1 /* -*- MODE: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mdd) routines
7 * Copyright (C) 2006 Cluster File Systems, Inc.
8 * Author: Wang Di <wangdi@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
29 # define EXPORT_SYMTAB
31 #define DEBUG_SUBSYSTEM S_MDS
33 #include <linux/module.h>
34 #include <linux/jbd.h>
36 #include <obd_class.h>
37 #include <lustre_ver.h>
38 #include <obd_support.h>
39 #include <lprocfs_status.h>
41 #include <linux/ldiskfs_fs.h>
42 #include <lustre_mds.h>
43 #include <lustre/lustre_idl.h>
45 #include "mdd_internal.h"
47 struct md_device_operations mdd_ops;
49 static const char *mdd_root_dir_name = "root";
50 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
51 const char *name, struct lu_device *next)
53 struct mdd_device *mdd = lu2mdd_dev(d);
57 mdd->mdd_child = lu2dt_dev(next);
59 /* Prepare transactions callbacks. */
60 mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
61 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
62 mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb;
63 mdd->mdd_txn_cb.dtc_cookie = mdd;
64 INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
65 rc = mdd_procfs_init(mdd, name);
69 static struct lu_device *mdd_device_fini(const struct lu_env *env,
72 struct mdd_device *mdd = lu2mdd_dev(d);
73 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
76 rc = mdd_procfs_fini(mdd);
78 CERROR("proc fini error %d \n", rc);
84 static int mdd_mount(const struct lu_env *env, struct mdd_device *mdd)
87 struct dt_object *root;
90 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
91 root = dt_store_open(env, mdd->mdd_child, mdd_root_dir_name,
94 LASSERT(root != NULL);
95 lu_object_put(env, &root->do_lu);
96 rc = orph_index_init(env, mdd);
103 static void mdd_device_shutdown(const struct lu_env *env,
104 struct mdd_device *m, struct lustre_cfg *cfg)
107 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
109 mdd_fini_obd(env, m, cfg);
110 orph_index_fini(env, m);
111 /* remove upcall device*/
112 md_upcall_fini(&m->mdd_md_dev);
116 static int mdd_process_config(const struct lu_env *env,
117 struct lu_device *d, struct lustre_cfg *cfg)
119 struct mdd_device *m = lu2mdd_dev(d);
120 struct dt_device *dt = m->mdd_child;
121 struct lu_device *next = &dt->dd_lu_dev;
125 switch (cfg->lcfg_command) {
127 rc = next->ld_ops->ldo_process_config(env, next, cfg);
130 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
132 rc = mdd_init_obd(env, m, cfg);
134 CERROR("lov init error %d \n", rc);
137 rc = mdd_mount(env, m);
140 rc = mdd_txn_init_credits(env, m);
143 mdd_device_shutdown(env, m, cfg);
145 rc = next->ld_ops->ldo_process_config(env, next, cfg);
152 static int mdd_lov_set_nextid(const struct lu_env *env,
153 struct mdd_device *mdd)
155 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
159 LASSERT(mds->mds_lov_objids != NULL);
160 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
161 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
162 mds->mds_lov_objids, NULL);
167 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
168 struct mdd_device *mdd)
170 /* XXX: to be implemented! */
175 static int mdd_recovery_complete(const struct lu_env *env,
178 struct mdd_device *mdd = lu2mdd_dev(d);
179 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
180 struct obd_device *obd = mdd2obd_dev(mdd);
184 LASSERT(mdd != NULL);
185 LASSERT(obd != NULL);
187 /* XXX: Do we need this in new stack? */
188 rc = mdd_lov_set_nextid(env, mdd);
190 CERROR("mdd_lov_set_nextid() failed %d\n",
195 /* XXX: cleanup unlink. */
196 rc = mdd_cleanup_unlink_llog(env, mdd);
198 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
203 /* Call that with obd_recovering = 1 just to update objids */
204 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
205 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
207 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
208 obd->obd_recovering = 0;
209 obd->obd_type->typ_dt_ops->o_postrecov(obd);
211 /* XXX: orphans handling. */
212 __mdd_orphan_cleanup(env, mdd);
213 rc = next->ld_ops->ldo_recovery_complete(env, next);
218 struct lu_device_operations mdd_lu_ops = {
219 .ldo_object_alloc = mdd_object_alloc,
220 .ldo_process_config = mdd_process_config,
221 .ldo_recovery_complete = mdd_recovery_complete
225 * No permission check is needed.
227 static int mdd_root_get(const struct lu_env *env,
228 struct md_device *m, struct lu_fid *f)
230 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
233 *f = mdd->mdd_root_fid;
238 * No permission check is needed.
240 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
243 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
248 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
254 * No permission check is needed.
256 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
257 int *md_size, int *cookie_size)
259 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
262 *md_size = mdd_lov_mdsize(env, mdd);
263 *cookie_size = mdd_lov_cookiesize(env, mdd);
268 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
269 int mode, unsigned long timeout, __u32 alg,
270 struct lustre_capa_key *keys)
272 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
273 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
277 mds->mds_capa_keys = keys;
278 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
283 static int mdd_update_capa_key(const struct lu_env *env,
285 struct lustre_capa_key *key)
287 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
288 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
292 rc = obd_set_info_async(lov_exp, strlen(KEY_CAPA_KEY), KEY_CAPA_KEY,
293 sizeof(*key), key, NULL);
297 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
298 struct lu_device_type *t,
299 struct lustre_cfg *lcfg)
302 struct mdd_device *m;
306 l = ERR_PTR(-ENOMEM);
308 md_device_init(&m->mdd_md_dev, t);
310 l->ld_ops = &mdd_lu_ops;
311 m->mdd_md_dev.md_ops = &mdd_ops;
312 md_upcall_init(&m->mdd_md_dev, NULL);
318 static void mdd_device_free(const struct lu_env *env,
319 struct lu_device *lu)
321 struct mdd_device *m = lu2mdd_dev(lu);
323 LASSERT(atomic_read(&lu->ld_ref) == 0);
324 md_device_fini(&m->mdd_md_dev);
328 static struct obd_ops mdd_obd_device_ops = {
329 .o_owner = THIS_MODULE
332 static void *mdd_ucred_key_init(const struct lu_context *ctx,
333 struct lu_context_key *key)
339 uc = ERR_PTR(-ENOMEM);
343 static void mdd_ucred_key_fini(const struct lu_context *ctx,
344 struct lu_context_key *key, void *data)
346 struct md_ucred *uc = data;
351 static struct lu_context_key mdd_ucred_key = {
352 .lct_tags = LCT_SESSION,
353 .lct_init = mdd_ucred_key_init,
354 .lct_fini = mdd_ucred_key_fini
357 struct md_ucred *md_ucred(const struct lu_env *env)
359 LASSERT(env->le_ses != NULL);
360 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
362 EXPORT_SYMBOL(md_ucred);
364 static void *mdd_capainfo_key_init(const struct lu_context *ctx,
365 struct lu_context_key *key)
367 struct md_capainfo *ci;
371 ci = ERR_PTR(-ENOMEM);
375 static void mdd_capainfo_key_fini(const struct lu_context *ctx,
376 struct lu_context_key *key, void *data)
378 struct md_capainfo *ci = data;
383 struct lu_context_key mdd_capainfo_key = {
384 .lct_tags = LCT_SESSION,
385 .lct_init = mdd_capainfo_key_init,
386 .lct_fini = mdd_capainfo_key_fini
389 struct md_capainfo *md_capainfo(const struct lu_env *env)
391 /* NB, in mdt_init0 */
392 if (env->le_ses == NULL)
394 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
396 EXPORT_SYMBOL(md_capainfo);
398 static int mdd_type_init(struct lu_device_type *t)
402 LU_CONTEXT_KEY_INIT(&mdd_thread_key);
403 result = lu_context_key_register(&mdd_thread_key);
405 LU_CONTEXT_KEY_INIT(&mdd_ucred_key);
406 result = lu_context_key_register(&mdd_ucred_key);
409 LU_CONTEXT_KEY_INIT(&mdd_capainfo_key);
410 result = lu_context_key_register(&mdd_capainfo_key);
415 static void mdd_type_fini(struct lu_device_type *t)
417 lu_context_key_degister(&mdd_capainfo_key);
418 lu_context_key_degister(&mdd_ucred_key);
419 lu_context_key_degister(&mdd_thread_key);
422 struct md_device_operations mdd_ops = {
423 .mdo_statfs = mdd_statfs,
424 .mdo_root_get = mdd_root_get,
425 .mdo_maxsize_get = mdd_maxsize_get,
426 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
427 .mdo_update_capa_key= mdd_update_capa_key,
430 static struct lu_device_type_operations mdd_device_type_ops = {
431 .ldto_init = mdd_type_init,
432 .ldto_fini = mdd_type_fini,
434 .ldto_device_alloc = mdd_device_alloc,
435 .ldto_device_free = mdd_device_free,
437 .ldto_device_init = mdd_device_init,
438 .ldto_device_fini = mdd_device_fini
441 static struct lu_device_type mdd_device_type = {
442 .ldt_tags = LU_DEVICE_MD,
443 .ldt_name = LUSTRE_MDD_NAME,
444 .ldt_ops = &mdd_device_type_ops,
445 .ldt_ctx_tags = LCT_MD_THREAD
448 static void *mdd_key_init(const struct lu_context *ctx,
449 struct lu_context_key *key)
451 struct mdd_thread_info *info;
455 info = ERR_PTR(-ENOMEM);
459 static void mdd_key_fini(const struct lu_context *ctx,
460 struct lu_context_key *key, void *data)
462 struct mdd_thread_info *info = data;
463 if (info->mti_max_lmm != NULL)
464 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
465 if (info->mti_max_cookie != NULL)
466 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
470 struct lu_context_key mdd_thread_key = {
471 .lct_tags = LCT_MD_THREAD,
472 .lct_init = mdd_key_init,
473 .lct_fini = mdd_key_fini
476 struct lprocfs_vars lprocfs_mdd_obd_vars[] = {
480 struct lprocfs_vars lprocfs_mdd_module_vars[] = {
484 LPROCFS_INIT_VARS(mdd, lprocfs_mdd_module_vars, lprocfs_mdd_obd_vars);
486 static int __init mdd_mod_init(void)
488 struct lprocfs_static_vars lvars;
489 lprocfs_init_vars(mdd, &lvars);
490 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
491 LUSTRE_MDD_NAME, &mdd_device_type);
494 static void __exit mdd_mod_exit(void)
496 class_unregister_type(LUSTRE_MDD_NAME);
499 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
500 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
501 MODULE_LICENSE("GPL");
503 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);