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 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);
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;
102 /* Slab for OSD object allocation */
103 cfs_mem_cache_t *lod_object_kmem;
105 static struct lu_kmem_descr lod_caches[] = {
107 .ckd_cache = &lod_object_kmem,
108 .ckd_name = "lod_obj",
109 .ckd_size = sizeof(struct lod_object)
116 static struct lu_device *lod_device_fini(const struct lu_env *env,
117 struct lu_device *d);
119 struct lu_object *lod_object_alloc(const struct lu_env *env,
120 const struct lu_object_header *hdr,
121 struct lu_device *dev)
123 struct lod_object *lod_obj;
124 struct lu_object *lu_obj;
125 const struct lu_fid *fid = &hdr->loh_fid;
130 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, CFS_ALLOC_IO);
132 RETURN(ERR_PTR(-ENOMEM));
134 rc = lod_fld_lookup(env, lu2lod_dev(dev), fid, &mds, LU_SEQ_RANGE_MDT);
136 OBD_SLAB_FREE_PTR(lod_obj, lod_object_kmem);
140 lod_obj->ldo_mds_num = mds;
141 lu_obj = lod2lu_obj(lod_obj);
142 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
143 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
144 if (likely(mds == lu_site2seq(dev->ld_site)->ss_node_id))
145 lu_obj->lo_ops = &lod_lu_obj_ops;
147 lu_obj->lo_ops = &lod_lu_robj_ops;
151 static int lod_cleanup_desc_tgts(const struct lu_env *env,
152 struct lod_device *lod,
153 struct lod_tgt_descs *ltd,
154 struct lustre_cfg *lcfg)
156 struct lu_device *next;
161 if (ltd->ltd_tgts_size <= 0) {
162 lod_putref(lod, ltd);
165 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
166 struct lod_tgt_desc *tgt;
169 tgt = LTD_TGT(ltd, i);
170 LASSERT(tgt && tgt->ltd_tgt);
171 next = &tgt->ltd_tgt->dd_lu_dev;
172 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
174 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
175 ": rc = %d\n", lod2obd(lod)->obd_name, i,
176 lcfg->lcfg_command, rc1);
180 lod_putref(lod, ltd);
184 static int lodname2mdt_index(char *lodname, int *index)
188 /* The lodname suppose to be fsname-MDTxxxx-mdtlov */
189 ptr = strrchr(lodname, '-');
191 CERROR("invalid MDT index in '%s'\n", lodname);
195 if (strncmp(ptr, "-mdtlov", 7) != 0) {
196 CERROR("invalid MDT index in '%s'\n", lodname);
200 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
201 CERROR("invalid MDT index in '%s'\n", lodname);
205 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
206 CERROR("invalid MDT index in '%s'\n", lodname);
210 *index = simple_strtol(ptr - 4, &tmp, 16);
211 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
212 CERROR("invalid MDT index in '%s'\n", lodname);
219 * Init client sequence manager which is used by local MDS to talk to sequence
220 * controller on remote node.
222 static int lod_seq_init_cli(const struct lu_env *env,
223 struct lod_device *lod,
224 char *tgtuuid, int index)
226 struct seq_server_site *ss;
227 struct obd_device *osp;
230 struct obd_uuid obd_uuid;
233 ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
236 /* check if this is adding the first MDC and controller is not yet
238 if (index != 0 || ss->ss_client_seq)
241 obd_str2uuid(&obd_uuid, tgtuuid);
242 osp = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
243 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
245 CERROR("%s: can't find %s device\n",
246 lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_name,
251 if (!osp->obd_set_up) {
252 CERROR("target %s not set up\n", osp->obd_name);
256 LASSERT(ss->ss_control_exp);
257 OBD_ALLOC_PTR(ss->ss_client_seq);
258 if (ss->ss_client_seq == NULL)
261 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
263 OBD_FREE_PTR(ss->ss_client_seq);
264 ss->ss_client_seq = NULL;
268 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", osp->obd_name);
269 rc = seq_client_init(ss->ss_client_seq, ss->ss_control_exp,
270 LUSTRE_SEQ_METADATA, prefix, NULL);
271 OBD_FREE(prefix, MAX_OBD_NAME + 5);
273 OBD_FREE_PTR(ss->ss_client_seq);
274 ss->ss_client_seq = NULL;
278 LASSERT(ss->ss_server_seq != NULL);
279 rc = seq_server_set_cli(ss->ss_server_seq, ss->ss_client_seq,
285 static void lod_seq_fini_cli(struct lod_device *lod)
287 struct seq_server_site *ss;
291 ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
297 if (ss->ss_server_seq)
298 seq_server_set_cli(ss->ss_server_seq,
301 if (ss->ss_control_exp) {
302 class_export_put(ss->ss_control_exp);
303 ss->ss_control_exp = NULL;
311 * Procss config log on LOD
312 * \param env environment info
313 * \param dev lod device
314 * \param lcfg config log
316 * Add osc config log,
317 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
318 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
319 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
320 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
321 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
322 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
325 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
326 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
327 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
328 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
329 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
330 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
332 static int lod_process_config(const struct lu_env *env,
333 struct lu_device *dev,
334 struct lustre_cfg *lcfg)
336 struct lod_device *lod = lu2lod_dev(dev);
337 struct lu_device *next = &lod->lod_child->dd_lu_dev;
342 switch(lcfg->lcfg_command) {
343 case LCFG_LOV_DEL_OBD:
344 case LCFG_LOV_ADD_INA:
345 case LCFG_LOV_ADD_OBD:
350 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
351 * modify_mdc_tgts add 0:lustre-MDT0001
352 * 1:lustre-MDT0001-mdc0002
354 arg1 = lustre_cfg_string(lcfg, 1);
356 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
357 GOTO(out, rc = -EINVAL);
358 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
359 GOTO(out, rc = -EINVAL);
361 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
363 mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
364 /* 1.8 configs don't have "-MDT0000" at the end */
368 rc = lodname2mdt_index(
369 lustre_cfg_string(lcfg, 0), &mdt_index);
373 rc = lod_add_device(env, lod, arg1, index, gen,
374 mdt_index, LUSTRE_OSC_NAME, 1);
375 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
377 rc = lod_add_device(env, lod, arg1, index, gen,
378 mdt_index, LUSTRE_MDC_NAME, 1);
380 rc = lod_seq_init_cli(env, lod, arg1,
382 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
383 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
385 rc = lod_add_device(env, lod, arg1, index, gen,
386 mdt_index, LUSTRE_OSC_NAME, 0);
388 rc = lod_del_device(env, lod,
397 struct lprocfs_static_vars v = { 0 };
398 struct obd_device *obd = lod2obd(lod);
400 lprocfs_lod_init_vars(&v);
402 rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
408 case LCFG_PRE_CLEANUP: {
409 lu_dev_del_linkage(dev->ld_site, dev);
410 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
411 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
413 lod_seq_fini_cli(lod);
415 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
418 * do cleanup on underlying storage only when
419 * all OSPs are cleaned up, as they use that OSD as well
421 next = &lod->lod_child->dd_lu_dev;
422 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
424 CERROR("%s: can't process %u: %d\n",
425 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
427 rc = obd_disconnect(lod->lod_child_exp);
429 CERROR("error in disconnect from storage: %d\n", rc);
433 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
443 static int lod_recovery_complete(const struct lu_env *env,
444 struct lu_device *dev)
446 struct lod_device *lod = lu2lod_dev(dev);
447 struct lu_device *next = &lod->lod_child->dd_lu_dev;
451 LASSERT(lod->lod_recovery_completed == 0);
452 lod->lod_recovery_completed = 1;
454 rc = next->ld_ops->ldo_recovery_complete(env, next);
456 lod_getref(&lod->lod_ost_descs);
457 if (lod->lod_osts_size > 0) {
458 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
459 struct lod_tgt_desc *tgt;
460 tgt = OST_TGT(lod, i);
461 LASSERT(tgt && tgt->ltd_tgt);
462 next = &tgt->ltd_ost->dd_lu_dev;
463 rc = next->ld_ops->ldo_recovery_complete(env, next);
465 CERROR("%s: can't complete recovery on #%d:"
466 "%d\n", lod2obd(lod)->obd_name, i, rc);
469 lod_putref(lod, &lod->lod_ost_descs);
473 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
474 struct lu_device *cdev)
476 struct lod_device *lod = lu2lod_dev(cdev);
477 struct lu_device *next = &lod->lod_child->dd_lu_dev;
481 rc = next->ld_ops->ldo_prepare(env, pdev, next);
483 CERROR("%s: prepare bottom error: rc = %d\n",
484 lod2obd(lod)->obd_name, rc);
488 lod->lod_initialized = 1;
493 const struct lu_device_operations lod_lu_ops = {
494 .ldo_object_alloc = lod_object_alloc,
495 .ldo_process_config = lod_process_config,
496 .ldo_recovery_complete = lod_recovery_complete,
497 .ldo_prepare = lod_prepare,
500 static int lod_root_get(const struct lu_env *env,
501 struct dt_device *dev, struct lu_fid *f)
503 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
506 static int lod_statfs(const struct lu_env *env,
507 struct dt_device *dev, struct obd_statfs *sfs)
509 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
512 static struct thandle *lod_trans_create(const struct lu_env *env,
513 struct dt_device *dev)
517 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
521 CFS_INIT_LIST_HEAD(&th->th_remote_update_list);
525 static int lod_remote_sync(const struct lu_env *env, struct dt_device *dev,
528 struct update_request *update;
532 if (cfs_list_empty(&th->th_remote_update_list))
535 cfs_list_for_each_entry(update, &th->th_remote_update_list,
537 /* In DNE phase I, there should be only one OSP
538 * here, so we will do send/receive one by one,
539 * instead of sending them parallel, will fix this
541 th->th_current_request = update;
542 rc = dt_trans_start(env, update->ur_dt, th);
544 /* FIXME how to revert the partial results
545 * once error happened? Resolved by 2 Phase commit */
554 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
557 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
560 rc = lod_remote_sync(env, dev, th);
564 return dt_trans_start(env, lod->lod_child, th);
567 static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
569 struct update_request *update;
570 struct update_request *tmp;
574 cfs_list_for_each_entry_safe(update, tmp,
575 &th->th_remote_update_list,
577 th->th_current_request = update;
578 rc2 = dt_trans_stop(env, update->ur_dt, th);
579 if (unlikely(rc2 != 0 && rc == 0))
583 rc2 = dt_trans_stop(env, th->th_dev, th);
585 return rc2 != 0 ? rc2 : rc;
588 static void lod_conf_get(const struct lu_env *env,
589 const struct dt_device *dev,
590 struct dt_device_param *param)
592 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
595 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
597 struct lod_device *lod = dt2lod_dev(dev);
598 struct lod_ost_desc *ost;
602 lod_getref(&lod->lod_ost_descs);
603 lod_foreach_ost(lod, i) {
604 ost = OST_TGT(lod, i);
605 LASSERT(ost && ost->ltd_ost);
606 rc = dt_sync(env, ost->ltd_ost);
608 CERROR("%s: can't sync %u: %d\n",
609 lod2obd(lod)->obd_name, i, rc);
613 lod_putref(lod, &lod->lod_ost_descs);
615 rc = dt_sync(env, lod->lod_child);
620 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
622 return dt_ro(env, dt2lod_dev(dev)->lod_child);
625 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
627 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
630 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
631 int mode, unsigned long timeout,
632 __u32 alg, struct lustre_capa_key *keys)
634 struct dt_device *next = dt2lod_dev(dev)->lod_child;
635 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
638 static const struct dt_device_operations lod_dt_ops = {
639 .dt_root_get = lod_root_get,
640 .dt_statfs = lod_statfs,
641 .dt_trans_create = lod_trans_create,
642 .dt_trans_start = lod_trans_start,
643 .dt_trans_stop = lod_trans_stop,
644 .dt_conf_get = lod_conf_get,
647 .dt_commit_async = lod_commit_async,
648 .dt_init_capa_ctxt = lod_init_capa_ctxt,
651 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
652 struct lustre_cfg *cfg)
654 struct obd_connect_data *data = NULL;
655 struct obd_device *obd;
656 char *nextdev = NULL, *p, *s;
661 LASSERT(lod->lod_child_exp == NULL);
663 /* compatibility hack: we still use old config logs
664 * which specify LOV, but we need to learn underlying
665 * OSD device, which is supposed to be:
666 * <fsname>-MDTxxxx-osd
668 * 2.x MGS generates lines like the following:
669 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
670 * 1.8 MGS generates lines like the following:
671 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
673 * we use "-MDT" to differentiate 2.x from 1.8 */
675 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
677 OBD_ALLOC(nextdev, len);
679 GOTO(out, rc = -ENOMEM);
682 s = strstr(nextdev, "-mdtlov");
683 if (unlikely(s == NULL)) {
684 CERROR("unable to parse device name %s\n",
685 lustre_cfg_string(cfg, 0));
686 GOTO(out, rc = -EINVAL);
689 if (strstr(nextdev, "-MDT")) {
694 strcpy(s, "-MDT0000-osd");
697 CERROR("unable to parse device name %s\n",
698 lustre_cfg_string(cfg, 0));
699 GOTO(out, rc = -EINVAL);
704 GOTO(out, rc = -ENOMEM);
706 obd = class_name2obd(nextdev);
708 CERROR("can not locate next device: %s\n", nextdev);
709 GOTO(out, rc = -ENOTCONN);
712 data->ocd_connect_flags = OBD_CONNECT_VERSION;
713 data->ocd_version = LUSTRE_VERSION_CODE;
715 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
718 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
722 lod->lod_dt_dev.dd_lu_dev.ld_site =
723 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
724 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
725 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
731 OBD_FREE(nextdev, len);
735 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
737 mutex_init(<d->ltd_mutex);
738 init_rwsem(<d->ltd_rw_sem);
740 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
741 * added to the LOD, see lod_add_device() */
742 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
743 if (ltd->ltd_tgt_bitmap == NULL)
746 ltd->ltd_tgts_size = 32;
749 ltd->ltd_death_row = 0;
750 ltd->ltd_refcount = 0;
754 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
755 struct lu_device_type *ldt, struct lustre_cfg *cfg)
757 struct dt_device_param ddp;
758 struct obd_device *obd;
762 obd = class_name2obd(lustre_cfg_string(cfg, 0));
764 CERROR("Cannot find obd with name %s\n",
765 lustre_cfg_string(cfg, 0));
769 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
770 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
771 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
772 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
774 rc = lod_connect_to_osd(env, lod, cfg);
778 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
779 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
781 /* setup obd to be used with old lov code */
782 rc = lod_pools_init(lod, cfg);
784 GOTO(out_disconnect, rc);
786 rc = lod_procfs_init(lod);
790 spin_lock_init(&lod->lod_desc_lock);
791 spin_lock_init(&lod->lod_connects_lock);
792 lod_tgt_desc_init(&lod->lod_mdt_descs);
793 lod_tgt_desc_init(&lod->lod_ost_descs);
800 obd_disconnect(lod->lod_child_exp);
804 static struct lu_device *lod_device_free(const struct lu_env *env,
805 struct lu_device *lu)
807 struct lod_device *lod = lu2lod_dev(lu);
808 struct lu_device *next = &lod->lod_child->dd_lu_dev;
811 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
812 dt_device_fini(&lod->lod_dt_dev);
817 static struct lu_device *lod_device_alloc(const struct lu_env *env,
818 struct lu_device_type *type,
819 struct lustre_cfg *lcfg)
821 struct lod_device *lod;
822 struct lu_device *lu_dev;
826 lu_dev = ERR_PTR(-ENOMEM);
830 lu_dev = lod2lu_dev(lod);
831 dt_device_init(&lod->lod_dt_dev, type);
832 rc = lod_init0(env, lod, type, lcfg);
834 lod_device_free(env, lu_dev);
835 lu_dev = ERR_PTR(rc);
842 static struct lu_device *lod_device_fini(const struct lu_env *env,
845 struct lod_device *lod = lu2lod_dev(d);
851 lod_procfs_fini(lod);
853 rc = lod_fini_tgt(lod, &lod->lod_ost_descs);
855 CERROR("%s:can not fini ost descs %d\n",
856 lod2obd(lod)->obd_name, rc);
858 rc = lod_fini_tgt(lod, &lod->lod_mdt_descs);
860 CERROR("%s:can not fini mdt descs %d\n",
861 lod2obd(lod)->obd_name, rc);
867 * we use exports to track all LOD users
869 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
870 struct obd_device *obd, struct obd_uuid *cluuid,
871 struct obd_connect_data *data, void *localdata)
873 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
874 struct lustre_handle conn;
878 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
880 rc = class_connect(&conn, obd, cluuid);
884 *exp = class_conn2export(&conn);
886 spin_lock(&lod->lod_connects_lock);
888 /* at the moment we expect the only user */
889 LASSERT(lod->lod_connects == 1);
890 spin_unlock(&lod->lod_connects_lock);
896 * once last export (we don't count self-export) disappeared
897 * lod can be released
899 static int lod_obd_disconnect(struct obd_export *exp)
901 struct obd_device *obd = exp->exp_obd;
902 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
906 /* Only disconnect the underlying layers on the final disconnect. */
907 spin_lock(&lod->lod_connects_lock);
909 if (lod->lod_connects != 0) {
910 /* why should there be more than 1 connect? */
911 spin_unlock(&lod->lod_connects_lock);
912 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
916 spin_unlock(&lod->lod_connects_lock);
918 /* the last user of lod has gone, let's release the device */
922 rc = class_disconnect(exp); /* bz 9811 */
924 if (rc == 0 && release)
925 class_manual_cleanup(obd);
929 LU_KEY_INIT(lod, struct lod_thread_info);
931 static void lod_key_fini(const struct lu_context *ctx,
932 struct lu_context_key *key, void *data)
934 struct lod_thread_info *info = data;
935 /* allocated in lod_get_lov_ea
936 * XXX: this is overload, a tread may have such store but used only
937 * once. Probably better would be pool of such stores per LOD.
939 if (info->lti_ea_store) {
940 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
941 info->lti_ea_store = NULL;
942 info->lti_ea_store_size = 0;
947 /* context key: lod_thread_key */
948 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
950 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
952 static struct lu_device_type_operations lod_device_type_ops = {
953 .ldto_init = lod_type_init,
954 .ldto_fini = lod_type_fini,
956 .ldto_start = lod_type_start,
957 .ldto_stop = lod_type_stop,
959 .ldto_device_alloc = lod_device_alloc,
960 .ldto_device_free = lod_device_free,
962 .ldto_device_fini = lod_device_fini
965 static struct lu_device_type lod_device_type = {
966 .ldt_tags = LU_DEVICE_DT,
967 .ldt_name = LUSTRE_LOD_NAME,
968 .ldt_ops = &lod_device_type_ops,
969 .ldt_ctx_tags = LCT_MD_THREAD,
972 static int lod_obd_health_check(const struct lu_env *env,
973 struct obd_device *obd)
975 struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
976 struct lod_ost_desc *ost;
981 lod_getref(&d->lod_ost_descs);
982 lod_foreach_ost(d, i) {
984 LASSERT(ost && ost->ltd_ost);
985 rc = obd_health_check(env, ost->ltd_exp->exp_obd);
986 /* one healthy device is enough */
990 lod_putref(d, &d->lod_ost_descs);
994 static struct obd_ops lod_obd_device_ops = {
995 .o_owner = THIS_MODULE,
996 .o_connect = lod_obd_connect,
997 .o_disconnect = lod_obd_disconnect,
998 .o_health_check = lod_obd_health_check,
999 .o_pool_new = lod_pool_new,
1000 .o_pool_rem = lod_pool_remove,
1001 .o_pool_add = lod_pool_add,
1002 .o_pool_del = lod_pool_del,
1005 static int __init lod_mod_init(void)
1007 struct lprocfs_static_vars lvars = { 0 };
1008 cfs_proc_dir_entry_t *lov_proc_dir;
1011 rc = lu_kmem_init(lod_caches);
1015 lprocfs_lod_init_vars(&lvars);
1017 rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
1018 LUSTRE_LOD_NAME, &lod_device_type);
1020 lu_kmem_fini(lod_caches);
1024 /* create "lov" entry in procfs for compatibility purposes */
1025 lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
1026 if (lov_proc_dir == NULL) {
1027 lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
1029 if (IS_ERR(lov_proc_dir))
1030 CERROR("lod: can't create compat entry \"lov\": %d\n",
1031 (int)PTR_ERR(lov_proc_dir));
1037 static void __exit lod_mod_exit(void)
1040 lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
1042 class_unregister_type(LUSTRE_LOD_NAME);
1043 lu_kmem_fini(lod_caches);
1046 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1047 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1048 MODULE_LICENSE("GPL");
1050 module_init(lod_mod_init);
1051 module_exit(lod_mod_exit);