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 target by FID.
104 * Lookup MDT/OST target index by FID. Type of the target can be
107 * \param[in] env LU environment provided by the caller
108 * \param[in] lod lod device
110 * \param[out] tgt result target index
111 * \param[in] type expected type of the target:
112 * LU_SEQ_RANGE_{MDT,OST,ANY}
114 * \retval 0 on success
115 * \retval negative negated errno on error
117 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
118 const struct lu_fid *fid, __u32 *tgt, int *type)
120 struct lu_seq_range range = { 0 };
121 struct lu_server_fld *server_fld;
125 if (!fid_is_sane(fid)) {
126 CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
131 if (fid_is_idif(fid)) {
132 *tgt = fid_idif_ost_idx(fid);
133 *type = LU_SEQ_RANGE_OST;
137 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
138 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
140 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
141 *type = LU_SEQ_RANGE_MDT;
145 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
146 fld_range_set_type(&range, *type);
147 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
151 *tgt = range.lsr_index;
152 *type = range.lsr_flags;
154 CDEBUG(D_INFO, "%s: got tgt %x for sequence: "LPX64"\n",
155 lod2obd(lod)->obd_name, *tgt, fid_seq(fid));
160 extern struct lu_object_operations lod_lu_obj_ops;
161 extern struct dt_object_operations lod_obj_ops;
163 /* Slab for OSD object allocation */
164 struct kmem_cache *lod_object_kmem;
166 static struct lu_kmem_descr lod_caches[] = {
168 .ckd_cache = &lod_object_kmem,
169 .ckd_name = "lod_obj",
170 .ckd_size = sizeof(struct lod_object)
177 static struct lu_device *lod_device_fini(const struct lu_env *env,
178 struct lu_device *d);
181 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
183 * Allocates and initializes LOD's slice in the given object.
185 * see include/lu_object.h for the details.
187 struct lu_object *lod_object_alloc(const struct lu_env *env,
188 const struct lu_object_header *hdr,
189 struct lu_device *dev)
191 struct lod_object *lod_obj;
192 struct lu_object *lu_obj;
195 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
197 RETURN(ERR_PTR(-ENOMEM));
199 lu_obj = lod2lu_obj(lod_obj);
200 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
201 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
202 lu_obj->lo_ops = &lod_lu_obj_ops;
208 * Cleanup table of target's descriptors.
210 * The function goes through all the targets in the given table
211 * and apply given configuration command on to the targets.
212 * Used to cleanup the targets at unmount.
214 * \param[in] env LU environment provided by the caller
215 * \param[in] lod lod device
216 * \param[in] ltd target's table to go through
217 * \param[in] lcfg configuration command to apply
219 * \retval 0 on success
220 * \retval negative negated errno on error
222 static int lod_cleanup_desc_tgts(const struct lu_env *env,
223 struct lod_device *lod,
224 struct lod_tgt_descs *ltd,
225 struct lustre_cfg *lcfg)
227 struct lu_device *next;
232 if (ltd->ltd_tgts_size <= 0) {
233 lod_putref(lod, ltd);
236 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
237 struct lod_tgt_desc *tgt;
240 tgt = LTD_TGT(ltd, i);
241 LASSERT(tgt && tgt->ltd_tgt);
242 next = &tgt->ltd_tgt->dd_lu_dev;
243 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
245 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
246 ": rc = %d\n", lod2obd(lod)->obd_name, i,
247 lcfg->lcfg_command, rc1);
251 lod_putref(lod, ltd);
256 * Extract MDT target index from a device name.
258 * a helper function to extract index from the given device name
259 * like "fsname-MDTxxxx-mdtlov"
261 * \param[in] lodname device name
262 * \param[out] index extracted index
264 * \retval 0 on success
265 * \retval -EINVAL if the name is invalid
267 static int lodname2mdt_index(char *lodname, long *index)
271 ptr = strrchr(lodname, '-');
273 CERROR("invalid MDT index in '%s'\n", lodname);
277 if (strncmp(ptr, "-mdtlov", 7) != 0) {
278 CERROR("invalid MDT index in '%s'\n", lodname);
282 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
283 CERROR("invalid MDT index in '%s'\n", lodname);
287 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
288 CERROR("invalid MDT index in '%s'\n", lodname);
292 *index = simple_strtol(ptr - 4, &tmp, 16);
293 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
294 CERROR("invalid MDT index in '%s'\n", lodname);
301 * Implementation of lu_device_operations::ldo_process_config() for LOD
303 * The method is called by the configuration subsystem during setup,
304 * cleanup and when the configuration changes. The method processes
305 * few specific commands like adding/removing the targets, changing
306 * the runtime parameters.
308 * \param[in] env LU environment provided by the caller
309 * \param[in] dev lod device
310 * \param[in] lcfg configuration command to apply
312 * \retval 0 on success
313 * \retval negative negated errno on error
315 * The examples are below.
317 * Add osc config log:
318 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
319 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
320 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
321 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
322 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
323 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
325 * Add mdc config log:
326 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
327 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
328 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
329 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
330 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
331 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
333 static int lod_process_config(const struct lu_env *env,
334 struct lu_device *dev,
335 struct lustre_cfg *lcfg)
337 struct lod_device *lod = lu2lod_dev(dev);
338 struct lu_device *next = &lod->lod_child->dd_lu_dev;
343 switch(lcfg->lcfg_command) {
344 case LCFG_LOV_DEL_OBD:
345 case LCFG_LOV_ADD_INA:
346 case LCFG_LOV_ADD_OBD:
351 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
352 * modify_mdc_tgts add 0:lustre-MDT0001
353 * 1:lustre-MDT0001-mdc0002
355 arg1 = lustre_cfg_string(lcfg, 1);
357 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
358 GOTO(out, rc = -EINVAL);
359 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
360 GOTO(out, rc = -EINVAL);
362 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
364 mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
365 /* 1.8 configs don't have "-MDT0000" at the end */
370 rc = lodname2mdt_index(
371 lustre_cfg_string(lcfg, 0),
375 mdt_index = long_index;
377 rc = lod_add_device(env, lod, arg1, index, gen,
378 mdt_index, LUSTRE_OSC_NAME, 1);
379 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
381 rc = lod_add_device(env, lod, arg1, index, gen,
382 mdt_index, LUSTRE_MDC_NAME, 1);
383 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
384 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
386 rc = lod_add_device(env, lod, arg1, index, gen,
387 mdt_index, LUSTRE_OSC_NAME, 0);
389 rc = lod_del_device(env, lod,
391 arg1, index, gen, true);
398 struct obd_device *obd = lod2obd(lod);
400 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
407 case LCFG_PRE_CLEANUP: {
408 lu_dev_del_linkage(dev->ld_site, dev);
409 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
410 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
411 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
414 * do cleanup on underlying storage only when
415 * all OSPs are cleaned up, as they use that OSD as well
417 next = &lod->lod_child->dd_lu_dev;
418 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
420 CERROR("%s: can't process %u: %d\n",
421 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
423 rc = obd_disconnect(lod->lod_child_exp);
425 CERROR("error in disconnect from storage: %d\n", rc);
429 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
440 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
442 * The method is called once the recovery is complete. This implementation
443 * distributes the notification to all the known targets.
445 * see include/lu_object.h for the details
447 static int lod_recovery_complete(const struct lu_env *env,
448 struct lu_device *dev)
450 struct lod_device *lod = lu2lod_dev(dev);
451 struct lu_device *next = &lod->lod_child->dd_lu_dev;
456 LASSERT(lod->lod_recovery_completed == 0);
457 lod->lod_recovery_completed = 1;
459 rc = next->ld_ops->ldo_recovery_complete(env, next);
461 lod_getref(&lod->lod_ost_descs);
462 if (lod->lod_osts_size > 0) {
463 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
464 struct lod_tgt_desc *tgt;
465 tgt = OST_TGT(lod, i);
466 LASSERT(tgt && tgt->ltd_tgt);
467 next = &tgt->ltd_ost->dd_lu_dev;
468 rc = next->ld_ops->ldo_recovery_complete(env, next);
470 CERROR("%s: can't complete recovery on #%d:"
471 "%d\n", lod2obd(lod)->obd_name, i, rc);
474 lod_putref(lod, &lod->lod_ost_descs);
479 * Implementation of lu_device_operations::ldo_prepare() for LOD
481 * see include/lu_object.h for the details.
483 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
484 struct lu_device *cdev)
486 struct lod_device *lod = lu2lod_dev(cdev);
487 struct lu_device *next = &lod->lod_child->dd_lu_dev;
491 rc = next->ld_ops->ldo_prepare(env, pdev, next);
493 CERROR("%s: prepare bottom error: rc = %d\n",
494 lod2obd(lod)->obd_name, rc);
498 lod->lod_initialized = 1;
503 const struct lu_device_operations lod_lu_ops = {
504 .ldo_object_alloc = lod_object_alloc,
505 .ldo_process_config = lod_process_config,
506 .ldo_recovery_complete = lod_recovery_complete,
507 .ldo_prepare = lod_prepare,
511 * Implementation of dt_device_operations::dt_root_get() for LOD
513 * see include/dt_object.h for the details.
515 static int lod_root_get(const struct lu_env *env,
516 struct dt_device *dev, struct lu_fid *f)
518 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
522 * Implementation of dt_device_operations::dt_statfs() for LOD
524 * see include/dt_object.h for the details.
526 static int lod_statfs(const struct lu_env *env,
527 struct dt_device *dev, struct obd_statfs *sfs)
529 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
533 * Implementation of dt_device_operations::dt_trans_create() for LOD
535 * Creates a transaction using local (to this node) OSD.
537 * see include/dt_object.h for the details.
539 static struct thandle *lod_trans_create(const struct lu_env *env,
540 struct dt_device *dev)
544 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
552 * Implementation of dt_device_operations::dt_trans_start() for LOD
554 * Starts the set of local transactions using the targets involved
555 * in declare phase. Initial support for the distributed transactions.
557 * see include/dt_object.h for the details.
559 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
562 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
565 if (unlikely(th->th_update != NULL)) {
566 struct thandle_update *tu = th->th_update;
567 struct dt_update_request *update;
569 list_for_each_entry(update, &tu->tu_remote_update_list,
571 LASSERT(update->dur_dt != NULL);
572 rc = dt_trans_start(env, update->dur_dt, th);
577 return dt_trans_start(env, lod->lod_child, th);
581 * Implementation of dt_device_operations::dt_trans_stop() for LOD
583 * Stops the set of local transactions using the targets involved
584 * in declare phase. Initial support for the distributed transactions.
586 * see include/dt_object.h for the details.
588 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
591 struct thandle_update *tu = th->th_update;
592 struct dt_update_request *update;
593 struct dt_update_request *tmp;
598 rc = dt_trans_stop(env, th->th_dev, th);
599 if (likely(tu == NULL))
602 list_for_each_entry_safe(update, tmp,
603 &tu->tu_remote_update_list,
605 /* update will be freed inside dt_trans_stop */
606 rc2 = dt_trans_stop(env, update->dur_dt, th);
607 if (unlikely(rc2 != 0 && rc == 0))
615 * Implementation of dt_device_operations::dt_conf_get() for LOD
617 * Currently returns the configuration provided by the local OSD.
619 * see include/dt_object.h for the details.
621 static void lod_conf_get(const struct lu_env *env,
622 const struct dt_device *dev,
623 struct dt_device_param *param)
625 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
629 * Implementation of dt_device_operations::dt_sync() for LOD
631 * Syncs all known OST targets. Very very expensive and used
632 * rarely by LFSCK now. Should not be used in general.
634 * see include/dt_object.h for the details.
636 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
638 struct lod_device *lod = dt2lod_dev(dev);
639 struct lod_ost_desc *ost;
644 lod_getref(&lod->lod_ost_descs);
645 lod_foreach_ost(lod, i) {
646 ost = OST_TGT(lod, i);
647 LASSERT(ost && ost->ltd_ost);
648 rc = dt_sync(env, ost->ltd_ost);
650 CERROR("%s: can't sync %u: %d\n",
651 lod2obd(lod)->obd_name, i, rc);
655 lod_putref(lod, &lod->lod_ost_descs);
657 rc = dt_sync(env, lod->lod_child);
663 * Implementation of dt_device_operations::dt_ro() for LOD
665 * Turns local OSD read-only, used for the testing only.
667 * see include/dt_object.h for the details.
669 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
671 return dt_ro(env, dt2lod_dev(dev)->lod_child);
675 * Implementation of dt_device_operations::dt_commit_async() for LOD
677 * Asks local OSD to commit sooner.
679 * see include/dt_object.h for the details.
681 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
683 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
689 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
690 int mode, unsigned long timeout,
691 __u32 alg, struct lustre_capa_key *keys)
693 struct dt_device *next = dt2lod_dev(dev)->lod_child;
694 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
697 static const struct dt_device_operations lod_dt_ops = {
698 .dt_root_get = lod_root_get,
699 .dt_statfs = lod_statfs,
700 .dt_trans_create = lod_trans_create,
701 .dt_trans_start = lod_trans_start,
702 .dt_trans_stop = lod_trans_stop,
703 .dt_conf_get = lod_conf_get,
706 .dt_commit_async = lod_commit_async,
707 .dt_init_capa_ctxt = lod_init_capa_ctxt,
711 * Connect to a local OSD.
713 * Used to connect to the local OSD at mount. OSD name is taken from the
714 * configuration command passed. This connection is used to identify LU
715 * site and pin the OSD from early removal.
717 * \param[in] env LU environment provided by the caller
718 * \param[in] lod lod device
719 * \param[in] cfg configuration command to apply
721 * \retval 0 on success
722 * \retval negative negated errno on error
724 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
725 struct lustre_cfg *cfg)
727 struct obd_connect_data *data = NULL;
728 struct obd_device *obd;
729 char *nextdev = NULL, *p, *s;
734 LASSERT(lod->lod_child_exp == NULL);
736 /* compatibility hack: we still use old config logs
737 * which specify LOV, but we need to learn underlying
738 * OSD device, which is supposed to be:
739 * <fsname>-MDTxxxx-osd
741 * 2.x MGS generates lines like the following:
742 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
743 * 1.8 MGS generates lines like the following:
744 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
746 * we use "-MDT" to differentiate 2.x from 1.8 */
748 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
750 OBD_ALLOC(nextdev, len);
752 GOTO(out, rc = -ENOMEM);
755 s = strstr(nextdev, "-mdtlov");
756 if (unlikely(s == NULL)) {
757 CERROR("unable to parse device name %s\n",
758 lustre_cfg_string(cfg, 0));
759 GOTO(out, rc = -EINVAL);
762 if (strstr(nextdev, "-MDT")) {
767 strcpy(s, "-MDT0000-osd");
770 CERROR("unable to parse device name %s\n",
771 lustre_cfg_string(cfg, 0));
772 GOTO(out, rc = -EINVAL);
777 GOTO(out, rc = -ENOMEM);
779 obd = class_name2obd(nextdev);
781 CERROR("can not locate next device: %s\n", nextdev);
782 GOTO(out, rc = -ENOTCONN);
785 data->ocd_connect_flags = OBD_CONNECT_VERSION;
786 data->ocd_version = LUSTRE_VERSION_CODE;
788 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
791 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
795 lod->lod_dt_dev.dd_lu_dev.ld_site =
796 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
797 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
798 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
804 OBD_FREE(nextdev, len);
809 * Allocate and initialize target table.
811 * A helper function to initialize the target table and allocate
812 * a bitmap of the available targets.
814 * \param[in] ltd target's table to initialize
816 * \retval 0 on success
817 * \retval negative negated errno on error
819 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
821 mutex_init(<d->ltd_mutex);
822 init_rwsem(<d->ltd_rw_sem);
824 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
825 * added to the LOD, see lod_add_device() */
826 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
827 if (ltd->ltd_tgt_bitmap == NULL)
830 ltd->ltd_tgts_size = 32;
833 ltd->ltd_death_row = 0;
834 ltd->ltd_refcount = 0;
839 * Initialize LOD device at setup.
841 * Initializes the given LOD device using the original configuration command.
842 * The function initiates a connection to the local OSD and initializes few
843 * internal structures like pools, target tables, etc.
845 * \param[in] env LU environment provided by the caller
846 * \param[in] lod lod device
847 * \param[in] ldt not used
848 * \param[in] cfg configuration command
850 * \retval 0 on success
851 * \retval negative negated errno on error
853 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
854 struct lu_device_type *ldt, struct lustre_cfg *cfg)
856 struct dt_device_param ddp;
857 struct obd_device *obd;
861 obd = class_name2obd(lustre_cfg_string(cfg, 0));
863 CERROR("Cannot find obd with name %s\n",
864 lustre_cfg_string(cfg, 0));
868 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
869 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
870 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
871 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
873 rc = lod_connect_to_osd(env, lod, cfg);
877 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
878 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
880 /* setup obd to be used with old lov code */
881 rc = lod_pools_init(lod, cfg);
883 GOTO(out_disconnect, rc);
885 rc = lod_procfs_init(lod);
889 spin_lock_init(&lod->lod_desc_lock);
890 spin_lock_init(&lod->lod_connects_lock);
891 lod_tgt_desc_init(&lod->lod_mdt_descs);
892 lod_tgt_desc_init(&lod->lod_ost_descs);
899 obd_disconnect(lod->lod_child_exp);
904 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
906 * Releases the memory allocated for LOD device.
908 * see include/lu_object.h for the details.
910 static struct lu_device *lod_device_free(const struct lu_env *env,
911 struct lu_device *lu)
913 struct lod_device *lod = lu2lod_dev(lu);
914 struct lu_device *next = &lod->lod_child->dd_lu_dev;
917 LASSERT(atomic_read(&lu->ld_ref) == 0);
918 dt_device_fini(&lod->lod_dt_dev);
924 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
926 * Allocates LOD device and calls the helpers to initialize it.
928 * see include/lu_object.h for the details.
930 static struct lu_device *lod_device_alloc(const struct lu_env *env,
931 struct lu_device_type *type,
932 struct lustre_cfg *lcfg)
934 struct lod_device *lod;
935 struct lu_device *lu_dev;
939 lu_dev = ERR_PTR(-ENOMEM);
943 lu_dev = lod2lu_dev(lod);
944 dt_device_init(&lod->lod_dt_dev, type);
945 rc = lod_init0(env, lod, type, lcfg);
947 lod_device_free(env, lu_dev);
948 lu_dev = ERR_PTR(rc);
956 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
958 * Releases the internal resources used by LOD device.
960 * see include/lu_object.h for the details.
962 static struct lu_device *lod_device_fini(const struct lu_env *env,
965 struct lod_device *lod = lu2lod_dev(d);
971 lod_procfs_fini(lod);
973 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
975 CERROR("%s:can not fini ost descs %d\n",
976 lod2obd(lod)->obd_name, rc);
978 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
980 CERROR("%s:can not fini mdt descs %d\n",
981 lod2obd(lod)->obd_name, rc);
987 * Implementation of obd_ops::o_connect() for LOD
989 * Used to track all the users of this specific LOD device,
990 * so the device stays up until the last user disconnected.
992 * \param[in] env LU environment provided by the caller
993 * \param[out] exp export the caller will be using to access LOD
994 * \param[in] obd OBD device representing LOD device
995 * \param[in] cluuid unique identifier of the caller
996 * \param[in] data not used
997 * \param[in] localdata not used
999 * \retval 0 on success
1000 * \retval negative negated errno on error
1002 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1003 struct obd_device *obd, struct obd_uuid *cluuid,
1004 struct obd_connect_data *data, void *localdata)
1006 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1007 struct lustre_handle conn;
1011 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1013 rc = class_connect(&conn, obd, cluuid);
1017 *exp = class_conn2export(&conn);
1019 spin_lock(&lod->lod_connects_lock);
1020 lod->lod_connects++;
1021 /* at the moment we expect the only user */
1022 LASSERT(lod->lod_connects == 1);
1023 spin_unlock(&lod->lod_connects_lock);
1030 * Implementation of obd_ops::o_disconnect() for LOD
1032 * When the caller doesn't need to use this LOD instance, it calls
1033 * obd_disconnect() and LOD releases corresponding export/reference count.
1034 * Once all the users gone, LOD device is released.
1036 * \param[in] exp export provided to the caller in obd_connect()
1038 * \retval 0 on success
1039 * \retval negative negated errno on error
1041 static int lod_obd_disconnect(struct obd_export *exp)
1043 struct obd_device *obd = exp->exp_obd;
1044 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1045 int rc, release = 0;
1048 /* Only disconnect the underlying layers on the final disconnect. */
1049 spin_lock(&lod->lod_connects_lock);
1050 lod->lod_connects--;
1051 if (lod->lod_connects != 0) {
1052 /* why should there be more than 1 connect? */
1053 spin_unlock(&lod->lod_connects_lock);
1054 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1058 spin_unlock(&lod->lod_connects_lock);
1060 /* the last user of lod has gone, let's release the device */
1064 rc = class_disconnect(exp); /* bz 9811 */
1066 if (rc == 0 && release)
1067 class_manual_cleanup(obd);
1071 LU_KEY_INIT(lod, struct lod_thread_info);
1073 static void lod_key_fini(const struct lu_context *ctx,
1074 struct lu_context_key *key, void *data)
1076 struct lod_thread_info *info = data;
1077 /* allocated in lod_get_lov_ea
1078 * XXX: this is overload, a tread may have such store but used only
1079 * once. Probably better would be pool of such stores per LOD.
1081 if (info->lti_ea_store) {
1082 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1083 info->lti_ea_store = NULL;
1084 info->lti_ea_store_size = 0;
1086 lu_buf_free(&info->lti_linkea_buf);
1090 /* context key: lod_thread_key */
1091 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1093 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1095 static struct lu_device_type_operations lod_device_type_ops = {
1096 .ldto_init = lod_type_init,
1097 .ldto_fini = lod_type_fini,
1099 .ldto_start = lod_type_start,
1100 .ldto_stop = lod_type_stop,
1102 .ldto_device_alloc = lod_device_alloc,
1103 .ldto_device_free = lod_device_free,
1105 .ldto_device_fini = lod_device_fini
1108 static struct lu_device_type lod_device_type = {
1109 .ldt_tags = LU_DEVICE_DT,
1110 .ldt_name = LUSTRE_LOD_NAME,
1111 .ldt_ops = &lod_device_type_ops,
1112 .ldt_ctx_tags = LCT_MD_THREAD,
1116 * Implementation of obd_ops::o_get_info() for LOD
1118 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1119 * the caller binary status whether LOD has seen connection to any OST target.
1121 * \param[in] env LU environment provided by the caller
1122 * \param[in] exp export of the caller
1123 * \param[in] keylen len of the key
1124 * \param[in] key the key
1125 * \param[in] vallen not used
1126 * \param[in] val not used
1127 * \param[in] lsm not used
1129 * \retval 0 if a connection was seen
1130 * \retval -EAGAIN if LOD isn't running yet or no
1131 * connection has been seen yet
1132 * \retval -EINVAL if not supported key is requested
1134 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1135 __u32 keylen, void *key, __u32 *vallen, void *val,
1136 struct lov_stripe_md *lsm)
1140 if (KEY_IS(KEY_OSP_CONNECTED)) {
1141 struct obd_device *obd = exp->exp_obd;
1142 struct lod_device *d;
1143 struct lod_ost_desc *ost;
1147 if (!obd->obd_set_up || obd->obd_stopping)
1150 d = lu2lod_dev(obd->obd_lu_dev);
1151 lod_getref(&d->lod_ost_descs);
1152 lod_foreach_ost(d, i) {
1153 ost = OST_TGT(d, i);
1154 LASSERT(ost && ost->ltd_ost);
1156 rc = obd_get_info(env, ost->ltd_exp, keylen, key,
1158 /* one healthy device is enough */
1162 lod_putref(d, &d->lod_ost_descs);
1169 static struct obd_ops lod_obd_device_ops = {
1170 .o_owner = THIS_MODULE,
1171 .o_connect = lod_obd_connect,
1172 .o_disconnect = lod_obd_disconnect,
1173 .o_get_info = lod_obd_get_info,
1174 .o_pool_new = lod_pool_new,
1175 .o_pool_rem = lod_pool_remove,
1176 .o_pool_add = lod_pool_add,
1177 .o_pool_del = lod_pool_del,
1180 static int __init lod_mod_init(void)
1182 struct obd_type *type;
1185 rc = lu_kmem_init(lod_caches);
1189 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
1190 LUSTRE_LOD_NAME, &lod_device_type);
1192 lu_kmem_fini(lod_caches);
1196 /* create "lov" entry in procfs for compatibility purposes */
1197 type = class_search_type(LUSTRE_LOV_NAME);
1198 if (type != NULL && type->typ_procroot != NULL)
1201 type = class_search_type(LUSTRE_LOD_NAME);
1202 type->typ_procsym = lprocfs_seq_register("lov", proc_lustre_root,
1204 if (IS_ERR(type->typ_procsym)) {
1205 CERROR("lod: can't create compat entry \"lov\": %d\n",
1206 (int)PTR_ERR(type->typ_procsym));
1207 type->typ_procsym = NULL;
1212 static void __exit lod_mod_exit(void)
1214 class_unregister_type(LUSTRE_LOD_NAME);
1215 lu_kmem_fini(lod_caches);
1218 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1219 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1220 MODULE_LICENSE("GPL");
1222 module_init(lod_mod_init);
1223 module_exit(lod_mod_exit);