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";
72 static const char mdd_dot_lustre_name[] = ".lustre";
74 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
75 const char *name, struct lu_device *next)
77 struct mdd_device *mdd = lu2mdd_dev(d);
81 mdd->mdd_child = lu2dt_dev(next);
83 /* Prepare transactions callbacks. */
84 mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
85 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
86 mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb;
87 mdd->mdd_txn_cb.dtc_cookie = mdd;
88 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
89 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
91 rc = mdd_procfs_init(mdd, name);
95 static struct lu_device *mdd_device_fini(const struct lu_env *env,
98 struct mdd_device *mdd = lu2mdd_dev(d);
99 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
102 rc = mdd_procfs_fini(mdd);
104 CERROR("proc fini error %d \n", rc);
110 static void mdd_changelog_fini(const struct lu_env *env,
111 struct mdd_device *mdd);
113 static void mdd_device_shutdown(const struct lu_env *env,
114 struct mdd_device *m, struct lustre_cfg *cfg)
117 mdd_changelog_fini(env, m);
118 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
119 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
120 mdd_object_put(env, m->mdd_dot_lustre);
122 mdd_fini_obd(env, m, cfg);
123 orph_index_fini(env, m);
124 /* remove upcall device*/
125 md_upcall_fini(&m->mdd_md_dev);
129 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
132 struct mdd_device *mdd = (struct mdd_device *)data;
133 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
136 if (!(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)) {
137 CERROR("log is not plain\n");
140 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
141 CERROR("Not a changelog rec? %d\n", rec->cr_hdr.lrh_type);
146 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
147 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
148 rec->cr_type, rec->cr_namelen, rec->cr_name,
149 llh->lgh_id.lgl_oid);
151 mdd->mdd_cl.mc_index = rec->cr_index;
152 RETURN(LLOG_PROC_BREAK);
155 static int mdd_changelog_llog_init(struct mdd_device *mdd)
157 struct obd_device *obd = mdd2obd_dev(mdd);
158 struct llog_ctxt *ctxt;
161 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
163 CERROR("no context\n");
166 if (!ctxt->loc_handle) {
167 CERROR("no handle\n");
170 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
174 CERROR("changelog init failed: %d\n", rc);
176 rc = 0; /* llog_proc_break is ok */
178 CDEBUG(D_INODE, "changelog_init index="LPU64"\n", mdd->mdd_cl.mc_index);
183 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
187 mdd->mdd_cl.mc_index = 0;
188 spin_lock_init(&mdd->mdd_cl.mc_lock);
189 cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
191 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
192 mdd->mdd_cl.mc_flags = 0; /* off by default */
193 mdd->mdd_cl.mc_mask = CL_DEFMASK;
194 rc = mdd_changelog_llog_init(mdd);
196 CERROR("Changelog setup during init failed %d\n", rc);
197 mdd->mdd_cl.mc_flags |= CLM_ERR;
202 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
204 mdd->mdd_cl.mc_flags = 0;
207 /** Add a changelog entry \a rec to the changelog llog
210 * \param handle - currently ignored since llogs start their own transaction;
211 * this will hopefully be fixed in llog rewrite
214 int mdd_changelog_llog_write(struct mdd_device *mdd,
215 struct llog_changelog_rec *rec,
216 struct thandle *handle)
218 struct obd_device *obd = mdd2obd_dev(mdd);
219 struct llog_ctxt *ctxt;
222 if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
225 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
226 /* llog_lvfs_write_rec sets the llog tail len */
227 rec->cr_hdr.lrh_type = CHANGELOG_REC;
228 rec->cr_time = cfs_time_current_64();
229 spin_lock(&mdd->mdd_cl.mc_lock);
230 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
231 but as long as the MDD transactions are ordered correctly for e.g.
232 rename conflicts, I don't think this should matter. */
233 rec->cr_index = ++mdd->mdd_cl.mc_index;
234 spin_unlock(&mdd->mdd_cl.mc_lock);
235 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
239 /* nested journal transaction */
240 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
243 cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
248 /** Remove entries with indicies up to and including \a endrec from the
254 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
256 struct obd_device *obd = mdd2obd_dev(mdd);
257 struct llog_ctxt *ctxt;
260 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
264 /* Some records purged; reset repeat-access time */
265 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
267 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
274 /** Add a CL_MARK record to the changelog
276 * \param markerflags - CLM_*
279 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
281 struct obd_device *obd = mdd2obd_dev(mdd);
282 struct llog_changelog_rec *rec;
284 int len = strlen(obd->obd_name);
288 reclen = llog_data_len(sizeof(*rec) + len);
289 OBD_ALLOC(rec, reclen);
293 rec->cr_flags = CLF_VERSION;
294 rec->cr_type = CL_MARK;
295 rec->cr_namelen = len;
296 memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
297 /* Status and action flags */
298 rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
300 rc = mdd_changelog_llog_write(mdd, rec, NULL);
302 /* assume on or off event; reset repeat-access time */
303 mdd->mdd_cl.mc_starttime = rec->cr_time;
305 OBD_FREE(rec, reclen);
310 * Create ".lustre" directory.
312 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
314 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
315 struct md_object *mdo;
318 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
319 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
320 mdd_root_dir_name, mdd_dot_lustre_name,
321 fid, &dt_directory_features);
322 /* .lustre dir may be already present */
323 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
325 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
326 mdd_dot_lustre_name, PFID(fid), rc);
333 static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
336 struct mdd_object *mdd_obj = md2mdd_obj(obj);
338 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
341 static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
342 const struct md_attr *ma)
347 static int dot_lustre_xattr_get(const struct lu_env *env,
348 struct md_object *obj, struct lu_buf *buf,
355 * Direct access to the ".lustre" directory is not allowed.
357 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
363 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
364 char *path, int pathlen, __u64 recno, int *linkno)
369 static struct md_object_operations mdd_dot_lustre_obj_ops = {
370 .moo_attr_get = dot_lustre_attr_get,
371 .moo_attr_set = dot_lustre_attr_set,
372 .moo_xattr_get = dot_lustre_xattr_get,
373 .moo_open = dot_lustre_mdd_open,
374 .moo_path = dot_lustre_path
377 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
378 const struct lu_name *lname, struct lu_fid *f,
379 struct md_op_spec *spec)
381 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
389 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
390 const struct lu_name *lname,
391 struct md_object *child, struct md_op_spec *spec,
397 static int dot_lustre_rename(const struct lu_env *env,
398 struct md_object *src_pobj,
399 struct md_object *tgt_pobj,
400 const struct lu_fid *lf,
401 const struct lu_name *lsname,
402 struct md_object *tobj,
403 const struct lu_name *ltname, struct md_attr *ma)
408 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
409 struct md_object *src_obj,
410 const struct lu_name *lname, struct md_attr *ma)
415 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
416 struct md_object *cobj, const struct lu_name *lname,
422 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
423 .mdo_lookup = dot_lustre_lookup,
424 .mdo_create = dot_lustre_create,
425 .mdo_rename = dot_lustre_rename,
426 .mdo_link = dot_lustre_link,
427 .mdo_unlink = dot_lustre_unlink,
430 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
435 if (ma->ma_need & MA_INODE) {
436 struct mdd_device *mdd = mdo2mdd(obj);
438 /* "fid" is a virtual object and hence does not have any "real"
439 * attributes. So we reuse attributes of .lustre for "fid" dir */
440 ma->ma_need |= MA_INODE;
441 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
444 ma->ma_valid |= MA_INODE;
447 /* "fid" directory does not have any striping information. */
448 if (ma->ma_need & MA_LOV) {
449 struct mdd_object *mdd_obj = md2mdd_obj(obj);
451 if (ma->ma_valid & MA_LOV)
454 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
455 S_ISDIR(mdd_object_type(mdd_obj))))
458 if (ma->ma_need & MA_LOV_DEF) {
459 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
462 ma->ma_valid |= MA_LOV;
471 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
472 const struct md_attr *ma)
477 static int obf_xattr_get(const struct lu_env *env,
478 struct md_object *obj, struct lu_buf *buf,
484 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
487 struct mdd_object *mdd_obj = md2mdd_obj(obj);
489 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
490 mdd_obj->mod_count++;
491 mdd_write_unlock(env, mdd_obj);
496 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
499 struct mdd_object *mdd_obj = md2mdd_obj(obj);
501 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
502 mdd_obj->mod_count--;
503 mdd_write_unlock(env, mdd_obj);
508 /** Nothing to list in "fid" directory */
509 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
510 const struct lu_rdpg *rdpg)
515 static int obf_path(const struct lu_env *env, struct md_object *obj,
516 char *path, int pathlen, __u64 recno, int *linkno)
521 static struct md_object_operations mdd_obf_obj_ops = {
522 .moo_attr_get = obf_attr_get,
523 .moo_attr_set = obf_attr_set,
524 .moo_xattr_get = obf_xattr_get,
525 .moo_open = obf_mdd_open,
526 .moo_close = obf_mdd_close,
527 .moo_readpage = obf_mdd_readpage,
532 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
533 * are valid. We also check if object with passed fid exists or not.
535 static int obf_lookup(const struct lu_env *env, struct md_object *p,
536 const struct lu_name *lname, struct lu_fid *f,
537 struct md_op_spec *spec)
539 char *name = (char *)lname->ln_name;
540 struct mdd_device *mdd = mdo2mdd(p);
541 struct mdd_object *child;
547 sscanf(name, SFID, &(f->f_seq), &(f->f_oid),
549 if (!fid_is_sane(f)) {
550 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
552 GOTO(out, rc = -EINVAL);
555 /* Check if object with this fid exists */
556 child = mdd_object_find(env, mdd, f);
560 GOTO(out, rc = PTR_ERR(child));
562 if (mdd_object_exists(child) == 0)
565 mdd_object_put(env, child);
571 static int obf_create(const struct lu_env *env, struct md_object *pobj,
572 const struct lu_name *lname, struct md_object *child,
573 struct md_op_spec *spec, struct md_attr* ma)
578 static int obf_rename(const struct lu_env *env,
579 struct md_object *src_pobj, struct md_object *tgt_pobj,
580 const struct lu_fid *lf, const struct lu_name *lsname,
581 struct md_object *tobj, const struct lu_name *ltname,
587 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
588 struct md_object *src_obj, const struct lu_name *lname,
594 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
595 struct md_object *cobj, const struct lu_name *lname,
601 static struct md_dir_operations mdd_obf_dir_ops = {
602 .mdo_lookup = obf_lookup,
603 .mdo_create = obf_create,
604 .mdo_rename = obf_rename,
605 .mdo_link = obf_link,
606 .mdo_unlink = obf_unlink
610 * Create special in-memory "fid" object for open-by-fid.
612 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
614 struct mdd_object *mdd_obf;
615 struct lu_object *obf_lu_obj;
618 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
620 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
621 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
622 GOTO(out, rc = -ENOENT);
624 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
625 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
626 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
627 /* Don't allow objects to be created in "fid" dir */
628 mdd_obf->mod_flags |= IMMUTE_OBJ;
630 obf_lu_obj = mdd2lu_obj(mdd_obf);
631 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
637 /** Setup ".lustre" directory object */
638 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
640 struct dt_object *dt_dot_lustre;
641 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
644 rc = create_dot_lustre_dir(env, m);
648 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
649 mdd_dot_lustre_name, fid);
650 if (IS_ERR(dt_dot_lustre)) {
651 rc = PTR_ERR(dt_dot_lustre);
655 /* references are released in mdd_device_shutdown() */
656 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
659 lu_object_put(env, &dt_dot_lustre->do_lu);
661 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
662 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
664 rc = mdd_obf_setup(env, m);
666 CERROR("Error initializing \"fid\" object - %d.\n", rc);
672 static int mdd_process_config(const struct lu_env *env,
673 struct lu_device *d, struct lustre_cfg *cfg)
675 struct mdd_device *m = lu2mdd_dev(d);
676 struct dt_device *dt = m->mdd_child;
677 struct lu_device *next = &dt->dd_lu_dev;
681 switch (cfg->lcfg_command) {
683 struct lprocfs_static_vars lvars;
685 lprocfs_mdd_init_vars(&lvars);
686 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
687 if (rc > 0 || rc == -ENOSYS)
688 /* we don't understand; pass it on */
689 rc = next->ld_ops->ldo_process_config(env, next, cfg);
693 rc = next->ld_ops->ldo_process_config(env, next, cfg);
696 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
698 rc = mdd_init_obd(env, m, cfg);
700 CERROR("lov init error %d \n", rc);
703 rc = mdd_txn_init_credits(env, m);
707 mdd_changelog_init(env, m);
710 mdd_device_shutdown(env, m, cfg);
712 rc = next->ld_ops->ldo_process_config(env, next, cfg);
720 static int mdd_lov_set_nextid(const struct lu_env *env,
721 struct mdd_device *mdd)
723 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
727 LASSERT(mds->mds_lov_objids != NULL);
728 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
729 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
730 mds->mds_lov_objids, NULL);
735 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
736 struct mdd_device *mdd)
738 /* XXX: to be implemented! */
743 static int mdd_recovery_complete(const struct lu_env *env,
746 struct mdd_device *mdd = lu2mdd_dev(d);
747 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
748 struct obd_device *obd = mdd2obd_dev(mdd);
752 LASSERT(mdd != NULL);
753 LASSERT(obd != NULL);
755 /* XXX: Do we need this in new stack? */
756 rc = mdd_lov_set_nextid(env, mdd);
758 CERROR("mdd_lov_set_nextid() failed %d\n",
763 /* XXX: cleanup unlink. */
764 rc = mdd_cleanup_unlink_llog(env, mdd);
766 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
771 /* Call that with obd_recovering = 1 just to update objids */
772 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
773 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
775 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
776 obd->obd_recovering = 0;
777 obd->obd_type->typ_dt_ops->o_postrecov(obd);
779 /* XXX: orphans handling. */
780 __mdd_orphan_cleanup(env, mdd);
781 rc = next->ld_ops->ldo_recovery_complete(env, next);
786 static int mdd_prepare(const struct lu_env *env,
787 struct lu_device *pdev,
788 struct lu_device *cdev)
790 struct mdd_device *mdd = lu2mdd_dev(cdev);
791 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
792 struct dt_object *root;
796 rc = next->ld_ops->ldo_prepare(env, cdev, next);
800 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
801 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
804 LASSERT(root != NULL);
805 lu_object_put(env, &root->do_lu);
806 rc = orph_index_init(env, mdd);
813 rc = mdd_dot_lustre_setup(env, mdd);
815 CERROR("Error(%d) initializing .lustre objects\n", rc);
823 const struct lu_device_operations mdd_lu_ops = {
824 .ldo_object_alloc = mdd_object_alloc,
825 .ldo_process_config = mdd_process_config,
826 .ldo_recovery_complete = mdd_recovery_complete,
827 .ldo_prepare = mdd_prepare,
831 * No permission check is needed.
833 static int mdd_root_get(const struct lu_env *env,
834 struct md_device *m, struct lu_fid *f)
836 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
839 *f = mdd->mdd_root_fid;
844 * No permission check is needed.
846 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
849 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
854 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
860 * No permission check is needed.
862 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
863 int *md_size, int *cookie_size)
865 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
868 *md_size = mdd_lov_mdsize(env, mdd);
869 *cookie_size = mdd_lov_cookiesize(env, mdd);
874 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
875 int mode, unsigned long timeout, __u32 alg,
876 struct lustre_capa_key *keys)
878 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
879 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
883 mds->mds_capa_keys = keys;
884 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
889 static int mdd_update_capa_key(const struct lu_env *env,
891 struct lustre_capa_key *key)
893 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
894 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
898 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
899 sizeof(*key), key, NULL);
903 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
904 struct lu_device_type *t,
905 struct lustre_cfg *lcfg)
908 struct mdd_device *m;
912 l = ERR_PTR(-ENOMEM);
914 md_device_init(&m->mdd_md_dev, t);
916 l->ld_ops = &mdd_lu_ops;
917 m->mdd_md_dev.md_ops = &mdd_ops;
918 md_upcall_init(&m->mdd_md_dev, NULL);
924 static struct lu_device *mdd_device_free(const struct lu_env *env,
925 struct lu_device *lu)
927 struct mdd_device *m = lu2mdd_dev(lu);
928 struct lu_device *next = &m->mdd_child->dd_lu_dev;
931 LASSERT(atomic_read(&lu->ld_ref) == 0);
932 md_device_fini(&m->mdd_md_dev);
937 static struct obd_ops mdd_obd_device_ops = {
938 .o_owner = THIS_MODULE
941 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
942 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
944 static struct lu_context_key mdd_ucred_key = {
945 .lct_tags = LCT_SESSION,
946 .lct_init = mdd_ucred_key_init,
947 .lct_fini = mdd_ucred_key_fini
950 struct md_ucred *md_ucred(const struct lu_env *env)
952 LASSERT(env->le_ses != NULL);
953 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
955 EXPORT_SYMBOL(md_ucred);
958 * context key constructor/destructor:
959 * mdd_capainfo_key_init, mdd_capainfo_key_fini
961 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
963 struct lu_context_key mdd_capainfo_key = {
964 .lct_tags = LCT_SESSION,
965 .lct_init = mdd_capainfo_key_init,
966 .lct_fini = mdd_capainfo_key_fini
969 struct md_capainfo *md_capainfo(const struct lu_env *env)
971 /* NB, in mdt_init0 */
972 if (env->le_ses == NULL)
974 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
976 EXPORT_SYMBOL(md_capainfo);
978 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
979 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
981 const struct md_device_operations mdd_ops = {
982 .mdo_statfs = mdd_statfs,
983 .mdo_root_get = mdd_root_get,
984 .mdo_maxsize_get = mdd_maxsize_get,
985 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
986 .mdo_update_capa_key= mdd_update_capa_key,
987 #ifdef HAVE_QUOTA_SUPPORT
989 .mqo_notify = mdd_quota_notify,
990 .mqo_setup = mdd_quota_setup,
991 .mqo_cleanup = mdd_quota_cleanup,
992 .mqo_recovery = mdd_quota_recovery,
993 .mqo_check = mdd_quota_check,
994 .mqo_on = mdd_quota_on,
995 .mqo_off = mdd_quota_off,
996 .mqo_setinfo = mdd_quota_setinfo,
997 .mqo_getinfo = mdd_quota_getinfo,
998 .mqo_setquota = mdd_quota_setquota,
999 .mqo_getquota = mdd_quota_getquota,
1000 .mqo_getoinfo = mdd_quota_getoinfo,
1001 .mqo_getoquota = mdd_quota_getoquota,
1002 .mqo_invalidate = mdd_quota_invalidate,
1003 .mqo_finvalidate = mdd_quota_finvalidate
1008 static struct lu_device_type_operations mdd_device_type_ops = {
1009 .ldto_init = mdd_type_init,
1010 .ldto_fini = mdd_type_fini,
1012 .ldto_start = mdd_type_start,
1013 .ldto_stop = mdd_type_stop,
1015 .ldto_device_alloc = mdd_device_alloc,
1016 .ldto_device_free = mdd_device_free,
1018 .ldto_device_init = mdd_device_init,
1019 .ldto_device_fini = mdd_device_fini
1022 static struct lu_device_type mdd_device_type = {
1023 .ldt_tags = LU_DEVICE_MD,
1024 .ldt_name = LUSTRE_MDD_NAME,
1025 .ldt_ops = &mdd_device_type_ops,
1026 .ldt_ctx_tags = LCT_MD_THREAD
1029 /* context key constructor: mdd_key_init */
1030 LU_KEY_INIT(mdd, struct mdd_thread_info);
1032 static void mdd_key_fini(const struct lu_context *ctx,
1033 struct lu_context_key *key, void *data)
1035 struct mdd_thread_info *info = data;
1036 if (info->mti_max_lmm != NULL)
1037 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1038 if (info->mti_max_cookie != NULL)
1039 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1040 mdd_buf_put(&info->mti_big_buf);
1045 /* context key: mdd_thread_key */
1046 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1048 static struct lu_local_obj_desc llod_capa_key = {
1049 .llod_name = CAPA_KEYS,
1050 .llod_oid = MDD_CAPA_KEYS_OID,
1054 static struct lu_local_obj_desc llod_mdd_orphan = {
1055 .llod_name = orph_index_name,
1056 .llod_oid = MDD_ORPHAN_OID,
1058 .llod_feat = &dt_directory_features,
1061 static struct lu_local_obj_desc llod_mdd_root = {
1062 .llod_name = mdd_root_dir_name,
1063 .llod_oid = MDD_ROOT_INDEX_OID,
1065 .llod_feat = &dt_directory_features,
1068 static int __init mdd_mod_init(void)
1070 struct lprocfs_static_vars lvars;
1071 lprocfs_mdd_init_vars(&lvars);
1073 llo_local_obj_register(&llod_capa_key);
1074 llo_local_obj_register(&llod_mdd_orphan);
1075 llo_local_obj_register(&llod_mdd_root);
1077 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1078 LUSTRE_MDD_NAME, &mdd_device_type);
1081 static void __exit mdd_mod_exit(void)
1083 llo_local_obj_unregister(&llod_capa_key);
1084 llo_local_obj_unregister(&llod_mdd_orphan);
1085 llo_local_obj_unregister(&llod_mdd_root);
1087 class_unregister_type(LUSTRE_MDD_NAME);
1090 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1091 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1092 MODULE_LICENSE("GPL");
1094 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);