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_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 flags 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 flags)
64 struct lu_seq_range range;
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 range.lsr_flags = flags;
83 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
85 CERROR("%s: Can't find tgt by seq "LPX64", rc %d\n",
86 lod2obd(lod)->obd_name, fid_seq(fid), rc);
90 *tgt = range.lsr_index;
92 CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
93 LPX64"\n", *tgt, fid_seq(fid));
98 extern struct lu_object_operations lod_lu_obj_ops;
99 extern struct lu_object_operations lod_lu_robj_ops;
100 extern struct dt_object_operations lod_obj_ops;
101 extern struct dt_lock_operations lod_lock_ops;
103 /* Slab for OSD object allocation */
104 cfs_mem_cache_t *lod_object_kmem;
106 static struct lu_kmem_descr lod_caches[] = {
108 .ckd_cache = &lod_object_kmem,
109 .ckd_name = "lod_obj",
110 .ckd_size = sizeof(struct lod_object)
117 static struct lu_device *lod_device_fini(const struct lu_env *env,
118 struct lu_device *d);
120 struct lu_object *lod_object_alloc(const struct lu_env *env,
121 const struct lu_object_header *hdr,
122 struct lu_device *dev)
124 struct lod_object *lod_obj;
125 struct lu_object *lu_obj;
126 const struct lu_fid *fid = &hdr->loh_fid;
131 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, CFS_ALLOC_IO);
133 RETURN(ERR_PTR(-ENOMEM));
135 rc = lod_fld_lookup(env, lu2lod_dev(dev), fid, &mds, LU_SEQ_RANGE_MDT);
137 OBD_SLAB_FREE_PTR(lod_obj, lod_object_kmem);
141 lod_obj->ldo_mds_num = mds;
142 lu_obj = lod2lu_obj(lod_obj);
143 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
144 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
145 lod_obj->ldo_obj.do_lock_ops = &lod_lock_ops;
146 if (likely(mds == lu_site2seq(dev->ld_site)->ss_node_id))
147 lu_obj->lo_ops = &lod_lu_obj_ops;
149 lu_obj->lo_ops = &lod_lu_robj_ops;
153 static int lod_cleanup_desc_tgts(const struct lu_env *env,
154 struct lod_device *lod,
155 struct lod_tgt_descs *ltd,
156 struct lustre_cfg *lcfg)
158 struct lu_device *next;
163 if (ltd->ltd_tgts_size <= 0) {
164 lod_putref(lod, ltd);
167 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
168 struct lod_tgt_desc *tgt;
171 tgt = LTD_TGT(ltd, i);
172 LASSERT(tgt && tgt->ltd_tgt);
173 next = &tgt->ltd_tgt->dd_lu_dev;
174 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
176 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
177 ": rc = %d\n", lod2obd(lod)->obd_name, i,
178 lcfg->lcfg_command, rc1);
182 lod_putref(lod, ltd);
186 static int lodname2mdt_index(char *lodname, int *index)
190 /* The lodname suppose to be fsname-MDTxxxx-mdtlov */
191 ptr = strrchr(lodname, '-');
193 CERROR("invalid MDT index in '%s'\n", lodname);
197 if (strncmp(ptr, "-mdtlov", 7) != 0) {
198 CERROR("invalid MDT index in '%s'\n", lodname);
202 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
203 CERROR("invalid MDT index in '%s'\n", lodname);
207 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
208 CERROR("invalid MDT index in '%s'\n", lodname);
212 *index = simple_strtol(ptr - 4, &tmp, 16);
213 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
214 CERROR("invalid MDT index in '%s'\n", lodname);
221 * Init client sequence manager which is used by local MDS to talk to sequence
222 * controller on remote node.
224 static int lod_seq_init_cli(const struct lu_env *env,
225 struct lod_device *lod,
226 char *tgtuuid, int index)
228 struct seq_server_site *ss;
229 struct obd_device *osp;
232 struct obd_uuid obd_uuid;
235 ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
238 /* check if this is adding the first MDC and controller is not yet
240 if (index != 0 || ss->ss_client_seq)
243 obd_str2uuid(&obd_uuid, tgtuuid);
244 osp = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
245 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
247 CERROR("%s: can't find %s device\n",
248 lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_name,
253 if (!osp->obd_set_up) {
254 CERROR("target %s not set up\n", osp->obd_name);
258 LASSERT(ss->ss_control_exp);
259 OBD_ALLOC_PTR(ss->ss_client_seq);
260 if (ss->ss_client_seq == NULL)
263 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
265 OBD_FREE_PTR(ss->ss_client_seq);
266 ss->ss_client_seq = NULL;
270 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", osp->obd_name);
271 rc = seq_client_init(ss->ss_client_seq, ss->ss_control_exp,
272 LUSTRE_SEQ_METADATA, prefix, NULL);
273 OBD_FREE(prefix, MAX_OBD_NAME + 5);
275 OBD_FREE_PTR(ss->ss_client_seq);
276 ss->ss_client_seq = NULL;
280 LASSERT(ss->ss_server_seq != NULL);
281 rc = seq_server_set_cli(ss->ss_server_seq, ss->ss_client_seq,
287 static void lod_seq_fini_cli(struct lod_device *lod)
289 struct seq_server_site *ss;
293 ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
299 if (ss->ss_server_seq)
300 seq_server_set_cli(ss->ss_server_seq,
303 if (ss->ss_control_exp) {
304 class_export_put(ss->ss_control_exp);
305 ss->ss_control_exp = NULL;
313 * Procss config log on LOD
314 * \param env environment info
315 * \param dev lod device
316 * \param lcfg config log
318 * Add osc config log,
319 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
320 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
321 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
322 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
323 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
324 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
327 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
328 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
329 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
330 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
331 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
332 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
334 static int lod_process_config(const struct lu_env *env,
335 struct lu_device *dev,
336 struct lustre_cfg *lcfg)
338 struct lod_device *lod = lu2lod_dev(dev);
339 struct lu_device *next = &lod->lod_child->dd_lu_dev;
344 switch(lcfg->lcfg_command) {
345 case LCFG_LOV_DEL_OBD:
346 case LCFG_LOV_ADD_INA:
347 case LCFG_LOV_ADD_OBD:
352 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
353 * modify_mdc_tgts add 0:lustre-MDT0001
354 * 1:lustre-MDT0001-mdc0002
356 arg1 = lustre_cfg_string(lcfg, 1);
358 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
359 GOTO(out, rc = -EINVAL);
360 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
361 GOTO(out, rc = -EINVAL);
363 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
365 mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
366 /* 1.8 configs don't have "-MDT0000" at the end */
370 rc = lodname2mdt_index(
371 lustre_cfg_string(lcfg, 0), &mdt_index);
375 rc = lod_add_device(env, lod, arg1, index, gen,
376 mdt_index, LUSTRE_OSC_NAME, 1);
377 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
379 rc = lod_add_device(env, lod, arg1, index, gen,
380 mdt_index, LUSTRE_MDC_NAME, 1);
382 rc = lod_seq_init_cli(env, lod, arg1,
384 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
385 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
387 rc = lod_add_device(env, lod, arg1, index, gen,
388 mdt_index, LUSTRE_OSC_NAME, 0);
390 rc = lod_del_device(env, lod,
399 struct lprocfs_static_vars v = { 0 };
400 struct obd_device *obd = lod2obd(lod);
402 lprocfs_lod_init_vars(&v);
404 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
410 case LCFG_PRE_CLEANUP: {
411 lu_dev_del_linkage(dev->ld_site, dev);
412 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
413 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
415 lod_seq_fini_cli(lod);
417 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
420 * do cleanup on underlying storage only when
421 * all OSPs are cleaned up, as they use that OSD as well
423 next = &lod->lod_child->dd_lu_dev;
424 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
426 CERROR("%s: can't process %u: %d\n",
427 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
429 rc = obd_disconnect(lod->lod_child_exp);
431 CERROR("error in disconnect from storage: %d\n", rc);
435 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
445 static int lod_recovery_complete(const struct lu_env *env,
446 struct lu_device *dev)
448 struct lod_device *lod = lu2lod_dev(dev);
449 struct lu_device *next = &lod->lod_child->dd_lu_dev;
453 LASSERT(lod->lod_recovery_completed == 0);
454 lod->lod_recovery_completed = 1;
456 rc = next->ld_ops->ldo_recovery_complete(env, next);
458 lod_getref(&lod->lod_ost_descs);
459 if (lod->lod_osts_size > 0) {
460 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
461 struct lod_tgt_desc *tgt;
462 tgt = OST_TGT(lod, i);
463 LASSERT(tgt && tgt->ltd_tgt);
464 next = &tgt->ltd_ost->dd_lu_dev;
465 rc = next->ld_ops->ldo_recovery_complete(env, next);
467 CERROR("%s: can't complete recovery on #%d:"
468 "%d\n", lod2obd(lod)->obd_name, i, rc);
471 lod_putref(lod, &lod->lod_ost_descs);
475 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
476 struct lu_device *cdev)
478 struct lod_device *lod = lu2lod_dev(cdev);
479 struct lu_device *next = &lod->lod_child->dd_lu_dev;
483 rc = next->ld_ops->ldo_prepare(env, pdev, next);
485 CERROR("%s: prepare bottom error: rc = %d\n",
486 lod2obd(lod)->obd_name, rc);
490 lod->lod_initialized = 1;
495 const struct lu_device_operations lod_lu_ops = {
496 .ldo_object_alloc = lod_object_alloc,
497 .ldo_process_config = lod_process_config,
498 .ldo_recovery_complete = lod_recovery_complete,
499 .ldo_prepare = lod_prepare,
502 static int lod_root_get(const struct lu_env *env,
503 struct dt_device *dev, struct lu_fid *f)
505 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
508 static int lod_statfs(const struct lu_env *env,
509 struct dt_device *dev, struct obd_statfs *sfs)
511 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
514 static struct thandle *lod_trans_create(const struct lu_env *env,
515 struct dt_device *dev)
519 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
523 CFS_INIT_LIST_HEAD(&th->th_remote_update_list);
527 static int lod_remote_sync(const struct lu_env *env, struct dt_device *dev,
530 struct update_request *update;
534 if (cfs_list_empty(&th->th_remote_update_list))
537 cfs_list_for_each_entry(update, &th->th_remote_update_list,
539 /* In DNE phase I, there should be only one OSP
540 * here, so we will do send/receive one by one,
541 * instead of sending them parallel, will fix this
543 th->th_current_request = update;
544 rc = dt_trans_start(env, update->ur_dt, th);
546 /* FIXME how to revert the partial results
547 * once error happened? Resolved by 2 Phase commit */
556 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
559 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
562 rc = lod_remote_sync(env, dev, th);
566 return dt_trans_start(env, lod->lod_child, th);
569 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
571 struct update_request *update;
572 struct update_request *tmp;
576 cfs_list_for_each_entry_safe(update, tmp,
577 &th->th_remote_update_list,
579 th->th_current_request = update;
580 rc2 = dt_trans_stop(env, update->ur_dt, th);
581 if (unlikely(rc2 != 0 && rc == 0))
585 rc2 = dt_trans_stop(env, th->th_dev, th);
587 return rc2 != 0 ? rc2 : rc;
590 static void lod_conf_get(const struct lu_env *env,
591 const struct dt_device *dev,
592 struct dt_device_param *param)
594 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
597 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
599 struct lod_device *lod = dt2lod_dev(dev);
600 struct lod_ost_desc *ost;
604 lod_getref(&lod->lod_ost_descs);
605 lod_foreach_ost(lod, i) {
606 ost = OST_TGT(lod, i);
607 LASSERT(ost && ost->ltd_ost);
608 rc = dt_sync(env, ost->ltd_ost);
610 CERROR("%s: can't sync %u: %d\n",
611 lod2obd(lod)->obd_name, i, rc);
615 lod_putref(lod, &lod->lod_ost_descs);
617 rc = dt_sync(env, lod->lod_child);
622 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
624 return dt_ro(env, dt2lod_dev(dev)->lod_child);
627 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
629 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
632 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
633 int mode, unsigned long timeout,
634 __u32 alg, struct lustre_capa_key *keys)
636 struct dt_device *next = dt2lod_dev(dev)->lod_child;
637 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
640 static const struct dt_device_operations lod_dt_ops = {
641 .dt_root_get = lod_root_get,
642 .dt_statfs = lod_statfs,
643 .dt_trans_create = lod_trans_create,
644 .dt_trans_start = lod_trans_start,
645 .dt_trans_stop = lod_trans_stop,
646 .dt_conf_get = lod_conf_get,
649 .dt_commit_async = lod_commit_async,
650 .dt_init_capa_ctxt = lod_init_capa_ctxt,
653 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
654 struct lustre_cfg *cfg)
656 struct obd_connect_data *data = NULL;
657 struct obd_device *obd;
658 char *nextdev = NULL, *p, *s;
663 LASSERT(lod->lod_child_exp == NULL);
665 /* compatibility hack: we still use old config logs
666 * which specify LOV, but we need to learn underlying
667 * OSD device, which is supposed to be:
668 * <fsname>-MDTxxxx-osd
670 * 2.x MGS generates lines like the following:
671 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
672 * 1.8 MGS generates lines like the following:
673 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
675 * we use "-MDT" to differentiate 2.x from 1.8 */
677 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
679 OBD_ALLOC(nextdev, len);
681 GOTO(out, rc = -ENOMEM);
684 s = strstr(nextdev, "-mdtlov");
685 if (unlikely(s == NULL)) {
686 CERROR("unable to parse device name %s\n",
687 lustre_cfg_string(cfg, 0));
688 GOTO(out, rc = -EINVAL);
691 if (strstr(nextdev, "-MDT")) {
696 strcpy(s, "-MDT0000-osd");
699 CERROR("unable to parse device name %s\n",
700 lustre_cfg_string(cfg, 0));
701 GOTO(out, rc = -EINVAL);
706 GOTO(out, rc = -ENOMEM);
708 obd = class_name2obd(nextdev);
710 CERROR("can not locate next device: %s\n", nextdev);
711 GOTO(out, rc = -ENOTCONN);
714 data->ocd_connect_flags = OBD_CONNECT_VERSION;
715 data->ocd_version = LUSTRE_VERSION_CODE;
717 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
720 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
724 lod->lod_dt_dev.dd_lu_dev.ld_site =
725 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
726 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
727 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
733 OBD_FREE(nextdev, len);
737 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
739 mutex_init(<d->ltd_mutex);
740 init_rwsem(<d->ltd_rw_sem);
742 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
743 * added to the LOD, see lod_add_device() */
744 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
745 if (ltd->ltd_tgt_bitmap == NULL)
748 ltd->ltd_tgts_size = 32;
751 ltd->ltd_death_row = 0;
752 ltd->ltd_refcount = 0;
756 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
757 struct lu_device_type *ldt, struct lustre_cfg *cfg)
759 struct dt_device_param ddp;
760 struct obd_device *obd;
764 obd = class_name2obd(lustre_cfg_string(cfg, 0));
766 CERROR("Cannot find obd with name %s\n",
767 lustre_cfg_string(cfg, 0));
771 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
772 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
773 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
774 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
776 rc = lod_connect_to_osd(env, lod, cfg);
780 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
781 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
783 /* setup obd to be used with old lov code */
784 rc = lod_pools_init(lod, cfg);
786 GOTO(out_disconnect, rc);
788 rc = lod_procfs_init(lod);
792 spin_lock_init(&lod->lod_desc_lock);
793 spin_lock_init(&lod->lod_connects_lock);
794 lod_tgt_desc_init(&lod->lod_mdt_descs);
795 lod_tgt_desc_init(&lod->lod_ost_descs);
802 obd_disconnect(lod->lod_child_exp);
806 static struct lu_device *lod_device_free(const struct lu_env *env,
807 struct lu_device *lu)
809 struct lod_device *lod = lu2lod_dev(lu);
810 struct lu_device *next = &lod->lod_child->dd_lu_dev;
813 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
814 dt_device_fini(&lod->lod_dt_dev);
819 static struct lu_device *lod_device_alloc(const struct lu_env *env,
820 struct lu_device_type *type,
821 struct lustre_cfg *lcfg)
823 struct lod_device *lod;
824 struct lu_device *lu_dev;
828 lu_dev = ERR_PTR(-ENOMEM);
832 lu_dev = lod2lu_dev(lod);
833 dt_device_init(&lod->lod_dt_dev, type);
834 rc = lod_init0(env, lod, type, lcfg);
836 lod_device_free(env, lu_dev);
837 lu_dev = ERR_PTR(rc);
844 static struct lu_device *lod_device_fini(const struct lu_env *env,
847 struct lod_device *lod = lu2lod_dev(d);
853 lod_procfs_fini(lod);
855 rc = lod_fini_tgt(lod, &lod->lod_ost_descs);
857 CERROR("%s:can not fini ost descs %d\n",
858 lod2obd(lod)->obd_name, rc);
860 rc = lod_fini_tgt(lod, &lod->lod_mdt_descs);
862 CERROR("%s:can not fini mdt descs %d\n",
863 lod2obd(lod)->obd_name, rc);
869 * we use exports to track all LOD users
871 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
872 struct obd_device *obd, struct obd_uuid *cluuid,
873 struct obd_connect_data *data, void *localdata)
875 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
876 struct lustre_handle conn;
880 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
882 rc = class_connect(&conn, obd, cluuid);
886 *exp = class_conn2export(&conn);
888 spin_lock(&lod->lod_connects_lock);
890 /* at the moment we expect the only user */
891 LASSERT(lod->lod_connects == 1);
892 spin_unlock(&lod->lod_connects_lock);
898 * once last export (we don't count self-export) disappeared
899 * lod can be released
901 static int lod_obd_disconnect(struct obd_export *exp)
903 struct obd_device *obd = exp->exp_obd;
904 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
908 /* Only disconnect the underlying layers on the final disconnect. */
909 spin_lock(&lod->lod_connects_lock);
911 if (lod->lod_connects != 0) {
912 /* why should there be more than 1 connect? */
913 spin_unlock(&lod->lod_connects_lock);
914 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
918 spin_unlock(&lod->lod_connects_lock);
920 /* the last user of lod has gone, let's release the device */
924 rc = class_disconnect(exp); /* bz 9811 */
926 if (rc == 0 && release)
927 class_manual_cleanup(obd);
931 LU_KEY_INIT(lod, struct lod_thread_info);
933 static void lod_key_fini(const struct lu_context *ctx,
934 struct lu_context_key *key, void *data)
936 struct lod_thread_info *info = data;
937 /* allocated in lod_get_lov_ea
938 * XXX: this is overload, a tread may have such store but used only
939 * once. Probably better would be pool of such stores per LOD.
941 if (info->lti_ea_store) {
942 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
943 info->lti_ea_store = NULL;
944 info->lti_ea_store_size = 0;
949 /* context key: lod_thread_key */
950 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
952 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
954 static struct lu_device_type_operations lod_device_type_ops = {
955 .ldto_init = lod_type_init,
956 .ldto_fini = lod_type_fini,
958 .ldto_start = lod_type_start,
959 .ldto_stop = lod_type_stop,
961 .ldto_device_alloc = lod_device_alloc,
962 .ldto_device_free = lod_device_free,
964 .ldto_device_fini = lod_device_fini
967 static struct lu_device_type lod_device_type = {
968 .ldt_tags = LU_DEVICE_DT,
969 .ldt_name = LUSTRE_LOD_NAME,
970 .ldt_ops = &lod_device_type_ops,
971 .ldt_ctx_tags = LCT_MD_THREAD,
974 static int lod_obd_health_check(const struct lu_env *env,
975 struct obd_device *obd)
977 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
978 struct lod_ost_desc *ost;
983 lod_getref(&d->lod_ost_descs);
984 lod_foreach_ost(d, i) {
986 LASSERT(ost && ost->ltd_ost);
987 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
988 /* one healthy device is enough */
992 lod_putref(d, &d->lod_ost_descs);
996 static struct obd_ops lod_obd_device_ops = {
997 .o_owner = THIS_MODULE,
998 .o_connect = lod_obd_connect,
999 .o_disconnect = lod_obd_disconnect,
1000 .o_health_check = lod_obd_health_check,
1001 .o_pool_new = lod_pool_new,
1002 .o_pool_rem = lod_pool_remove,
1003 .o_pool_add = lod_pool_add,
1004 .o_pool_del = lod_pool_del,
1007 static int __init lod_mod_init(void)
1009 struct lprocfs_static_vars lvars = { 0 };
1010 cfs_proc_dir_entry_t *lov_proc_dir;
1013 rc = lu_kmem_init(lod_caches);
1017 lprocfs_lod_init_vars(&lvars);
1019 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
1020 LUSTRE_LOD_NAME, &lod_device_type);
1022 lu_kmem_fini(lod_caches);
1026 /* create "lov" entry in procfs for compatibility purposes */
1027 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
1028 if (lov_proc_dir == NULL) {
1029 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
1031 if (IS_ERR(lov_proc_dir))
1032 CERROR("lod: can't create compat entry \"lov\": %d\n",
1033 (int)PTR_ERR(lov_proc_dir));
1039 static void __exit lod_mod_exit(void)
1042 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
1044 class_unregister_type(LUSTRE_LOD_NAME);
1045 lu_kmem_fini(lod_caches);
1048 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1049 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1050 MODULE_LICENSE("GPL");
1052 module_init(lod_mod_init);
1053 module_exit(lod_mod_exit);