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, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #ifdef HAVE_EXT4_LDISKFS
47 #include <ldiskfs/ldiskfs_jbd2.h>
49 #include <linux/jbd.h>
52 #include <obd_class.h>
53 #include <lustre_ver.h>
54 #include <obd_support.h>
55 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <lustre_fid.h>
59 #ifdef HAVE_EXT4_LDISKFS
60 #include <ldiskfs/ldiskfs.h>
62 #include <linux/ldiskfs_fs.h>
64 #include <lustre_mds.h>
65 #include <lustre/lustre_idl.h>
66 #include <lustre_disk.h> /* for changelogs */
67 #include <lustre_param.h>
68 #include <lustre_fid.h>
70 #include "mdd_internal.h"
72 const struct md_device_operations mdd_ops;
73 static struct lu_device_type mdd_device_type;
75 static const char mdd_root_dir_name[] = "ROOT";
76 static const char mdd_obf_dir_name[] = "fid";
78 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
79 const char *name, struct lu_device *next)
81 struct mdd_device *mdd = lu2mdd_dev(d);
85 mdd->mdd_child = lu2dt_dev(next);
87 /* Prepare transactions callbacks. */
88 mdd->mdd_txn_cb.dtc_txn_start = NULL;
89 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
90 mdd->mdd_txn_cb.dtc_txn_commit = NULL;
91 mdd->mdd_txn_cb.dtc_cookie = mdd;
92 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
93 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
94 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
95 /* sync permission changes */
96 mdd->mdd_sync_permission = 1;
98 rc = mdd_procfs_init(mdd, name);
102 static struct lu_device *mdd_device_fini(const struct lu_env *env,
105 struct mdd_device *mdd = lu2mdd_dev(d);
106 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
109 rc = mdd_procfs_fini(mdd);
111 CERROR("proc fini error %d \n", rc);
117 static void mdd_changelog_fini(const struct lu_env *env,
118 struct mdd_device *mdd);
120 static void mdd_device_shutdown(const struct lu_env *env,
121 struct mdd_device *m, struct lustre_cfg *cfg)
124 mdd_changelog_fini(env, m);
125 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
126 if (m->mdd_dot_lustre_objs.mdd_obf)
127 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
128 if (m->mdd_dot_lustre)
129 mdd_object_put(env, m->mdd_dot_lustre);
131 mdd_fini_obd(env, m, cfg);
132 orph_index_fini(env, m);
133 if (m->mdd_capa != NULL) {
134 lu_object_put(env, &m->mdd_capa->do_lu);
137 /* remove upcall device*/
138 md_upcall_fini(&m->mdd_md_dev);
142 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
145 struct mdd_device *mdd = (struct mdd_device *)data;
146 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
149 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
150 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
153 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
154 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
155 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
156 llh->lgh_id.lgl_oid);
158 mdd->mdd_cl.mc_index = rec->cr.cr_index;
159 RETURN(LLOG_PROC_BREAK);
162 static int changelog_user_init_cb(struct llog_handle *llh,
163 struct llog_rec_hdr *hdr, void *data)
165 struct mdd_device *mdd = (struct mdd_device *)data;
166 struct llog_changelog_user_rec *rec =
167 (struct llog_changelog_user_rec *)hdr;
170 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
171 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
173 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
174 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
175 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
177 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
178 mdd->mdd_cl.mc_lastuser = rec->cur_id;
179 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
181 RETURN(LLOG_PROC_BREAK);
185 static int mdd_changelog_llog_init(struct mdd_device *mdd)
187 struct obd_device *obd = mdd2obd_dev(mdd);
188 struct llog_ctxt *ctxt;
191 /* Find last changelog entry number */
192 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
194 CERROR("no changelog context\n");
197 if (!ctxt->loc_handle) {
202 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
206 CERROR("changelog init failed: %d\n", rc);
209 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
210 mdd->mdd_cl.mc_index);
212 /* Find last changelog user id */
213 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
215 CERROR("no changelog user context\n");
218 if (!ctxt->loc_handle) {
223 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
228 CERROR("changelog user init failed: %d\n", rc);
232 /* If we have registered users, assume we want changelogs on */
233 if (mdd->mdd_cl.mc_lastuser > 0)
234 rc = mdd_changelog_on(mdd, 1);
239 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
243 mdd->mdd_cl.mc_index = 0;
244 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
245 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
246 mdd->mdd_cl.mc_flags = 0; /* off by default */
247 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
248 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
249 mdd->mdd_cl.mc_lastuser = 0;
251 rc = mdd_changelog_llog_init(mdd);
253 CERROR("Changelog setup during init failed %d\n", rc);
254 mdd->mdd_cl.mc_flags |= CLM_ERR;
260 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
262 mdd->mdd_cl.mc_flags = 0;
265 /* Start / stop recording */
266 int mdd_changelog_on(struct mdd_device *mdd, int on)
270 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
271 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
272 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
273 CERROR("Changelogs cannot be enabled due to error "
274 "condition (see %s log).\n",
275 mdd2obd_dev(mdd)->obd_name);
278 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
279 mdd->mdd_cl.mc_flags |= CLM_ON;
280 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
281 rc = mdd_changelog_write_header(mdd, CLM_START);
283 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
284 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
285 rc = mdd_changelog_write_header(mdd, CLM_FINI);
286 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
287 mdd->mdd_cl.mc_flags &= ~CLM_ON;
288 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
293 static __u64 cl_time(void) {
296 cfs_fs_time_current(&time);
297 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
300 /** Add a changelog entry \a rec to the changelog llog
303 * \param handle - currently ignored since llogs start their own transaction;
304 * this will hopefully be fixed in llog rewrite
307 int mdd_changelog_llog_write(struct mdd_device *mdd,
308 struct llog_changelog_rec *rec,
309 struct thandle *handle)
311 struct obd_device *obd = mdd2obd_dev(mdd);
312 struct llog_ctxt *ctxt;
315 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
316 /* llog_lvfs_write_rec sets the llog tail len */
317 rec->cr_hdr.lrh_type = CHANGELOG_REC;
318 rec->cr.cr_time = cl_time();
319 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
320 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
321 but as long as the MDD transactions are ordered correctly for e.g.
322 rename conflicts, I don't think this should matter. */
323 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
324 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
325 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
329 /* nested journal transaction */
330 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
336 /** Remove entries with indicies up to and including \a endrec from the
342 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
344 struct obd_device *obd = mdd2obd_dev(mdd);
345 struct llog_ctxt *ctxt;
346 long long unsigned cur;
349 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
353 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
354 cur = (long long)mdd->mdd_cl.mc_index;
355 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
359 /* purge to "0" is shorthand for everything */
363 /* If purging all records, write a header entry so we don't have an
364 empty catalog and we're sure to have a valid starting index next
365 time. In case of crash, we just restart with old log so we're
368 /* XXX: transaction is started by llog itself */
369 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
374 /* Some records were purged, so reset repeat-access time (so we
375 record new mtime update records, so users can see a file has been
376 changed since the last purge) */
377 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
379 /* XXX: transaction is started by llog itself */
380 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
386 /** Add a CL_MARK record to the changelog
388 * \param markerflags - CLM_*
391 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
393 struct obd_device *obd = mdd2obd_dev(mdd);
394 struct llog_changelog_rec *rec;
396 int len = strlen(obd->obd_name);
400 reclen = llog_data_len(sizeof(*rec) + len);
401 OBD_ALLOC(rec, reclen);
405 rec->cr.cr_flags = CLF_VERSION;
406 rec->cr.cr_type = CL_MARK;
407 rec->cr.cr_namelen = len;
408 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
409 /* Status and action flags */
410 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
412 /* XXX: transaction is started by llog itself */
413 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
414 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
416 /* assume on or off event; reset repeat-access time */
417 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
419 OBD_FREE(rec, reclen);
424 * Create ".lustre" directory.
426 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
428 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
429 struct md_object *mdo;
432 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
433 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
434 mdd_root_dir_name, dot_lustre_name,
435 fid, &dt_directory_features);
436 /* .lustre dir may be already present */
437 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
439 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
440 dot_lustre_name, PFID(fid), rc);
445 lu_object_put(env, &mdo->mo_lu);
451 static int dot_lustre_mdd_permission(const struct lu_env *env,
452 struct md_object *pobj,
453 struct md_object *cobj,
454 struct md_attr *attr, int mask)
456 if (mask & ~(MAY_READ | MAY_EXEC))
462 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
463 struct md_object *obj, struct lu_buf *buf,
469 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
470 struct md_object *obj, struct lu_buf *buf)
475 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
476 struct md_object *obj,
477 const struct lu_buf *buf, const char *name,
483 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
484 struct md_object *obj,
490 static int dot_lustre_mdd_readlink(const struct lu_env *env,
491 struct md_object *obj, struct lu_buf *buf)
496 static int dot_lustre_mdd_object_create(const struct lu_env *env,
497 struct md_object *obj,
498 const struct md_op_spec *spec,
504 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
505 struct md_object *obj,
506 const struct md_attr *ma)
511 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
512 struct md_object *obj,
518 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
521 struct mdd_object *mdd_obj = md2mdd_obj(obj);
523 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
524 mdd_obj->mod_count++;
525 mdd_write_unlock(env, mdd_obj);
530 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
531 struct md_attr *ma, int mode)
533 struct mdd_object *mdd_obj = md2mdd_obj(obj);
535 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
536 mdd_obj->mod_count--;
537 mdd_write_unlock(env, mdd_obj);
542 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
543 struct md_object *obj)
548 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
549 char *path, int pathlen, __u64 *recno, int *linkno)
554 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
555 struct lov_mds_md *lmm, struct ldlm_extent *extent,
556 struct lustre_handle *lockh)
561 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
562 struct lov_mds_md *lmm, struct lustre_handle *lockh)
567 static struct md_object_operations mdd_dot_lustre_obj_ops = {
568 .moo_permission = dot_lustre_mdd_permission,
569 .moo_attr_get = mdd_attr_get,
570 .moo_attr_set = mdd_attr_set,
571 .moo_xattr_get = dot_lustre_mdd_xattr_get,
572 .moo_xattr_list = dot_lustre_mdd_xattr_list,
573 .moo_xattr_set = dot_lustre_mdd_xattr_set,
574 .moo_xattr_del = dot_lustre_mdd_xattr_del,
575 .moo_readpage = mdd_readpage,
576 .moo_readlink = dot_lustre_mdd_readlink,
577 .moo_object_create = dot_lustre_mdd_object_create,
578 .moo_ref_add = dot_lustre_mdd_ref_add,
579 .moo_ref_del = dot_lustre_mdd_ref_del,
580 .moo_open = dot_lustre_mdd_open,
581 .moo_close = dot_lustre_mdd_close,
582 .moo_capa_get = mdd_capa_get,
583 .moo_object_sync = dot_lustre_mdd_object_sync,
584 .moo_path = dot_lustre_mdd_path,
585 .moo_file_lock = dot_file_lock,
586 .moo_file_unlock = dot_file_unlock,
590 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
591 const struct lu_name *lname, struct lu_fid *f,
592 struct md_op_spec *spec)
594 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
602 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
603 struct md_object *obj,
609 static int dot_lustre_mdd_create(const struct lu_env *env,
610 struct md_object *pobj,
611 const struct lu_name *lname,
612 struct md_object *child,
613 struct md_op_spec *spec,
619 static int dot_lustre_mdd_create_data(const struct lu_env *env,
622 const struct md_op_spec *spec,
628 static int dot_lustre_mdd_rename(const struct lu_env *env,
629 struct md_object *src_pobj,
630 struct md_object *tgt_pobj,
631 const struct lu_fid *lf,
632 const struct lu_name *lsname,
633 struct md_object *tobj,
634 const struct lu_name *ltname,
640 static int dot_lustre_mdd_link(const struct lu_env *env,
641 struct md_object *tgt_obj,
642 struct md_object *src_obj,
643 const struct lu_name *lname,
649 static int dot_lustre_mdd_unlink(const struct lu_env *env,
650 struct md_object *pobj,
651 struct md_object *cobj,
652 const struct lu_name *lname,
658 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
659 struct md_object *obj,
660 const struct lu_name *lname,
661 const struct lu_fid *fid,
662 const struct md_attr *ma)
667 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
668 struct md_object *obj,
669 const struct lu_name *lname,
670 const struct md_attr *ma)
675 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
676 struct md_object *pobj,
677 struct md_object *tobj,
678 const struct lu_fid *fid,
679 const struct lu_name *lname,
686 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
687 .mdo_is_subdir = mdd_is_subdir,
688 .mdo_lookup = dot_lustre_mdd_lookup,
689 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
690 .mdo_create = dot_lustre_mdd_create,
691 .mdo_create_data = dot_lustre_mdd_create_data,
692 .mdo_rename = dot_lustre_mdd_rename,
693 .mdo_link = dot_lustre_mdd_link,
694 .mdo_unlink = dot_lustre_mdd_unlink,
695 .mdo_name_insert = dot_lustre_mdd_name_insert,
696 .mdo_name_remove = dot_lustre_mdd_name_remove,
697 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
700 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
705 if (ma->ma_need & MA_INODE) {
706 struct mdd_device *mdd = mdo2mdd(obj);
708 /* "fid" is a virtual object and hence does not have any "real"
709 * attributes. So we reuse attributes of .lustre for "fid" dir */
710 ma->ma_need |= MA_INODE;
711 rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
714 ma->ma_valid |= MA_INODE;
717 /* "fid" directory does not have any striping information. */
718 if (ma->ma_need & MA_LOV) {
719 struct mdd_object *mdd_obj = md2mdd_obj(obj);
721 if (ma->ma_valid & MA_LOV)
724 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
725 S_ISDIR(mdd_object_type(mdd_obj))))
728 if (ma->ma_need & MA_LOV_DEF) {
729 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
731 ma->ma_lmm_size = rc;
732 ma->ma_valid |= MA_LOV;
741 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
742 const struct md_attr *ma)
747 static int obf_xattr_list(const struct lu_env *env,
748 struct md_object *obj, struct lu_buf *buf)
753 static int obf_xattr_get(const struct lu_env *env,
754 struct md_object *obj, struct lu_buf *buf,
760 static int obf_xattr_set(const struct lu_env *env,
761 struct md_object *obj,
762 const struct lu_buf *buf, const char *name,
768 static int obf_xattr_del(const struct lu_env *env,
769 struct md_object *obj,
775 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
778 struct mdd_object *mdd_obj = md2mdd_obj(obj);
780 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
781 mdd_obj->mod_count++;
782 mdd_write_unlock(env, mdd_obj);
787 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
788 struct md_attr *ma, int mode)
790 struct mdd_object *mdd_obj = md2mdd_obj(obj);
792 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
793 mdd_obj->mod_count--;
794 mdd_write_unlock(env, mdd_obj);
799 /** Nothing to list in "fid" directory */
800 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
801 const struct lu_rdpg *rdpg)
806 static int obf_path(const struct lu_env *env, struct md_object *obj,
807 char *path, int pathlen, __u64 *recno, int *linkno)
812 static dt_obj_version_t obf_version_get(const struct lu_env *env,
813 struct md_object *obj)
818 static struct md_object_operations mdd_obf_obj_ops = {
819 .moo_attr_get = obf_attr_get,
820 .moo_attr_set = obf_attr_set,
821 .moo_xattr_list = obf_xattr_list,
822 .moo_xattr_get = obf_xattr_get,
823 .moo_xattr_set = obf_xattr_set,
824 .moo_xattr_del = obf_xattr_del,
825 .moo_open = obf_mdd_open,
826 .moo_close = obf_mdd_close,
827 .moo_readpage = obf_mdd_readpage,
828 .moo_version_get = obf_version_get,
833 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
834 * are valid. We also check if object with passed fid exists or not.
836 static int obf_lookup(const struct lu_env *env, struct md_object *p,
837 const struct lu_name *lname, struct lu_fid *f,
838 struct md_op_spec *spec)
840 char *name = (char *)lname->ln_name;
841 struct mdd_device *mdd = mdo2mdd(p);
842 struct mdd_object *child;
848 sscanf(name, SFID, RFID(f));
849 if (!fid_is_sane(f)) {
850 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
852 GOTO(out, rc = -EINVAL);
855 /* Check if object with this fid exists */
856 child = mdd_object_find(env, mdd, f);
860 GOTO(out, rc = PTR_ERR(child));
862 if (mdd_object_exists(child) == 0)
865 mdd_object_put(env, child);
871 static int obf_create(const struct lu_env *env, struct md_object *pobj,
872 const struct lu_name *lname, struct md_object *child,
873 struct md_op_spec *spec, struct md_attr* ma)
878 static int obf_rename(const struct lu_env *env,
879 struct md_object *src_pobj, struct md_object *tgt_pobj,
880 const struct lu_fid *lf, const struct lu_name *lsname,
881 struct md_object *tobj, const struct lu_name *ltname,
887 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
888 struct md_object *src_obj, const struct lu_name *lname,
894 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
895 struct md_object *cobj, const struct lu_name *lname,
901 static struct md_dir_operations mdd_obf_dir_ops = {
902 .mdo_lookup = obf_lookup,
903 .mdo_create = obf_create,
904 .mdo_rename = obf_rename,
905 .mdo_link = obf_link,
906 .mdo_unlink = obf_unlink
910 * Create special in-memory "fid" object for open-by-fid.
912 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
914 struct mdd_object *mdd_obf;
915 struct lu_object *obf_lu_obj;
918 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
920 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
921 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
922 GOTO(out, rc = -ENOENT);
924 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
925 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
926 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
927 /* Don't allow objects to be created in "fid" dir */
928 mdd_obf->mod_flags |= IMMUTE_OBJ;
930 obf_lu_obj = mdd2lu_obj(mdd_obf);
931 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
937 /** Setup ".lustre" directory object */
938 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
940 struct dt_object *dt_dot_lustre;
941 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
944 rc = create_dot_lustre_dir(env, m);
948 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
949 dot_lustre_name, fid);
950 if (IS_ERR(dt_dot_lustre)) {
951 rc = PTR_ERR(dt_dot_lustre);
955 /* references are released in mdd_device_shutdown() */
956 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
959 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
960 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
962 rc = mdd_obf_setup(env, m);
964 CERROR("Error initializing \"fid\" object - %d.\n", rc);
970 static int mdd_process_config(const struct lu_env *env,
971 struct lu_device *d, struct lustre_cfg *cfg)
973 struct mdd_device *m = lu2mdd_dev(d);
974 struct dt_device *dt = m->mdd_child;
975 struct lu_device *next = &dt->dd_lu_dev;
979 switch (cfg->lcfg_command) {
981 struct lprocfs_static_vars lvars;
983 lprocfs_mdd_init_vars(&lvars);
984 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
985 if (rc > 0 || rc == -ENOSYS)
986 /* we don't understand; pass it on */
987 rc = next->ld_ops->ldo_process_config(env, next, cfg);
991 rc = next->ld_ops->ldo_process_config(env, next, cfg);
994 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
996 rc = mdd_init_obd(env, m, cfg);
998 CERROR("lov init error %d\n", rc);
1002 mdd_changelog_init(env, m);
1005 mdd_device_shutdown(env, m, cfg);
1007 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1015 static int mdd_lov_set_nextid(const struct lu_env *env,
1016 struct mdd_device *mdd)
1018 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1022 LASSERT(mds->mds_lov_objids != NULL);
1023 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1024 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1025 mds->mds_lov_objids, NULL);
1030 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1031 struct mdd_device *mdd)
1033 /* XXX: to be implemented! */
1038 static int mdd_recovery_complete(const struct lu_env *env,
1039 struct lu_device *d)
1041 struct mdd_device *mdd = lu2mdd_dev(d);
1042 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1043 struct obd_device *obd = mdd2obd_dev(mdd);
1047 LASSERT(mdd != NULL);
1048 LASSERT(obd != NULL);
1050 /* XXX: Do we need this in new stack? */
1051 rc = mdd_lov_set_nextid(env, mdd);
1053 CERROR("mdd_lov_set_nextid() failed %d\n",
1058 /* XXX: cleanup unlink. */
1059 rc = mdd_cleanup_unlink_llog(env, mdd);
1061 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1066 /* Call that with obd_recovering = 1 just to update objids */
1067 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1068 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1070 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1071 cfs_spin_lock(&obd->obd_dev_lock);
1072 obd->obd_recovering = 0;
1073 cfs_spin_unlock(&obd->obd_dev_lock);
1074 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1076 /* XXX: orphans handling. */
1077 __mdd_orphan_cleanup(env, mdd);
1078 rc = next->ld_ops->ldo_recovery_complete(env, next);
1083 static int mdd_prepare(const struct lu_env *env,
1084 struct lu_device *pdev,
1085 struct lu_device *cdev)
1087 struct mdd_device *mdd = lu2mdd_dev(cdev);
1088 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1089 struct dt_object *root;
1094 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1098 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1099 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1100 &mdd->mdd_root_fid);
1101 if (!IS_ERR(root)) {
1102 LASSERT(root != NULL);
1103 lu_object_put(env, &root->do_lu);
1104 rc = orph_index_init(env, mdd);
1111 rc = mdd_dot_lustre_setup(env, mdd);
1113 CERROR("Error(%d) initializing .lustre objects\n", rc);
1117 /* we use capa file to declare llog changes,
1118 * will be fixed with new llog in 2.3 */
1119 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1121 mdd->mdd_capa = root;
1129 const struct lu_device_operations mdd_lu_ops = {
1130 .ldo_object_alloc = mdd_object_alloc,
1131 .ldo_process_config = mdd_process_config,
1132 .ldo_recovery_complete = mdd_recovery_complete,
1133 .ldo_prepare = mdd_prepare,
1137 * No permission check is needed.
1139 static int mdd_root_get(const struct lu_env *env,
1140 struct md_device *m, struct lu_fid *f)
1142 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1145 *f = mdd->mdd_root_fid;
1150 * No permission check is needed.
1152 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1153 struct obd_statfs *sfs)
1155 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1160 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1166 * No permission check is needed.
1168 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1169 int *md_size, int *cookie_size)
1171 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1174 *md_size = mdd_lov_mdsize(env, mdd);
1175 *cookie_size = mdd_lov_cookiesize(env, mdd);
1180 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1181 int mode, unsigned long timeout, __u32 alg,
1182 struct lustre_capa_key *keys)
1184 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1185 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1189 /* need barrier for mds_capa_keys access. */
1190 cfs_down_write(&mds->mds_notify_lock);
1191 mds->mds_capa_keys = keys;
1192 cfs_up_write(&mds->mds_notify_lock);
1194 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1195 timeout, alg, keys);
1199 static int mdd_update_capa_key(const struct lu_env *env,
1200 struct md_device *m,
1201 struct lustre_capa_key *key)
1203 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1204 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1205 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1209 rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
1210 KEY_CAPA_KEY, sizeof(info), &info, NULL);
1214 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1217 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1219 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1220 return (*h == NULL ? -ENOENT : 0);
1223 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1224 struct lu_device_type *t,
1225 struct lustre_cfg *lcfg)
1227 struct lu_device *l;
1228 struct mdd_device *m;
1232 l = ERR_PTR(-ENOMEM);
1234 md_device_init(&m->mdd_md_dev, t);
1236 l->ld_ops = &mdd_lu_ops;
1237 m->mdd_md_dev.md_ops = &mdd_ops;
1238 md_upcall_init(&m->mdd_md_dev, NULL);
1244 static struct lu_device *mdd_device_free(const struct lu_env *env,
1245 struct lu_device *lu)
1247 struct mdd_device *m = lu2mdd_dev(lu);
1248 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1251 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1252 md_device_fini(&m->mdd_md_dev);
1257 static struct obd_ops mdd_obd_device_ops = {
1258 .o_owner = THIS_MODULE
1261 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1262 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1264 static struct lu_context_key mdd_ucred_key = {
1265 .lct_tags = LCT_SESSION,
1266 .lct_init = mdd_ucred_key_init,
1267 .lct_fini = mdd_ucred_key_fini
1270 struct md_ucred *md_ucred(const struct lu_env *env)
1272 LASSERT(env->le_ses != NULL);
1273 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1275 EXPORT_SYMBOL(md_ucred);
1278 * context key constructor/destructor:
1279 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1281 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1283 struct lu_context_key mdd_capainfo_key = {
1284 .lct_tags = LCT_SESSION,
1285 .lct_init = mdd_capainfo_key_init,
1286 .lct_fini = mdd_capainfo_key_fini
1289 struct md_capainfo *md_capainfo(const struct lu_env *env)
1291 /* NB, in mdt_init0 */
1292 if (env->le_ses == NULL)
1294 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1296 EXPORT_SYMBOL(md_capainfo);
1299 * context key constructor/destructor:
1300 * mdd_quota_key_init, mdd_quota_key_fini
1302 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1304 struct lu_context_key mdd_quota_key = {
1305 .lct_tags = LCT_SESSION,
1306 .lct_init = mdd_quota_key_init,
1307 .lct_fini = mdd_quota_key_fini
1310 struct md_quota *md_quota(const struct lu_env *env)
1312 LASSERT(env->le_ses != NULL);
1313 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1315 EXPORT_SYMBOL(md_quota);
1317 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1319 struct llog_ctxt *ctxt;
1320 struct llog_changelog_user_rec *rec;
1324 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1330 llog_ctxt_put(ctxt);
1334 /* Assume we want it on since somebody registered */
1335 rc = mdd_changelog_on(mdd, 1);
1339 rec->cur_hdr.lrh_len = sizeof(*rec);
1340 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1341 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1342 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1343 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1344 CERROR("Maximum number of changelog users exceeded!\n");
1345 GOTO(out, rc = -EOVERFLOW);
1347 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1348 rec->cur_endrec = mdd->mdd_cl.mc_index;
1349 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1351 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1353 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1356 llog_ctxt_put(ctxt);
1360 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1361 struct thandle *handle)
1366 /* XXX: this is a temporary solution to declare llog changes
1367 * will be fixed in 2.3 with new llog implementation */
1369 LASSERT(mdd->mdd_capa);
1371 /* the llog record could be canceled either by modifying
1372 * the plain llog's header or by destroying the llog itself
1373 * when this record is the last one in it, it can't be known
1374 * here, but the catlog's header will also be modified for
1375 * the second case, then the first case can be covered and
1376 * is no need to declare it */
1378 /* destroy empty plain log */
1379 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1383 /* record the catlog's header if an empty plain log was destroyed */
1384 rc = dt_declare_record_write(env, mdd->mdd_capa,
1385 sizeof(struct llog_logid_rec), 0, handle);
1389 struct mdd_changelog_user_data {
1390 __u64 mcud_endrec; /**< purge record for this user */
1391 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1393 __u32 mcud_minid; /**< user id with lowest rec reference */
1394 __u32 mcud_usercount;
1396 struct mdd_device *mcud_mdd;
1397 const struct lu_env *mcud_env;
1399 #define MCUD_UNREGISTER -1LL
1402 * 1. Find the smallest record everyone is willing to purge
1403 * 2. Update the last purgeable record for this user
1405 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1406 struct llog_rec_hdr *hdr, void *data)
1408 struct llog_changelog_user_rec *rec;
1409 struct mdd_changelog_user_data *mcud =
1410 (struct mdd_changelog_user_data *)data;
1414 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1416 rec = (struct llog_changelog_user_rec *)hdr;
1418 mcud->mcud_usercount++;
1420 /* If we have a new endrec for this id, use it for the following
1421 min check instead of its old value */
1422 if (rec->cur_id == mcud->mcud_id)
1423 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1425 /* Track the minimum referenced record */
1426 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1427 mcud->mcud_minid = rec->cur_id;
1428 mcud->mcud_minrec = rec->cur_endrec;
1431 if (rec->cur_id != mcud->mcud_id)
1434 /* Update this user's record */
1435 mcud->mcud_found = 1;
1437 /* Special case: unregister this user */
1438 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1439 struct llog_cookie cookie;
1441 struct mdd_device *mdd = mcud->mcud_mdd;
1443 cookie.lgc_lgl = llh->lgh_id;
1444 cookie.lgc_index = hdr->lrh_index;
1446 /* XXX This is a workaround for the deadlock of changelog
1447 * adding vs. changelog cancelling. LU-81. */
1448 th = mdd_trans_create(mcud->mcud_env, mdd);
1450 CERROR("Cannot get thandle\n");
1454 rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
1458 rc = mdd_trans_start(mcud->mcud_env, mdd, th);
1462 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1465 mcud->mcud_usercount--;
1468 mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
1472 /* Update the endrec */
1473 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1474 mcud->mcud_id, rec->cur_endrec);
1476 /* hdr+1 is loc of data */
1477 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1478 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1484 static int mdd_changelog_user_purge(const struct lu_env *env,
1485 struct mdd_device *mdd, int id,
1488 struct mdd_changelog_user_data data;
1489 struct llog_ctxt *ctxt;
1493 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1496 data.mcud_minid = 0;
1497 data.mcud_minrec = 0;
1498 data.mcud_usercount = 0;
1499 data.mcud_endrec = endrec;
1500 data.mcud_mdd = mdd;
1501 data.mcud_env = env;
1502 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1503 endrec = mdd->mdd_cl.mc_index;
1504 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1505 if ((data.mcud_endrec == 0) ||
1506 ((data.mcud_endrec > endrec) &&
1507 (data.mcud_endrec != MCUD_UNREGISTER)))
1508 data.mcud_endrec = endrec;
1510 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1513 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1515 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1516 (void *)&data, 0, 0);
1517 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1518 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1519 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1520 data.mcud_minrec, data.mcud_minid);
1521 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1523 CWARN("Could not determine changelog records to purge; rc=%d\n",
1527 llog_ctxt_put(ctxt);
1529 if (!data.mcud_found) {
1530 CWARN("No entry for user %d. Last changelog reference is "
1531 LPD64" by changelog user %d\n", data.mcud_id,
1532 data.mcud_minrec, data.mcud_minid);
1536 if (!rc && data.mcud_usercount == 0)
1537 /* No more users; turn changelogs off */
1538 rc = mdd_changelog_on(mdd, 0);
1544 * May be called remotely from mdt_iocontrol_handle or locally from
1545 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1546 * an obd_ioctl_data format (but local ioctl handler does).
1548 * \param len - data len
1549 * \param karg - ioctl data, in kernel space
1551 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1552 unsigned int cmd, int len, void *karg)
1554 struct mdd_device *mdd;
1555 struct obd_ioctl_data *data = karg;
1559 mdd = lu2mdd_dev(&m->md_lu_dev);
1561 /* Doesn't use obd_ioctl_data */
1563 case OBD_IOC_CHANGELOG_CLEAR: {
1564 struct changelog_setinfo *cs = karg;
1565 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1569 case OBD_IOC_GET_MNTOPT: {
1570 mntopt_t *mntopts = (mntopt_t *)karg;
1571 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1576 /* Below ioctls use obd_ioctl_data */
1577 if (len != sizeof(*data)) {
1578 CERROR("Bad ioctl size %d\n", len);
1581 if (data->ioc_version != OBD_IOCTL_VERSION) {
1582 CERROR("Bad magic %x != %x\n", data->ioc_version,
1588 case OBD_IOC_CHANGELOG_REG:
1589 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1591 case OBD_IOC_CHANGELOG_DEREG:
1592 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1602 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1603 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1606 const struct md_device_operations mdd_ops = {
1607 .mdo_statfs = mdd_statfs,
1608 .mdo_root_get = mdd_root_get,
1609 .mdo_maxsize_get = mdd_maxsize_get,
1610 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1611 .mdo_update_capa_key= mdd_update_capa_key,
1612 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1613 .mdo_iocontrol = mdd_iocontrol,
1614 #ifdef HAVE_QUOTA_SUPPORT
1616 .mqo_notify = mdd_quota_notify,
1617 .mqo_setup = mdd_quota_setup,
1618 .mqo_cleanup = mdd_quota_cleanup,
1619 .mqo_recovery = mdd_quota_recovery,
1620 .mqo_check = mdd_quota_check,
1621 .mqo_on = mdd_quota_on,
1622 .mqo_off = mdd_quota_off,
1623 .mqo_setinfo = mdd_quota_setinfo,
1624 .mqo_getinfo = mdd_quota_getinfo,
1625 .mqo_setquota = mdd_quota_setquota,
1626 .mqo_getquota = mdd_quota_getquota,
1627 .mqo_getoinfo = mdd_quota_getoinfo,
1628 .mqo_getoquota = mdd_quota_getoquota,
1629 .mqo_invalidate = mdd_quota_invalidate,
1630 .mqo_finvalidate = mdd_quota_finvalidate
1635 static struct lu_device_type_operations mdd_device_type_ops = {
1636 .ldto_init = mdd_type_init,
1637 .ldto_fini = mdd_type_fini,
1639 .ldto_start = mdd_type_start,
1640 .ldto_stop = mdd_type_stop,
1642 .ldto_device_alloc = mdd_device_alloc,
1643 .ldto_device_free = mdd_device_free,
1645 .ldto_device_init = mdd_device_init,
1646 .ldto_device_fini = mdd_device_fini
1649 static struct lu_device_type mdd_device_type = {
1650 .ldt_tags = LU_DEVICE_MD,
1651 .ldt_name = LUSTRE_MDD_NAME,
1652 .ldt_ops = &mdd_device_type_ops,
1653 .ldt_ctx_tags = LCT_MD_THREAD
1656 /* context key constructor: mdd_key_init */
1657 LU_KEY_INIT(mdd, struct mdd_thread_info);
1659 static void mdd_key_fini(const struct lu_context *ctx,
1660 struct lu_context_key *key, void *data)
1662 struct mdd_thread_info *info = data;
1663 if (info->mti_max_lmm != NULL)
1664 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1665 if (info->mti_max_cookie != NULL)
1666 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1667 mdd_buf_put(&info->mti_big_buf);
1672 /* context key: mdd_thread_key */
1673 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1675 static struct lu_local_obj_desc llod_capa_key = {
1676 .llod_name = CAPA_KEYS,
1677 .llod_oid = MDD_CAPA_KEYS_OID,
1681 static struct lu_local_obj_desc llod_mdd_orphan = {
1682 .llod_name = orph_index_name,
1683 .llod_oid = MDD_ORPHAN_OID,
1685 .llod_feat = &dt_directory_features,
1688 static struct lu_local_obj_desc llod_mdd_root = {
1689 .llod_name = mdd_root_dir_name,
1690 .llod_oid = MDD_ROOT_INDEX_OID,
1692 .llod_feat = &dt_directory_features,
1695 static int __init mdd_mod_init(void)
1697 struct lprocfs_static_vars lvars;
1698 lprocfs_mdd_init_vars(&lvars);
1700 llo_local_obj_register(&llod_capa_key);
1701 llo_local_obj_register(&llod_mdd_orphan);
1702 llo_local_obj_register(&llod_mdd_root);
1704 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1705 LUSTRE_MDD_NAME, &mdd_device_type);
1708 static void __exit mdd_mod_exit(void)
1710 llo_local_obj_unregister(&llod_capa_key);
1711 llo_local_obj_unregister(&llod_mdd_orphan);
1712 llo_local_obj_unregister(&llod_mdd_root);
1714 class_unregister_type(LUSTRE_MDD_NAME);
1717 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1718 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1719 MODULE_LICENSE("GPL");
1721 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);