Whamcloud - gitweb
LU-2785 osc: remove unused obd methods
[fs/lustre-release.git] / lustre / lov / lov_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/lov/lov_obd.c
37  *
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Peter Braam <braam@clusterfs.com>
40  * Author: Mike Shaver <shaver@clusterfs.com>
41  * Author: Nathan Rutman <nathan@clusterfs.com>
42  */
43
44 #define DEBUG_SUBSYSTEM S_LOV
45 #ifdef __KERNEL__
46 #include <libcfs/libcfs.h>
47 #else
48 #include <liblustre.h>
49 #endif
50
51 #include <obd_support.h>
52 #include <lustre_lib.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_dlm.h>
56 #include <lustre_mds.h>
57 #include <obd_class.h>
58 #include <lprocfs_status.h>
59 #include <lustre_param.h>
60 #include <cl_object.h>
61 #include <lclient.h>
62 #include <lustre/ll_fiemap.h>
63 #include <lustre_fid.h>
64
65 #include "lov_internal.h"
66
67 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
68    Any function that expects lov_tgts to remain stationary must take a ref. */
69 static void lov_getref(struct obd_device *obd)
70 {
71         struct lov_obd *lov = &obd->u.lov;
72
73         /* nobody gets through here until lov_putref is done */
74         mutex_lock(&lov->lov_lock);
75         cfs_atomic_inc(&lov->lov_refcount);
76         mutex_unlock(&lov->lov_lock);
77         return;
78 }
79
80 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
81
82 static void lov_putref(struct obd_device *obd)
83 {
84         struct lov_obd *lov = &obd->u.lov;
85
86         mutex_lock(&lov->lov_lock);
87         /* ok to dec to 0 more than once -- ltd_exp's will be null */
88         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
89                 CFS_LIST_HEAD(kill);
90                 int i;
91                 struct lov_tgt_desc *tgt, *n;
92                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
93                        lov->lov_death_row);
94                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
95                         tgt = lov->lov_tgts[i];
96
97                         if (!tgt || !tgt->ltd_reap)
98                                 continue;
99                         cfs_list_add(&tgt->ltd_kill, &kill);
100                         /* XXX - right now there is a dependency on ld_tgt_count
101                          * being the maximum tgt index for computing the
102                          * mds_max_easize. So we can't shrink it. */
103                         lov_ost_pool_remove(&lov->lov_packed, i);
104                         lov->lov_tgts[i] = NULL;
105                         lov->lov_death_row--;
106                 }
107                 mutex_unlock(&lov->lov_lock);
108
109                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
110                         cfs_list_del(&tgt->ltd_kill);
111                         /* Disconnect */
112                         __lov_del_obd(obd, tgt);
113                 }
114         } else {
115                 mutex_unlock(&lov->lov_lock);
116         }
117 }
118
119 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
120                               enum obd_notify_event ev);
121 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
122                       enum obd_notify_event ev, void *data);
123
124 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
125                     struct obd_connect_data *data)
126 {
127         struct lov_obd *lov = &obd->u.lov;
128         struct obd_uuid *tgt_uuid;
129         struct obd_device *tgt_obd;
130         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
131         struct obd_import *imp;
132 #ifdef __KERNEL__
133         struct proc_dir_entry *lov_proc_dir;
134 #endif
135         int rc;
136         ENTRY;
137
138         if (!lov->lov_tgts[index])
139                 RETURN(-EINVAL);
140
141         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
142         tgt_obd = lov->lov_tgts[index]->ltd_obd;
143
144         if (!tgt_obd->obd_set_up) {
145                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
146                 RETURN(-EINVAL);
147         }
148
149         /* override the sp_me from lov */
150         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
151
152         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
153                 data->ocd_index = index;
154
155         /*
156          * Divine LOV knows that OBDs under it are OSCs.
157          */
158         imp = tgt_obd->u.cli.cl_import;
159
160         if (activate) {
161                 tgt_obd->obd_no_recov = 0;
162                 /* FIXME this is probably supposed to be
163                    ptlrpc_set_import_active.  Horrible naming. */
164                 ptlrpc_activate_import(imp);
165         }
166
167         rc = obd_register_observer(tgt_obd, obd);
168         if (rc) {
169                 CERROR("Target %s register_observer error %d\n",
170                        obd_uuid2str(tgt_uuid), rc);
171                 RETURN(rc);
172         }
173
174
175         if (imp->imp_invalid) {
176                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
177                        "disabled\n", obd_uuid2str(tgt_uuid));
178                 RETURN(0);
179         }
180
181         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
182                          &lov_osc_uuid, data, NULL);
183         if (rc || !lov->lov_tgts[index]->ltd_exp) {
184                 CERROR("Target %s connect error %d\n",
185                        obd_uuid2str(tgt_uuid), rc);
186                 RETURN(-ENODEV);
187         }
188
189         lov->lov_tgts[index]->ltd_reap = 0;
190
191         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
192                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
193
194 #ifdef __KERNEL__
195         lov_proc_dir = obd->obd_proc_private;
196         if (lov_proc_dir) {
197                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
198                 struct proc_dir_entry *osc_symlink;
199
200                 LASSERT(osc_obd != NULL);
201                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
202                 LASSERT(osc_obd->obd_type->typ_name != NULL);
203
204                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
205                                                   lov_proc_dir,
206                                                   "../../../%s/%s",
207                                                   osc_obd->obd_type->typ_name,
208                                                   osc_obd->obd_name);
209                 if (osc_symlink == NULL) {
210                         CERROR("could not register LOV target "
211                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
212                                 obd->obd_type->typ_name, obd->obd_name,
213                                 osc_obd->obd_name);
214                         lprocfs_remove(&lov_proc_dir);
215                         obd->obd_proc_private = NULL;
216                 }
217         }
218 #endif
219
220         RETURN(0);
221 }
222
223 static int lov_connect(const struct lu_env *env,
224                        struct obd_export **exp, struct obd_device *obd,
225                        struct obd_uuid *cluuid, struct obd_connect_data *data,
226                        void *localdata)
227 {
228         struct lov_obd *lov = &obd->u.lov;
229         struct lov_tgt_desc *tgt;
230         struct lustre_handle conn;
231         int i, rc;
232         ENTRY;
233
234         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
235
236         rc = class_connect(&conn, obd, cluuid);
237         if (rc)
238                 RETURN(rc);
239
240         *exp = class_conn2export(&conn);
241
242         /* Why should there ever be more than 1 connect? */
243         lov->lov_connects++;
244         LASSERT(lov->lov_connects == 1);
245
246         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
247         if (data)
248                 lov->lov_ocd = *data;
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 proc_dir_entry *lov_proc_dir;
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         lov_proc_dir = obd->obd_proc_private;
297         if (lov_proc_dir)
298                 lprocfs_remove_proc_entry(osc_obd->obd_name, lov_proc_dir);
299
300         if (osc_obd) {
301                 /* Pass it on to our clients.
302                  * XXX This should be an argument to disconnect,
303                  * XXX not a back-door flag on the OBD.  Ah well.
304                  */
305                 osc_obd->obd_force = obd->obd_force;
306                 osc_obd->obd_fail = obd->obd_fail;
307                 osc_obd->obd_no_recov = obd->obd_no_recov;
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 out:
354         rc = class_disconnect(exp); /* bz 9811 */
355         RETURN(rc);
356 }
357
358 /* Error codes:
359  *
360  *  -EINVAL  : UUID can't be found in the LOV's target list
361  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
362  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
363  *  any >= 0 : is log target index
364  */
365 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
366                               enum obd_notify_event ev)
367 {
368         struct lov_obd *lov = &obd->u.lov;
369         struct lov_tgt_desc *tgt;
370         int index, activate, active;
371         ENTRY;
372
373         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
374                lov, uuid->uuid, ev);
375
376         obd_getref(obd);
377         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
378                 tgt = lov->lov_tgts[index];
379                 if (!tgt)
380                         continue;
381                 /*
382                  * LU-642, initially inactive OSC could miss the obd_connect,
383                  * we make up for it here.
384                  */
385                 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
386                     obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
387                         struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
388
389                         obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
390                                     &lov_osc_uuid, &lov->lov_ocd, NULL);
391                 }
392                 if (!tgt->ltd_exp)
393                         continue;
394
395                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
396                        index, obd_uuid2str(&tgt->ltd_uuid),
397                        tgt->ltd_exp->exp_handle.h_cookie);
398                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
399                         break;
400         }
401
402         if (index == lov->desc.ld_tgt_count)
403                 GOTO(out, index = -EINVAL);
404
405         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
406                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
407
408                 if (lov->lov_tgts[index]->ltd_activate == activate) {
409                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
410                                uuid->uuid, activate ? "" : "de");
411                 } else {
412                         lov->lov_tgts[index]->ltd_activate = activate;
413                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
414                                activate ? "" : "de", obd_uuid2str(uuid));
415                 }
416
417         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
418                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
419
420                 if (lov->lov_tgts[index]->ltd_active == active) {
421                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
422                                uuid->uuid, active ? "" : "in");
423                         GOTO(out, index);
424                 } else {
425                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
426                                obd_uuid2str(uuid), active ? "" : "in");
427                 }
428
429                 lov->lov_tgts[index]->ltd_active = active;
430                 if (active) {
431                         lov->desc.ld_active_tgt_count++;
432                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
433                 } else {
434                         lov->desc.ld_active_tgt_count--;
435                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
436                 }
437         } else {
438                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
439         }
440
441  out:
442         obd_putref(obd);
443         RETURN(index);
444 }
445
446 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
447                       enum obd_notify_event ev, void *data)
448 {
449         int rc = 0;
450         struct lov_obd *lov = &obd->u.lov;
451         ENTRY;
452
453         down_read(&lov->lov_notify_lock);
454         if (!lov->lov_connects) {
455                 up_read(&lov->lov_notify_lock);
456                 RETURN(rc);
457         }
458
459         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
460             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
461                 struct obd_uuid *uuid;
462
463                 LASSERT(watched);
464
465                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
466                         up_read(&lov->lov_notify_lock);
467                         CERROR("unexpected notification of %s %s!\n",
468                                watched->obd_type->typ_name,
469                                watched->obd_name);
470                         RETURN(-EINVAL);
471                 }
472                 uuid = &watched->u.cli.cl_target_uuid;
473
474                 /* Set OSC as active before notifying the observer, so the
475                  * observer can use the OSC normally.
476                  */
477                 rc = lov_set_osc_active(obd, uuid, ev);
478                 if (rc < 0) {
479                         up_read(&lov->lov_notify_lock);
480                         CERROR("event(%d) of %s failed: %d\n", ev,
481                                obd_uuid2str(uuid), rc);
482                         RETURN(rc);
483                 }
484                 /* active event should be pass lov target index as data */
485                 data = &rc;
486         }
487
488         /* Pass the notification up the chain. */
489         if (watched) {
490                 rc = obd_notify_observer(obd, watched, ev, data);
491         } else {
492                 /* NULL watched means all osc's in the lov (only for syncs) */
493                 /* sync event should be send lov idx as data */
494                 struct lov_obd *lov = &obd->u.lov;
495                 int i, is_sync;
496
497                 data = &i;
498                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
499                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
500
501                 obd_getref(obd);
502                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
503                         if (!lov->lov_tgts[i])
504                                 continue;
505
506                         /* don't send sync event if target not
507                          * connected/activated */
508                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
509                                 continue;
510
511                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
512                                                  ev, data);
513                         if (rc) {
514                                 CERROR("%s: notify %s of %s failed %d\n",
515                                        obd->obd_name,
516                                        obd->obd_observer->obd_name,
517                                        lov->lov_tgts[i]->ltd_obd->obd_name,
518                                        rc);
519                         }
520                 }
521                 obd_putref(obd);
522         }
523
524         up_read(&lov->lov_notify_lock);
525         RETURN(rc);
526 }
527
528 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
529                           __u32 index, int gen, int active)
530 {
531         struct lov_obd *lov = &obd->u.lov;
532         struct lov_tgt_desc *tgt;
533         struct obd_device *tgt_obd;
534         int rc;
535         ENTRY;
536
537         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
538                uuidp->uuid, index, gen, active);
539
540         if (gen <= 0) {
541                 CERROR("request to add OBD %s with invalid generation: %d\n",
542                        uuidp->uuid, gen);
543                 RETURN(-EINVAL);
544         }
545
546         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
547                                         &obd->obd_uuid);
548         if (tgt_obd == NULL)
549                 RETURN(-EINVAL);
550
551         mutex_lock(&lov->lov_lock);
552
553         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
554                 tgt = lov->lov_tgts[index];
555                 CERROR("UUID %s already assigned at LOV target index %d\n",
556                        obd_uuid2str(&tgt->ltd_uuid), index);
557                 mutex_unlock(&lov->lov_lock);
558                 RETURN(-EEXIST);
559         }
560
561         if (index >= lov->lov_tgt_size) {
562                 /* We need to reallocate the lov target array. */
563                 struct lov_tgt_desc **newtgts, **old = NULL;
564                 __u32 newsize, oldsize = 0;
565
566                 newsize = max(lov->lov_tgt_size, (__u32)2);
567                 while (newsize < index + 1)
568                         newsize = newsize << 1;
569                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
570                 if (newtgts == NULL) {
571                         mutex_unlock(&lov->lov_lock);
572                         RETURN(-ENOMEM);
573                 }
574
575                 if (lov->lov_tgt_size) {
576                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
577                                lov->lov_tgt_size);
578                         old = lov->lov_tgts;
579                         oldsize = lov->lov_tgt_size;
580                 }
581
582                 lov->lov_tgts = newtgts;
583                 lov->lov_tgt_size = newsize;
584                 smp_rmb();
585                 if (old)
586                         OBD_FREE(old, sizeof(*old) * oldsize);
587
588                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
589                        lov->lov_tgts, lov->lov_tgt_size);
590         }
591
592         OBD_ALLOC_PTR(tgt);
593         if (!tgt) {
594                 mutex_unlock(&lov->lov_lock);
595                 RETURN(-ENOMEM);
596         }
597
598         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
599         if (rc) {
600                 mutex_unlock(&lov->lov_lock);
601                 OBD_FREE_PTR(tgt);
602                 RETURN(rc);
603         }
604
605         tgt->ltd_uuid = *uuidp;
606         tgt->ltd_obd = tgt_obd;
607         /* XXX - add a sanity check on the generation number. */
608         tgt->ltd_gen = gen;
609         tgt->ltd_index = index;
610         tgt->ltd_activate = active;
611         lov->lov_tgts[index] = tgt;
612         if (index >= lov->desc.ld_tgt_count)
613                 lov->desc.ld_tgt_count = index + 1;
614
615         mutex_unlock(&lov->lov_lock);
616
617         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
618                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
619
620         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
621
622         if (lov->lov_connects == 0) {
623                 /* lov_connect hasn't been called yet. We'll do the
624                    lov_connect_obd on this target when that fn first runs,
625                    because we don't know the connect flags yet. */
626                 RETURN(0);
627         }
628
629         obd_getref(obd);
630
631         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
632         if (rc)
633                 GOTO(out, rc);
634
635         /* connect to administrative disabled ost */
636         if (!tgt->ltd_exp)
637                 GOTO(out, rc = 0);
638
639         if (lov->lov_cache != NULL) {
640                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
641                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
642                                 sizeof(struct cl_client_cache), lov->lov_cache,
643                                 NULL);
644                 if (rc < 0)
645                         GOTO(out, rc);
646         }
647
648         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
649                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
650                         (void *)&index);
651
652 out:
653         if (rc) {
654                 CERROR("add failed (%d), deleting %s\n", rc,
655                        obd_uuid2str(&tgt->ltd_uuid));
656                 lov_del_target(obd, index, 0, 0);
657         }
658         obd_putref(obd);
659         RETURN(rc);
660 }
661
662 /* Schedule a target for deletion */
663 int lov_del_target(struct obd_device *obd, __u32 index,
664                    struct obd_uuid *uuidp, int gen)
665 {
666         struct lov_obd *lov = &obd->u.lov;
667         int count = lov->desc.ld_tgt_count;
668         int rc = 0;
669         ENTRY;
670
671         if (index >= count) {
672                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
673                        index, count);
674                 RETURN(-EINVAL);
675         }
676
677         /* to make sure there's no ongoing lov_notify() now */
678         down_write(&lov->lov_notify_lock);
679         obd_getref(obd);
680
681         if (!lov->lov_tgts[index]) {
682                 CERROR("LOV target at index %d is not setup.\n", index);
683                 GOTO(out, rc = -EINVAL);
684         }
685
686         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
687                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
688                        lov_uuid2str(lov, index), index,
689                        obd_uuid2str(uuidp));
690                 GOTO(out, rc = -EINVAL);
691         }
692
693         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
694                lov_uuid2str(lov, index), index,
695                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
696                lov->lov_tgts[index]->ltd_active);
697
698         lov->lov_tgts[index]->ltd_reap = 1;
699         lov->lov_death_row++;
700         /* we really delete it from obd_putref */
701 out:
702         obd_putref(obd);
703         up_write(&lov->lov_notify_lock);
704
705         RETURN(rc);
706 }
707
708 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
709 {
710         struct obd_device *osc_obd;
711
712         LASSERT(tgt);
713         LASSERT(tgt->ltd_reap);
714
715         osc_obd = class_exp2obd(tgt->ltd_exp);
716
717         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
718                tgt->ltd_uuid.uuid,
719                osc_obd ? osc_obd->obd_name : "<no obd>");
720
721         if (tgt->ltd_exp)
722                 lov_disconnect_obd(obd, tgt);
723
724         OBD_FREE_PTR(tgt);
725
726         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
727            do it ourselves. And we can't do it from lov_cleanup,
728            because we just lost our only reference to it. */
729         if (osc_obd)
730                 class_manual_cleanup(osc_obd);
731 }
732
733 void lov_fix_desc_stripe_size(__u64 *val)
734 {
735         if (*val < LOV_MIN_STRIPE_SIZE) {
736                 if (*val != 0)
737                         LCONSOLE_INFO("Increasing default stripe size to "
738                                       "minimum %u\n",
739                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
740                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
741         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
742                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
743                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
744                               "multiple of %u)\n",
745                               *val, LOV_MIN_STRIPE_SIZE);
746         }
747 }
748
749 void lov_fix_desc_stripe_count(__u32 *val)
750 {
751         if (*val == 0)
752                 *val = 1;
753 }
754
755 void lov_fix_desc_pattern(__u32 *val)
756 {
757         /* from lov_setstripe */
758         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
759                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
760                 *val = 0;
761         }
762 }
763
764 void lov_fix_desc_qos_maxage(__u32 *val)
765 {
766         if (*val == 0)
767                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
768 }
769
770 void lov_fix_desc(struct lov_desc *desc)
771 {
772         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
773         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
774         lov_fix_desc_pattern(&desc->ld_pattern);
775         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
776 }
777
778 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
779 {
780         struct lov_desc *desc;
781         struct lov_obd *lov = &obd->u.lov;
782         int rc;
783         ENTRY;
784
785         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
786                 CERROR("LOV setup requires a descriptor\n");
787                 RETURN(-EINVAL);
788         }
789
790         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
791
792         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
793                 CERROR("descriptor size wrong: %d > %d\n",
794                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
795                 RETURN(-EINVAL);
796         }
797
798         if (desc->ld_magic != LOV_DESC_MAGIC) {
799                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
800                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
801                                    obd->obd_name, desc);
802                             lustre_swab_lov_desc(desc);
803                 } else {
804                         CERROR("%s: Bad lov desc magic: %#x\n",
805                                obd->obd_name, desc->ld_magic);
806                         RETURN(-EINVAL);
807                 }
808         }
809
810         lov_fix_desc(desc);
811
812         desc->ld_active_tgt_count = 0;
813         lov->desc = *desc;
814         lov->lov_tgt_size = 0;
815
816         mutex_init(&lov->lov_lock);
817         cfs_atomic_set(&lov->lov_refcount, 0);
818         lov->lov_sp_me = LUSTRE_SP_CLI;
819
820         init_rwsem(&lov->lov_notify_lock);
821
822         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
823                                                    HASH_POOLS_MAX_BITS,
824                                                    HASH_POOLS_BKT_BITS, 0,
825                                                    CFS_HASH_MIN_THETA,
826                                                    CFS_HASH_MAX_THETA,
827                                                    &pool_hash_operations,
828                                                    CFS_HASH_DEFAULT);
829         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
830         lov->lov_pool_count = 0;
831         rc = lov_ost_pool_init(&lov->lov_packed, 0);
832         if (rc)
833                 GOTO(out, rc);
834
835 #ifdef LPROCFS
836         obd->obd_vars = lprocfs_lov_obd_vars;
837         lprocfs_seq_obd_setup(obd);
838         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", 0444,
839                                 &lov_proc_target_fops, obd);
840         if (rc)
841                 CWARN("Error adding the target_obd file\n");
842
843         lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
844                                                         obd->obd_proc_entry,
845                                                         NULL, NULL);
846 #endif
847         RETURN(0);
848
849 out:
850         return rc;
851 }
852
853 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
854 {
855         int rc = 0;
856         struct lov_obd *lov = &obd->u.lov;
857
858         ENTRY;
859
860         switch (stage) {
861         case OBD_CLEANUP_EARLY: {
862                 int i;
863                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
864                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
865                                 continue;
866                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
867                                        OBD_CLEANUP_EARLY);
868                 }
869                 break;
870         }
871         default:
872                 break;
873         }
874
875         RETURN(rc);
876 }
877
878 static int lov_cleanup(struct obd_device *obd)
879 {
880         struct lov_obd *lov = &obd->u.lov;
881         cfs_list_t *pos, *tmp;
882         struct pool_desc *pool;
883         ENTRY;
884
885         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
886                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
887                 /* free pool structs */
888                 CDEBUG(D_INFO, "delete pool %p\n", pool);
889                 /* In the function below, .hs_keycmp resolves to
890                  * pool_hashkey_keycmp() */
891                 /* coverity[overrun-buffer-val] */
892                 lov_pool_del(obd, pool->pool_name);
893         }
894         cfs_hash_putref(lov->lov_pools_hash_body);
895         lov_ost_pool_free(&lov->lov_packed);
896
897         lprocfs_obd_cleanup(obd);
898         if (lov->lov_tgts) {
899                 int i;
900                 obd_getref(obd);
901                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
902                         if (!lov->lov_tgts[i])
903                                 continue;
904
905                         /* Inactive targets may never have connected */
906                         if (lov->lov_tgts[i]->ltd_active ||
907                             cfs_atomic_read(&lov->lov_refcount))
908                             /* We should never get here - these
909                                should have been removed in the
910                              disconnect. */
911                                 CERROR("lov tgt %d not cleaned!"
912                                        " deathrow=%d, lovrc=%d\n",
913                                        i, lov->lov_death_row,
914                                        cfs_atomic_read(&lov->lov_refcount));
915                         lov_del_target(obd, i, 0, 0);
916                 }
917                 obd_putref(obd);
918                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
919                          lov->lov_tgt_size);
920                 lov->lov_tgt_size = 0;
921         }
922         RETURN(0);
923 }
924
925 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
926                             __u32 *indexp, int *genp)
927 {
928         struct obd_uuid obd_uuid;
929         int cmd;
930         int rc = 0;
931         ENTRY;
932
933         switch(cmd = lcfg->lcfg_command) {
934         case LCFG_LOV_ADD_OBD:
935         case LCFG_LOV_ADD_INA:
936         case LCFG_LOV_DEL_OBD: {
937                 __u32 index;
938                 int gen;
939                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
940                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
941                         GOTO(out, rc = -EINVAL);
942
943                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
944
945                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
946                         GOTO(out, rc = -EINVAL);
947                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
948                         GOTO(out, rc = -EINVAL);
949                 index = *indexp;
950                 gen = *genp;
951                 if (cmd == LCFG_LOV_ADD_OBD)
952                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
953                 else if (cmd == LCFG_LOV_ADD_INA)
954                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
955                 else
956                         rc = lov_del_target(obd, index, &obd_uuid, gen);
957                 GOTO(out, rc);
958         }
959         case LCFG_PARAM: {
960                 struct lov_desc *desc = &(obd->u.lov.desc);
961
962                 if (!desc)
963                         GOTO(out, rc = -EINVAL);
964
965                 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
966                                                   lcfg, obd);
967                 if (rc > 0)
968                         rc = 0;
969                 GOTO(out, rc);
970         }
971         case LCFG_POOL_NEW:
972         case LCFG_POOL_ADD:
973         case LCFG_POOL_DEL:
974         case LCFG_POOL_REM:
975                 GOTO(out, rc);
976
977         default: {
978                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
979                 GOTO(out, rc = -EINVAL);
980
981         }
982         }
983 out:
984         RETURN(rc);
985 }
986
987 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
988                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
989 {
990         struct lov_stripe_md *obj_mdp, *lsm;
991         struct lov_obd *lov = &exp->exp_obd->u.lov;
992         unsigned ost_idx;
993         int rc, i;
994         ENTRY;
995
996         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
997                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
998
999         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1000         if (obj_mdp == NULL)
1001                 RETURN(-ENOMEM);
1002
1003         ost_idx = src_oa->o_nlink;
1004         lsm = *ea;
1005         if (lsm == NULL)
1006                 GOTO(out, rc = -EINVAL);
1007         if (ost_idx >= lov->desc.ld_tgt_count ||
1008             !lov->lov_tgts[ost_idx])
1009                 GOTO(out, rc = -EINVAL);
1010
1011         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1012                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1013                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1014                                         ostid_id(&src_oa->o_oi))
1015                                 GOTO(out, rc = -EINVAL);
1016                         break;
1017                 }
1018         }
1019         if (i == lsm->lsm_stripe_count)
1020                 GOTO(out, rc = -EINVAL);
1021
1022         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1023                         src_oa, &obj_mdp, oti);
1024 out:
1025         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1026         RETURN(rc);
1027 }
1028
1029 /* the LOV expects oa->o_id to be set to the LOV object id */
1030 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1031                       struct obdo *src_oa, struct lov_stripe_md **ea,
1032                       struct obd_trans_info *oti)
1033 {
1034         struct lov_obd *lov;
1035         int rc = 0;
1036         ENTRY;
1037
1038         LASSERT(ea != NULL);
1039         if (exp == NULL)
1040                 RETURN(-EINVAL);
1041
1042         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1043             src_oa->o_flags == OBD_FL_DELORPHAN) {
1044                 /* should be used with LOV anymore */
1045                 LBUG();
1046         }
1047
1048         lov = &exp->exp_obd->u.lov;
1049         if (!lov->desc.ld_active_tgt_count)
1050                 RETURN(-EIO);
1051
1052         obd_getref(exp->exp_obd);
1053         /* Recreate a specific object id at the given OST index */
1054         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1055             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1056                  rc = lov_recreate(exp, src_oa, ea, oti);
1057         }
1058
1059         obd_putref(exp->exp_obd);
1060         RETURN(rc);
1061 }
1062
1063 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1064 do {                                                                            \
1065         LASSERT((lsmp) != NULL);                                                \
1066         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1067                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1068                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1069 } while (0)
1070
1071 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1072                        struct obdo *oa, struct lov_stripe_md *lsm,
1073                        struct obd_trans_info *oti, struct obd_export *md_exp,
1074                        void *capa)
1075 {
1076         struct lov_request_set *set;
1077         struct obd_info oinfo;
1078         struct lov_request *req;
1079         cfs_list_t *pos;
1080         struct lov_obd *lov;
1081         int rc = 0, err = 0;
1082         ENTRY;
1083
1084         ASSERT_LSM_MAGIC(lsm);
1085
1086         if (!exp || !exp->exp_obd)
1087                 RETURN(-ENODEV);
1088
1089         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1090                 LASSERT(oti);
1091                 LASSERT(oti->oti_logcookies);
1092         }
1093
1094         lov = &exp->exp_obd->u.lov;
1095         obd_getref(exp->exp_obd);
1096         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1097         if (rc)
1098                 GOTO(out, rc);
1099
1100         cfs_list_for_each (pos, &set->set_list) {
1101                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1102
1103                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1104                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1105
1106                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1107                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1108                 err = lov_update_common_set(set, req, err);
1109                 if (err) {
1110                         CERROR("%s: destroying objid "DOSTID" subobj "
1111                                DOSTID" on OST idx %d: rc = %d\n",
1112                                exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1113                                POSTID(&req->rq_oi.oi_oa->o_oi),
1114                                req->rq_idx, err);
1115                         if (!rc)
1116                                 rc = err;
1117                 }
1118         }
1119
1120         if (rc == 0) {
1121                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1122                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1123         }
1124         err = lov_fini_destroy_set(set);
1125 out:
1126         obd_putref(exp->exp_obd);
1127         RETURN(rc ? rc : err);
1128 }
1129
1130 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1131                                  void *data, int rc)
1132 {
1133         struct lov_request_set *lovset = (struct lov_request_set *)data;
1134         int err;
1135         ENTRY;
1136
1137         /* don't do attribute merge if this aysnc op failed */
1138         if (rc)
1139                 cfs_atomic_set(&lovset->set_completes, 0);
1140         err = lov_fini_getattr_set(lovset);
1141         RETURN(rc ? rc : err);
1142 }
1143
1144 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1145                               struct ptlrpc_request_set *rqset)
1146 {
1147         struct lov_request_set *lovset;
1148         struct lov_obd *lov;
1149         cfs_list_t *pos;
1150         struct lov_request *req;
1151         int rc = 0, err;
1152         ENTRY;
1153
1154         LASSERT(oinfo);
1155         ASSERT_LSM_MAGIC(oinfo->oi_md);
1156
1157         if (!exp || !exp->exp_obd)
1158                 RETURN(-ENODEV);
1159
1160         lov = &exp->exp_obd->u.lov;
1161
1162         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1163         if (rc)
1164                 RETURN(rc);
1165
1166         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1167                POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1168                oinfo->oi_md->lsm_stripe_size);
1169
1170         cfs_list_for_each(pos, &lovset->set_list) {
1171                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1172
1173                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1174                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1175                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1176                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1177                                        &req->rq_oi, rqset);
1178                 if (rc) {
1179                         CERROR("%s: getattr objid "DOSTID" subobj"
1180                                DOSTID" on OST idx %d: rc = %d\n",
1181                                exp->exp_obd->obd_name,
1182                                POSTID(&oinfo->oi_oa->o_oi),
1183                                POSTID(&req->rq_oi.oi_oa->o_oi),
1184                                req->rq_idx, rc);
1185                         GOTO(out, rc);
1186                 }
1187         }
1188
1189         if (!cfs_list_empty(&rqset->set_requests)) {
1190                 LASSERT(rc == 0);
1191                 LASSERT (rqset->set_interpret == NULL);
1192                 rqset->set_interpret = lov_getattr_interpret;
1193                 rqset->set_arg = (void *)lovset;
1194                 RETURN(rc);
1195         }
1196 out:
1197         if (rc)
1198                 cfs_atomic_set(&lovset->set_completes, 0);
1199         err = lov_fini_getattr_set(lovset);
1200         RETURN(rc ? rc : err);
1201 }
1202
1203 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1204                                  void *data, int rc)
1205 {
1206         struct lov_request_set *lovset = (struct lov_request_set *)data;
1207         int err;
1208         ENTRY;
1209
1210         if (rc)
1211                 cfs_atomic_set(&lovset->set_completes, 0);
1212         err = lov_fini_setattr_set(lovset);
1213         RETURN(rc ? rc : err);
1214 }
1215
1216 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1217    needed. Otherwise, a client is waiting for responses. */
1218 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1219                              struct obd_trans_info *oti,
1220                              struct ptlrpc_request_set *rqset)
1221 {
1222         struct lov_request_set *set;
1223         struct lov_request *req;
1224         cfs_list_t *pos;
1225         struct lov_obd *lov;
1226         int rc = 0;
1227         ENTRY;
1228
1229         LASSERT(oinfo);
1230         ASSERT_LSM_MAGIC(oinfo->oi_md);
1231         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1232                 LASSERT(oti);
1233                 LASSERT(oti->oti_logcookies);
1234         }
1235
1236         if (!exp || !exp->exp_obd)
1237                 RETURN(-ENODEV);
1238
1239         lov = &exp->exp_obd->u.lov;
1240         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1241         if (rc)
1242                 RETURN(rc);
1243
1244         CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1245                POSTID(&oinfo->oi_md->lsm_oi),
1246                oinfo->oi_md->lsm_stripe_count,
1247                oinfo->oi_md->lsm_stripe_size);
1248
1249         cfs_list_for_each(pos, &set->set_list) {
1250                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1251
1252                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1253                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1254
1255                 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1256                        "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1257                        POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1258
1259                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1260                                        &req->rq_oi, oti, rqset);
1261                 if (rc) {
1262                         CERROR("error: setattr objid "DOSTID" subobj"
1263                                DOSTID" on OST idx %d: rc = %d\n",
1264                                POSTID(&set->set_oi->oi_oa->o_oi),
1265                                POSTID(&req->rq_oi.oi_oa->o_oi),
1266                                req->rq_idx, rc);
1267                         break;
1268                 }
1269         }
1270
1271         /* If we are not waiting for responses on async requests, return. */
1272         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1273                 int err;
1274                 if (rc)
1275                         cfs_atomic_set(&set->set_completes, 0);
1276                 err = lov_fini_setattr_set(set);
1277                 RETURN(rc ? rc : err);
1278         }
1279
1280         LASSERT(rqset->set_interpret == NULL);
1281         rqset->set_interpret = lov_setattr_interpret;
1282         rqset->set_arg = (void *)set;
1283
1284         RETURN(0);
1285 }
1286
1287 static int lov_change_cbdata(struct obd_export *exp,
1288                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1289                              void *data)
1290 {
1291         struct lov_obd *lov;
1292         int rc = 0, i;
1293         ENTRY;
1294
1295         ASSERT_LSM_MAGIC(lsm);
1296
1297         if (!exp || !exp->exp_obd)
1298                 RETURN(-ENODEV);
1299
1300         lov = &exp->exp_obd->u.lov;
1301         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1302                 struct lov_stripe_md submd;
1303                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1304
1305                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1306                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1307                         continue;
1308                 }
1309
1310                 submd.lsm_oi = loi->loi_oi;
1311                 submd.lsm_stripe_count = 0;
1312                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1313                                        &submd, it, data);
1314         }
1315         RETURN(rc);
1316 }
1317
1318 /* find any ldlm lock of the inode in lov
1319  * return 0    not find
1320  *        1    find one
1321  *      < 0    error */
1322 static int lov_find_cbdata(struct obd_export *exp,
1323                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1324                            void *data)
1325 {
1326         struct lov_obd *lov;
1327         int rc = 0, i;
1328         ENTRY;
1329
1330         ASSERT_LSM_MAGIC(lsm);
1331
1332         if (!exp || !exp->exp_obd)
1333                 RETURN(-ENODEV);
1334
1335         lov = &exp->exp_obd->u.lov;
1336         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1337                 struct lov_stripe_md submd;
1338                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1339
1340                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1341                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1342                         continue;
1343                 }
1344                 submd.lsm_oi = loi->loi_oi;
1345                 submd.lsm_stripe_count = 0;
1346                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1347                                      &submd, it, data);
1348                 if (rc != 0)
1349                         RETURN(rc);
1350         }
1351         RETURN(rc);
1352 }
1353
1354 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1355 {
1356         struct lov_request_set *lovset = (struct lov_request_set *)data;
1357         int err;
1358         ENTRY;
1359
1360         if (rc)
1361                 cfs_atomic_set(&lovset->set_completes, 0);
1362
1363         err = lov_fini_statfs_set(lovset);
1364         RETURN(rc ? rc : err);
1365 }
1366
1367 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1368                             __u64 max_age, struct ptlrpc_request_set *rqset)
1369 {
1370         struct obd_device      *obd = class_exp2obd(exp);
1371         struct lov_request_set *set;
1372         struct lov_request *req;
1373         cfs_list_t *pos;
1374         struct lov_obd *lov;
1375         int rc = 0;
1376         ENTRY;
1377
1378         LASSERT(oinfo != NULL);
1379         LASSERT(oinfo->oi_osfs != NULL);
1380
1381         lov = &obd->u.lov;
1382         rc = lov_prep_statfs_set(obd, oinfo, &set);
1383         if (rc)
1384                 RETURN(rc);
1385
1386         cfs_list_for_each (pos, &set->set_list) {
1387                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1388                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1389                                       &req->rq_oi, max_age, rqset);
1390                 if (rc)
1391                         break;
1392         }
1393
1394         if (rc || cfs_list_empty(&rqset->set_requests)) {
1395                 int err;
1396                 if (rc)
1397                         cfs_atomic_set(&set->set_completes, 0);
1398                 err = lov_fini_statfs_set(set);
1399                 RETURN(rc ? rc : err);
1400         }
1401
1402         LASSERT(rqset->set_interpret == NULL);
1403         rqset->set_interpret = lov_statfs_interpret;
1404         rqset->set_arg = (void *)set;
1405         RETURN(0);
1406 }
1407
1408 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1409                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1410 {
1411         struct ptlrpc_request_set *set = NULL;
1412         struct obd_info oinfo = { { { 0 } } };
1413         int rc = 0;
1414         ENTRY;
1415
1416
1417         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1418          * statfs requests */
1419         set = ptlrpc_prep_set();
1420         if (set == NULL)
1421                 RETURN(-ENOMEM);
1422
1423         oinfo.oi_osfs = osfs;
1424         oinfo.oi_flags = flags;
1425         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1426         if (rc == 0)
1427                 rc = ptlrpc_set_wait(set);
1428         ptlrpc_set_destroy(set);
1429
1430         RETURN(rc);
1431 }
1432
1433 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1434                          void *karg, void *uarg)
1435 {
1436         struct obd_device *obddev = class_exp2obd(exp);
1437         struct lov_obd *lov = &obddev->u.lov;
1438         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1439         struct obd_uuid *uuidp;
1440         ENTRY;
1441
1442         switch (cmd) {
1443         case IOC_OBD_STATFS: {
1444                 struct obd_ioctl_data *data = karg;
1445                 struct obd_device *osc_obd;
1446                 struct obd_statfs stat_buf = {0};
1447                 __u32 index;
1448                 __u32 flags;
1449
1450                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1451                 if ((index >= count))
1452                         RETURN(-ENODEV);
1453
1454                 if (!lov->lov_tgts[index])
1455                         /* Try again with the next index */
1456                         RETURN(-EAGAIN);
1457                 if (!lov->lov_tgts[index]->ltd_active)
1458                         RETURN(-ENODATA);
1459
1460                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1461                 if (!osc_obd)
1462                         RETURN(-EINVAL);
1463
1464                 /* copy UUID */
1465                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1466                                  min((int)data->ioc_plen2,
1467                                      (int)sizeof(struct obd_uuid))))
1468                         RETURN(-EFAULT);
1469
1470                 flags = uarg ? *(__u32*)uarg : 0;
1471                 /* got statfs data */
1472                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1473                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1474                                 flags);
1475                 if (rc)
1476                         RETURN(rc);
1477                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1478                                      min((int) data->ioc_plen1,
1479                                          (int) sizeof(stat_buf))))
1480                         RETURN(-EFAULT);
1481                 break;
1482         }
1483         case OBD_IOC_LOV_GET_CONFIG: {
1484                 struct obd_ioctl_data *data;
1485                 struct lov_desc *desc;
1486                 char *buf = NULL;
1487                 __u32 *genp;
1488
1489                 len = 0;
1490                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1491                         RETURN(-EINVAL);
1492
1493                 data = (struct obd_ioctl_data *)buf;
1494
1495                 if (sizeof(*desc) > data->ioc_inllen1) {
1496                         obd_ioctl_freedata(buf, len);
1497                         RETURN(-EINVAL);
1498                 }
1499
1500                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1501                         obd_ioctl_freedata(buf, len);
1502                         RETURN(-EINVAL);
1503                 }
1504
1505                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1506                         obd_ioctl_freedata(buf, len);
1507                         RETURN(-EINVAL);
1508                 }
1509
1510                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1511                 memcpy(desc, &(lov->desc), sizeof(*desc));
1512
1513                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1514                 genp = (__u32 *)data->ioc_inlbuf3;
1515                 /* the uuid will be empty for deleted OSTs */
1516                 for (i = 0; i < count; i++, uuidp++, genp++) {
1517                         if (!lov->lov_tgts[i])
1518                                 continue;
1519                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1520                         *genp = lov->lov_tgts[i]->ltd_gen;
1521                 }
1522
1523                 if (copy_to_user((void *)uarg, buf, len))
1524                         rc = -EFAULT;
1525                 obd_ioctl_freedata(buf, len);
1526                 break;
1527         }
1528         case LL_IOC_LOV_GETSTRIPE:
1529                 rc = lov_getstripe(exp, karg, uarg);
1530                 break;
1531         case OBD_IOC_QUOTACTL: {
1532                 struct if_quotactl *qctl = karg;
1533                 struct lov_tgt_desc *tgt = NULL;
1534                 struct obd_quotactl *oqctl;
1535
1536                 if (qctl->qc_valid == QC_OSTIDX) {
1537                         if (count <= qctl->qc_idx)
1538                                 RETURN(-EINVAL);
1539
1540                         tgt = lov->lov_tgts[qctl->qc_idx];
1541                         if (!tgt || !tgt->ltd_exp)
1542                                 RETURN(-EINVAL);
1543                 } else if (qctl->qc_valid == QC_UUID) {
1544                         for (i = 0; i < count; i++) {
1545                                 tgt = lov->lov_tgts[i];
1546                                 if (!tgt ||
1547                                     !obd_uuid_equals(&tgt->ltd_uuid,
1548                                                      &qctl->obd_uuid))
1549                                         continue;
1550
1551                                 if (tgt->ltd_exp == NULL)
1552                                         RETURN(-EINVAL);
1553
1554                                 break;
1555                         }
1556                 } else {
1557                         RETURN(-EINVAL);
1558                 }
1559
1560                 if (i >= count)
1561                         RETURN(-EAGAIN);
1562
1563                 LASSERT(tgt && tgt->ltd_exp);
1564                 OBD_ALLOC_PTR(oqctl);
1565                 if (!oqctl)
1566                         RETURN(-ENOMEM);
1567
1568                 QCTL_COPY(oqctl, qctl);
1569                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1570                 if (rc == 0) {
1571                         QCTL_COPY(qctl, oqctl);
1572                         qctl->qc_valid = QC_OSTIDX;
1573                         qctl->obd_uuid = tgt->ltd_uuid;
1574                 }
1575                 OBD_FREE_PTR(oqctl);
1576                 break;
1577         }
1578         default: {
1579                 int set = 0;
1580
1581                 if (count == 0)
1582                         RETURN(-ENOTTY);
1583
1584                 for (i = 0; i < count; i++) {
1585                         int err;
1586                         struct obd_device *osc_obd;
1587
1588                         /* OST was disconnected */
1589                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1590                                 continue;
1591
1592                         /* ll_umount_begin() sets force flag but for lov, not
1593                          * osc. Let's pass it through */
1594                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1595                         osc_obd->obd_force = obddev->obd_force;
1596                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1597                                             len, karg, uarg);
1598                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1599                                 RETURN(err);
1600                         } else if (err) {
1601                                 if (lov->lov_tgts[i]->ltd_active) {
1602                                         CDEBUG(err == -ENOTTY ?
1603                                                D_IOCTL : D_WARNING,
1604                                                "iocontrol OSC %s on OST "
1605                                                "idx %d cmd %x: err = %d\n",
1606                                                lov_uuid2str(lov, i),
1607                                                i, cmd, err);
1608                                         if (!rc)
1609                                                 rc = err;
1610                                 }
1611                         } else {
1612                                 set = 1;
1613                         }
1614                 }
1615                 if (!set && !rc)
1616                         rc = -EIO;
1617         }
1618         }
1619
1620         RETURN(rc);
1621 }
1622
1623 #define FIEMAP_BUFFER_SIZE 4096
1624
1625 /**
1626  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1627  * call. The local end offset and the device are sent in the first
1628  * fm_extent. This function calculates the stripe number from the index.
1629  * This function returns a stripe_no on which mapping is to be restarted.
1630  *
1631  * This function returns fm_end_offset which is the in-OST offset at which
1632  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1633  * will re-calculate proper offset in next stripe.
1634  * Note that the first extent is passed to lov_get_info via the value field.
1635  *
1636  * \param fiemap fiemap request header
1637  * \param lsm striping information for the file
1638  * \param fm_start logical start of mapping
1639  * \param fm_end logical end of mapping
1640  * \param start_stripe starting stripe will be returned in this
1641  */
1642 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1643                                    struct lov_stripe_md *lsm, obd_size fm_start,
1644                                    obd_size fm_end, int *start_stripe)
1645 {
1646         obd_size local_end = fiemap->fm_extents[0].fe_logical;
1647         obd_off lun_start, lun_end;
1648         obd_size fm_end_offset;
1649         int stripe_no = -1, i;
1650
1651         if (fiemap->fm_extent_count == 0 ||
1652             fiemap->fm_extents[0].fe_logical == 0)
1653                 return 0;
1654
1655         /* Find out stripe_no from ost_index saved in the fe_device */
1656         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1657                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
1658                                         fiemap->fm_extents[0].fe_device) {
1659                         stripe_no = i;
1660                         break;
1661                 }
1662         }
1663         if (stripe_no == -1)
1664                 return -EINVAL;
1665
1666         /* If we have finished mapping on previous device, shift logical
1667          * offset to start of next device */
1668         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1669                                    &lun_start, &lun_end)) != 0 &&
1670                                    local_end < lun_end) {
1671                 fm_end_offset = local_end;
1672                 *start_stripe = stripe_no;
1673         } else {
1674                 /* This is a special value to indicate that caller should
1675                  * calculate offset in next stripe. */
1676                 fm_end_offset = 0;
1677                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1678         }
1679
1680         return fm_end_offset;
1681 }
1682
1683 /**
1684  * We calculate on which OST the mapping will end. If the length of mapping
1685  * is greater than (stripe_size * stripe_count) then the last_stripe will
1686  * will be one just before start_stripe. Else we check if the mapping
1687  * intersects each OST and find last_stripe.
1688  * This function returns the last_stripe and also sets the stripe_count
1689  * over which the mapping is spread
1690  *
1691  * \param lsm striping information for the file
1692  * \param fm_start logical start of mapping
1693  * \param fm_end logical end of mapping
1694  * \param start_stripe starting stripe of the mapping
1695  * \param stripe_count the number of stripes across which to map is returned
1696  *
1697  * \retval last_stripe return the last stripe of the mapping
1698  */
1699 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1700                             obd_size fm_end, int start_stripe,
1701                             int *stripe_count)
1702 {
1703         int last_stripe;
1704         obd_off obd_start, obd_end;
1705         int i, j;
1706
1707         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1708                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1709                                                               start_stripe - 1);
1710                 *stripe_count = lsm->lsm_stripe_count;
1711         } else {
1712                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1713                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
1714                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1715                                                    &obd_start, &obd_end)) == 0)
1716                                 break;
1717                 }
1718                 *stripe_count = j;
1719                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1720         }
1721
1722         return last_stripe;
1723 }
1724
1725 /**
1726  * Set fe_device and copy extents from local buffer into main return buffer.
1727  *
1728  * \param fiemap fiemap request header
1729  * \param lcl_fm_ext array of local fiemap extents to be copied
1730  * \param ost_index OST index to be written into the fm_device field for each
1731                     extent
1732  * \param ext_count number of extents to be copied
1733  * \param current_extent where to start copying in main extent array
1734  */
1735 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1736                                   struct ll_fiemap_extent *lcl_fm_ext,
1737                                   int ost_index, unsigned int ext_count,
1738                                   int current_extent)
1739 {
1740         char *to;
1741         int ext;
1742
1743         for (ext = 0; ext < ext_count; ext++) {
1744                 lcl_fm_ext[ext].fe_device = ost_index;
1745                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1746         }
1747
1748         /* Copy fm_extent's from fm_local to return buffer */
1749         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1750         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1751 }
1752
1753 /**
1754  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1755  * This also handles the restarting of FIEMAP calls in case mapping overflows
1756  * the available number of extents in single call.
1757  */
1758 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1759                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1760 {
1761         struct ll_fiemap_info_key *fm_key = key;
1762         struct ll_user_fiemap *fiemap = val;
1763         struct ll_user_fiemap *fm_local = NULL;
1764         struct ll_fiemap_extent *lcl_fm_ext;
1765         int count_local;
1766         unsigned int get_num_extents = 0;
1767         int ost_index = 0, actual_start_stripe, start_stripe;
1768         obd_size fm_start, fm_end, fm_length, fm_end_offset;
1769         obd_size curr_loc;
1770         int current_extent = 0, rc = 0, i;
1771         int ost_eof = 0; /* EOF for object */
1772         int ost_done = 0; /* done with required mapping for this OST? */
1773         int last_stripe;
1774         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1775         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1776
1777         if (!lsm_has_objects(lsm)) {
1778                 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1779                     fm_key->oa.o_size)) {
1780                         /* released file, return a minimal FIEMAP if
1781                          * request fits in file-size.
1782                          */
1783                         fiemap->fm_mapped_extents = 1;
1784                         fiemap->fm_extents[0].fe_logical =
1785                                                 fm_key->fiemap.fm_start;
1786                         if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1787                             fm_key->oa.o_size)
1788                                 fiemap->fm_extents[0].fe_length =
1789                                                 fm_key->fiemap.fm_length;
1790                         else
1791                                 fiemap->fm_extents[0].fe_length =
1792                                                 fm_key->oa.o_size -
1793                                                 fm_key->fiemap.fm_start;
1794                         fiemap->fm_extents[0].fe_flags |=
1795                                                 (FIEMAP_EXTENT_UNKNOWN |
1796                                                  FIEMAP_EXTENT_LAST);
1797                 }
1798                 GOTO(out, rc = 0);
1799         }
1800
1801         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1802                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1803
1804         OBD_ALLOC_LARGE(fm_local, buffer_size);
1805         if (fm_local == NULL)
1806                 GOTO(out, rc = -ENOMEM);
1807         lcl_fm_ext = &fm_local->fm_extents[0];
1808
1809         count_local = fiemap_size_to_count(buffer_size);
1810
1811         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1812         fm_start = fiemap->fm_start;
1813         fm_length = fiemap->fm_length;
1814         /* Calculate start stripe, last stripe and length of mapping */
1815         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1816         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1817                                                 fm_start + fm_length - 1);
1818         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1819         if (fm_end > fm_key->oa.o_size)
1820                 fm_end = fm_key->oa.o_size;
1821
1822         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1823                                             actual_start_stripe, &stripe_count);
1824
1825         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1826                                                   fm_end, &start_stripe);
1827         if (fm_end_offset == -EINVAL)
1828                 GOTO(out, rc = -EINVAL);
1829
1830         if (fiemap->fm_extent_count == 0) {
1831                 get_num_extents = 1;
1832                 count_local = 0;
1833         }
1834
1835         /* Check each stripe */
1836         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1837              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1838                 obd_size req_fm_len; /* Stores length of required mapping */
1839                 obd_size len_mapped_single_call;
1840                 obd_off lun_start, lun_end, obd_object_end;
1841                 unsigned int ext_count;
1842
1843                 cur_stripe_wrap = cur_stripe;
1844
1845                 /* Find out range of mapping on this stripe */
1846                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1847                                            &lun_start, &obd_object_end)) == 0)
1848                         continue;
1849
1850                 /* If this is a continuation FIEMAP call and we are on
1851                  * starting stripe then lun_start needs to be set to
1852                  * fm_end_offset */
1853                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1854                         lun_start = fm_end_offset;
1855
1856                 if (fm_length != ~0ULL) {
1857                         /* Handle fm_start + fm_length overflow */
1858                         if (fm_start + fm_length < fm_start)
1859                                 fm_length = ~0ULL - fm_start;
1860                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1861                                                      cur_stripe);
1862                 } else {
1863                         lun_end = ~0ULL;
1864                 }
1865
1866                 if (lun_start == lun_end)
1867                         continue;
1868
1869                 req_fm_len = obd_object_end - lun_start;
1870                 fm_local->fm_length = 0;
1871                 len_mapped_single_call = 0;
1872
1873                 /* If the output buffer is very large and the objects have many
1874                  * extents we may need to loop on a single OST repeatedly */
1875                 ost_eof = 0;
1876                 ost_done = 0;
1877                 do {
1878                         if (get_num_extents == 0) {
1879                                 /* Don't get too many extents. */
1880                                 if (current_extent + count_local >
1881                                     fiemap->fm_extent_count)
1882                                         count_local = fiemap->fm_extent_count -
1883                                                                  current_extent;
1884                         }
1885
1886                         lun_start += len_mapped_single_call;
1887                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
1888                         req_fm_len = fm_local->fm_length;
1889                         fm_local->fm_extent_count = count_local;
1890                         fm_local->fm_mapped_extents = 0;
1891                         fm_local->fm_flags = fiemap->fm_flags;
1892
1893                         fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1894                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1895
1896                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1897                                 GOTO(out, rc = -EINVAL);
1898
1899                         /* If OST is inactive, return extent with UNKNOWN flag */
1900                         if (!lov->lov_tgts[ost_index]->ltd_active) {
1901                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1902                                 fm_local->fm_mapped_extents = 1;
1903
1904                                 lcl_fm_ext[0].fe_logical = lun_start;
1905                                 lcl_fm_ext[0].fe_length = obd_object_end -
1906                                                                       lun_start;
1907                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1908
1909                                 goto inactive_tgt;
1910                         }
1911
1912                         fm_local->fm_start = lun_start;
1913                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1914                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1915                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1916                         rc = obd_get_info(NULL,
1917                                           lov->lov_tgts[ost_index]->ltd_exp,
1918                                           keylen, key, vallen, fm_local, lsm);
1919                         if (rc != 0)
1920                                 GOTO(out, rc);
1921
1922 inactive_tgt:
1923                         ext_count = fm_local->fm_mapped_extents;
1924                         if (ext_count == 0) {
1925                                 ost_done = 1;
1926                                 /* If last stripe has hole at the end,
1927                                  * then we need to return */
1928                                 if (cur_stripe_wrap == last_stripe) {
1929                                         fiemap->fm_mapped_extents = 0;
1930                                         goto finish;
1931                                 }
1932                                 break;
1933                         }
1934
1935                         /* If we just need num of extents then go to next device */
1936                         if (get_num_extents) {
1937                                 current_extent += ext_count;
1938                                 break;
1939                         }
1940
1941                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1942                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1943
1944                         /* Have we finished mapping on this device? */
1945                         if (req_fm_len <= len_mapped_single_call)
1946                                 ost_done = 1;
1947
1948                         /* Clear the EXTENT_LAST flag which can be present on
1949                          * last extent */
1950                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1951                                 lcl_fm_ext[ext_count - 1].fe_flags &=
1952                                                             ~FIEMAP_EXTENT_LAST;
1953
1954                         curr_loc = lov_stripe_size(lsm,
1955                                            lcl_fm_ext[ext_count - 1].fe_logical+
1956                                            lcl_fm_ext[ext_count - 1].fe_length,
1957                                            cur_stripe);
1958                         if (curr_loc >= fm_key->oa.o_size)
1959                                 ost_eof = 1;
1960
1961                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1962                                                      ost_index, ext_count,
1963                                                      current_extent);
1964
1965                         current_extent += ext_count;
1966
1967                         /* Ran out of available extents? */
1968                         if (current_extent >= fiemap->fm_extent_count)
1969                                 goto finish;
1970                 } while (ost_done == 0 && ost_eof == 0);
1971
1972                 if (cur_stripe_wrap == last_stripe)
1973                         goto finish;
1974         }
1975
1976 finish:
1977         /* Indicate that we are returning device offsets unless file just has
1978          * single stripe */
1979         if (lsm->lsm_stripe_count > 1)
1980                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1981
1982         if (get_num_extents)
1983                 goto skip_last_device_calc;
1984
1985         /* Check if we have reached the last stripe and whether mapping for that
1986          * stripe is done. */
1987         if (cur_stripe_wrap == last_stripe) {
1988                 if (ost_done || ost_eof)
1989                         fiemap->fm_extents[current_extent - 1].fe_flags |=
1990                                                              FIEMAP_EXTENT_LAST;
1991         }
1992
1993 skip_last_device_calc:
1994         fiemap->fm_mapped_extents = current_extent;
1995
1996 out:
1997         if (fm_local)
1998                 OBD_FREE_LARGE(fm_local, buffer_size);
1999         return rc;
2000 }
2001
2002 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2003                         __u32 keylen, void *key, __u32 *vallen, void *val,
2004                         struct lov_stripe_md *lsm)
2005 {
2006         struct obd_device *obddev = class_exp2obd(exp);
2007         struct lov_obd *lov = &obddev->u.lov;
2008         int i, rc;
2009         ENTRY;
2010
2011         if (!vallen || !val)
2012                 RETURN(-EFAULT);
2013
2014         obd_getref(obddev);
2015
2016         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2017                 struct {
2018                         char name[16];
2019                         struct ldlm_lock *lock;
2020                 } *data = key;
2021                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2022                 struct lov_oinfo *loi;
2023                 __u32 *stripe = val;
2024
2025                 if (*vallen < sizeof(*stripe))
2026                         GOTO(out, rc = -EFAULT);
2027                 *vallen = sizeof(*stripe);
2028
2029                 /* XXX This is another one of those bits that will need to
2030                  * change if we ever actually support nested LOVs.  It uses
2031                  * the lock's export to find out which stripe it is. */
2032                 /* XXX - it's assumed all the locks for deleted OSTs have
2033                  * been cancelled. Also, the export for deleted OSTs will
2034                  * be NULL and won't match the lock's export. */
2035                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2036                         loi = lsm->lsm_oinfo[i];
2037                         if (!lov->lov_tgts[loi->loi_ost_idx])
2038                                 continue;
2039                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2040                             data->lock->l_conn_export &&
2041                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2042                                 *stripe = i;
2043                                 GOTO(out, rc = 0);
2044                         }
2045                 }
2046                 LDLM_ERROR(data->lock, "lock on inode without such object");
2047                 dump_lsm(D_ERROR, lsm);
2048                 GOTO(out, rc = -ENXIO);
2049         } else if (KEY_IS(KEY_LAST_ID)) {
2050                 struct obd_id_info *info = val;
2051                 __u32 size = sizeof(obd_id);
2052                 struct lov_tgt_desc *tgt;
2053
2054                 LASSERT(*vallen == sizeof(struct obd_id_info));
2055                 tgt = lov->lov_tgts[info->idx];
2056
2057                 if (!tgt || !tgt->ltd_active)
2058                         GOTO(out, rc = -ESRCH);
2059
2060                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2061                                   &size, info->data, NULL);
2062                 GOTO(out, rc = 0);
2063         } else if (KEY_IS(KEY_LOVDESC)) {
2064                 struct lov_desc *desc_ret = val;
2065                 *desc_ret = lov->desc;
2066
2067                 GOTO(out, rc = 0);
2068         } else if (KEY_IS(KEY_FIEMAP)) {
2069                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2070                 GOTO(out, rc);
2071         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2072                 struct lov_tgt_desc *tgt;
2073                 __u64 ost_idx = *((__u64*)val);
2074
2075                 LASSERT(*vallen == sizeof(__u64));
2076                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2077                 tgt = lov->lov_tgts[ost_idx];
2078
2079                 if (!tgt || !tgt->ltd_exp)
2080                         GOTO(out, rc = -ESRCH);
2081
2082                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2083                 GOTO(out, rc = 0);
2084         } else if (KEY_IS(KEY_TGT_COUNT)) {
2085                 *((int *)val) = lov->desc.ld_tgt_count;
2086                 GOTO(out, rc = 0);
2087         }
2088
2089         rc = -EINVAL;
2090
2091 out:
2092         obd_putref(obddev);
2093         RETURN(rc);
2094 }
2095
2096 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2097                               obd_count keylen, void *key, obd_count vallen,
2098                               void *val, struct ptlrpc_request_set *set)
2099 {
2100         struct obd_device *obddev = class_exp2obd(exp);
2101         struct lov_obd *lov = &obddev->u.lov;
2102         obd_count count;
2103         int i, rc = 0, err;
2104         struct lov_tgt_desc *tgt;
2105         unsigned incr, check_uuid,
2106                  do_inactive, no_set;
2107         unsigned next_id = 0,  mds_con = 0, capa = 0;
2108         ENTRY;
2109
2110         incr = check_uuid = do_inactive = no_set = 0;
2111         if (set == NULL) {
2112                 no_set = 1;
2113                 set = ptlrpc_prep_set();
2114                 if (!set)
2115                         RETURN(-ENOMEM);
2116         }
2117
2118         obd_getref(obddev);
2119         count = lov->desc.ld_tgt_count;
2120
2121         if (KEY_IS(KEY_NEXT_ID)) {
2122                 count = vallen / sizeof(struct obd_id_info);
2123                 vallen = sizeof(obd_id);
2124                 incr = sizeof(struct obd_id_info);
2125                 do_inactive = 1;
2126                 next_id = 1;
2127         } else if (KEY_IS(KEY_CHECKSUM)) {
2128                 do_inactive = 1;
2129         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2130                 /* use defaults:  do_inactive = incr = 0; */
2131         } else if (KEY_IS(KEY_MDS_CONN)) {
2132                 mds_con = 1;
2133         } else if (KEY_IS(KEY_CAPA_KEY)) {
2134                 capa = 1;
2135         } else if (KEY_IS(KEY_CACHE_SET)) {
2136                 LASSERT(lov->lov_cache == NULL);
2137                 lov->lov_cache = val;
2138                 do_inactive = 1;
2139         }
2140
2141         for (i = 0; i < count; i++, val = (char *)val + incr) {
2142                 if (next_id) {
2143                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2144                 } else {
2145                         tgt = lov->lov_tgts[i];
2146                 }
2147                 /* OST was disconnected */
2148                 if (!tgt || !tgt->ltd_exp)
2149                         continue;
2150
2151                 /* OST is inactive and we don't want inactive OSCs */
2152                 if (!tgt->ltd_active && !do_inactive)
2153                         continue;
2154
2155                 if (mds_con) {
2156                         struct mds_group_info *mgi;
2157
2158                         LASSERT(vallen == sizeof(*mgi));
2159                         mgi = (struct mds_group_info *)val;
2160
2161                         /* Only want a specific OSC */
2162                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2163                                                 &tgt->ltd_uuid))
2164                                 continue;
2165
2166                         err = obd_set_info_async(env, tgt->ltd_exp,
2167                                          keylen, key, sizeof(int),
2168                                          &mgi->group, set);
2169                 } else if (next_id) {
2170                         err = obd_set_info_async(env, tgt->ltd_exp,
2171                                          keylen, key, vallen,
2172                                          ((struct obd_id_info*)val)->data, set);
2173                 } else if (capa) {
2174                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2175
2176                         LASSERT(vallen == sizeof(*info));
2177
2178                          /* Only want a specific OSC */
2179                         if (info->uuid &&
2180                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2181                                 continue;
2182
2183                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2184                                                  key, sizeof(*info->capa),
2185                                                  info->capa, set);
2186                 } else {
2187                         /* Only want a specific OSC */
2188                         if (check_uuid &&
2189                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2190                                 continue;
2191
2192                         err = obd_set_info_async(env, tgt->ltd_exp,
2193                                          keylen, key, vallen, val, set);
2194                 }
2195
2196                 if (!rc)
2197                         rc = err;
2198         }
2199
2200         obd_putref(obddev);
2201         if (no_set) {
2202                 err = ptlrpc_set_wait(set);
2203                 if (!rc)
2204                         rc = err;
2205                 ptlrpc_set_destroy(set);
2206         }
2207         RETURN(rc);
2208 }
2209
2210 void lov_stripe_lock(struct lov_stripe_md *md)
2211 {
2212         LASSERT(md->lsm_lock_owner != current_pid());
2213         spin_lock(&md->lsm_lock);
2214         LASSERT(md->lsm_lock_owner == 0);
2215         md->lsm_lock_owner = current_pid();
2216 }
2217
2218 void lov_stripe_unlock(struct lov_stripe_md *md)
2219 {
2220         LASSERT(md->lsm_lock_owner == current_pid());
2221         md->lsm_lock_owner = 0;
2222         spin_unlock(&md->lsm_lock);
2223 }
2224
2225 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2226                         struct obd_quotactl *oqctl)
2227 {
2228         struct lov_obd      *lov = &obd->u.lov;
2229         struct lov_tgt_desc *tgt;
2230         __u64                curspace = 0;
2231         __u64                bhardlimit = 0;
2232         int                  i, rc = 0;
2233         ENTRY;
2234
2235         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2236             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2237             oqctl->qc_cmd != Q_GETOQUOTA &&
2238             oqctl->qc_cmd != Q_INITQUOTA &&
2239             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2240             oqctl->qc_cmd != Q_FINVALIDATE) {
2241                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2242                 RETURN(-EFAULT);
2243         }
2244
2245         /* for lov tgt */
2246         obd_getref(obd);
2247         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2248                 int err;
2249
2250                 tgt = lov->lov_tgts[i];
2251
2252                 if (!tgt)
2253                         continue;
2254
2255                 if (!tgt->ltd_active || tgt->ltd_reap) {
2256                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2257                             lov->lov_tgts[i]->ltd_activate) {
2258                                 rc = -EREMOTEIO;
2259                                 CERROR("ost %d is inactive\n", i);
2260                         } else {
2261                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2262                         }
2263                         continue;
2264                 }
2265
2266                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2267                 if (err) {
2268                         if (tgt->ltd_active && !rc)
2269                                 rc = err;
2270                         continue;
2271                 }
2272
2273                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2274                         curspace += oqctl->qc_dqblk.dqb_curspace;
2275                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2276                 }
2277         }
2278         obd_putref(obd);
2279
2280         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2281                 oqctl->qc_dqblk.dqb_curspace = curspace;
2282                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2283         }
2284         RETURN(rc);
2285 }
2286
2287 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2288                           struct obd_quotactl *oqctl)
2289 {
2290         struct lov_obd *lov = &obd->u.lov;
2291         int             i, rc = 0;
2292         ENTRY;
2293
2294         obd_getref(obd);
2295
2296         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2297                 if (!lov->lov_tgts[i])
2298                         continue;
2299
2300                 /* Skip quota check on the administratively disabled OSTs. */
2301                 if (!lov->lov_tgts[i]->ltd_activate) {
2302                         CWARN("lov idx %d was administratively disabled, "
2303                               "skip quotacheck on it.\n", i);
2304                         continue;
2305                 }
2306
2307                 if (!lov->lov_tgts[i]->ltd_active) {
2308                         CERROR("lov idx %d inactive\n", i);
2309                         rc = -EIO;
2310                         goto out;
2311                 }
2312         }
2313
2314         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2315                 int err;
2316
2317                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2318                         continue;
2319
2320                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2321                 if (err && !rc)
2322                         rc = err;
2323         }
2324
2325 out:
2326         obd_putref(obd);
2327
2328         RETURN(rc);
2329 }
2330
2331 static struct obd_ops lov_obd_ops = {
2332         .o_owner                = THIS_MODULE,
2333         .o_setup                = lov_setup,
2334         .o_precleanup           = lov_precleanup,
2335         .o_cleanup              = lov_cleanup,
2336         .o_connect              = lov_connect,
2337         .o_disconnect           = lov_disconnect,
2338         .o_statfs               = lov_statfs,
2339         .o_statfs_async         = lov_statfs_async,
2340         .o_packmd               = lov_packmd,
2341         .o_unpackmd             = lov_unpackmd,
2342         .o_create               = lov_create,
2343         .o_destroy              = lov_destroy,
2344         .o_getattr_async        = lov_getattr_async,
2345         .o_setattr_async        = lov_setattr_async,
2346         .o_change_cbdata        = lov_change_cbdata,
2347         .o_find_cbdata          = lov_find_cbdata,
2348         .o_iocontrol            = lov_iocontrol,
2349         .o_get_info             = lov_get_info,
2350         .o_set_info_async       = lov_set_info_async,
2351         .o_notify               = lov_notify,
2352         .o_pool_new             = lov_pool_new,
2353         .o_pool_rem             = lov_pool_remove,
2354         .o_pool_add             = lov_pool_add,
2355         .o_pool_del             = lov_pool_del,
2356         .o_getref               = lov_getref,
2357         .o_putref               = lov_putref,
2358         .o_quotactl             = lov_quotactl,
2359         .o_quotacheck           = lov_quotacheck,
2360 };
2361
2362 struct kmem_cache *lov_oinfo_slab;
2363
2364 extern struct lu_kmem_descr lov_caches[];
2365
2366 int __init lov_init(void)
2367 {
2368         int rc;
2369         ENTRY;
2370
2371         /* print an address of _any_ initialized kernel symbol from this
2372          * module, to allow debugging with gdb that doesn't support data
2373          * symbols from modules.*/
2374         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2375
2376         rc = lu_kmem_init(lov_caches);
2377         if (rc)
2378                 return rc;
2379
2380         lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2381                                            sizeof(struct lov_oinfo), 0,
2382                                            SLAB_HWCACHE_ALIGN, NULL);
2383         if (lov_oinfo_slab == NULL) {
2384                 lu_kmem_fini(lov_caches);
2385                 return -ENOMEM;
2386         }
2387
2388         rc = class_register_type(&lov_obd_ops, NULL, NULL,
2389 #ifndef HAVE_ONLY_PROCFS_SEQ
2390                                 NULL,
2391 #endif
2392                                 LUSTRE_LOV_NAME, &lov_device_type);
2393
2394         if (rc) {
2395                 kmem_cache_destroy(lov_oinfo_slab);
2396                 lu_kmem_fini(lov_caches);
2397         }
2398
2399         RETURN(rc);
2400 }
2401
2402 #ifdef __KERNEL__
2403 static void /*__exit*/ lov_exit(void)
2404 {
2405         class_unregister_type(LUSTRE_LOV_NAME);
2406         kmem_cache_destroy(lov_oinfo_slab);
2407         lu_kmem_fini(lov_caches);
2408 }
2409
2410 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2411 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2412 MODULE_LICENSE("GPL");
2413
2414 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2415 #endif