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, 2013, 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_fid.h>
48 #include <lustre_param.h>
49 #include <lustre_update.h>
51 #include "lod_internal.h"
54 * Lookup MDT/OST index \a tgt by FID \a fid.
56 * \param lod LOD to be lookup at.
57 * \param fid FID of object to find MDT/OST.
58 * \param tgt MDT/OST index to return.
59 * \param type indidcate the FID is on MDS or OST.
61 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
62 const struct lu_fid *fid, __u32 *tgt, int type)
64 struct lu_seq_range range = { 0 };
65 struct lu_server_fld *server_fld;
69 LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
70 if (fid_is_idif(fid)) {
71 *tgt = fid_idif_ost_idx(fid);
75 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
76 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
77 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
81 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
82 fld_range_set_type(&range, type);
83 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
87 *tgt = range.lsr_index;
89 CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
90 LPX64"\n", *tgt, fid_seq(fid));
95 extern struct lu_object_operations lod_lu_obj_ops;
96 extern struct lu_object_operations lod_lu_robj_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 lod_object *lod_obj;
121 struct lu_object *lu_obj;
122 const struct lu_fid *fid = &hdr->loh_fid;
127 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, CFS_ALLOC_IO);
129 RETURN(ERR_PTR(-ENOMEM));
131 rc = lod_fld_lookup(env, lu2lod_dev(dev), fid, &mds, LU_SEQ_RANGE_MDT);
133 OBD_SLAB_FREE_PTR(lod_obj, lod_object_kmem);
137 lod_obj->ldo_mds_num = mds;
138 lu_obj = lod2lu_obj(lod_obj);
139 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
140 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
141 if (likely(mds == lu_site2seq(dev->ld_site)->ss_node_id))
142 lu_obj->lo_ops = &lod_lu_obj_ops;
144 lu_obj->lo_ops = &lod_lu_robj_ops;
148 static int lod_cleanup_desc_tgts(const struct lu_env *env,
149 struct lod_device *lod,
150 struct lod_tgt_descs *ltd,
151 struct lustre_cfg *lcfg)
153 struct lu_device *next;
158 if (ltd->ltd_tgts_size <= 0) {
159 lod_putref(lod, ltd);
162 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
163 struct lod_tgt_desc *tgt;
166 tgt = LTD_TGT(ltd, i);
167 LASSERT(tgt && tgt->ltd_tgt);
168 next = &tgt->ltd_tgt->dd_lu_dev;
169 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
171 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
172 ": rc = %d\n", lod2obd(lod)->obd_name, i,
173 lcfg->lcfg_command, rc1);
177 lod_putref(lod, ltd);
181 static int lodname2mdt_index(char *lodname, int *index)
185 /* The lodname suppose to be fsname-MDTxxxx-mdtlov */
186 ptr = strrchr(lodname, '-');
188 CERROR("invalid MDT index in '%s'\n", lodname);
192 if (strncmp(ptr, "-mdtlov", 7) != 0) {
193 CERROR("invalid MDT index in '%s'\n", lodname);
197 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
198 CERROR("invalid MDT index in '%s'\n", lodname);
202 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
203 CERROR("invalid MDT index in '%s'\n", lodname);
207 *index = simple_strtol(ptr - 4, &tmp, 16);
208 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
209 CERROR("invalid MDT index in '%s'\n", lodname);
216 * Init client sequence manager which is used by local MDS to talk to sequence
217 * controller on remote node.
219 static int lod_seq_init_cli(const struct lu_env *env,
220 struct lod_device *lod,
221 char *tgtuuid, int index)
223 struct seq_server_site *ss;
224 struct obd_device *osp;
227 struct obd_uuid obd_uuid;
230 ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
233 /* check if this is adding the first MDC and controller is not yet
235 if (index != 0 || ss->ss_client_seq)
238 obd_str2uuid(&obd_uuid, tgtuuid);
239 osp = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
240 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
242 CERROR("%s: can't find %s device\n",
243 lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_name,
248 if (!osp->obd_set_up) {
249 CERROR("target %s not set up\n", osp->obd_name);
253 LASSERT(ss->ss_control_exp);
254 OBD_ALLOC_PTR(ss->ss_client_seq);
255 if (ss->ss_client_seq == NULL)
258 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
260 OBD_FREE_PTR(ss->ss_client_seq);
261 ss->ss_client_seq = NULL;
265 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", osp->obd_name);
266 rc = seq_client_init(ss->ss_client_seq, ss->ss_control_exp,
267 LUSTRE_SEQ_METADATA, prefix, NULL);
268 OBD_FREE(prefix, MAX_OBD_NAME + 5);
270 OBD_FREE_PTR(ss->ss_client_seq);
271 ss->ss_client_seq = NULL;
275 LASSERT(ss->ss_server_seq != NULL);
276 rc = seq_server_set_cli(ss->ss_server_seq, ss->ss_client_seq,
282 static void lod_seq_fini_cli(struct lod_device *lod)
284 struct seq_server_site *ss;
288 ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
294 if (ss->ss_server_seq)
295 seq_server_set_cli(ss->ss_server_seq,
298 if (ss->ss_control_exp) {
299 class_export_put(ss->ss_control_exp);
300 ss->ss_control_exp = NULL;
308 * Procss config log on LOD
309 * \param env environment info
310 * \param dev lod device
311 * \param lcfg config log
313 * Add osc config log,
314 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
315 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
316 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
317 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
318 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
319 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
322 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
323 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
324 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
325 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
326 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
327 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
329 static int lod_process_config(const struct lu_env *env,
330 struct lu_device *dev,
331 struct lustre_cfg *lcfg)
333 struct lod_device *lod = lu2lod_dev(dev);
334 struct lu_device *next = &lod->lod_child->dd_lu_dev;
339 switch(lcfg->lcfg_command) {
340 case LCFG_LOV_DEL_OBD:
341 case LCFG_LOV_ADD_INA:
342 case LCFG_LOV_ADD_OBD:
347 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
348 * modify_mdc_tgts add 0:lustre-MDT0001
349 * 1:lustre-MDT0001-mdc0002
351 arg1 = lustre_cfg_string(lcfg, 1);
353 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
354 GOTO(out, rc = -EINVAL);
355 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
356 GOTO(out, rc = -EINVAL);
358 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
360 mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
361 /* 1.8 configs don't have "-MDT0000" at the end */
365 rc = lodname2mdt_index(
366 lustre_cfg_string(lcfg, 0), &mdt_index);
370 rc = lod_add_device(env, lod, arg1, index, gen,
371 mdt_index, LUSTRE_OSC_NAME, 1);
372 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
374 rc = lod_add_device(env, lod, arg1, index, gen,
375 mdt_index, LUSTRE_MDC_NAME, 1);
377 rc = lod_seq_init_cli(env, lod, arg1,
379 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
380 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
382 rc = lod_add_device(env, lod, arg1, index, gen,
383 mdt_index, LUSTRE_OSC_NAME, 0);
385 rc = lod_del_device(env, lod,
394 struct lprocfs_static_vars v = { 0 };
395 struct obd_device *obd = lod2obd(lod);
397 lprocfs_lod_init_vars(&v);
399 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
405 case LCFG_PRE_CLEANUP: {
406 lu_dev_del_linkage(dev->ld_site, dev);
407 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
408 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
410 lod_seq_fini_cli(lod);
412 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
415 * do cleanup on underlying storage only when
416 * all OSPs are cleaned up, as they use that OSD as well
418 next = &lod->lod_child->dd_lu_dev;
419 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
421 CERROR("%s: can't process %u: %d\n",
422 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
424 rc = obd_disconnect(lod->lod_child_exp);
426 CERROR("error in disconnect from storage: %d\n", rc);
430 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
440 static int lod_recovery_complete(const struct lu_env *env,
441 struct lu_device *dev)
443 struct lod_device *lod = lu2lod_dev(dev);
444 struct lu_device *next = &lod->lod_child->dd_lu_dev;
448 LASSERT(lod->lod_recovery_completed == 0);
449 lod->lod_recovery_completed = 1;
451 rc = next->ld_ops->ldo_recovery_complete(env, next);
453 lod_getref(&lod->lod_ost_descs);
454 if (lod->lod_osts_size > 0) {
455 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
456 struct lod_tgt_desc *tgt;
457 tgt = OST_TGT(lod, i);
458 LASSERT(tgt && tgt->ltd_tgt);
459 next = &tgt->ltd_ost->dd_lu_dev;
460 rc = next->ld_ops->ldo_recovery_complete(env, next);
462 CERROR("%s: can't complete recovery on #%d:"
463 "%d\n", lod2obd(lod)->obd_name, i, rc);
466 lod_putref(lod, &lod->lod_ost_descs);
470 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
471 struct lu_device *cdev)
473 struct lod_device *lod = lu2lod_dev(cdev);
474 struct lu_device *next = &lod->lod_child->dd_lu_dev;
478 rc = next->ld_ops->ldo_prepare(env, pdev, next);
480 CERROR("%s: prepare bottom error: rc = %d\n",
481 lod2obd(lod)->obd_name, rc);
485 lod->lod_initialized = 1;
490 const struct lu_device_operations lod_lu_ops = {
491 .ldo_object_alloc = lod_object_alloc,
492 .ldo_process_config = lod_process_config,
493 .ldo_recovery_complete = lod_recovery_complete,
494 .ldo_prepare = lod_prepare,
497 static int lod_root_get(const struct lu_env *env,
498 struct dt_device *dev, struct lu_fid *f)
500 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
503 static int lod_statfs(const struct lu_env *env,
504 struct dt_device *dev, struct obd_statfs *sfs)
506 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
509 static struct thandle *lod_trans_create(const struct lu_env *env,
510 struct dt_device *dev)
514 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
518 CFS_INIT_LIST_HEAD(&th->th_remote_update_list);
522 static int lod_remote_sync(const struct lu_env *env, struct dt_device *dev,
525 struct update_request *update;
529 if (cfs_list_empty(&th->th_remote_update_list))
532 cfs_list_for_each_entry(update, &th->th_remote_update_list,
534 /* In DNE phase I, there should be only one OSP
535 * here, so we will do send/receive one by one,
536 * instead of sending them parallel, will fix this
538 th->th_current_request = update;
539 rc = dt_trans_start(env, update->ur_dt, th);
541 /* FIXME how to revert the partial results
542 * once error happened? Resolved by 2 Phase commit */
551 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
554 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
557 rc = lod_remote_sync(env, dev, th);
561 return dt_trans_start(env, lod->lod_child, th);
564 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
566 struct update_request *update;
567 struct update_request *tmp;
571 cfs_list_for_each_entry_safe(update, tmp,
572 &th->th_remote_update_list,
574 th->th_current_request = update;
575 rc2 = dt_trans_stop(env, update->ur_dt, th);
576 if (unlikely(rc2 != 0 && rc == 0))
580 rc2 = dt_trans_stop(env, th->th_dev, th);
582 return rc2 != 0 ? rc2 : rc;
585 static void lod_conf_get(const struct lu_env *env,
586 const struct dt_device *dev,
587 struct dt_device_param *param)
589 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
592 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
594 struct lod_device *lod = dt2lod_dev(dev);
595 struct lod_ost_desc *ost;
599 lod_getref(&lod->lod_ost_descs);
600 lod_foreach_ost(lod, i) {
601 ost = OST_TGT(lod, i);
602 LASSERT(ost && ost->ltd_ost);
603 rc = dt_sync(env, ost->ltd_ost);
605 CERROR("%s: can't sync %u: %d\n",
606 lod2obd(lod)->obd_name, i, rc);
610 lod_putref(lod, &lod->lod_ost_descs);
612 rc = dt_sync(env, lod->lod_child);
617 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
619 return dt_ro(env, dt2lod_dev(dev)->lod_child);
622 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
624 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
627 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
628 int mode, unsigned long timeout,
629 __u32 alg, struct lustre_capa_key *keys)
631 struct dt_device *next = dt2lod_dev(dev)->lod_child;
632 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
635 static const struct dt_device_operations lod_dt_ops = {
636 .dt_root_get = lod_root_get,
637 .dt_statfs = lod_statfs,
638 .dt_trans_create = lod_trans_create,
639 .dt_trans_start = lod_trans_start,
640 .dt_trans_stop = lod_trans_stop,
641 .dt_conf_get = lod_conf_get,
644 .dt_commit_async = lod_commit_async,
645 .dt_init_capa_ctxt = lod_init_capa_ctxt,
648 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
649 struct lustre_cfg *cfg)
651 struct obd_connect_data *data = NULL;
652 struct obd_device *obd;
653 char *nextdev = NULL, *p, *s;
658 LASSERT(lod->lod_child_exp == NULL);
660 /* compatibility hack: we still use old config logs
661 * which specify LOV, but we need to learn underlying
662 * OSD device, which is supposed to be:
663 * <fsname>-MDTxxxx-osd
665 * 2.x MGS generates lines like the following:
666 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
667 * 1.8 MGS generates lines like the following:
668 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
670 * we use "-MDT" to differentiate 2.x from 1.8 */
672 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
674 OBD_ALLOC(nextdev, len);
676 GOTO(out, rc = -ENOMEM);
679 s = strstr(nextdev, "-mdtlov");
680 if (unlikely(s == NULL)) {
681 CERROR("unable to parse device name %s\n",
682 lustre_cfg_string(cfg, 0));
683 GOTO(out, rc = -EINVAL);
686 if (strstr(nextdev, "-MDT")) {
691 strcpy(s, "-MDT0000-osd");
694 CERROR("unable to parse device name %s\n",
695 lustre_cfg_string(cfg, 0));
696 GOTO(out, rc = -EINVAL);
701 GOTO(out, rc = -ENOMEM);
703 obd = class_name2obd(nextdev);
705 CERROR("can not locate next device: %s\n", nextdev);
706 GOTO(out, rc = -ENOTCONN);
709 data->ocd_connect_flags = OBD_CONNECT_VERSION;
710 data->ocd_version = LUSTRE_VERSION_CODE;
712 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
715 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
719 lod->lod_dt_dev.dd_lu_dev.ld_site =
720 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
721 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
722 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
728 OBD_FREE(nextdev, len);
732 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
734 mutex_init(<d->ltd_mutex);
735 init_rwsem(<d->ltd_rw_sem);
737 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
738 * added to the LOD, see lod_add_device() */
739 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
740 if (ltd->ltd_tgt_bitmap == NULL)
743 ltd->ltd_tgts_size = 32;
746 ltd->ltd_death_row = 0;
747 ltd->ltd_refcount = 0;
751 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
752 struct lu_device_type *ldt, struct lustre_cfg *cfg)
754 struct dt_device_param ddp;
755 struct obd_device *obd;
759 obd = class_name2obd(lustre_cfg_string(cfg, 0));
761 CERROR("Cannot find obd with name %s\n",
762 lustre_cfg_string(cfg, 0));
766 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
767 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
768 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
769 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
771 rc = lod_connect_to_osd(env, lod, cfg);
775 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
776 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
778 /* setup obd to be used with old lov code */
779 rc = lod_pools_init(lod, cfg);
781 GOTO(out_disconnect, rc);
783 rc = lod_procfs_init(lod);
787 spin_lock_init(&lod->lod_desc_lock);
788 spin_lock_init(&lod->lod_connects_lock);
789 lod_tgt_desc_init(&lod->lod_mdt_descs);
790 lod_tgt_desc_init(&lod->lod_ost_descs);
797 obd_disconnect(lod->lod_child_exp);
801 static struct lu_device *lod_device_free(const struct lu_env *env,
802 struct lu_device *lu)
804 struct lod_device *lod = lu2lod_dev(lu);
805 struct lu_device *next = &lod->lod_child->dd_lu_dev;
808 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
809 dt_device_fini(&lod->lod_dt_dev);
814 static struct lu_device *lod_device_alloc(const struct lu_env *env,
815 struct lu_device_type *type,
816 struct lustre_cfg *lcfg)
818 struct lod_device *lod;
819 struct lu_device *lu_dev;
823 lu_dev = ERR_PTR(-ENOMEM);
827 lu_dev = lod2lu_dev(lod);
828 dt_device_init(&lod->lod_dt_dev, type);
829 rc = lod_init0(env, lod, type, lcfg);
831 lod_device_free(env, lu_dev);
832 lu_dev = ERR_PTR(rc);
839 static struct lu_device *lod_device_fini(const struct lu_env *env,
842 struct lod_device *lod = lu2lod_dev(d);
848 lod_procfs_fini(lod);
850 rc = lod_fini_tgt(lod, &lod->lod_ost_descs);
852 CERROR("%s:can not fini ost descs %d\n",
853 lod2obd(lod)->obd_name, rc);
855 rc = lod_fini_tgt(lod, &lod->lod_mdt_descs);
857 CERROR("%s:can not fini mdt descs %d\n",
858 lod2obd(lod)->obd_name, rc);
864 * we use exports to track all LOD users
866 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
867 struct obd_device *obd, struct obd_uuid *cluuid,
868 struct obd_connect_data *data, void *localdata)
870 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
871 struct lustre_handle conn;
875 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
877 rc = class_connect(&conn, obd, cluuid);
881 *exp = class_conn2export(&conn);
883 spin_lock(&lod->lod_connects_lock);
885 /* at the moment we expect the only user */
886 LASSERT(lod->lod_connects == 1);
887 spin_unlock(&lod->lod_connects_lock);
893 * once last export (we don't count self-export) disappeared
894 * lod can be released
896 static int lod_obd_disconnect(struct obd_export *exp)
898 struct obd_device *obd = exp->exp_obd;
899 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
903 /* Only disconnect the underlying layers on the final disconnect. */
904 spin_lock(&lod->lod_connects_lock);
906 if (lod->lod_connects != 0) {
907 /* why should there be more than 1 connect? */
908 spin_unlock(&lod->lod_connects_lock);
909 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
913 spin_unlock(&lod->lod_connects_lock);
915 /* the last user of lod has gone, let's release the device */
919 rc = class_disconnect(exp); /* bz 9811 */
921 if (rc == 0 && release)
922 class_manual_cleanup(obd);
926 LU_KEY_INIT(lod, struct lod_thread_info);
928 static void lod_key_fini(const struct lu_context *ctx,
929 struct lu_context_key *key, void *data)
931 struct lod_thread_info *info = data;
932 /* allocated in lod_get_lov_ea
933 * XXX: this is overload, a tread may have such store but used only
934 * once. Probably better would be pool of such stores per LOD.
936 if (info->lti_ea_store) {
937 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
938 info->lti_ea_store = NULL;
939 info->lti_ea_store_size = 0;
944 /* context key: lod_thread_key */
945 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
947 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
949 static struct lu_device_type_operations lod_device_type_ops = {
950 .ldto_init = lod_type_init,
951 .ldto_fini = lod_type_fini,
953 .ldto_start = lod_type_start,
954 .ldto_stop = lod_type_stop,
956 .ldto_device_alloc = lod_device_alloc,
957 .ldto_device_free = lod_device_free,
959 .ldto_device_fini = lod_device_fini
962 static struct lu_device_type lod_device_type = {
963 .ldt_tags = LU_DEVICE_DT,
964 .ldt_name = LUSTRE_LOD_NAME,
965 .ldt_ops = &lod_device_type_ops,
966 .ldt_ctx_tags = LCT_MD_THREAD,
969 static int lod_obd_health_check(const struct lu_env *env,
970 struct obd_device *obd)
972 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
973 struct lod_ost_desc *ost;
978 lod_getref(&d->lod_ost_descs);
979 lod_foreach_ost(d, i) {
981 LASSERT(ost && ost->ltd_ost);
982 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
983 /* one healthy device is enough */
987 lod_putref(d, &d->lod_ost_descs);
991 static struct obd_ops lod_obd_device_ops = {
992 .o_owner = THIS_MODULE,
993 .o_connect = lod_obd_connect,
994 .o_disconnect = lod_obd_disconnect,
995 .o_health_check = lod_obd_health_check,
996 .o_pool_new = lod_pool_new,
997 .o_pool_rem = lod_pool_remove,
998 .o_pool_add = lod_pool_add,
999 .o_pool_del = lod_pool_del,
1002 static int __init lod_mod_init(void)
1004 struct lprocfs_static_vars lvars = { 0 };
1005 cfs_proc_dir_entry_t *lov_proc_dir;
1008 rc = lu_kmem_init(lod_caches);
1012 lprocfs_lod_init_vars(&lvars);
1014 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
1015 LUSTRE_LOD_NAME, &lod_device_type);
1017 lu_kmem_fini(lod_caches);
1021 /* create "lov" entry in procfs for compatibility purposes */
1022 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
1023 if (lov_proc_dir == NULL) {
1024 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
1026 if (IS_ERR(lov_proc_dir))
1027 CERROR("lod: can't create compat entry \"lov\": %d\n",
1028 (int)PTR_ERR(lov_proc_dir));
1034 static void __exit lod_mod_exit(void)
1037 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
1039 class_unregister_type(LUSTRE_LOD_NAME);
1040 lu_kmem_fini(lod_caches);
1043 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1044 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1045 MODULE_LICENSE("GPL");
1047 module_init(lod_mod_init);
1048 module_exit(lod_mod_exit);