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) 2012, Intel Corporation.
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"
52 * Lookup MDT/OST index \a tgt by FID \a fid.
54 * \param lod LOD to be lookup at.
55 * \param fid FID of object to find MDT/OST.
56 * \param tgt MDT/OST index to return.
57 * \param flags indidcate the FID is on MDS or OST.
59 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
60 const struct lu_fid *fid, __u32 *tgt, int flags)
62 struct lu_seq_range range;
63 struct lu_server_fld *server_fld;
67 LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
68 if (fid_is_idif(fid)) {
69 *tgt = fid_idif_ost_idx(fid);
73 if (!lod->lod_initialized || !fid_is_norm(fid)) {
74 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
75 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
79 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
80 range.lsr_flags = flags;
81 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
83 CERROR("%s: Can't find tgt by seq "LPX64", rc %d\n",
84 lod2obd(lod)->obd_name, fid_seq(fid), rc);
88 *tgt = range.lsr_index;
90 CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
91 LPX64"\n", *tgt, fid_seq(fid));
96 extern struct lu_object_operations lod_lu_obj_ops;
97 extern struct dt_object_operations lod_obj_ops;
99 /* Slab for OSD object allocation */
100 cfs_mem_cache_t *lod_object_kmem;
102 static struct lu_kmem_descr lod_caches[] = {
104 .ckd_cache = &lod_object_kmem,
105 .ckd_name = "lod_obj",
106 .ckd_size = sizeof(struct lod_object)
113 static struct lu_device *lod_device_fini(const struct lu_env *env,
114 struct lu_device *d);
116 struct lu_object *lod_object_alloc(const struct lu_env *env,
117 const struct lu_object_header *hdr,
118 struct lu_device *dev)
120 struct lu_object *lu_obj;
121 struct lod_object *lo;
123 OBD_SLAB_ALLOC_PTR_GFP(lo, lod_object_kmem, CFS_ALLOC_IO);
127 lu_obj = lod2lu_obj(lo);
128 dt_object_init(&lo->ldo_obj, NULL, dev);
129 lo->ldo_obj.do_ops = &lod_obj_ops;
130 lu_obj->lo_ops = &lod_lu_obj_ops;
135 static int lod_process_config(const struct lu_env *env,
136 struct lu_device *dev,
137 struct lustre_cfg *lcfg)
139 struct lod_device *lod = lu2lod_dev(dev);
140 struct lu_device *next = &lod->lod_child->dd_lu_dev;
145 switch(lcfg->lcfg_command) {
147 case LCFG_LOV_DEL_OBD:
148 case LCFG_LOV_ADD_INA:
149 case LCFG_LOV_ADD_OBD: {
152 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1 */
153 arg1 = lustre_cfg_string(lcfg, 1);
155 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
156 GOTO(out, rc = -EINVAL);
157 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
158 GOTO(out, rc = -EINVAL);
160 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD)
161 rc = lod_add_device(env, lod, arg1, index, gen, 1);
162 else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA)
163 rc = lod_add_device(env, lod, arg1, index, gen, 0);
165 rc = lod_del_device(env, lod, arg1, index, gen);
171 struct lprocfs_static_vars v = { 0 };
172 struct obd_device *obd = lod2obd(lod);
174 lprocfs_lod_init_vars(&v);
176 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
183 lu_dev_del_linkage(dev->ld_site, dev);
185 lod_foreach_ost(lod, i) {
186 struct lod_ost_desc *ost;
187 ost = OST_TGT(lod, i);
188 LASSERT(ost && ost->ltd_ost);
189 next = &ost->ltd_ost->dd_lu_dev;
190 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
192 CERROR("%s: can't process %u: %d\n",
193 lod2obd(lod)->obd_name,
194 lcfg->lcfg_command, rc);
199 * do cleanup on underlying storage only when
200 * all OSPs are cleaned up, as they use that OSD as well
202 next = &lod->lod_child->dd_lu_dev;
203 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
205 CERROR("%s: can't process %u: %d\n",
206 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
208 rc = obd_disconnect(lod->lod_child_exp);
210 CERROR("error in disconnect from storage: %d\n", rc);
214 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
224 static int lod_recovery_complete(const struct lu_env *env,
225 struct lu_device *dev)
227 struct lod_device *lod = lu2lod_dev(dev);
228 struct lu_device *next = &lod->lod_child->dd_lu_dev;
229 struct lod_ost_desc *ost;
233 LASSERT(lod->lod_recovery_completed == 0);
234 lod->lod_recovery_completed = 1;
236 rc = next->ld_ops->ldo_recovery_complete(env, next);
239 lod_foreach_ost(lod, i) {
240 ost = OST_TGT(lod, i);
241 LASSERT(ost && ost->ltd_ost);
242 next = &ost->ltd_ost->dd_lu_dev;
243 rc = next->ld_ops->ldo_recovery_complete(env, next);
245 CERROR("%s: can't complete recovery on #%d: %d\n",
246 lod2obd(lod)->obd_name, i, rc);
253 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
254 struct lu_device *cdev)
256 struct lod_device *lod = lu2lod_dev(cdev);
257 struct lu_device *next = &lod->lod_child->dd_lu_dev;
261 rc = next->ld_ops->ldo_prepare(env, pdev, next);
263 CERROR("%s: prepare bottom error: rc = %d\n",
264 lod2obd(lod)->obd_name, rc);
268 lod->lod_initialized = 1;
273 const struct lu_device_operations lod_lu_ops = {
274 .ldo_object_alloc = lod_object_alloc,
275 .ldo_process_config = lod_process_config,
276 .ldo_recovery_complete = lod_recovery_complete,
277 .ldo_prepare = lod_prepare,
280 static int lod_root_get(const struct lu_env *env,
281 struct dt_device *dev, struct lu_fid *f)
283 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
286 static int lod_statfs(const struct lu_env *env,
287 struct dt_device *dev, struct obd_statfs *sfs)
289 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
292 static struct thandle *lod_trans_create(const struct lu_env *env,
293 struct dt_device *dev)
295 return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
298 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
301 return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
304 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
306 /* XXX: we don't know next device, will be fixed with DNE */
307 return dt_trans_stop(env, th->th_dev, th);
310 static void lod_conf_get(const struct lu_env *env,
311 const struct dt_device *dev,
312 struct dt_device_param *param)
314 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
317 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
319 struct lod_device *lod = dt2lod_dev(dev);
320 struct lod_ost_desc *ost;
325 lod_foreach_ost(lod, i) {
326 ost = OST_TGT(lod, i);
327 LASSERT(ost && ost->ltd_ost);
328 rc = dt_sync(env, ost->ltd_ost);
330 CERROR("%s: can't sync %u: %d\n",
331 lod2obd(lod)->obd_name, i, rc);
337 rc = dt_sync(env, lod->lod_child);
342 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
344 return dt_ro(env, dt2lod_dev(dev)->lod_child);
347 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
349 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
352 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
353 int mode, unsigned long timeout,
354 __u32 alg, struct lustre_capa_key *keys)
356 struct dt_device *next = dt2lod_dev(dev)->lod_child;
357 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
360 static const struct dt_device_operations lod_dt_ops = {
361 .dt_root_get = lod_root_get,
362 .dt_statfs = lod_statfs,
363 .dt_trans_create = lod_trans_create,
364 .dt_trans_start = lod_trans_start,
365 .dt_trans_stop = lod_trans_stop,
366 .dt_conf_get = lod_conf_get,
369 .dt_commit_async = lod_commit_async,
370 .dt_init_capa_ctxt = lod_init_capa_ctxt,
373 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
374 struct lustre_cfg *cfg)
376 struct obd_connect_data *data = NULL;
377 struct obd_device *obd;
378 char *nextdev = NULL, *p, *s;
383 LASSERT(lod->lod_child_exp == NULL);
385 /* compatibility hack: we still use old config logs
386 * which specify LOV, but we need to learn underlying
387 * OSD device, which is supposed to be:
388 * <fsname>-MDTxxxx-osd
390 * 2.x MGS generates lines like the following:
391 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
392 * 1.8 MGS generates lines like the following:
393 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
395 * we use "-MDT" to differentiate 2.x from 1.8 */
397 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
399 OBD_ALLOC(nextdev, len);
401 GOTO(out, rc = -ENOMEM);
404 s = strstr(nextdev, "-mdtlov");
405 if (unlikely(s == NULL)) {
406 CERROR("unable to parse device name %s\n",
407 lustre_cfg_string(cfg, 0));
408 GOTO(out, rc = -EINVAL);
411 if (strstr(nextdev, "-MDT")) {
416 strcpy(s, "-MDT0000-osd");
419 CERROR("unable to parse device name %s\n",
420 lustre_cfg_string(cfg, 0));
421 GOTO(out, rc = -EINVAL);
426 GOTO(out, rc = -ENOMEM);
428 obd = class_name2obd(nextdev);
430 CERROR("can not locate next device: %s\n", nextdev);
431 GOTO(out, rc = -ENOTCONN);
434 data->ocd_connect_flags = OBD_CONNECT_VERSION;
435 data->ocd_version = LUSTRE_VERSION_CODE;
437 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
440 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
444 lod->lod_dt_dev.dd_lu_dev.ld_site =
445 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
446 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
447 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
453 OBD_FREE(nextdev, len);
457 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
458 struct lu_device_type *ldt, struct lustre_cfg *cfg)
460 struct dt_device_param ddp;
461 struct obd_device *obd;
465 obd = class_name2obd(lustre_cfg_string(cfg, 0));
467 CERROR("Cannot find obd with name %s\n",
468 lustre_cfg_string(cfg, 0));
472 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
473 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
474 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
475 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
477 rc = lod_connect_to_osd(env, lod, cfg);
481 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
482 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
484 /* setup obd to be used with old lov code */
485 rc = lod_pools_init(lod, cfg);
487 GOTO(out_disconnect, rc);
489 rc = lod_procfs_init(lod);
493 mutex_init(&lod->lod_mutex);
494 init_rwsem(&lod->lod_rw_sem);
495 spin_lock_init(&lod->lod_desc_lock);
502 obd_disconnect(lod->lod_child_exp);
506 static struct lu_device *lod_device_free(const struct lu_env *env,
507 struct lu_device *lu)
509 struct lod_device *lod = lu2lod_dev(lu);
510 struct lu_device *next = &lod->lod_child->dd_lu_dev;
513 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
514 dt_device_fini(&lod->lod_dt_dev);
519 static struct lu_device *lod_device_alloc(const struct lu_env *env,
520 struct lu_device_type *type,
521 struct lustre_cfg *lcfg)
523 struct lod_device *lod;
524 struct lu_device *lu_dev;
528 lu_dev = ERR_PTR(-ENOMEM);
532 lu_dev = lod2lu_dev(lod);
533 dt_device_init(&lod->lod_dt_dev, type);
534 rc = lod_init0(env, lod, type, lcfg);
536 lod_device_free(env, lu_dev);
537 lu_dev = ERR_PTR(rc);
544 static struct lu_device *lod_device_fini(const struct lu_env *env,
547 struct lod_device *lod = lu2lod_dev(d);
552 lod_procfs_fini(lod);
558 * we use exports to track all LOD users
560 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
561 struct obd_device *obd, struct obd_uuid *cluuid,
562 struct obd_connect_data *data, void *localdata)
564 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
565 struct lustre_handle conn;
569 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
571 rc = class_connect(&conn, obd, cluuid);
575 *exp = class_conn2export(&conn);
577 mutex_lock(&lod->lod_mutex);
579 /* at the moment we expect the only user */
580 LASSERT(lod->lod_connects == 1);
581 mutex_unlock(&lod->lod_mutex);
587 * once last export (we don't count self-export) disappeared
588 * lod can be released
590 static int lod_obd_disconnect(struct obd_export *exp)
592 struct obd_device *obd = exp->exp_obd;
593 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
597 /* Only disconnect the underlying layers on the final disconnect. */
598 mutex_lock(&lod->lod_mutex);
600 if (lod->lod_connects != 0) {
601 /* why should there be more than 1 connect? */
602 mutex_unlock(&lod->lod_mutex);
603 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
607 mutex_unlock(&lod->lod_mutex);
609 /* the last user of lod has gone, let's release the device */
613 rc = class_disconnect(exp); /* bz 9811 */
615 if (rc == 0 && release)
616 class_manual_cleanup(obd);
620 LU_KEY_INIT(lod, struct lod_thread_info);
622 static void lod_key_fini(const struct lu_context *ctx,
623 struct lu_context_key *key, void *data)
625 struct lod_thread_info *info = data;
626 /* allocated in lod_get_lov_ea
627 * XXX: this is overload, a tread may have such store but used only
628 * once. Probably better would be pool of such stores per LOD.
630 if (info->lti_ea_store) {
631 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
632 info->lti_ea_store = NULL;
633 info->lti_ea_store_size = 0;
638 /* context key: lod_thread_key */
639 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
641 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
643 static struct lu_device_type_operations lod_device_type_ops = {
644 .ldto_init = lod_type_init,
645 .ldto_fini = lod_type_fini,
647 .ldto_start = lod_type_start,
648 .ldto_stop = lod_type_stop,
650 .ldto_device_alloc = lod_device_alloc,
651 .ldto_device_free = lod_device_free,
653 .ldto_device_fini = lod_device_fini
656 static struct lu_device_type lod_device_type = {
657 .ldt_tags = LU_DEVICE_DT,
658 .ldt_name = LUSTRE_LOD_NAME,
659 .ldt_ops = &lod_device_type_ops,
660 .ldt_ctx_tags = LCT_MD_THREAD,
663 static int lod_obd_health_check(const struct lu_env *env,
664 struct obd_device *obd)
666 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
667 struct lod_ost_desc *ost;
673 lod_foreach_ost(d, i) {
675 LASSERT(ost && ost->ltd_ost);
676 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
677 /* one healthy device is enough */
685 static struct obd_ops lod_obd_device_ops = {
686 .o_owner = THIS_MODULE,
687 .o_connect = lod_obd_connect,
688 .o_disconnect = lod_obd_disconnect,
689 .o_health_check = lod_obd_health_check,
690 .o_pool_new = lod_pool_new,
691 .o_pool_rem = lod_pool_remove,
692 .o_pool_add = lod_pool_add,
693 .o_pool_del = lod_pool_del,
696 static int __init lod_mod_init(void)
698 struct lprocfs_static_vars lvars = { 0 };
699 cfs_proc_dir_entry_t *lov_proc_dir;
702 rc = lu_kmem_init(lod_caches);
706 lprocfs_lod_init_vars(&lvars);
708 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
709 LUSTRE_LOD_NAME, &lod_device_type);
711 lu_kmem_fini(lod_caches);
715 /* create "lov" entry in procfs for compatibility purposes */
716 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
717 if (lov_proc_dir == NULL) {
718 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
720 if (IS_ERR(lov_proc_dir))
721 CERROR("lod: can't create compat entry \"lov\": %d\n",
722 (int)PTR_ERR(lov_proc_dir));
728 static void __exit lod_mod_exit(void)
731 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
733 class_unregister_type(LUSTRE_LOD_NAME);
734 lu_kmem_fini(lod_caches);
737 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
738 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
739 MODULE_LICENSE("GPL");
741 module_init(lod_mod_init);
742 module_exit(lod_mod_exit);