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, 2014, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/lod/lod_dev.c
34 * Lustre Logical Object Device
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mikhail Pershin <mike.pershin@intel.com>
40 * The Logical Object Device (LOD) layer manages access to striped
41 * objects (both regular files and directories). It implements the DT
42 * device and object APIs and is responsible for creating, storing,
43 * and loading striping information as an extended attribute of the
44 * underlying OSD object. LOD is the server side analog of the LOV and
45 * LMV layers on the client side.
47 * Metadata LU object stack (layers of the same compound LU object,
48 * all have the same FID):
58 * During LOD object initialization the localness or remoteness of the
59 * object FID dictates the choice between OSD and OSP.
61 * An LOD object (file or directory) with N stripes (each has a
68 * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
70 * When upper layers must access an object's stripes (which are
71 * themselves OST or MDT LU objects) LOD finds these objects by their
72 * FIDs and stores them as an array of DT object pointers on the
73 * object. Declarations and operations on LOD objects are received by
74 * LOD (as DT object operations) and performed on the underlying
75 * OS[DP] object and (as needed) on the stripes. From the perspective
76 * of LOD, a stripe-less file (created by mknod() or open with
77 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
78 * while a non-striped directory (created by mkdir()) is an object
79 * which will never have stripes.
81 * The LOD layer also implements a small subset of the OBD device API
82 * to support MDT stack initialization and finalization (an MDD device
83 * connects and disconnects itself to and from the underlying LOD
84 * device), and pool management. In turn LOD uses the OBD device API
85 * to connect it self to the underlying OSD, and to connect itself to
86 * OSP devices representing the MDTs and OSTs that bear the stripes of
90 #define DEBUG_SUBSYSTEM S_MDS
92 #include <obd_class.h>
93 #include <md_object.h>
94 #include <lustre_fid.h>
95 #include <lustre_param.h>
96 #include <lustre_update.h>
98 #include "lod_internal.h"
101 * Lookup target by FID.
103 * Lookup MDT/OST target index by FID. Type of the target can be
106 * \param[in] env LU environment provided by the caller
107 * \param[in] lod lod device
109 * \param[out] tgt result target index
110 * \param[in] type expected type of the target:
111 * LU_SEQ_RANGE_{MDT,OST,ANY}
113 * \retval 0 on success
114 * \retval negative negated errno on error
116 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
117 const struct lu_fid *fid, __u32 *tgt, int *type)
119 struct lu_seq_range range = { 0 };
120 struct lu_server_fld *server_fld;
124 if (!fid_is_sane(fid)) {
125 CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
130 if (fid_is_idif(fid)) {
131 *tgt = fid_idif_ost_idx(fid);
132 *type = LU_SEQ_RANGE_OST;
136 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
137 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
139 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
140 *type = LU_SEQ_RANGE_MDT;
144 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
145 fld_range_set_type(&range, *type);
146 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
150 *tgt = range.lsr_index;
151 *type = range.lsr_flags;
153 CDEBUG(D_INFO, "%s: got tgt %x for sequence: "LPX64"\n",
154 lod2obd(lod)->obd_name, *tgt, fid_seq(fid));
159 extern struct lu_object_operations lod_lu_obj_ops;
160 extern struct dt_object_operations lod_obj_ops;
162 /* Slab for OSD object allocation */
163 struct kmem_cache *lod_object_kmem;
165 static struct lu_kmem_descr lod_caches[] = {
167 .ckd_cache = &lod_object_kmem,
168 .ckd_name = "lod_obj",
169 .ckd_size = sizeof(struct lod_object)
176 static struct lu_device *lod_device_fini(const struct lu_env *env,
177 struct lu_device *d);
180 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
182 * Allocates and initializes LOD's slice in the given object.
184 * see include/lu_object.h for the details.
186 struct lu_object *lod_object_alloc(const struct lu_env *env,
187 const struct lu_object_header *hdr,
188 struct lu_device *dev)
190 struct lod_object *lod_obj;
191 struct lu_object *lu_obj;
194 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
196 RETURN(ERR_PTR(-ENOMEM));
198 lu_obj = lod2lu_obj(lod_obj);
199 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
200 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
201 lu_obj->lo_ops = &lod_lu_obj_ops;
207 * Cleanup table of target's descriptors.
209 * The function goes through all the targets in the given table
210 * and apply given configuration command on to the targets.
211 * Used to cleanup the targets at unmount.
213 * \param[in] env LU environment provided by the caller
214 * \param[in] lod lod device
215 * \param[in] ltd target's table to go through
216 * \param[in] lcfg configuration command to apply
218 * \retval 0 on success
219 * \retval negative negated errno on error
221 static int lod_cleanup_desc_tgts(const struct lu_env *env,
222 struct lod_device *lod,
223 struct lod_tgt_descs *ltd,
224 struct lustre_cfg *lcfg)
226 struct lu_device *next;
231 if (ltd->ltd_tgts_size <= 0) {
232 lod_putref(lod, ltd);
235 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
236 struct lod_tgt_desc *tgt;
239 tgt = LTD_TGT(ltd, i);
240 LASSERT(tgt && tgt->ltd_tgt);
241 next = &tgt->ltd_tgt->dd_lu_dev;
242 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
244 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
245 ": rc = %d\n", lod2obd(lod)->obd_name, i,
246 lcfg->lcfg_command, rc1);
250 lod_putref(lod, ltd);
255 * Extract MDT target index from a device name.
257 * a helper function to extract index from the given device name
258 * like "fsname-MDTxxxx-mdtlov"
260 * \param[in] lodname device name
261 * \param[out] index extracted index
263 * \retval 0 on success
264 * \retval -EINVAL if the name is invalid
266 static int lodname2mdt_index(char *lodname, long *index)
270 ptr = strrchr(lodname, '-');
272 CERROR("invalid MDT index in '%s'\n", lodname);
276 if (strncmp(ptr, "-mdtlov", 7) != 0) {
277 CERROR("invalid MDT index in '%s'\n", lodname);
281 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
282 CERROR("invalid MDT index in '%s'\n", lodname);
286 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
287 CERROR("invalid MDT index in '%s'\n", lodname);
291 *index = simple_strtol(ptr - 4, &tmp, 16);
292 if (*tmp != '-' || *index > INT_MAX || *index < 0) {
293 CERROR("invalid MDT index in '%s'\n", lodname);
300 * Implementation of lu_device_operations::ldo_process_config() for LOD
302 * The method is called by the configuration subsystem during setup,
303 * cleanup and when the configuration changes. The method processes
304 * few specific commands like adding/removing the targets, changing
305 * the runtime parameters.
307 * \param[in] env LU environment provided by the caller
308 * \param[in] dev lod device
309 * \param[in] lcfg configuration command to apply
311 * \retval 0 on success
312 * \retval negative negated errno on error
314 * The examples are below.
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'
324 * Add mdc config log:
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 */
369 rc = lodname2mdt_index(
370 lustre_cfg_string(lcfg, 0),
374 mdt_index = long_index;
376 rc = lod_add_device(env, lod, arg1, index, gen,
377 mdt_index, LUSTRE_OSC_NAME, 1);
378 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
380 rc = lod_add_device(env, lod, arg1, index, gen,
381 mdt_index, LUSTRE_MDC_NAME, 1);
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,
390 arg1, index, gen, true);
397 struct obd_device *obd = lod2obd(lod);
399 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
406 case LCFG_PRE_CLEANUP: {
407 lu_dev_del_linkage(dev->ld_site, dev);
408 lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
409 lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
410 if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
413 * do cleanup on underlying storage only when
414 * all OSPs are cleaned up, as they use that OSD as well
416 next = &lod->lod_child->dd_lu_dev;
417 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
419 CERROR("%s: can't process %u: %d\n",
420 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
422 rc = obd_disconnect(lod->lod_child_exp);
424 CERROR("error in disconnect from storage: %d\n", rc);
428 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
439 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
441 * The method is called once the recovery is complete. This implementation
442 * distributes the notification to all the known targets.
444 * see include/lu_object.h for the details
446 static int lod_recovery_complete(const struct lu_env *env,
447 struct lu_device *dev)
449 struct lod_device *lod = lu2lod_dev(dev);
450 struct lu_device *next = &lod->lod_child->dd_lu_dev;
455 LASSERT(lod->lod_recovery_completed == 0);
456 lod->lod_recovery_completed = 1;
458 rc = next->ld_ops->ldo_recovery_complete(env, next);
460 lod_getref(&lod->lod_ost_descs);
461 if (lod->lod_osts_size > 0) {
462 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
463 struct lod_tgt_desc *tgt;
464 tgt = OST_TGT(lod, i);
465 LASSERT(tgt && tgt->ltd_tgt);
466 next = &tgt->ltd_ost->dd_lu_dev;
467 rc = next->ld_ops->ldo_recovery_complete(env, next);
469 CERROR("%s: can't complete recovery on #%d:"
470 "%d\n", lod2obd(lod)->obd_name, i, rc);
473 lod_putref(lod, &lod->lod_ost_descs);
478 * Implementation of lu_device_operations::ldo_prepare() for LOD
480 * see include/lu_object.h for the details.
482 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
483 struct lu_device *cdev)
485 struct lod_device *lod = lu2lod_dev(cdev);
486 struct lu_device *next = &lod->lod_child->dd_lu_dev;
490 rc = next->ld_ops->ldo_prepare(env, pdev, next);
492 CERROR("%s: prepare bottom error: rc = %d\n",
493 lod2obd(lod)->obd_name, rc);
497 lod->lod_initialized = 1;
502 const struct lu_device_operations lod_lu_ops = {
503 .ldo_object_alloc = lod_object_alloc,
504 .ldo_process_config = lod_process_config,
505 .ldo_recovery_complete = lod_recovery_complete,
506 .ldo_prepare = lod_prepare,
510 * Implementation of dt_device_operations::dt_root_get() for LOD
512 * see include/dt_object.h for the details.
514 static int lod_root_get(const struct lu_env *env,
515 struct dt_device *dev, struct lu_fid *f)
517 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
521 * Implementation of dt_device_operations::dt_statfs() for LOD
523 * see include/dt_object.h for the details.
525 static int lod_statfs(const struct lu_env *env,
526 struct dt_device *dev, struct obd_statfs *sfs)
528 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
532 * Implementation of dt_device_operations::dt_trans_create() for LOD
534 * Creates a transaction using local (to this node) OSD.
536 * see include/dt_object.h for the details.
538 static struct thandle *lod_trans_create(const struct lu_env *env,
539 struct dt_device *dev)
543 th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
551 * Implementation of dt_device_operations::dt_trans_start() for LOD
553 * Starts the set of local transactions using the targets involved
554 * in declare phase. Initial support for the distributed transactions.
556 * see include/dt_object.h for the details.
558 static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
561 struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
564 if (unlikely(th->th_update != NULL)) {
565 struct thandle_update *tu = th->th_update;
566 struct dt_update_request *update;
568 list_for_each_entry(update, &tu->tu_remote_update_list,
570 LASSERT(update->dur_dt != NULL);
571 rc = dt_trans_start(env, update->dur_dt, th);
576 return dt_trans_start(env, lod->lod_child, th);
580 * Implementation of dt_device_operations::dt_trans_stop() for LOD
582 * Stops the set of local transactions using the targets involved
583 * in declare phase. Initial support for the distributed transactions.
585 * see include/dt_object.h for the details.
587 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
590 struct thandle_update *tu = th->th_update;
591 struct dt_update_request *update;
592 struct dt_update_request *tmp;
597 rc = dt_trans_stop(env, th->th_dev, th);
598 if (likely(tu == NULL))
601 list_for_each_entry_safe(update, tmp,
602 &tu->tu_remote_update_list,
604 /* update will be freed inside dt_trans_stop */
605 rc2 = dt_trans_stop(env, update->dur_dt, th);
606 if (unlikely(rc2 != 0 && rc == 0))
614 * Implementation of dt_device_operations::dt_conf_get() for LOD
616 * Currently returns the configuration provided by the local OSD.
618 * see include/dt_object.h for the details.
620 static void lod_conf_get(const struct lu_env *env,
621 const struct dt_device *dev,
622 struct dt_device_param *param)
624 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
628 * Implementation of dt_device_operations::dt_sync() for LOD
630 * Syncs all known OST targets. Very very expensive and used
631 * rarely by LFSCK now. Should not be used in general.
633 * see include/dt_object.h for the details.
635 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
637 struct lod_device *lod = dt2lod_dev(dev);
638 struct lod_ost_desc *ost;
643 lod_getref(&lod->lod_ost_descs);
644 lod_foreach_ost(lod, i) {
645 ost = OST_TGT(lod, i);
646 LASSERT(ost && ost->ltd_ost);
647 rc = dt_sync(env, ost->ltd_ost);
649 CERROR("%s: can't sync %u: %d\n",
650 lod2obd(lod)->obd_name, i, rc);
654 lod_putref(lod, &lod->lod_ost_descs);
656 rc = dt_sync(env, lod->lod_child);
662 * Implementation of dt_device_operations::dt_ro() for LOD
664 * Turns local OSD read-only, used for the testing only.
666 * see include/dt_object.h for the details.
668 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
670 return dt_ro(env, dt2lod_dev(dev)->lod_child);
674 * Implementation of dt_device_operations::dt_commit_async() for LOD
676 * Asks local OSD to commit sooner.
678 * see include/dt_object.h for the details.
680 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
682 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
688 static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
689 int mode, unsigned long timeout,
690 __u32 alg, struct lustre_capa_key *keys)
692 struct dt_device *next = dt2lod_dev(dev)->lod_child;
693 return dt_init_capa_ctxt(env, next, mode, timeout, alg, keys);
696 static const struct dt_device_operations lod_dt_ops = {
697 .dt_root_get = lod_root_get,
698 .dt_statfs = lod_statfs,
699 .dt_trans_create = lod_trans_create,
700 .dt_trans_start = lod_trans_start,
701 .dt_trans_stop = lod_trans_stop,
702 .dt_conf_get = lod_conf_get,
705 .dt_commit_async = lod_commit_async,
706 .dt_init_capa_ctxt = lod_init_capa_ctxt,
710 * Connect to a local OSD.
712 * Used to connect to the local OSD at mount. OSD name is taken from the
713 * configuration command passed. This connection is used to identify LU
714 * site and pin the OSD from early removal.
716 * \param[in] env LU environment provided by the caller
717 * \param[in] lod lod device
718 * \param[in] cfg configuration command to apply
720 * \retval 0 on success
721 * \retval negative negated errno on error
723 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
724 struct lustre_cfg *cfg)
726 struct obd_connect_data *data = NULL;
727 struct obd_device *obd;
728 char *nextdev = NULL, *p, *s;
733 LASSERT(lod->lod_child_exp == NULL);
735 /* compatibility hack: we still use old config logs
736 * which specify LOV, but we need to learn underlying
737 * OSD device, which is supposed to be:
738 * <fsname>-MDTxxxx-osd
740 * 2.x MGS generates lines like the following:
741 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
742 * 1.8 MGS generates lines like the following:
743 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
745 * we use "-MDT" to differentiate 2.x from 1.8 */
747 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
749 OBD_ALLOC(nextdev, len);
751 GOTO(out, rc = -ENOMEM);
754 s = strstr(nextdev, "-mdtlov");
755 if (unlikely(s == NULL)) {
756 CERROR("unable to parse device name %s\n",
757 lustre_cfg_string(cfg, 0));
758 GOTO(out, rc = -EINVAL);
761 if (strstr(nextdev, "-MDT")) {
766 strcpy(s, "-MDT0000-osd");
769 CERROR("unable to parse device name %s\n",
770 lustre_cfg_string(cfg, 0));
771 GOTO(out, rc = -EINVAL);
776 GOTO(out, rc = -ENOMEM);
778 obd = class_name2obd(nextdev);
780 CERROR("can not locate next device: %s\n", nextdev);
781 GOTO(out, rc = -ENOTCONN);
784 data->ocd_connect_flags = OBD_CONNECT_VERSION;
785 data->ocd_version = LUSTRE_VERSION_CODE;
787 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
790 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
794 lod->lod_dt_dev.dd_lu_dev.ld_site =
795 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
796 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
797 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
803 OBD_FREE(nextdev, len);
808 * Allocate and initialize target table.
810 * A helper function to initialize the target table and allocate
811 * a bitmap of the available targets.
813 * \param[in] ltd target's table to initialize
815 * \retval 0 on success
816 * \retval negative negated errno on error
818 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
820 mutex_init(<d->ltd_mutex);
821 init_rwsem(<d->ltd_rw_sem);
823 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
824 * added to the LOD, see lod_add_device() */
825 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
826 if (ltd->ltd_tgt_bitmap == NULL)
829 ltd->ltd_tgts_size = 32;
832 ltd->ltd_death_row = 0;
833 ltd->ltd_refcount = 0;
838 * Initialize LOD device at setup.
840 * Initializes the given LOD device using the original configuration command.
841 * The function initiates a connection to the local OSD and initializes few
842 * internal structures like pools, target tables, etc.
844 * \param[in] env LU environment provided by the caller
845 * \param[in] lod lod device
846 * \param[in] ldt not used
847 * \param[in] cfg configuration command
849 * \retval 0 on success
850 * \retval negative negated errno on error
852 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
853 struct lu_device_type *ldt, struct lustre_cfg *cfg)
855 struct dt_device_param ddp;
856 struct obd_device *obd;
860 obd = class_name2obd(lustre_cfg_string(cfg, 0));
862 CERROR("Cannot find obd with name %s\n",
863 lustre_cfg_string(cfg, 0));
867 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
868 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
869 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
870 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
872 rc = lod_connect_to_osd(env, lod, cfg);
876 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
877 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
879 /* setup obd to be used with old lov code */
880 rc = lod_pools_init(lod, cfg);
882 GOTO(out_disconnect, rc);
884 rc = lod_procfs_init(lod);
888 spin_lock_init(&lod->lod_desc_lock);
889 spin_lock_init(&lod->lod_connects_lock);
890 lod_tgt_desc_init(&lod->lod_mdt_descs);
891 lod_tgt_desc_init(&lod->lod_ost_descs);
898 obd_disconnect(lod->lod_child_exp);
903 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
905 * Releases the memory allocated for LOD device.
907 * see include/lu_object.h for the details.
909 static struct lu_device *lod_device_free(const struct lu_env *env,
910 struct lu_device *lu)
912 struct lod_device *lod = lu2lod_dev(lu);
913 struct lu_device *next = &lod->lod_child->dd_lu_dev;
916 LASSERT(atomic_read(&lu->ld_ref) == 0);
917 dt_device_fini(&lod->lod_dt_dev);
923 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
925 * Allocates LOD device and calls the helpers to initialize it.
927 * see include/lu_object.h for the details.
929 static struct lu_device *lod_device_alloc(const struct lu_env *env,
930 struct lu_device_type *type,
931 struct lustre_cfg *lcfg)
933 struct lod_device *lod;
934 struct lu_device *lu_dev;
938 lu_dev = ERR_PTR(-ENOMEM);
942 lu_dev = lod2lu_dev(lod);
943 dt_device_init(&lod->lod_dt_dev, type);
944 rc = lod_init0(env, lod, type, lcfg);
946 lod_device_free(env, lu_dev);
947 lu_dev = ERR_PTR(rc);
955 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
957 * Releases the internal resources used by LOD device.
959 * see include/lu_object.h for the details.
961 static struct lu_device *lod_device_fini(const struct lu_env *env,
964 struct lod_device *lod = lu2lod_dev(d);
970 lod_procfs_fini(lod);
972 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
974 CERROR("%s:can not fini ost descs %d\n",
975 lod2obd(lod)->obd_name, rc);
977 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
979 CERROR("%s:can not fini mdt descs %d\n",
980 lod2obd(lod)->obd_name, rc);
986 * Implementation of obd_ops::o_connect() for LOD
988 * Used to track all the users of this specific LOD device,
989 * so the device stays up until the last user disconnected.
991 * \param[in] env LU environment provided by the caller
992 * \param[out] exp export the caller will be using to access LOD
993 * \param[in] obd OBD device representing LOD device
994 * \param[in] cluuid unique identifier of the caller
995 * \param[in] data not used
996 * \param[in] localdata not used
998 * \retval 0 on success
999 * \retval negative negated errno on error
1001 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1002 struct obd_device *obd, struct obd_uuid *cluuid,
1003 struct obd_connect_data *data, void *localdata)
1005 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1006 struct lustre_handle conn;
1010 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1012 rc = class_connect(&conn, obd, cluuid);
1016 *exp = class_conn2export(&conn);
1018 spin_lock(&lod->lod_connects_lock);
1019 lod->lod_connects++;
1020 /* at the moment we expect the only user */
1021 LASSERT(lod->lod_connects == 1);
1022 spin_unlock(&lod->lod_connects_lock);
1029 * Implementation of obd_ops::o_disconnect() for LOD
1031 * When the caller doesn't need to use this LOD instance, it calls
1032 * obd_disconnect() and LOD releases corresponding export/reference count.
1033 * Once all the users gone, LOD device is released.
1035 * \param[in] exp export provided to the caller in obd_connect()
1037 * \retval 0 on success
1038 * \retval negative negated errno on error
1040 static int lod_obd_disconnect(struct obd_export *exp)
1042 struct obd_device *obd = exp->exp_obd;
1043 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1044 int rc, release = 0;
1047 /* Only disconnect the underlying layers on the final disconnect. */
1048 spin_lock(&lod->lod_connects_lock);
1049 lod->lod_connects--;
1050 if (lod->lod_connects != 0) {
1051 /* why should there be more than 1 connect? */
1052 spin_unlock(&lod->lod_connects_lock);
1053 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1057 spin_unlock(&lod->lod_connects_lock);
1059 /* the last user of lod has gone, let's release the device */
1063 rc = class_disconnect(exp); /* bz 9811 */
1065 if (rc == 0 && release)
1066 class_manual_cleanup(obd);
1070 LU_KEY_INIT(lod, struct lod_thread_info);
1072 static void lod_key_fini(const struct lu_context *ctx,
1073 struct lu_context_key *key, void *data)
1075 struct lod_thread_info *info = data;
1076 /* allocated in lod_get_lov_ea
1077 * XXX: this is overload, a tread may have such store but used only
1078 * once. Probably better would be pool of such stores per LOD.
1080 if (info->lti_ea_store) {
1081 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1082 info->lti_ea_store = NULL;
1083 info->lti_ea_store_size = 0;
1085 lu_buf_free(&info->lti_linkea_buf);
1089 /* context key: lod_thread_key */
1090 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1092 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1094 static struct lu_device_type_operations lod_device_type_ops = {
1095 .ldto_init = lod_type_init,
1096 .ldto_fini = lod_type_fini,
1098 .ldto_start = lod_type_start,
1099 .ldto_stop = lod_type_stop,
1101 .ldto_device_alloc = lod_device_alloc,
1102 .ldto_device_free = lod_device_free,
1104 .ldto_device_fini = lod_device_fini
1107 static struct lu_device_type lod_device_type = {
1108 .ldt_tags = LU_DEVICE_DT,
1109 .ldt_name = LUSTRE_LOD_NAME,
1110 .ldt_ops = &lod_device_type_ops,
1111 .ldt_ctx_tags = LCT_MD_THREAD,
1115 * Implementation of obd_ops::o_get_info() for LOD
1117 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1118 * the caller binary status whether LOD has seen connection to any OST target.
1120 * \param[in] env LU environment provided by the caller
1121 * \param[in] exp export of the caller
1122 * \param[in] keylen len of the key
1123 * \param[in] key the key
1124 * \param[in] vallen not used
1125 * \param[in] val not used
1126 * \param[in] lsm not used
1128 * \retval 0 if a connection was seen
1129 * \retval -EAGAIN if LOD isn't running yet or no
1130 * connection has been seen yet
1131 * \retval -EINVAL if not supported key is requested
1133 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1134 __u32 keylen, void *key, __u32 *vallen, void *val,
1135 struct lov_stripe_md *lsm)
1139 if (KEY_IS(KEY_OSP_CONNECTED)) {
1140 struct obd_device *obd = exp->exp_obd;
1141 struct lod_device *d;
1142 struct lod_ost_desc *ost;
1146 if (!obd->obd_set_up || obd->obd_stopping)
1149 d = lu2lod_dev(obd->obd_lu_dev);
1150 lod_getref(&d->lod_ost_descs);
1151 lod_foreach_ost(d, i) {
1152 ost = OST_TGT(d, i);
1153 LASSERT(ost && ost->ltd_ost);
1155 rc = obd_get_info(env, ost->ltd_exp, keylen, key,
1157 /* one healthy device is enough */
1161 lod_putref(d, &d->lod_ost_descs);
1168 static struct obd_ops lod_obd_device_ops = {
1169 .o_owner = THIS_MODULE,
1170 .o_connect = lod_obd_connect,
1171 .o_disconnect = lod_obd_disconnect,
1172 .o_get_info = lod_obd_get_info,
1173 .o_pool_new = lod_pool_new,
1174 .o_pool_rem = lod_pool_remove,
1175 .o_pool_add = lod_pool_add,
1176 .o_pool_del = lod_pool_del,
1179 static int __init lod_mod_init(void)
1181 struct obd_type *type;
1184 rc = lu_kmem_init(lod_caches);
1188 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
1189 LUSTRE_LOD_NAME, &lod_device_type);
1191 lu_kmem_fini(lod_caches);
1195 /* create "lov" entry in procfs for compatibility purposes */
1196 type = class_search_type(LUSTRE_LOV_NAME);
1197 if (type != NULL && type->typ_procroot != NULL)
1200 type = class_search_type(LUSTRE_LOD_NAME);
1201 type->typ_procsym = lprocfs_seq_register("lov", proc_lustre_root,
1203 if (IS_ERR(type->typ_procsym)) {
1204 CERROR("lod: can't create compat entry \"lov\": %d\n",
1205 (int)PTR_ERR(type->typ_procsym));
1206 type->typ_procsym = NULL;
1211 static void __exit lod_mod_exit(void)
1213 class_unregister_type(LUSTRE_LOD_NAME);
1214 lu_kmem_fini(lod_caches);
1217 MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
1218 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1219 MODULE_LICENSE("GPL");
1221 module_init(lod_mod_init);
1222 module_exit(lod_mod_exit);