1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_device.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #include <linux/jbd.h>
51 #include <obd_class.h>
52 #include <lustre_ver.h>
53 #include <obd_support.h>
54 #include <lprocfs_status.h>
56 #include <lustre_disk.h>
57 #include <lustre_fid.h>
58 #include <linux/ldiskfs_fs.h>
59 #include <lustre_mds.h>
60 #include <lustre/lustre_idl.h>
61 #include <lustre_disk.h> /* for changelogs */
62 #include <lustre_param.h>
63 #include <lustre_fid.h>
65 #include "mdd_internal.h"
67 const struct md_device_operations mdd_ops;
68 static struct lu_device_type mdd_device_type;
70 static const char mdd_root_dir_name[] = "ROOT";
71 static const char mdd_obf_dir_name[] = "fid";
73 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
74 const char *name, struct lu_device *next)
76 struct mdd_device *mdd = lu2mdd_dev(d);
80 mdd->mdd_child = lu2dt_dev(next);
82 /* Prepare transactions callbacks. */
83 mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
84 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
85 mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb;
86 mdd->mdd_txn_cb.dtc_cookie = mdd;
87 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
88 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
90 rc = mdd_procfs_init(mdd, name);
94 static struct lu_device *mdd_device_fini(const struct lu_env *env,
97 struct mdd_device *mdd = lu2mdd_dev(d);
98 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
101 rc = mdd_procfs_fini(mdd);
103 CERROR("proc fini error %d \n", rc);
109 static void mdd_changelog_fini(const struct lu_env *env,
110 struct mdd_device *mdd);
112 static void mdd_device_shutdown(const struct lu_env *env,
113 struct mdd_device *m, struct lustre_cfg *cfg)
116 mdd_changelog_fini(env, m);
117 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
118 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
119 mdd_object_put(env, m->mdd_dot_lustre);
121 mdd_fini_obd(env, m, cfg);
122 orph_index_fini(env, m);
123 /* remove upcall device*/
124 md_upcall_fini(&m->mdd_md_dev);
128 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
131 struct mdd_device *mdd = (struct mdd_device *)data;
132 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
135 if (!(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)) {
136 CERROR("log is not plain\n");
139 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
140 CERROR("Not a changelog rec? %d\n", rec->cr_hdr.lrh_type);
145 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
146 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
147 rec->cr_type, rec->cr_namelen, rec->cr_name,
148 llh->lgh_id.lgl_oid);
150 mdd->mdd_cl.mc_index = rec->cr_index;
151 RETURN(LLOG_PROC_BREAK);
154 static int mdd_changelog_llog_init(struct mdd_device *mdd)
156 struct obd_device *obd = mdd2obd_dev(mdd);
157 struct llog_ctxt *ctxt;
160 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
162 CERROR("no context\n");
165 if (!ctxt->loc_handle) {
166 CERROR("no handle\n");
169 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
173 CERROR("changelog init failed: %d\n", rc);
175 rc = 0; /* llog_proc_break is ok */
177 CDEBUG(D_INODE, "changelog_init index="LPU64"\n", mdd->mdd_cl.mc_index);
182 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
186 mdd->mdd_cl.mc_index = 0;
187 spin_lock_init(&mdd->mdd_cl.mc_lock);
188 cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
190 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
191 mdd->mdd_cl.mc_flags = 0; /* off by default */
192 mdd->mdd_cl.mc_mask = CL_DEFMASK;
193 rc = mdd_changelog_llog_init(mdd);
195 CERROR("Changelog setup during init failed %d\n", rc);
196 mdd->mdd_cl.mc_flags |= CLM_ERR;
201 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
203 mdd->mdd_cl.mc_flags = 0;
206 /** Add a changelog entry \a rec to the changelog llog
209 * \param handle - currently ignored since llogs start their own transaction;
210 * this will hopefully be fixed in llog rewrite
213 int mdd_changelog_llog_write(struct mdd_device *mdd,
214 struct llog_changelog_rec *rec,
215 struct thandle *handle)
217 struct obd_device *obd = mdd2obd_dev(mdd);
218 struct llog_ctxt *ctxt;
221 if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
224 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
225 /* llog_lvfs_write_rec sets the llog tail len */
226 rec->cr_hdr.lrh_type = CHANGELOG_REC;
227 rec->cr_time = cfs_time_current_64();
228 spin_lock(&mdd->mdd_cl.mc_lock);
229 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
230 but as long as the MDD transactions are ordered correctly for e.g.
231 rename conflicts, I don't think this should matter. */
232 rec->cr_index = ++mdd->mdd_cl.mc_index;
233 spin_unlock(&mdd->mdd_cl.mc_lock);
234 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
238 /* nested journal transaction */
239 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
242 cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
247 /** Remove entries with indicies up to and including \a endrec from the
253 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
255 struct obd_device *obd = mdd2obd_dev(mdd);
256 struct llog_ctxt *ctxt;
259 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
263 /* Some records purged; reset repeat-access time */
264 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
266 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
273 /** Add a CL_MARK record to the changelog
275 * \param markerflags - CLM_*
278 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
280 struct obd_device *obd = mdd2obd_dev(mdd);
281 struct llog_changelog_rec *rec;
283 int len = strlen(obd->obd_name);
287 reclen = llog_data_len(sizeof(*rec) + len);
288 OBD_ALLOC(rec, reclen);
292 rec->cr_flags = CLF_VERSION;
293 rec->cr_type = CL_MARK;
294 rec->cr_namelen = len;
295 memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
296 /* Status and action flags */
297 rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
299 rc = mdd_changelog_llog_write(mdd, rec, NULL);
301 /* assume on or off event; reset repeat-access time */
302 mdd->mdd_cl.mc_starttime = rec->cr_time;
304 OBD_FREE(rec, reclen);
309 * Create ".lustre" directory.
311 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
313 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
314 struct md_object *mdo;
317 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
318 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
319 mdd_root_dir_name, mdd_dot_lustre_name,
320 fid, &dt_directory_features);
321 /* .lustre dir may be already present */
322 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
324 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
325 mdd_dot_lustre_name, PFID(fid), rc);
330 lu_object_put(env, &mdo->mo_lu);
335 static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
338 struct mdd_object *mdd_obj = md2mdd_obj(obj);
340 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
343 static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
344 const struct md_attr *ma)
349 static int dot_lustre_xattr_get(const struct lu_env *env,
350 struct md_object *obj, struct lu_buf *buf,
356 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
359 struct mdd_object *mdd_obj = md2mdd_obj(obj);
361 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
362 mdd_obj->mod_count++;
363 mdd_write_unlock(env, mdd_obj);
368 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
369 char *path, int pathlen, __u64 recno, int *linkno)
374 static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
377 struct mdd_object *mdd_obj = md2mdd_obj(obj);
379 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
380 mdd_obj->mod_count--;
381 mdd_write_unlock(env, mdd_obj);
386 static struct md_object_operations mdd_dot_lustre_obj_ops = {
387 .moo_attr_get = dot_lustre_attr_get,
388 .moo_attr_set = dot_lustre_attr_set,
389 .moo_xattr_get = dot_lustre_xattr_get,
390 .moo_open = dot_lustre_mdd_open,
391 .moo_close = dot_lustre_close,
392 .moo_readpage = mdd_readpage,
393 .moo_path = dot_lustre_path
396 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
397 const struct lu_name *lname, struct lu_fid *f,
398 struct md_op_spec *spec)
400 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
408 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
409 const struct lu_name *lname,
410 struct md_object *child, struct md_op_spec *spec,
416 static int dot_lustre_rename(const struct lu_env *env,
417 struct md_object *src_pobj,
418 struct md_object *tgt_pobj,
419 const struct lu_fid *lf,
420 const struct lu_name *lsname,
421 struct md_object *tobj,
422 const struct lu_name *ltname, struct md_attr *ma)
427 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
428 struct md_object *src_obj,
429 const struct lu_name *lname, struct md_attr *ma)
434 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
435 struct md_object *cobj, const struct lu_name *lname,
441 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
442 .mdo_lookup = dot_lustre_lookup,
443 .mdo_create = dot_lustre_create,
444 .mdo_rename = dot_lustre_rename,
445 .mdo_link = dot_lustre_link,
446 .mdo_unlink = dot_lustre_unlink,
449 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
454 if (ma->ma_need & MA_INODE) {
455 struct mdd_device *mdd = mdo2mdd(obj);
457 /* "fid" is a virtual object and hence does not have any "real"
458 * attributes. So we reuse attributes of .lustre for "fid" dir */
459 ma->ma_need |= MA_INODE;
460 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
463 ma->ma_valid |= MA_INODE;
466 /* "fid" directory does not have any striping information. */
467 if (ma->ma_need & MA_LOV) {
468 struct mdd_object *mdd_obj = md2mdd_obj(obj);
470 if (ma->ma_valid & MA_LOV)
473 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
474 S_ISDIR(mdd_object_type(mdd_obj))))
477 if (ma->ma_need & MA_LOV_DEF) {
478 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
481 ma->ma_valid |= MA_LOV;
490 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
491 const struct md_attr *ma)
496 static int obf_xattr_get(const struct lu_env *env,
497 struct md_object *obj, struct lu_buf *buf,
503 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
506 struct mdd_object *mdd_obj = md2mdd_obj(obj);
508 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
509 mdd_obj->mod_count++;
510 mdd_write_unlock(env, mdd_obj);
515 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
518 struct mdd_object *mdd_obj = md2mdd_obj(obj);
520 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
521 mdd_obj->mod_count--;
522 mdd_write_unlock(env, mdd_obj);
527 /** Nothing to list in "fid" directory */
528 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
529 const struct lu_rdpg *rdpg)
534 static int obf_path(const struct lu_env *env, struct md_object *obj,
535 char *path, int pathlen, __u64 recno, int *linkno)
540 static struct md_object_operations mdd_obf_obj_ops = {
541 .moo_attr_get = obf_attr_get,
542 .moo_attr_set = obf_attr_set,
543 .moo_xattr_get = obf_xattr_get,
544 .moo_open = obf_mdd_open,
545 .moo_close = obf_mdd_close,
546 .moo_readpage = obf_mdd_readpage,
551 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
552 * are valid. We also check if object with passed fid exists or not.
554 static int obf_lookup(const struct lu_env *env, struct md_object *p,
555 const struct lu_name *lname, struct lu_fid *f,
556 struct md_op_spec *spec)
558 char *name = (char *)lname->ln_name;
559 struct mdd_device *mdd = mdo2mdd(p);
560 struct mdd_object *child;
566 sscanf(name, SFID, &(f->f_seq), &(f->f_oid),
568 if (!fid_is_sane(f)) {
569 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
571 GOTO(out, rc = -EINVAL);
574 /* Check if object with this fid exists */
575 child = mdd_object_find(env, mdd, f);
579 GOTO(out, rc = PTR_ERR(child));
581 if (mdd_object_exists(child) == 0)
584 mdd_object_put(env, child);
590 static int obf_create(const struct lu_env *env, struct md_object *pobj,
591 const struct lu_name *lname, struct md_object *child,
592 struct md_op_spec *spec, struct md_attr* ma)
597 static int obf_rename(const struct lu_env *env,
598 struct md_object *src_pobj, struct md_object *tgt_pobj,
599 const struct lu_fid *lf, const struct lu_name *lsname,
600 struct md_object *tobj, const struct lu_name *ltname,
606 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
607 struct md_object *src_obj, const struct lu_name *lname,
613 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
614 struct md_object *cobj, const struct lu_name *lname,
620 static struct md_dir_operations mdd_obf_dir_ops = {
621 .mdo_lookup = obf_lookup,
622 .mdo_create = obf_create,
623 .mdo_rename = obf_rename,
624 .mdo_link = obf_link,
625 .mdo_unlink = obf_unlink
629 * Create special in-memory "fid" object for open-by-fid.
631 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
633 struct mdd_object *mdd_obf;
634 struct lu_object *obf_lu_obj;
637 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
639 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
640 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
641 GOTO(out, rc = -ENOENT);
643 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
644 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
645 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
646 /* Don't allow objects to be created in "fid" dir */
647 mdd_obf->mod_flags |= IMMUTE_OBJ;
649 obf_lu_obj = mdd2lu_obj(mdd_obf);
650 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
656 /** Setup ".lustre" directory object */
657 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
659 struct dt_object *dt_dot_lustre;
660 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
663 rc = create_dot_lustre_dir(env, m);
667 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
668 mdd_dot_lustre_name, fid);
669 if (IS_ERR(dt_dot_lustre)) {
670 rc = PTR_ERR(dt_dot_lustre);
674 /* references are released in mdd_device_shutdown() */
675 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
678 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
679 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
681 rc = mdd_obf_setup(env, m);
683 CERROR("Error initializing \"fid\" object - %d.\n", rc);
689 static int mdd_process_config(const struct lu_env *env,
690 struct lu_device *d, struct lustre_cfg *cfg)
692 struct mdd_device *m = lu2mdd_dev(d);
693 struct dt_device *dt = m->mdd_child;
694 struct lu_device *next = &dt->dd_lu_dev;
698 switch (cfg->lcfg_command) {
700 struct lprocfs_static_vars lvars;
702 lprocfs_mdd_init_vars(&lvars);
703 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
704 if (rc > 0 || rc == -ENOSYS)
705 /* we don't understand; pass it on */
706 rc = next->ld_ops->ldo_process_config(env, next, cfg);
710 rc = next->ld_ops->ldo_process_config(env, next, cfg);
713 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
715 rc = mdd_init_obd(env, m, cfg);
717 CERROR("lov init error %d \n", rc);
720 rc = mdd_txn_init_credits(env, m);
724 mdd_changelog_init(env, m);
727 mdd_device_shutdown(env, m, cfg);
729 rc = next->ld_ops->ldo_process_config(env, next, cfg);
737 static int mdd_lov_set_nextid(const struct lu_env *env,
738 struct mdd_device *mdd)
740 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
744 LASSERT(mds->mds_lov_objids != NULL);
745 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
746 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
747 mds->mds_lov_objids, NULL);
752 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
753 struct mdd_device *mdd)
755 /* XXX: to be implemented! */
760 static int mdd_recovery_complete(const struct lu_env *env,
763 struct mdd_device *mdd = lu2mdd_dev(d);
764 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
765 struct obd_device *obd = mdd2obd_dev(mdd);
769 LASSERT(mdd != NULL);
770 LASSERT(obd != NULL);
772 /* XXX: Do we need this in new stack? */
773 rc = mdd_lov_set_nextid(env, mdd);
775 CERROR("mdd_lov_set_nextid() failed %d\n",
780 /* XXX: cleanup unlink. */
781 rc = mdd_cleanup_unlink_llog(env, mdd);
783 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
788 /* Call that with obd_recovering = 1 just to update objids */
789 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
790 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
792 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
793 obd->obd_recovering = 0;
794 obd->obd_type->typ_dt_ops->o_postrecov(obd);
796 /* XXX: orphans handling. */
797 __mdd_orphan_cleanup(env, mdd);
798 rc = next->ld_ops->ldo_recovery_complete(env, next);
803 static int mdd_prepare(const struct lu_env *env,
804 struct lu_device *pdev,
805 struct lu_device *cdev)
807 struct mdd_device *mdd = lu2mdd_dev(cdev);
808 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
809 struct dt_object *root;
813 rc = next->ld_ops->ldo_prepare(env, cdev, next);
817 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
818 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
821 LASSERT(root != NULL);
822 lu_object_put(env, &root->do_lu);
823 rc = orph_index_init(env, mdd);
830 rc = mdd_dot_lustre_setup(env, mdd);
832 CERROR("Error(%d) initializing .lustre objects\n", rc);
840 const struct lu_device_operations mdd_lu_ops = {
841 .ldo_object_alloc = mdd_object_alloc,
842 .ldo_process_config = mdd_process_config,
843 .ldo_recovery_complete = mdd_recovery_complete,
844 .ldo_prepare = mdd_prepare,
848 * No permission check is needed.
850 static int mdd_root_get(const struct lu_env *env,
851 struct md_device *m, struct lu_fid *f)
853 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
856 *f = mdd->mdd_root_fid;
861 * No permission check is needed.
863 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
866 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
871 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
877 * No permission check is needed.
879 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
880 int *md_size, int *cookie_size)
882 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
885 *md_size = mdd_lov_mdsize(env, mdd);
886 *cookie_size = mdd_lov_cookiesize(env, mdd);
891 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
892 int mode, unsigned long timeout, __u32 alg,
893 struct lustre_capa_key *keys)
895 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
896 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
900 mds->mds_capa_keys = keys;
901 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
906 static int mdd_update_capa_key(const struct lu_env *env,
908 struct lustre_capa_key *key)
910 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
911 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
915 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
916 sizeof(*key), key, NULL);
920 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
921 struct lu_device_type *t,
922 struct lustre_cfg *lcfg)
925 struct mdd_device *m;
929 l = ERR_PTR(-ENOMEM);
931 md_device_init(&m->mdd_md_dev, t);
933 l->ld_ops = &mdd_lu_ops;
934 m->mdd_md_dev.md_ops = &mdd_ops;
935 md_upcall_init(&m->mdd_md_dev, NULL);
941 static struct lu_device *mdd_device_free(const struct lu_env *env,
942 struct lu_device *lu)
944 struct mdd_device *m = lu2mdd_dev(lu);
945 struct lu_device *next = &m->mdd_child->dd_lu_dev;
948 LASSERT(atomic_read(&lu->ld_ref) == 0);
949 md_device_fini(&m->mdd_md_dev);
954 static struct obd_ops mdd_obd_device_ops = {
955 .o_owner = THIS_MODULE
958 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
959 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
961 static struct lu_context_key mdd_ucred_key = {
962 .lct_tags = LCT_SESSION,
963 .lct_init = mdd_ucred_key_init,
964 .lct_fini = mdd_ucred_key_fini
967 struct md_ucred *md_ucred(const struct lu_env *env)
969 LASSERT(env->le_ses != NULL);
970 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
972 EXPORT_SYMBOL(md_ucred);
975 * context key constructor/destructor:
976 * mdd_capainfo_key_init, mdd_capainfo_key_fini
978 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
980 struct lu_context_key mdd_capainfo_key = {
981 .lct_tags = LCT_SESSION,
982 .lct_init = mdd_capainfo_key_init,
983 .lct_fini = mdd_capainfo_key_fini
986 struct md_capainfo *md_capainfo(const struct lu_env *env)
988 /* NB, in mdt_init0 */
989 if (env->le_ses == NULL)
991 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
993 EXPORT_SYMBOL(md_capainfo);
995 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
996 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
998 const struct md_device_operations mdd_ops = {
999 .mdo_statfs = mdd_statfs,
1000 .mdo_root_get = mdd_root_get,
1001 .mdo_maxsize_get = mdd_maxsize_get,
1002 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1003 .mdo_update_capa_key= mdd_update_capa_key,
1004 #ifdef HAVE_QUOTA_SUPPORT
1006 .mqo_notify = mdd_quota_notify,
1007 .mqo_setup = mdd_quota_setup,
1008 .mqo_cleanup = mdd_quota_cleanup,
1009 .mqo_recovery = mdd_quota_recovery,
1010 .mqo_check = mdd_quota_check,
1011 .mqo_on = mdd_quota_on,
1012 .mqo_off = mdd_quota_off,
1013 .mqo_setinfo = mdd_quota_setinfo,
1014 .mqo_getinfo = mdd_quota_getinfo,
1015 .mqo_setquota = mdd_quota_setquota,
1016 .mqo_getquota = mdd_quota_getquota,
1017 .mqo_getoinfo = mdd_quota_getoinfo,
1018 .mqo_getoquota = mdd_quota_getoquota,
1019 .mqo_invalidate = mdd_quota_invalidate,
1020 .mqo_finvalidate = mdd_quota_finvalidate
1025 static struct lu_device_type_operations mdd_device_type_ops = {
1026 .ldto_init = mdd_type_init,
1027 .ldto_fini = mdd_type_fini,
1029 .ldto_start = mdd_type_start,
1030 .ldto_stop = mdd_type_stop,
1032 .ldto_device_alloc = mdd_device_alloc,
1033 .ldto_device_free = mdd_device_free,
1035 .ldto_device_init = mdd_device_init,
1036 .ldto_device_fini = mdd_device_fini
1039 static struct lu_device_type mdd_device_type = {
1040 .ldt_tags = LU_DEVICE_MD,
1041 .ldt_name = LUSTRE_MDD_NAME,
1042 .ldt_ops = &mdd_device_type_ops,
1043 .ldt_ctx_tags = LCT_MD_THREAD
1046 /* context key constructor: mdd_key_init */
1047 LU_KEY_INIT(mdd, struct mdd_thread_info);
1049 static void mdd_key_fini(const struct lu_context *ctx,
1050 struct lu_context_key *key, void *data)
1052 struct mdd_thread_info *info = data;
1053 if (info->mti_max_lmm != NULL)
1054 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1055 if (info->mti_max_cookie != NULL)
1056 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1057 mdd_buf_put(&info->mti_big_buf);
1062 /* context key: mdd_thread_key */
1063 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1065 static struct lu_local_obj_desc llod_capa_key = {
1066 .llod_name = CAPA_KEYS,
1067 .llod_oid = MDD_CAPA_KEYS_OID,
1071 static struct lu_local_obj_desc llod_mdd_orphan = {
1072 .llod_name = orph_index_name,
1073 .llod_oid = MDD_ORPHAN_OID,
1075 .llod_feat = &dt_directory_features,
1078 static struct lu_local_obj_desc llod_mdd_root = {
1079 .llod_name = mdd_root_dir_name,
1080 .llod_oid = MDD_ROOT_INDEX_OID,
1082 .llod_feat = &dt_directory_features,
1085 static int __init mdd_mod_init(void)
1087 struct lprocfs_static_vars lvars;
1088 lprocfs_mdd_init_vars(&lvars);
1090 llo_local_obj_register(&llod_capa_key);
1091 llo_local_obj_register(&llod_mdd_orphan);
1092 llo_local_obj_register(&llod_mdd_root);
1094 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1095 LUSTRE_MDD_NAME, &mdd_device_type);
1098 static void __exit mdd_mod_exit(void)
1100 llo_local_obj_unregister(&llod_capa_key);
1101 llo_local_obj_unregister(&llod_mdd_orphan);
1102 llo_local_obj_unregister(&llod_mdd_root);
1104 class_unregister_type(LUSTRE_MDD_NAME);
1107 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1108 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1109 MODULE_LICENSE("GPL");
1111 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);