projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging: lustre: remove ENTRY/EXIT_NESTING stuff
[deliverable/linux.git] / drivers / staging / lustre / lustre / lov / lov_obd.c
CommitLineData
d7e09d03
PT		 1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * lustre/lov/lov_obd.c
37 *
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
42 */
43
44#define DEBUG_SUBSYSTEM S_LOV
45#include <linux/libcfs/libcfs.h>
46
47#include <obd_support.h>
48#include <lustre_lib.h>
49#include <lustre_net.h>
50#include <lustre/lustre_idl.h>
51#include <lustre_dlm.h>
52#include <lustre_mds.h>
53#include <lustre_debug.h>
54#include <obd_class.h>
55#include <obd_lov.h>
56#include <obd_ost.h>
57#include <lprocfs_status.h>
58#include <lustre_param.h>
59#include <cl_object.h>
c52f69c5 60#include <lclient.h> /* for cl_client_lru */
d7e09d03
PT		 61#include <lustre/ll_fiemap.h>
62#include <lustre_log.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. */
69static 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 atomic_inc(&lov->lov_refcount);
76 mutex_unlock(&lov->lov_lock);
77 return;
78}
79
80static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
81
82static 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 (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
89 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 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 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
110 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
119static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
120 enum obd_notify_event ev);
121static int lov_notify(struct obd_device *obd, struct obd_device *watched,
122 enum obd_notify_event ev, void *data);
123
124
125#define MAX_STRING_SIZE 128
126int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
127 struct obd_connect_data *data)
128{
129 struct lov_obd *lov = &obd->u.lov;
130 struct obd_uuid *tgt_uuid;
131 struct obd_device *tgt_obd;
132 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
133 struct obd_import *imp;
134 proc_dir_entry_t *lov_proc_dir;
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
73bb1da6 194 lov_proc_dir = obd->obd_proc_private;
d7e09d03
PT		 195 if (lov_proc_dir) {
196 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
197 proc_dir_entry_t *osc_symlink;
198
199 LASSERT(osc_obd != NULL);
200 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
201 LASSERT(osc_obd->obd_type->typ_name != NULL);
202
203 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
204 lov_proc_dir,
205 "../../../%s/%s",
206 osc_obd->obd_type->typ_name,
207 osc_obd->obd_name);
208 if (osc_symlink == NULL) {
209 CERROR("could not register LOV target "
210 "/proc/fs/lustre/%s/%s/target_obds/%s.",
211 obd->obd_type->typ_name, obd->obd_name,
212 osc_obd->obd_name);
213 lprocfs_remove(&lov_proc_dir);
73bb1da6 214 obd->obd_proc_private = NULL;
d7e09d03
PT		 215 }
216 }
217
218 RETURN(0);
219}
220
221static int lov_connect(const struct lu_env *env,
222 struct obd_export **exp, struct obd_device *obd,
223 struct obd_uuid *cluuid, struct obd_connect_data *data,
224 void *localdata)
225{
226 struct lov_obd *lov = &obd->u.lov;
227 struct lov_tgt_desc *tgt;
228 struct lustre_handle conn;
229 int i, rc;
230 ENTRY;
231
232 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
233
234 rc = class_connect(&conn, obd, cluuid);
235 if (rc)
236 RETURN(rc);
237
238 *exp = class_conn2export(&conn);
239
240 /* Why should there ever be more than 1 connect? */
241 lov->lov_connects++;
242 LASSERT(lov->lov_connects == 1);
243
244 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
245 if (data)
246 lov->lov_ocd = *data;
247
248 obd_getref(obd);
249 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
250 tgt = lov->lov_tgts[i];
251 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
252 continue;
253 /* Flags will be lowest common denominator */
254 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
255 if (rc) {
256 CERROR("%s: lov connect tgt %d failed: %d\n",
257 obd->obd_name, i, rc);
258 continue;
259 }
260 /* connect to administrative disabled ost */
261 if (!lov->lov_tgts[i]->ltd_exp)
262 continue;
263
264 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
265 OBD_NOTIFY_CONNECT, (void *)&i);
266 if (rc) {
267 CERROR("%s error sending notify %d\n",
268 obd->obd_name, rc);
269 }
270 }
271 obd_putref(obd);
272
273 RETURN(0);
274}
275
276static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
277{
278 proc_dir_entry_t *lov_proc_dir;
279 struct lov_obd *lov = &obd->u.lov;
280 struct obd_device *osc_obd;
281 int rc;
282 ENTRY;
283
284 osc_obd = class_exp2obd(tgt->ltd_exp);
285 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
286 obd->obd_name, osc_obd->obd_name);
287
288 if (tgt->ltd_active) {
289 tgt->ltd_active = 0;
290 lov->desc.ld_active_tgt_count--;
291 tgt->ltd_exp->exp_obd->obd_inactive = 1;
292 }
293
73bb1da6
PT		 294 lov_proc_dir = obd->obd_proc_private;
295 if (lov_proc_dir)
296 lprocfs_remove_proc_entry(osc_obd->obd_name, lov_proc_dir);
d7e09d03
PT		 297
298 if (osc_obd) {
299 /* Pass it on to our clients.
300 * XXX This should be an argument to disconnect,
301 * XXX not a back-door flag on the OBD. Ah well.
302 */
303 osc_obd->obd_force = obd->obd_force;
304 osc_obd->obd_fail = obd->obd_fail;
305 osc_obd->obd_no_recov = obd->obd_no_recov;
306 }
307
308 obd_register_observer(osc_obd, NULL);
309
310 rc = obd_disconnect(tgt->ltd_exp);
311 if (rc) {
312 CERROR("Target %s disconnect error %d\n",
313 tgt->ltd_uuid.uuid, rc);
314 rc = 0;
315 }
316
317 tgt->ltd_exp = NULL;
318 RETURN(0);
319}
320
321static int lov_disconnect(struct obd_export *exp)
322{
323 struct obd_device *obd = class_exp2obd(exp);
324 struct lov_obd *lov = &obd->u.lov;
325 int i, rc;
326 ENTRY;
327
328 if (!lov->lov_tgts)
329 goto out;
330
331 /* Only disconnect the underlying layers on the final disconnect. */
332 lov->lov_connects--;
333 if (lov->lov_connects != 0) {
334 /* why should there be more than 1 connect? */
335 CERROR("disconnect #%d\n", lov->lov_connects);
336 goto out;
337 }
338
339 /* Let's hold another reference so lov_del_obd doesn't spin through
340 putref every time */
341 obd_getref(obd);
342
343 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
344 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
345 /* Disconnection is the last we know about an obd */
346 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
347 }
348 }
349 obd_putref(obd);
350
351out:
352 rc = class_disconnect(exp); /* bz 9811 */
353 RETURN(rc);
354}
355
356/* Error codes:
357 *
358 * -EINVAL : UUID can't be found in the LOV's target list
359 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
360 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
361 * any >= 0 : is log target index
362 */
363static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
364 enum obd_notify_event ev)
365{
366 struct lov_obd *lov = &obd->u.lov;
367 struct lov_tgt_desc *tgt;
368 int index, activate, active;
369 ENTRY;
370
371 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
372 lov, uuid->uuid, ev);
373
374 obd_getref(obd);
375 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
376 tgt = lov->lov_tgts[index];
377 if (!tgt)
378 continue;
379 /*
380 * LU-642, initially inactive OSC could miss the obd_connect,
381 * we make up for it here.
382 */
383 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
384 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
385 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
386
387 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
388 &lov_osc_uuid, &lov->lov_ocd, NULL);
389 }
390 if (!tgt->ltd_exp)
391 continue;
392
393 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
394 index, obd_uuid2str(&tgt->ltd_uuid),
395 tgt->ltd_exp->exp_handle.h_cookie);
396 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
397 break;
398 }
399
400 if (index == lov->desc.ld_tgt_count)
401 GOTO(out, index = -EINVAL);
402
403 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
404 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
405
406 if (lov->lov_tgts[index]->ltd_activate == activate) {
407 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
408 uuid->uuid, activate ? "" : "de");
409 } else {
410 lov->lov_tgts[index]->ltd_activate = activate;
411 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
412 activate ? "" : "de", obd_uuid2str(uuid));
413 }
414
415 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
416 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
417
418 if (lov->lov_tgts[index]->ltd_active == active) {
419 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
420 uuid->uuid, active ? "" : "in");
421 GOTO(out, index);
422 } else {
423 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
424 obd_uuid2str(uuid), active ? "" : "in");
425 }
426
427 lov->lov_tgts[index]->ltd_active = active;
428 if (active) {
429 lov->desc.ld_active_tgt_count++;
430 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
431 } else {
432 lov->desc.ld_active_tgt_count--;
433 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
434 }
435 } else {
436 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
437 }
438
439 out:
440 obd_putref(obd);
441 RETURN(index);
442}
443
444static int lov_notify(struct obd_device *obd, struct obd_device *watched,
445 enum obd_notify_event ev, void *data)
446{
447 int rc = 0;
448 struct lov_obd *lov = &obd->u.lov;
449 ENTRY;
450
451 down_read(&lov->lov_notify_lock);
452 if (!lov->lov_connects) {
453 up_read(&lov->lov_notify_lock);
454 RETURN(rc);
455 }
456
457 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
458 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
459 struct obd_uuid *uuid;
460
461 LASSERT(watched);
462
463 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
464 up_read(&lov->lov_notify_lock);
465 CERROR("unexpected notification of %s %s!\n",
466 watched->obd_type->typ_name,
467 watched->obd_name);
468 RETURN(-EINVAL);
469 }
470 uuid = &watched->u.cli.cl_target_uuid;
471
472 /* Set OSC as active before notifying the observer, so the
473 * observer can use the OSC normally.
474 */
475 rc = lov_set_osc_active(obd, uuid, ev);
476 if (rc < 0) {
477 up_read(&lov->lov_notify_lock);
478 CERROR("event(%d) of %s failed: %d\n", ev,
479 obd_uuid2str(uuid), rc);
480 RETURN(rc);
481 }
482 /* active event should be pass lov target index as data */
483 data = &rc;
484 }
485
486 /* Pass the notification up the chain. */
487 if (watched) {
488 rc = obd_notify_observer(obd, watched, ev, data);
489 } else {
490 /* NULL watched means all osc's in the lov (only for syncs) */
491 /* sync event should be send lov idx as data */
492 struct lov_obd *lov = &obd->u.lov;
493 int i, is_sync;
494
495 data = &i;
496 is_sync = (ev == OBD_NOTIFY_SYNC) ||
497 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
498
499 obd_getref(obd);
500 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
501 if (!lov->lov_tgts[i])
502 continue;
503
504 /* don't send sync event if target not
505 * connected/activated */
506 if (is_sync && !lov->lov_tgts[i]->ltd_active)
507 continue;
508
509 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
510 ev, data);
511 if (rc) {
512 CERROR("%s: notify %s of %s failed %d\n",
513 obd->obd_name,
514 obd->obd_observer->obd_name,
515 lov->lov_tgts[i]->ltd_obd->obd_name,
516 rc);
517 }
518 }
519 obd_putref(obd);
520 }
521
522 up_read(&lov->lov_notify_lock);
523 RETURN(rc);
524}
525
526static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
527 __u32 index, int gen, int active)
528{
529 struct lov_obd *lov = &obd->u.lov;
530 struct lov_tgt_desc *tgt;
531 struct obd_device *tgt_obd;
532 int rc;
533 ENTRY;
534
535 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
536 uuidp->uuid, index, gen, active);
537
538 if (gen <= 0) {
539 CERROR("request to add OBD %s with invalid generation: %d\n",
540 uuidp->uuid, gen);
541 RETURN(-EINVAL);
542 }
543
544 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
545 &obd->obd_uuid);
546 if (tgt_obd == NULL)
547 RETURN(-EINVAL);
548
549 mutex_lock(&lov->lov_lock);
550
551 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
552 tgt = lov->lov_tgts[index];
553 CERROR("UUID %s already assigned at LOV target index %d\n",
554 obd_uuid2str(&tgt->ltd_uuid), index);
555 mutex_unlock(&lov->lov_lock);
556 RETURN(-EEXIST);
557 }
558
559 if (index >= lov->lov_tgt_size) {
560 /* We need to reallocate the lov target array. */
561 struct lov_tgt_desc **newtgts, **old = NULL;
562 __u32 newsize, oldsize = 0;
563
564 newsize = max(lov->lov_tgt_size, (__u32)2);
565 while (newsize < index + 1)
566 newsize = newsize << 1;
567 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
568 if (newtgts == NULL) {
569 mutex_unlock(&lov->lov_lock);
570 RETURN(-ENOMEM);
571 }
572
573 if (lov->lov_tgt_size) {
574 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
575 lov->lov_tgt_size);
576 old = lov->lov_tgts;
577 oldsize = lov->lov_tgt_size;
578 }
579
580 lov->lov_tgts = newtgts;
581 lov->lov_tgt_size = newsize;
582 smp_rmb();
583 if (old)
584 OBD_FREE(old, sizeof(*old) * oldsize);
585
586 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
587 lov->lov_tgts, lov->lov_tgt_size);
588 }
589
590 OBD_ALLOC_PTR(tgt);
591 if (!tgt) {
592 mutex_unlock(&lov->lov_lock);
593 RETURN(-ENOMEM);
594 }
595
596 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
597 if (rc) {
598 mutex_unlock(&lov->lov_lock);
599 OBD_FREE_PTR(tgt);
600 RETURN(rc);
601 }
602
603 tgt->ltd_uuid = *uuidp;
604 tgt->ltd_obd = tgt_obd;
605 /* XXX - add a sanity check on the generation number. */
606 tgt->ltd_gen = gen;
607 tgt->ltd_index = index;
608 tgt->ltd_activate = active;
609 lov->lov_tgts[index] = tgt;
610 if (index >= lov->desc.ld_tgt_count)
611 lov->desc.ld_tgt_count = index + 1;
612
613 mutex_unlock(&lov->lov_lock);
614
615 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
616 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
617
618 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
619
620 if (lov->lov_connects == 0) {
621 /* lov_connect hasn't been called yet. We'll do the
622 lov_connect_obd on this target when that fn first runs,
623 because we don't know the connect flags yet. */
624 RETURN(0);
625 }
626
627 obd_getref(obd);
628
629 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
630 if (rc)
631 GOTO(out, rc);
632
633 /* connect to administrative disabled ost */
634 if (!tgt->ltd_exp)
635 GOTO(out, rc = 0);
636
637 if (lov->lov_cache != NULL) {
638 rc = obd_set_info_async(NULL, tgt->ltd_exp,
639 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
640 sizeof(struct cl_client_cache), lov->lov_cache,
641 NULL);
642 if (rc < 0)
643 GOTO(out, rc);
644 }
645
646 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
647 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
648 (void *)&index);
649
650out:
651 if (rc) {
652 CERROR("add failed (%d), deleting %s\n", rc,
653 obd_uuid2str(&tgt->ltd_uuid));
654 lov_del_target(obd, index, 0, 0);
655 }
656 obd_putref(obd);
657 RETURN(rc);
658}
659
660/* Schedule a target for deletion */
661int lov_del_target(struct obd_device *obd, __u32 index,
662 struct obd_uuid *uuidp, int gen)
663{
664 struct lov_obd *lov = &obd->u.lov;
665 int count = lov->desc.ld_tgt_count;
666 int rc = 0;
667 ENTRY;
668
669 if (index >= count) {
670 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
671 index, count);
672 RETURN(-EINVAL);
673 }
674
675 /* to make sure there's no ongoing lov_notify() now */
676 down_write(&lov->lov_notify_lock);
677 obd_getref(obd);
678
679 if (!lov->lov_tgts[index]) {
680 CERROR("LOV target at index %d is not setup.\n", index);
681 GOTO(out, rc = -EINVAL);
682 }
683
684 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
685 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
686 lov_uuid2str(lov, index), index,
687 obd_uuid2str(uuidp));
688 GOTO(out, rc = -EINVAL);
689 }
690
691 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
692 lov_uuid2str(lov, index), index,
693 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
694 lov->lov_tgts[index]->ltd_active);
695
696 lov->lov_tgts[index]->ltd_reap = 1;
697 lov->lov_death_row++;
698 /* we really delete it from obd_putref */
699out:
700 obd_putref(obd);
701 up_write(&lov->lov_notify_lock);
702
703 RETURN(rc);
704}
705
706static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
707{
708 struct obd_device *osc_obd;
709
710 LASSERT(tgt);
711 LASSERT(tgt->ltd_reap);
712
713 osc_obd = class_exp2obd(tgt->ltd_exp);
714
715 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
716 tgt->ltd_uuid.uuid,
717 osc_obd ? osc_obd->obd_name : "<no obd>");
718
719 if (tgt->ltd_exp)
720 lov_disconnect_obd(obd, tgt);
721
722 OBD_FREE_PTR(tgt);
723
724 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
725 do it ourselves. And we can't do it from lov_cleanup,
726 because we just lost our only reference to it. */
727 if (osc_obd)
728 class_manual_cleanup(osc_obd);
729}
730
731void lov_fix_desc_stripe_size(__u64 *val)
732{
50dc198a
AD		 733 if (*val < LOV_MIN_STRIPE_SIZE) {
734 if (*val != 0)
735 LCONSOLE_INFO("Increasing default stripe size to "
736 "minimum %u\n",
737 LOV_DEFAULT_STRIPE_SIZE);
d7e09d03
PT		 738 *val = LOV_DEFAULT_STRIPE_SIZE;
739 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
740 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
741 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
742 "multiple of %u)\n",
743 *val, LOV_MIN_STRIPE_SIZE);
744 }
745}
746
747void lov_fix_desc_stripe_count(__u32 *val)
748{
749 if (*val == 0)
750 *val = 1;
751}
752
753void lov_fix_desc_pattern(__u32 *val)
754{
755 /* from lov_setstripe */
756 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
757 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
758 *val = 0;
759 }
760}
761
762void lov_fix_desc_qos_maxage(__u32 *val)
763{
764 /* fix qos_maxage */
765 if (*val == 0)
766 *val = QOS_DEFAULT_MAXAGE;
767}
768
769void lov_fix_desc(struct lov_desc *desc)
770{
771 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
772 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
773 lov_fix_desc_pattern(&desc->ld_pattern);
774 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
775}
776
777int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
778{
779 struct lprocfs_static_vars lvars = { 0 };
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 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 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 lprocfs_lov_init_vars(&lvars);
836 lprocfs_obd_setup(obd, lvars.obd_vars);
837#ifdef LPROCFS
838 {
5907838a 839 int rc1;
d7e09d03 840
5907838a 841 rc1 = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
d7e09d03 842 0444, &lov_proc_target_fops, obd);
5907838a 843 if (rc1)
d7e09d03
PT		 844 CWARN("Error adding the target_obd file\n");
845 }
846#endif
847 lov->lov_pool_proc_entry = lprocfs_register("pools",
848 obd->obd_proc_entry,
849 NULL, NULL);
850
851 RETURN(0);
852
853out:
854 return rc;
855}
856
857static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
858{
859 int rc = 0;
860 struct lov_obd *lov = &obd->u.lov;
861
862 ENTRY;
863
864 switch (stage) {
865 case OBD_CLEANUP_EARLY: {
866 int i;
867 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
868 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
869 continue;
870 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
871 OBD_CLEANUP_EARLY);
872 }
873 break;
874 }
875 case OBD_CLEANUP_EXPORTS:
876 rc = obd_llog_finish(obd, 0);
877 if (rc != 0)
878 CERROR("failed to cleanup llogging subsystems\n");
879 break;
880 }
881 RETURN(rc);
882}
883
884static int lov_cleanup(struct obd_device *obd)
885{
886 struct lov_obd *lov = &obd->u.lov;
887 struct list_head *pos, *tmp;
888 struct pool_desc *pool;
889 ENTRY;
890
891 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
892 pool = list_entry(pos, struct pool_desc, pool_list);
893 /* free pool structs */
894 CDEBUG(D_INFO, "delete pool %p\n", pool);
895 /* In the function below, .hs_keycmp resolves to
896 * pool_hashkey_keycmp() */
897 /* coverity[overrun-buffer-val] */
898 lov_pool_del(obd, pool->pool_name);
899 }
900 cfs_hash_putref(lov->lov_pools_hash_body);
901 lov_ost_pool_free(&lov->lov_packed);
902
903 lprocfs_obd_cleanup(obd);
904 if (lov->lov_tgts) {
905 int i;
906 obd_getref(obd);
907 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
908 if (!lov->lov_tgts[i])
909 continue;
910
911 /* Inactive targets may never have connected */
912 if (lov->lov_tgts[i]->ltd_active ||
913 atomic_read(&lov->lov_refcount))
914 /* We should never get here - these
915 should have been removed in the
916 disconnect. */
917 CERROR("lov tgt %d not cleaned!"
918 " deathrow=%d, lovrc=%d\n",
919 i, lov->lov_death_row,
920 atomic_read(&lov->lov_refcount));
921 lov_del_target(obd, i, 0, 0);
922 }
923 obd_putref(obd);
924 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
925 lov->lov_tgt_size);
926 lov->lov_tgt_size = 0;
927 }
928 RETURN(0);
929}
930
931int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
932 __u32 *indexp, int *genp)
933{
934 struct obd_uuid obd_uuid;
935 int cmd;
936 int rc = 0;
937 ENTRY;
938
939 switch(cmd = lcfg->lcfg_command) {
940 case LCFG_LOV_ADD_OBD:
941 case LCFG_LOV_ADD_INA:
942 case LCFG_LOV_DEL_OBD: {
943 __u32 index;
944 int gen;
945 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
946 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
947 GOTO(out, rc = -EINVAL);
948
949 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
950
951 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
952 GOTO(out, rc = -EINVAL);
953 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
954 GOTO(out, rc = -EINVAL);
955 index = *indexp;
956 gen = *genp;
957 if (cmd == LCFG_LOV_ADD_OBD)
958 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
959 else if (cmd == LCFG_LOV_ADD_INA)
960 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
961 else
962 rc = lov_del_target(obd, index, &obd_uuid, gen);
963 GOTO(out, rc);
964 }
965 case LCFG_PARAM: {
966 struct lprocfs_static_vars lvars = { 0 };
967 struct lov_desc *desc = &(obd->u.lov.desc);
968
969 if (!desc)
970 GOTO(out, rc = -EINVAL);
971
972 lprocfs_lov_init_vars(&lvars);
973
974 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
975 lcfg, obd);
976 if (rc > 0)
977 rc = 0;
978 GOTO(out, rc);
979 }
980 case LCFG_POOL_NEW:
981 case LCFG_POOL_ADD:
982 case LCFG_POOL_DEL:
983 case LCFG_POOL_REM:
984 GOTO(out, rc);
985
986 default: {
987 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
988 GOTO(out, rc = -EINVAL);
989
990 }
991 }
992out:
993 RETURN(rc);
994}
995
996static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
997 struct lov_stripe_md **ea, struct obd_trans_info *oti)
998{
999 struct lov_stripe_md *obj_mdp, *lsm;
1000 struct lov_obd *lov = &exp->exp_obd->u.lov;
1001 unsigned ost_idx;
1002 int rc, i;
1003 ENTRY;
1004
1005 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1006 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1007
1008 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1009 if (obj_mdp == NULL)
1010 RETURN(-ENOMEM);
1011
1012 ost_idx = src_oa->o_nlink;
1013 lsm = *ea;
1014 if (lsm == NULL)
1015 GOTO(out, rc = -EINVAL);
1016 if (ost_idx >= lov->desc.ld_tgt_count ||
1017 !lov->lov_tgts[ost_idx])
1018 GOTO(out, rc = -EINVAL);
1019
1020 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1021 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1022 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1023 ostid_id(&src_oa->o_oi))
1024 GOTO(out, rc = -EINVAL);
1025 break;
1026 }
1027 }
1028 if (i == lsm->lsm_stripe_count)
1029 GOTO(out, rc = -EINVAL);
1030
1031 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1032 src_oa, &obj_mdp, oti);
1033out:
1034 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1035 RETURN(rc);
1036}
1037
1038/* the LOV expects oa->o_id to be set to the LOV object id */
1039static int lov_create(const struct lu_env *env, struct obd_export *exp,
1040 struct obdo *src_oa, struct lov_stripe_md **ea,
1041 struct obd_trans_info *oti)
1042{
1043 struct lov_obd *lov;
1044 int rc = 0;
1045 ENTRY;
1046
1047 LASSERT(ea != NULL);
1048 if (exp == NULL)
1049 RETURN(-EINVAL);
1050
1051 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1052 src_oa->o_flags == OBD_FL_DELORPHAN) {
1053 /* should be used with LOV anymore */
1054 LBUG();
1055 }
1056
1057 lov = &exp->exp_obd->u.lov;
1058 if (!lov->desc.ld_active_tgt_count)
1059 RETURN(-EIO);
1060
1061 obd_getref(exp->exp_obd);
1062 /* Recreate a specific object id at the given OST index */
1063 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1064 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1065 rc = lov_recreate(exp, src_oa, ea, oti);
1066 }
1067
1068 obd_putref(exp->exp_obd);
1069 RETURN(rc);
1070}
1071
1072#define ASSERT_LSM_MAGIC(lsmp) \
1073do { \
1074 LASSERT((lsmp) != NULL); \
1075 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1076 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1077 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1078} while (0)
1079
1080static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1081 struct obdo *oa, struct lov_stripe_md *lsm,
1082 struct obd_trans_info *oti, struct obd_export *md_exp,
1083 void *capa)
1084{
1085 struct lov_request_set *set;
1086 struct obd_info oinfo;
1087 struct lov_request *req;
1088 struct list_head *pos;
1089 struct lov_obd *lov;
1090 int rc = 0, err = 0;
1091 ENTRY;
1092
1093 ASSERT_LSM_MAGIC(lsm);
1094
1095 if (!exp || !exp->exp_obd)
1096 RETURN(-ENODEV);
1097
1098 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1099 LASSERT(oti);
1100 LASSERT(oti->oti_logcookies);
1101 }
1102
1103 lov = &exp->exp_obd->u.lov;
1104 obd_getref(exp->exp_obd);
1105 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1106 if (rc)
1107 GOTO(out, rc);
1108
1109 list_for_each (pos, &set->set_list) {
1110 req = list_entry(pos, struct lov_request, rq_link);
1111
1112 if (oa->o_valid & OBD_MD_FLCOOKIE)
1113 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1114
1115 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1116 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1117 err = lov_update_common_set(set, req, err);
1118 if (err) {
1119 CERROR("%s: destroying objid "DOSTID" subobj "
1120 DOSTID" on OST idx %d: rc = %d\n",
1121 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1122 POSTID(&req->rq_oi.oi_oa->o_oi),
1123 req->rq_idx, err);
1124 if (!rc)
1125 rc = err;
1126 }
1127 }
1128
1129 if (rc == 0) {
1130 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1131 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1132 }
1133 err = lov_fini_destroy_set(set);
1134out:
1135 obd_putref(exp->exp_obd);
1136 RETURN(rc ? rc : err);
1137}
1138
1139static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1140 struct obd_info *oinfo)
1141{
1142 struct lov_request_set *set;
1143 struct lov_request *req;
1144 struct list_head *pos;
1145 struct lov_obd *lov;
1146 int err = 0, rc = 0;
1147 ENTRY;
1148
1149 LASSERT(oinfo);
1150 ASSERT_LSM_MAGIC(oinfo->oi_md);
1151
1152 if (!exp || !exp->exp_obd)
1153 RETURN(-ENODEV);
1154
1155 lov = &exp->exp_obd->u.lov;
1156
1157 rc = lov_prep_getattr_set(exp, oinfo, &set);
1158 if (rc)
1159 RETURN(rc);
1160
1161 list_for_each (pos, &set->set_list) {
1162 req = list_entry(pos, struct lov_request, rq_link);
1163
1164 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1165 " %u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1166 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1167
1168 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1169 &req->rq_oi);
1170 err = lov_update_common_set(set, req, rc);
1171 if (err) {
1172 CERROR("%s: getattr objid "DOSTID" subobj "
1173 DOSTID" on OST idx %d: rc = %d\n",
1174 exp->exp_obd->obd_name,
1175 POSTID(&oinfo->oi_oa->o_oi),
1176 POSTID(&req->rq_oi.oi_oa->o_oi),
1177 req->rq_idx, err);
1178 break;
1179 }
1180 }
1181
1182 rc = lov_fini_getattr_set(set);
1183 if (err)
1184 rc = err;
1185 RETURN(rc);
1186}
1187
1188static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1189 void *data, int rc)
1190{
1191 struct lov_request_set *lovset = (struct lov_request_set *)data;
1192 int err;
1193 ENTRY;
1194
1195 /* don't do attribute merge if this aysnc op failed */
1196 if (rc)
1197 atomic_set(&lovset->set_completes, 0);
1198 err = lov_fini_getattr_set(lovset);
1199 RETURN(rc ? rc : err);
1200}
1201
1202static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1203 struct ptlrpc_request_set *rqset)
1204{
1205 struct lov_request_set *lovset;
1206 struct lov_obd *lov;
1207 struct list_head *pos;
1208 struct lov_request *req;
1209 int rc = 0, err;
1210 ENTRY;
1211
1212 LASSERT(oinfo);
1213 ASSERT_LSM_MAGIC(oinfo->oi_md);
1214
1215 if (!exp || !exp->exp_obd)
1216 RETURN(-ENODEV);
1217
1218 lov = &exp->exp_obd->u.lov;
1219
1220 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1221 if (rc)
1222 RETURN(rc);
1223
1224 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1225 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1226 oinfo->oi_md->lsm_stripe_size);
1227
1228 list_for_each(pos, &lovset->set_list) {
1229 req = list_entry(pos, struct lov_request, rq_link);
1230
1231 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1232 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1233 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1234 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1235 &req->rq_oi, rqset);
1236 if (rc) {
1237 CERROR("%s: getattr objid "DOSTID" subobj"
1238 DOSTID" on OST idx %d: rc = %d\n",
1239 exp->exp_obd->obd_name,
1240 POSTID(&oinfo->oi_oa->o_oi),
1241 POSTID(&req->rq_oi.oi_oa->o_oi),
1242 req->rq_idx, rc);
1243 GOTO(out, rc);
1244 }
1245 }
1246
1247 if (!list_empty(&rqset->set_requests)) {
1248 LASSERT(rc == 0);
1249 LASSERT (rqset->set_interpret == NULL);
1250 rqset->set_interpret = lov_getattr_interpret;
1251 rqset->set_arg = (void *)lovset;
1252 RETURN(rc);
1253 }
1254out:
1255 if (rc)
1256 atomic_set(&lovset->set_completes, 0);
1257 err = lov_fini_getattr_set(lovset);
1258 RETURN(rc ? rc : err);
1259}
1260
1261static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1262 struct obd_info *oinfo, struct obd_trans_info *oti)
1263{
1264 struct lov_request_set *set;
1265 struct lov_obd *lov;
1266 struct list_head *pos;
1267 struct lov_request *req;
1268 int err = 0, rc = 0;
1269 ENTRY;
1270
1271 LASSERT(oinfo);
1272 ASSERT_LSM_MAGIC(oinfo->oi_md);
1273
1274 if (!exp || !exp->exp_obd)
1275 RETURN(-ENODEV);
1276
1277 /* for now, we only expect the following updates here */
1278 LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1279 OBD_MD_FLMODE | OBD_MD_FLATIME |
1280 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1281 OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1282 OBD_MD_FLGROUP | OBD_MD_FLUID |
1283 OBD_MD_FLGID | OBD_MD_FLFID |
1284 OBD_MD_FLGENER)));
1285 lov = &exp->exp_obd->u.lov;
1286 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1287 if (rc)
1288 RETURN(rc);
1289
1290 list_for_each (pos, &set->set_list) {
1291 req = list_entry(pos, struct lov_request, rq_link);
1292
1293 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1294 &req->rq_oi, NULL);
1295 err = lov_update_setattr_set(set, req, rc);
1296 if (err) {
1297 CERROR("%s: setattr objid "DOSTID" subobj "
1298 DOSTID" on OST idx %d: rc = %d\n",
1299 exp->exp_obd->obd_name,
1300 POSTID(&set->set_oi->oi_oa->o_oi),
1301 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1302 err);
1303 if (!rc)
1304 rc = err;
1305 }
1306 }
1307 err = lov_fini_setattr_set(set);
1308 if (!rc)
1309 rc = err;
1310 RETURN(rc);
1311}
1312
1313static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1314 void *data, int rc)
1315{
1316 struct lov_request_set *lovset = (struct lov_request_set *)data;
1317 int err;
1318 ENTRY;
1319
1320 if (rc)
1321 atomic_set(&lovset->set_completes, 0);
1322 err = lov_fini_setattr_set(lovset);
1323 RETURN(rc ? rc : err);
1324}
1325
1326/* If @oti is given, the request goes from MDS and responses from OSTs are not
1327 needed. Otherwise, a client is waiting for responses. */
1328static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1329 struct obd_trans_info *oti,
1330 struct ptlrpc_request_set *rqset)
1331{
1332 struct lov_request_set *set;
1333 struct lov_request *req;
1334 struct list_head *pos;
1335 struct lov_obd *lov;
1336 int rc = 0;
1337 ENTRY;
1338
1339 LASSERT(oinfo);
1340 ASSERT_LSM_MAGIC(oinfo->oi_md);
1341 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1342 LASSERT(oti);
1343 LASSERT(oti->oti_logcookies);
1344 }
1345
1346 if (!exp || !exp->exp_obd)
1347 RETURN(-ENODEV);
1348
1349 lov = &exp->exp_obd->u.lov;
1350 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1351 if (rc)
1352 RETURN(rc);
1353
1354 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1355 POSTID(&oinfo->oi_md->lsm_oi),
1356 oinfo->oi_md->lsm_stripe_count,
1357 oinfo->oi_md->lsm_stripe_size);
1358
1359 list_for_each(pos, &set->set_list) {
1360 req = list_entry(pos, struct lov_request, rq_link);
1361
1362 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1363 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1364
1365 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1366 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1367 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1368
1369 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1370 &req->rq_oi, oti, rqset);
1371 if (rc) {
1372 CERROR("error: setattr objid "DOSTID" subobj"
1373 DOSTID" on OST idx %d: rc = %d\n",
1374 POSTID(&set->set_oi->oi_oa->o_oi),
1375 POSTID(&req->rq_oi.oi_oa->o_oi),
1376 req->rq_idx, rc);
1377 break;
1378 }
1379 }
1380
1381 /* If we are not waiting for responses on async requests, return. */
1382 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1383 int err;
1384 if (rc)
1385 atomic_set(&set->set_completes, 0);
1386 err = lov_fini_setattr_set(set);
1387 RETURN(rc ? rc : err);
1388 }
1389
1390 LASSERT(rqset->set_interpret == NULL);
1391 rqset->set_interpret = lov_setattr_interpret;
1392 rqset->set_arg = (void *)set;
1393
1394 RETURN(0);
1395}
1396
1397static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1398 void *data, int rc)
1399{
1400 struct lov_request_set *lovset = (struct lov_request_set *)data;
1401 int err;
1402 ENTRY;
1403
1404 if (rc)
1405 atomic_set(&lovset->set_completes, 0);
1406 err = lov_fini_punch_set(lovset);
1407 RETURN(rc ? rc : err);
1408}
1409
1410/* FIXME: maybe we'll just make one node the authoritative attribute node, then
1411 * we can send this 'punch' to just the authoritative node and the nodes
1412 * that the punch will affect. */
1413static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1414 struct obd_info *oinfo, struct obd_trans_info *oti,
1415 struct ptlrpc_request_set *rqset)
1416{
1417 struct lov_request_set *set;
1418 struct lov_obd *lov;
1419 struct list_head *pos;
1420 struct lov_request *req;
1421 int rc = 0;
1422 ENTRY;
1423
1424 LASSERT(oinfo);
1425 ASSERT_LSM_MAGIC(oinfo->oi_md);
1426
1427 if (!exp || !exp->exp_obd)
1428 RETURN(-ENODEV);
1429
1430 lov = &exp->exp_obd->u.lov;
1431 rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1432 if (rc)
1433 RETURN(rc);
1434
1435 list_for_each (pos, &set->set_list) {
1436 req = list_entry(pos, struct lov_request, rq_link);
1437
1438 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1439 &req->rq_oi, NULL, rqset);
1440 if (rc) {
1441 CERROR("%s: punch objid "DOSTID" subobj "DOSTID
1442 " on OST idx %d: rc = %d\n",
1443 exp->exp_obd->obd_name,
1444 POSTID(&set->set_oi->oi_oa->o_oi),
1445 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, rc);
1446 break;
1447 }
1448 }
1449
1450 if (rc || list_empty(&rqset->set_requests)) {
1451 int err;
1452 err = lov_fini_punch_set(set);
1453 RETURN(rc ? rc : err);
1454 }
1455
1456 LASSERT(rqset->set_interpret == NULL);
1457 rqset->set_interpret = lov_punch_interpret;
1458 rqset->set_arg = (void *)set;
1459
1460 RETURN(0);
1461}
1462
1463static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1464 void *data, int rc)
1465{
1466 struct lov_request_set *lovset = data;
1467 int err;
1468 ENTRY;
1469
1470 if (rc)
1471 atomic_set(&lovset->set_completes, 0);
1472 err = lov_fini_sync_set(lovset);
1473 RETURN(rc ?: err);
1474}
1475
1476static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1477 struct obd_info *oinfo, obd_off start, obd_off end,
1478 struct ptlrpc_request_set *rqset)
1479{
1480 struct lov_request_set *set = NULL;
1481 struct lov_obd *lov;
1482 struct list_head *pos;
1483 struct lov_request *req;
1484 int rc = 0;
1485 ENTRY;
1486
1487 ASSERT_LSM_MAGIC(oinfo->oi_md);
1488 LASSERT(rqset != NULL);
1489
1490 if (!exp->exp_obd)
1491 RETURN(-ENODEV);
1492
1493 lov = &exp->exp_obd->u.lov;
1494 rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1495 if (rc)
1496 RETURN(rc);
1497
1498 CDEBUG(D_INFO, "fsync objid "DOSTID" ["LPX64", "LPX64"]\n",
1499 POSTID(&set->set_oi->oi_oa->o_oi), start, end);
1500
1501 list_for_each (pos, &set->set_list) {
1502 req = list_entry(pos, struct lov_request, rq_link);
1503
1504 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1505 &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1506 req->rq_oi.oi_policy.l_extent.end, rqset);
1507 if (rc) {
1508 CERROR("%s: fsync objid "DOSTID" subobj "DOSTID
1509 " on OST idx %d: rc = %d\n",
1510 exp->exp_obd->obd_name,
1511 POSTID(&set->set_oi->oi_oa->o_oi),
1512 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx,
1513 rc);
1514 break;
1515 }
1516 }
1517
1518 /* If we are not waiting for responses on async requests, return. */
1519 if (rc || list_empty(&rqset->set_requests)) {
1520 int err = lov_fini_sync_set(set);
1521
1522 RETURN(rc ?: err);
1523 }
1524
1525 LASSERT(rqset->set_interpret == NULL);
1526 rqset->set_interpret = lov_sync_interpret;
1527 rqset->set_arg = (void *)set;
1528
1529 RETURN(0);
1530}
1531
1532static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1533 obd_count oa_bufs, struct brw_page *pga)
1534{
1535 struct obd_info oinfo = { { { 0 } } };
1536 int i, rc = 0;
1537
1538 oinfo.oi_oa = lov_oinfo->oi_oa;
1539
1540 /* The caller just wants to know if there's a chance that this
1541 * I/O can succeed */
1542 for (i = 0; i < oa_bufs; i++) {
1543 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1544 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1545 obd_off start, end;
1546
1547 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1548 pga[i].off + pga[i].count - 1,
1549 &start, &end))
1550 continue;
1551
1552 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1553 CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1554 return -EIO;
1555 }
1556
1557 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1558 1, &pga[i], NULL);
1559 if (rc)
1560 break;
1561 }
1562 return rc;
1563}
1564
1565static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1566 obd_count oa_bufs, struct brw_page *pga,
1567 struct obd_trans_info *oti)
1568{
1569 struct lov_request_set *set;
1570 struct lov_request *req;
1571 struct list_head *pos;
1572 struct lov_obd *lov = &exp->exp_obd->u.lov;
1573 int err, rc = 0;
1574 ENTRY;
1575
1576 ASSERT_LSM_MAGIC(oinfo->oi_md);
1577
1578 if (cmd == OBD_BRW_CHECK) {
1579 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1580 RETURN(rc);
1581 }
1582
1583 rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1584 if (rc)
1585 RETURN(rc);
1586
1587 list_for_each (pos, &set->set_list) {
1588 struct obd_export *sub_exp;
1589 struct brw_page *sub_pga;
1590 req = list_entry(pos, struct lov_request, rq_link);
1591
1592 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1593 sub_pga = set->set_pga + req->rq_pgaidx;
1594 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1595 sub_pga, oti);
1596 if (rc)
1597 break;
1598 lov_update_common_set(set, req, rc);
1599 }
1600
1601 err = lov_fini_brw_set(set);
1602 if (!rc)
1603 rc = err;
1604 RETURN(rc);
1605}
1606
1607static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1608 void *data, int rc)
1609{
1610 struct lov_request_set *lovset = (struct lov_request_set *)data;
1611 ENTRY;
1612 rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1613 RETURN(rc);
1614}
1615
1616static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1617 struct ldlm_enqueue_info *einfo,
1618 struct ptlrpc_request_set *rqset)
1619{
1620 ldlm_mode_t mode = einfo->ei_mode;
1621 struct lov_request_set *set;
1622 struct lov_request *req;
1623 struct list_head *pos;
1624 struct lov_obd *lov;
1625 ldlm_error_t rc;
1626 ENTRY;
1627
1628 LASSERT(oinfo);
1629 ASSERT_LSM_MAGIC(oinfo->oi_md);
1630 LASSERT(mode == (mode & -mode));
1631
1632 /* we should never be asked to replay a lock this way. */
1633 LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1634
1635 if (!exp || !exp->exp_obd)
1636 RETURN(-ENODEV);
1637
1638 lov = &exp->exp_obd->u.lov;
1639 rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1640 if (rc)
1641 RETURN(rc);
1642
1643 list_for_each (pos, &set->set_list) {
1644 req = list_entry(pos, struct lov_request, rq_link);
1645
1646 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1647 &req->rq_oi, einfo, rqset);
1648 if (rc != ELDLM_OK)
1649 GOTO(out, rc);
1650 }
1651
1652 if (rqset && !list_empty(&rqset->set_requests)) {
1653 LASSERT(rc == 0);
1654 LASSERT(rqset->set_interpret == NULL);
1655 rqset->set_interpret = lov_enqueue_interpret;
1656 rqset->set_arg = (void *)set;
1657 RETURN(rc);
1658 }
1659out:
1660 rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1661 RETURN(rc);
1662}
1663
1664static int lov_change_cbdata(struct obd_export *exp,
1665 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1666 void *data)
1667{
1668 struct lov_obd *lov;
1669 int rc = 0, i;
1670 ENTRY;
1671
1672 ASSERT_LSM_MAGIC(lsm);
1673
1674 if (!exp || !exp->exp_obd)
1675 RETURN(-ENODEV);
1676
1677 lov = &exp->exp_obd->u.lov;
1678 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1679 struct lov_stripe_md submd;
1680 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1681
1682 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1683 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1684 continue;
1685 }
1686
1687 submd.lsm_oi = loi->loi_oi;
1688 submd.lsm_stripe_count = 0;
1689 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1690 &submd, it, data);
1691 }
1692 RETURN(rc);
1693}
1694
1695/* find any ldlm lock of the inode in lov
1696 * return 0 not find
1697 * 1 find one
1698 * < 0 error */
1699static int lov_find_cbdata(struct obd_export *exp,
1700 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1701 void *data)
1702{
1703 struct lov_obd *lov;
1704 int rc = 0, i;
1705 ENTRY;
1706
1707 ASSERT_LSM_MAGIC(lsm);
1708
1709 if (!exp || !exp->exp_obd)
1710 RETURN(-ENODEV);
1711
1712 lov = &exp->exp_obd->u.lov;
1713 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1714 struct lov_stripe_md submd;
1715 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1716
1717 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1718 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1719 continue;
1720 }
1721 submd.lsm_oi = loi->loi_oi;
1722 submd.lsm_stripe_count = 0;
1723 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1724 &submd, it, data);
1725 if (rc != 0)
1726 RETURN(rc);
1727 }
1728 RETURN(rc);
1729}
1730
1731static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1732 __u32 mode, struct lustre_handle *lockh)
1733{
1734 struct lov_request_set *set;
1735 struct obd_info oinfo;
1736 struct lov_request *req;
1737 struct list_head *pos;
1738 struct lov_obd *lov;
1739 struct lustre_handle *lov_lockhp;
1740 int err = 0, rc = 0;
1741 ENTRY;
1742
1743 ASSERT_LSM_MAGIC(lsm);
1744
1745 if (!exp || !exp->exp_obd)
1746 RETURN(-ENODEV);
1747
1748 LASSERT(lockh);
1749 lov = &exp->exp_obd->u.lov;
1750 rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1751 if (rc)
1752 RETURN(rc);
1753
1754 list_for_each(pos, &set->set_list) {
1755 req = list_entry(pos, struct lov_request, rq_link);
1756 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1757
1758 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1759 req->rq_oi.oi_md, mode, lov_lockhp);
1760 rc = lov_update_common_set(set, req, rc);
1761 if (rc) {
1762 CERROR("%s: cancel objid "DOSTID" subobj "
1763 DOSTID" on OST idx %d: rc = %d\n",
1764 exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1765 POSTID(&req->rq_oi.oi_md->lsm_oi),
1766 req->rq_idx, rc);
1767 err = rc;
1768 }
1769
1770 }
1771 lov_fini_cancel_set(set);
1772 RETURN(err);
1773}
1774
1775static int lov_cancel_unused(struct obd_export *exp,
1776 struct lov_stripe_md *lsm,
1777 ldlm_cancel_flags_t flags, void *opaque)
1778{
1779 struct lov_obd *lov;
1780 int rc = 0, i;
1781 ENTRY;
1782
1783 if (!exp || !exp->exp_obd)
1784 RETURN(-ENODEV);
1785
1786 lov = &exp->exp_obd->u.lov;
1787 if (lsm == NULL) {
1788 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1789 int err;
1790 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1791 continue;
1792
1793 err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1794 flags, opaque);
1795 if (!rc)
1796 rc = err;
1797 }
1798 RETURN(rc);
1799 }
1800
1801 ASSERT_LSM_MAGIC(lsm);
1802
1803 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1804 struct lov_stripe_md submd;
1805 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1806 int idx = loi->loi_ost_idx;
1807 int err;
1808
1809 if (!lov->lov_tgts[idx]) {
1810 CDEBUG(D_HA, "lov idx %d NULL\n", idx);
1811 continue;
1812 }
1813
1814 if (!lov->lov_tgts[idx]->ltd_active)
1815 CDEBUG(D_HA, "lov idx %d inactive\n", idx);
1816
1817 submd.lsm_oi = loi->loi_oi;
1818 submd.lsm_stripe_count = 0;
1819 err = obd_cancel_unused(lov->lov_tgts[idx]->ltd_exp,
1820 &submd, flags, opaque);
1821 if (err && lov->lov_tgts[idx]->ltd_active) {
1822 CERROR("%s: cancel unused objid "DOSTID
1823 " subobj "DOSTID" on OST idx %d: rc = %d\n",
1824 exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi),
1825 POSTID(&loi->loi_oi), idx, err);
1826 if (!rc)
1827 rc = err;
1828 }
1829 }
1830 RETURN(rc);
1831}
1832
1833int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1834{
1835 struct lov_request_set *lovset = (struct lov_request_set *)data;
1836 int err;
1837 ENTRY;
1838
1839 if (rc)
1840 atomic_set(&lovset->set_completes, 0);
1841
1842 err = lov_fini_statfs_set(lovset);
1843 RETURN(rc ? rc : err);
1844}
1845
1846static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1847 __u64 max_age, struct ptlrpc_request_set *rqset)
1848{
1849 struct obd_device *obd = class_exp2obd(exp);
1850 struct lov_request_set *set;
1851 struct lov_request *req;
1852 struct list_head *pos;
1853 struct lov_obd *lov;
1854 int rc = 0;
1855 ENTRY;
1856
1857 LASSERT(oinfo != NULL);
1858 LASSERT(oinfo->oi_osfs != NULL);
1859
1860 lov = &obd->u.lov;
1861 rc = lov_prep_statfs_set(obd, oinfo, &set);
1862 if (rc)
1863 RETURN(rc);
1864
1865 list_for_each (pos, &set->set_list) {
1866 req = list_entry(pos, struct lov_request, rq_link);
1867 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1868 &req->rq_oi, max_age, rqset);
1869 if (rc)
1870 break;
1871 }
1872
1873 if (rc || list_empty(&rqset->set_requests)) {
1874 int err;
1875 if (rc)
1876 atomic_set(&set->set_completes, 0);
1877 err = lov_fini_statfs_set(set);
1878 RETURN(rc ? rc : err);
1879 }
1880
1881 LASSERT(rqset->set_interpret == NULL);
1882 rqset->set_interpret = lov_statfs_interpret;
1883 rqset->set_arg = (void *)set;
1884 RETURN(0);
1885}
1886
1887static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1888 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1889{
1890 struct ptlrpc_request_set *set = NULL;
1891 struct obd_info oinfo = { { { 0 } } };
1892 int rc = 0;
1893 ENTRY;
1894
1895
1896 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1897 * statfs requests */
1898 set = ptlrpc_prep_set();
1899 if (set == NULL)
1900 RETURN(-ENOMEM);
1901
1902 oinfo.oi_osfs = osfs;
1903 oinfo.oi_flags = flags;
1904 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1905 if (rc == 0)
1906 rc = ptlrpc_set_wait(set);
1907 ptlrpc_set_destroy(set);
1908
1909 RETURN(rc);
1910}
1911
1912static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1913 void *karg, void *uarg)
1914{
1915 struct obd_device *obddev = class_exp2obd(exp);
1916 struct lov_obd *lov = &obddev->u.lov;
1917 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1918 struct obd_uuid *uuidp;
1919 ENTRY;
1920
1921 switch (cmd) {
1922 case IOC_OBD_STATFS: {
1923 struct obd_ioctl_data *data = karg;
1924 struct obd_device *osc_obd;
1925 struct obd_statfs stat_buf = {0};
1926 __u32 index;
1927 __u32 flags;
1928
1929 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1930 if ((index >= count))
1931 RETURN(-ENODEV);
1932
1933 if (!lov->lov_tgts[index])
1934 /* Try again with the next index */
1935 RETURN(-EAGAIN);
1936 if (!lov->lov_tgts[index]->ltd_active)
1937 RETURN(-ENODATA);
1938
1939 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1940 if (!osc_obd)
1941 RETURN(-EINVAL);
1942
1943 /* copy UUID */
1944 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1945 min((int) data->ioc_plen2,
1946 (int) sizeof(struct obd_uuid))))
1947 RETURN(-EFAULT);
1948
1949 flags = uarg ? *(__u32*)uarg : 0;
1950 /* got statfs data */
1951 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1952 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1953 flags);
1954 if (rc)
1955 RETURN(rc);
1956 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1957 min((int) data->ioc_plen1,
1958 (int) sizeof(stat_buf))))
1959 RETURN(-EFAULT);
1960 break;
1961 }
1962 case OBD_IOC_LOV_GET_CONFIG: {
1963 struct obd_ioctl_data *data;
1964 struct lov_desc *desc;
1965 char *buf = NULL;
1966 __u32 *genp;
1967
1968 len = 0;
1969 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1970 RETURN(-EINVAL);
1971
1972 data = (struct obd_ioctl_data *)buf;
1973
1974 if (sizeof(*desc) > data->ioc_inllen1) {
1975 obd_ioctl_freedata(buf, len);
1976 RETURN(-EINVAL);
1977 }
1978
1979 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1980 obd_ioctl_freedata(buf, len);
1981 RETURN(-EINVAL);
1982 }
1983
1984 if (sizeof(__u32) * count > data->ioc_inllen3) {
1985 obd_ioctl_freedata(buf, len);
1986 RETURN(-EINVAL);
1987 }
1988
1989 desc = (struct lov_desc *)data->ioc_inlbuf1;
1990 memcpy(desc, &(lov->desc), sizeof(*desc));
1991
1992 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1993 genp = (__u32 *)data->ioc_inlbuf3;
1994 /* the uuid will be empty for deleted OSTs */
1995 for (i = 0; i < count; i++, uuidp++, genp++) {
1996 if (!lov->lov_tgts[i])
1997 continue;
1998 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1999 *genp = lov->lov_tgts[i]->ltd_gen;
2000 }
2001
2002 if (copy_to_user((void *)uarg, buf, len))
2003 rc = -EFAULT;
2004 obd_ioctl_freedata(buf, len);
2005 break;
2006 }
2007 case LL_IOC_LOV_SETSTRIPE:
2008 rc = lov_setstripe(exp, len, karg, uarg);
2009 break;
2010 case LL_IOC_LOV_GETSTRIPE:
2011 rc = lov_getstripe(exp, karg, uarg);
2012 break;
2013 case LL_IOC_LOV_SETEA:
2014 rc = lov_setea(exp, karg, uarg);
2015 break;
2016 case OBD_IOC_QUOTACTL: {
2017 struct if_quotactl *qctl = karg;
2018 struct lov_tgt_desc *tgt = NULL;
2019 struct obd_quotactl *oqctl;
2020
2021 if (qctl->qc_valid == QC_OSTIDX) {
2022 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2023 RETURN(-EINVAL);
2024
2025 tgt = lov->lov_tgts[qctl->qc_idx];
2026 if (!tgt || !tgt->ltd_exp)
2027 RETURN(-EINVAL);
2028 } else if (qctl->qc_valid == QC_UUID) {
2029 for (i = 0; i < count; i++) {
2030 tgt = lov->lov_tgts[i];
2031 if (!tgt ||
2032 !obd_uuid_equals(&tgt->ltd_uuid,
2033 &qctl->obd_uuid))
2034 continue;
2035
2036 if (tgt->ltd_exp == NULL)
2037 RETURN(-EINVAL);
2038
2039 break;
2040 }
2041 } else {
2042 RETURN(-EINVAL);
2043 }
2044
2045 if (i >= count)
2046 RETURN(-EAGAIN);
2047
2048 LASSERT(tgt && tgt->ltd_exp);
2049 OBD_ALLOC_PTR(oqctl);
2050 if (!oqctl)
2051 RETURN(-ENOMEM);
2052
2053 QCTL_COPY(oqctl, qctl);
2054 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2055 if (rc == 0) {
2056 QCTL_COPY(qctl, oqctl);
2057 qctl->qc_valid = QC_OSTIDX;
2058 qctl->obd_uuid = tgt->ltd_uuid;
2059 }
2060 OBD_FREE_PTR(oqctl);
2061 break;
2062 }
2063 default: {
2064 int set = 0;
2065
2066 if (count == 0)
2067 RETURN(-ENOTTY);
2068
2069 for (i = 0; i < count; i++) {
2070 int err;
2071 struct obd_device *osc_obd;
2072
2073 /* OST was disconnected */
2074 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2075 continue;
2076
2077 /* ll_umount_begin() sets force flag but for lov, not
2078 * osc. Let's pass it through */
2079 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2080 osc_obd->obd_force = obddev->obd_force;
2081 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2082 len, karg, uarg);
2083 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2084 RETURN(err);
2085 } else if (err) {
2086 if (lov->lov_tgts[i]->ltd_active) {
2087 CDEBUG(err == -ENOTTY ?
2088 D_IOCTL : D_WARNING,
2089 "iocontrol OSC %s on OST "
2090 "idx %d cmd %x: err = %d\n",
2091 lov_uuid2str(lov, i),
2092 i, cmd, err);
2093 if (!rc)
2094 rc = err;
2095 }
2096 } else {
2097 set = 1;
2098 }
2099 }
2100 if (!set && !rc)
2101 rc = -EIO;
2102 }
2103 }
2104
2105 RETURN(rc);
2106}
2107
2108#define FIEMAP_BUFFER_SIZE 4096
2109
2110/**
2111 * Non-zero fe_logical indicates that this is a continuation FIEMAP
2112 * call. The local end offset and the device are sent in the first
2113 * fm_extent. This function calculates the stripe number from the index.
2114 * This function returns a stripe_no on which mapping is to be restarted.
2115 *
2116 * This function returns fm_end_offset which is the in-OST offset at which
2117 * mapping should be restarted. If fm_end_offset=0 is returned then caller
2118 * will re-calculate proper offset in next stripe.
2119 * Note that the first extent is passed to lov_get_info via the value field.
2120 *
2121 * \param fiemap fiemap request header
2122 * \param lsm striping information for the file
2123 * \param fm_start logical start of mapping
2124 * \param fm_end logical end of mapping
2125 * \param start_stripe starting stripe will be returned in this
2126 */
2127obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2128 struct lov_stripe_md *lsm, obd_size fm_start,
2129 obd_size fm_end, int *start_stripe)
2130{
2131 obd_size local_end = fiemap->fm_extents[0].fe_logical;
2132 obd_off lun_start, lun_end;
2133 obd_size fm_end_offset;
2134 int stripe_no = -1, i;
2135
2136 if (fiemap->fm_extent_count == 0 ||
2137 fiemap->fm_extents[0].fe_logical == 0)
2138 return 0;
2139
2140 /* Find out stripe_no from ost_index saved in the fe_device */
2141 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2142 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2143 fiemap->fm_extents[0].fe_device) {
2144 stripe_no = i;
2145 break;
2146 }
2147 }
2148 if (stripe_no == -1)
2149 return -EINVAL;
2150
2151 /* If we have finished mapping on previous device, shift logical
2152 * offset to start of next device */
2153 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2154 &lun_start, &lun_end)) != 0 &&
2155 local_end < lun_end) {
2156 fm_end_offset = local_end;
2157 *start_stripe = stripe_no;
2158 } else {
2159 /* This is a special value to indicate that caller should
2160 * calculate offset in next stripe. */
2161 fm_end_offset = 0;
2162 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2163 }
2164
2165 return fm_end_offset;
2166}
2167
2168/**
2169 * We calculate on which OST the mapping will end. If the length of mapping
2170 * is greater than (stripe_size * stripe_count) then the last_stripe will
2171 * will be one just before start_stripe. Else we check if the mapping
2172 * intersects each OST and find last_stripe.
2173 * This function returns the last_stripe and also sets the stripe_count
2174 * over which the mapping is spread
2175 *
2176 * \param lsm striping information for the file
2177 * \param fm_start logical start of mapping
2178 * \param fm_end logical end of mapping
2179 * \param start_stripe starting stripe of the mapping
2180 * \param stripe_count the number of stripes across which to map is returned
2181 *
2182 * \retval last_stripe return the last stripe of the mapping
2183 */
2184int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2185 obd_size fm_end, int start_stripe,
2186 int *stripe_count)
2187{
2188 int last_stripe;
2189 obd_off obd_start, obd_end;
2190 int i, j;
2191
2192 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2193 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2194 start_stripe - 1);
2195 *stripe_count = lsm->lsm_stripe_count;
2196 } else {
2197 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2198 i = (i + 1) % lsm->lsm_stripe_count, j++) {
2199 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2200 &obd_start, &obd_end)) == 0)
2201 break;
2202 }
2203 *stripe_count = j;
2204 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2205 }
2206
2207 return last_stripe;
2208}
2209
2210/**
2211 * Set fe_device and copy extents from local buffer into main return buffer.
2212 *
2213 * \param fiemap fiemap request header
2214 * \param lcl_fm_ext array of local fiemap extents to be copied
2215 * \param ost_index OST index to be written into the fm_device field for each
2216 extent
2217 * \param ext_count number of extents to be copied
2218 * \param current_extent where to start copying in main extent array
2219 */
2220void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2221 struct ll_fiemap_extent *lcl_fm_ext,
2222 int ost_index, unsigned int ext_count,
2223 int current_extent)
2224{
2225 char *to;
2226 int ext;
2227
2228 for (ext = 0; ext < ext_count; ext++) {
2229 lcl_fm_ext[ext].fe_device = ost_index;
2230 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2231 }
2232
2233 /* Copy fm_extent's from fm_local to return buffer */
2234 to = (char *)fiemap + fiemap_count_to_size(current_extent);
2235 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2236}
2237
2238/**
2239 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2240 * This also handles the restarting of FIEMAP calls in case mapping overflows
2241 * the available number of extents in single call.
2242 */
2243static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2244 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2245{
2246 struct ll_fiemap_info_key *fm_key = key;
2247 struct ll_user_fiemap *fiemap = val;
2248 struct ll_user_fiemap *fm_local = NULL;
2249 struct ll_fiemap_extent *lcl_fm_ext;
2250 int count_local;
2251 unsigned int get_num_extents = 0;
2252 int ost_index = 0, actual_start_stripe, start_stripe;
2253 obd_size fm_start, fm_end, fm_length, fm_end_offset;
2254 obd_size curr_loc;
2255 int current_extent = 0, rc = 0, i;
2256 int ost_eof = 0; /* EOF for object */
2257 int ost_done = 0; /* done with required mapping for this OST? */
2258 int last_stripe;
2259 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2260 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2261
5dd16419 2262 if (!lsm_has_objects(lsm))
d7e09d03
PT		 2263 GOTO(out, rc = 0);
2264
2265 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2266 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2267
2268 OBD_ALLOC_LARGE(fm_local, buffer_size);
2269 if (fm_local == NULL)
2270 GOTO(out, rc = -ENOMEM);
2271 lcl_fm_ext = &fm_local->fm_extents[0];
2272
2273 count_local = fiemap_size_to_count(buffer_size);
2274
2275 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2276 fm_start = fiemap->fm_start;
2277 fm_length = fiemap->fm_length;
2278 /* Calculate start stripe, last stripe and length of mapping */
2279 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2280 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2281 fm_start + fm_length - 1);
2282 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2283 if (fm_end > fm_key->oa.o_size)
2284 fm_end = fm_key->oa.o_size;
2285
2286 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2287 actual_start_stripe, &stripe_count);
2288
2289 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2290 fm_end, &start_stripe);
2291 if (fm_end_offset == -EINVAL)
2292 GOTO(out, rc = -EINVAL);
2293
2294 if (fiemap->fm_extent_count == 0) {
2295 get_num_extents = 1;
2296 count_local = 0;
2297 }
2298
2299 /* Check each stripe */
2300 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2301 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2302 obd_size req_fm_len; /* Stores length of required mapping */
2303 obd_size len_mapped_single_call;
2304 obd_off lun_start, lun_end, obd_object_end;
2305 unsigned int ext_count;
2306
2307 cur_stripe_wrap = cur_stripe;
2308
2309 /* Find out range of mapping on this stripe */
2310 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2311 &lun_start, &obd_object_end)) == 0)
2312 continue;
2313
2314 /* If this is a continuation FIEMAP call and we are on
2315 * starting stripe then lun_start needs to be set to
2316 * fm_end_offset */
2317 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2318 lun_start = fm_end_offset;
2319
2320 if (fm_length != ~0ULL) {
2321 /* Handle fm_start + fm_length overflow */
2322 if (fm_start + fm_length < fm_start)
2323 fm_length = ~0ULL - fm_start;
2324 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2325 cur_stripe);
2326 } else {
2327 lun_end = ~0ULL;
2328 }
2329
2330 if (lun_start == lun_end)
2331 continue;
2332
2333 req_fm_len = obd_object_end - lun_start;
2334 fm_local->fm_length = 0;
2335 len_mapped_single_call = 0;
2336
2337 /* If the output buffer is very large and the objects have many
2338 * extents we may need to loop on a single OST repeatedly */
2339 ost_eof = 0;
2340 ost_done = 0;
2341 do {
2342 if (get_num_extents == 0) {
2343 /* Don't get too many extents. */
2344 if (current_extent + count_local >
2345 fiemap->fm_extent_count)
2346 count_local = fiemap->fm_extent_count -
2347 current_extent;
2348 }
2349
2350 lun_start += len_mapped_single_call;
2351 fm_local->fm_length = req_fm_len - len_mapped_single_call;
2352 req_fm_len = fm_local->fm_length;
2353 fm_local->fm_extent_count = count_local;
2354 fm_local->fm_mapped_extents = 0;
2355 fm_local->fm_flags = fiemap->fm_flags;
2356
2357 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
2358 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2359
2360 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2361 GOTO(out, rc = -EINVAL);
2362
2363 /* If OST is inactive, return extent with UNKNOWN flag */
2364 if (!lov->lov_tgts[ost_index]->ltd_active) {
2365 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2366 fm_local->fm_mapped_extents = 1;
2367
2368 lcl_fm_ext[0].fe_logical = lun_start;
2369 lcl_fm_ext[0].fe_length = obd_object_end -
2370 lun_start;
2371 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2372
2373 goto inactive_tgt;
2374 }
2375
2376 fm_local->fm_start = lun_start;
2377 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2378 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2379 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2380 rc = obd_get_info(NULL,
2381 lov->lov_tgts[ost_index]->ltd_exp,
2382 keylen, key, vallen, fm_local, lsm);
2383 if (rc != 0)
2384 GOTO(out, rc);
2385
2386inactive_tgt:
2387 ext_count = fm_local->fm_mapped_extents;
2388 if (ext_count == 0) {
2389 ost_done = 1;
2390 /* If last stripe has hole at the end,
2391 * then we need to return */
2392 if (cur_stripe_wrap == last_stripe) {
2393 fiemap->fm_mapped_extents = 0;
2394 goto finish;
2395 }
2396 break;
2397 }
2398
2399 /* If we just need num of extents then go to next device */
2400 if (get_num_extents) {
2401 current_extent += ext_count;
2402 break;
2403 }
2404
2405 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2406 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2407
2408 /* Have we finished mapping on this device? */
2409 if (req_fm_len <= len_mapped_single_call)
2410 ost_done = 1;
2411
2412 /* Clear the EXTENT_LAST flag which can be present on
2413 * last extent */
2414 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2415 lcl_fm_ext[ext_count - 1].fe_flags &=
2416 ~FIEMAP_EXTENT_LAST;
2417
2418 curr_loc = lov_stripe_size(lsm,
2419 lcl_fm_ext[ext_count - 1].fe_logical+
2420 lcl_fm_ext[ext_count - 1].fe_length,
2421 cur_stripe);
2422 if (curr_loc >= fm_key->oa.o_size)
2423 ost_eof = 1;
2424
2425 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2426 ost_index, ext_count,
2427 current_extent);
2428
2429 current_extent += ext_count;
2430
2431 /* Ran out of available extents? */
2432 if (current_extent >= fiemap->fm_extent_count)
2433 goto finish;
2434 } while (ost_done == 0 && ost_eof == 0);
2435
2436 if (cur_stripe_wrap == last_stripe)
2437 goto finish;
2438 }
2439
2440finish:
2441 /* Indicate that we are returning device offsets unless file just has
2442 * single stripe */
2443 if (lsm->lsm_stripe_count > 1)
2444 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2445
2446 if (get_num_extents)
2447 goto skip_last_device_calc;
2448
2449 /* Check if we have reached the last stripe and whether mapping for that
2450 * stripe is done. */
2451 if (cur_stripe_wrap == last_stripe) {
2452 if (ost_done || ost_eof)
2453 fiemap->fm_extents[current_extent - 1].fe_flags |=
2454 FIEMAP_EXTENT_LAST;
2455 }
2456
2457skip_last_device_calc:
2458 fiemap->fm_mapped_extents = current_extent;
2459
2460out:
2461 OBD_FREE_LARGE(fm_local, buffer_size);
2462 return rc;
2463}
2464
2465static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2466 __u32 keylen, void *key, __u32 *vallen, void *val,
2467 struct lov_stripe_md *lsm)
2468{
2469 struct obd_device *obddev = class_exp2obd(exp);
2470 struct lov_obd *lov = &obddev->u.lov;
2471 int i, rc;
2472 ENTRY;
2473
2474 if (!vallen || !val)
2475 RETURN(-EFAULT);
2476
2477 obd_getref(obddev);
2478
2479 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2480 struct {
2481 char name[16];
2482 struct ldlm_lock *lock;
2483 } *data = key;
2484 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2485 struct lov_oinfo *loi;
2486 __u32 *stripe = val;
2487
2488 if (*vallen < sizeof(*stripe))
2489 GOTO(out, rc = -EFAULT);
2490 *vallen = sizeof(*stripe);
2491
2492 /* XXX This is another one of those bits that will need to
2493 * change if we ever actually support nested LOVs. It uses
2494 * the lock's export to find out which stripe it is. */
2495 /* XXX - it's assumed all the locks for deleted OSTs have
2496 * been cancelled. Also, the export for deleted OSTs will
2497 * be NULL and won't match the lock's export. */
2498 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2499 loi = lsm->lsm_oinfo[i];
2500 if (!lov->lov_tgts[loi->loi_ost_idx])
2501 continue;
2502 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2503 data->lock->l_conn_export &&
2504 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2505 *stripe = i;
2506 GOTO(out, rc = 0);
2507 }
2508 }
2509 LDLM_ERROR(data->lock, "lock on inode without such object");
2510 dump_lsm(D_ERROR, lsm);
2511 GOTO(out, rc = -ENXIO);
2512 } else if (KEY_IS(KEY_LAST_ID)) {
2513 struct obd_id_info *info = val;
2514 __u32 size = sizeof(obd_id);
2515 struct lov_tgt_desc *tgt;
2516
2517 LASSERT(*vallen == sizeof(struct obd_id_info));
2518 tgt = lov->lov_tgts[info->idx];
2519
2520 if (!tgt || !tgt->ltd_active)
2521 GOTO(out, rc = -ESRCH);
2522
2523 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2524 &size, info->data, NULL);
2525 GOTO(out, rc = 0);
2526 } else if (KEY_IS(KEY_LOVDESC)) {
2527 struct lov_desc *desc_ret = val;
2528 *desc_ret = lov->desc;
2529
2530 GOTO(out, rc = 0);
2531 } else if (KEY_IS(KEY_FIEMAP)) {
2532 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2533 GOTO(out, rc);
2534 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2535 struct lov_tgt_desc *tgt;
2536 __u64 ost_idx = *((__u64*)val);
2537
2538 LASSERT(*vallen == sizeof(__u64));
2539 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2540 tgt = lov->lov_tgts[ost_idx];
2541
2542 if (!tgt || !tgt->ltd_exp)
2543 GOTO(out, rc = -ESRCH);
2544
2545 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2546 GOTO(out, rc = 0);
2547 } else if (KEY_IS(KEY_TGT_COUNT)) {
2548 *((int *)val) = lov->desc.ld_tgt_count;
2549 GOTO(out, rc = 0);
2550 }
2551
2552 rc = -EINVAL;
2553
2554out:
2555 obd_putref(obddev);
2556 RETURN(rc);
2557}
2558
2559static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2560 obd_count keylen, void *key, obd_count vallen,
2561 void *val, struct ptlrpc_request_set *set)
2562{
2563 struct obd_device *obddev = class_exp2obd(exp);
2564 struct lov_obd *lov = &obddev->u.lov;
2565 obd_count count;
2566 int i, rc = 0, err;
2567 struct lov_tgt_desc *tgt;
2568 unsigned incr, check_uuid,
2569 do_inactive, no_set;
2570 unsigned next_id = 0, mds_con = 0, capa = 0;
2571 ENTRY;
2572
2573 incr = check_uuid = do_inactive = no_set = 0;
2574 if (set == NULL) {
2575 no_set = 1;
2576 set = ptlrpc_prep_set();
2577 if (!set)
2578 RETURN(-ENOMEM);
2579 }
2580
2581 obd_getref(obddev);
2582 count = lov->desc.ld_tgt_count;
2583
2584 if (KEY_IS(KEY_NEXT_ID)) {
2585 count = vallen / sizeof(struct obd_id_info);
2586 vallen = sizeof(obd_id);
2587 incr = sizeof(struct obd_id_info);
2588 do_inactive = 1;
2589 next_id = 1;
2590 } else if (KEY_IS(KEY_CHECKSUM)) {
2591 do_inactive = 1;
2592 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2593 /* use defaults: do_inactive = incr = 0; */
2594 } else if (KEY_IS(KEY_MDS_CONN)) {
2595 mds_con = 1;
2596 } else if (KEY_IS(KEY_CAPA_KEY)) {
2597 capa = 1;
2598 } else if (KEY_IS(KEY_CACHE_SET)) {
2599 LASSERT(lov->lov_cache == NULL);
2600 lov->lov_cache = val;
2601 do_inactive = 1;
2602 }
2603
2604 for (i = 0; i < count; i++, val = (char *)val + incr) {
2605 if (next_id) {
2606 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2607 } else {
2608 tgt = lov->lov_tgts[i];
2609 }
2610 /* OST was disconnected */
2611 if (!tgt || !tgt->ltd_exp)
2612 continue;
2613
2614 /* OST is inactive and we don't want inactive OSCs */
2615 if (!tgt->ltd_active && !do_inactive)
2616 continue;
2617
2618 if (mds_con) {
2619 struct mds_group_info *mgi;
2620
2621 LASSERT(vallen == sizeof(*mgi));
2622 mgi = (struct mds_group_info *)val;
2623
2624 /* Only want a specific OSC */
2625 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2626 &tgt->ltd_uuid))
2627 continue;
2628
2629 err = obd_set_info_async(env, tgt->ltd_exp,
2630 keylen, key, sizeof(int),
2631 &mgi->group, set);
2632 } else if (next_id) {
2633 err = obd_set_info_async(env, tgt->ltd_exp,
2634 keylen, key, vallen,
2635 ((struct obd_id_info*)val)->data, set);
2636 } else if (capa) {
2637 struct mds_capa_info *info = (struct mds_capa_info*)val;
2638
2639 LASSERT(vallen == sizeof(*info));
2640
2641 /* Only want a specific OSC */
2642 if (info->uuid &&
2643 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2644 continue;
2645
2646 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2647 key, sizeof(*info->capa),
2648 info->capa, set);
2649 } else {
2650 /* Only want a specific OSC */
2651 if (check_uuid &&
2652 !obd_uuid_equals(val, &tgt->ltd_uuid))
2653 continue;
2654
2655 err = obd_set_info_async(env, tgt->ltd_exp,
2656 keylen, key, vallen, val, set);
2657 }
2658
2659 if (!rc)
2660 rc = err;
2661 }
2662
2663 obd_putref(obddev);
2664 if (no_set) {
2665 err = ptlrpc_set_wait(set);
2666 if (!rc)
2667 rc = err;
2668 ptlrpc_set_destroy(set);
2669 }
2670 RETURN(rc);
2671}
2672
2673static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2674 int cmd, __u64 *offset)
2675{
2676 __u32 ssize = lsm->lsm_stripe_size;
2677 __u64 start;
2678
2679 start = *offset;
2680 lov_do_div64(start, ssize);
2681 start = start * ssize;
2682
2683 CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2684 ", end "LPU64"\n", *offset, ssize, start,
2685 start + ssize - 1);
2686 if (cmd == OBD_CALC_STRIPE_END) {
2687 *offset = start + ssize - 1;
2688 } else if (cmd == OBD_CALC_STRIPE_START) {
2689 *offset = start;
2690 } else {
2691 LBUG();
2692 }
2693
2694 RETURN(0);
2695}
2696
2697void lov_stripe_lock(struct lov_stripe_md *md)
2698{
2699 LASSERT(md->lsm_lock_owner != current_pid());
2700 spin_lock(&md->lsm_lock);
2701 LASSERT(md->lsm_lock_owner == 0);
2702 md->lsm_lock_owner = current_pid();
2703}
2704EXPORT_SYMBOL(lov_stripe_lock);
2705
2706void lov_stripe_unlock(struct lov_stripe_md *md)
2707{
2708 LASSERT(md->lsm_lock_owner == current_pid());
2709 md->lsm_lock_owner = 0;
2710 spin_unlock(&md->lsm_lock);
2711}
2712EXPORT_SYMBOL(lov_stripe_unlock);
2713
2714static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2715 struct obd_quotactl *oqctl)
2716{
2717 struct lov_obd *lov = &obd->u.lov;
2718 struct lov_tgt_desc *tgt;
2719 __u64 curspace = 0;
2720 __u64 bhardlimit = 0;
2721 int i, rc = 0;
2722 ENTRY;
2723
2724 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2725 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2726 oqctl->qc_cmd != Q_GETOQUOTA &&
2727 oqctl->qc_cmd != Q_INITQUOTA &&
2728 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2729 oqctl->qc_cmd != Q_FINVALIDATE) {
2730 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2731 RETURN(-EFAULT);
2732 }
2733
2734 /* for lov tgt */
2735 obd_getref(obd);
2736 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2737 int err;
2738
2739 tgt = lov->lov_tgts[i];
2740
2741 if (!tgt)
2742 continue;
2743
2744 if (!tgt->ltd_active || tgt->ltd_reap) {
2745 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2746 lov->lov_tgts[i]->ltd_activate) {
2747 rc = -EREMOTEIO;
2748 CERROR("ost %d is inactive\n", i);
2749 } else {
2750 CDEBUG(D_HA, "ost %d is inactive\n", i);
2751 }
2752 continue;
2753 }
2754
2755 err = obd_quotactl(tgt->ltd_exp, oqctl);
2756 if (err) {
2757 if (tgt->ltd_active && !rc)
2758 rc = err;
2759 continue;
2760 }
2761
2762 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2763 curspace += oqctl->qc_dqblk.dqb_curspace;
2764 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2765 }
2766 }
2767 obd_putref(obd);
2768
2769 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2770 oqctl->qc_dqblk.dqb_curspace = curspace;
2771 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2772 }
2773 RETURN(rc);
2774}
2775
2776static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2777 struct obd_quotactl *oqctl)
2778{
2779 struct lov_obd *lov = &obd->u.lov;
2780 int i, rc = 0;
2781 ENTRY;
2782
2783 obd_getref(obd);
2784
2785 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2786 if (!lov->lov_tgts[i])
2787 continue;
2788
2789 /* Skip quota check on the administratively disabled OSTs. */
2790 if (!lov->lov_tgts[i]->ltd_activate) {
2791 CWARN("lov idx %d was administratively disabled, "
2792 "skip quotacheck on it.\n", i);
2793 continue;
2794 }
2795
2796 if (!lov->lov_tgts[i]->ltd_active) {
2797 CERROR("lov idx %d inactive\n", i);
2798 rc = -EIO;
2799 goto out;
2800 }
2801 }
2802
2803 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2804 int err;
2805
2806 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2807 continue;
2808
2809 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2810 if (err && !rc)
2811 rc = err;
2812 }
2813
2814out:
2815 obd_putref(obd);
2816
2817 RETURN(rc);
2818}
2819
2820struct obd_ops lov_obd_ops = {
2821 .o_owner = THIS_MODULE,
2822 .o_setup = lov_setup,
2823 .o_precleanup = lov_precleanup,
2824 .o_cleanup = lov_cleanup,
2825 //.o_process_config = lov_process_config,
2826 .o_connect = lov_connect,
2827 .o_disconnect = lov_disconnect,
2828 .o_statfs = lov_statfs,
2829 .o_statfs_async = lov_statfs_async,
2830 .o_packmd = lov_packmd,
2831 .o_unpackmd = lov_unpackmd,
2832 .o_create = lov_create,
2833 .o_destroy = lov_destroy,
2834 .o_getattr = lov_getattr,
2835 .o_getattr_async = lov_getattr_async,
2836 .o_setattr = lov_setattr,
2837 .o_setattr_async = lov_setattr_async,
2838 .o_brw = lov_brw,
2839 .o_merge_lvb = lov_merge_lvb,
2840 .o_adjust_kms = lov_adjust_kms,
2841 .o_punch = lov_punch,
2842 .o_sync = lov_sync,
2843 .o_enqueue = lov_enqueue,
2844 .o_change_cbdata = lov_change_cbdata,
2845 .o_find_cbdata = lov_find_cbdata,
2846 .o_cancel = lov_cancel,
2847 .o_cancel_unused = lov_cancel_unused,
2848 .o_iocontrol = lov_iocontrol,
2849 .o_get_info = lov_get_info,
2850 .o_set_info_async = lov_set_info_async,
2851 .o_extent_calc = lov_extent_calc,
2852 .o_llog_init = lov_llog_init,
2853 .o_llog_finish = lov_llog_finish,
2854 .o_notify = lov_notify,
2855 .o_pool_new = lov_pool_new,
2856 .o_pool_rem = lov_pool_remove,
2857 .o_pool_add = lov_pool_add,
2858 .o_pool_del = lov_pool_del,
2859 .o_getref = lov_getref,
2860 .o_putref = lov_putref,
2861 .o_quotactl = lov_quotactl,
2862 .o_quotacheck = lov_quotacheck,
2863};
2864
2865struct kmem_cache *lov_oinfo_slab;
2866
2867extern struct lu_kmem_descr lov_caches[];
2868
2869int __init lov_init(void)
2870{
2871 struct lprocfs_static_vars lvars = { 0 };
2872 int rc;
2873 ENTRY;
2874
2875 /* print an address of _any_ initialized kernel symbol from this
2876 * module, to allow debugging with gdb that doesn't support data
2877 * symbols from modules.*/
2878 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2879
2880 rc = lu_kmem_init(lov_caches);
2881 if (rc)
2882 return rc;
2883
2884 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2885 sizeof(struct lov_oinfo),
2886 0, SLAB_HWCACHE_ALIGN, NULL);
2887 if (lov_oinfo_slab == NULL) {
2888 lu_kmem_fini(lov_caches);
2889 return -ENOMEM;
2890 }
2891 lprocfs_lov_init_vars(&lvars);
2892
2893 rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2894 LUSTRE_LOV_NAME, &lov_device_type);
2895
2896 if (rc) {
2897 kmem_cache_destroy(lov_oinfo_slab);
2898 lu_kmem_fini(lov_caches);
2899 }
2900
2901 RETURN(rc);
2902}
2903
2904static void /*__exit*/ lov_exit(void)
2905{
2906 class_unregister_type(LUSTRE_LOV_NAME);
2907 kmem_cache_destroy(lov_oinfo_slab);
2908
2909 lu_kmem_fini(lov_caches);
2910}
2911
2912MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2913MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2914MODULE_LICENSE("GPL");
2915
2916cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
Linux kernel - Deliverables
This page took 0.184606 seconds and 5 git commands to generate.