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 extern struct lu_object_operations lod_lu_obj_ops;
52 extern struct dt_object_operations lod_obj_ops;
54 /* Slab for OSD object allocation */
55 cfs_mem_cache_t *lod_object_kmem;
57 static struct lu_kmem_descr lod_caches[] = {
59 .ckd_cache = &lod_object_kmem,
60 .ckd_name = "lod_obj",
61 .ckd_size = sizeof(struct lod_object)
68 static struct lu_device *lod_device_fini(const struct lu_env *env,
71 struct lu_object *lod_object_alloc(const struct lu_env *env,
72 const struct lu_object_header *hdr,
73 struct lu_device *dev)
75 struct lu_object *lu_obj;
76 struct lod_object *lo;
78 OBD_SLAB_ALLOC_PTR_GFP(lo, lod_object_kmem, CFS_ALLOC_IO);
82 lu_obj = lod2lu_obj(lo);
83 dt_object_init(&lo->ldo_obj, NULL, dev);
84 lo->ldo_obj.do_ops = &lod_obj_ops;
85 lu_obj->lo_ops = &lod_lu_obj_ops;
90 static int lod_process_config(const struct lu_env *env,
91 struct lu_device *dev,
92 struct lustre_cfg *lcfg)
94 struct lod_device *lod = lu2lod_dev(dev);
95 struct lu_device *next = &lod->lod_child->dd_lu_dev;
100 switch(lcfg->lcfg_command) {
102 case LCFG_LOV_DEL_OBD:
103 case LCFG_LOV_ADD_INA:
104 case LCFG_LOV_ADD_OBD: {
107 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1 */
108 arg1 = lustre_cfg_string(lcfg, 1);
110 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
111 GOTO(out, rc = -EINVAL);
112 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
113 GOTO(out, rc = -EINVAL);
115 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD)
116 rc = lod_add_device(env, lod, arg1, index, gen, 1);
117 else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA)
118 rc = lod_add_device(env, lod, arg1, index, gen, 0);
120 rc = lod_del_device(env, lod, arg1, index, gen);
126 struct lprocfs_static_vars v = { 0 };
127 struct obd_device *obd = lod2obd(lod);
129 lprocfs_lod_init_vars(&v);
131 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
138 lu_dev_del_linkage(dev->ld_site, dev);
140 lod_foreach_ost(lod, i) {
141 struct lod_ost_desc *ost;
142 ost = OST_TGT(lod, i);
143 LASSERT(ost && ost->ltd_ost);
144 next = &ost->ltd_ost->dd_lu_dev;
145 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
147 CERROR("%s: can't process %u: %d\n",
148 lod2obd(lod)->obd_name,
149 lcfg->lcfg_command, rc);
154 * do cleanup on underlying storage only when
155 * all OSPs are cleaned up, as they use that OSD as well
157 next = &lod->lod_child->dd_lu_dev;
158 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
160 CERROR("%s: can't process %u: %d\n",
161 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
163 rc = obd_disconnect(lod->lod_child_exp);
165 CERROR("error in disconnect from storage: %d\n", rc);
169 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
179 static int lod_recovery_complete(const struct lu_env *env,
180 struct lu_device *dev)
182 struct lod_device *lod = lu2lod_dev(dev);
183 struct lu_device *next = &lod->lod_child->dd_lu_dev;
184 struct lod_ost_desc *ost;
188 LASSERT(lod->lod_recovery_completed == 0);
189 lod->lod_recovery_completed = 1;
191 rc = next->ld_ops->ldo_recovery_complete(env, next);
194 lod_foreach_ost(lod, i) {
195 ost = OST_TGT(lod, i);
196 LASSERT(ost && ost->ltd_ost);
197 next = &ost->ltd_ost->dd_lu_dev;
198 rc = next->ld_ops->ldo_recovery_complete(env, next);
200 CERROR("%s: can't complete recovery on #%d: %d\n",
201 lod2obd(lod)->obd_name, i, rc);
208 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
209 struct lu_device *cdev)
211 struct lod_device *lod = lu2lod_dev(cdev);
212 struct lu_device *next = &lod->lod_child->dd_lu_dev;
216 rc = next->ld_ops->ldo_prepare(env, pdev, next);
221 const struct lu_device_operations lod_lu_ops = {
222 .ldo_object_alloc = lod_object_alloc,
223 .ldo_process_config = lod_process_config,
224 .ldo_recovery_complete = lod_recovery_complete,
225 .ldo_prepare = lod_prepare,
228 static int lod_root_get(const struct lu_env *env,
229 struct dt_device *dev, struct lu_fid *f)
231 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
234 static int lod_statfs(const struct lu_env *env,
235 struct dt_device *dev, struct obd_statfs *sfs)
237 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
240 static struct thandle *lod_trans_create(const struct lu_env *env,
241 struct dt_device *dev)
243 return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
246 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
249 return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
252 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
254 /* XXX: we don't know next device, will be fixed with DNE */
255 return dt_trans_stop(env, th->th_dev, th);
258 static void lod_conf_get(const struct lu_env *env,
259 const struct dt_device *dev,
260 struct dt_device_param *param)
262 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
265 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
267 struct lod_device *lod = dt2lod_dev(dev);
268 struct lod_ost_desc *ost;
273 lod_foreach_ost(lod, i) {
274 ost = OST_TGT(lod, i);
275 LASSERT(ost && ost->ltd_ost);
276 rc = dt_sync(env, ost->ltd_ost);
278 CERROR("%s: can't sync %u: %d\n",
279 lod2obd(lod)->obd_name, i, rc);
285 rc = dt_sync(env, lod->lod_child);
290 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
292 return dt_ro(env, dt2lod_dev(dev)->lod_child);
295 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
297 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
300 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
301 int mode, unsigned long timeout,
302 __u32 alg, struct lustre_capa_key *keys)
304 struct dt_device *next = dt2lod_dev(dev)->lod_child;
305 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
308 static const struct dt_device_operations lod_dt_ops = {
309 .dt_root_get = lod_root_get,
310 .dt_statfs = lod_statfs,
311 .dt_trans_create = lod_trans_create,
312 .dt_trans_start = lod_trans_start,
313 .dt_trans_stop = lod_trans_stop,
314 .dt_conf_get = lod_conf_get,
317 .dt_commit_async = lod_commit_async,
318 .dt_init_capa_ctxt = lod_init_capa_ctxt,
321 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
322 struct lustre_cfg *cfg)
324 struct obd_connect_data *data = NULL;
325 struct obd_device *obd;
326 char *nextdev = NULL, *p, *s;
331 LASSERT(lod->lod_child_exp == NULL);
333 /* compatibility hack: we still use old config logs
334 * which specify LOV, but we need to learn underlying
335 * OSD device, which is supposed to be:
336 * <fsname>-MDTxxxx-osd
338 * 2.x MGS generates lines like the following:
339 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
340 * 1.8 MGS generates lines like the following:
341 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
343 * we use "-MDT" to differentiate 2.x from 1.8 */
345 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
347 OBD_ALLOC(nextdev, len);
349 GOTO(out, rc = -ENOMEM);
352 s = strstr(nextdev, "-mdtlov");
353 if (unlikely(s == NULL)) {
354 CERROR("unable to parse device name %s\n",
355 lustre_cfg_string(cfg, 0));
356 GOTO(out, rc = -EINVAL);
359 if (strstr(nextdev, "-MDT")) {
364 strcpy(s, "-MDT0000-osd");
367 CERROR("unable to parse device name %s\n",
368 lustre_cfg_string(cfg, 0));
369 GOTO(out, rc = -EINVAL);
374 GOTO(out, rc = -ENOMEM);
376 obd = class_name2obd(nextdev);
378 CERROR("can not locate next device: %s\n", nextdev);
379 GOTO(out, rc = -ENOTCONN);
382 data->ocd_connect_flags = OBD_CONNECT_VERSION;
383 data->ocd_version = LUSTRE_VERSION_CODE;
385 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
388 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
392 lod->lod_dt_dev.dd_lu_dev.ld_site =
393 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
394 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
395 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
401 OBD_FREE(nextdev, len);
405 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
406 struct lu_device_type *ldt, struct lustre_cfg *cfg)
408 struct dt_device_param ddp;
409 struct obd_device *obd;
413 obd = class_name2obd(lustre_cfg_string(cfg, 0));
415 CERROR("Cannot find obd with name %s\n",
416 lustre_cfg_string(cfg, 0));
420 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
421 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
422 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
423 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
425 rc = lod_connect_to_osd(env, lod, cfg);
429 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
430 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
432 /* setup obd to be used with old lov code */
433 rc = lod_pools_init(lod, cfg);
435 GOTO(out_disconnect, rc);
437 rc = lod_procfs_init(lod);
441 mutex_init(&lod->lod_mutex);
442 init_rwsem(&lod->lod_rw_sem);
443 spin_lock_init(&lod->lod_desc_lock);
450 obd_disconnect(lod->lod_child_exp);
454 static struct lu_device *lod_device_free(const struct lu_env *env,
455 struct lu_device *lu)
457 struct lod_device *lod = lu2lod_dev(lu);
458 struct lu_device *next = &lod->lod_child->dd_lu_dev;
461 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
462 dt_device_fini(&lod->lod_dt_dev);
467 static struct lu_device *lod_device_alloc(const struct lu_env *env,
468 struct lu_device_type *type,
469 struct lustre_cfg *lcfg)
471 struct lod_device *lod;
472 struct lu_device *lu_dev;
476 lu_dev = ERR_PTR(-ENOMEM);
480 lu_dev = lod2lu_dev(lod);
481 dt_device_init(&lod->lod_dt_dev, type);
482 rc = lod_init0(env, lod, type, lcfg);
484 lod_device_free(env, lu_dev);
485 lu_dev = ERR_PTR(rc);
492 static struct lu_device *lod_device_fini(const struct lu_env *env,
495 struct lod_device *lod = lu2lod_dev(d);
500 lod_procfs_fini(lod);
506 * we use exports to track all LOD users
508 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
509 struct obd_device *obd, struct obd_uuid *cluuid,
510 struct obd_connect_data *data, void *localdata)
512 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
513 struct lustre_handle conn;
517 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
519 rc = class_connect(&conn, obd, cluuid);
523 *exp = class_conn2export(&conn);
525 mutex_lock(&lod->lod_mutex);
527 /* at the moment we expect the only user */
528 LASSERT(lod->lod_connects == 1);
529 mutex_unlock(&lod->lod_mutex);
535 * once last export (we don't count self-export) disappeared
536 * lod can be released
538 static int lod_obd_disconnect(struct obd_export *exp)
540 struct obd_device *obd = exp->exp_obd;
541 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
545 /* Only disconnect the underlying layers on the final disconnect. */
546 mutex_lock(&lod->lod_mutex);
548 if (lod->lod_connects != 0) {
549 /* why should there be more than 1 connect? */
550 mutex_unlock(&lod->lod_mutex);
551 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
555 mutex_unlock(&lod->lod_mutex);
557 /* the last user of lod has gone, let's release the device */
561 rc = class_disconnect(exp); /* bz 9811 */
563 if (rc == 0 && release)
564 class_manual_cleanup(obd);
568 LU_KEY_INIT(lod, struct lod_thread_info);
570 static void lod_key_fini(const struct lu_context *ctx,
571 struct lu_context_key *key, void *data)
573 struct lod_thread_info *info = data;
574 /* allocated in lod_get_lov_ea
575 * XXX: this is overload, a tread may have such store but used only
576 * once. Probably better would be pool of such stores per LOD.
578 if (info->lti_ea_store) {
579 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
580 info->lti_ea_store = NULL;
581 info->lti_ea_store_size = 0;
586 /* context key: lod_thread_key */
587 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
589 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
591 static struct lu_device_type_operations lod_device_type_ops = {
592 .ldto_init = lod_type_init,
593 .ldto_fini = lod_type_fini,
595 .ldto_start = lod_type_start,
596 .ldto_stop = lod_type_stop,
598 .ldto_device_alloc = lod_device_alloc,
599 .ldto_device_free = lod_device_free,
601 .ldto_device_fini = lod_device_fini
604 static struct lu_device_type lod_device_type = {
605 .ldt_tags = LU_DEVICE_DT,
606 .ldt_name = LUSTRE_LOD_NAME,
607 .ldt_ops = &lod_device_type_ops,
608 .ldt_ctx_tags = LCT_MD_THREAD,
611 static int lod_obd_health_check(const struct lu_env *env,
612 struct obd_device *obd)
614 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
615 struct lod_ost_desc *ost;
621 lod_foreach_ost(d, i) {
623 LASSERT(ost && ost->ltd_ost);
624 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
625 /* one healthy device is enough */
633 static struct obd_ops lod_obd_device_ops = {
634 .o_owner = THIS_MODULE,
635 .o_connect = lod_obd_connect,
636 .o_disconnect = lod_obd_disconnect,
637 .o_health_check = lod_obd_health_check,
638 .o_pool_new = lod_pool_new,
639 .o_pool_rem = lod_pool_remove,
640 .o_pool_add = lod_pool_add,
641 .o_pool_del = lod_pool_del,
644 static int __init lod_mod_init(void)
646 struct lprocfs_static_vars lvars = { 0 };
647 cfs_proc_dir_entry_t *lov_proc_dir;
650 rc = lu_kmem_init(lod_caches);
654 lprocfs_lod_init_vars(&lvars);
656 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
657 LUSTRE_LOD_NAME, &lod_device_type);
659 lu_kmem_fini(lod_caches);
663 /* create "lov" entry in procfs for compatibility purposes */
664 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
665 if (lov_proc_dir == NULL) {
666 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
668 if (IS_ERR(lov_proc_dir))
669 CERROR("lod: can't create compat entry \"lov\": %d\n",
670 (int)PTR_ERR(lov_proc_dir));
676 static void __exit lod_mod_exit(void)
679 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
681 class_unregister_type(LUSTRE_LOD_NAME);
682 lu_kmem_fini(lod_caches);
685 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
686 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
687 MODULE_LICENSE("GPL");
689 module_init(lod_mod_init);
690 module_exit(lod_mod_exit);