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 struct lu_device *lod_device_fini(const struct lu_env *env,
68 static int lod_process_config(const struct lu_env *env,
69 struct lu_device *dev,
70 struct lustre_cfg *lcfg)
72 struct lod_device *lod = lu2lod_dev(dev);
73 struct lu_device *next = &lod->lod_child->dd_lu_dev;
78 switch(lcfg->lcfg_command) {
80 case LCFG_LOV_DEL_OBD:
81 case LCFG_LOV_ADD_INA:
82 case LCFG_LOV_ADD_OBD: {
85 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1 */
86 arg1 = lustre_cfg_string(lcfg, 1);
88 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
89 GOTO(out, rc = -EINVAL);
90 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
91 GOTO(out, rc = -EINVAL);
93 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD)
94 rc = lod_add_device(env, lod, arg1, index, gen, 1);
95 else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA)
96 rc = lod_add_device(env, lod, arg1, index, gen, 0);
98 rc = lod_del_device(env, lod, arg1, index, gen);
104 struct lprocfs_static_vars v = { 0 };
105 struct obd_device *obd = lod2obd(lod);
107 lprocfs_lod_init_vars(&v);
109 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
116 lu_dev_del_linkage(dev->ld_site, dev);
118 lod_foreach_ost(lod, i) {
119 struct lod_ost_desc *ost;
120 ost = OST_TGT(lod, i);
121 LASSERT(ost && ost->ltd_ost);
122 next = &ost->ltd_ost->dd_lu_dev;
123 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
125 CERROR("%s: can't process %u: %d\n",
126 lod2obd(lod)->obd_name,
127 lcfg->lcfg_command, rc);
132 * do cleanup on underlying storage only when
133 * all OSPs are cleaned up, as they use that OSD as well
135 next = &lod->lod_child->dd_lu_dev;
136 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
138 CERROR("%s: can't process %u: %d\n",
139 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
141 rc = obd_disconnect(lod->lod_child_exp);
143 CERROR("error in disconnect from storage: %d\n", rc);
147 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
157 static int lod_recovery_complete(const struct lu_env *env,
158 struct lu_device *dev)
160 struct lod_device *lod = lu2lod_dev(dev);
161 struct lu_device *next = &lod->lod_child->dd_lu_dev;
162 struct lod_ost_desc *ost;
166 LASSERT(lod->lod_recovery_completed == 0);
167 lod->lod_recovery_completed = 1;
169 rc = next->ld_ops->ldo_recovery_complete(env, next);
172 lod_foreach_ost(lod, i) {
173 ost = OST_TGT(lod, i);
174 LASSERT(ost && ost->ltd_ost);
175 next = &ost->ltd_ost->dd_lu_dev;
176 rc = next->ld_ops->ldo_recovery_complete(env, next);
178 CERROR("%s: can't complete recovery on #%d: %d\n",
179 lod2obd(lod)->obd_name, i, rc);
186 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
187 struct lu_device *cdev)
189 struct lod_device *lod = lu2lod_dev(cdev);
190 struct lu_device *next = &lod->lod_child->dd_lu_dev;
194 rc = next->ld_ops->ldo_prepare(env, pdev, next);
199 const struct lu_device_operations lod_lu_ops = {
200 .ldo_process_config = lod_process_config,
201 .ldo_recovery_complete = lod_recovery_complete,
202 .ldo_prepare = lod_prepare,
205 static int lod_root_get(const struct lu_env *env,
206 struct dt_device *dev, struct lu_fid *f)
208 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
211 static int lod_statfs(const struct lu_env *env,
212 struct dt_device *dev, struct obd_statfs *sfs)
214 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
217 static struct thandle *lod_trans_create(const struct lu_env *env,
218 struct dt_device *dev)
220 return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
223 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
226 return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
229 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
231 /* XXX: we don't know next device, will be fixed with DNE */
232 return dt_trans_stop(env, th->th_dev, th);
235 static void lod_conf_get(const struct lu_env *env,
236 const struct dt_device *dev,
237 struct dt_device_param *param)
239 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
242 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
244 struct lod_device *lod = dt2lod_dev(dev);
245 struct lod_ost_desc *ost;
250 lod_foreach_ost(lod, i) {
251 ost = OST_TGT(lod, i);
252 LASSERT(ost && ost->ltd_ost);
253 rc = dt_sync(env, ost->ltd_ost);
255 CERROR("%s: can't sync %u: %d\n",
256 lod2obd(lod)->obd_name, i, rc);
262 rc = dt_sync(env, lod->lod_child);
267 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
269 return dt_ro(env, dt2lod_dev(dev)->lod_child);
272 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
274 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
277 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
278 int mode, unsigned long timeout,
279 __u32 alg, struct lustre_capa_key *keys)
281 struct dt_device *next = dt2lod_dev(dev)->lod_child;
282 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
285 static const struct dt_device_operations lod_dt_ops = {
286 .dt_root_get = lod_root_get,
287 .dt_statfs = lod_statfs,
288 .dt_trans_create = lod_trans_create,
289 .dt_trans_start = lod_trans_start,
290 .dt_trans_stop = lod_trans_stop,
291 .dt_conf_get = lod_conf_get,
294 .dt_commit_async = lod_commit_async,
295 .dt_init_capa_ctxt = lod_init_capa_ctxt,
298 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
299 struct lustre_cfg *cfg)
301 struct obd_connect_data *data = NULL;
302 struct obd_device *obd;
303 char *nextdev = NULL, *p, *s;
308 LASSERT(lod->lod_child_exp == NULL);
310 /* compatibility hack: we still use old config logs
311 * which specify LOV, but we need to learn underlying
312 * OSD device, which is supposed to be:
313 * <fsname>-MDTxxxx-osd
315 * 2.x MGS generates lines like the following:
316 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
317 * 1.8 MGS generates lines like the following:
318 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
320 * we use "-MDT" to differentiate 2.x from 1.8 */
322 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
324 OBD_ALLOC(nextdev, len);
326 GOTO(out, rc = -ENOMEM);
329 s = strstr(nextdev, "-mdtlov");
330 if (unlikely(s == NULL)) {
331 CERROR("unable to parse device name %s\n",
332 lustre_cfg_string(cfg, 0));
333 GOTO(out, rc = -EINVAL);
336 if (strstr(nextdev, "-MDT")) {
341 strcpy(s, "-MDT0000-osd");
344 CERROR("unable to parse device name %s\n",
345 lustre_cfg_string(cfg, 0));
346 GOTO(out, rc = -EINVAL);
351 GOTO(out, rc = -ENOMEM);
353 obd = class_name2obd(nextdev);
355 CERROR("can not locate next device: %s\n", nextdev);
356 GOTO(out, rc = -ENOTCONN);
359 data->ocd_connect_flags = OBD_CONNECT_VERSION;
360 data->ocd_version = LUSTRE_VERSION_CODE;
362 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
365 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
369 lod->lod_dt_dev.dd_lu_dev.ld_site =
370 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
371 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
372 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
378 OBD_FREE(nextdev, len);
382 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
383 struct lu_device_type *ldt, struct lustre_cfg *cfg)
385 struct dt_device_param ddp;
386 struct proc_dir_entry *lov_proc_dir;
387 struct obd_device *obd;
391 obd = class_name2obd(lustre_cfg_string(cfg, 0));
393 CERROR("Cannot find obd with name %s\n",
394 lustre_cfg_string(cfg, 0));
398 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
399 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
400 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
401 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
403 rc = lod_connect_to_osd(env, lod, cfg);
407 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
408 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
410 /* setup obd to be used with old lov code */
411 rc = lod_pools_init(lod, cfg);
413 GOTO(out_disconnect, rc);
415 /* for compatibility we link old procfs's OSC entries to osp ones */
416 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
418 cfs_proc_dir_entry_t *symlink = NULL;
420 OBD_ALLOC(name, strlen(obd->obd_name) + 1);
422 strcpy(name, obd->obd_name);
423 if (strstr(name, "lov"))
424 symlink = lprocfs_add_symlink(name,
428 OBD_FREE(name, strlen(obd->obd_name) + 1);
429 lod->lod_symlink = symlink;
433 cfs_mutex_init(&lod->lod_mutex);
434 cfs_init_rwsem(&lod->lod_rw_sem);
435 cfs_spin_lock_init(&lod->lod_desc_lock);
440 obd_disconnect(lod->lod_child_exp);
444 static struct lu_device *lod_device_free(const struct lu_env *env,
445 struct lu_device *lu)
447 struct lod_device *lod = lu2lod_dev(lu);
448 struct lu_device *next = &lod->lod_child->dd_lu_dev;
451 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
452 dt_device_fini(&lod->lod_dt_dev);
457 static struct lu_device *lod_device_alloc(const struct lu_env *env,
458 struct lu_device_type *type,
459 struct lustre_cfg *lcfg)
461 struct lod_device *lod;
462 struct lu_device *lu_dev;
466 lu_dev = ERR_PTR(-ENOMEM);
470 lu_dev = lod2lu_dev(lod);
471 dt_device_init(&lod->lod_dt_dev, type);
472 rc = lod_init0(env, lod, type, lcfg);
474 lod_device_free(env, lu_dev);
475 lu_dev = ERR_PTR(rc);
482 static struct lu_device *lod_device_fini(const struct lu_env *env,
485 struct lod_device *lod = lu2lod_dev(d);
488 if (lod->lod_symlink)
489 lprocfs_remove(&lod->lod_symlink);
497 * we use exports to track all LOD users
499 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
500 struct obd_device *obd, struct obd_uuid *cluuid,
501 struct obd_connect_data *data, void *localdata)
503 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
504 struct lustre_handle conn;
508 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
510 rc = class_connect(&conn, obd, cluuid);
514 *exp = class_conn2export(&conn);
516 cfs_mutex_lock(&lod->lod_mutex);
518 /* at the moment we expect the only user */
519 LASSERT(lod->lod_connects == 1);
520 cfs_mutex_unlock(&lod->lod_mutex);
526 * once last export (we don't count self-export) disappeared
527 * lod can be released
529 static int lod_obd_disconnect(struct obd_export *exp)
531 struct obd_device *obd = exp->exp_obd;
532 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
536 /* Only disconnect the underlying layers on the final disconnect. */
537 cfs_mutex_lock(&lod->lod_mutex);
539 if (lod->lod_connects != 0) {
540 /* why should there be more than 1 connect? */
541 cfs_mutex_unlock(&lod->lod_mutex);
542 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
546 cfs_mutex_unlock(&lod->lod_mutex);
548 /* the last user of lod has gone, let's release the device */
552 rc = class_disconnect(exp); /* bz 9811 */
554 if (rc == 0 && release)
555 class_manual_cleanup(obd);
559 LU_KEY_INIT(lod, struct lod_thread_info);
561 static void lod_key_fini(const struct lu_context *ctx,
562 struct lu_context_key *key, void *data)
564 struct lod_thread_info *info = data;
565 /* allocated in lod_get_lov_ea
566 * XXX: this is overload, a tread may have such store but used only
567 * once. Probably better would be pool of such stores per LOD.
569 if (info->lti_ea_store) {
570 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
571 info->lti_ea_store = NULL;
572 info->lti_ea_store_size = 0;
577 /* context key: lod_thread_key */
578 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
580 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
582 static struct lu_device_type_operations lod_device_type_ops = {
583 .ldto_init = lod_type_init,
584 .ldto_fini = lod_type_fini,
586 .ldto_start = lod_type_start,
587 .ldto_stop = lod_type_stop,
589 .ldto_device_alloc = lod_device_alloc,
590 .ldto_device_free = lod_device_free,
592 .ldto_device_fini = lod_device_fini
595 static struct lu_device_type lod_device_type = {
596 .ldt_tags = LU_DEVICE_DT,
597 .ldt_name = LUSTRE_LOD_NAME,
598 .ldt_ops = &lod_device_type_ops,
599 .ldt_ctx_tags = LCT_MD_THREAD,
602 static int lod_obd_health_check(const struct lu_env *env,
603 struct obd_device *obd)
605 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
606 struct lod_ost_desc *ost;
612 lod_foreach_ost(d, i) {
614 LASSERT(ost && ost->ltd_ost);
615 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
616 /* one healthy device is enough */
624 static struct obd_ops lod_obd_device_ops = {
625 .o_owner = THIS_MODULE,
626 .o_connect = lod_obd_connect,
627 .o_disconnect = lod_obd_disconnect,
628 .o_health_check = lod_obd_health_check,
629 .o_pool_new = lod_pool_new,
630 .o_pool_rem = lod_pool_remove,
631 .o_pool_add = lod_pool_add,
632 .o_pool_del = lod_pool_del,
635 static int __init lod_mod_init(void)
637 struct lprocfs_static_vars lvars = { 0 };
638 cfs_proc_dir_entry_t *lov_proc_dir;
641 rc = lu_kmem_init(lod_caches);
645 lprocfs_lod_init_vars(&lvars);
647 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
648 LUSTRE_LOD_NAME, &lod_device_type);
650 lu_kmem_fini(lod_caches);
654 /* create "lov" entry in procfs for compatibility purposes */
655 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
656 if (lov_proc_dir == NULL) {
657 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
659 if (IS_ERR(lov_proc_dir))
660 CERROR("lod: can't create compat entry \"lov\": %d\n",
661 (int)PTR_ERR(lov_proc_dir));
667 static void __exit lod_mod_exit(void)
670 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
672 class_unregister_type(LUSTRE_LOD_NAME);
673 lu_kmem_fini(lod_caches);
676 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
677 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
678 MODULE_LICENSE("GPL");
680 module_init(lod_mod_init);
681 module_exit(lod_mod_exit);