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>
41 * The Logical Object Device (LOD) layer manages access to striped
42 * objects (both regular files and directories). It implements the DT
43 * device and object APIs and is responsible for creating, storing,
44 * and loading striping information as an extended attribute of the
45 * underlying OSD object. LOD is the server side analog of the LOV and
46 * LMV layers on the client side.
48 * Metadata LU object stack (layers of the same compound LU object,
49 * all have the same FID):
59 * During LOD object initialization the localness or remoteness of the
60 * object FID dictates the choice between OSD and OSP.
62 * An LOD object (file or directory) with N stripes (each has a
69 * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
71 * When upper layers must access an object's stripes (which are
72 * themselves OST or MDT LU objects) LOD finds these objects by their
73 * FIDs and stores them as an array of DT object pointers on the
74 * object. Declarations and operations on LOD objects are received by
75 * LOD (as DT object operations) and performed on the underlying
76 * OS[DP] object and (as needed) on the stripes. From the perspective
77 * of LOD, a stripe-less file (created by mknod() or open with
78 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
79 * while a non-striped directory (created by mkdir()) is an object
80 * which will never have stripes.
82 * The LOD layer also implements a small subset of the OBD device API
83 * to support MDT stack initialization and finalization (an MDD device
84 * connects and disconnects itself to and from the underlying LOD
85 * device), and pool management. In turn LOD uses the OBD device API
86 * to connect it self to the underlying OSD, and to connect itself to
87 * OSP devices representing the MDTs and OSTs that bear the stripes of
91 #define DEBUG_SUBSYSTEM S_MDS
93 #include <obd_class.h>
94 #include <md_object.h>
95 #include <lustre_fid.h>
96 #include <lustre_param.h>
97 #include <lustre_update.h>
99 #include "lod_internal.h"
102 * Lookup MDT/OST index \a tgt by FID \a fid.
104 * \param lod LOD to be lookup at.
105 * \param fid FID of object to find MDT/OST.
106 * \param tgt MDT/OST index to return.
107 * \param type indidcate the FID is on MDS or OST.
109 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
110 const struct lu_fid *fid, __u32 *tgt, int *type)
112 struct lu_seq_range range = { 0 };
113 struct lu_server_fld *server_fld;
117 LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
119 if (fid_is_idif(fid)) {
120 *tgt = fid_idif_ost_idx(fid);
121 *type = LU_SEQ_RANGE_OST;
125 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
126 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
128 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
129 *type = LU_SEQ_RANGE_MDT;
133 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
134 fld_range_set_type(&range, *type);
135 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
139 *tgt = range.lsr_index;
140 *type = range.lsr_flags;
142 CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
143 LPX64"\n", *tgt, fid_seq(fid));
148 extern struct lu_object_operations lod_lu_obj_ops;
149 extern struct dt_object_operations lod_obj_ops;
151 /* Slab for OSD object allocation */
152 struct kmem_cache *lod_object_kmem;
154 static struct lu_kmem_descr lod_caches[] = {
156 .ckd_cache = &lod_object_kmem,
157 .ckd_name = "lod_obj",
158 .ckd_size = sizeof(struct lod_object)
165 static struct lu_device *lod_device_fini(const struct lu_env *env,
166 struct lu_device *d);
168 struct lu_object *lod_object_alloc(const struct lu_env *env,
169 const struct lu_object_header *hdr,
170 struct lu_device *dev)
172 struct lod_object *lod_obj;
173 struct lu_object *lu_obj;
176 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
178 RETURN(ERR_PTR(-ENOMEM));
180 lu_obj = lod2lu_obj(lod_obj);
181 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
182 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
183 lu_obj->lo_ops = &lod_lu_obj_ops;
188 static int lod_cleanup_desc_tgts(const struct lu_env *env,
189 struct lod_device *lod,
190 struct lod_tgt_descs *ltd,
191 struct lustre_cfg *lcfg)
193 struct lu_device *next;
198 if (ltd->ltd_tgts_size <= 0) {
199 lod_putref(lod, ltd);
202 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
203 struct lod_tgt_desc *tgt;
206 tgt = LTD_TGT(ltd, i);
207 LASSERT(tgt && tgt->ltd_tgt);
208 next = &tgt->ltd_tgt->dd_lu_dev;
209 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
211 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
212 ": rc = %d\n", lod2obd(lod)->obd_name, i,
213 lcfg->lcfg_command, rc1);
217 lod_putref(lod, ltd);
221 static int lodname2mdt_index(char *lodname, long *index)
225 /* The lodname suppose to be fsname-MDTxxxx-mdtlov */
226 ptr = strrchr(lodname, '-');
228 CERROR("invalid MDT index in '%s'\n", lodname);
232 if (strncmp(ptr, "-mdtlov", 7) != 0) {
233 CERROR("invalid MDT index in '%s'\n", lodname);
237 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
238 CERROR("invalid MDT index in '%s'\n", lodname);
242 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
243 CERROR("invalid MDT index in '%s'\n", lodname);
247 *index = simple_strtol(ptr - 4, &tmp, 16);
248 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
249 CERROR("invalid MDT index in '%s'\n", lodname);
256 * Procss config log on LOD
257 * \param env environment info
258 * \param dev lod device
259 * \param lcfg config log
261 * Add osc config log,
262 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
263 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
264 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
265 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
266 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
267 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
270 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
271 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
272 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
273 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
274 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
275 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
277 static int lod_process_config(const struct lu_env *env,
278 struct lu_device *dev,
279 struct lustre_cfg *lcfg)
281 struct lod_device *lod = lu2lod_dev(dev);
282 struct lu_device *next = &lod->lod_child->dd_lu_dev;
287 switch(lcfg->lcfg_command) {
288 case LCFG_LOV_DEL_OBD:
289 case LCFG_LOV_ADD_INA:
290 case LCFG_LOV_ADD_OBD:
295 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
296 * modify_mdc_tgts add 0:lustre-MDT0001
297 * 1:lustre-MDT0001-mdc0002
299 arg1 = lustre_cfg_string(lcfg, 1);
301 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
302 GOTO(out, rc = -EINVAL);
303 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
304 GOTO(out, rc = -EINVAL);
306 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
308 mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
309 /* 1.8 configs don't have "-MDT0000" at the end */
314 rc = lodname2mdt_index(
315 lustre_cfg_string(lcfg, 0),
319 mdt_index = long_index;
321 rc = lod_add_device(env, lod, arg1, index, gen,
322 mdt_index, LUSTRE_OSC_NAME, 1);
323 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
325 rc = lod_add_device(env, lod, arg1, index, gen,
326 mdt_index, LUSTRE_MDC_NAME, 1);
327 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
328 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
330 rc = lod_add_device(env, lod, arg1, index, gen,
331 mdt_index, LUSTRE_OSC_NAME, 0);
333 rc = lod_del_device(env, lod,
335 arg1, index, gen, true);
342 struct obd_device *obd = lod2obd(lod);
344 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
351 case LCFG_PRE_CLEANUP: {
352 lu_dev_del_linkage(dev->ld_site, dev);
353 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
354 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
355 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
358 * do cleanup on underlying storage only when
359 * all OSPs are cleaned up, as they use that OSD as well
361 next = &lod->lod_child->dd_lu_dev;
362 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
364 CERROR("%s: can't process %u: %d\n",
365 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
367 rc = obd_disconnect(lod->lod_child_exp);
369 CERROR("error in disconnect from storage: %d\n", rc);
373 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
383 static int lod_recovery_complete(const struct lu_env *env,
384 struct lu_device *dev)
386 struct lod_device *lod = lu2lod_dev(dev);
387 struct lu_device *next = &lod->lod_child->dd_lu_dev;
392 LASSERT(lod->lod_recovery_completed == 0);
393 lod->lod_recovery_completed = 1;
395 rc = next->ld_ops->ldo_recovery_complete(env, next);
397 lod_getref(&lod->lod_ost_descs);
398 if (lod->lod_osts_size > 0) {
399 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
400 struct lod_tgt_desc *tgt;
401 tgt = OST_TGT(lod, i);
402 LASSERT(tgt && tgt->ltd_tgt);
403 next = &tgt->ltd_ost->dd_lu_dev;
404 rc = next->ld_ops->ldo_recovery_complete(env, next);
406 CERROR("%s: can't complete recovery on #%d:"
407 "%d\n", lod2obd(lod)->obd_name, i, rc);
410 lod_putref(lod, &lod->lod_ost_descs);
414 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
415 struct lu_device *cdev)
417 struct lod_device *lod = lu2lod_dev(cdev);
418 struct lu_device *next = &lod->lod_child->dd_lu_dev;
422 rc = next->ld_ops->ldo_prepare(env, pdev, next);
424 CERROR("%s: prepare bottom error: rc = %d\n",
425 lod2obd(lod)->obd_name, rc);
429 lod->lod_initialized = 1;
434 const struct lu_device_operations lod_lu_ops = {
435 .ldo_object_alloc = lod_object_alloc,
436 .ldo_process_config = lod_process_config,
437 .ldo_recovery_complete = lod_recovery_complete,
438 .ldo_prepare = lod_prepare,
441 static int lod_root_get(const struct lu_env *env,
442 struct dt_device *dev, struct lu_fid *f)
444 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
447 static int lod_statfs(const struct lu_env *env,
448 struct dt_device *dev, struct obd_statfs *sfs)
450 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
453 static struct thandle *lod_trans_create(const struct lu_env *env,
454 struct dt_device *dev)
458 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
465 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
468 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
471 if (unlikely(th->th_update != NULL)) {
472 struct thandle_update *tu = th->th_update;
473 struct dt_update_request *update;
475 list_for_each_entry(update, &tu->tu_remote_update_list,
477 LASSERT(update->dur_dt != NULL);
478 rc = dt_trans_start(env, update->dur_dt, th);
483 return dt_trans_start(env, lod->lod_child, th);
486 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
489 struct thandle_update *tu = th->th_update;
490 struct dt_update_request *update;
491 struct dt_update_request *tmp;
496 rc = dt_trans_stop(env, th->th_dev, th);
497 if (likely(tu == NULL))
500 list_for_each_entry_safe(update, tmp,
501 &tu->tu_remote_update_list,
503 /* update will be freed inside dt_trans_stop */
504 rc2 = dt_trans_stop(env, update->dur_dt, th);
505 if (unlikely(rc2 != 0 && rc == 0))
512 static void lod_conf_get(const struct lu_env *env,
513 const struct dt_device *dev,
514 struct dt_device_param *param)
516 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
519 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
521 struct lod_device *lod = dt2lod_dev(dev);
522 struct lod_ost_desc *ost;
527 lod_getref(&lod->lod_ost_descs);
528 lod_foreach_ost(lod, i) {
529 ost = OST_TGT(lod, i);
530 LASSERT(ost && ost->ltd_ost);
531 rc = dt_sync(env, ost->ltd_ost);
533 CERROR("%s: can't sync %u: %d\n",
534 lod2obd(lod)->obd_name, i, rc);
538 lod_putref(lod, &lod->lod_ost_descs);
540 rc = dt_sync(env, lod->lod_child);
545 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
547 return dt_ro(env, dt2lod_dev(dev)->lod_child);
550 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
552 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
555 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
556 int mode, unsigned long timeout,
557 __u32 alg, struct lustre_capa_key *keys)
559 struct dt_device *next = dt2lod_dev(dev)->lod_child;
560 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
563 static const struct dt_device_operations lod_dt_ops = {
564 .dt_root_get = lod_root_get,
565 .dt_statfs = lod_statfs,
566 .dt_trans_create = lod_trans_create,
567 .dt_trans_start = lod_trans_start,
568 .dt_trans_stop = lod_trans_stop,
569 .dt_conf_get = lod_conf_get,
572 .dt_commit_async = lod_commit_async,
573 .dt_init_capa_ctxt = lod_init_capa_ctxt,
576 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
577 struct lustre_cfg *cfg)
579 struct obd_connect_data *data = NULL;
580 struct obd_device *obd;
581 char *nextdev = NULL, *p, *s;
586 LASSERT(lod->lod_child_exp == NULL);
588 /* compatibility hack: we still use old config logs
589 * which specify LOV, but we need to learn underlying
590 * OSD device, which is supposed to be:
591 * <fsname>-MDTxxxx-osd
593 * 2.x MGS generates lines like the following:
594 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
595 * 1.8 MGS generates lines like the following:
596 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
598 * we use "-MDT" to differentiate 2.x from 1.8 */
600 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
602 OBD_ALLOC(nextdev, len);
604 GOTO(out, rc = -ENOMEM);
607 s = strstr(nextdev, "-mdtlov");
608 if (unlikely(s == NULL)) {
609 CERROR("unable to parse device name %s\n",
610 lustre_cfg_string(cfg, 0));
611 GOTO(out, rc = -EINVAL);
614 if (strstr(nextdev, "-MDT")) {
619 strcpy(s, "-MDT0000-osd");
622 CERROR("unable to parse device name %s\n",
623 lustre_cfg_string(cfg, 0));
624 GOTO(out, rc = -EINVAL);
629 GOTO(out, rc = -ENOMEM);
631 obd = class_name2obd(nextdev);
633 CERROR("can not locate next device: %s\n", nextdev);
634 GOTO(out, rc = -ENOTCONN);
637 data->ocd_connect_flags = OBD_CONNECT_VERSION;
638 data->ocd_version = LUSTRE_VERSION_CODE;
640 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
643 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
647 lod->lod_dt_dev.dd_lu_dev.ld_site =
648 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
649 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
650 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
656 OBD_FREE(nextdev, len);
660 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
662 mutex_init(<d->ltd_mutex);
663 init_rwsem(<d->ltd_rw_sem);
665 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
666 * added to the LOD, see lod_add_device() */
667 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
668 if (ltd->ltd_tgt_bitmap == NULL)
671 ltd->ltd_tgts_size = 32;
674 ltd->ltd_death_row = 0;
675 ltd->ltd_refcount = 0;
679 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
680 struct lu_device_type *ldt, struct lustre_cfg *cfg)
682 struct dt_device_param ddp;
683 struct obd_device *obd;
687 obd = class_name2obd(lustre_cfg_string(cfg, 0));
689 CERROR("Cannot find obd with name %s\n",
690 lustre_cfg_string(cfg, 0));
694 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
695 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
696 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
697 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
699 rc = lod_connect_to_osd(env, lod, cfg);
703 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
704 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
706 /* setup obd to be used with old lov code */
707 rc = lod_pools_init(lod, cfg);
709 GOTO(out_disconnect, rc);
711 rc = lod_procfs_init(lod);
715 spin_lock_init(&lod->lod_desc_lock);
716 spin_lock_init(&lod->lod_connects_lock);
717 lod_tgt_desc_init(&lod->lod_mdt_descs);
718 lod_tgt_desc_init(&lod->lod_ost_descs);
725 obd_disconnect(lod->lod_child_exp);
729 static struct lu_device *lod_device_free(const struct lu_env *env,
730 struct lu_device *lu)
732 struct lod_device *lod = lu2lod_dev(lu);
733 struct lu_device *next = &lod->lod_child->dd_lu_dev;
736 LASSERT(atomic_read(&lu->ld_ref) == 0);
737 dt_device_fini(&lod->lod_dt_dev);
742 static struct lu_device *lod_device_alloc(const struct lu_env *env,
743 struct lu_device_type *type,
744 struct lustre_cfg *lcfg)
746 struct lod_device *lod;
747 struct lu_device *lu_dev;
751 lu_dev = ERR_PTR(-ENOMEM);
755 lu_dev = lod2lu_dev(lod);
756 dt_device_init(&lod->lod_dt_dev, type);
757 rc = lod_init0(env, lod, type, lcfg);
759 lod_device_free(env, lu_dev);
760 lu_dev = ERR_PTR(rc);
767 static struct lu_device *lod_device_fini(const struct lu_env *env,
770 struct lod_device *lod = lu2lod_dev(d);
776 lod_procfs_fini(lod);
778 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
780 CERROR("%s:can not fini ost descs %d\n",
781 lod2obd(lod)->obd_name, rc);
783 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
785 CERROR("%s:can not fini mdt descs %d\n",
786 lod2obd(lod)->obd_name, rc);
792 * we use exports to track all LOD users
794 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
795 struct obd_device *obd, struct obd_uuid *cluuid,
796 struct obd_connect_data *data, void *localdata)
798 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
799 struct lustre_handle conn;
803 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
805 rc = class_connect(&conn, obd, cluuid);
809 *exp = class_conn2export(&conn);
811 spin_lock(&lod->lod_connects_lock);
813 /* at the moment we expect the only user */
814 LASSERT(lod->lod_connects == 1);
815 spin_unlock(&lod->lod_connects_lock);
821 * once last export (we don't count self-export) disappeared
822 * lod can be released
824 static int lod_obd_disconnect(struct obd_export *exp)
826 struct obd_device *obd = exp->exp_obd;
827 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
831 /* Only disconnect the underlying layers on the final disconnect. */
832 spin_lock(&lod->lod_connects_lock);
834 if (lod->lod_connects != 0) {
835 /* why should there be more than 1 connect? */
836 spin_unlock(&lod->lod_connects_lock);
837 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
841 spin_unlock(&lod->lod_connects_lock);
843 /* the last user of lod has gone, let's release the device */
847 rc = class_disconnect(exp); /* bz 9811 */
849 if (rc == 0 && release)
850 class_manual_cleanup(obd);
854 LU_KEY_INIT(lod, struct lod_thread_info);
856 static void lod_key_fini(const struct lu_context *ctx,
857 struct lu_context_key *key, void *data)
859 struct lod_thread_info *info = data;
860 /* allocated in lod_get_lov_ea
861 * XXX: this is overload, a tread may have such store but used only
862 * once. Probably better would be pool of such stores per LOD.
864 if (info->lti_ea_store) {
865 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
866 info->lti_ea_store = NULL;
867 info->lti_ea_store_size = 0;
869 lu_buf_free(&info->lti_linkea_buf);
873 /* context key: lod_thread_key */
874 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
876 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
878 static struct lu_device_type_operations lod_device_type_ops = {
879 .ldto_init = lod_type_init,
880 .ldto_fini = lod_type_fini,
882 .ldto_start = lod_type_start,
883 .ldto_stop = lod_type_stop,
885 .ldto_device_alloc = lod_device_alloc,
886 .ldto_device_free = lod_device_free,
888 .ldto_device_fini = lod_device_fini
891 static struct lu_device_type lod_device_type = {
892 .ldt_tags = LU_DEVICE_DT,
893 .ldt_name = LUSTRE_LOD_NAME,
894 .ldt_ops = &lod_device_type_ops,
895 .ldt_ctx_tags = LCT_MD_THREAD,
898 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
899 __u32 keylen, void *key, __u32 *vallen, void *val,
900 struct lov_stripe_md *lsm)
904 if (KEY_IS(KEY_OSP_CONNECTED)) {
905 struct obd_device *obd = exp->exp_obd;
906 struct lod_device *d;
907 struct lod_ost_desc *ost;
911 if (!obd->obd_set_up || obd->obd_stopping)
914 d = lu2lod_dev(obd->obd_lu_dev);
915 lod_getref(&d->lod_ost_descs);
916 lod_foreach_ost(d, i) {
918 LASSERT(ost && ost->ltd_ost);
920 rc = obd_get_info(env, ost->ltd_exp, keylen, key,
922 /* one healthy device is enough */
926 lod_putref(d, &d->lod_ost_descs);
933 static struct obd_ops lod_obd_device_ops = {
934 .o_owner = THIS_MODULE,
935 .o_connect = lod_obd_connect,
936 .o_disconnect = lod_obd_disconnect,
937 .o_get_info = lod_obd_get_info,
938 .o_pool_new = lod_pool_new,
939 .o_pool_rem = lod_pool_remove,
940 .o_pool_add = lod_pool_add,
941 .o_pool_del = lod_pool_del,
944 static int __init lod_mod_init(void)
946 struct obd_type *type;
949 rc = lu_kmem_init(lod_caches);
953 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
954 LUSTRE_LOD_NAME, &lod_device_type);
956 lu_kmem_fini(lod_caches);
960 /* create "lov" entry in procfs for compatibility purposes */
961 type = class_search_type(LUSTRE_LOV_NAME);
962 if (type != NULL && type->typ_procroot != NULL)
965 type = class_search_type(LUSTRE_LOD_NAME);
966 type->typ_procsym = lprocfs_seq_register("lov", proc_lustre_root,
968 if (IS_ERR(type->typ_procsym)) {
969 CERROR("lod: can't create compat entry \"lov\": %d\n",
970 (int)PTR_ERR(type->typ_procsym));
971 type->typ_procsym = NULL;
976 static void __exit lod_mod_exit(void)
978 class_unregister_type(LUSTRE_LOD_NAME);
979 lu_kmem_fini(lod_caches);
982 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
983 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
984 MODULE_LICENSE("GPL");
986 module_init(lod_mod_init);
987 module_exit(lod_mod_exit);