1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/cmm/cmm_device.c
38 * Lustre Cluster Metadata Manager (cmm)
40 * Author: Mike Pershin <tappro@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
51 #include <obd_class.h>
52 #include <lprocfs_status.h>
53 #include <lustre_ver.h>
54 #include "cmm_internal.h"
55 #include "mdc_internal.h"
56 #ifdef HAVE_QUOTA_SUPPORT
57 # include <lustre_quota.h>
60 static struct obd_ops cmm_obd_device_ops = {
61 .o_owner = THIS_MODULE
64 static const struct lu_device_operations cmm_lu_ops;
66 static inline int lu_device_is_cmm(struct lu_device *d)
68 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &cmm_lu_ops);
71 int cmm_root_get(const struct lu_env *env, struct md_device *md,
74 struct cmm_device *cmm_dev = md2cmm_dev(md);
75 /* valid only on master MDS */
76 if (cmm_dev->cmm_local_num == 0)
77 return cmm_child_ops(cmm_dev)->mdo_root_get(env,
78 cmm_dev->cmm_child, fid);
83 static int cmm_statfs(const struct lu_env *env, struct md_device *md,
86 struct cmm_device *cmm_dev = md2cmm_dev(md);
90 rc = cmm_child_ops(cmm_dev)->mdo_statfs(env,
91 cmm_dev->cmm_child, sfs);
95 static int cmm_maxsize_get(const struct lu_env *env, struct md_device *md,
96 int *md_size, int *cookie_size)
98 struct cmm_device *cmm_dev = md2cmm_dev(md);
101 rc = cmm_child_ops(cmm_dev)->mdo_maxsize_get(env, cmm_dev->cmm_child,
102 md_size, cookie_size);
106 static int cmm_init_capa_ctxt(const struct lu_env *env, struct md_device *md,
107 int mode , unsigned long timeout, __u32 alg,
108 struct lustre_capa_key *keys)
110 struct cmm_device *cmm_dev = md2cmm_dev(md);
113 LASSERT(cmm_child_ops(cmm_dev)->mdo_init_capa_ctxt);
114 rc = cmm_child_ops(cmm_dev)->mdo_init_capa_ctxt(env, cmm_dev->cmm_child,
120 static int cmm_update_capa_key(const struct lu_env *env,
121 struct md_device *md,
122 struct lustre_capa_key *key)
124 struct cmm_device *cmm_dev = md2cmm_dev(md);
127 rc = cmm_child_ops(cmm_dev)->mdo_update_capa_key(env,
133 #ifdef HAVE_QUOTA_SUPPORT
134 static int cmm_quota_notify(const struct lu_env *env, struct md_device *m)
136 struct cmm_device *cmm_dev = md2cmm_dev(m);
140 /* disable quota for CMD case temporary. */
141 if (cmm_dev->cmm_tgt_count)
144 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_notify(env,
149 static int cmm_quota_setup(const struct lu_env *env, struct md_device *m,
152 struct cmm_device *cmm_dev = md2cmm_dev(m);
156 /* disable quota for CMD case temporary. */
157 if (cmm_dev->cmm_tgt_count)
160 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_setup(env,
166 static int cmm_quota_cleanup(const struct lu_env *env, struct md_device *m)
168 struct cmm_device *cmm_dev = md2cmm_dev(m);
172 /* disable quota for CMD case temporary. */
173 if (cmm_dev->cmm_tgt_count)
176 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_cleanup(env,
181 static int cmm_quota_recovery(const struct lu_env *env, struct md_device *m)
183 struct cmm_device *cmm_dev = md2cmm_dev(m);
187 /* disable quota for CMD case temporary. */
188 if (cmm_dev->cmm_tgt_count)
191 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_recovery(env,
196 static int cmm_quota_check(const struct lu_env *env, struct md_device *m,
197 struct obd_export *exp, __u32 type)
199 struct cmm_device *cmm_dev = md2cmm_dev(m);
203 /* disable quota for CMD case temporary. */
204 if (cmm_dev->cmm_tgt_count)
207 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_check(env,
213 static int cmm_quota_on(const struct lu_env *env, struct md_device *m,
214 __u32 type, __u32 id)
216 struct cmm_device *cmm_dev = md2cmm_dev(m);
220 /* disable quota for CMD case temporary. */
221 if (cmm_dev->cmm_tgt_count)
224 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_on(env,
230 static int cmm_quota_off(const struct lu_env *env, struct md_device *m,
231 __u32 type, __u32 id)
233 struct cmm_device *cmm_dev = md2cmm_dev(m);
237 /* disable quota for CMD case temporary. */
238 if (cmm_dev->cmm_tgt_count)
241 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_off(env,
247 static int cmm_quota_setinfo(const struct lu_env *env, struct md_device *m,
248 __u32 type, __u32 id, struct obd_dqinfo *dqinfo)
250 struct cmm_device *cmm_dev = md2cmm_dev(m);
254 /* disable quota for CMD case temporary. */
255 if (cmm_dev->cmm_tgt_count)
258 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_setinfo(env,
264 static int cmm_quota_getinfo(const struct lu_env *env,
265 const struct md_device *m,
266 __u32 type, __u32 id, struct obd_dqinfo *dqinfo)
268 struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
272 /* disable quota for CMD case temporary. */
273 if (cmm_dev->cmm_tgt_count)
276 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getinfo(env,
282 static int cmm_quota_setquota(const struct lu_env *env, struct md_device *m,
283 __u32 type, __u32 id, struct obd_dqblk *dqblk)
285 struct cmm_device *cmm_dev = md2cmm_dev(m);
289 /* disable quota for CMD case temporary. */
290 if (cmm_dev->cmm_tgt_count)
293 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_setquota(env,
299 static int cmm_quota_getquota(const struct lu_env *env,
300 const struct md_device *m,
301 __u32 type, __u32 id, struct obd_dqblk *dqblk)
303 struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
307 /* disable quota for CMD case temporary. */
308 if (cmm_dev->cmm_tgt_count)
311 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getquota(env,
317 static int cmm_quota_getoinfo(const struct lu_env *env,
318 const struct md_device *m,
319 __u32 type, __u32 id, struct obd_dqinfo *dqinfo)
321 struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
325 /* disable quota for CMD case temporary. */
326 if (cmm_dev->cmm_tgt_count)
329 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getoinfo(env,
335 static int cmm_quota_getoquota(const struct lu_env *env,
336 const struct md_device *m,
337 __u32 type, __u32 id, struct obd_dqblk *dqblk)
339 struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
343 /* disable quota for CMD case temporary. */
344 if (cmm_dev->cmm_tgt_count)
347 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getoquota(env,
353 static int cmm_quota_invalidate(const struct lu_env *env, struct md_device *m,
356 struct cmm_device *cmm_dev = md2cmm_dev(m);
360 /* disable quota for CMD case temporary. */
361 if (cmm_dev->cmm_tgt_count)
364 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_invalidate(env,
370 static int cmm_quota_finvalidate(const struct lu_env *env, struct md_device *m,
373 struct cmm_device *cmm_dev = md2cmm_dev(m);
377 /* disable quota for CMD case temporary. */
378 if (cmm_dev->cmm_tgt_count)
381 rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_finvalidate(env,
388 static const struct md_device_operations cmm_md_ops = {
389 .mdo_statfs = cmm_statfs,
390 .mdo_root_get = cmm_root_get,
391 .mdo_maxsize_get = cmm_maxsize_get,
392 .mdo_init_capa_ctxt = cmm_init_capa_ctxt,
393 .mdo_update_capa_key = cmm_update_capa_key,
394 #ifdef HAVE_QUOTA_SUPPORT
396 .mqo_notify = cmm_quota_notify,
397 .mqo_setup = cmm_quota_setup,
398 .mqo_cleanup = cmm_quota_cleanup,
399 .mqo_recovery = cmm_quota_recovery,
400 .mqo_check = cmm_quota_check,
401 .mqo_on = cmm_quota_on,
402 .mqo_off = cmm_quota_off,
403 .mqo_setinfo = cmm_quota_setinfo,
404 .mqo_getinfo = cmm_quota_getinfo,
405 .mqo_setquota = cmm_quota_setquota,
406 .mqo_getquota = cmm_quota_getquota,
407 .mqo_getoinfo = cmm_quota_getoinfo,
408 .mqo_getoquota = cmm_quota_getoquota,
409 .mqo_invalidate = cmm_quota_invalidate,
410 .mqo_finvalidate = cmm_quota_finvalidate
415 extern struct lu_device_type mdc_device_type;
417 static int cmm_post_init_mdc(const struct lu_env *env,
418 struct cmm_device *cmm)
420 int max_mdsize, max_cookiesize, rc;
421 struct mdc_device *mc, *tmp;
423 /* get the max mdsize and cookiesize from lower layer */
424 rc = cmm_maxsize_get(env, &cmm->cmm_md_dev, &max_mdsize,
429 spin_lock(&cmm->cmm_tgt_guard);
430 list_for_each_entry_safe(mc, tmp, &cmm->cmm_targets,
432 cmm_mdc_init_ea_size(env, mc, max_mdsize, max_cookiesize);
434 spin_unlock(&cmm->cmm_tgt_guard);
438 /* --- cmm_lu_operations --- */
439 /* add new MDC to the CMM, create MDC lu_device and connect it to mdc_obd */
440 static int cmm_add_mdc(const struct lu_env *env,
441 struct cmm_device *cm, struct lustre_cfg *cfg)
443 struct lu_device_type *ldt = &mdc_device_type;
444 char *p, *num = lustre_cfg_string(cfg, 2);
445 struct mdc_device *mc, *tmp;
446 struct lu_fld_target target;
447 struct lu_device *ld;
448 struct lu_device *cmm_lu = cmm2lu_dev(cm);
453 /* find out that there is no such mdc */
455 mdc_num = simple_strtol(num, &p, 10);
457 CERROR("Invalid index in lustre_cgf, offset 2\n");
461 spin_lock(&cm->cmm_tgt_guard);
462 list_for_each_entry_safe(mc, tmp, &cm->cmm_targets,
464 if (mc->mc_num == mdc_num) {
465 spin_unlock(&cm->cmm_tgt_guard);
469 spin_unlock(&cm->cmm_tgt_guard);
470 ld = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
474 ld->ld_site = cmm2lu_dev(cm)->ld_site;
476 rc = ldt->ldt_ops->ldto_device_init(env, ld, NULL, NULL);
478 ldt->ldt_ops->ldto_device_free(env, ld);
481 /* pass config to the just created MDC */
482 rc = ld->ld_ops->ldo_process_config(env, ld, cfg);
484 ldt->ldt_ops->ldto_device_fini(env, ld);
485 ldt->ldt_ops->ldto_device_free(env, ld);
489 spin_lock(&cm->cmm_tgt_guard);
490 list_for_each_entry_safe(mc, tmp, &cm->cmm_targets,
492 if (mc->mc_num == mdc_num) {
493 spin_unlock(&cm->cmm_tgt_guard);
494 ldt->ldt_ops->ldto_device_fini(env, ld);
495 ldt->ldt_ops->ldto_device_free(env, ld);
500 list_add_tail(&mc->mc_linkage, &cm->cmm_targets);
502 spin_unlock(&cm->cmm_tgt_guard);
504 lu_device_get(cmm_lu);
505 lu_ref_add(&cmm_lu->ld_reference, "mdc-child", ld);
507 target.ft_srv = NULL;
508 target.ft_idx = mc->mc_num;
509 target.ft_exp = mc->mc_desc.cl_exp;
510 fld_client_add_target(cm->cmm_fld, &target);
512 /* Set max md size for the mdc. */
513 rc = cmm_post_init_mdc(env, cm);
517 static void cmm_device_shutdown(const struct lu_env *env,
518 struct cmm_device *cm,
519 struct lustre_cfg *cfg)
521 struct mdc_device *mc, *tmp;
524 /* Remove local target from FLD. */
525 fld_client_del_target(cm->cmm_fld, cm->cmm_local_num);
527 /* Finish all mdc devices. */
528 spin_lock(&cm->cmm_tgt_guard);
529 list_for_each_entry_safe(mc, tmp, &cm->cmm_targets, mc_linkage) {
530 struct lu_device *ld_m = mdc2lu_dev(mc);
531 fld_client_del_target(cm->cmm_fld, mc->mc_num);
532 ld_m->ld_ops->ldo_process_config(env, ld_m, cfg);
534 spin_unlock(&cm->cmm_tgt_guard);
536 /* remove upcall device*/
537 md_upcall_fini(&cm->cmm_md_dev);
542 static int cmm_device_mount(const struct lu_env *env,
543 struct cmm_device *m, struct lustre_cfg *cfg)
545 const char *index = lustre_cfg_string(cfg, 2);
548 LASSERT(index != NULL);
550 m->cmm_local_num = simple_strtol(index, &p, 10);
552 CERROR("Invalid index in lustre_cgf\n");
559 static int cmm_process_config(const struct lu_env *env,
560 struct lu_device *d, struct lustre_cfg *cfg)
562 struct cmm_device *m = lu2cmm_dev(d);
563 struct lu_device *next = md2lu_dev(m->cmm_child);
567 switch(cfg->lcfg_command) {
569 /* On first ADD_MDC add also local target. */
570 if (!(m->cmm_flags & CMM_INITIALIZED)) {
571 struct lu_site *ls = cmm2lu_dev(m)->ld_site;
572 struct lu_fld_target target;
574 target.ft_srv = lu_site2md(ls)->ms_server_fld;
575 target.ft_idx = m->cmm_local_num;
576 target.ft_exp = NULL;
578 fld_client_add_target(m->cmm_fld, &target);
580 err = cmm_add_mdc(env, m, cfg);
582 /* The first ADD_MDC can be counted as setup is finished. */
583 if (!(m->cmm_flags & CMM_INITIALIZED))
584 m->cmm_flags |= CMM_INITIALIZED;
589 /* lower layers should be set up at first */
590 err = next->ld_ops->ldo_process_config(env, next, cfg);
592 err = cmm_device_mount(env, m, cfg);
597 cmm_device_shutdown(env, m, cfg);
600 err = next->ld_ops->ldo_process_config(env, next, cfg);
605 static int cmm_recovery_complete(const struct lu_env *env,
608 struct cmm_device *m = lu2cmm_dev(d);
609 struct lu_device *next = md2lu_dev(m->cmm_child);
612 rc = next->ld_ops->ldo_recovery_complete(env, next);
616 static const struct lu_device_operations cmm_lu_ops = {
617 .ldo_object_alloc = cmm_object_alloc,
618 .ldo_process_config = cmm_process_config,
619 .ldo_recovery_complete = cmm_recovery_complete
622 /* --- lu_device_type operations --- */
623 int cmm_upcall(const struct lu_env *env, struct md_device *md,
624 enum md_upcall_event ev)
631 rc = cmm_post_init_mdc(env, md2cmm_dev(md));
633 CERROR("can not init md size %d\n", rc);
636 rc = md_do_upcall(env, md, ev);
641 static struct lu_device *cmm_device_free(const struct lu_env *env,
644 struct cmm_device *m = lu2cmm_dev(d);
645 struct lu_device *next = md2lu_dev(m->cmm_child);
648 LASSERT(m->cmm_tgt_count == 0);
649 LASSERT(list_empty(&m->cmm_targets));
650 if (m->cmm_fld != NULL) {
651 OBD_FREE_PTR(m->cmm_fld);
654 md_device_fini(&m->cmm_md_dev);
659 static struct lu_device *cmm_device_alloc(const struct lu_env *env,
660 struct lu_device_type *t,
661 struct lustre_cfg *cfg)
664 struct cmm_device *m;
669 l = ERR_PTR(-ENOMEM);
671 md_device_init(&m->cmm_md_dev, t);
672 m->cmm_md_dev.md_ops = &cmm_md_ops;
673 md_upcall_init(&m->cmm_md_dev, cmm_upcall);
675 l->ld_ops = &cmm_lu_ops;
677 OBD_ALLOC_PTR(m->cmm_fld);
679 cmm_device_free(env, l);
680 l = ERR_PTR(-ENOMEM);
686 /* context key constructor/destructor: cmm_key_init, cmm_key_fini */
687 LU_KEY_INIT_FINI(cmm, struct cmm_thread_info);
689 /* context key: cmm_thread_key */
690 LU_CONTEXT_KEY_DEFINE(cmm, LCT_MD_THREAD);
692 struct cmm_thread_info *cmm_env_info(const struct lu_env *env)
694 struct cmm_thread_info *info;
696 info = lu_context_key_get(&env->le_ctx, &cmm_thread_key);
697 LASSERT(info != NULL);
701 /* type constructor/destructor: cmm_type_init/cmm_type_fini */
702 LU_TYPE_INIT_FINI(cmm, &cmm_thread_key);
704 static int cmm_device_init(const struct lu_env *env, struct lu_device *d,
705 const char *name, struct lu_device *next)
707 struct cmm_device *m = lu2cmm_dev(d);
712 spin_lock_init(&m->cmm_tgt_guard);
713 CFS_INIT_LIST_HEAD(&m->cmm_targets);
714 m->cmm_tgt_count = 0;
715 m->cmm_child = lu2md_dev(next);
717 err = fld_client_init(m->cmm_fld, name,
718 LUSTRE_CLI_FLD_HASH_DHT);
720 CERROR("Can't init FLD, err %d\n", err);
724 /* Assign site's fld client ref, needed for asserts in osd. */
725 ls = cmm2lu_dev(m)->ld_site;
726 lu_site2md(ls)->ms_client_fld = m->cmm_fld;
727 err = cmm_procfs_init(m, name);
732 static struct lu_device *cmm_device_fini(const struct lu_env *env,
733 struct lu_device *ld)
735 struct cmm_device *cm = lu2cmm_dev(ld);
736 struct mdc_device *mc, *tmp;
740 /* Finish all mdc devices */
741 spin_lock(&cm->cmm_tgt_guard);
742 list_for_each_entry_safe(mc, tmp, &cm->cmm_targets, mc_linkage) {
743 struct lu_device *ld_m = mdc2lu_dev(mc);
744 struct lu_device *ld_c = cmm2lu_dev(cm);
746 list_del_init(&mc->mc_linkage);
747 lu_ref_del(&ld_c->ld_reference, "mdc-child", ld_m);
749 ld_m->ld_type->ldt_ops->ldto_device_fini(env, ld_m);
750 ld_m->ld_type->ldt_ops->ldto_device_free(env, ld_m);
753 spin_unlock(&cm->cmm_tgt_guard);
755 fld_client_fini(cm->cmm_fld);
756 ls = cmm2lu_dev(cm)->ld_site;
757 lu_site2md(ls)->ms_client_fld = NULL;
760 RETURN (md2lu_dev(cm->cmm_child));
763 static struct lu_device_type_operations cmm_device_type_ops = {
764 .ldto_init = cmm_type_init,
765 .ldto_fini = cmm_type_fini,
767 .ldto_start = cmm_type_start,
768 .ldto_stop = cmm_type_stop,
770 .ldto_device_alloc = cmm_device_alloc,
771 .ldto_device_free = cmm_device_free,
773 .ldto_device_init = cmm_device_init,
774 .ldto_device_fini = cmm_device_fini
777 static struct lu_device_type cmm_device_type = {
778 .ldt_tags = LU_DEVICE_MD,
779 .ldt_name = LUSTRE_CMM_NAME,
780 .ldt_ops = &cmm_device_type_ops,
781 .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD
784 struct lprocfs_vars lprocfs_cmm_obd_vars[] = {
788 struct lprocfs_vars lprocfs_cmm_module_vars[] = {
792 static void lprocfs_cmm_init_vars(struct lprocfs_static_vars *lvars)
794 lvars->module_vars = lprocfs_cmm_module_vars;
795 lvars->obd_vars = lprocfs_cmm_obd_vars;
798 static int __init cmm_mod_init(void)
800 struct lprocfs_static_vars lvars;
802 lprocfs_cmm_init_vars(&lvars);
803 return class_register_type(&cmm_obd_device_ops, NULL, lvars.module_vars,
804 LUSTRE_CMM_NAME, &cmm_device_type);
807 static void __exit cmm_mod_exit(void)
809 class_unregister_type(LUSTRE_CMM_NAME);
812 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
813 MODULE_DESCRIPTION("Lustre Clustered Metadata Manager ("LUSTRE_CMM_NAME")");
814 MODULE_LICENSE("GPL");
816 cfs_module(cmm, "0.1.0", cmm_mod_init, cmm_mod_exit);