4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2012, Intel, Inc.
30 * This file is part of Lustre, http://www.lustre.org/
31 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/lod/lod_dev.c
35 * Lustre Logical Object Device
37 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
38 * Author: Mikhail Pershin <mike.pershin@intel.com>
42 # define EXPORT_SYMTAB
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <obd_class.h>
47 #include <lustre_param.h>
49 #include "lod_internal.h"
51 /* Slab for OSD object allocation */
52 cfs_mem_cache_t *lod_object_kmem;
54 static struct lu_kmem_descr lod_caches[] = {
56 .ckd_cache = &lod_object_kmem,
57 .ckd_name = "lod_obj",
58 .ckd_size = sizeof(struct lod_object)
65 static int lod_process_config(const struct lu_env *env,
66 struct lu_device *dev,
67 struct lustre_cfg *lcfg)
69 struct lod_device *lod = lu2lod_dev(dev);
70 struct lu_device *next = &lod->lod_child->dd_lu_dev;
75 switch(lcfg->lcfg_command) {
77 case LCFG_LOV_DEL_OBD:
78 case LCFG_LOV_ADD_INA:
79 case LCFG_LOV_ADD_OBD: {
82 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1 */
83 arg1 = lustre_cfg_string(lcfg, 1);
85 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
86 GOTO(out, rc = -EINVAL);
87 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
88 GOTO(out, rc = -EINVAL);
95 struct lprocfs_static_vars v = { 0 };
96 struct obd_device *obd = lod2obd(lod);
98 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
105 lu_dev_del_linkage(dev->ld_site, dev);
107 lod_foreach_ost(lod, i) {
108 struct lod_ost_desc *ost;
109 ost = OST_TGT(lod, i);
110 LASSERT(ost && ost->ltd_ost);
111 next = &ost->ltd_ost->dd_lu_dev;
112 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
114 CERROR("%s: can't process %u: %d\n",
115 lod2obd(lod)->obd_name,
116 lcfg->lcfg_command, rc);
121 * do cleanup on underlying storage only when
122 * all OSPs are cleaned up, as they use that OSD as well
124 next = &lod->lod_child->dd_lu_dev;
125 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
127 CERROR("%s: can't process %u: %d\n",
128 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
132 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
142 static int lod_recovery_complete(const struct lu_env *env,
143 struct lu_device *dev)
145 struct lod_device *lod = lu2lod_dev(dev);
146 struct lu_device *next = &lod->lod_child->dd_lu_dev;
147 struct lod_ost_desc *ost;
151 LASSERT(lod->lod_recovery_completed == 0);
152 lod->lod_recovery_completed = 1;
154 rc = next->ld_ops->ldo_recovery_complete(env, next);
157 lod_foreach_ost(lod, i) {
158 ost = OST_TGT(lod, i);
159 LASSERT(ost && ost->ltd_ost);
160 next = &ost->ltd_ost->dd_lu_dev;
161 rc = next->ld_ops->ldo_recovery_complete(env, next);
163 CERROR("%s: can't complete recovery on #%d: %d\n",
164 lod2obd(lod)->obd_name, i, rc);
171 const struct lu_device_operations lod_lu_ops = {
172 .ldo_process_config = lod_process_config,
173 .ldo_recovery_complete = lod_recovery_complete,
176 static int lod_root_get(const struct lu_env *env,
177 struct dt_device *dev, struct lu_fid *f)
179 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
182 static int lod_statfs(const struct lu_env *env,
183 struct dt_device *dev, struct obd_statfs *sfs)
185 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
188 static struct thandle *lod_trans_create(const struct lu_env *env,
189 struct dt_device *dev)
191 return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
194 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
197 return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
200 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
202 /* XXX: we don't know next device, will be fixed with DNE */
203 return dt_trans_stop(env, th->th_dev, th);
206 static void lod_conf_get(const struct lu_env *env,
207 const struct dt_device *dev,
208 struct dt_device_param *param)
210 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
213 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
215 struct lod_device *lod = dt2lod_dev(dev);
216 struct lod_ost_desc *ost;
221 lod_foreach_ost(lod, i) {
222 ost = OST_TGT(lod, i);
223 LASSERT(ost && ost->ltd_ost);
224 rc = dt_sync(env, ost->ltd_ost);
226 CERROR("%s: can't sync %u: %d\n",
227 lod2obd(lod)->obd_name, i, rc);
233 rc = dt_sync(env, lod->lod_child);
238 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
240 return dt_ro(env, dt2lod_dev(dev)->lod_child);
243 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
245 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
248 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
249 int mode, unsigned long timeout,
250 __u32 alg, struct lustre_capa_key *keys)
252 struct dt_device *next = dt2lod_dev(dev)->lod_child;
253 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
256 static const struct dt_device_operations lod_dt_ops = {
257 .dt_root_get = lod_root_get,
258 .dt_statfs = lod_statfs,
259 .dt_trans_create = lod_trans_create,
260 .dt_trans_start = lod_trans_start,
261 .dt_trans_stop = lod_trans_stop,
262 .dt_conf_get = lod_conf_get,
265 .dt_commit_async = lod_commit_async,
266 .dt_init_capa_ctxt = lod_init_capa_ctxt,
269 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
270 struct lustre_cfg *cfg)
272 struct obd_connect_data *data = NULL;
273 struct obd_device *obd;
274 char *nextdev = NULL, *p, *s;
279 LASSERT(lod->lod_child_exp == NULL);
281 /* compatibility hack: we still use old config logs
282 * which specify LOV, but we need to learn underlying
283 * OSD device, which is supposed to be:
284 * <fsname>-MDTxxxx-osd
286 * 2.x MGS generates lines like the following:
287 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
288 * 1.8 MGS generates lines like the following:
289 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
291 * we use "-MDT" to differentiate 2.x from 1.8 */
293 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
295 OBD_ALLOC(nextdev, len);
297 GOTO(out, rc = -ENOMEM);
300 s = strstr(nextdev, "-mdtlov");
301 if (unlikely(s == NULL)) {
302 CERROR("unable to parse device name %s\n",
303 lustre_cfg_string(cfg, 0));
304 GOTO(out, rc = -EINVAL);
307 if (strstr(nextdev, "-MDT")) {
312 strcpy(s, "-MDT0000-osd");
315 CERROR("unable to parse device name %s\n",
316 lustre_cfg_string(cfg, 0));
317 GOTO(out, rc = -EINVAL);
322 GOTO(out, rc = -ENOMEM);
324 obd = class_name2obd(nextdev);
326 CERROR("can not locate next device: %s\n", nextdev);
327 GOTO(out, rc = -ENOTCONN);
330 data->ocd_connect_flags = OBD_CONNECT_VERSION;
331 data->ocd_version = LUSTRE_VERSION_CODE;
333 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
336 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
340 lod->lod_dt_dev.dd_lu_dev.ld_site =
341 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
342 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
343 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
349 OBD_FREE(nextdev, len);
353 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
354 struct lu_device_type *ldt, struct lustre_cfg *cfg)
356 struct dt_device_param ddp;
357 struct proc_dir_entry *lov_proc_dir;
358 struct obd_device *obd;
362 obd = class_name2obd(lustre_cfg_string(cfg, 0));
364 CERROR("Cannot find obd with name %s\n",
365 lustre_cfg_string(cfg, 0));
369 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
370 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
372 rc = lod_connect_to_osd(env, lod, cfg);
376 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
377 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
379 /* for compatibility we link old procfs's OSC entries to osp ones */
380 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
382 cfs_proc_dir_entry_t *symlink = NULL;
384 OBD_ALLOC(name, strlen(obd->obd_name) + 1);
386 strcpy(name, obd->obd_name);
387 if (strstr(name, "lov"))
388 symlink = lprocfs_add_symlink(name,
392 OBD_FREE(name, strlen(obd->obd_name) + 1);
393 lod->lod_symlink = symlink;
397 cfs_mutex_init(&lod->lod_mutex);
398 cfs_init_rwsem(&lod->lod_rw_sem);
399 cfs_spin_lock_init(&lod->lod_desc_lock);
403 obd_disconnect(lod->lod_child_exp);
407 static struct lu_device *lod_device_free(const struct lu_env *env,
408 struct lu_device *lu)
410 struct lod_device *lod = lu2lod_dev(lu);
411 struct lu_device *next = &lod->lod_child->dd_lu_dev;
414 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
415 dt_device_fini(&lod->lod_dt_dev);
420 static struct lu_device *lod_device_alloc(const struct lu_env *env,
421 struct lu_device_type *type,
422 struct lustre_cfg *lcfg)
424 struct lod_device *lod;
425 struct lu_device *lu_dev;
429 lu_dev = ERR_PTR(-ENOMEM);
433 lu_dev = lod2lu_dev(lod);
434 dt_device_init(&lod->lod_dt_dev, type);
435 rc = lod_init0(env, lod, type, lcfg);
437 lod_device_free(env, lu_dev);
438 lu_dev = ERR_PTR(rc);
445 static struct lu_device *lod_device_fini(const struct lu_env *env,
448 struct lod_device *lod = lu2lod_dev(d);
452 if (lod->lod_symlink)
453 lprocfs_remove(&lod->lod_symlink);
455 rc = obd_disconnect(lod->lod_child_exp);
457 CERROR("error in disconnect from storage: %d\n", rc);
463 * we use exports to track all LOD users
465 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
466 struct obd_device *obd, struct obd_uuid *cluuid,
467 struct obd_connect_data *data, void *localdata)
469 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
470 struct lustre_handle conn;
474 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
476 rc = class_connect(&conn, obd, cluuid);
480 *exp = class_conn2export(&conn);
482 cfs_mutex_lock(&lod->lod_mutex);
484 /* at the moment we expect the only user */
485 LASSERT(lod->lod_connects == 1);
486 cfs_mutex_unlock(&lod->lod_mutex);
492 * once last export (we don't count self-export) disappeared
493 * lod can be released
495 static int lod_obd_disconnect(struct obd_export *exp)
497 struct obd_device *obd = exp->exp_obd;
498 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
502 /* Only disconnect the underlying layers on the final disconnect. */
503 cfs_mutex_lock(&lod->lod_mutex);
505 if (lod->lod_connects != 0) {
506 /* why should there be more than 1 connect? */
507 cfs_mutex_unlock(&lod->lod_mutex);
508 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
512 cfs_mutex_unlock(&lod->lod_mutex);
514 /* the last user of lod has gone, let's release the device */
518 rc = class_disconnect(exp); /* bz 9811 */
520 if (rc == 0 && release)
521 class_manual_cleanup(obd);
525 LU_KEY_INIT(lod, struct lod_thread_info);
527 static void lod_key_fini(const struct lu_context *ctx,
528 struct lu_context_key *key, void *data)
530 struct lod_thread_info *info = data;
531 /* allocated in lod_get_lov_ea
532 * XXX: this is overload, a tread may have such store but used only
533 * once. Probably better would be pool of such stores per LOD.
535 if (info->lti_ea_store) {
536 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
537 info->lti_ea_store = NULL;
538 info->lti_ea_store_size = 0;
543 /* context key: lod_thread_key */
544 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
546 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
548 static struct lu_device_type_operations lod_device_type_ops = {
549 .ldto_init = lod_type_init,
550 .ldto_fini = lod_type_fini,
552 .ldto_start = lod_type_start,
553 .ldto_stop = lod_type_stop,
555 .ldto_device_alloc = lod_device_alloc,
556 .ldto_device_free = lod_device_free,
558 .ldto_device_fini = lod_device_fini
561 static struct lu_device_type lod_device_type = {
562 .ldt_tags = LU_DEVICE_DT,
563 .ldt_name = LUSTRE_LOD_NAME,
564 .ldt_ops = &lod_device_type_ops,
565 .ldt_ctx_tags = LCT_MD_THREAD,
568 static int lod_obd_health_check(const struct lu_env *env,
569 struct obd_device *obd)
571 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
572 struct lod_ost_desc *ost;
578 cfs_foreach_bit(d->lod_ost_bitmap, i) {
580 LASSERT(ost && ost->ltd_ost);
581 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
582 /* one healthy device is enough */
590 static struct obd_ops lod_obd_device_ops = {
591 .o_owner = THIS_MODULE,
592 .o_connect = lod_obd_connect,
593 .o_disconnect = lod_obd_disconnect,
594 .o_health_check = lod_obd_health_check,
597 static int __init lod_mod_init(void)
599 struct lprocfs_static_vars lvars = { 0 };
600 cfs_proc_dir_entry_t *lov_proc_dir;
603 rc = lu_kmem_init(lod_caches);
607 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
608 LUSTRE_LOD_NAME, &lod_device_type);
610 lu_kmem_fini(lod_caches);
614 /* create "lov" entry in procfs for compatibility purposes */
615 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
616 if (lov_proc_dir == NULL) {
617 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
619 if (IS_ERR(lov_proc_dir))
620 CERROR("lod: can't create compat entry \"lov\": %d\n",
621 (int)PTR_ERR(lov_proc_dir));
627 static void __exit lod_mod_exit(void)
630 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
632 class_unregister_type(LUSTRE_LOD_NAME);
633 lu_kmem_fini(lod_caches);
636 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
637 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
638 MODULE_LICENSE("GPL");
640 module_init(lod_mod_init);
641 module_exit(lod_mod_exit);