]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/staging/lustre/lustre/lov/lov_obd.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
staging: lustre: update the MODULE_DESCRIPTION for all lustre modules
[mirror_ubuntu-bionic-kernel.git] / drivers / staging / lustre / 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, 2015, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * lustre/lov/lov_obd.c
37 *
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
42 */
43
44 #define DEBUG_SUBSYSTEM S_LOV
45 #include "../../include/linux/libcfs/libcfs.h"
46
47 #include "../include/obd_support.h"
48 #include "../include/lustre_lib.h"
49 #include "../include/lustre_net.h"
50 #include "../include/lustre/lustre_idl.h"
51 #include "../include/lustre_dlm.h"
52 #include "../include/lustre_mds.h"
53 #include "../include/obd_class.h"
54 #include "../include/lprocfs_status.h"
55 #include "../include/lustre_param.h"
56 #include "../include/cl_object.h"
57 #include "../include/lclient.h" /* for cl_client_lru */
58 #include "../include/lustre/ll_fiemap.h"
59 #include "../include/lustre_fid.h"
60
61 #include "lov_internal.h"
62
63 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
64 * Any function that expects lov_tgts to remain stationary must take a ref.
65 */
66 static void lov_getref(struct obd_device *obd)
67 {
68 struct lov_obd *lov = &obd->u.lov;
69
70 /* nobody gets through here until lov_putref is done */
71 mutex_lock(&lov->lov_lock);
72 atomic_inc(&lov->lov_refcount);
73 mutex_unlock(&lov->lov_lock);
74 return;
75 }
76
77 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
78
79 static void lov_putref(struct obd_device *obd)
80 {
81 struct lov_obd *lov = &obd->u.lov;
82
83 mutex_lock(&lov->lov_lock);
84 /* ok to dec to 0 more than once -- ltd_exp's will be null */
85 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
86 LIST_HEAD(kill);
87 int i;
88 struct lov_tgt_desc *tgt, *n;
89
90 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
91 lov->lov_death_row);
92 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
93 tgt = lov->lov_tgts[i];
94
95 if (!tgt || !tgt->ltd_reap)
96 continue;
97 list_add(&tgt->ltd_kill, &kill);
98 /* XXX - right now there is a dependency on ld_tgt_count
99 * being the maximum tgt index for computing the
100 * mds_max_easize. So we can't shrink it.
101 */
102 lov_ost_pool_remove(&lov->lov_packed, i);
103 lov->lov_tgts[i] = NULL;
104 lov->lov_death_row--;
105 }
106 mutex_unlock(&lov->lov_lock);
107
108 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
109 list_del(&tgt->ltd_kill);
110 /* Disconnect */
111 __lov_del_obd(obd, tgt);
112 }
113
114 if (lov->lov_tgts_kobj)
115 kobject_put(lov->lov_tgts_kobj);
116
117 } else {
118 mutex_unlock(&lov->lov_lock);
119 }
120 }
121
122 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
123 enum obd_notify_event ev);
124 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
125 enum obd_notify_event ev, void *data);
126
127 #define MAX_STRING_SIZE 128
128 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
129 struct obd_connect_data *data)
130 {
131 struct lov_obd *lov = &obd->u.lov;
132 struct obd_uuid *tgt_uuid;
133 struct obd_device *tgt_obd;
134 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
135 struct obd_import *imp;
136 int rc;
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 */
165 ptlrpc_activate_import(imp);
166 }
167
168 rc = obd_register_observer(tgt_obd, obd);
169 if (rc) {
170 CERROR("Target %s register_observer error %d\n",
171 obd_uuid2str(tgt_uuid), rc);
172 return rc;
173 }
174
175 if (imp->imp_invalid) {
176 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
177 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 if (lov->lov_tgts_kobj)
195 /* Even if we failed, that's ok */
196 rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj,
197 tgt_obd->obd_name);
198
199 return 0;
200 }
201
202 static int lov_connect(const struct lu_env *env,
203 struct obd_export **exp, struct obd_device *obd,
204 struct obd_uuid *cluuid, struct obd_connect_data *data,
205 void *localdata)
206 {
207 struct lov_obd *lov = &obd->u.lov;
208 struct lov_tgt_desc *tgt;
209 struct lustre_handle conn;
210 int i, rc;
211
212 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
213
214 rc = class_connect(&conn, obd, cluuid);
215 if (rc)
216 return rc;
217
218 *exp = class_conn2export(&conn);
219
220 /* Why should there ever be more than 1 connect? */
221 lov->lov_connects++;
222 LASSERT(lov->lov_connects == 1);
223
224 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
225 if (data)
226 lov->lov_ocd = *data;
227
228 obd_getref(obd);
229
230 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
231 &obd->obd_kobj);
232
233 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
234 tgt = lov->lov_tgts[i];
235 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
236 continue;
237 /* Flags will be lowest common denominator */
238 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
239 if (rc) {
240 CERROR("%s: lov connect tgt %d failed: %d\n",
241 obd->obd_name, i, rc);
242 continue;
243 }
244 /* connect to administrative disabled ost */
245 if (!lov->lov_tgts[i]->ltd_exp)
246 continue;
247
248 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
249 OBD_NOTIFY_CONNECT, (void *)&i);
250 if (rc) {
251 CERROR("%s error sending notify %d\n",
252 obd->obd_name, rc);
253 }
254 }
255 obd_putref(obd);
256
257 return 0;
258 }
259
260 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
261 {
262 struct lov_obd *lov = &obd->u.lov;
263 struct obd_device *osc_obd;
264 int rc;
265
266 osc_obd = class_exp2obd(tgt->ltd_exp);
267 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
268 obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
269
270 if (tgt->ltd_active) {
271 tgt->ltd_active = 0;
272 lov->desc.ld_active_tgt_count--;
273 tgt->ltd_exp->exp_obd->obd_inactive = 1;
274 }
275
276 if (osc_obd) {
277 if (lov->lov_tgts_kobj)
278 sysfs_remove_link(lov->lov_tgts_kobj,
279 osc_obd->obd_name);
280
281 /* Pass it on to our clients.
282 * XXX This should be an argument to disconnect,
283 * XXX not a back-door flag on the OBD. Ah well.
284 */
285 osc_obd->obd_force = obd->obd_force;
286 osc_obd->obd_fail = obd->obd_fail;
287 osc_obd->obd_no_recov = obd->obd_no_recov;
288 }
289
290 obd_register_observer(osc_obd, NULL);
291
292 rc = obd_disconnect(tgt->ltd_exp);
293 if (rc) {
294 CERROR("Target %s disconnect error %d\n",
295 tgt->ltd_uuid.uuid, rc);
296 rc = 0;
297 }
298
299 tgt->ltd_exp = NULL;
300 return 0;
301 }
302
303 static int lov_disconnect(struct obd_export *exp)
304 {
305 struct obd_device *obd = class_exp2obd(exp);
306 struct lov_obd *lov = &obd->u.lov;
307 int i, rc;
308
309 if (!lov->lov_tgts)
310 goto out;
311
312 /* Only disconnect the underlying layers on the final disconnect. */
313 lov->lov_connects--;
314 if (lov->lov_connects != 0) {
315 /* why should there be more than 1 connect? */
316 CERROR("disconnect #%d\n", lov->lov_connects);
317 goto out;
318 }
319
320 /* Let's hold another reference so lov_del_obd doesn't spin through
321 * putref every time
322 */
323 obd_getref(obd);
324
325 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
326 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
327 /* Disconnection is the last we know about an obd */
328 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
329 }
330 }
331
332 obd_putref(obd);
333
334 out:
335 rc = class_disconnect(exp); /* bz 9811 */
336 return rc;
337 }
338
339 /* Error codes:
340 *
341 * -EINVAL : UUID can't be found in the LOV's target list
342 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
343 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
344 * any >= 0 : is log target index
345 */
346 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
347 enum obd_notify_event ev)
348 {
349 struct lov_obd *lov = &obd->u.lov;
350 struct lov_tgt_desc *tgt;
351 int index, activate, active;
352
353 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
354 lov, uuid->uuid, ev);
355
356 obd_getref(obd);
357 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
358 tgt = lov->lov_tgts[index];
359 if (!tgt)
360 continue;
361 /*
362 * LU-642, initially inactive OSC could miss the obd_connect,
363 * we make up for it here.
364 */
365 if (ev == OBD_NOTIFY_ACTIVATE && !tgt->ltd_exp &&
366 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
367 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
368
369 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
370 &lov_osc_uuid, &lov->lov_ocd, NULL);
371 }
372 if (!tgt->ltd_exp)
373 continue;
374
375 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
376 index, obd_uuid2str(&tgt->ltd_uuid),
377 tgt->ltd_exp->exp_handle.h_cookie);
378 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
379 break;
380 }
381
382 if (index == lov->desc.ld_tgt_count) {
383 index = -EINVAL;
384 goto out;
385 }
386
387 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
388 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
389
390 if (lov->lov_tgts[index]->ltd_activate == activate) {
391 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
392 uuid->uuid, activate ? "" : "de");
393 } else {
394 lov->lov_tgts[index]->ltd_activate = activate;
395 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
396 activate ? "" : "de", obd_uuid2str(uuid));
397 }
398
399 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
400 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
401
402 if (lov->lov_tgts[index]->ltd_active == active) {
403 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
404 uuid->uuid, active ? "" : "in");
405 goto out;
406 }
407 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
408 obd_uuid2str(uuid), active ? "" : "in");
409
410 lov->lov_tgts[index]->ltd_active = active;
411 if (active) {
412 lov->desc.ld_active_tgt_count++;
413 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
414 } else {
415 lov->desc.ld_active_tgt_count--;
416 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
417 }
418 } else {
419 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
420 }
421
422 out:
423 obd_putref(obd);
424 return index;
425 }
426
427 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
428 enum obd_notify_event ev, void *data)
429 {
430 int rc = 0;
431 struct lov_obd *lov = &obd->u.lov;
432
433 down_read(&lov->lov_notify_lock);
434 if (!lov->lov_connects) {
435 up_read(&lov->lov_notify_lock);
436 return rc;
437 }
438
439 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
440 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
441 struct obd_uuid *uuid;
442
443 LASSERT(watched);
444
445 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
446 up_read(&lov->lov_notify_lock);
447 CERROR("unexpected notification of %s %s!\n",
448 watched->obd_type->typ_name,
449 watched->obd_name);
450 return -EINVAL;
451 }
452 uuid = &watched->u.cli.cl_target_uuid;
453
454 /* Set OSC as active before notifying the observer, so the
455 * observer can use the OSC normally.
456 */
457 rc = lov_set_osc_active(obd, uuid, ev);
458 if (rc < 0) {
459 up_read(&lov->lov_notify_lock);
460 CERROR("event(%d) of %s failed: %d\n", ev,
461 obd_uuid2str(uuid), rc);
462 return rc;
463 }
464 /* active event should be pass lov target index as data */
465 data = &rc;
466 }
467
468 /* Pass the notification up the chain. */
469 if (watched) {
470 rc = obd_notify_observer(obd, watched, ev, data);
471 } else {
472 /* NULL watched means all osc's in the lov (only for syncs) */
473 /* sync event should be send lov idx as data */
474 struct lov_obd *lov = &obd->u.lov;
475 int i, is_sync;
476
477 data = &i;
478 is_sync = (ev == OBD_NOTIFY_SYNC) ||
479 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
480
481 obd_getref(obd);
482 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
483 if (!lov->lov_tgts[i])
484 continue;
485
486 /* don't send sync event if target not
487 * connected/activated
488 */
489 if (is_sync && !lov->lov_tgts[i]->ltd_active)
490 continue;
491
492 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
493 ev, data);
494 if (rc) {
495 CERROR("%s: notify %s of %s failed %d\n",
496 obd->obd_name,
497 obd->obd_observer->obd_name,
498 lov->lov_tgts[i]->ltd_obd->obd_name,
499 rc);
500 }
501 }
502 obd_putref(obd);
503 }
504
505 up_read(&lov->lov_notify_lock);
506 return rc;
507 }
508
509 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
510 __u32 index, int gen, int active)
511 {
512 struct lov_obd *lov = &obd->u.lov;
513 struct lov_tgt_desc *tgt;
514 struct obd_device *tgt_obd;
515 int rc;
516
517 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
518 uuidp->uuid, index, gen, active);
519
520 if (gen <= 0) {
521 CERROR("request to add OBD %s with invalid generation: %d\n",
522 uuidp->uuid, gen);
523 return -EINVAL;
524 }
525
526 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
527 &obd->obd_uuid);
528 if (!tgt_obd)
529 return -EINVAL;
530
531 mutex_lock(&lov->lov_lock);
532
533 if ((index < lov->lov_tgt_size) && lov->lov_tgts[index]) {
534 tgt = lov->lov_tgts[index];
535 CERROR("UUID %s already assigned at LOV target index %d\n",
536 obd_uuid2str(&tgt->ltd_uuid), index);
537 mutex_unlock(&lov->lov_lock);
538 return -EEXIST;
539 }
540
541 if (index >= lov->lov_tgt_size) {
542 /* We need to reallocate the lov target array. */
543 struct lov_tgt_desc **newtgts, **old = NULL;
544 __u32 newsize, oldsize = 0;
545
546 newsize = max_t(__u32, lov->lov_tgt_size, 2);
547 while (newsize < index + 1)
548 newsize <<= 1;
549 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
550 if (!newtgts) {
551 mutex_unlock(&lov->lov_lock);
552 return -ENOMEM;
553 }
554
555 if (lov->lov_tgt_size) {
556 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
557 lov->lov_tgt_size);
558 old = lov->lov_tgts;
559 oldsize = lov->lov_tgt_size;
560 }
561
562 lov->lov_tgts = newtgts;
563 lov->lov_tgt_size = newsize;
564 smp_rmb();
565 kfree(old);
566
567 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
568 lov->lov_tgts, lov->lov_tgt_size);
569 }
570
571 tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
572 if (!tgt) {
573 mutex_unlock(&lov->lov_lock);
574 return -ENOMEM;
575 }
576
577 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
578 if (rc) {
579 mutex_unlock(&lov->lov_lock);
580 kfree(tgt);
581 return rc;
582 }
583
584 tgt->ltd_uuid = *uuidp;
585 tgt->ltd_obd = tgt_obd;
586 /* XXX - add a sanity check on the generation number. */
587 tgt->ltd_gen = gen;
588 tgt->ltd_index = index;
589 tgt->ltd_activate = active;
590 lov->lov_tgts[index] = tgt;
591 if (index >= lov->desc.ld_tgt_count)
592 lov->desc.ld_tgt_count = index + 1;
593
594 mutex_unlock(&lov->lov_lock);
595
596 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
597 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
598
599 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
600
601 if (lov->lov_connects == 0) {
602 /* lov_connect hasn't been called yet. We'll do the
603 * lov_connect_obd on this target when that fn first runs,
604 * because we don't know the connect flags yet.
605 */
606 return 0;
607 }
608
609 obd_getref(obd);
610
611 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
612 if (rc)
613 goto out;
614
615 /* connect to administrative disabled ost */
616 if (!tgt->ltd_exp) {
617 rc = 0;
618 goto out;
619 }
620
621 if (lov->lov_cache) {
622 rc = obd_set_info_async(NULL, tgt->ltd_exp,
623 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
624 sizeof(struct cl_client_cache),
625 lov->lov_cache, NULL);
626 if (rc < 0)
627 goto out;
628 }
629
630 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
631 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
632 (void *)&index);
633
634 out:
635 if (rc) {
636 CERROR("add failed (%d), deleting %s\n", rc,
637 obd_uuid2str(&tgt->ltd_uuid));
638 lov_del_target(obd, index, NULL, 0);
639 }
640 obd_putref(obd);
641 return rc;
642 }
643
644 /* Schedule a target for deletion */
645 int lov_del_target(struct obd_device *obd, __u32 index,
646 struct obd_uuid *uuidp, int gen)
647 {
648 struct lov_obd *lov = &obd->u.lov;
649 int count = lov->desc.ld_tgt_count;
650 int rc = 0;
651
652 if (index >= count) {
653 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
654 index, count);
655 return -EINVAL;
656 }
657
658 /* to make sure there's no ongoing lov_notify() now */
659 down_write(&lov->lov_notify_lock);
660 obd_getref(obd);
661
662 if (!lov->lov_tgts[index]) {
663 CERROR("LOV target at index %d is not setup.\n", index);
664 rc = -EINVAL;
665 goto out;
666 }
667
668 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
669 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
670 lov_uuid2str(lov, index), index,
671 obd_uuid2str(uuidp));
672 rc = -EINVAL;
673 goto out;
674 }
675
676 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
677 lov_uuid2str(lov, index), index,
678 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
679 lov->lov_tgts[index]->ltd_active);
680
681 lov->lov_tgts[index]->ltd_reap = 1;
682 lov->lov_death_row++;
683 /* we really delete it from obd_putref */
684 out:
685 obd_putref(obd);
686 up_write(&lov->lov_notify_lock);
687
688 return rc;
689 }
690
691 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
692 {
693 struct obd_device *osc_obd;
694
695 LASSERT(tgt);
696 LASSERT(tgt->ltd_reap);
697
698 osc_obd = class_exp2obd(tgt->ltd_exp);
699
700 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
701 tgt->ltd_uuid.uuid,
702 osc_obd ? osc_obd->obd_name : "<no obd>");
703
704 if (tgt->ltd_exp)
705 lov_disconnect_obd(obd, tgt);
706
707 kfree(tgt);
708
709 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
710 * do it ourselves. And we can't do it from lov_cleanup,
711 * because we just lost our only reference to it.
712 */
713 if (osc_obd)
714 class_manual_cleanup(osc_obd);
715 }
716
717 void lov_fix_desc_stripe_size(__u64 *val)
718 {
719 if (*val < LOV_MIN_STRIPE_SIZE) {
720 if (*val != 0)
721 LCONSOLE_INFO("Increasing default stripe size to minimum %u\n",
722 LOV_DESC_STRIPE_SIZE_DEFAULT);
723 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
724 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
725 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
726 LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n",
727 *val, LOV_MIN_STRIPE_SIZE);
728 }
729 }
730
731 void lov_fix_desc_stripe_count(__u32 *val)
732 {
733 if (*val == 0)
734 *val = 1;
735 }
736
737 void lov_fix_desc_pattern(__u32 *val)
738 {
739 /* from lov_setstripe */
740 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
741 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
742 *val = 0;
743 }
744 }
745
746 void lov_fix_desc_qos_maxage(__u32 *val)
747 {
748 if (*val == 0)
749 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
750 }
751
752 void lov_fix_desc(struct lov_desc *desc)
753 {
754 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
755 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
756 lov_fix_desc_pattern(&desc->ld_pattern);
757 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
758 }
759
760 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
761 {
762 struct lprocfs_static_vars lvars = { NULL };
763 struct lov_desc *desc;
764 struct lov_obd *lov = &obd->u.lov;
765 int rc;
766
767 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
768 CERROR("LOV setup requires a descriptor\n");
769 return -EINVAL;
770 }
771
772 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
773
774 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
775 CERROR("descriptor size wrong: %d > %d\n",
776 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
777 return -EINVAL;
778 }
779
780 if (desc->ld_magic != LOV_DESC_MAGIC) {
781 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
782 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
783 obd->obd_name, desc);
784 lustre_swab_lov_desc(desc);
785 } else {
786 CERROR("%s: Bad lov desc magic: %#x\n",
787 obd->obd_name, desc->ld_magic);
788 return -EINVAL;
789 }
790 }
791
792 lov_fix_desc(desc);
793
794 desc->ld_active_tgt_count = 0;
795 lov->desc = *desc;
796 lov->lov_tgt_size = 0;
797
798 mutex_init(&lov->lov_lock);
799 atomic_set(&lov->lov_refcount, 0);
800 lov->lov_sp_me = LUSTRE_SP_CLI;
801
802 init_rwsem(&lov->lov_notify_lock);
803
804 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
805 HASH_POOLS_MAX_BITS,
806 HASH_POOLS_BKT_BITS, 0,
807 CFS_HASH_MIN_THETA,
808 CFS_HASH_MAX_THETA,
809 &pool_hash_operations,
810 CFS_HASH_DEFAULT);
811 INIT_LIST_HEAD(&lov->lov_pool_list);
812 lov->lov_pool_count = 0;
813 rc = lov_ost_pool_init(&lov->lov_packed, 0);
814 if (rc)
815 goto out;
816
817 lprocfs_lov_init_vars(&lvars);
818 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
819
820 rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd",
821 0444, &lov_proc_target_fops, obd);
822 if (rc)
823 CWARN("Error adding the target_obd file\n");
824
825 lov->lov_pool_debugfs_entry = ldebugfs_register("pools",
826 obd->obd_debugfs_entry,
827 NULL, NULL);
828 return 0;
829
830 out:
831 return rc;
832 }
833
834 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
835 {
836 struct lov_obd *lov = &obd->u.lov;
837
838 switch (stage) {
839 case OBD_CLEANUP_EARLY: {
840 int i;
841
842 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
843 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
844 continue;
845 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
846 OBD_CLEANUP_EARLY);
847 }
848 break;
849 }
850 default:
851 break;
852 }
853
854 return 0;
855 }
856
857 static int lov_cleanup(struct obd_device *obd)
858 {
859 struct lov_obd *lov = &obd->u.lov;
860 struct list_head *pos, *tmp;
861 struct pool_desc *pool;
862
863 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
864 pool = list_entry(pos, struct pool_desc, pool_list);
865 /* free pool structs */
866 CDEBUG(D_INFO, "delete pool %p\n", pool);
867 /* In the function below, .hs_keycmp resolves to
868 * pool_hashkey_keycmp()
869 */
870 /* coverity[overrun-buffer-val] */
871 lov_pool_del(obd, pool->pool_name);
872 }
873 cfs_hash_putref(lov->lov_pools_hash_body);
874 lov_ost_pool_free(&lov->lov_packed);
875
876 lprocfs_obd_cleanup(obd);
877 if (lov->lov_tgts) {
878 int i;
879
880 obd_getref(obd);
881 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
882 if (!lov->lov_tgts[i])
883 continue;
884
885 /* Inactive targets may never have connected */
886 if (lov->lov_tgts[i]->ltd_active ||
887 atomic_read(&lov->lov_refcount))
888 /* We should never get here - these
889 * should have been removed in the
890 * disconnect.
891 */
892 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
893 i, lov->lov_death_row,
894 atomic_read(&lov->lov_refcount));
895 lov_del_target(obd, i, NULL, 0);
896 }
897 obd_putref(obd);
898 kfree(lov->lov_tgts);
899 lov->lov_tgt_size = 0;
900 }
901 return 0;
902 }
903
904 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
905 __u32 *indexp, int *genp)
906 {
907 struct obd_uuid obd_uuid;
908 int cmd;
909 int rc = 0;
910
911 switch (cmd = lcfg->lcfg_command) {
912 case LCFG_LOV_ADD_OBD:
913 case LCFG_LOV_ADD_INA:
914 case LCFG_LOV_DEL_OBD: {
915 __u32 index;
916 int gen;
917 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
918 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
919 rc = -EINVAL;
920 goto out;
921 }
922
923 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
924
925 rc = kstrtoint(lustre_cfg_buf(lcfg, 2), 10, indexp);
926 if (rc < 0)
927 goto out;
928 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
929 if (rc < 0)
930 goto out;
931 index = *indexp;
932 gen = *genp;
933 if (cmd == LCFG_LOV_ADD_OBD)
934 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
935 else if (cmd == LCFG_LOV_ADD_INA)
936 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
937 else
938 rc = lov_del_target(obd, index, &obd_uuid, gen);
939 goto out;
940 }
941 case LCFG_PARAM: {
942 struct lprocfs_static_vars lvars = { NULL };
943 struct lov_desc *desc = &(obd->u.lov.desc);
944
945 if (!desc) {
946 rc = -EINVAL;
947 goto out;
948 }
949
950 lprocfs_lov_init_vars(&lvars);
951
952 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
953 lcfg, obd);
954 if (rc > 0)
955 rc = 0;
956 goto out;
957 }
958 case LCFG_POOL_NEW:
959 case LCFG_POOL_ADD:
960 case LCFG_POOL_DEL:
961 case LCFG_POOL_REM:
962 goto out;
963
964 default: {
965 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
966 rc = -EINVAL;
967 goto out;
968
969 }
970 }
971 out:
972 return rc;
973 }
974
975 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
976 struct lov_stripe_md **ea, struct obd_trans_info *oti)
977 {
978 struct lov_stripe_md *obj_mdp, *lsm;
979 struct lov_obd *lov = &exp->exp_obd->u.lov;
980 unsigned ost_idx;
981 int rc, i;
982
983 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
984 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
985
986 obj_mdp = kzalloc(sizeof(*obj_mdp), GFP_NOFS);
987 if (!obj_mdp)
988 return -ENOMEM;
989
990 ost_idx = src_oa->o_nlink;
991 lsm = *ea;
992 if (!lsm) {
993 rc = -EINVAL;
994 goto out;
995 }
996 if (ost_idx >= lov->desc.ld_tgt_count ||
997 !lov->lov_tgts[ost_idx]) {
998 rc = -EINVAL;
999 goto out;
1000 }
1001
1002 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1003 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1004
1005 if (lov_oinfo_is_dummy(loi))
1006 continue;
1007
1008 if (loi->loi_ost_idx == ost_idx) {
1009 if (ostid_id(&loi->loi_oi) != ostid_id(&src_oa->o_oi)) {
1010 rc = -EINVAL;
1011 goto out;
1012 }
1013 break;
1014 }
1015 }
1016 if (i == lsm->lsm_stripe_count) {
1017 rc = -EINVAL;
1018 goto out;
1019 }
1020
1021 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1022 src_oa, &obj_mdp, oti);
1023 out:
1024 kfree(obj_mdp);
1025 return rc;
1026 }
1027
1028 /* the LOV expects oa->o_id to be set to the LOV object id */
1029 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1030 struct obdo *src_oa, struct lov_stripe_md **ea,
1031 struct obd_trans_info *oti)
1032 {
1033 struct lov_obd *lov;
1034 int rc = 0;
1035
1036 LASSERT(ea);
1037 if (!exp)
1038 return -EINVAL;
1039
1040 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1041 src_oa->o_flags == OBD_FL_DELORPHAN) {
1042 /* should be used with LOV anymore */
1043 LBUG();
1044 }
1045
1046 lov = &exp->exp_obd->u.lov;
1047 if (!lov->desc.ld_active_tgt_count)
1048 return -EIO;
1049
1050 obd_getref(exp->exp_obd);
1051 /* Recreate a specific object id at the given OST index */
1052 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1053 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1054 rc = lov_recreate(exp, src_oa, ea, oti);
1055 }
1056
1057 obd_putref(exp->exp_obd);
1058 return rc;
1059 }
1060
1061 #define ASSERT_LSM_MAGIC(lsmp) \
1062 do { \
1063 LASSERT((lsmp)); \
1064 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1065 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1066 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1067 } while (0)
1068
1069 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1070 struct obdo *oa, struct lov_stripe_md *lsm,
1071 struct obd_trans_info *oti, struct obd_export *md_exp)
1072 {
1073 struct lov_request_set *set;
1074 struct obd_info oinfo;
1075 struct lov_request *req;
1076 struct lov_obd *lov;
1077 int rc = 0, err = 0;
1078
1079 ASSERT_LSM_MAGIC(lsm);
1080
1081 if (!exp || !exp->exp_obd)
1082 return -ENODEV;
1083
1084 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1085 LASSERT(oti);
1086 LASSERT(oti->oti_logcookies);
1087 }
1088
1089 lov = &exp->exp_obd->u.lov;
1090 obd_getref(exp->exp_obd);
1091 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1092 if (rc)
1093 goto out;
1094
1095 list_for_each_entry(req, &set->set_list, rq_link) {
1096 if (oa->o_valid & OBD_MD_FLCOOKIE)
1097 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1098
1099 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1100 req->rq_oi.oi_oa, NULL, oti, NULL);
1101 err = lov_update_common_set(set, req, err);
1102 if (err) {
1103 CERROR("%s: destroying objid "DOSTID" subobj "
1104 DOSTID" on OST idx %d: rc = %d\n",
1105 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1106 POSTID(&req->rq_oi.oi_oa->o_oi),
1107 req->rq_idx, err);
1108 if (!rc)
1109 rc = err;
1110 }
1111 }
1112
1113 if (rc == 0)
1114 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1115
1116 err = lov_fini_destroy_set(set);
1117 out:
1118 obd_putref(exp->exp_obd);
1119 return rc ? rc : err;
1120 }
1121
1122 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1123 void *data, int rc)
1124 {
1125 struct lov_request_set *lovset = (struct lov_request_set *)data;
1126 int err;
1127
1128 /* don't do attribute merge if this async op failed */
1129 if (rc)
1130 atomic_set(&lovset->set_completes, 0);
1131 err = lov_fini_getattr_set(lovset);
1132 return rc ? rc : err;
1133 }
1134
1135 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1136 struct ptlrpc_request_set *rqset)
1137 {
1138 struct lov_request_set *lovset;
1139 struct lov_obd *lov;
1140 struct lov_request *req;
1141 int rc = 0, err;
1142
1143 LASSERT(oinfo);
1144 ASSERT_LSM_MAGIC(oinfo->oi_md);
1145
1146 if (!exp || !exp->exp_obd)
1147 return -ENODEV;
1148
1149 lov = &exp->exp_obd->u.lov;
1150
1151 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1152 if (rc)
1153 return rc;
1154
1155 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1156 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1157 oinfo->oi_md->lsm_stripe_size);
1158
1159 list_for_each_entry(req, &lovset->set_list, rq_link) {
1160 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1161 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1162 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1163 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1164 &req->rq_oi, rqset);
1165 if (rc) {
1166 CERROR("%s: getattr objid "DOSTID" subobj"
1167 DOSTID" on OST idx %d: rc = %d\n",
1168 exp->exp_obd->obd_name,
1169 POSTID(&oinfo->oi_oa->o_oi),
1170 POSTID(&req->rq_oi.oi_oa->o_oi),
1171 req->rq_idx, rc);
1172 goto out;
1173 }
1174 }
1175
1176 if (!list_empty(&rqset->set_requests)) {
1177 LASSERT(rc == 0);
1178 LASSERT(!rqset->set_interpret);
1179 rqset->set_interpret = lov_getattr_interpret;
1180 rqset->set_arg = (void *)lovset;
1181 return rc;
1182 }
1183 out:
1184 if (rc)
1185 atomic_set(&lovset->set_completes, 0);
1186 err = lov_fini_getattr_set(lovset);
1187 return rc ? rc : err;
1188 }
1189
1190 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1191 void *data, int rc)
1192 {
1193 struct lov_request_set *lovset = (struct lov_request_set *)data;
1194 int err;
1195
1196 if (rc)
1197 atomic_set(&lovset->set_completes, 0);
1198 err = lov_fini_setattr_set(lovset);
1199 return rc ? rc : err;
1200 }
1201
1202 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1203 * needed. Otherwise, a client is waiting for responses.
1204 */
1205 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1206 struct obd_trans_info *oti,
1207 struct ptlrpc_request_set *rqset)
1208 {
1209 struct lov_request_set *set;
1210 struct lov_request *req;
1211 struct lov_obd *lov;
1212 int rc = 0;
1213
1214 LASSERT(oinfo);
1215 ASSERT_LSM_MAGIC(oinfo->oi_md);
1216 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1217 LASSERT(oti);
1218 LASSERT(oti->oti_logcookies);
1219 }
1220
1221 if (!exp || !exp->exp_obd)
1222 return -ENODEV;
1223
1224 lov = &exp->exp_obd->u.lov;
1225 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1226 if (rc)
1227 return rc;
1228
1229 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1230 POSTID(&oinfo->oi_md->lsm_oi),
1231 oinfo->oi_md->lsm_stripe_count,
1232 oinfo->oi_md->lsm_stripe_size);
1233
1234 list_for_each_entry(req, &set->set_list, rq_link) {
1235 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1236 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1237
1238 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1239 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1240 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1241
1242 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1243 &req->rq_oi, oti, rqset);
1244 if (rc) {
1245 CERROR("error: setattr objid "DOSTID" subobj"
1246 DOSTID" on OST idx %d: rc = %d\n",
1247 POSTID(&set->set_oi->oi_oa->o_oi),
1248 POSTID(&req->rq_oi.oi_oa->o_oi),
1249 req->rq_idx, rc);
1250 break;
1251 }
1252 }
1253
1254 /* If we are not waiting for responses on async requests, return. */
1255 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1256 int err;
1257
1258 if (rc)
1259 atomic_set(&set->set_completes, 0);
1260 err = lov_fini_setattr_set(set);
1261 return rc ? rc : err;
1262 }
1263
1264 LASSERT(!rqset->set_interpret);
1265 rqset->set_interpret = lov_setattr_interpret;
1266 rqset->set_arg = (void *)set;
1267
1268 return 0;
1269 }
1270
1271 /* find any ldlm lock of the inode in lov
1272 * return 0 not find
1273 * 1 find one
1274 * < 0 error
1275 */
1276 static int lov_find_cbdata(struct obd_export *exp,
1277 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1278 void *data)
1279 {
1280 struct lov_obd *lov;
1281 int rc = 0, i;
1282
1283 ASSERT_LSM_MAGIC(lsm);
1284
1285 if (!exp || !exp->exp_obd)
1286 return -ENODEV;
1287
1288 lov = &exp->exp_obd->u.lov;
1289 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1290 struct lov_stripe_md submd;
1291 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1292
1293 if (lov_oinfo_is_dummy(loi))
1294 continue;
1295
1296 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1297 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1298 continue;
1299 }
1300
1301 submd.lsm_oi = loi->loi_oi;
1302 submd.lsm_stripe_count = 0;
1303 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1304 &submd, it, data);
1305 if (rc != 0)
1306 return rc;
1307 }
1308 return rc;
1309 }
1310
1311 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1312 {
1313 struct lov_request_set *lovset = (struct lov_request_set *)data;
1314 int err;
1315
1316 if (rc)
1317 atomic_set(&lovset->set_completes, 0);
1318
1319 err = lov_fini_statfs_set(lovset);
1320 return rc ? rc : err;
1321 }
1322
1323 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1324 __u64 max_age, struct ptlrpc_request_set *rqset)
1325 {
1326 struct obd_device *obd = class_exp2obd(exp);
1327 struct lov_request_set *set;
1328 struct lov_request *req;
1329 struct lov_obd *lov;
1330 int rc = 0;
1331
1332 LASSERT(oinfo->oi_osfs);
1333
1334 lov = &obd->u.lov;
1335 rc = lov_prep_statfs_set(obd, oinfo, &set);
1336 if (rc)
1337 return rc;
1338
1339 list_for_each_entry(req, &set->set_list, rq_link) {
1340 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1341 &req->rq_oi, max_age, rqset);
1342 if (rc)
1343 break;
1344 }
1345
1346 if (rc || list_empty(&rqset->set_requests)) {
1347 int err;
1348
1349 if (rc)
1350 atomic_set(&set->set_completes, 0);
1351 err = lov_fini_statfs_set(set);
1352 return rc ? rc : err;
1353 }
1354
1355 LASSERT(!rqset->set_interpret);
1356 rqset->set_interpret = lov_statfs_interpret;
1357 rqset->set_arg = (void *)set;
1358 return 0;
1359 }
1360
1361 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1362 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1363 {
1364 struct ptlrpc_request_set *set = NULL;
1365 struct obd_info oinfo = { };
1366 int rc = 0;
1367
1368 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1369 * statfs requests
1370 */
1371 set = ptlrpc_prep_set();
1372 if (!set)
1373 return -ENOMEM;
1374
1375 oinfo.oi_osfs = osfs;
1376 oinfo.oi_flags = flags;
1377 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1378 if (rc == 0)
1379 rc = ptlrpc_set_wait(set);
1380 ptlrpc_set_destroy(set);
1381
1382 return rc;
1383 }
1384
1385 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1386 void *karg, void __user *uarg)
1387 {
1388 struct obd_device *obddev = class_exp2obd(exp);
1389 struct lov_obd *lov = &obddev->u.lov;
1390 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1391 struct obd_uuid *uuidp;
1392
1393 switch (cmd) {
1394 case IOC_OBD_STATFS: {
1395 struct obd_ioctl_data *data = karg;
1396 struct obd_device *osc_obd;
1397 struct obd_statfs stat_buf = {0};
1398 __u32 index;
1399 __u32 flags;
1400
1401 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1402 if (index >= count)
1403 return -ENODEV;
1404
1405 if (!lov->lov_tgts[index])
1406 /* Try again with the next index */
1407 return -EAGAIN;
1408 if (!lov->lov_tgts[index]->ltd_active)
1409 return -ENODATA;
1410
1411 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1412 if (!osc_obd)
1413 return -EINVAL;
1414
1415 /* copy UUID */
1416 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1417 min((int)data->ioc_plen2,
1418 (int)sizeof(struct obd_uuid))))
1419 return -EFAULT;
1420
1421 memcpy(&flags, data->ioc_inlbuf1, sizeof(__u32));
1422 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1423
1424 /* got statfs data */
1425 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1426 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1427 flags);
1428 if (rc)
1429 return rc;
1430 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1431 min((int)data->ioc_plen1,
1432 (int)sizeof(stat_buf))))
1433 return -EFAULT;
1434 break;
1435 }
1436 case OBD_IOC_LOV_GET_CONFIG: {
1437 struct obd_ioctl_data *data;
1438 struct lov_desc *desc;
1439 char *buf = NULL;
1440 __u32 *genp;
1441
1442 len = 0;
1443 if (obd_ioctl_getdata(&buf, &len, uarg))
1444 return -EINVAL;
1445
1446 data = (struct obd_ioctl_data *)buf;
1447
1448 if (sizeof(*desc) > data->ioc_inllen1) {
1449 obd_ioctl_freedata(buf, len);
1450 return -EINVAL;
1451 }
1452
1453 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1454 obd_ioctl_freedata(buf, len);
1455 return -EINVAL;
1456 }
1457
1458 if (sizeof(__u32) * count > data->ioc_inllen3) {
1459 obd_ioctl_freedata(buf, len);
1460 return -EINVAL;
1461 }
1462
1463 desc = (struct lov_desc *)data->ioc_inlbuf1;
1464 memcpy(desc, &(lov->desc), sizeof(*desc));
1465
1466 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1467 genp = (__u32 *)data->ioc_inlbuf3;
1468 /* the uuid will be empty for deleted OSTs */
1469 for (i = 0; i < count; i++, uuidp++, genp++) {
1470 if (!lov->lov_tgts[i])
1471 continue;
1472 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1473 *genp = lov->lov_tgts[i]->ltd_gen;
1474 }
1475
1476 if (copy_to_user(uarg, buf, len))
1477 rc = -EFAULT;
1478 obd_ioctl_freedata(buf, len);
1479 break;
1480 }
1481 case LL_IOC_LOV_GETSTRIPE:
1482 rc = lov_getstripe(exp, karg, uarg);
1483 break;
1484 case OBD_IOC_QUOTACTL: {
1485 struct if_quotactl *qctl = karg;
1486 struct lov_tgt_desc *tgt = NULL;
1487 struct obd_quotactl *oqctl;
1488
1489 if (qctl->qc_valid == QC_OSTIDX) {
1490 if (count <= qctl->qc_idx)
1491 return -EINVAL;
1492
1493 tgt = lov->lov_tgts[qctl->qc_idx];
1494 if (!tgt || !tgt->ltd_exp)
1495 return -EINVAL;
1496 } else if (qctl->qc_valid == QC_UUID) {
1497 for (i = 0; i < count; i++) {
1498 tgt = lov->lov_tgts[i];
1499 if (!tgt ||
1500 !obd_uuid_equals(&tgt->ltd_uuid,
1501 &qctl->obd_uuid))
1502 continue;
1503
1504 if (!tgt->ltd_exp)
1505 return -EINVAL;
1506
1507 break;
1508 }
1509 } else {
1510 return -EINVAL;
1511 }
1512
1513 if (i >= count)
1514 return -EAGAIN;
1515
1516 LASSERT(tgt && tgt->ltd_exp);
1517 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1518 if (!oqctl)
1519 return -ENOMEM;
1520
1521 QCTL_COPY(oqctl, qctl);
1522 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1523 if (rc == 0) {
1524 QCTL_COPY(qctl, oqctl);
1525 qctl->qc_valid = QC_OSTIDX;
1526 qctl->obd_uuid = tgt->ltd_uuid;
1527 }
1528 kfree(oqctl);
1529 break;
1530 }
1531 default: {
1532 int set = 0;
1533
1534 if (count == 0)
1535 return -ENOTTY;
1536
1537 for (i = 0; i < count; i++) {
1538 int err;
1539 struct obd_device *osc_obd;
1540
1541 /* OST was disconnected */
1542 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1543 continue;
1544
1545 /* ll_umount_begin() sets force flag but for lov, not
1546 * osc. Let's pass it through
1547 */
1548 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1549 osc_obd->obd_force = obddev->obd_force;
1550 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1551 len, karg, uarg);
1552 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK)
1553 return err;
1554 if (err) {
1555 if (lov->lov_tgts[i]->ltd_active) {
1556 CDEBUG(err == -ENOTTY ?
1557 D_IOCTL : D_WARNING,
1558 "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1559 lov_uuid2str(lov, i),
1560 i, cmd, err);
1561 if (!rc)
1562 rc = err;
1563 }
1564 } else {
1565 set = 1;
1566 }
1567 }
1568 if (!set && !rc)
1569 rc = -EIO;
1570 }
1571 }
1572
1573 return rc;
1574 }
1575
1576 #define FIEMAP_BUFFER_SIZE 4096
1577
1578 /**
1579 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1580 * call. The local end offset and the device are sent in the first
1581 * fm_extent. This function calculates the stripe number from the index.
1582 * This function returns a stripe_no on which mapping is to be restarted.
1583 *
1584 * This function returns fm_end_offset which is the in-OST offset at which
1585 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1586 * will re-calculate proper offset in next stripe.
1587 * Note that the first extent is passed to lov_get_info via the value field.
1588 *
1589 * \param fiemap fiemap request header
1590 * \param lsm striping information for the file
1591 * \param fm_start logical start of mapping
1592 * \param fm_end logical end of mapping
1593 * \param start_stripe starting stripe will be returned in this
1594 */
1595 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1596 struct lov_stripe_md *lsm, u64 fm_start,
1597 u64 fm_end, int *start_stripe)
1598 {
1599 u64 local_end = fiemap->fm_extents[0].fe_logical;
1600 u64 lun_start, lun_end;
1601 u64 fm_end_offset;
1602 int stripe_no = -1, i;
1603
1604 if (fiemap->fm_extent_count == 0 ||
1605 fiemap->fm_extents[0].fe_logical == 0)
1606 return 0;
1607
1608 /* Find out stripe_no from ost_index saved in the fe_device */
1609 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1610 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1611
1612 if (lov_oinfo_is_dummy(oinfo))
1613 continue;
1614
1615 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1616 stripe_no = i;
1617 break;
1618 }
1619 }
1620 if (stripe_no == -1)
1621 return -EINVAL;
1622
1623 /* If we have finished mapping on previous device, shift logical
1624 * offset to start of next device
1625 */
1626 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1627 &lun_start, &lun_end)) != 0 &&
1628 local_end < lun_end) {
1629 fm_end_offset = local_end;
1630 *start_stripe = stripe_no;
1631 } else {
1632 /* This is a special value to indicate that caller should
1633 * calculate offset in next stripe.
1634 */
1635 fm_end_offset = 0;
1636 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1637 }
1638
1639 return fm_end_offset;
1640 }
1641
1642 /**
1643 * We calculate on which OST the mapping will end. If the length of mapping
1644 * is greater than (stripe_size * stripe_count) then the last_stripe will
1645 * will be one just before start_stripe. Else we check if the mapping
1646 * intersects each OST and find last_stripe.
1647 * This function returns the last_stripe and also sets the stripe_count
1648 * over which the mapping is spread
1649 *
1650 * \param lsm striping information for the file
1651 * \param fm_start logical start of mapping
1652 * \param fm_end logical end of mapping
1653 * \param start_stripe starting stripe of the mapping
1654 * \param stripe_count the number of stripes across which to map is returned
1655 *
1656 * \retval last_stripe return the last stripe of the mapping
1657 */
1658 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1659 u64 fm_end, int start_stripe,
1660 int *stripe_count)
1661 {
1662 int last_stripe;
1663 u64 obd_start, obd_end;
1664 int i, j;
1665
1666 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1667 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1668 start_stripe - 1;
1669 *stripe_count = lsm->lsm_stripe_count;
1670 } else {
1671 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1672 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1673 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1674 &obd_start, &obd_end)) == 0)
1675 break;
1676 }
1677 *stripe_count = j;
1678 last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count;
1679 }
1680
1681 return last_stripe;
1682 }
1683
1684 /**
1685 * Set fe_device and copy extents from local buffer into main return buffer.
1686 *
1687 * \param fiemap fiemap request header
1688 * \param lcl_fm_ext array of local fiemap extents to be copied
1689 * \param ost_index OST index to be written into the fm_device field for each
1690 extent
1691 * \param ext_count number of extents to be copied
1692 * \param current_extent where to start copying in main extent array
1693 */
1694 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1695 struct ll_fiemap_extent *lcl_fm_ext,
1696 int ost_index, unsigned int ext_count,
1697 int current_extent)
1698 {
1699 char *to;
1700 int ext;
1701
1702 for (ext = 0; ext < ext_count; ext++) {
1703 lcl_fm_ext[ext].fe_device = ost_index;
1704 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1705 }
1706
1707 /* Copy fm_extent's from fm_local to return buffer */
1708 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1709 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1710 }
1711
1712 /**
1713 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1714 * This also handles the restarting of FIEMAP calls in case mapping overflows
1715 * the available number of extents in single call.
1716 */
1717 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1718 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1719 {
1720 struct ll_fiemap_info_key *fm_key = key;
1721 struct ll_user_fiemap *fiemap = val;
1722 struct ll_user_fiemap *fm_local = NULL;
1723 struct ll_fiemap_extent *lcl_fm_ext;
1724 int count_local;
1725 unsigned int get_num_extents = 0;
1726 int ost_index = 0, actual_start_stripe, start_stripe;
1727 u64 fm_start, fm_end, fm_length, fm_end_offset;
1728 u64 curr_loc;
1729 int current_extent = 0, rc = 0, i;
1730 int ost_eof = 0; /* EOF for object */
1731 int ost_done = 0; /* done with required mapping for this OST? */
1732 int last_stripe;
1733 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1734 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1735
1736 if (!lsm_has_objects(lsm)) {
1737 rc = 0;
1738 goto out;
1739 }
1740
1741 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1742 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1743
1744 fm_local = libcfs_kvzalloc(buffer_size, GFP_NOFS);
1745 if (!fm_local) {
1746 rc = -ENOMEM;
1747 goto out;
1748 }
1749 lcl_fm_ext = &fm_local->fm_extents[0];
1750
1751 count_local = fiemap_size_to_count(buffer_size);
1752
1753 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1754 fm_start = fiemap->fm_start;
1755 fm_length = fiemap->fm_length;
1756 /* Calculate start stripe, last stripe and length of mapping */
1757 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1758 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1759 fm_start + fm_length - 1);
1760 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1761 if (fm_end > fm_key->oa.o_size)
1762 fm_end = fm_key->oa.o_size;
1763
1764 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1765 actual_start_stripe,
1766 &stripe_count);
1767
1768 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1769 fm_end, &start_stripe);
1770 if (fm_end_offset == -EINVAL) {
1771 rc = -EINVAL;
1772 goto out;
1773 }
1774
1775 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1776 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1777 if (fiemap->fm_extent_count == 0) {
1778 get_num_extents = 1;
1779 count_local = 0;
1780 }
1781 /* Check each stripe */
1782 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1783 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1784 u64 req_fm_len; /* Stores length of required mapping */
1785 u64 len_mapped_single_call;
1786 u64 lun_start, lun_end, obd_object_end;
1787 unsigned int ext_count;
1788
1789 cur_stripe_wrap = cur_stripe;
1790
1791 /* Find out range of mapping on this stripe */
1792 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1793 &lun_start, &obd_object_end)) == 0)
1794 continue;
1795
1796 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1797 rc = -EIO;
1798 goto out;
1799 }
1800
1801 /* If this is a continuation FIEMAP call and we are on
1802 * starting stripe then lun_start needs to be set to
1803 * fm_end_offset
1804 */
1805 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1806 lun_start = fm_end_offset;
1807
1808 if (fm_length != ~0ULL) {
1809 /* Handle fm_start + fm_length overflow */
1810 if (fm_start + fm_length < fm_start)
1811 fm_length = ~0ULL - fm_start;
1812 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1813 cur_stripe);
1814 } else {
1815 lun_end = ~0ULL;
1816 }
1817
1818 if (lun_start == lun_end)
1819 continue;
1820
1821 req_fm_len = obd_object_end - lun_start;
1822 fm_local->fm_length = 0;
1823 len_mapped_single_call = 0;
1824
1825 /* If the output buffer is very large and the objects have many
1826 * extents we may need to loop on a single OST repeatedly
1827 */
1828 ost_eof = 0;
1829 ost_done = 0;
1830 do {
1831 if (get_num_extents == 0) {
1832 /* Don't get too many extents. */
1833 if (current_extent + count_local >
1834 fiemap->fm_extent_count)
1835 count_local = fiemap->fm_extent_count -
1836 current_extent;
1837 }
1838
1839 lun_start += len_mapped_single_call;
1840 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1841 req_fm_len = fm_local->fm_length;
1842 fm_local->fm_extent_count = count_local;
1843 fm_local->fm_mapped_extents = 0;
1844 fm_local->fm_flags = fiemap->fm_flags;
1845
1846 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1847 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1848
1849 if (ost_index < 0 ||
1850 ost_index >= lov->desc.ld_tgt_count) {
1851 rc = -EINVAL;
1852 goto out;
1853 }
1854
1855 /* If OST is inactive, return extent with UNKNOWN flag */
1856 if (!lov->lov_tgts[ost_index]->ltd_active) {
1857 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1858 fm_local->fm_mapped_extents = 1;
1859
1860 lcl_fm_ext[0].fe_logical = lun_start;
1861 lcl_fm_ext[0].fe_length = obd_object_end -
1862 lun_start;
1863 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1864
1865 goto inactive_tgt;
1866 }
1867
1868 fm_local->fm_start = lun_start;
1869 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1870 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1871 *vallen = fiemap_count_to_size(fm_local->fm_extent_count);
1872 rc = obd_get_info(NULL,
1873 lov->lov_tgts[ost_index]->ltd_exp,
1874 keylen, key, vallen, fm_local, lsm);
1875 if (rc != 0)
1876 goto out;
1877
1878 inactive_tgt:
1879 ext_count = fm_local->fm_mapped_extents;
1880 if (ext_count == 0) {
1881 ost_done = 1;
1882 /* If last stripe has hole at the end,
1883 * then we need to return
1884 */
1885 if (cur_stripe_wrap == last_stripe) {
1886 fiemap->fm_mapped_extents = 0;
1887 goto finish;
1888 }
1889 break;
1890 }
1891
1892 /* If we just need num of extents then go to next device */
1893 if (get_num_extents) {
1894 current_extent += ext_count;
1895 break;
1896 }
1897
1898 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1899 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1900
1901 /* Have we finished mapping on this device? */
1902 if (req_fm_len <= len_mapped_single_call)
1903 ost_done = 1;
1904
1905 /* Clear the EXTENT_LAST flag which can be present on
1906 * last extent
1907 */
1908 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1909 lcl_fm_ext[ext_count - 1].fe_flags &=
1910 ~FIEMAP_EXTENT_LAST;
1911
1912 curr_loc = lov_stripe_size(lsm,
1913 lcl_fm_ext[ext_count - 1].fe_logical+
1914 lcl_fm_ext[ext_count - 1].fe_length,
1915 cur_stripe);
1916 if (curr_loc >= fm_key->oa.o_size)
1917 ost_eof = 1;
1918
1919 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1920 ost_index, ext_count,
1921 current_extent);
1922
1923 current_extent += ext_count;
1924
1925 /* Ran out of available extents? */
1926 if (current_extent >= fiemap->fm_extent_count)
1927 goto finish;
1928 } while (ost_done == 0 && ost_eof == 0);
1929
1930 if (cur_stripe_wrap == last_stripe)
1931 goto finish;
1932 }
1933
1934 finish:
1935 /* Indicate that we are returning device offsets unless file just has
1936 * single stripe
1937 */
1938 if (lsm->lsm_stripe_count > 1)
1939 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1940
1941 if (get_num_extents)
1942 goto skip_last_device_calc;
1943
1944 /* Check if we have reached the last stripe and whether mapping for that
1945 * stripe is done.
1946 */
1947 if (cur_stripe_wrap == last_stripe) {
1948 if (ost_done || ost_eof)
1949 fiemap->fm_extents[current_extent - 1].fe_flags |=
1950 FIEMAP_EXTENT_LAST;
1951 }
1952
1953 skip_last_device_calc:
1954 fiemap->fm_mapped_extents = current_extent;
1955
1956 out:
1957 kvfree(fm_local);
1958 return rc;
1959 }
1960
1961 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1962 __u32 keylen, void *key, __u32 *vallen, void *val,
1963 struct lov_stripe_md *lsm)
1964 {
1965 struct obd_device *obddev = class_exp2obd(exp);
1966 struct lov_obd *lov = &obddev->u.lov;
1967 int i, rc;
1968
1969 if (!vallen || !val)
1970 return -EFAULT;
1971
1972 obd_getref(obddev);
1973
1974 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
1975 struct {
1976 char name[16];
1977 struct ldlm_lock *lock;
1978 } *data = key;
1979 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
1980 struct lov_oinfo *loi;
1981 __u32 *stripe = val;
1982
1983 if (*vallen < sizeof(*stripe)) {
1984 rc = -EFAULT;
1985 goto out;
1986 }
1987 *vallen = sizeof(*stripe);
1988
1989 /* XXX This is another one of those bits that will need to
1990 * change if we ever actually support nested LOVs. It uses
1991 * the lock's export to find out which stripe it is.
1992 */
1993 /* XXX - it's assumed all the locks for deleted OSTs have
1994 * been cancelled. Also, the export for deleted OSTs will
1995 * be NULL and won't match the lock's export.
1996 */
1997 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1998 loi = lsm->lsm_oinfo[i];
1999 if (lov_oinfo_is_dummy(loi))
2000 continue;
2001
2002 if (!lov->lov_tgts[loi->loi_ost_idx])
2003 continue;
2004 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2005 data->lock->l_conn_export &&
2006 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2007 *stripe = i;
2008 rc = 0;
2009 goto out;
2010 }
2011 }
2012 LDLM_ERROR(data->lock, "lock on inode without such object");
2013 dump_lsm(D_ERROR, lsm);
2014 rc = -ENXIO;
2015 goto out;
2016 } else if (KEY_IS(KEY_LAST_ID)) {
2017 struct obd_id_info *info = val;
2018 __u32 size = sizeof(u64);
2019 struct lov_tgt_desc *tgt;
2020
2021 LASSERT(*vallen == sizeof(struct obd_id_info));
2022 tgt = lov->lov_tgts[info->idx];
2023
2024 if (!tgt || !tgt->ltd_active) {
2025 rc = -ESRCH;
2026 goto out;
2027 }
2028
2029 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2030 &size, info->data, NULL);
2031 rc = 0;
2032 goto out;
2033 } else if (KEY_IS(KEY_LOVDESC)) {
2034 struct lov_desc *desc_ret = val;
2035 *desc_ret = lov->desc;
2036
2037 rc = 0;
2038 goto out;
2039 } else if (KEY_IS(KEY_FIEMAP)) {
2040 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2041 goto out;
2042 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2043 struct lov_tgt_desc *tgt;
2044 __u64 ost_idx = *((__u64 *)val);
2045
2046 LASSERT(*vallen == sizeof(__u64));
2047 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2048 tgt = lov->lov_tgts[ost_idx];
2049
2050 if (!tgt || !tgt->ltd_exp) {
2051 rc = -ESRCH;
2052 goto out;
2053 }
2054
2055 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2056 rc = 0;
2057 goto out;
2058 } else if (KEY_IS(KEY_TGT_COUNT)) {
2059 *((int *)val) = lov->desc.ld_tgt_count;
2060 rc = 0;
2061 goto out;
2062 }
2063
2064 rc = -EINVAL;
2065
2066 out:
2067 obd_putref(obddev);
2068 return rc;
2069 }
2070
2071 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2072 u32 keylen, void *key, u32 vallen,
2073 void *val, struct ptlrpc_request_set *set)
2074 {
2075 struct obd_device *obddev = class_exp2obd(exp);
2076 struct lov_obd *lov = &obddev->u.lov;
2077 u32 count;
2078 int i, rc = 0, err;
2079 struct lov_tgt_desc *tgt;
2080 unsigned incr, check_uuid,
2081 do_inactive, no_set;
2082 unsigned next_id = 0, mds_con = 0;
2083
2084 incr = check_uuid = do_inactive = no_set = 0;
2085 if (!set) {
2086 no_set = 1;
2087 set = ptlrpc_prep_set();
2088 if (!set)
2089 return -ENOMEM;
2090 }
2091
2092 obd_getref(obddev);
2093 count = lov->desc.ld_tgt_count;
2094
2095 if (KEY_IS(KEY_NEXT_ID)) {
2096 count = vallen / sizeof(struct obd_id_info);
2097 vallen = sizeof(u64);
2098 incr = sizeof(struct obd_id_info);
2099 do_inactive = 1;
2100 next_id = 1;
2101 } else if (KEY_IS(KEY_CHECKSUM)) {
2102 do_inactive = 1;
2103 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2104 /* use defaults: do_inactive = incr = 0; */
2105 } else if (KEY_IS(KEY_MDS_CONN)) {
2106 mds_con = 1;
2107 } else if (KEY_IS(KEY_CACHE_SET)) {
2108 LASSERT(!lov->lov_cache);
2109 lov->lov_cache = val;
2110 do_inactive = 1;
2111 }
2112
2113 for (i = 0; i < count; i++, val = (char *)val + incr) {
2114 if (next_id)
2115 tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx];
2116 else
2117 tgt = lov->lov_tgts[i];
2118 /* OST was disconnected */
2119 if (!tgt || !tgt->ltd_exp)
2120 continue;
2121
2122 /* OST is inactive and we don't want inactive OSCs */
2123 if (!tgt->ltd_active && !do_inactive)
2124 continue;
2125
2126 if (mds_con) {
2127 struct mds_group_info *mgi;
2128
2129 LASSERT(vallen == sizeof(*mgi));
2130 mgi = (struct mds_group_info *)val;
2131
2132 /* Only want a specific OSC */
2133 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2134 &tgt->ltd_uuid))
2135 continue;
2136
2137 err = obd_set_info_async(env, tgt->ltd_exp,
2138 keylen, key, sizeof(int),
2139 &mgi->group, set);
2140 } else if (next_id) {
2141 err = obd_set_info_async(env, tgt->ltd_exp,
2142 keylen, key, vallen,
2143 ((struct obd_id_info *)val)->data, set);
2144 } else {
2145 /* Only want a specific OSC */
2146 if (check_uuid &&
2147 !obd_uuid_equals(val, &tgt->ltd_uuid))
2148 continue;
2149
2150 err = obd_set_info_async(env, tgt->ltd_exp,
2151 keylen, key, vallen, val, set);
2152 }
2153
2154 if (!rc)
2155 rc = err;
2156 }
2157
2158 obd_putref(obddev);
2159 if (no_set) {
2160 err = ptlrpc_set_wait(set);
2161 if (!rc)
2162 rc = err;
2163 ptlrpc_set_destroy(set);
2164 }
2165 return rc;
2166 }
2167
2168 void lov_stripe_lock(struct lov_stripe_md *md)
2169 __acquires(&md->lsm_lock)
2170 {
2171 LASSERT(md->lsm_lock_owner != current_pid());
2172 spin_lock(&md->lsm_lock);
2173 LASSERT(md->lsm_lock_owner == 0);
2174 md->lsm_lock_owner = current_pid();
2175 }
2176 EXPORT_SYMBOL(lov_stripe_lock);
2177
2178 void lov_stripe_unlock(struct lov_stripe_md *md)
2179 __releases(&md->lsm_lock)
2180 {
2181 LASSERT(md->lsm_lock_owner == current_pid());
2182 md->lsm_lock_owner = 0;
2183 spin_unlock(&md->lsm_lock);
2184 }
2185 EXPORT_SYMBOL(lov_stripe_unlock);
2186
2187 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2188 struct obd_quotactl *oqctl)
2189 {
2190 struct lov_obd *lov = &obd->u.lov;
2191 struct lov_tgt_desc *tgt;
2192 __u64 curspace = 0;
2193 __u64 bhardlimit = 0;
2194 int i, rc = 0;
2195
2196 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2197 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2198 oqctl->qc_cmd != Q_GETOQUOTA &&
2199 oqctl->qc_cmd != Q_INITQUOTA &&
2200 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2201 oqctl->qc_cmd != Q_FINVALIDATE) {
2202 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2203 return -EFAULT;
2204 }
2205
2206 /* for lov tgt */
2207 obd_getref(obd);
2208 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2209 int err;
2210
2211 tgt = lov->lov_tgts[i];
2212
2213 if (!tgt)
2214 continue;
2215
2216 if (!tgt->ltd_active || tgt->ltd_reap) {
2217 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2218 lov->lov_tgts[i]->ltd_activate) {
2219 rc = -EREMOTEIO;
2220 CERROR("ost %d is inactive\n", i);
2221 } else {
2222 CDEBUG(D_HA, "ost %d is inactive\n", i);
2223 }
2224 continue;
2225 }
2226
2227 err = obd_quotactl(tgt->ltd_exp, oqctl);
2228 if (err) {
2229 if (tgt->ltd_active && !rc)
2230 rc = err;
2231 continue;
2232 }
2233
2234 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2235 curspace += oqctl->qc_dqblk.dqb_curspace;
2236 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2237 }
2238 }
2239 obd_putref(obd);
2240
2241 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2242 oqctl->qc_dqblk.dqb_curspace = curspace;
2243 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2244 }
2245 return rc;
2246 }
2247
2248 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2249 struct obd_quotactl *oqctl)
2250 {
2251 struct lov_obd *lov = &obd->u.lov;
2252 int i, rc = 0;
2253
2254 obd_getref(obd);
2255
2256 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2257 if (!lov->lov_tgts[i])
2258 continue;
2259
2260 /* Skip quota check on the administratively disabled OSTs. */
2261 if (!lov->lov_tgts[i]->ltd_activate) {
2262 CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
2263 i);
2264 continue;
2265 }
2266
2267 if (!lov->lov_tgts[i]->ltd_active) {
2268 CERROR("lov idx %d inactive\n", i);
2269 rc = -EIO;
2270 goto out;
2271 }
2272 }
2273
2274 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2275 int err;
2276
2277 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2278 continue;
2279
2280 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2281 if (err && !rc)
2282 rc = err;
2283 }
2284
2285 out:
2286 obd_putref(obd);
2287
2288 return rc;
2289 }
2290
2291 static struct obd_ops lov_obd_ops = {
2292 .owner = THIS_MODULE,
2293 .setup = lov_setup,
2294 .precleanup = lov_precleanup,
2295 .cleanup = lov_cleanup,
2296 /*.process_config = lov_process_config,*/
2297 .connect = lov_connect,
2298 .disconnect = lov_disconnect,
2299 .statfs = lov_statfs,
2300 .statfs_async = lov_statfs_async,
2301 .packmd = lov_packmd,
2302 .unpackmd = lov_unpackmd,
2303 .create = lov_create,
2304 .destroy = lov_destroy,
2305 .getattr_async = lov_getattr_async,
2306 .setattr_async = lov_setattr_async,
2307 .adjust_kms = lov_adjust_kms,
2308 .find_cbdata = lov_find_cbdata,
2309 .iocontrol = lov_iocontrol,
2310 .get_info = lov_get_info,
2311 .set_info_async = lov_set_info_async,
2312 .notify = lov_notify,
2313 .pool_new = lov_pool_new,
2314 .pool_rem = lov_pool_remove,
2315 .pool_add = lov_pool_add,
2316 .pool_del = lov_pool_del,
2317 .getref = lov_getref,
2318 .putref = lov_putref,
2319 .quotactl = lov_quotactl,
2320 .quotacheck = lov_quotacheck,
2321 };
2322
2323 struct kmem_cache *lov_oinfo_slab;
2324
2325 static int __init lov_init(void)
2326 {
2327 struct lprocfs_static_vars lvars = { NULL };
2328 int rc;
2329
2330 /* print an address of _any_ initialized kernel symbol from this
2331 * module, to allow debugging with gdb that doesn't support data
2332 * symbols from modules.
2333 */
2334 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2335
2336 rc = lu_kmem_init(lov_caches);
2337 if (rc)
2338 return rc;
2339
2340 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2341 sizeof(struct lov_oinfo),
2342 0, SLAB_HWCACHE_ALIGN, NULL);
2343 if (!lov_oinfo_slab) {
2344 lu_kmem_fini(lov_caches);
2345 return -ENOMEM;
2346 }
2347 lprocfs_lov_init_vars(&lvars);
2348
2349 rc = class_register_type(&lov_obd_ops, NULL,
2350 LUSTRE_LOV_NAME, &lov_device_type);
2351
2352 if (rc) {
2353 kmem_cache_destroy(lov_oinfo_slab);
2354 lu_kmem_fini(lov_caches);
2355 }
2356
2357 return rc;
2358 }
2359
2360 static void /*__exit*/ lov_exit(void)
2361 {
2362 class_unregister_type(LUSTRE_LOV_NAME);
2363 kmem_cache_destroy(lov_oinfo_slab);
2364
2365 lu_kmem_fini(lov_caches);
2366 }
2367
2368 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2369 MODULE_DESCRIPTION("Lustre Logical Object Volume");
2370 MODULE_LICENSE("GPL");
2371 MODULE_VERSION(LUSTRE_VERSION_STRING);
2372
2373 module_init(lov_init);
2374 module_exit(lov_exit);
ubuntu-bionic-kernel mirror