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_cleanup_desc_tgts(const struct lu_env *env,
136 struct lod_device *lod,
137 struct lod_tgt_descs *ltd,
138 struct lustre_cfg *lcfg)
140 struct lu_device *next;
145 if (ltd->ltd_tgts_size <= 0) {
146 lod_putref(lod, ltd);
149 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
150 struct lod_tgt_desc *tgt;
153 tgt = LTD_TGT(ltd, i);
154 LASSERT(tgt && tgt->ltd_tgt);
155 next = &tgt->ltd_tgt->dd_lu_dev;
156 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
158 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
159 ": rc = %d\n", lod2obd(lod)->obd_name, i,
160 lcfg->lcfg_command, rc1);
164 lod_putref(lod, ltd);
168 static int lod_process_config(const struct lu_env *env,
169 struct lu_device *dev,
170 struct lustre_cfg *lcfg)
172 struct lod_device *lod = lu2lod_dev(dev);
173 struct lu_device *next = &lod->lod_child->dd_lu_dev;
178 switch(lcfg->lcfg_command) {
179 case LCFG_LOV_DEL_OBD:
180 case LCFG_LOV_ADD_INA:
181 case LCFG_LOV_ADD_OBD: {
184 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1 */
185 arg1 = lustre_cfg_string(lcfg, 1);
187 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
188 GOTO(out, rc = -EINVAL);
189 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
190 GOTO(out, rc = -EINVAL);
192 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD)
193 rc = lod_add_device(env, lod, arg1, index, gen, 1);
194 else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA)
195 rc = lod_add_device(env, lod, arg1, index, gen, 0);
197 rc = lod_del_device(env, lod,
205 struct lprocfs_static_vars v = { 0 };
206 struct obd_device *obd = lod2obd(lod);
208 lprocfs_lod_init_vars(&v);
210 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
216 lu_dev_del_linkage(dev->ld_site, dev);
217 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
218 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
220 * do cleanup on underlying storage only when
221 * all OSPs are cleaned up, as they use that OSD as well
223 next = &lod->lod_child->dd_lu_dev;
224 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
226 CERROR("%s: can't process %u: %d\n",
227 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
229 rc = obd_disconnect(lod->lod_child_exp);
231 CERROR("error in disconnect from storage: %d\n", rc);
235 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
245 static int lod_recovery_complete(const struct lu_env *env,
246 struct lu_device *dev)
248 struct lod_device *lod = lu2lod_dev(dev);
249 struct lu_device *next = &lod->lod_child->dd_lu_dev;
253 LASSERT(lod->lod_recovery_completed == 0);
254 lod->lod_recovery_completed = 1;
256 rc = next->ld_ops->ldo_recovery_complete(env, next);
258 lod_getref(&lod->lod_ost_descs);
259 if (lod->lod_osts_size > 0) {
260 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
261 struct lod_tgt_desc *tgt;
262 tgt = OST_TGT(lod, i);
263 LASSERT(tgt && tgt->ltd_tgt);
264 next = &tgt->ltd_ost->dd_lu_dev;
265 rc = next->ld_ops->ldo_recovery_complete(env, next);
267 CERROR("%s: can't complete recovery on #%d:"
268 "%d\n", lod2obd(lod)->obd_name, i, rc);
271 lod_putref(lod, &lod->lod_ost_descs);
275 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
276 struct lu_device *cdev)
278 struct lod_device *lod = lu2lod_dev(cdev);
279 struct lu_device *next = &lod->lod_child->dd_lu_dev;
283 rc = next->ld_ops->ldo_prepare(env, pdev, next);
285 CERROR("%s: prepare bottom error: rc = %d\n",
286 lod2obd(lod)->obd_name, rc);
290 lod->lod_initialized = 1;
295 const struct lu_device_operations lod_lu_ops = {
296 .ldo_object_alloc = lod_object_alloc,
297 .ldo_process_config = lod_process_config,
298 .ldo_recovery_complete = lod_recovery_complete,
299 .ldo_prepare = lod_prepare,
302 static int lod_root_get(const struct lu_env *env,
303 struct dt_device *dev, struct lu_fid *f)
305 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
308 static int lod_statfs(const struct lu_env *env,
309 struct dt_device *dev, struct obd_statfs *sfs)
311 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
314 static struct thandle *lod_trans_create(const struct lu_env *env,
315 struct dt_device *dev)
317 return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
320 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
323 return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
326 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
328 /* XXX: we don't know next device, will be fixed with DNE */
329 return dt_trans_stop(env, th->th_dev, th);
332 static void lod_conf_get(const struct lu_env *env,
333 const struct dt_device *dev,
334 struct dt_device_param *param)
336 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
339 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
341 struct lod_device *lod = dt2lod_dev(dev);
342 struct lod_ost_desc *ost;
346 lod_getref(&lod->lod_ost_descs);
347 lod_foreach_ost(lod, i) {
348 ost = OST_TGT(lod, i);
349 LASSERT(ost && ost->ltd_ost);
350 rc = dt_sync(env, ost->ltd_ost);
352 CERROR("%s: can't sync %u: %d\n",
353 lod2obd(lod)->obd_name, i, rc);
357 lod_putref(lod, &lod->lod_ost_descs);
359 rc = dt_sync(env, lod->lod_child);
364 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
366 return dt_ro(env, dt2lod_dev(dev)->lod_child);
369 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
371 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
374 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
375 int mode, unsigned long timeout,
376 __u32 alg, struct lustre_capa_key *keys)
378 struct dt_device *next = dt2lod_dev(dev)->lod_child;
379 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
382 static const struct dt_device_operations lod_dt_ops = {
383 .dt_root_get = lod_root_get,
384 .dt_statfs = lod_statfs,
385 .dt_trans_create = lod_trans_create,
386 .dt_trans_start = lod_trans_start,
387 .dt_trans_stop = lod_trans_stop,
388 .dt_conf_get = lod_conf_get,
391 .dt_commit_async = lod_commit_async,
392 .dt_init_capa_ctxt = lod_init_capa_ctxt,
395 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
396 struct lustre_cfg *cfg)
398 struct obd_connect_data *data = NULL;
399 struct obd_device *obd;
400 char *nextdev = NULL, *p, *s;
405 LASSERT(lod->lod_child_exp == NULL);
407 /* compatibility hack: we still use old config logs
408 * which specify LOV, but we need to learn underlying
409 * OSD device, which is supposed to be:
410 * <fsname>-MDTxxxx-osd
412 * 2.x MGS generates lines like the following:
413 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
414 * 1.8 MGS generates lines like the following:
415 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
417 * we use "-MDT" to differentiate 2.x from 1.8 */
419 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
421 OBD_ALLOC(nextdev, len);
423 GOTO(out, rc = -ENOMEM);
426 s = strstr(nextdev, "-mdtlov");
427 if (unlikely(s == NULL)) {
428 CERROR("unable to parse device name %s\n",
429 lustre_cfg_string(cfg, 0));
430 GOTO(out, rc = -EINVAL);
433 if (strstr(nextdev, "-MDT")) {
438 strcpy(s, "-MDT0000-osd");
441 CERROR("unable to parse device name %s\n",
442 lustre_cfg_string(cfg, 0));
443 GOTO(out, rc = -EINVAL);
448 GOTO(out, rc = -ENOMEM);
450 obd = class_name2obd(nextdev);
452 CERROR("can not locate next device: %s\n", nextdev);
453 GOTO(out, rc = -ENOTCONN);
456 data->ocd_connect_flags = OBD_CONNECT_VERSION;
457 data->ocd_version = LUSTRE_VERSION_CODE;
459 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
462 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
466 lod->lod_dt_dev.dd_lu_dev.ld_site =
467 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
468 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
469 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
475 OBD_FREE(nextdev, len);
479 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
481 mutex_init(<d->ltd_mutex);
482 init_rwsem(<d->ltd_rw_sem);
484 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
485 * added to the LOD, see lod_add_device() */
486 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
487 if (ltd->ltd_tgt_bitmap == NULL)
490 ltd->ltd_tgts_size = 32;
493 ltd->ltd_death_row = 0;
494 ltd->ltd_refcount = 0;
498 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
499 struct lu_device_type *ldt, struct lustre_cfg *cfg)
501 struct dt_device_param ddp;
502 struct obd_device *obd;
506 obd = class_name2obd(lustre_cfg_string(cfg, 0));
508 CERROR("Cannot find obd with name %s\n",
509 lustre_cfg_string(cfg, 0));
513 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
514 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
515 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
516 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
518 rc = lod_connect_to_osd(env, lod, cfg);
522 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
523 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
525 /* setup obd to be used with old lov code */
526 rc = lod_pools_init(lod, cfg);
528 GOTO(out_disconnect, rc);
530 rc = lod_procfs_init(lod);
534 spin_lock_init(&lod->lod_desc_lock);
535 spin_lock_init(&lod->lod_connects_lock);
536 lod_tgt_desc_init(&lod->lod_mdt_descs);
537 lod_tgt_desc_init(&lod->lod_ost_descs);
544 obd_disconnect(lod->lod_child_exp);
548 static struct lu_device *lod_device_free(const struct lu_env *env,
549 struct lu_device *lu)
551 struct lod_device *lod = lu2lod_dev(lu);
552 struct lu_device *next = &lod->lod_child->dd_lu_dev;
555 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
556 dt_device_fini(&lod->lod_dt_dev);
561 static struct lu_device *lod_device_alloc(const struct lu_env *env,
562 struct lu_device_type *type,
563 struct lustre_cfg *lcfg)
565 struct lod_device *lod;
566 struct lu_device *lu_dev;
570 lu_dev = ERR_PTR(-ENOMEM);
574 lu_dev = lod2lu_dev(lod);
575 dt_device_init(&lod->lod_dt_dev, type);
576 rc = lod_init0(env, lod, type, lcfg);
578 lod_device_free(env, lu_dev);
579 lu_dev = ERR_PTR(rc);
586 static struct lu_device *lod_device_fini(const struct lu_env *env,
589 struct lod_device *lod = lu2lod_dev(d);
595 lod_procfs_fini(lod);
597 rc = lod_fini_tgt(lod, &lod->lod_ost_descs);
599 CERROR("%s:can not fini ost descs %d\n",
600 lod2obd(lod)->obd_name, rc);
602 rc = lod_fini_tgt(lod, &lod->lod_mdt_descs);
604 CERROR("%s:can not fini mdt descs %d\n",
605 lod2obd(lod)->obd_name, rc);
611 * we use exports to track all LOD users
613 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
614 struct obd_device *obd, struct obd_uuid *cluuid,
615 struct obd_connect_data *data, void *localdata)
617 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
618 struct lustre_handle conn;
622 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
624 rc = class_connect(&conn, obd, cluuid);
628 *exp = class_conn2export(&conn);
630 spin_lock(&lod->lod_connects_lock);
632 /* at the moment we expect the only user */
633 LASSERT(lod->lod_connects == 1);
634 spin_unlock(&lod->lod_connects_lock);
640 * once last export (we don't count self-export) disappeared
641 * lod can be released
643 static int lod_obd_disconnect(struct obd_export *exp)
645 struct obd_device *obd = exp->exp_obd;
646 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
650 /* Only disconnect the underlying layers on the final disconnect. */
651 spin_lock(&lod->lod_connects_lock);
653 if (lod->lod_connects != 0) {
654 /* why should there be more than 1 connect? */
655 spin_unlock(&lod->lod_connects_lock);
656 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
660 spin_unlock(&lod->lod_connects_lock);
662 /* the last user of lod has gone, let's release the device */
666 rc = class_disconnect(exp); /* bz 9811 */
668 if (rc == 0 && release)
669 class_manual_cleanup(obd);
673 LU_KEY_INIT(lod, struct lod_thread_info);
675 static void lod_key_fini(const struct lu_context *ctx,
676 struct lu_context_key *key, void *data)
678 struct lod_thread_info *info = data;
679 /* allocated in lod_get_lov_ea
680 * XXX: this is overload, a tread may have such store but used only
681 * once. Probably better would be pool of such stores per LOD.
683 if (info->lti_ea_store) {
684 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
685 info->lti_ea_store = NULL;
686 info->lti_ea_store_size = 0;
691 /* context key: lod_thread_key */
692 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
694 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
696 static struct lu_device_type_operations lod_device_type_ops = {
697 .ldto_init = lod_type_init,
698 .ldto_fini = lod_type_fini,
700 .ldto_start = lod_type_start,
701 .ldto_stop = lod_type_stop,
703 .ldto_device_alloc = lod_device_alloc,
704 .ldto_device_free = lod_device_free,
706 .ldto_device_fini = lod_device_fini
709 static struct lu_device_type lod_device_type = {
710 .ldt_tags = LU_DEVICE_DT,
711 .ldt_name = LUSTRE_LOD_NAME,
712 .ldt_ops = &lod_device_type_ops,
713 .ldt_ctx_tags = LCT_MD_THREAD,
716 static int lod_obd_health_check(const struct lu_env *env,
717 struct obd_device *obd)
719 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
720 struct lod_ost_desc *ost;
725 lod_getref(&d->lod_ost_descs);
726 lod_foreach_ost(d, i) {
728 LASSERT(ost && ost->ltd_ost);
729 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
730 /* one healthy device is enough */
734 lod_putref(d, &d->lod_ost_descs);
738 static struct obd_ops lod_obd_device_ops = {
739 .o_owner = THIS_MODULE,
740 .o_connect = lod_obd_connect,
741 .o_disconnect = lod_obd_disconnect,
742 .o_health_check = lod_obd_health_check,
743 .o_pool_new = lod_pool_new,
744 .o_pool_rem = lod_pool_remove,
745 .o_pool_add = lod_pool_add,
746 .o_pool_del = lod_pool_del,
749 static int __init lod_mod_init(void)
751 struct lprocfs_static_vars lvars = { 0 };
752 cfs_proc_dir_entry_t *lov_proc_dir;
755 rc = lu_kmem_init(lod_caches);
759 lprocfs_lod_init_vars(&lvars);
761 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
762 LUSTRE_LOD_NAME, &lod_device_type);
764 lu_kmem_fini(lod_caches);
768 /* create "lov" entry in procfs for compatibility purposes */
769 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
770 if (lov_proc_dir == NULL) {
771 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
773 if (IS_ERR(lov_proc_dir))
774 CERROR("lod: can't create compat entry \"lov\": %d\n",
775 (int)PTR_ERR(lov_proc_dir));
781 static void __exit lod_mod_exit(void)
784 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
786 class_unregister_type(LUSTRE_LOD_NAME);
787 lu_kmem_fini(lod_caches);
790 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
791 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
792 MODULE_LICENSE("GPL");
794 module_init(lod_mod_init);
795 module_exit(lod_mod_exit);