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 LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
127 if (fid_is_idif(fid)) {
128 *tgt = fid_idif_ost_idx(fid);
129 *type = LU_SEQ_RANGE_OST;
133 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
134 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
136 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
137 *type = LU_SEQ_RANGE_MDT;
141 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
142 fld_range_set_type(&range, *type);
143 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
147 *tgt = range.lsr_index;
148 *type = range.lsr_flags;
150 CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
151 LPX64"\n", *tgt, fid_seq(fid));
156 extern struct lu_object_operations lod_lu_obj_ops;
157 extern struct dt_object_operations lod_obj_ops;
159 /* Slab for OSD object allocation */
160 struct kmem_cache *lod_object_kmem;
162 static struct lu_kmem_descr lod_caches[] = {
164 .ckd_cache = &lod_object_kmem,
165 .ckd_name = "lod_obj",
166 .ckd_size = sizeof(struct lod_object)
173 static struct lu_device *lod_device_fini(const struct lu_env *env,
174 struct lu_device *d);
177 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
179 * Allocates and initializes LOD's slice in the given object.
181 * see include/lu_object.h for the details.
183 struct lu_object *lod_object_alloc(const struct lu_env *env,
184 const struct lu_object_header *hdr,
185 struct lu_device *dev)
187 struct lod_object *lod_obj;
188 struct lu_object *lu_obj;
191 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
193 RETURN(ERR_PTR(-ENOMEM));
195 lu_obj = lod2lu_obj(lod_obj);
196 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
197 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
198 lu_obj->lo_ops = &lod_lu_obj_ops;
204 * Cleanup table of target's descriptors.
206 * The function goes through all the targets in the given table
207 * and apply given configuration command on to the targets.
208 * Used to cleanup the targets at unmount.
210 * \param[in] env LU environment provided by the caller
211 * \param[in] lod lod device
212 * \param[in] ltd target's table to go through
213 * \param[in] lcfg configuration command to apply
215 * \retval 0 on success
216 * \retval negative negated errno on error
218 static int lod_cleanup_desc_tgts(const struct lu_env *env,
219 struct lod_device *lod,
220 struct lod_tgt_descs *ltd,
221 struct lustre_cfg *lcfg)
223 struct lu_device *next;
228 if (ltd->ltd_tgts_size <= 0) {
229 lod_putref(lod, ltd);
232 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
233 struct lod_tgt_desc *tgt;
236 tgt = LTD_TGT(ltd, i);
237 LASSERT(tgt && tgt->ltd_tgt);
238 next = &tgt->ltd_tgt->dd_lu_dev;
239 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
241 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
242 ": rc = %d\n", lod2obd(lod)->obd_name, i,
243 lcfg->lcfg_command, rc1);
247 lod_putref(lod, ltd);
252 * Extract MDT target index from a device name.
254 * a helper function to extract index from the given device name
255 * like "fsname-MDTxxxx-mdtlov"
257 * \param[in] lodname device name
258 * \param[out] index extracted index
260 * \retval 0 on success
261 * \retval -EINVAL if the name is invalid
263 static int lodname2mdt_index(char *lodname, long *index)
267 ptr = strrchr(lodname, '-');
269 CERROR("invalid MDT index in '%s'\n", lodname);
273 if (strncmp(ptr, "-mdtlov", 7) != 0) {
274 CERROR("invalid MDT index in '%s'\n", lodname);
278 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
279 CERROR("invalid MDT index in '%s'\n", lodname);
283 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
284 CERROR("invalid MDT index in '%s'\n", lodname);
288 *index = simple_strtol(ptr - 4, &tmp, 16);
289 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
290 CERROR("invalid MDT index in '%s'\n", lodname);
297 * Implementation of lu_device_operations::ldo_process_config() for LOD
299 * The method is called by the configuration subsystem during setup,
300 * cleanup and when the configuration changes. The method processes
301 * few specific commands like adding/removing the targets, changing
302 * the runtime parameters.
304 * \param[in] env LU environment provided by the caller
305 * \param[in] dev lod device
306 * \param[in] lcfg configuration command to apply
308 * \retval 0 on success
309 * \retval negative negated errno on error
311 * The examples are below.
313 * Add osc config log:
314 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
315 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
316 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
317 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
318 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
319 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
321 * Add mdc config log:
322 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
323 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
324 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
325 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
326 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
327 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
329 static int lod_process_config(const struct lu_env *env,
330 struct lu_device *dev,
331 struct lustre_cfg *lcfg)
333 struct lod_device *lod = lu2lod_dev(dev);
334 struct lu_device *next = &lod->lod_child->dd_lu_dev;
339 switch(lcfg->lcfg_command) {
340 case LCFG_LOV_DEL_OBD:
341 case LCFG_LOV_ADD_INA:
342 case LCFG_LOV_ADD_OBD:
347 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
348 * modify_mdc_tgts add 0:lustre-MDT0001
349 * 1:lustre-MDT0001-mdc0002
351 arg1 = lustre_cfg_string(lcfg, 1);
353 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
354 GOTO(out, rc = -EINVAL);
355 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
356 GOTO(out, rc = -EINVAL);
358 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
360 mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
361 /* 1.8 configs don't have "-MDT0000" at the end */
366 rc = lodname2mdt_index(
367 lustre_cfg_string(lcfg, 0),
371 mdt_index = long_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);
379 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
380 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
382 rc = lod_add_device(env, lod, arg1, index, gen,
383 mdt_index, LUSTRE_OSC_NAME, 0);
385 rc = lod_del_device(env, lod,
387 arg1, index, gen, true);
394 struct obd_device *obd = lod2obd(lod);
396 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
403 case LCFG_PRE_CLEANUP: {
404 lu_dev_del_linkage(dev->ld_site, dev);
405 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
406 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
407 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
410 * do cleanup on underlying storage only when
411 * all OSPs are cleaned up, as they use that OSD as well
413 next = &lod->lod_child->dd_lu_dev;
414 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
416 CERROR("%s: can't process %u: %d\n",
417 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
419 rc = obd_disconnect(lod->lod_child_exp);
421 CERROR("error in disconnect from storage: %d\n", rc);
425 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
436 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
438 * The method is called once the recovery is complete. This implementation
439 * distributes the notification to all the known targets.
441 * see include/lu_object.h for the details
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;
452 LASSERT(lod->lod_recovery_completed == 0);
453 lod->lod_recovery_completed = 1;
455 rc = next->ld_ops->ldo_recovery_complete(env, next);
457 lod_getref(&lod->lod_ost_descs);
458 if (lod->lod_osts_size > 0) {
459 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
460 struct lod_tgt_desc *tgt;
461 tgt = OST_TGT(lod, i);
462 LASSERT(tgt && tgt->ltd_tgt);
463 next = &tgt->ltd_ost->dd_lu_dev;
464 rc = next->ld_ops->ldo_recovery_complete(env, next);
466 CERROR("%s: can't complete recovery on #%d:"
467 "%d\n", lod2obd(lod)->obd_name, i, rc);
470 lod_putref(lod, &lod->lod_ost_descs);
475 * Implementation of lu_device_operations::ldo_prepare() for LOD
477 * see include/lu_object.h for the details.
479 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
480 struct lu_device *cdev)
482 struct lod_device *lod = lu2lod_dev(cdev);
483 struct lu_device *next = &lod->lod_child->dd_lu_dev;
487 rc = next->ld_ops->ldo_prepare(env, pdev, next);
489 CERROR("%s: prepare bottom error: rc = %d\n",
490 lod2obd(lod)->obd_name, rc);
494 lod->lod_initialized = 1;
499 const struct lu_device_operations lod_lu_ops = {
500 .ldo_object_alloc = lod_object_alloc,
501 .ldo_process_config = lod_process_config,
502 .ldo_recovery_complete = lod_recovery_complete,
503 .ldo_prepare = lod_prepare,
507 * Implementation of dt_device_operations::dt_root_get() for LOD
509 * see include/dt_object.h for the details.
511 static int lod_root_get(const struct lu_env *env,
512 struct dt_device *dev, struct lu_fid *f)
514 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
518 * Implementation of dt_device_operations::dt_statfs() for LOD
520 * see include/dt_object.h for the details.
522 static int lod_statfs(const struct lu_env *env,
523 struct dt_device *dev, struct obd_statfs *sfs)
525 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
529 * Implementation of dt_device_operations::dt_trans_create() for LOD
531 * Creates a transaction using local (to this node) OSD.
533 * see include/dt_object.h for the details.
535 static struct thandle *lod_trans_create(const struct lu_env *env,
536 struct dt_device *dev)
540 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
548 * Implementation of dt_device_operations::dt_trans_start() for LOD
550 * Starts the set of local transactions using the targets involved
551 * in declare phase. Initial support for the distributed transactions.
553 * see include/dt_object.h for the details.
555 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
558 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
561 if (unlikely(th->th_update != NULL)) {
562 struct thandle_update *tu = th->th_update;
563 struct dt_update_request *update;
565 list_for_each_entry(update, &tu->tu_remote_update_list,
567 LASSERT(update->dur_dt != NULL);
568 rc = dt_trans_start(env, update->dur_dt, th);
573 return dt_trans_start(env, lod->lod_child, th);
577 * Implementation of dt_device_operations::dt_trans_stop() for LOD
579 * Stops the set of local transactions using the targets involved
580 * in declare phase. Initial support for the distributed transactions.
582 * see include/dt_object.h for the details.
584 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
587 struct thandle_update *tu = th->th_update;
588 struct dt_update_request *update;
589 struct dt_update_request *tmp;
594 rc = dt_trans_stop(env, th->th_dev, th);
595 if (likely(tu == NULL))
598 list_for_each_entry_safe(update, tmp,
599 &tu->tu_remote_update_list,
601 /* update will be freed inside dt_trans_stop */
602 rc2 = dt_trans_stop(env, update->dur_dt, th);
603 if (unlikely(rc2 != 0 && rc == 0))
611 * Implementation of dt_device_operations::dt_conf_get() for LOD
613 * Currently returns the configuration provided by the local OSD.
615 * see include/dt_object.h for the details.
617 static void lod_conf_get(const struct lu_env *env,
618 const struct dt_device *dev,
619 struct dt_device_param *param)
621 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
625 * Implementation of dt_device_operations::dt_sync() for LOD
627 * Syncs all known OST targets. Very very expensive and used
628 * rarely by LFSCK now. Should not be used in general.
630 * see include/dt_object.h for the details.
632 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
634 struct lod_device *lod = dt2lod_dev(dev);
635 struct lod_ost_desc *ost;
640 lod_getref(&lod->lod_ost_descs);
641 lod_foreach_ost(lod, i) {
642 ost = OST_TGT(lod, i);
643 LASSERT(ost && ost->ltd_ost);
644 rc = dt_sync(env, ost->ltd_ost);
646 CERROR("%s: can't sync %u: %d\n",
647 lod2obd(lod)->obd_name, i, rc);
651 lod_putref(lod, &lod->lod_ost_descs);
653 rc = dt_sync(env, lod->lod_child);
659 * Implementation of dt_device_operations::dt_ro() for LOD
661 * Turns local OSD read-only, used for the testing only.
663 * see include/dt_object.h for the details.
665 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
667 return dt_ro(env, dt2lod_dev(dev)->lod_child);
671 * Implementation of dt_device_operations::dt_commit_async() for LOD
673 * Asks local OSD to commit sooner.
675 * see include/dt_object.h for the details.
677 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
679 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
685 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
686 int mode, unsigned long timeout,
687 __u32 alg, struct lustre_capa_key *keys)
689 struct dt_device *next = dt2lod_dev(dev)->lod_child;
690 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
693 static const struct dt_device_operations lod_dt_ops = {
694 .dt_root_get = lod_root_get,
695 .dt_statfs = lod_statfs,
696 .dt_trans_create = lod_trans_create,
697 .dt_trans_start = lod_trans_start,
698 .dt_trans_stop = lod_trans_stop,
699 .dt_conf_get = lod_conf_get,
702 .dt_commit_async = lod_commit_async,
703 .dt_init_capa_ctxt = lod_init_capa_ctxt,
707 * Connect to a local OSD.
709 * Used to connect to the local OSD at mount. OSD name is taken from the
710 * configuration command passed. This connection is used to identify LU
711 * site and pin the OSD from early removal.
713 * \param[in] env LU environment provided by the caller
714 * \param[in] lod lod device
715 * \param[in] cfg configuration command to apply
717 * \retval 0 on success
718 * \retval negative negated errno on error
720 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
721 struct lustre_cfg *cfg)
723 struct obd_connect_data *data = NULL;
724 struct obd_device *obd;
725 char *nextdev = NULL, *p, *s;
730 LASSERT(lod->lod_child_exp == NULL);
732 /* compatibility hack: we still use old config logs
733 * which specify LOV, but we need to learn underlying
734 * OSD device, which is supposed to be:
735 * <fsname>-MDTxxxx-osd
737 * 2.x MGS generates lines like the following:
738 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
739 * 1.8 MGS generates lines like the following:
740 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
742 * we use "-MDT" to differentiate 2.x from 1.8 */
744 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
746 OBD_ALLOC(nextdev, len);
748 GOTO(out, rc = -ENOMEM);
751 s = strstr(nextdev, "-mdtlov");
752 if (unlikely(s == NULL)) {
753 CERROR("unable to parse device name %s\n",
754 lustre_cfg_string(cfg, 0));
755 GOTO(out, rc = -EINVAL);
758 if (strstr(nextdev, "-MDT")) {
763 strcpy(s, "-MDT0000-osd");
766 CERROR("unable to parse device name %s\n",
767 lustre_cfg_string(cfg, 0));
768 GOTO(out, rc = -EINVAL);
773 GOTO(out, rc = -ENOMEM);
775 obd = class_name2obd(nextdev);
777 CERROR("can not locate next device: %s\n", nextdev);
778 GOTO(out, rc = -ENOTCONN);
781 data->ocd_connect_flags = OBD_CONNECT_VERSION;
782 data->ocd_version = LUSTRE_VERSION_CODE;
784 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
787 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
791 lod->lod_dt_dev.dd_lu_dev.ld_site =
792 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
793 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
794 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
800 OBD_FREE(nextdev, len);
805 * Allocate and initialize target table.
807 * A helper function to initialize the target table and allocate
808 * a bitmap of the available targets.
810 * \param[in] ltd target's table to initialize
812 * \retval 0 on success
813 * \retval negative negated errno on error
815 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
817 mutex_init(<d->ltd_mutex);
818 init_rwsem(<d->ltd_rw_sem);
820 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
821 * added to the LOD, see lod_add_device() */
822 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
823 if (ltd->ltd_tgt_bitmap == NULL)
826 ltd->ltd_tgts_size = 32;
829 ltd->ltd_death_row = 0;
830 ltd->ltd_refcount = 0;
835 * Initialize LOD device at setup.
837 * Initializes the given LOD device using the original configuration command.
838 * The function initiates a connection to the local OSD and initializes few
839 * internal structures like pools, target tables, etc.
841 * \param[in] env LU environment provided by the caller
842 * \param[in] lod lod device
843 * \param[in] ldt not used
844 * \param[in] cfg configuration command
846 * \retval 0 on success
847 * \retval negative negated errno on error
849 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
850 struct lu_device_type *ldt, struct lustre_cfg *cfg)
852 struct dt_device_param ddp;
853 struct obd_device *obd;
857 obd = class_name2obd(lustre_cfg_string(cfg, 0));
859 CERROR("Cannot find obd with name %s\n",
860 lustre_cfg_string(cfg, 0));
864 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
865 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
866 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
867 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
869 rc = lod_connect_to_osd(env, lod, cfg);
873 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
874 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
876 /* setup obd to be used with old lov code */
877 rc = lod_pools_init(lod, cfg);
879 GOTO(out_disconnect, rc);
881 rc = lod_procfs_init(lod);
885 spin_lock_init(&lod->lod_desc_lock);
886 spin_lock_init(&lod->lod_connects_lock);
887 lod_tgt_desc_init(&lod->lod_mdt_descs);
888 lod_tgt_desc_init(&lod->lod_ost_descs);
895 obd_disconnect(lod->lod_child_exp);
900 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
902 * Releases the memory allocated for LOD device.
904 * see include/lu_object.h for the details.
906 static struct lu_device *lod_device_free(const struct lu_env *env,
907 struct lu_device *lu)
909 struct lod_device *lod = lu2lod_dev(lu);
910 struct lu_device *next = &lod->lod_child->dd_lu_dev;
913 LASSERT(atomic_read(&lu->ld_ref) == 0);
914 dt_device_fini(&lod->lod_dt_dev);
920 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
922 * Allocates LOD device and calls the helpers to initialize it.
924 * see include/lu_object.h for the details.
926 static struct lu_device *lod_device_alloc(const struct lu_env *env,
927 struct lu_device_type *type,
928 struct lustre_cfg *lcfg)
930 struct lod_device *lod;
931 struct lu_device *lu_dev;
935 lu_dev = ERR_PTR(-ENOMEM);
939 lu_dev = lod2lu_dev(lod);
940 dt_device_init(&lod->lod_dt_dev, type);
941 rc = lod_init0(env, lod, type, lcfg);
943 lod_device_free(env, lu_dev);
944 lu_dev = ERR_PTR(rc);
952 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
954 * Releases the internal resources used by LOD device.
956 * see include/lu_object.h for the details.
958 static struct lu_device *lod_device_fini(const struct lu_env *env,
961 struct lod_device *lod = lu2lod_dev(d);
967 lod_procfs_fini(lod);
969 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
971 CERROR("%s:can not fini ost descs %d\n",
972 lod2obd(lod)->obd_name, rc);
974 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
976 CERROR("%s:can not fini mdt descs %d\n",
977 lod2obd(lod)->obd_name, rc);
983 * Implementation of obd_ops::o_connect() for LOD
985 * Used to track all the users of this specific LOD device,
986 * so the device stays up until the last user disconnected.
988 * \param[in] env LU environment provided by the caller
989 * \param[out] exp export the caller will be using to access LOD
990 * \param[in] obd OBD device representing LOD device
991 * \param[in] cluuid unique identifier of the caller
992 * \param[in] data not used
993 * \param[in] localdata not used
995 * \retval 0 on success
996 * \retval negative negated errno on error
998 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
999 struct obd_device *obd, struct obd_uuid *cluuid,
1000 struct obd_connect_data *data, void *localdata)
1002 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1003 struct lustre_handle conn;
1007 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1009 rc = class_connect(&conn, obd, cluuid);
1013 *exp = class_conn2export(&conn);
1015 spin_lock(&lod->lod_connects_lock);
1016 lod->lod_connects++;
1017 /* at the moment we expect the only user */
1018 LASSERT(lod->lod_connects == 1);
1019 spin_unlock(&lod->lod_connects_lock);
1026 * Implementation of obd_ops::o_disconnect() for LOD
1028 * When the caller doesn't need to use this LOD instance, it calls
1029 * obd_disconnect() and LOD releases corresponding export/reference count.
1030 * Once all the users gone, LOD device is released.
1032 * \param[in] exp export provided to the caller in obd_connect()
1034 * \retval 0 on success
1035 * \retval negative negated errno on error
1037 static int lod_obd_disconnect(struct obd_export *exp)
1039 struct obd_device *obd = exp->exp_obd;
1040 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1041 int rc, release = 0;
1044 /* Only disconnect the underlying layers on the final disconnect. */
1045 spin_lock(&lod->lod_connects_lock);
1046 lod->lod_connects--;
1047 if (lod->lod_connects != 0) {
1048 /* why should there be more than 1 connect? */
1049 spin_unlock(&lod->lod_connects_lock);
1050 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1054 spin_unlock(&lod->lod_connects_lock);
1056 /* the last user of lod has gone, let's release the device */
1060 rc = class_disconnect(exp); /* bz 9811 */
1062 if (rc == 0 && release)
1063 class_manual_cleanup(obd);
1067 LU_KEY_INIT(lod, struct lod_thread_info);
1069 static void lod_key_fini(const struct lu_context *ctx,
1070 struct lu_context_key *key, void *data)
1072 struct lod_thread_info *info = data;
1073 /* allocated in lod_get_lov_ea
1074 * XXX: this is overload, a tread may have such store but used only
1075 * once. Probably better would be pool of such stores per LOD.
1077 if (info->lti_ea_store) {
1078 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1079 info->lti_ea_store = NULL;
1080 info->lti_ea_store_size = 0;
1082 lu_buf_free(&info->lti_linkea_buf);
1086 /* context key: lod_thread_key */
1087 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1089 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1091 static struct lu_device_type_operations lod_device_type_ops = {
1092 .ldto_init = lod_type_init,
1093 .ldto_fini = lod_type_fini,
1095 .ldto_start = lod_type_start,
1096 .ldto_stop = lod_type_stop,
1098 .ldto_device_alloc = lod_device_alloc,
1099 .ldto_device_free = lod_device_free,
1101 .ldto_device_fini = lod_device_fini
1104 static struct lu_device_type lod_device_type = {
1105 .ldt_tags = LU_DEVICE_DT,
1106 .ldt_name = LUSTRE_LOD_NAME,
1107 .ldt_ops = &lod_device_type_ops,
1108 .ldt_ctx_tags = LCT_MD_THREAD,
1112 * Implementation of obd_ops::o_get_info() for LOD
1114 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1115 * the caller binary status whether LOD has seen connection to any OST target.
1117 * \param[in] env LU environment provided by the caller
1118 * \param[in] exp export of the caller
1119 * \param[in] keylen len of the key
1120 * \param[in] key the key
1121 * \param[in] vallen not used
1122 * \param[in] val not used
1123 * \param[in] lsm not used
1125 * \retval 0 if a connection was seen
1126 * \retval -EAGAIN if LOD isn't running yet or no
1127 * connection has been seen yet
1128 * \retval -EINVAL if not supported key is requested
1130 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1131 __u32 keylen, void *key, __u32 *vallen, void *val,
1132 struct lov_stripe_md *lsm)
1136 if (KEY_IS(KEY_OSP_CONNECTED)) {
1137 struct obd_device *obd = exp->exp_obd;
1138 struct lod_device *d;
1139 struct lod_ost_desc *ost;
1143 if (!obd->obd_set_up || obd->obd_stopping)
1146 d = lu2lod_dev(obd->obd_lu_dev);
1147 lod_getref(&d->lod_ost_descs);
1148 lod_foreach_ost(d, i) {
1149 ost = OST_TGT(d, i);
1150 LASSERT(ost && ost->ltd_ost);
1152 rc = obd_get_info(env, ost->ltd_exp, keylen, key,
1154 /* one healthy device is enough */
1158 lod_putref(d, &d->lod_ost_descs);
1165 static struct obd_ops lod_obd_device_ops = {
1166 .o_owner = THIS_MODULE,
1167 .o_connect = lod_obd_connect,
1168 .o_disconnect = lod_obd_disconnect,
1169 .o_get_info = lod_obd_get_info,
1170 .o_pool_new = lod_pool_new,
1171 .o_pool_rem = lod_pool_remove,
1172 .o_pool_add = lod_pool_add,
1173 .o_pool_del = lod_pool_del,
1176 static int __init lod_mod_init(void)
1178 struct obd_type *type;
1181 rc = lu_kmem_init(lod_caches);
1185 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
1186 LUSTRE_LOD_NAME, &lod_device_type);
1188 lu_kmem_fini(lod_caches);
1192 /* create "lov" entry in procfs for compatibility purposes */
1193 type = class_search_type(LUSTRE_LOV_NAME);
1194 if (type != NULL && type->typ_procroot != NULL)
1197 type = class_search_type(LUSTRE_LOD_NAME);
1198 type->typ_procsym = lprocfs_seq_register("lov", proc_lustre_root,
1200 if (IS_ERR(type->typ_procsym)) {
1201 CERROR("lod: can't create compat entry \"lov\": %d\n",
1202 (int)PTR_ERR(type->typ_procsym));
1203 type->typ_procsym = NULL;
1208 static void __exit lod_mod_exit(void)
1210 class_unregister_type(LUSTRE_LOD_NAME);
1211 lu_kmem_fini(lod_caches);
1214 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1215 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1216 MODULE_LICENSE("GPL");
1218 module_init(lod_mod_init);
1219 module_exit(lod_mod_exit);