         lu_kmem_fini(lov_caches);
2209 }
2210
2211 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2212 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2213 MODULE_LICENSE("GPL");
2214
2215 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);