Commit | Line | Data |
---|---|---|
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> | |
60 | #include <lclient.h> | |
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. */ | |
69 | static void lov_getref(struct obd_device *obd) | |
70 | { | |
71 | struct lov_obd *lov = &obd->u.lov; | |
72 | ||
73 | /* nobody gets through here until lov_putref is done */ | |
74 | mutex_lock(&lov->lov_lock); | |
75 | atomic_inc(&lov->lov_refcount); | |
76 | mutex_unlock(&lov->lov_lock); | |
77 | return; | |
78 | } | |
79 | ||
80 | static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt); | |
81 | ||
82 | static void lov_putref(struct obd_device *obd) | |
83 | { | |
84 | struct lov_obd *lov = &obd->u.lov; | |
85 | ||
86 | mutex_lock(&lov->lov_lock); | |
87 | /* ok to dec to 0 more than once -- ltd_exp's will be null */ | |
88 | if (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 | ||
119 | static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid, | |
120 | enum obd_notify_event ev); | |
121 | static int lov_notify(struct obd_device *obd, struct obd_device *watched, | |
122 | enum obd_notify_event ev, void *data); | |
123 | ||
124 | ||
125 | #define MAX_STRING_SIZE 128 | |
126 | int 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 | ||
221 | static 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 | ||
276 | static 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 | ||
321 | static 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 | ||
351 | out: | |
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 | */ | |
363 | static 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 | ||
444 | static 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 | ||
526 | static 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 | ||
650 | out: | |
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 */ | |
661 | int 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 */ | |
699 | out: | |
700 | obd_putref(obd); | |
701 | up_write(&lov->lov_notify_lock); | |
702 | ||
703 | RETURN(rc); | |
704 | } | |
705 | ||
706 | static 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 | ||
731 | void lov_fix_desc_stripe_size(__u64 *val) | |
732 | { | |
733 | if (*val < LOV_DEFAULT_STRIPE_SIZE) { | |
734 | LCONSOLE_WARN("Increasing default stripe size to min %u\n", | |
735 | LOV_DEFAULT_STRIPE_SIZE); | |
736 | *val = LOV_DEFAULT_STRIPE_SIZE; | |
737 | } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) { | |
738 | *val &= ~(LOV_MIN_STRIPE_SIZE - 1); | |
739 | LCONSOLE_WARN("Changing default stripe size to "LPU64" (a " | |
740 | "multiple of %u)\n", | |
741 | *val, LOV_MIN_STRIPE_SIZE); | |
742 | } | |
743 | } | |
744 | ||
745 | void lov_fix_desc_stripe_count(__u32 *val) | |
746 | { | |
747 | if (*val == 0) | |
748 | *val = 1; | |
749 | } | |
750 | ||
751 | void lov_fix_desc_pattern(__u32 *val) | |
752 | { | |
753 | /* from lov_setstripe */ | |
754 | if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) { | |
755 | LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val); | |
756 | *val = 0; | |
757 | } | |
758 | } | |
759 | ||
760 | void lov_fix_desc_qos_maxage(__u32 *val) | |
761 | { | |
762 | /* fix qos_maxage */ | |
763 | if (*val == 0) | |
764 | *val = QOS_DEFAULT_MAXAGE; | |
765 | } | |
766 | ||
767 | void lov_fix_desc(struct lov_desc *desc) | |
768 | { | |
769 | lov_fix_desc_stripe_size(&desc->ld_default_stripe_size); | |
770 | lov_fix_desc_stripe_count(&desc->ld_default_stripe_count); | |
771 | lov_fix_desc_pattern(&desc->ld_pattern); | |
772 | lov_fix_desc_qos_maxage(&desc->ld_qos_maxage); | |
773 | } | |
774 | ||
775 | int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg) | |
776 | { | |
777 | struct lprocfs_static_vars lvars = { 0 }; | |
778 | struct lov_desc *desc; | |
779 | struct lov_obd *lov = &obd->u.lov; | |
780 | int rc; | |
781 | ENTRY; | |
782 | ||
783 | if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { | |
784 | CERROR("LOV setup requires a descriptor\n"); | |
785 | RETURN(-EINVAL); | |
786 | } | |
787 | ||
788 | desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1); | |
789 | ||
790 | if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) { | |
791 | CERROR("descriptor size wrong: %d > %d\n", | |
792 | (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1)); | |
793 | RETURN(-EINVAL); | |
794 | } | |
795 | ||
796 | if (desc->ld_magic != LOV_DESC_MAGIC) { | |
797 | if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) { | |
798 | CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n", | |
799 | obd->obd_name, desc); | |
800 | lustre_swab_lov_desc(desc); | |
801 | } else { | |
802 | CERROR("%s: Bad lov desc magic: %#x\n", | |
803 | obd->obd_name, desc->ld_magic); | |
804 | RETURN(-EINVAL); | |
805 | } | |
806 | } | |
807 | ||
808 | lov_fix_desc(desc); | |
809 | ||
810 | desc->ld_active_tgt_count = 0; | |
811 | lov->desc = *desc; | |
812 | lov->lov_tgt_size = 0; | |
813 | ||
814 | mutex_init(&lov->lov_lock); | |
815 | atomic_set(&lov->lov_refcount, 0); | |
816 | lov->lov_sp_me = LUSTRE_SP_CLI; | |
817 | ||
818 | init_rwsem(&lov->lov_notify_lock); | |
819 | ||
820 | lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS, | |
821 | HASH_POOLS_MAX_BITS, | |
822 | HASH_POOLS_BKT_BITS, 0, | |
823 | CFS_HASH_MIN_THETA, | |
824 | CFS_HASH_MAX_THETA, | |
825 | &pool_hash_operations, | |
826 | CFS_HASH_DEFAULT); | |
827 | INIT_LIST_HEAD(&lov->lov_pool_list); | |
828 | lov->lov_pool_count = 0; | |
829 | rc = lov_ost_pool_init(&lov->lov_packed, 0); | |
830 | if (rc) | |
831 | GOTO(out, rc); | |
832 | ||
833 | lprocfs_lov_init_vars(&lvars); | |
834 | lprocfs_obd_setup(obd, lvars.obd_vars); | |
835 | #ifdef LPROCFS | |
836 | { | |
837 | int rc; | |
838 | ||
839 | rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", | |
840 | 0444, &lov_proc_target_fops, obd); | |
841 | if (rc) | |
842 | CWARN("Error adding the target_obd file\n"); | |
843 | } | |
844 | #endif | |
845 | lov->lov_pool_proc_entry = lprocfs_register("pools", | |
846 | obd->obd_proc_entry, | |
847 | NULL, NULL); | |
848 | ||
849 | RETURN(0); | |
850 | ||
851 | out: | |
852 | return rc; | |
853 | } | |
854 | ||
855 | static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) | |
856 | { | |
857 | int rc = 0; | |
858 | struct lov_obd *lov = &obd->u.lov; | |
859 | ||
860 | ENTRY; | |
861 | ||
862 | switch (stage) { | |
863 | case OBD_CLEANUP_EARLY: { | |
864 | int i; | |
865 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { | |
866 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) | |
867 | continue; | |
868 | obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp), | |
869 | OBD_CLEANUP_EARLY); | |
870 | } | |
871 | break; | |
872 | } | |
873 | case OBD_CLEANUP_EXPORTS: | |
874 | rc = obd_llog_finish(obd, 0); | |
875 | if (rc != 0) | |
876 | CERROR("failed to cleanup llogging subsystems\n"); | |
877 | break; | |
878 | } | |
879 | RETURN(rc); | |
880 | } | |
881 | ||
882 | static int lov_cleanup(struct obd_device *obd) | |
883 | { | |
884 | struct lov_obd *lov = &obd->u.lov; | |
885 | struct list_head *pos, *tmp; | |
886 | struct pool_desc *pool; | |
887 | ENTRY; | |
888 | ||
889 | list_for_each_safe(pos, tmp, &lov->lov_pool_list) { | |
890 | pool = list_entry(pos, struct pool_desc, pool_list); | |
891 | /* free pool structs */ | |
892 | CDEBUG(D_INFO, "delete pool %p\n", pool); | |
893 | /* In the function below, .hs_keycmp resolves to | |
894 | * pool_hashkey_keycmp() */ | |
895 | /* coverity[overrun-buffer-val] */ | |
896 | lov_pool_del(obd, pool->pool_name); | |
897 | } | |
898 | cfs_hash_putref(lov->lov_pools_hash_body); | |
899 | lov_ost_pool_free(&lov->lov_packed); | |
900 | ||
901 | lprocfs_obd_cleanup(obd); | |
902 | if (lov->lov_tgts) { | |
903 | int i; | |
904 | obd_getref(obd); | |
905 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { | |
906 | if (!lov->lov_tgts[i]) | |
907 | continue; | |
908 | ||
909 | /* Inactive targets may never have connected */ | |
910 | if (lov->lov_tgts[i]->ltd_active || | |
911 | atomic_read(&lov->lov_refcount)) | |
912 | /* We should never get here - these | |
913 | should have been removed in the | |
914 | disconnect. */ | |
915 | CERROR("lov tgt %d not cleaned!" | |
916 | " deathrow=%d, lovrc=%d\n", | |
917 | i, lov->lov_death_row, | |
918 | atomic_read(&lov->lov_refcount)); | |
919 | lov_del_target(obd, i, 0, 0); | |
920 | } | |
921 | obd_putref(obd); | |
922 | OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) * | |
923 | lov->lov_tgt_size); | |
924 | lov->lov_tgt_size = 0; | |
925 | } | |
926 | RETURN(0); | |
927 | } | |
928 | ||
929 | int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg, | |
930 | __u32 *indexp, int *genp) | |
931 | { | |
932 | struct obd_uuid obd_uuid; | |
933 | int cmd; | |
934 | int rc = 0; | |
935 | ENTRY; | |
936 | ||
937 | switch(cmd = lcfg->lcfg_command) { | |
938 | case LCFG_LOV_ADD_OBD: | |
939 | case LCFG_LOV_ADD_INA: | |
940 | case LCFG_LOV_DEL_OBD: { | |
941 | __u32 index; | |
942 | int gen; | |
943 | /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */ | |
944 | if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) | |
945 | GOTO(out, rc = -EINVAL); | |
946 | ||
947 | obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1)); | |
948 | ||
949 | if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1) | |
950 | GOTO(out, rc = -EINVAL); | |
951 | if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1) | |
952 | GOTO(out, rc = -EINVAL); | |
953 | index = *indexp; | |
954 | gen = *genp; | |
955 | if (cmd == LCFG_LOV_ADD_OBD) | |
956 | rc = lov_add_target(obd, &obd_uuid, index, gen, 1); | |
957 | else if (cmd == LCFG_LOV_ADD_INA) | |
958 | rc = lov_add_target(obd, &obd_uuid, index, gen, 0); | |
959 | else | |
960 | rc = lov_del_target(obd, index, &obd_uuid, gen); | |
961 | GOTO(out, rc); | |
962 | } | |
963 | case LCFG_PARAM: { | |
964 | struct lprocfs_static_vars lvars = { 0 }; | |
965 | struct lov_desc *desc = &(obd->u.lov.desc); | |
966 | ||
967 | if (!desc) | |
968 | GOTO(out, rc = -EINVAL); | |
969 | ||
970 | lprocfs_lov_init_vars(&lvars); | |
971 | ||
972 | rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars, | |
973 | lcfg, obd); | |
974 | if (rc > 0) | |
975 | rc = 0; | |
976 | GOTO(out, rc); | |
977 | } | |
978 | case LCFG_POOL_NEW: | |
979 | case LCFG_POOL_ADD: | |
980 | case LCFG_POOL_DEL: | |
981 | case LCFG_POOL_REM: | |
982 | GOTO(out, rc); | |
983 | ||
984 | default: { | |
985 | CERROR("Unknown command: %d\n", lcfg->lcfg_command); | |
986 | GOTO(out, rc = -EINVAL); | |
987 | ||
988 | } | |
989 | } | |
990 | out: | |
991 | RETURN(rc); | |
992 | } | |
993 | ||
994 | static int lov_recreate(struct obd_export *exp, struct obdo *src_oa, | |
995 | struct lov_stripe_md **ea, struct obd_trans_info *oti) | |
996 | { | |
997 | struct lov_stripe_md *obj_mdp, *lsm; | |
998 | struct lov_obd *lov = &exp->exp_obd->u.lov; | |
999 | unsigned ost_idx; | |
1000 | int rc, i; | |
1001 | ENTRY; | |
1002 | ||
1003 | LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS && | |
1004 | src_oa->o_flags & OBD_FL_RECREATE_OBJS); | |
1005 | ||
1006 | OBD_ALLOC(obj_mdp, sizeof(*obj_mdp)); | |
1007 | if (obj_mdp == NULL) | |
1008 | RETURN(-ENOMEM); | |
1009 | ||
1010 | ost_idx = src_oa->o_nlink; | |
1011 | lsm = *ea; | |
1012 | if (lsm == NULL) | |
1013 | GOTO(out, rc = -EINVAL); | |
1014 | if (ost_idx >= lov->desc.ld_tgt_count || | |
1015 | !lov->lov_tgts[ost_idx]) | |
1016 | GOTO(out, rc = -EINVAL); | |
1017 | ||
1018 | for (i = 0; i < lsm->lsm_stripe_count; i++) { | |
1019 | if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) { | |
1020 | if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) != | |
1021 | ostid_id(&src_oa->o_oi)) | |
1022 | GOTO(out, rc = -EINVAL); | |
1023 | break; | |
1024 | } | |
1025 | } | |
1026 | if (i == lsm->lsm_stripe_count) | |
1027 | GOTO(out, rc = -EINVAL); | |
1028 | ||
1029 | rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp, | |
1030 | src_oa, &obj_mdp, oti); | |
1031 | out: | |
1032 | OBD_FREE(obj_mdp, sizeof(*obj_mdp)); | |
1033 | RETURN(rc); | |
1034 | } | |
1035 | ||
1036 | /* the LOV expects oa->o_id to be set to the LOV object id */ | |
1037 | static int lov_create(const struct lu_env *env, struct obd_export *exp, | |
1038 | struct obdo *src_oa, struct lov_stripe_md **ea, | |
1039 | struct obd_trans_info *oti) | |
1040 | { | |
1041 | struct lov_obd *lov; | |
1042 | int rc = 0; | |
1043 | ENTRY; | |
1044 | ||
1045 | LASSERT(ea != NULL); | |
1046 | if (exp == NULL) | |
1047 | RETURN(-EINVAL); | |
1048 | ||
1049 | if ((src_oa->o_valid & OBD_MD_FLFLAGS) && | |
1050 | src_oa->o_flags == OBD_FL_DELORPHAN) { | |
1051 | /* should be used with LOV anymore */ | |
1052 | LBUG(); | |
1053 | } | |
1054 | ||
1055 | lov = &exp->exp_obd->u.lov; | |
1056 | if (!lov->desc.ld_active_tgt_count) | |
1057 | RETURN(-EIO); | |
1058 | ||
1059 | obd_getref(exp->exp_obd); | |
1060 | /* Recreate a specific object id at the given OST index */ | |
1061 | if ((src_oa->o_valid & OBD_MD_FLFLAGS) && | |
1062 | (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) { | |
1063 | rc = lov_recreate(exp, src_oa, ea, oti); | |
1064 | } | |
1065 | ||
1066 | obd_putref(exp->exp_obd); | |
1067 | RETURN(rc); | |
1068 | } | |
1069 | ||
1070 | #define ASSERT_LSM_MAGIC(lsmp) \ | |
1071 | do { \ | |
1072 | LASSERT((lsmp) != NULL); \ | |
1073 | LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \ | |
1074 | (lsmp)->lsm_magic == LOV_MAGIC_V3), \ | |
1075 | "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \ | |
1076 | } while (0) | |
1077 | ||
1078 | static int lov_destroy(const struct lu_env *env, struct obd_export *exp, | |
1079 | struct obdo *oa, struct lov_stripe_md *lsm, | |
1080 | struct obd_trans_info *oti, struct obd_export *md_exp, | |
1081 | void *capa) | |
1082 | { | |
1083 | struct lov_request_set *set; | |
1084 | struct obd_info oinfo; | |
1085 | struct lov_request *req; | |
1086 | struct list_head *pos; | |
1087 | struct lov_obd *lov; | |
1088 | int rc = 0, err = 0; | |
1089 | ENTRY; | |
1090 | ||
1091 | ASSERT_LSM_MAGIC(lsm); | |
1092 | ||
1093 | if (!exp || !exp->exp_obd) | |
1094 | RETURN(-ENODEV); | |
1095 | ||
1096 | if (oa->o_valid & OBD_MD_FLCOOKIE) { | |
1097 | LASSERT(oti); | |
1098 | LASSERT(oti->oti_logcookies); | |
1099 | } | |
1100 | ||
1101 | lov = &exp->exp_obd->u.lov; | |
1102 | obd_getref(exp->exp_obd); | |
1103 | rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set); | |
1104 | if (rc) | |
1105 | GOTO(out, rc); | |
1106 | ||
1107 | list_for_each (pos, &set->set_list) { | |
1108 | req = list_entry(pos, struct lov_request, rq_link); | |
1109 | ||
1110 | if (oa->o_valid & OBD_MD_FLCOOKIE) | |
1111 | oti->oti_logcookies = set->set_cookies + req->rq_stripe; | |
1112 | ||
1113 | err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1114 | req->rq_oi.oi_oa, NULL, oti, NULL, capa); | |
1115 | err = lov_update_common_set(set, req, err); | |
1116 | if (err) { | |
1117 | CERROR("%s: destroying objid "DOSTID" subobj " | |
1118 | DOSTID" on OST idx %d: rc = %d\n", | |
1119 | exp->exp_obd->obd_name, POSTID(&oa->o_oi), | |
1120 | POSTID(&req->rq_oi.oi_oa->o_oi), | |
1121 | req->rq_idx, err); | |
1122 | if (!rc) | |
1123 | rc = err; | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | if (rc == 0) { | |
1128 | LASSERT(lsm_op_find(lsm->lsm_magic) != NULL); | |
1129 | rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp); | |
1130 | } | |
1131 | err = lov_fini_destroy_set(set); | |
1132 | out: | |
1133 | obd_putref(exp->exp_obd); | |
1134 | RETURN(rc ? rc : err); | |
1135 | } | |
1136 | ||
1137 | static int lov_getattr(const struct lu_env *env, struct obd_export *exp, | |
1138 | struct obd_info *oinfo) | |
1139 | { | |
1140 | struct lov_request_set *set; | |
1141 | struct lov_request *req; | |
1142 | struct list_head *pos; | |
1143 | struct lov_obd *lov; | |
1144 | int err = 0, rc = 0; | |
1145 | ENTRY; | |
1146 | ||
1147 | LASSERT(oinfo); | |
1148 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1149 | ||
1150 | if (!exp || !exp->exp_obd) | |
1151 | RETURN(-ENODEV); | |
1152 | ||
1153 | lov = &exp->exp_obd->u.lov; | |
1154 | ||
1155 | rc = lov_prep_getattr_set(exp, oinfo, &set); | |
1156 | if (rc) | |
1157 | RETURN(rc); | |
1158 | ||
1159 | list_for_each (pos, &set->set_list) { | |
1160 | req = list_entry(pos, struct lov_request, rq_link); | |
1161 | ||
1162 | CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx" | |
1163 | " %u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe, | |
1164 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx); | |
1165 | ||
1166 | rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1167 | &req->rq_oi); | |
1168 | err = lov_update_common_set(set, req, rc); | |
1169 | if (err) { | |
1170 | CERROR("%s: getattr objid "DOSTID" subobj " | |
1171 | DOSTID" on OST idx %d: rc = %d\n", | |
1172 | exp->exp_obd->obd_name, | |
1173 | POSTID(&oinfo->oi_oa->o_oi), | |
1174 | POSTID(&req->rq_oi.oi_oa->o_oi), | |
1175 | req->rq_idx, err); | |
1176 | break; | |
1177 | } | |
1178 | } | |
1179 | ||
1180 | rc = lov_fini_getattr_set(set); | |
1181 | if (err) | |
1182 | rc = err; | |
1183 | RETURN(rc); | |
1184 | } | |
1185 | ||
1186 | static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, | |
1187 | void *data, int rc) | |
1188 | { | |
1189 | struct lov_request_set *lovset = (struct lov_request_set *)data; | |
1190 | int err; | |
1191 | ENTRY; | |
1192 | ||
1193 | /* don't do attribute merge if this aysnc op failed */ | |
1194 | if (rc) | |
1195 | atomic_set(&lovset->set_completes, 0); | |
1196 | err = lov_fini_getattr_set(lovset); | |
1197 | RETURN(rc ? rc : err); | |
1198 | } | |
1199 | ||
1200 | static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo, | |
1201 | struct ptlrpc_request_set *rqset) | |
1202 | { | |
1203 | struct lov_request_set *lovset; | |
1204 | struct lov_obd *lov; | |
1205 | struct list_head *pos; | |
1206 | struct lov_request *req; | |
1207 | int rc = 0, err; | |
1208 | ENTRY; | |
1209 | ||
1210 | LASSERT(oinfo); | |
1211 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1212 | ||
1213 | if (!exp || !exp->exp_obd) | |
1214 | RETURN(-ENODEV); | |
1215 | ||
1216 | lov = &exp->exp_obd->u.lov; | |
1217 | ||
1218 | rc = lov_prep_getattr_set(exp, oinfo, &lovset); | |
1219 | if (rc) | |
1220 | RETURN(rc); | |
1221 | ||
1222 | CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n", | |
1223 | POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count, | |
1224 | oinfo->oi_md->lsm_stripe_size); | |
1225 | ||
1226 | list_for_each(pos, &lovset->set_list) { | |
1227 | req = list_entry(pos, struct lov_request, rq_link); | |
1228 | ||
1229 | CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx" | |
1230 | "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe, | |
1231 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx); | |
1232 | rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1233 | &req->rq_oi, rqset); | |
1234 | if (rc) { | |
1235 | CERROR("%s: getattr objid "DOSTID" subobj" | |
1236 | DOSTID" on OST idx %d: rc = %d\n", | |
1237 | exp->exp_obd->obd_name, | |
1238 | POSTID(&oinfo->oi_oa->o_oi), | |
1239 | POSTID(&req->rq_oi.oi_oa->o_oi), | |
1240 | req->rq_idx, rc); | |
1241 | GOTO(out, rc); | |
1242 | } | |
1243 | } | |
1244 | ||
1245 | if (!list_empty(&rqset->set_requests)) { | |
1246 | LASSERT(rc == 0); | |
1247 | LASSERT (rqset->set_interpret == NULL); | |
1248 | rqset->set_interpret = lov_getattr_interpret; | |
1249 | rqset->set_arg = (void *)lovset; | |
1250 | RETURN(rc); | |
1251 | } | |
1252 | out: | |
1253 | if (rc) | |
1254 | atomic_set(&lovset->set_completes, 0); | |
1255 | err = lov_fini_getattr_set(lovset); | |
1256 | RETURN(rc ? rc : err); | |
1257 | } | |
1258 | ||
1259 | static int lov_setattr(const struct lu_env *env, struct obd_export *exp, | |
1260 | struct obd_info *oinfo, struct obd_trans_info *oti) | |
1261 | { | |
1262 | struct lov_request_set *set; | |
1263 | struct lov_obd *lov; | |
1264 | struct list_head *pos; | |
1265 | struct lov_request *req; | |
1266 | int err = 0, rc = 0; | |
1267 | ENTRY; | |
1268 | ||
1269 | LASSERT(oinfo); | |
1270 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1271 | ||
1272 | if (!exp || !exp->exp_obd) | |
1273 | RETURN(-ENODEV); | |
1274 | ||
1275 | /* for now, we only expect the following updates here */ | |
1276 | LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE | | |
1277 | OBD_MD_FLMODE | OBD_MD_FLATIME | | |
1278 | OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
1279 | OBD_MD_FLFLAGS | OBD_MD_FLSIZE | | |
1280 | OBD_MD_FLGROUP | OBD_MD_FLUID | | |
1281 | OBD_MD_FLGID | OBD_MD_FLFID | | |
1282 | OBD_MD_FLGENER))); | |
1283 | lov = &exp->exp_obd->u.lov; | |
1284 | rc = lov_prep_setattr_set(exp, oinfo, oti, &set); | |
1285 | if (rc) | |
1286 | RETURN(rc); | |
1287 | ||
1288 | list_for_each (pos, &set->set_list) { | |
1289 | req = list_entry(pos, struct lov_request, rq_link); | |
1290 | ||
1291 | rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1292 | &req->rq_oi, NULL); | |
1293 | err = lov_update_setattr_set(set, req, rc); | |
1294 | if (err) { | |
1295 | CERROR("%s: setattr objid "DOSTID" subobj " | |
1296 | DOSTID" on OST idx %d: rc = %d\n", | |
1297 | exp->exp_obd->obd_name, | |
1298 | POSTID(&set->set_oi->oi_oa->o_oi), | |
1299 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, | |
1300 | err); | |
1301 | if (!rc) | |
1302 | rc = err; | |
1303 | } | |
1304 | } | |
1305 | err = lov_fini_setattr_set(set); | |
1306 | if (!rc) | |
1307 | rc = err; | |
1308 | RETURN(rc); | |
1309 | } | |
1310 | ||
1311 | static int lov_setattr_interpret(struct ptlrpc_request_set *rqset, | |
1312 | void *data, int rc) | |
1313 | { | |
1314 | struct lov_request_set *lovset = (struct lov_request_set *)data; | |
1315 | int err; | |
1316 | ENTRY; | |
1317 | ||
1318 | if (rc) | |
1319 | atomic_set(&lovset->set_completes, 0); | |
1320 | err = lov_fini_setattr_set(lovset); | |
1321 | RETURN(rc ? rc : err); | |
1322 | } | |
1323 | ||
1324 | /* If @oti is given, the request goes from MDS and responses from OSTs are not | |
1325 | needed. Otherwise, a client is waiting for responses. */ | |
1326 | static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo, | |
1327 | struct obd_trans_info *oti, | |
1328 | struct ptlrpc_request_set *rqset) | |
1329 | { | |
1330 | struct lov_request_set *set; | |
1331 | struct lov_request *req; | |
1332 | struct list_head *pos; | |
1333 | struct lov_obd *lov; | |
1334 | int rc = 0; | |
1335 | ENTRY; | |
1336 | ||
1337 | LASSERT(oinfo); | |
1338 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1339 | if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) { | |
1340 | LASSERT(oti); | |
1341 | LASSERT(oti->oti_logcookies); | |
1342 | } | |
1343 | ||
1344 | if (!exp || !exp->exp_obd) | |
1345 | RETURN(-ENODEV); | |
1346 | ||
1347 | lov = &exp->exp_obd->u.lov; | |
1348 | rc = lov_prep_setattr_set(exp, oinfo, oti, &set); | |
1349 | if (rc) | |
1350 | RETURN(rc); | |
1351 | ||
1352 | CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n", | |
1353 | POSTID(&oinfo->oi_md->lsm_oi), | |
1354 | oinfo->oi_md->lsm_stripe_count, | |
1355 | oinfo->oi_md->lsm_stripe_size); | |
1356 | ||
1357 | list_for_each(pos, &set->set_list) { | |
1358 | req = list_entry(pos, struct lov_request, rq_link); | |
1359 | ||
1360 | if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) | |
1361 | oti->oti_logcookies = set->set_cookies + req->rq_stripe; | |
1362 | ||
1363 | CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx" | |
1364 | "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe, | |
1365 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx); | |
1366 | ||
1367 | rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1368 | &req->rq_oi, oti, rqset); | |
1369 | if (rc) { | |
1370 | CERROR("error: setattr objid "DOSTID" subobj" | |
1371 | DOSTID" on OST idx %d: rc = %d\n", | |
1372 | POSTID(&set->set_oi->oi_oa->o_oi), | |
1373 | POSTID(&req->rq_oi.oi_oa->o_oi), | |
1374 | req->rq_idx, rc); | |
1375 | break; | |
1376 | } | |
1377 | } | |
1378 | ||
1379 | /* If we are not waiting for responses on async requests, return. */ | |
1380 | if (rc || !rqset || list_empty(&rqset->set_requests)) { | |
1381 | int err; | |
1382 | if (rc) | |
1383 | atomic_set(&set->set_completes, 0); | |
1384 | err = lov_fini_setattr_set(set); | |
1385 | RETURN(rc ? rc : err); | |
1386 | } | |
1387 | ||
1388 | LASSERT(rqset->set_interpret == NULL); | |
1389 | rqset->set_interpret = lov_setattr_interpret; | |
1390 | rqset->set_arg = (void *)set; | |
1391 | ||
1392 | RETURN(0); | |
1393 | } | |
1394 | ||
1395 | static int lov_punch_interpret(struct ptlrpc_request_set *rqset, | |
1396 | void *data, int rc) | |
1397 | { | |
1398 | struct lov_request_set *lovset = (struct lov_request_set *)data; | |
1399 | int err; | |
1400 | ENTRY; | |
1401 | ||
1402 | if (rc) | |
1403 | atomic_set(&lovset->set_completes, 0); | |
1404 | err = lov_fini_punch_set(lovset); | |
1405 | RETURN(rc ? rc : err); | |
1406 | } | |
1407 | ||
1408 | /* FIXME: maybe we'll just make one node the authoritative attribute node, then | |
1409 | * we can send this 'punch' to just the authoritative node and the nodes | |
1410 | * that the punch will affect. */ | |
1411 | static int lov_punch(const struct lu_env *env, struct obd_export *exp, | |
1412 | struct obd_info *oinfo, struct obd_trans_info *oti, | |
1413 | struct ptlrpc_request_set *rqset) | |
1414 | { | |
1415 | struct lov_request_set *set; | |
1416 | struct lov_obd *lov; | |
1417 | struct list_head *pos; | |
1418 | struct lov_request *req; | |
1419 | int rc = 0; | |
1420 | ENTRY; | |
1421 | ||
1422 | LASSERT(oinfo); | |
1423 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1424 | ||
1425 | if (!exp || !exp->exp_obd) | |
1426 | RETURN(-ENODEV); | |
1427 | ||
1428 | lov = &exp->exp_obd->u.lov; | |
1429 | rc = lov_prep_punch_set(exp, oinfo, oti, &set); | |
1430 | if (rc) | |
1431 | RETURN(rc); | |
1432 | ||
1433 | list_for_each (pos, &set->set_list) { | |
1434 | req = list_entry(pos, struct lov_request, rq_link); | |
1435 | ||
1436 | rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1437 | &req->rq_oi, NULL, rqset); | |
1438 | if (rc) { | |
1439 | CERROR("%s: punch objid "DOSTID" subobj "DOSTID | |
1440 | " on OST idx %d: rc = %d\n", | |
1441 | exp->exp_obd->obd_name, | |
1442 | POSTID(&set->set_oi->oi_oa->o_oi), | |
1443 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, rc); | |
1444 | break; | |
1445 | } | |
1446 | } | |
1447 | ||
1448 | if (rc || list_empty(&rqset->set_requests)) { | |
1449 | int err; | |
1450 | err = lov_fini_punch_set(set); | |
1451 | RETURN(rc ? rc : err); | |
1452 | } | |
1453 | ||
1454 | LASSERT(rqset->set_interpret == NULL); | |
1455 | rqset->set_interpret = lov_punch_interpret; | |
1456 | rqset->set_arg = (void *)set; | |
1457 | ||
1458 | RETURN(0); | |
1459 | } | |
1460 | ||
1461 | static int lov_sync_interpret(struct ptlrpc_request_set *rqset, | |
1462 | void *data, int rc) | |
1463 | { | |
1464 | struct lov_request_set *lovset = data; | |
1465 | int err; | |
1466 | ENTRY; | |
1467 | ||
1468 | if (rc) | |
1469 | atomic_set(&lovset->set_completes, 0); | |
1470 | err = lov_fini_sync_set(lovset); | |
1471 | RETURN(rc ?: err); | |
1472 | } | |
1473 | ||
1474 | static int lov_sync(const struct lu_env *env, struct obd_export *exp, | |
1475 | struct obd_info *oinfo, obd_off start, obd_off end, | |
1476 | struct ptlrpc_request_set *rqset) | |
1477 | { | |
1478 | struct lov_request_set *set = NULL; | |
1479 | struct lov_obd *lov; | |
1480 | struct list_head *pos; | |
1481 | struct lov_request *req; | |
1482 | int rc = 0; | |
1483 | ENTRY; | |
1484 | ||
1485 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1486 | LASSERT(rqset != NULL); | |
1487 | ||
1488 | if (!exp->exp_obd) | |
1489 | RETURN(-ENODEV); | |
1490 | ||
1491 | lov = &exp->exp_obd->u.lov; | |
1492 | rc = lov_prep_sync_set(exp, oinfo, start, end, &set); | |
1493 | if (rc) | |
1494 | RETURN(rc); | |
1495 | ||
1496 | CDEBUG(D_INFO, "fsync objid "DOSTID" ["LPX64", "LPX64"]\n", | |
1497 | POSTID(&set->set_oi->oi_oa->o_oi), start, end); | |
1498 | ||
1499 | list_for_each (pos, &set->set_list) { | |
1500 | req = list_entry(pos, struct lov_request, rq_link); | |
1501 | ||
1502 | rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1503 | &req->rq_oi, req->rq_oi.oi_policy.l_extent.start, | |
1504 | req->rq_oi.oi_policy.l_extent.end, rqset); | |
1505 | if (rc) { | |
1506 | CERROR("%s: fsync objid "DOSTID" subobj "DOSTID | |
1507 | " on OST idx %d: rc = %d\n", | |
1508 | exp->exp_obd->obd_name, | |
1509 | POSTID(&set->set_oi->oi_oa->o_oi), | |
1510 | POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx, | |
1511 | rc); | |
1512 | break; | |
1513 | } | |
1514 | } | |
1515 | ||
1516 | /* If we are not waiting for responses on async requests, return. */ | |
1517 | if (rc || list_empty(&rqset->set_requests)) { | |
1518 | int err = lov_fini_sync_set(set); | |
1519 | ||
1520 | RETURN(rc ?: err); | |
1521 | } | |
1522 | ||
1523 | LASSERT(rqset->set_interpret == NULL); | |
1524 | rqset->set_interpret = lov_sync_interpret; | |
1525 | rqset->set_arg = (void *)set; | |
1526 | ||
1527 | RETURN(0); | |
1528 | } | |
1529 | ||
1530 | static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo, | |
1531 | obd_count oa_bufs, struct brw_page *pga) | |
1532 | { | |
1533 | struct obd_info oinfo = { { { 0 } } }; | |
1534 | int i, rc = 0; | |
1535 | ||
1536 | oinfo.oi_oa = lov_oinfo->oi_oa; | |
1537 | ||
1538 | /* The caller just wants to know if there's a chance that this | |
1539 | * I/O can succeed */ | |
1540 | for (i = 0; i < oa_bufs; i++) { | |
1541 | int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off); | |
1542 | int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx; | |
1543 | obd_off start, end; | |
1544 | ||
1545 | if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off, | |
1546 | pga[i].off + pga[i].count - 1, | |
1547 | &start, &end)) | |
1548 | continue; | |
1549 | ||
1550 | if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) { | |
1551 | CDEBUG(D_HA, "lov idx %d inactive\n", ost); | |
1552 | return -EIO; | |
1553 | } | |
1554 | ||
1555 | rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo, | |
1556 | 1, &pga[i], NULL); | |
1557 | if (rc) | |
1558 | break; | |
1559 | } | |
1560 | return rc; | |
1561 | } | |
1562 | ||
1563 | static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo, | |
1564 | obd_count oa_bufs, struct brw_page *pga, | |
1565 | struct obd_trans_info *oti) | |
1566 | { | |
1567 | struct lov_request_set *set; | |
1568 | struct lov_request *req; | |
1569 | struct list_head *pos; | |
1570 | struct lov_obd *lov = &exp->exp_obd->u.lov; | |
1571 | int err, rc = 0; | |
1572 | ENTRY; | |
1573 | ||
1574 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1575 | ||
1576 | if (cmd == OBD_BRW_CHECK) { | |
1577 | rc = lov_brw_check(lov, oinfo, oa_bufs, pga); | |
1578 | RETURN(rc); | |
1579 | } | |
1580 | ||
1581 | rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set); | |
1582 | if (rc) | |
1583 | RETURN(rc); | |
1584 | ||
1585 | list_for_each (pos, &set->set_list) { | |
1586 | struct obd_export *sub_exp; | |
1587 | struct brw_page *sub_pga; | |
1588 | req = list_entry(pos, struct lov_request, rq_link); | |
1589 | ||
1590 | sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp; | |
1591 | sub_pga = set->set_pga + req->rq_pgaidx; | |
1592 | rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs, | |
1593 | sub_pga, oti); | |
1594 | if (rc) | |
1595 | break; | |
1596 | lov_update_common_set(set, req, rc); | |
1597 | } | |
1598 | ||
1599 | err = lov_fini_brw_set(set); | |
1600 | if (!rc) | |
1601 | rc = err; | |
1602 | RETURN(rc); | |
1603 | } | |
1604 | ||
1605 | static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset, | |
1606 | void *data, int rc) | |
1607 | { | |
1608 | struct lov_request_set *lovset = (struct lov_request_set *)data; | |
1609 | ENTRY; | |
1610 | rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset); | |
1611 | RETURN(rc); | |
1612 | } | |
1613 | ||
1614 | static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo, | |
1615 | struct ldlm_enqueue_info *einfo, | |
1616 | struct ptlrpc_request_set *rqset) | |
1617 | { | |
1618 | ldlm_mode_t mode = einfo->ei_mode; | |
1619 | struct lov_request_set *set; | |
1620 | struct lov_request *req; | |
1621 | struct list_head *pos; | |
1622 | struct lov_obd *lov; | |
1623 | ldlm_error_t rc; | |
1624 | ENTRY; | |
1625 | ||
1626 | LASSERT(oinfo); | |
1627 | ASSERT_LSM_MAGIC(oinfo->oi_md); | |
1628 | LASSERT(mode == (mode & -mode)); | |
1629 | ||
1630 | /* we should never be asked to replay a lock this way. */ | |
1631 | LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0); | |
1632 | ||
1633 | if (!exp || !exp->exp_obd) | |
1634 | RETURN(-ENODEV); | |
1635 | ||
1636 | lov = &exp->exp_obd->u.lov; | |
1637 | rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set); | |
1638 | if (rc) | |
1639 | RETURN(rc); | |
1640 | ||
1641 | list_for_each (pos, &set->set_list) { | |
1642 | req = list_entry(pos, struct lov_request, rq_link); | |
1643 | ||
1644 | rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1645 | &req->rq_oi, einfo, rqset); | |
1646 | if (rc != ELDLM_OK) | |
1647 | GOTO(out, rc); | |
1648 | } | |
1649 | ||
1650 | if (rqset && !list_empty(&rqset->set_requests)) { | |
1651 | LASSERT(rc == 0); | |
1652 | LASSERT(rqset->set_interpret == NULL); | |
1653 | rqset->set_interpret = lov_enqueue_interpret; | |
1654 | rqset->set_arg = (void *)set; | |
1655 | RETURN(rc); | |
1656 | } | |
1657 | out: | |
1658 | rc = lov_fini_enqueue_set(set, mode, rc, rqset); | |
1659 | RETURN(rc); | |
1660 | } | |
1661 | ||
1662 | static int lov_change_cbdata(struct obd_export *exp, | |
1663 | struct lov_stripe_md *lsm, ldlm_iterator_t it, | |
1664 | void *data) | |
1665 | { | |
1666 | struct lov_obd *lov; | |
1667 | int rc = 0, i; | |
1668 | ENTRY; | |
1669 | ||
1670 | ASSERT_LSM_MAGIC(lsm); | |
1671 | ||
1672 | if (!exp || !exp->exp_obd) | |
1673 | RETURN(-ENODEV); | |
1674 | ||
1675 | lov = &exp->exp_obd->u.lov; | |
1676 | for (i = 0; i < lsm->lsm_stripe_count; i++) { | |
1677 | struct lov_stripe_md submd; | |
1678 | struct lov_oinfo *loi = lsm->lsm_oinfo[i]; | |
1679 | ||
1680 | if (!lov->lov_tgts[loi->loi_ost_idx]) { | |
1681 | CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx); | |
1682 | continue; | |
1683 | } | |
1684 | ||
1685 | submd.lsm_oi = loi->loi_oi; | |
1686 | submd.lsm_stripe_count = 0; | |
1687 | rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, | |
1688 | &submd, it, data); | |
1689 | } | |
1690 | RETURN(rc); | |
1691 | } | |
1692 | ||
1693 | /* find any ldlm lock of the inode in lov | |
1694 | * return 0 not find | |
1695 | * 1 find one | |
1696 | * < 0 error */ | |
1697 | static int lov_find_cbdata(struct obd_export *exp, | |
1698 | struct lov_stripe_md *lsm, ldlm_iterator_t it, | |
1699 | void *data) | |
1700 | { | |
1701 | struct lov_obd *lov; | |
1702 | int rc = 0, i; | |
1703 | ENTRY; | |
1704 | ||
1705 | ASSERT_LSM_MAGIC(lsm); | |
1706 | ||
1707 | if (!exp || !exp->exp_obd) | |
1708 | RETURN(-ENODEV); | |
1709 | ||
1710 | lov = &exp->exp_obd->u.lov; | |
1711 | for (i = 0; i < lsm->lsm_stripe_count; i++) { | |
1712 | struct lov_stripe_md submd; | |
1713 | struct lov_oinfo *loi = lsm->lsm_oinfo[i]; | |
1714 | ||
1715 | if (!lov->lov_tgts[loi->loi_ost_idx]) { | |
1716 | CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx); | |
1717 | continue; | |
1718 | } | |
1719 | submd.lsm_oi = loi->loi_oi; | |
1720 | submd.lsm_stripe_count = 0; | |
1721 | rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, | |
1722 | &submd, it, data); | |
1723 | if (rc != 0) | |
1724 | RETURN(rc); | |
1725 | } | |
1726 | RETURN(rc); | |
1727 | } | |
1728 | ||
1729 | static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm, | |
1730 | __u32 mode, struct lustre_handle *lockh) | |
1731 | { | |
1732 | struct lov_request_set *set; | |
1733 | struct obd_info oinfo; | |
1734 | struct lov_request *req; | |
1735 | struct list_head *pos; | |
1736 | struct lov_obd *lov; | |
1737 | struct lustre_handle *lov_lockhp; | |
1738 | int err = 0, rc = 0; | |
1739 | ENTRY; | |
1740 | ||
1741 | ASSERT_LSM_MAGIC(lsm); | |
1742 | ||
1743 | if (!exp || !exp->exp_obd) | |
1744 | RETURN(-ENODEV); | |
1745 | ||
1746 | LASSERT(lockh); | |
1747 | lov = &exp->exp_obd->u.lov; | |
1748 | rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set); | |
1749 | if (rc) | |
1750 | RETURN(rc); | |
1751 | ||
1752 | list_for_each(pos, &set->set_list) { | |
1753 | req = list_entry(pos, struct lov_request, rq_link); | |
1754 | lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe; | |
1755 | ||
1756 | rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1757 | req->rq_oi.oi_md, mode, lov_lockhp); | |
1758 | rc = lov_update_common_set(set, req, rc); | |
1759 | if (rc) { | |
1760 | CERROR("%s: cancel objid "DOSTID" subobj " | |
1761 | DOSTID" on OST idx %d: rc = %d\n", | |
1762 | exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi), | |
1763 | POSTID(&req->rq_oi.oi_md->lsm_oi), | |
1764 | req->rq_idx, rc); | |
1765 | err = rc; | |
1766 | } | |
1767 | ||
1768 | } | |
1769 | lov_fini_cancel_set(set); | |
1770 | RETURN(err); | |
1771 | } | |
1772 | ||
1773 | static int lov_cancel_unused(struct obd_export *exp, | |
1774 | struct lov_stripe_md *lsm, | |
1775 | ldlm_cancel_flags_t flags, void *opaque) | |
1776 | { | |
1777 | struct lov_obd *lov; | |
1778 | int rc = 0, i; | |
1779 | ENTRY; | |
1780 | ||
1781 | if (!exp || !exp->exp_obd) | |
1782 | RETURN(-ENODEV); | |
1783 | ||
1784 | lov = &exp->exp_obd->u.lov; | |
1785 | if (lsm == NULL) { | |
1786 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { | |
1787 | int err; | |
1788 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp) | |
1789 | continue; | |
1790 | ||
1791 | err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL, | |
1792 | flags, opaque); | |
1793 | if (!rc) | |
1794 | rc = err; | |
1795 | } | |
1796 | RETURN(rc); | |
1797 | } | |
1798 | ||
1799 | ASSERT_LSM_MAGIC(lsm); | |
1800 | ||
1801 | for (i = 0; i < lsm->lsm_stripe_count; i++) { | |
1802 | struct lov_stripe_md submd; | |
1803 | struct lov_oinfo *loi = lsm->lsm_oinfo[i]; | |
1804 | int idx = loi->loi_ost_idx; | |
1805 | int err; | |
1806 | ||
1807 | if (!lov->lov_tgts[idx]) { | |
1808 | CDEBUG(D_HA, "lov idx %d NULL\n", idx); | |
1809 | continue; | |
1810 | } | |
1811 | ||
1812 | if (!lov->lov_tgts[idx]->ltd_active) | |
1813 | CDEBUG(D_HA, "lov idx %d inactive\n", idx); | |
1814 | ||
1815 | submd.lsm_oi = loi->loi_oi; | |
1816 | submd.lsm_stripe_count = 0; | |
1817 | err = obd_cancel_unused(lov->lov_tgts[idx]->ltd_exp, | |
1818 | &submd, flags, opaque); | |
1819 | if (err && lov->lov_tgts[idx]->ltd_active) { | |
1820 | CERROR("%s: cancel unused objid "DOSTID | |
1821 | " subobj "DOSTID" on OST idx %d: rc = %d\n", | |
1822 | exp->exp_obd->obd_name, POSTID(&lsm->lsm_oi), | |
1823 | POSTID(&loi->loi_oi), idx, err); | |
1824 | if (!rc) | |
1825 | rc = err; | |
1826 | } | |
1827 | } | |
1828 | RETURN(rc); | |
1829 | } | |
1830 | ||
1831 | int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc) | |
1832 | { | |
1833 | struct lov_request_set *lovset = (struct lov_request_set *)data; | |
1834 | int err; | |
1835 | ENTRY; | |
1836 | ||
1837 | if (rc) | |
1838 | atomic_set(&lovset->set_completes, 0); | |
1839 | ||
1840 | err = lov_fini_statfs_set(lovset); | |
1841 | RETURN(rc ? rc : err); | |
1842 | } | |
1843 | ||
1844 | static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo, | |
1845 | __u64 max_age, struct ptlrpc_request_set *rqset) | |
1846 | { | |
1847 | struct obd_device *obd = class_exp2obd(exp); | |
1848 | struct lov_request_set *set; | |
1849 | struct lov_request *req; | |
1850 | struct list_head *pos; | |
1851 | struct lov_obd *lov; | |
1852 | int rc = 0; | |
1853 | ENTRY; | |
1854 | ||
1855 | LASSERT(oinfo != NULL); | |
1856 | LASSERT(oinfo->oi_osfs != NULL); | |
1857 | ||
1858 | lov = &obd->u.lov; | |
1859 | rc = lov_prep_statfs_set(obd, oinfo, &set); | |
1860 | if (rc) | |
1861 | RETURN(rc); | |
1862 | ||
1863 | list_for_each (pos, &set->set_list) { | |
1864 | req = list_entry(pos, struct lov_request, rq_link); | |
1865 | rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp, | |
1866 | &req->rq_oi, max_age, rqset); | |
1867 | if (rc) | |
1868 | break; | |
1869 | } | |
1870 | ||
1871 | if (rc || list_empty(&rqset->set_requests)) { | |
1872 | int err; | |
1873 | if (rc) | |
1874 | atomic_set(&set->set_completes, 0); | |
1875 | err = lov_fini_statfs_set(set); | |
1876 | RETURN(rc ? rc : err); | |
1877 | } | |
1878 | ||
1879 | LASSERT(rqset->set_interpret == NULL); | |
1880 | rqset->set_interpret = lov_statfs_interpret; | |
1881 | rqset->set_arg = (void *)set; | |
1882 | RETURN(0); | |
1883 | } | |
1884 | ||
1885 | static int lov_statfs(const struct lu_env *env, struct obd_export *exp, | |
1886 | struct obd_statfs *osfs, __u64 max_age, __u32 flags) | |
1887 | { | |
1888 | struct ptlrpc_request_set *set = NULL; | |
1889 | struct obd_info oinfo = { { { 0 } } }; | |
1890 | int rc = 0; | |
1891 | ENTRY; | |
1892 | ||
1893 | ||
1894 | /* for obdclass we forbid using obd_statfs_rqset, but prefer using async | |
1895 | * statfs requests */ | |
1896 | set = ptlrpc_prep_set(); | |
1897 | if (set == NULL) | |
1898 | RETURN(-ENOMEM); | |
1899 | ||
1900 | oinfo.oi_osfs = osfs; | |
1901 | oinfo.oi_flags = flags; | |
1902 | rc = lov_statfs_async(exp, &oinfo, max_age, set); | |
1903 | if (rc == 0) | |
1904 | rc = ptlrpc_set_wait(set); | |
1905 | ptlrpc_set_destroy(set); | |
1906 | ||
1907 | RETURN(rc); | |
1908 | } | |
1909 | ||
1910 | static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len, | |
1911 | void *karg, void *uarg) | |
1912 | { | |
1913 | struct obd_device *obddev = class_exp2obd(exp); | |
1914 | struct lov_obd *lov = &obddev->u.lov; | |
1915 | int i = 0, rc = 0, count = lov->desc.ld_tgt_count; | |
1916 | struct obd_uuid *uuidp; | |
1917 | ENTRY; | |
1918 | ||
1919 | switch (cmd) { | |
1920 | case IOC_OBD_STATFS: { | |
1921 | struct obd_ioctl_data *data = karg; | |
1922 | struct obd_device *osc_obd; | |
1923 | struct obd_statfs stat_buf = {0}; | |
1924 | __u32 index; | |
1925 | __u32 flags; | |
1926 | ||
1927 | memcpy(&index, data->ioc_inlbuf2, sizeof(__u32)); | |
1928 | if ((index >= count)) | |
1929 | RETURN(-ENODEV); | |
1930 | ||
1931 | if (!lov->lov_tgts[index]) | |
1932 | /* Try again with the next index */ | |
1933 | RETURN(-EAGAIN); | |
1934 | if (!lov->lov_tgts[index]->ltd_active) | |
1935 | RETURN(-ENODATA); | |
1936 | ||
1937 | osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp); | |
1938 | if (!osc_obd) | |
1939 | RETURN(-EINVAL); | |
1940 | ||
1941 | /* copy UUID */ | |
1942 | if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd), | |
1943 | min((int) data->ioc_plen2, | |
1944 | (int) sizeof(struct obd_uuid)))) | |
1945 | RETURN(-EFAULT); | |
1946 | ||
1947 | flags = uarg ? *(__u32*)uarg : 0; | |
1948 | /* got statfs data */ | |
1949 | rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf, | |
1950 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), | |
1951 | flags); | |
1952 | if (rc) | |
1953 | RETURN(rc); | |
1954 | if (copy_to_user(data->ioc_pbuf1, &stat_buf, | |
1955 | min((int) data->ioc_plen1, | |
1956 | (int) sizeof(stat_buf)))) | |
1957 | RETURN(-EFAULT); | |
1958 | break; | |
1959 | } | |
1960 | case OBD_IOC_LOV_GET_CONFIG: { | |
1961 | struct obd_ioctl_data *data; | |
1962 | struct lov_desc *desc; | |
1963 | char *buf = NULL; | |
1964 | __u32 *genp; | |
1965 | ||
1966 | len = 0; | |
1967 | if (obd_ioctl_getdata(&buf, &len, (void *)uarg)) | |
1968 | RETURN(-EINVAL); | |
1969 | ||
1970 | data = (struct obd_ioctl_data *)buf; | |
1971 | ||
1972 | if (sizeof(*desc) > data->ioc_inllen1) { | |
1973 | obd_ioctl_freedata(buf, len); | |
1974 | RETURN(-EINVAL); | |
1975 | } | |
1976 | ||
1977 | if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) { | |
1978 | obd_ioctl_freedata(buf, len); | |
1979 | RETURN(-EINVAL); | |
1980 | } | |
1981 | ||
1982 | if (sizeof(__u32) * count > data->ioc_inllen3) { | |
1983 | obd_ioctl_freedata(buf, len); | |
1984 | RETURN(-EINVAL); | |
1985 | } | |
1986 | ||
1987 | desc = (struct lov_desc *)data->ioc_inlbuf1; | |
1988 | memcpy(desc, &(lov->desc), sizeof(*desc)); | |
1989 | ||
1990 | uuidp = (struct obd_uuid *)data->ioc_inlbuf2; | |
1991 | genp = (__u32 *)data->ioc_inlbuf3; | |
1992 | /* the uuid will be empty for deleted OSTs */ | |
1993 | for (i = 0; i < count; i++, uuidp++, genp++) { | |
1994 | if (!lov->lov_tgts[i]) | |
1995 | continue; | |
1996 | *uuidp = lov->lov_tgts[i]->ltd_uuid; | |
1997 | *genp = lov->lov_tgts[i]->ltd_gen; | |
1998 | } | |
1999 | ||
2000 | if (copy_to_user((void *)uarg, buf, len)) | |
2001 | rc = -EFAULT; | |
2002 | obd_ioctl_freedata(buf, len); | |
2003 | break; | |
2004 | } | |
2005 | case LL_IOC_LOV_SETSTRIPE: | |
2006 | rc = lov_setstripe(exp, len, karg, uarg); | |
2007 | break; | |
2008 | case LL_IOC_LOV_GETSTRIPE: | |
2009 | rc = lov_getstripe(exp, karg, uarg); | |
2010 | break; | |
2011 | case LL_IOC_LOV_SETEA: | |
2012 | rc = lov_setea(exp, karg, uarg); | |
2013 | break; | |
2014 | case OBD_IOC_QUOTACTL: { | |
2015 | struct if_quotactl *qctl = karg; | |
2016 | struct lov_tgt_desc *tgt = NULL; | |
2017 | struct obd_quotactl *oqctl; | |
2018 | ||
2019 | if (qctl->qc_valid == QC_OSTIDX) { | |
2020 | if (qctl->qc_idx < 0 || count <= qctl->qc_idx) | |
2021 | RETURN(-EINVAL); | |
2022 | ||
2023 | tgt = lov->lov_tgts[qctl->qc_idx]; | |
2024 | if (!tgt || !tgt->ltd_exp) | |
2025 | RETURN(-EINVAL); | |
2026 | } else if (qctl->qc_valid == QC_UUID) { | |
2027 | for (i = 0; i < count; i++) { | |
2028 | tgt = lov->lov_tgts[i]; | |
2029 | if (!tgt || | |
2030 | !obd_uuid_equals(&tgt->ltd_uuid, | |
2031 | &qctl->obd_uuid)) | |
2032 | continue; | |
2033 | ||
2034 | if (tgt->ltd_exp == NULL) | |
2035 | RETURN(-EINVAL); | |
2036 | ||
2037 | break; | |
2038 | } | |
2039 | } else { | |
2040 | RETURN(-EINVAL); | |
2041 | } | |
2042 | ||
2043 | if (i >= count) | |
2044 | RETURN(-EAGAIN); | |
2045 | ||
2046 | LASSERT(tgt && tgt->ltd_exp); | |
2047 | OBD_ALLOC_PTR(oqctl); | |
2048 | if (!oqctl) | |
2049 | RETURN(-ENOMEM); | |
2050 | ||
2051 | QCTL_COPY(oqctl, qctl); | |
2052 | rc = obd_quotactl(tgt->ltd_exp, oqctl); | |
2053 | if (rc == 0) { | |
2054 | QCTL_COPY(qctl, oqctl); | |
2055 | qctl->qc_valid = QC_OSTIDX; | |
2056 | qctl->obd_uuid = tgt->ltd_uuid; | |
2057 | } | |
2058 | OBD_FREE_PTR(oqctl); | |
2059 | break; | |
2060 | } | |
2061 | default: { | |
2062 | int set = 0; | |
2063 | ||
2064 | if (count == 0) | |
2065 | RETURN(-ENOTTY); | |
2066 | ||
2067 | for (i = 0; i < count; i++) { | |
2068 | int err; | |
2069 | struct obd_device *osc_obd; | |
2070 | ||
2071 | /* OST was disconnected */ | |
2072 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp) | |
2073 | continue; | |
2074 | ||
2075 | /* ll_umount_begin() sets force flag but for lov, not | |
2076 | * osc. Let's pass it through */ | |
2077 | osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp); | |
2078 | osc_obd->obd_force = obddev->obd_force; | |
2079 | err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp, | |
2080 | len, karg, uarg); | |
2081 | if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) { | |
2082 | RETURN(err); | |
2083 | } else if (err) { | |
2084 | if (lov->lov_tgts[i]->ltd_active) { | |
2085 | CDEBUG(err == -ENOTTY ? | |
2086 | D_IOCTL : D_WARNING, | |
2087 | "iocontrol OSC %s on OST " | |
2088 | "idx %d cmd %x: err = %d\n", | |
2089 | lov_uuid2str(lov, i), | |
2090 | i, cmd, err); | |
2091 | if (!rc) | |
2092 | rc = err; | |
2093 | } | |
2094 | } else { | |
2095 | set = 1; | |
2096 | } | |
2097 | } | |
2098 | if (!set && !rc) | |
2099 | rc = -EIO; | |
2100 | } | |
2101 | } | |
2102 | ||
2103 | RETURN(rc); | |
2104 | } | |
2105 | ||
2106 | #define FIEMAP_BUFFER_SIZE 4096 | |
2107 | ||
2108 | /** | |
2109 | * Non-zero fe_logical indicates that this is a continuation FIEMAP | |
2110 | * call. The local end offset and the device are sent in the first | |
2111 | * fm_extent. This function calculates the stripe number from the index. | |
2112 | * This function returns a stripe_no on which mapping is to be restarted. | |
2113 | * | |
2114 | * This function returns fm_end_offset which is the in-OST offset at which | |
2115 | * mapping should be restarted. If fm_end_offset=0 is returned then caller | |
2116 | * will re-calculate proper offset in next stripe. | |
2117 | * Note that the first extent is passed to lov_get_info via the value field. | |
2118 | * | |
2119 | * \param fiemap fiemap request header | |
2120 | * \param lsm striping information for the file | |
2121 | * \param fm_start logical start of mapping | |
2122 | * \param fm_end logical end of mapping | |
2123 | * \param start_stripe starting stripe will be returned in this | |
2124 | */ | |
2125 | obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap, | |
2126 | struct lov_stripe_md *lsm, obd_size fm_start, | |
2127 | obd_size fm_end, int *start_stripe) | |
2128 | { | |
2129 | obd_size local_end = fiemap->fm_extents[0].fe_logical; | |
2130 | obd_off lun_start, lun_end; | |
2131 | obd_size fm_end_offset; | |
2132 | int stripe_no = -1, i; | |
2133 | ||
2134 | if (fiemap->fm_extent_count == 0 || | |
2135 | fiemap->fm_extents[0].fe_logical == 0) | |
2136 | return 0; | |
2137 | ||
2138 | /* Find out stripe_no from ost_index saved in the fe_device */ | |
2139 | for (i = 0; i < lsm->lsm_stripe_count; i++) { | |
2140 | if (lsm->lsm_oinfo[i]->loi_ost_idx == | |
2141 | fiemap->fm_extents[0].fe_device) { | |
2142 | stripe_no = i; | |
2143 | break; | |
2144 | } | |
2145 | } | |
2146 | if (stripe_no == -1) | |
2147 | return -EINVAL; | |
2148 | ||
2149 | /* If we have finished mapping on previous device, shift logical | |
2150 | * offset to start of next device */ | |
2151 | if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end, | |
2152 | &lun_start, &lun_end)) != 0 && | |
2153 | local_end < lun_end) { | |
2154 | fm_end_offset = local_end; | |
2155 | *start_stripe = stripe_no; | |
2156 | } else { | |
2157 | /* This is a special value to indicate that caller should | |
2158 | * calculate offset in next stripe. */ | |
2159 | fm_end_offset = 0; | |
2160 | *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count; | |
2161 | } | |
2162 | ||
2163 | return fm_end_offset; | |
2164 | } | |
2165 | ||
2166 | /** | |
2167 | * We calculate on which OST the mapping will end. If the length of mapping | |
2168 | * is greater than (stripe_size * stripe_count) then the last_stripe will | |
2169 | * will be one just before start_stripe. Else we check if the mapping | |
2170 | * intersects each OST and find last_stripe. | |
2171 | * This function returns the last_stripe and also sets the stripe_count | |
2172 | * over which the mapping is spread | |
2173 | * | |
2174 | * \param lsm striping information for the file | |
2175 | * \param fm_start logical start of mapping | |
2176 | * \param fm_end logical end of mapping | |
2177 | * \param start_stripe starting stripe of the mapping | |
2178 | * \param stripe_count the number of stripes across which to map is returned | |
2179 | * | |
2180 | * \retval last_stripe return the last stripe of the mapping | |
2181 | */ | |
2182 | int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start, | |
2183 | obd_size fm_end, int start_stripe, | |
2184 | int *stripe_count) | |
2185 | { | |
2186 | int last_stripe; | |
2187 | obd_off obd_start, obd_end; | |
2188 | int i, j; | |
2189 | ||
2190 | if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) { | |
2191 | last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 : | |
2192 | start_stripe - 1); | |
2193 | *stripe_count = lsm->lsm_stripe_count; | |
2194 | } else { | |
2195 | for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count; | |
2196 | i = (i + 1) % lsm->lsm_stripe_count, j++) { | |
2197 | if ((lov_stripe_intersects(lsm, i, fm_start, fm_end, | |
2198 | &obd_start, &obd_end)) == 0) | |
2199 | break; | |
2200 | } | |
2201 | *stripe_count = j; | |
2202 | last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count; | |
2203 | } | |
2204 | ||
2205 | return last_stripe; | |
2206 | } | |
2207 | ||
2208 | /** | |
2209 | * Set fe_device and copy extents from local buffer into main return buffer. | |
2210 | * | |
2211 | * \param fiemap fiemap request header | |
2212 | * \param lcl_fm_ext array of local fiemap extents to be copied | |
2213 | * \param ost_index OST index to be written into the fm_device field for each | |
2214 | extent | |
2215 | * \param ext_count number of extents to be copied | |
2216 | * \param current_extent where to start copying in main extent array | |
2217 | */ | |
2218 | void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap, | |
2219 | struct ll_fiemap_extent *lcl_fm_ext, | |
2220 | int ost_index, unsigned int ext_count, | |
2221 | int current_extent) | |
2222 | { | |
2223 | char *to; | |
2224 | int ext; | |
2225 | ||
2226 | for (ext = 0; ext < ext_count; ext++) { | |
2227 | lcl_fm_ext[ext].fe_device = ost_index; | |
2228 | lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET; | |
2229 | } | |
2230 | ||
2231 | /* Copy fm_extent's from fm_local to return buffer */ | |
2232 | to = (char *)fiemap + fiemap_count_to_size(current_extent); | |
2233 | memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent)); | |
2234 | } | |
2235 | ||
2236 | /** | |
2237 | * Break down the FIEMAP request and send appropriate calls to individual OSTs. | |
2238 | * This also handles the restarting of FIEMAP calls in case mapping overflows | |
2239 | * the available number of extents in single call. | |
2240 | */ | |
2241 | static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key, | |
2242 | __u32 *vallen, void *val, struct lov_stripe_md *lsm) | |
2243 | { | |
2244 | struct ll_fiemap_info_key *fm_key = key; | |
2245 | struct ll_user_fiemap *fiemap = val; | |
2246 | struct ll_user_fiemap *fm_local = NULL; | |
2247 | struct ll_fiemap_extent *lcl_fm_ext; | |
2248 | int count_local; | |
2249 | unsigned int get_num_extents = 0; | |
2250 | int ost_index = 0, actual_start_stripe, start_stripe; | |
2251 | obd_size fm_start, fm_end, fm_length, fm_end_offset; | |
2252 | obd_size curr_loc; | |
2253 | int current_extent = 0, rc = 0, i; | |
2254 | int ost_eof = 0; /* EOF for object */ | |
2255 | int ost_done = 0; /* done with required mapping for this OST? */ | |
2256 | int last_stripe; | |
2257 | int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count; | |
2258 | unsigned int buffer_size = FIEMAP_BUFFER_SIZE; | |
2259 | ||
2260 | if (lsm == NULL) | |
2261 | GOTO(out, rc = 0); | |
2262 | ||
2263 | if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size) | |
2264 | buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count); | |
2265 | ||
2266 | OBD_ALLOC_LARGE(fm_local, buffer_size); | |
2267 | if (fm_local == NULL) | |
2268 | GOTO(out, rc = -ENOMEM); | |
2269 | lcl_fm_ext = &fm_local->fm_extents[0]; | |
2270 | ||
2271 | count_local = fiemap_size_to_count(buffer_size); | |
2272 | ||
2273 | memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap)); | |
2274 | fm_start = fiemap->fm_start; | |
2275 | fm_length = fiemap->fm_length; | |
2276 | /* Calculate start stripe, last stripe and length of mapping */ | |
2277 | actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start); | |
2278 | fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size : | |
2279 | fm_start + fm_length - 1); | |
2280 | /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */ | |
2281 | if (fm_end > fm_key->oa.o_size) | |
2282 | fm_end = fm_key->oa.o_size; | |
2283 | ||
2284 | last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end, | |
2285 | actual_start_stripe, &stripe_count); | |
2286 | ||
2287 | fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, | |
2288 | fm_end, &start_stripe); | |
2289 | if (fm_end_offset == -EINVAL) | |
2290 | GOTO(out, rc = -EINVAL); | |
2291 | ||
2292 | if (fiemap->fm_extent_count == 0) { | |
2293 | get_num_extents = 1; | |
2294 | count_local = 0; | |
2295 | } | |
2296 | ||
2297 | /* Check each stripe */ | |
2298 | for (cur_stripe = start_stripe, i = 0; i < stripe_count; | |
2299 | i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) { | |
2300 | obd_size req_fm_len; /* Stores length of required mapping */ | |
2301 | obd_size len_mapped_single_call; | |
2302 | obd_off lun_start, lun_end, obd_object_end; | |
2303 | unsigned int ext_count; | |
2304 | ||
2305 | cur_stripe_wrap = cur_stripe; | |
2306 | ||
2307 | /* Find out range of mapping on this stripe */ | |
2308 | if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end, | |
2309 | &lun_start, &obd_object_end)) == 0) | |
2310 | continue; | |
2311 | ||
2312 | /* If this is a continuation FIEMAP call and we are on | |
2313 | * starting stripe then lun_start needs to be set to | |
2314 | * fm_end_offset */ | |
2315 | if (fm_end_offset != 0 && cur_stripe == start_stripe) | |
2316 | lun_start = fm_end_offset; | |
2317 | ||
2318 | if (fm_length != ~0ULL) { | |
2319 | /* Handle fm_start + fm_length overflow */ | |
2320 | if (fm_start + fm_length < fm_start) | |
2321 | fm_length = ~0ULL - fm_start; | |
2322 | lun_end = lov_size_to_stripe(lsm, fm_start + fm_length, | |
2323 | cur_stripe); | |
2324 | } else { | |
2325 | lun_end = ~0ULL; | |
2326 | } | |
2327 | ||
2328 | if (lun_start == lun_end) | |
2329 | continue; | |
2330 | ||
2331 | req_fm_len = obd_object_end - lun_start; | |
2332 | fm_local->fm_length = 0; | |
2333 | len_mapped_single_call = 0; | |
2334 | ||
2335 | /* If the output buffer is very large and the objects have many | |
2336 | * extents we may need to loop on a single OST repeatedly */ | |
2337 | ost_eof = 0; | |
2338 | ost_done = 0; | |
2339 | do { | |
2340 | if (get_num_extents == 0) { | |
2341 | /* Don't get too many extents. */ | |
2342 | if (current_extent + count_local > | |
2343 | fiemap->fm_extent_count) | |
2344 | count_local = fiemap->fm_extent_count - | |
2345 | current_extent; | |
2346 | } | |
2347 | ||
2348 | lun_start += len_mapped_single_call; | |
2349 | fm_local->fm_length = req_fm_len - len_mapped_single_call; | |
2350 | req_fm_len = fm_local->fm_length; | |
2351 | fm_local->fm_extent_count = count_local; | |
2352 | fm_local->fm_mapped_extents = 0; | |
2353 | fm_local->fm_flags = fiemap->fm_flags; | |
2354 | ||
2355 | fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi; | |
2356 | ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx; | |
2357 | ||
2358 | if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count) | |
2359 | GOTO(out, rc = -EINVAL); | |
2360 | ||
2361 | /* If OST is inactive, return extent with UNKNOWN flag */ | |
2362 | if (!lov->lov_tgts[ost_index]->ltd_active) { | |
2363 | fm_local->fm_flags |= FIEMAP_EXTENT_LAST; | |
2364 | fm_local->fm_mapped_extents = 1; | |
2365 | ||
2366 | lcl_fm_ext[0].fe_logical = lun_start; | |
2367 | lcl_fm_ext[0].fe_length = obd_object_end - | |
2368 | lun_start; | |
2369 | lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN; | |
2370 | ||
2371 | goto inactive_tgt; | |
2372 | } | |
2373 | ||
2374 | fm_local->fm_start = lun_start; | |
2375 | fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER; | |
2376 | memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local)); | |
2377 | *vallen=fiemap_count_to_size(fm_local->fm_extent_count); | |
2378 | rc = obd_get_info(NULL, | |
2379 | lov->lov_tgts[ost_index]->ltd_exp, | |
2380 | keylen, key, vallen, fm_local, lsm); | |
2381 | if (rc != 0) | |
2382 | GOTO(out, rc); | |
2383 | ||
2384 | inactive_tgt: | |
2385 | ext_count = fm_local->fm_mapped_extents; | |
2386 | if (ext_count == 0) { | |
2387 | ost_done = 1; | |
2388 | /* If last stripe has hole at the end, | |
2389 | * then we need to return */ | |
2390 | if (cur_stripe_wrap == last_stripe) { | |
2391 | fiemap->fm_mapped_extents = 0; | |
2392 | goto finish; | |
2393 | } | |
2394 | break; | |
2395 | } | |
2396 | ||
2397 | /* If we just need num of extents then go to next device */ | |
2398 | if (get_num_extents) { | |
2399 | current_extent += ext_count; | |
2400 | break; | |
2401 | } | |
2402 | ||
2403 | len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical - | |
2404 | lun_start + lcl_fm_ext[ext_count - 1].fe_length; | |
2405 | ||
2406 | /* Have we finished mapping on this device? */ | |
2407 | if (req_fm_len <= len_mapped_single_call) | |
2408 | ost_done = 1; | |
2409 | ||
2410 | /* Clear the EXTENT_LAST flag which can be present on | |
2411 | * last extent */ | |
2412 | if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST) | |
2413 | lcl_fm_ext[ext_count - 1].fe_flags &= | |
2414 | ~FIEMAP_EXTENT_LAST; | |
2415 | ||
2416 | curr_loc = lov_stripe_size(lsm, | |
2417 | lcl_fm_ext[ext_count - 1].fe_logical+ | |
2418 | lcl_fm_ext[ext_count - 1].fe_length, | |
2419 | cur_stripe); | |
2420 | if (curr_loc >= fm_key->oa.o_size) | |
2421 | ost_eof = 1; | |
2422 | ||
2423 | fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext, | |
2424 | ost_index, ext_count, | |
2425 | current_extent); | |
2426 | ||
2427 | current_extent += ext_count; | |
2428 | ||
2429 | /* Ran out of available extents? */ | |
2430 | if (current_extent >= fiemap->fm_extent_count) | |
2431 | goto finish; | |
2432 | } while (ost_done == 0 && ost_eof == 0); | |
2433 | ||
2434 | if (cur_stripe_wrap == last_stripe) | |
2435 | goto finish; | |
2436 | } | |
2437 | ||
2438 | finish: | |
2439 | /* Indicate that we are returning device offsets unless file just has | |
2440 | * single stripe */ | |
2441 | if (lsm->lsm_stripe_count > 1) | |
2442 | fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER; | |
2443 | ||
2444 | if (get_num_extents) | |
2445 | goto skip_last_device_calc; | |
2446 | ||
2447 | /* Check if we have reached the last stripe and whether mapping for that | |
2448 | * stripe is done. */ | |
2449 | if (cur_stripe_wrap == last_stripe) { | |
2450 | if (ost_done || ost_eof) | |
2451 | fiemap->fm_extents[current_extent - 1].fe_flags |= | |
2452 | FIEMAP_EXTENT_LAST; | |
2453 | } | |
2454 | ||
2455 | skip_last_device_calc: | |
2456 | fiemap->fm_mapped_extents = current_extent; | |
2457 | ||
2458 | out: | |
2459 | OBD_FREE_LARGE(fm_local, buffer_size); | |
2460 | return rc; | |
2461 | } | |
2462 | ||
2463 | static int lov_get_info(const struct lu_env *env, struct obd_export *exp, | |
2464 | __u32 keylen, void *key, __u32 *vallen, void *val, | |
2465 | struct lov_stripe_md *lsm) | |
2466 | { | |
2467 | struct obd_device *obddev = class_exp2obd(exp); | |
2468 | struct lov_obd *lov = &obddev->u.lov; | |
2469 | int i, rc; | |
2470 | ENTRY; | |
2471 | ||
2472 | if (!vallen || !val) | |
2473 | RETURN(-EFAULT); | |
2474 | ||
2475 | obd_getref(obddev); | |
2476 | ||
2477 | if (KEY_IS(KEY_LOCK_TO_STRIPE)) { | |
2478 | struct { | |
2479 | char name[16]; | |
2480 | struct ldlm_lock *lock; | |
2481 | } *data = key; | |
2482 | struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name; | |
2483 | struct lov_oinfo *loi; | |
2484 | __u32 *stripe = val; | |
2485 | ||
2486 | if (*vallen < sizeof(*stripe)) | |
2487 | GOTO(out, rc = -EFAULT); | |
2488 | *vallen = sizeof(*stripe); | |
2489 | ||
2490 | /* XXX This is another one of those bits that will need to | |
2491 | * change if we ever actually support nested LOVs. It uses | |
2492 | * the lock's export to find out which stripe it is. */ | |
2493 | /* XXX - it's assumed all the locks for deleted OSTs have | |
2494 | * been cancelled. Also, the export for deleted OSTs will | |
2495 | * be NULL and won't match the lock's export. */ | |
2496 | for (i = 0; i < lsm->lsm_stripe_count; i++) { | |
2497 | loi = lsm->lsm_oinfo[i]; | |
2498 | if (!lov->lov_tgts[loi->loi_ost_idx]) | |
2499 | continue; | |
2500 | if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp == | |
2501 | data->lock->l_conn_export && | |
2502 | ostid_res_name_eq(&loi->loi_oi, res_id)) { | |
2503 | *stripe = i; | |
2504 | GOTO(out, rc = 0); | |
2505 | } | |
2506 | } | |
2507 | LDLM_ERROR(data->lock, "lock on inode without such object"); | |
2508 | dump_lsm(D_ERROR, lsm); | |
2509 | GOTO(out, rc = -ENXIO); | |
2510 | } else if (KEY_IS(KEY_LAST_ID)) { | |
2511 | struct obd_id_info *info = val; | |
2512 | __u32 size = sizeof(obd_id); | |
2513 | struct lov_tgt_desc *tgt; | |
2514 | ||
2515 | LASSERT(*vallen == sizeof(struct obd_id_info)); | |
2516 | tgt = lov->lov_tgts[info->idx]; | |
2517 | ||
2518 | if (!tgt || !tgt->ltd_active) | |
2519 | GOTO(out, rc = -ESRCH); | |
2520 | ||
2521 | rc = obd_get_info(env, tgt->ltd_exp, keylen, key, | |
2522 | &size, info->data, NULL); | |
2523 | GOTO(out, rc = 0); | |
2524 | } else if (KEY_IS(KEY_LOVDESC)) { | |
2525 | struct lov_desc *desc_ret = val; | |
2526 | *desc_ret = lov->desc; | |
2527 | ||
2528 | GOTO(out, rc = 0); | |
2529 | } else if (KEY_IS(KEY_FIEMAP)) { | |
2530 | rc = lov_fiemap(lov, keylen, key, vallen, val, lsm); | |
2531 | GOTO(out, rc); | |
2532 | } else if (KEY_IS(KEY_CONNECT_FLAG)) { | |
2533 | struct lov_tgt_desc *tgt; | |
2534 | __u64 ost_idx = *((__u64*)val); | |
2535 | ||
2536 | LASSERT(*vallen == sizeof(__u64)); | |
2537 | LASSERT(ost_idx < lov->desc.ld_tgt_count); | |
2538 | tgt = lov->lov_tgts[ost_idx]; | |
2539 | ||
2540 | if (!tgt || !tgt->ltd_exp) | |
2541 | GOTO(out, rc = -ESRCH); | |
2542 | ||
2543 | *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp); | |
2544 | GOTO(out, rc = 0); | |
2545 | } else if (KEY_IS(KEY_TGT_COUNT)) { | |
2546 | *((int *)val) = lov->desc.ld_tgt_count; | |
2547 | GOTO(out, rc = 0); | |
2548 | } | |
2549 | ||
2550 | rc = -EINVAL; | |
2551 | ||
2552 | out: | |
2553 | obd_putref(obddev); | |
2554 | RETURN(rc); | |
2555 | } | |
2556 | ||
2557 | static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp, | |
2558 | obd_count keylen, void *key, obd_count vallen, | |
2559 | void *val, struct ptlrpc_request_set *set) | |
2560 | { | |
2561 | struct obd_device *obddev = class_exp2obd(exp); | |
2562 | struct lov_obd *lov = &obddev->u.lov; | |
2563 | obd_count count; | |
2564 | int i, rc = 0, err; | |
2565 | struct lov_tgt_desc *tgt; | |
2566 | unsigned incr, check_uuid, | |
2567 | do_inactive, no_set; | |
2568 | unsigned next_id = 0, mds_con = 0, capa = 0; | |
2569 | ENTRY; | |
2570 | ||
2571 | incr = check_uuid = do_inactive = no_set = 0; | |
2572 | if (set == NULL) { | |
2573 | no_set = 1; | |
2574 | set = ptlrpc_prep_set(); | |
2575 | if (!set) | |
2576 | RETURN(-ENOMEM); | |
2577 | } | |
2578 | ||
2579 | obd_getref(obddev); | |
2580 | count = lov->desc.ld_tgt_count; | |
2581 | ||
2582 | if (KEY_IS(KEY_NEXT_ID)) { | |
2583 | count = vallen / sizeof(struct obd_id_info); | |
2584 | vallen = sizeof(obd_id); | |
2585 | incr = sizeof(struct obd_id_info); | |
2586 | do_inactive = 1; | |
2587 | next_id = 1; | |
2588 | } else if (KEY_IS(KEY_CHECKSUM)) { | |
2589 | do_inactive = 1; | |
2590 | } else if (KEY_IS(KEY_EVICT_BY_NID)) { | |
2591 | /* use defaults: do_inactive = incr = 0; */ | |
2592 | } else if (KEY_IS(KEY_MDS_CONN)) { | |
2593 | mds_con = 1; | |
2594 | } else if (KEY_IS(KEY_CAPA_KEY)) { | |
2595 | capa = 1; | |
2596 | } else if (KEY_IS(KEY_CACHE_SET)) { | |
2597 | LASSERT(lov->lov_cache == NULL); | |
2598 | lov->lov_cache = val; | |
2599 | do_inactive = 1; | |
2600 | } | |
2601 | ||
2602 | for (i = 0; i < count; i++, val = (char *)val + incr) { | |
2603 | if (next_id) { | |
2604 | tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx]; | |
2605 | } else { | |
2606 | tgt = lov->lov_tgts[i]; | |
2607 | } | |
2608 | /* OST was disconnected */ | |
2609 | if (!tgt || !tgt->ltd_exp) | |
2610 | continue; | |
2611 | ||
2612 | /* OST is inactive and we don't want inactive OSCs */ | |
2613 | if (!tgt->ltd_active && !do_inactive) | |
2614 | continue; | |
2615 | ||
2616 | if (mds_con) { | |
2617 | struct mds_group_info *mgi; | |
2618 | ||
2619 | LASSERT(vallen == sizeof(*mgi)); | |
2620 | mgi = (struct mds_group_info *)val; | |
2621 | ||
2622 | /* Only want a specific OSC */ | |
2623 | if (mgi->uuid && !obd_uuid_equals(mgi->uuid, | |
2624 | &tgt->ltd_uuid)) | |
2625 | continue; | |
2626 | ||
2627 | err = obd_set_info_async(env, tgt->ltd_exp, | |
2628 | keylen, key, sizeof(int), | |
2629 | &mgi->group, set); | |
2630 | } else if (next_id) { | |
2631 | err = obd_set_info_async(env, tgt->ltd_exp, | |
2632 | keylen, key, vallen, | |
2633 | ((struct obd_id_info*)val)->data, set); | |
2634 | } else if (capa) { | |
2635 | struct mds_capa_info *info = (struct mds_capa_info*)val; | |
2636 | ||
2637 | LASSERT(vallen == sizeof(*info)); | |
2638 | ||
2639 | /* Only want a specific OSC */ | |
2640 | if (info->uuid && | |
2641 | !obd_uuid_equals(info->uuid, &tgt->ltd_uuid)) | |
2642 | continue; | |
2643 | ||
2644 | err = obd_set_info_async(env, tgt->ltd_exp, keylen, | |
2645 | key, sizeof(*info->capa), | |
2646 | info->capa, set); | |
2647 | } else { | |
2648 | /* Only want a specific OSC */ | |
2649 | if (check_uuid && | |
2650 | !obd_uuid_equals(val, &tgt->ltd_uuid)) | |
2651 | continue; | |
2652 | ||
2653 | err = obd_set_info_async(env, tgt->ltd_exp, | |
2654 | keylen, key, vallen, val, set); | |
2655 | } | |
2656 | ||
2657 | if (!rc) | |
2658 | rc = err; | |
2659 | } | |
2660 | ||
2661 | obd_putref(obddev); | |
2662 | if (no_set) { | |
2663 | err = ptlrpc_set_wait(set); | |
2664 | if (!rc) | |
2665 | rc = err; | |
2666 | ptlrpc_set_destroy(set); | |
2667 | } | |
2668 | RETURN(rc); | |
2669 | } | |
2670 | ||
2671 | static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm, | |
2672 | int cmd, __u64 *offset) | |
2673 | { | |
2674 | __u32 ssize = lsm->lsm_stripe_size; | |
2675 | __u64 start; | |
2676 | ||
2677 | start = *offset; | |
2678 | lov_do_div64(start, ssize); | |
2679 | start = start * ssize; | |
2680 | ||
2681 | CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64 | |
2682 | ", end "LPU64"\n", *offset, ssize, start, | |
2683 | start + ssize - 1); | |
2684 | if (cmd == OBD_CALC_STRIPE_END) { | |
2685 | *offset = start + ssize - 1; | |
2686 | } else if (cmd == OBD_CALC_STRIPE_START) { | |
2687 | *offset = start; | |
2688 | } else { | |
2689 | LBUG(); | |
2690 | } | |
2691 | ||
2692 | RETURN(0); | |
2693 | } | |
2694 | ||
2695 | void lov_stripe_lock(struct lov_stripe_md *md) | |
2696 | { | |
2697 | LASSERT(md->lsm_lock_owner != current_pid()); | |
2698 | spin_lock(&md->lsm_lock); | |
2699 | LASSERT(md->lsm_lock_owner == 0); | |
2700 | md->lsm_lock_owner = current_pid(); | |
2701 | } | |
2702 | EXPORT_SYMBOL(lov_stripe_lock); | |
2703 | ||
2704 | void lov_stripe_unlock(struct lov_stripe_md *md) | |
2705 | { | |
2706 | LASSERT(md->lsm_lock_owner == current_pid()); | |
2707 | md->lsm_lock_owner = 0; | |
2708 | spin_unlock(&md->lsm_lock); | |
2709 | } | |
2710 | EXPORT_SYMBOL(lov_stripe_unlock); | |
2711 | ||
2712 | static int lov_quotactl(struct obd_device *obd, struct obd_export *exp, | |
2713 | struct obd_quotactl *oqctl) | |
2714 | { | |
2715 | struct lov_obd *lov = &obd->u.lov; | |
2716 | struct lov_tgt_desc *tgt; | |
2717 | __u64 curspace = 0; | |
2718 | __u64 bhardlimit = 0; | |
2719 | int i, rc = 0; | |
2720 | ENTRY; | |
2721 | ||
2722 | if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON && | |
2723 | oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF && | |
2724 | oqctl->qc_cmd != Q_GETOQUOTA && | |
2725 | oqctl->qc_cmd != Q_INITQUOTA && | |
2726 | oqctl->qc_cmd != LUSTRE_Q_SETQUOTA && | |
2727 | oqctl->qc_cmd != Q_FINVALIDATE) { | |
2728 | CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd); | |
2729 | RETURN(-EFAULT); | |
2730 | } | |
2731 | ||
2732 | /* for lov tgt */ | |
2733 | obd_getref(obd); | |
2734 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { | |
2735 | int err; | |
2736 | ||
2737 | tgt = lov->lov_tgts[i]; | |
2738 | ||
2739 | if (!tgt) | |
2740 | continue; | |
2741 | ||
2742 | if (!tgt->ltd_active || tgt->ltd_reap) { | |
2743 | if (oqctl->qc_cmd == Q_GETOQUOTA && | |
2744 | lov->lov_tgts[i]->ltd_activate) { | |
2745 | rc = -EREMOTEIO; | |
2746 | CERROR("ost %d is inactive\n", i); | |
2747 | } else { | |
2748 | CDEBUG(D_HA, "ost %d is inactive\n", i); | |
2749 | } | |
2750 | continue; | |
2751 | } | |
2752 | ||
2753 | err = obd_quotactl(tgt->ltd_exp, oqctl); | |
2754 | if (err) { | |
2755 | if (tgt->ltd_active && !rc) | |
2756 | rc = err; | |
2757 | continue; | |
2758 | } | |
2759 | ||
2760 | if (oqctl->qc_cmd == Q_GETOQUOTA) { | |
2761 | curspace += oqctl->qc_dqblk.dqb_curspace; | |
2762 | bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit; | |
2763 | } | |
2764 | } | |
2765 | obd_putref(obd); | |
2766 | ||
2767 | if (oqctl->qc_cmd == Q_GETOQUOTA) { | |
2768 | oqctl->qc_dqblk.dqb_curspace = curspace; | |
2769 | oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit; | |
2770 | } | |
2771 | RETURN(rc); | |
2772 | } | |
2773 | ||
2774 | static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp, | |
2775 | struct obd_quotactl *oqctl) | |
2776 | { | |
2777 | struct lov_obd *lov = &obd->u.lov; | |
2778 | int i, rc = 0; | |
2779 | ENTRY; | |
2780 | ||
2781 | obd_getref(obd); | |
2782 | ||
2783 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { | |
2784 | if (!lov->lov_tgts[i]) | |
2785 | continue; | |
2786 | ||
2787 | /* Skip quota check on the administratively disabled OSTs. */ | |
2788 | if (!lov->lov_tgts[i]->ltd_activate) { | |
2789 | CWARN("lov idx %d was administratively disabled, " | |
2790 | "skip quotacheck on it.\n", i); | |
2791 | continue; | |
2792 | } | |
2793 | ||
2794 | if (!lov->lov_tgts[i]->ltd_active) { | |
2795 | CERROR("lov idx %d inactive\n", i); | |
2796 | rc = -EIO; | |
2797 | goto out; | |
2798 | } | |
2799 | } | |
2800 | ||
2801 | for (i = 0; i < lov->desc.ld_tgt_count; i++) { | |
2802 | int err; | |
2803 | ||
2804 | if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate) | |
2805 | continue; | |
2806 | ||
2807 | err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl); | |
2808 | if (err && !rc) | |
2809 | rc = err; | |
2810 | } | |
2811 | ||
2812 | out: | |
2813 | obd_putref(obd); | |
2814 | ||
2815 | RETURN(rc); | |
2816 | } | |
2817 | ||
2818 | struct obd_ops lov_obd_ops = { | |
2819 | .o_owner = THIS_MODULE, | |
2820 | .o_setup = lov_setup, | |
2821 | .o_precleanup = lov_precleanup, | |
2822 | .o_cleanup = lov_cleanup, | |
2823 | //.o_process_config = lov_process_config, | |
2824 | .o_connect = lov_connect, | |
2825 | .o_disconnect = lov_disconnect, | |
2826 | .o_statfs = lov_statfs, | |
2827 | .o_statfs_async = lov_statfs_async, | |
2828 | .o_packmd = lov_packmd, | |
2829 | .o_unpackmd = lov_unpackmd, | |
2830 | .o_create = lov_create, | |
2831 | .o_destroy = lov_destroy, | |
2832 | .o_getattr = lov_getattr, | |
2833 | .o_getattr_async = lov_getattr_async, | |
2834 | .o_setattr = lov_setattr, | |
2835 | .o_setattr_async = lov_setattr_async, | |
2836 | .o_brw = lov_brw, | |
2837 | .o_merge_lvb = lov_merge_lvb, | |
2838 | .o_adjust_kms = lov_adjust_kms, | |
2839 | .o_punch = lov_punch, | |
2840 | .o_sync = lov_sync, | |
2841 | .o_enqueue = lov_enqueue, | |
2842 | .o_change_cbdata = lov_change_cbdata, | |
2843 | .o_find_cbdata = lov_find_cbdata, | |
2844 | .o_cancel = lov_cancel, | |
2845 | .o_cancel_unused = lov_cancel_unused, | |
2846 | .o_iocontrol = lov_iocontrol, | |
2847 | .o_get_info = lov_get_info, | |
2848 | .o_set_info_async = lov_set_info_async, | |
2849 | .o_extent_calc = lov_extent_calc, | |
2850 | .o_llog_init = lov_llog_init, | |
2851 | .o_llog_finish = lov_llog_finish, | |
2852 | .o_notify = lov_notify, | |
2853 | .o_pool_new = lov_pool_new, | |
2854 | .o_pool_rem = lov_pool_remove, | |
2855 | .o_pool_add = lov_pool_add, | |
2856 | .o_pool_del = lov_pool_del, | |
2857 | .o_getref = lov_getref, | |
2858 | .o_putref = lov_putref, | |
2859 | .o_quotactl = lov_quotactl, | |
2860 | .o_quotacheck = lov_quotacheck, | |
2861 | }; | |
2862 | ||
2863 | struct kmem_cache *lov_oinfo_slab; | |
2864 | ||
2865 | extern struct lu_kmem_descr lov_caches[]; | |
2866 | ||
2867 | int __init lov_init(void) | |
2868 | { | |
2869 | struct lprocfs_static_vars lvars = { 0 }; | |
2870 | int rc; | |
2871 | ENTRY; | |
2872 | ||
2873 | /* print an address of _any_ initialized kernel symbol from this | |
2874 | * module, to allow debugging with gdb that doesn't support data | |
2875 | * symbols from modules.*/ | |
2876 | CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches); | |
2877 | ||
2878 | rc = lu_kmem_init(lov_caches); | |
2879 | if (rc) | |
2880 | return rc; | |
2881 | ||
2882 | lov_oinfo_slab = kmem_cache_create("lov_oinfo", | |
2883 | sizeof(struct lov_oinfo), | |
2884 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
2885 | if (lov_oinfo_slab == NULL) { | |
2886 | lu_kmem_fini(lov_caches); | |
2887 | return -ENOMEM; | |
2888 | } | |
2889 | lprocfs_lov_init_vars(&lvars); | |
2890 | ||
2891 | rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars, | |
2892 | LUSTRE_LOV_NAME, &lov_device_type); | |
2893 | ||
2894 | if (rc) { | |
2895 | kmem_cache_destroy(lov_oinfo_slab); | |
2896 | lu_kmem_fini(lov_caches); | |
2897 | } | |
2898 | ||
2899 | RETURN(rc); | |
2900 | } | |
2901 | ||
2902 | static void /*__exit*/ lov_exit(void) | |
2903 | { | |
2904 | class_unregister_type(LUSTRE_LOV_NAME); | |
2905 | kmem_cache_destroy(lov_oinfo_slab); | |
2906 | ||
2907 | lu_kmem_fini(lov_caches); | |
2908 | } | |
2909 | ||
2910 | MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>"); | |
2911 | MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver"); | |
2912 | MODULE_LICENSE("GPL"); | |
2913 | ||
2914 | cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit); |