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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/mdd/mdd_device.c
41 * Lustre Metadata Server (mdd) routines
43 * Author: Wang Di <wangdi@clusterfs.com>
47 # define EXPORT_SYMTAB
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
52 #ifdef HAVE_EXT4_LDISKFS
53 #include <ldiskfs/ldiskfs_jbd2.h>
55 #include <linux/jbd.h>
58 #include <obd_class.h>
59 #include <lustre_ver.h>
60 #include <obd_support.h>
61 #include <lprocfs_status.h>
63 #include <lustre_disk.h>
64 #include <lustre_fid.h>
65 #ifdef HAVE_EXT4_LDISKFS
66 #include <ldiskfs/ldiskfs.h>
68 #include <linux/ldiskfs_fs.h>
70 #include <lustre_mds.h>
71 #include <lustre/lustre_idl.h>
72 #include <lustre_disk.h> /* for changelogs */
73 #include <lustre_param.h>
74 #include <lustre_fid.h>
76 #include "mdd_internal.h"
78 const struct md_device_operations mdd_ops;
79 static struct lu_device_type mdd_device_type;
81 static const char mdd_root_dir_name[] = "ROOT";
82 static const char mdd_obf_dir_name[] = "fid";
84 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
85 const char *name, struct lu_device *next)
87 struct mdd_device *mdd = lu2mdd_dev(d);
91 mdd->mdd_child = lu2dt_dev(next);
93 /* Prepare transactions callbacks. */
94 mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
95 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
96 mdd->mdd_txn_cb.dtc_txn_commit = NULL;
97 mdd->mdd_txn_cb.dtc_cookie = mdd;
98 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
99 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
100 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
101 /* sync permission changes */
102 mdd->mdd_sync_permission = 1;
104 rc = mdd_procfs_init(mdd, name);
108 static struct lu_device *mdd_device_fini(const struct lu_env *env,
111 struct mdd_device *mdd = lu2mdd_dev(d);
112 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
115 rc = mdd_procfs_fini(mdd);
117 CERROR("proc fini error %d \n", rc);
123 static void mdd_changelog_fini(const struct lu_env *env,
124 struct mdd_device *mdd);
126 static void mdd_device_shutdown(const struct lu_env *env,
127 struct mdd_device *m, struct lustre_cfg *cfg)
130 mdd_changelog_fini(env, m);
131 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
132 if (m->mdd_dot_lustre_objs.mdd_obf)
133 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
134 if (m->mdd_dot_lustre)
135 mdd_object_put(env, m->mdd_dot_lustre);
137 mdd_fini_obd(env, m, cfg);
138 orph_index_fini(env, m);
139 /* remove upcall device*/
140 md_upcall_fini(&m->mdd_md_dev);
144 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
147 struct mdd_device *mdd = (struct mdd_device *)data;
148 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
151 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
152 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
155 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
156 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
157 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
158 llh->lgh_id.lgl_oid);
160 mdd->mdd_cl.mc_index = rec->cr.cr_index;
161 RETURN(LLOG_PROC_BREAK);
164 static int changelog_user_init_cb(struct llog_handle *llh,
165 struct llog_rec_hdr *hdr, void *data)
167 struct mdd_device *mdd = (struct mdd_device *)data;
168 struct llog_changelog_user_rec *rec =
169 (struct llog_changelog_user_rec *)hdr;
172 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
173 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
175 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
176 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
177 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
179 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
180 mdd->mdd_cl.mc_lastuser = rec->cur_id;
181 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
183 RETURN(LLOG_PROC_BREAK);
187 static int mdd_changelog_llog_init(struct mdd_device *mdd)
189 struct obd_device *obd = mdd2obd_dev(mdd);
190 struct llog_ctxt *ctxt;
193 /* Find last changelog entry number */
194 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
196 CERROR("no changelog context\n");
199 if (!ctxt->loc_handle) {
204 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
208 CERROR("changelog init failed: %d\n", rc);
211 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
212 mdd->mdd_cl.mc_index);
214 /* Find last changelog user id */
215 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
217 CERROR("no changelog user context\n");
220 if (!ctxt->loc_handle) {
225 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
230 CERROR("changelog user init failed: %d\n", rc);
234 /* If we have registered users, assume we want changelogs on */
235 if (mdd->mdd_cl.mc_lastuser > 0)
236 rc = mdd_changelog_on(mdd, 1);
241 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
245 mdd->mdd_cl.mc_index = 0;
246 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
247 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
248 mdd->mdd_cl.mc_flags = 0; /* off by default */
249 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
250 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
251 mdd->mdd_cl.mc_lastuser = 0;
253 rc = mdd_changelog_llog_init(mdd);
255 CERROR("Changelog setup during init failed %d\n", rc);
256 mdd->mdd_cl.mc_flags |= CLM_ERR;
262 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
264 mdd->mdd_cl.mc_flags = 0;
267 /* Start / stop recording */
268 int mdd_changelog_on(struct mdd_device *mdd, int on)
272 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
273 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
274 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
275 CERROR("Changelogs cannot be enabled due to error "
276 "condition (see %s log).\n",
277 mdd2obd_dev(mdd)->obd_name);
280 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
281 mdd->mdd_cl.mc_flags |= CLM_ON;
282 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
283 rc = mdd_changelog_write_header(mdd, CLM_START);
285 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
286 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
287 rc = mdd_changelog_write_header(mdd, CLM_FINI);
288 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
289 mdd->mdd_cl.mc_flags &= ~CLM_ON;
290 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
295 static __u64 cl_time(void) {
298 cfs_fs_time_current(&time);
299 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
302 /** Add a changelog entry \a rec to the changelog llog
305 * \param handle - currently ignored since llogs start their own transaction;
306 * this will hopefully be fixed in llog rewrite
309 int mdd_changelog_llog_write(struct mdd_device *mdd,
310 struct llog_changelog_rec *rec,
311 struct thandle *handle)
313 struct obd_device *obd = mdd2obd_dev(mdd);
314 struct llog_ctxt *ctxt;
317 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
318 /* llog_lvfs_write_rec sets the llog tail len */
319 rec->cr_hdr.lrh_type = CHANGELOG_REC;
320 rec->cr.cr_time = cl_time();
321 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
322 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
323 but as long as the MDD transactions are ordered correctly for e.g.
324 rename conflicts, I don't think this should matter. */
325 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
326 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
327 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
331 /* nested journal transaction */
332 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
338 /** Remove entries with indicies up to and including \a endrec from the
344 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
346 struct obd_device *obd = mdd2obd_dev(mdd);
347 struct llog_ctxt *ctxt;
348 long long unsigned cur;
351 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
355 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
356 cur = (long long)mdd->mdd_cl.mc_index;
357 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
361 /* purge to "0" is shorthand for everything */
365 /* If purging all records, write a header entry so we don't have an
366 empty catalog and we're sure to have a valid starting index next
367 time. In case of crash, we just restart with old log so we're
370 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
375 /* Some records were purged, so reset repeat-access time (so we
376 record new mtime update records, so users can see a file has been
377 changed since the last purge) */
378 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
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 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
413 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
415 /* assume on or off event; reset repeat-access time */
416 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
418 OBD_FREE(rec, reclen);
423 * Create ".lustre" directory.
425 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
427 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
428 struct md_object *mdo;
431 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
432 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
433 mdd_root_dir_name, dot_lustre_name,
434 fid, &dt_directory_features);
435 /* .lustre dir may be already present */
436 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
438 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
439 dot_lustre_name, PFID(fid), rc);
444 lu_object_put(env, &mdo->mo_lu);
450 static int dot_lustre_mdd_permission(const struct lu_env *env,
451 struct md_object *pobj,
452 struct md_object *cobj,
453 struct md_attr *attr, int mask)
455 if (mask & ~(MAY_READ | MAY_EXEC))
461 static int dot_lustre_mdd_attr_get(const struct lu_env *env,
462 struct md_object *obj, struct md_attr *ma)
464 struct mdd_object *mdd_obj = md2mdd_obj(obj);
466 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
469 static int dot_lustre_mdd_attr_set(const struct lu_env *env,
470 struct md_object *obj,
471 const struct md_attr *ma)
476 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
477 struct md_object *obj, struct lu_buf *buf,
483 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
484 struct md_object *obj, struct lu_buf *buf)
489 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
490 struct md_object *obj,
491 const struct lu_buf *buf, const char *name,
497 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
498 struct md_object *obj,
504 static int dot_lustre_mdd_readlink(const struct lu_env *env,
505 struct md_object *obj, struct lu_buf *buf)
510 static int dot_lustre_mdd_object_create(const struct lu_env *env,
511 struct md_object *obj,
512 const struct md_op_spec *spec,
518 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
519 struct md_object *obj,
520 const struct md_attr *ma)
525 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
526 struct md_object *obj,
532 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
535 struct mdd_object *mdd_obj = md2mdd_obj(obj);
537 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
538 mdd_obj->mod_count++;
539 mdd_write_unlock(env, mdd_obj);
544 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
545 struct md_attr *ma, int mode)
547 struct mdd_object *mdd_obj = md2mdd_obj(obj);
549 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
550 mdd_obj->mod_count--;
551 mdd_write_unlock(env, mdd_obj);
556 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
557 struct md_object *obj)
562 static dt_obj_version_t dot_lustre_mdd_version_get(const struct lu_env *env,
563 struct md_object *obj)
568 static void dot_lustre_mdd_version_set(const struct lu_env *env,
569 struct md_object *obj,
570 dt_obj_version_t version)
575 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
576 char *path, int pathlen, __u64 *recno, int *linkno)
581 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
582 struct lov_mds_md *lmm, struct ldlm_extent *extent,
583 struct lustre_handle *lockh)
588 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
589 struct lov_mds_md *lmm, struct lustre_handle *lockh)
594 static struct md_object_operations mdd_dot_lustre_obj_ops = {
595 .moo_permission = dot_lustre_mdd_permission,
596 .moo_attr_get = dot_lustre_mdd_attr_get,
597 .moo_attr_set = dot_lustre_mdd_attr_set,
598 .moo_xattr_get = dot_lustre_mdd_xattr_get,
599 .moo_xattr_list = dot_lustre_mdd_xattr_list,
600 .moo_xattr_set = dot_lustre_mdd_xattr_set,
601 .moo_xattr_del = dot_lustre_mdd_xattr_del,
602 .moo_readpage = mdd_readpage,
603 .moo_readlink = dot_lustre_mdd_readlink,
604 .moo_object_create = dot_lustre_mdd_object_create,
605 .moo_ref_add = dot_lustre_mdd_ref_add,
606 .moo_ref_del = dot_lustre_mdd_ref_del,
607 .moo_open = dot_lustre_mdd_open,
608 .moo_close = dot_lustre_mdd_close,
609 .moo_capa_get = mdd_capa_get,
610 .moo_object_sync = dot_lustre_mdd_object_sync,
611 .moo_version_get = dot_lustre_mdd_version_get,
612 .moo_version_set = dot_lustre_mdd_version_set,
613 .moo_path = dot_lustre_mdd_path,
614 .moo_file_lock = dot_file_lock,
615 .moo_file_unlock = dot_file_unlock,
619 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
620 const struct lu_name *lname, struct lu_fid *f,
621 struct md_op_spec *spec)
623 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
631 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
632 struct md_object *obj,
638 static int dot_lustre_mdd_create(const struct lu_env *env,
639 struct md_object *pobj,
640 const struct lu_name *lname,
641 struct md_object *child,
642 struct md_op_spec *spec,
648 static int dot_lustre_mdd_create_data(const struct lu_env *env,
651 const struct md_op_spec *spec,
657 static int dot_lustre_mdd_rename(const struct lu_env *env,
658 struct md_object *src_pobj,
659 struct md_object *tgt_pobj,
660 const struct lu_fid *lf,
661 const struct lu_name *lsname,
662 struct md_object *tobj,
663 const struct lu_name *ltname,
669 static int dot_lustre_mdd_link(const struct lu_env *env,
670 struct md_object *tgt_obj,
671 struct md_object *src_obj,
672 const struct lu_name *lname,
678 static int dot_lustre_mdd_unlink(const struct lu_env *env,
679 struct md_object *pobj,
680 struct md_object *cobj,
681 const struct lu_name *lname,
687 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
688 struct md_object *obj,
689 const struct lu_name *lname,
690 const struct lu_fid *fid,
691 const struct md_attr *ma)
696 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
697 struct md_object *obj,
698 const struct lu_name *lname,
699 const struct md_attr *ma)
704 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
705 struct md_object *pobj,
706 struct md_object *tobj,
707 const struct lu_fid *fid,
708 const struct lu_name *lname,
715 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
716 .mdo_is_subdir = mdd_is_subdir,
717 .mdo_lookup = dot_lustre_mdd_lookup,
718 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
719 .mdo_create = dot_lustre_mdd_create,
720 .mdo_create_data = dot_lustre_mdd_create_data,
721 .mdo_rename = dot_lustre_mdd_rename,
722 .mdo_link = dot_lustre_mdd_link,
723 .mdo_unlink = dot_lustre_mdd_unlink,
724 .mdo_name_insert = dot_lustre_mdd_name_insert,
725 .mdo_name_remove = dot_lustre_mdd_name_remove,
726 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
729 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
734 if (ma->ma_need & MA_INODE) {
735 struct mdd_device *mdd = mdo2mdd(obj);
737 /* "fid" is a virtual object and hence does not have any "real"
738 * attributes. So we reuse attributes of .lustre for "fid" dir */
739 ma->ma_need |= MA_INODE;
740 rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
744 ma->ma_valid |= MA_INODE;
747 /* "fid" directory does not have any striping information. */
748 if (ma->ma_need & MA_LOV) {
749 struct mdd_object *mdd_obj = md2mdd_obj(obj);
751 if (ma->ma_valid & MA_LOV)
754 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
755 S_ISDIR(mdd_object_type(mdd_obj))))
758 if (ma->ma_need & MA_LOV_DEF) {
759 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
761 ma->ma_lmm_size = rc;
762 ma->ma_valid |= MA_LOV;
771 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
772 const struct md_attr *ma)
777 static int obf_xattr_get(const struct lu_env *env,
778 struct md_object *obj, struct lu_buf *buf,
784 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
787 struct mdd_object *mdd_obj = md2mdd_obj(obj);
789 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
790 mdd_obj->mod_count++;
791 mdd_write_unlock(env, mdd_obj);
796 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
797 struct md_attr *ma, int mode)
799 struct mdd_object *mdd_obj = md2mdd_obj(obj);
801 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
802 mdd_obj->mod_count--;
803 mdd_write_unlock(env, mdd_obj);
808 /** Nothing to list in "fid" directory */
809 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
810 const struct lu_rdpg *rdpg)
815 static int obf_path(const struct lu_env *env, struct md_object *obj,
816 char *path, int pathlen, __u64 *recno, int *linkno)
821 static struct md_object_operations mdd_obf_obj_ops = {
822 .moo_attr_get = obf_attr_get,
823 .moo_attr_set = obf_attr_set,
824 .moo_xattr_get = obf_xattr_get,
825 .moo_open = obf_mdd_open,
826 .moo_close = obf_mdd_close,
827 .moo_readpage = obf_mdd_readpage,
832 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
833 * are valid. We also check if object with passed fid exists or not.
835 static int obf_lookup(const struct lu_env *env, struct md_object *p,
836 const struct lu_name *lname, struct lu_fid *f,
837 struct md_op_spec *spec)
839 char *name = (char *)lname->ln_name;
840 struct mdd_device *mdd = mdo2mdd(p);
841 struct mdd_object *child;
847 sscanf(name, SFID, RFID(f));
848 if (!fid_is_sane(f)) {
849 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
851 GOTO(out, rc = -EINVAL);
854 /* Check if object with this fid exists */
855 child = mdd_object_find(env, mdd, f);
859 GOTO(out, rc = PTR_ERR(child));
861 if (mdd_object_exists(child) == 0)
864 mdd_object_put(env, child);
870 static int obf_create(const struct lu_env *env, struct md_object *pobj,
871 const struct lu_name *lname, struct md_object *child,
872 struct md_op_spec *spec, struct md_attr* ma)
877 static int obf_rename(const struct lu_env *env,
878 struct md_object *src_pobj, struct md_object *tgt_pobj,
879 const struct lu_fid *lf, const struct lu_name *lsname,
880 struct md_object *tobj, const struct lu_name *ltname,
886 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
887 struct md_object *src_obj, const struct lu_name *lname,
893 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
894 struct md_object *cobj, const struct lu_name *lname,
900 static struct md_dir_operations mdd_obf_dir_ops = {
901 .mdo_lookup = obf_lookup,
902 .mdo_create = obf_create,
903 .mdo_rename = obf_rename,
904 .mdo_link = obf_link,
905 .mdo_unlink = obf_unlink
909 * Create special in-memory "fid" object for open-by-fid.
911 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
913 struct mdd_object *mdd_obf;
914 struct lu_object *obf_lu_obj;
917 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
919 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
920 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
921 GOTO(out, rc = -ENOENT);
923 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
924 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
925 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
926 /* Don't allow objects to be created in "fid" dir */
927 mdd_obf->mod_flags |= IMMUTE_OBJ;
929 obf_lu_obj = mdd2lu_obj(mdd_obf);
930 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
936 /** Setup ".lustre" directory object */
937 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
939 struct dt_object *dt_dot_lustre;
940 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
943 rc = create_dot_lustre_dir(env, m);
947 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
948 dot_lustre_name, fid);
949 if (IS_ERR(dt_dot_lustre)) {
950 rc = PTR_ERR(dt_dot_lustre);
954 /* references are released in mdd_device_shutdown() */
955 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
958 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
959 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
961 rc = mdd_obf_setup(env, m);
963 CERROR("Error initializing \"fid\" object - %d.\n", rc);
969 static int mdd_process_config(const struct lu_env *env,
970 struct lu_device *d, struct lustre_cfg *cfg)
972 struct mdd_device *m = lu2mdd_dev(d);
973 struct dt_device *dt = m->mdd_child;
974 struct lu_device *next = &dt->dd_lu_dev;
978 switch (cfg->lcfg_command) {
980 struct lprocfs_static_vars lvars;
982 lprocfs_mdd_init_vars(&lvars);
983 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
984 if (rc > 0 || rc == -ENOSYS)
985 /* we don't understand; pass it on */
986 rc = next->ld_ops->ldo_process_config(env, next, cfg);
990 rc = next->ld_ops->ldo_process_config(env, next, cfg);
993 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
995 rc = mdd_init_obd(env, m, cfg);
997 CERROR("lov init error %d \n", rc);
1000 rc = mdd_txn_init_credits(env, m);
1004 mdd_changelog_init(env, m);
1007 mdd_device_shutdown(env, m, cfg);
1009 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1017 static int mdd_lov_set_nextid(const struct lu_env *env,
1018 struct mdd_device *mdd)
1020 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1024 LASSERT(mds->mds_lov_objids != NULL);
1025 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1026 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1027 mds->mds_lov_objids, NULL);
1032 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1033 struct mdd_device *mdd)
1035 /* XXX: to be implemented! */
1040 static int mdd_recovery_complete(const struct lu_env *env,
1041 struct lu_device *d)
1043 struct mdd_device *mdd = lu2mdd_dev(d);
1044 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1045 struct obd_device *obd = mdd2obd_dev(mdd);
1049 LASSERT(mdd != NULL);
1050 LASSERT(obd != NULL);
1052 /* XXX: Do we need this in new stack? */
1053 rc = mdd_lov_set_nextid(env, mdd);
1055 CERROR("mdd_lov_set_nextid() failed %d\n",
1060 /* XXX: cleanup unlink. */
1061 rc = mdd_cleanup_unlink_llog(env, mdd);
1063 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1068 /* Call that with obd_recovering = 1 just to update objids */
1069 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1070 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1072 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1073 cfs_spin_lock(&obd->obd_dev_lock);
1074 obd->obd_recovering = 0;
1075 cfs_spin_unlock(&obd->obd_dev_lock);
1076 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1078 /* XXX: orphans handling. */
1079 __mdd_orphan_cleanup(env, mdd);
1080 rc = next->ld_ops->ldo_recovery_complete(env, next);
1085 static int mdd_prepare(const struct lu_env *env,
1086 struct lu_device *pdev,
1087 struct lu_device *cdev)
1089 struct mdd_device *mdd = lu2mdd_dev(cdev);
1090 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1091 struct dt_object *root;
1095 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1099 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1100 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1101 &mdd->mdd_root_fid);
1102 if (!IS_ERR(root)) {
1103 LASSERT(root != NULL);
1104 lu_object_put(env, &root->do_lu);
1105 rc = orph_index_init(env, mdd);
1112 rc = mdd_dot_lustre_setup(env, mdd);
1114 CERROR("Error(%d) initializing .lustre objects\n", rc);
1122 const struct lu_device_operations mdd_lu_ops = {
1123 .ldo_object_alloc = mdd_object_alloc,
1124 .ldo_process_config = mdd_process_config,
1125 .ldo_recovery_complete = mdd_recovery_complete,
1126 .ldo_prepare = mdd_prepare,
1130 * No permission check is needed.
1132 static int mdd_root_get(const struct lu_env *env,
1133 struct md_device *m, struct lu_fid *f)
1135 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1138 *f = mdd->mdd_root_fid;
1143 * No permission check is needed.
1145 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1148 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1153 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1159 * No permission check is needed.
1161 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1162 int *md_size, int *cookie_size)
1164 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1167 *md_size = mdd_lov_mdsize(env, mdd);
1168 *cookie_size = mdd_lov_cookiesize(env, mdd);
1173 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1174 int mode, unsigned long timeout, __u32 alg,
1175 struct lustre_capa_key *keys)
1177 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1178 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1182 /* need barrier for mds_capa_keys access. */
1183 cfs_down_write(&mds->mds_notify_lock);
1184 mds->mds_capa_keys = keys;
1185 cfs_up_write(&mds->mds_notify_lock);
1187 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1188 timeout, alg, keys);
1192 static int mdd_update_capa_key(const struct lu_env *env,
1193 struct md_device *m,
1194 struct lustre_capa_key *key)
1196 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1197 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1198 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1202 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1203 sizeof(info), &info, NULL);
1207 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1210 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1212 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1213 return (*h == NULL ? -ENOENT : 0);
1216 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1217 struct lu_device_type *t,
1218 struct lustre_cfg *lcfg)
1220 struct lu_device *l;
1221 struct mdd_device *m;
1225 l = ERR_PTR(-ENOMEM);
1227 md_device_init(&m->mdd_md_dev, t);
1229 l->ld_ops = &mdd_lu_ops;
1230 m->mdd_md_dev.md_ops = &mdd_ops;
1231 md_upcall_init(&m->mdd_md_dev, NULL);
1237 static struct lu_device *mdd_device_free(const struct lu_env *env,
1238 struct lu_device *lu)
1240 struct mdd_device *m = lu2mdd_dev(lu);
1241 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1244 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1245 md_device_fini(&m->mdd_md_dev);
1250 static struct obd_ops mdd_obd_device_ops = {
1251 .o_owner = THIS_MODULE
1254 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1255 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1257 static struct lu_context_key mdd_ucred_key = {
1258 .lct_tags = LCT_SESSION,
1259 .lct_init = mdd_ucred_key_init,
1260 .lct_fini = mdd_ucred_key_fini
1263 struct md_ucred *md_ucred(const struct lu_env *env)
1265 LASSERT(env->le_ses != NULL);
1266 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1268 EXPORT_SYMBOL(md_ucred);
1271 * context key constructor/destructor:
1272 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1274 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1276 struct lu_context_key mdd_capainfo_key = {
1277 .lct_tags = LCT_SESSION,
1278 .lct_init = mdd_capainfo_key_init,
1279 .lct_fini = mdd_capainfo_key_fini
1282 struct md_capainfo *md_capainfo(const struct lu_env *env)
1284 /* NB, in mdt_init0 */
1285 if (env->le_ses == NULL)
1287 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1289 EXPORT_SYMBOL(md_capainfo);
1292 * context key constructor/destructor:
1293 * mdd_quota_key_init, mdd_quota_key_fini
1295 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1297 struct lu_context_key mdd_quota_key = {
1298 .lct_tags = LCT_SESSION,
1299 .lct_init = mdd_quota_key_init,
1300 .lct_fini = mdd_quota_key_fini
1303 struct md_quota *md_quota(const struct lu_env *env)
1305 LASSERT(env->le_ses != NULL);
1306 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1308 EXPORT_SYMBOL(md_quota);
1310 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1312 struct llog_ctxt *ctxt;
1313 struct llog_changelog_user_rec *rec;
1317 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1323 llog_ctxt_put(ctxt);
1327 /* Assume we want it on since somebody registered */
1328 rc = mdd_changelog_on(mdd, 1);
1332 rec->cur_hdr.lrh_len = sizeof(*rec);
1333 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1334 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1335 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1336 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1337 CERROR("Maximum number of changelog users exceeded!\n");
1338 GOTO(out, rc = -EOVERFLOW);
1340 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1341 rec->cur_endrec = mdd->mdd_cl.mc_index;
1342 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1344 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1346 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1349 llog_ctxt_put(ctxt);
1353 struct mdd_changelog_user_data {
1354 __u64 mcud_endrec; /**< purge record for this user */
1355 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1357 __u32 mcud_minid; /**< user id with lowest rec reference */
1358 __u32 mcud_usercount;
1360 struct mdd_device *mcud_mdd;
1361 const struct lu_env *mcud_env;
1363 #define MCUD_UNREGISTER -1LL
1366 * 1. Find the smallest record everyone is willing to purge
1367 * 2. Update the last purgeable record for this user
1369 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1370 struct llog_rec_hdr *hdr, void *data)
1372 struct llog_changelog_user_rec *rec;
1373 struct mdd_changelog_user_data *mcud =
1374 (struct mdd_changelog_user_data *)data;
1378 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1380 rec = (struct llog_changelog_user_rec *)hdr;
1382 mcud->mcud_usercount++;
1384 /* If we have a new endrec for this id, use it for the following
1385 min check instead of its old value */
1386 if (rec->cur_id == mcud->mcud_id)
1387 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1389 /* Track the minimum referenced record */
1390 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1391 mcud->mcud_minid = rec->cur_id;
1392 mcud->mcud_minrec = rec->cur_endrec;
1395 if (rec->cur_id != mcud->mcud_id)
1398 /* Update this user's record */
1399 mcud->mcud_found = 1;
1401 /* Special case: unregister this user */
1402 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1403 struct llog_cookie cookie;
1405 struct mdd_device *mdd = mcud->mcud_mdd;
1407 cookie.lgc_lgl = llh->lgh_id;
1408 cookie.lgc_index = hdr->lrh_index;
1410 /* XXX This is a workaround for the deadlock of changelog
1411 * adding vs. changelog cancelling. LU-81. */
1412 mdd_txn_param_build(mcud->mcud_env, mdd, MDD_TXN_UNLINK_OP, 0);
1413 trans_h = mdd_trans_start(mcud->mcud_env, mdd);
1414 if (IS_ERR(trans_h)) {
1415 CERROR("fsfilt_start_log failed: %ld\n",
1417 RETURN(PTR_ERR(trans_h));
1420 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1423 mcud->mcud_usercount--;
1425 mdd_trans_stop(mcud->mcud_env, mdd, rc, trans_h);
1429 /* Update the endrec */
1430 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1431 mcud->mcud_id, rec->cur_endrec);
1433 /* hdr+1 is loc of data */
1434 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1435 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1441 static int mdd_changelog_user_purge(const struct lu_env *env,
1442 struct mdd_device *mdd, int id,
1445 struct mdd_changelog_user_data data;
1446 struct llog_ctxt *ctxt;
1450 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1453 data.mcud_minid = 0;
1454 data.mcud_minrec = 0;
1455 data.mcud_usercount = 0;
1456 data.mcud_endrec = endrec;
1457 data.mcud_mdd = mdd;
1458 data.mcud_env = env;
1459 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1460 endrec = mdd->mdd_cl.mc_index;
1461 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1462 if ((data.mcud_endrec == 0) ||
1463 ((data.mcud_endrec > endrec) &&
1464 (data.mcud_endrec != MCUD_UNREGISTER)))
1465 data.mcud_endrec = endrec;
1467 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1470 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1472 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1473 (void *)&data, 0, 0);
1474 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1475 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1476 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1477 data.mcud_minrec, data.mcud_minid);
1478 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1480 CWARN("Could not determine changelog records to purge; rc=%d\n",
1484 llog_ctxt_put(ctxt);
1486 if (!data.mcud_found) {
1487 CWARN("No entry for user %d. Last changelog reference is "
1488 LPD64" by changelog user %d\n", data.mcud_id,
1489 data.mcud_minrec, data.mcud_minid);
1493 if (!rc && data.mcud_usercount == 0)
1494 /* No more users; turn changelogs off */
1495 rc = mdd_changelog_on(mdd, 0);
1501 * May be called remotely from mdt_iocontrol_handle or locally from
1502 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1503 * an obd_ioctl_data format (but local ioctl handler does).
1505 * \param len - data len
1506 * \param karg - ioctl data, in kernel space
1508 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1509 unsigned int cmd, int len, void *karg)
1511 struct mdd_device *mdd;
1512 struct obd_ioctl_data *data = karg;
1516 mdd = lu2mdd_dev(&m->md_lu_dev);
1518 /* Doesn't use obd_ioctl_data */
1520 case OBD_IOC_CHANGELOG_CLEAR: {
1521 struct changelog_setinfo *cs = karg;
1522 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1526 case OBD_IOC_GET_MNTOPT: {
1527 mntopt_t *mntopts = (mntopt_t *)karg;
1528 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1533 /* Below ioctls use obd_ioctl_data */
1534 if (len != sizeof(*data)) {
1535 CERROR("Bad ioctl size %d\n", len);
1538 if (data->ioc_version != OBD_IOCTL_VERSION) {
1539 CERROR("Bad magic %x != %x\n", data->ioc_version,
1545 case OBD_IOC_CHANGELOG_REG:
1546 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1548 case OBD_IOC_CHANGELOG_DEREG:
1549 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1559 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1560 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1563 const struct md_device_operations mdd_ops = {
1564 .mdo_statfs = mdd_statfs,
1565 .mdo_root_get = mdd_root_get,
1566 .mdo_maxsize_get = mdd_maxsize_get,
1567 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1568 .mdo_update_capa_key= mdd_update_capa_key,
1569 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1570 .mdo_iocontrol = mdd_iocontrol,
1571 #ifdef HAVE_QUOTA_SUPPORT
1573 .mqo_notify = mdd_quota_notify,
1574 .mqo_setup = mdd_quota_setup,
1575 .mqo_cleanup = mdd_quota_cleanup,
1576 .mqo_recovery = mdd_quota_recovery,
1577 .mqo_check = mdd_quota_check,
1578 .mqo_on = mdd_quota_on,
1579 .mqo_off = mdd_quota_off,
1580 .mqo_setinfo = mdd_quota_setinfo,
1581 .mqo_getinfo = mdd_quota_getinfo,
1582 .mqo_setquota = mdd_quota_setquota,
1583 .mqo_getquota = mdd_quota_getquota,
1584 .mqo_getoinfo = mdd_quota_getoinfo,
1585 .mqo_getoquota = mdd_quota_getoquota,
1586 .mqo_invalidate = mdd_quota_invalidate,
1587 .mqo_finvalidate = mdd_quota_finvalidate
1592 static struct lu_device_type_operations mdd_device_type_ops = {
1593 .ldto_init = mdd_type_init,
1594 .ldto_fini = mdd_type_fini,
1596 .ldto_start = mdd_type_start,
1597 .ldto_stop = mdd_type_stop,
1599 .ldto_device_alloc = mdd_device_alloc,
1600 .ldto_device_free = mdd_device_free,
1602 .ldto_device_init = mdd_device_init,
1603 .ldto_device_fini = mdd_device_fini
1606 static struct lu_device_type mdd_device_type = {
1607 .ldt_tags = LU_DEVICE_MD,
1608 .ldt_name = LUSTRE_MDD_NAME,
1609 .ldt_ops = &mdd_device_type_ops,
1610 .ldt_ctx_tags = LCT_MD_THREAD
1613 /* context key constructor: mdd_key_init */
1614 LU_KEY_INIT(mdd, struct mdd_thread_info);
1616 static void mdd_key_fini(const struct lu_context *ctx,
1617 struct lu_context_key *key, void *data)
1619 struct mdd_thread_info *info = data;
1620 if (info->mti_max_lmm != NULL)
1621 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1622 if (info->mti_max_cookie != NULL)
1623 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1624 mdd_buf_put(&info->mti_big_buf);
1629 /* context key: mdd_thread_key */
1630 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1632 static struct lu_local_obj_desc llod_capa_key = {
1633 .llod_name = CAPA_KEYS,
1634 .llod_oid = MDD_CAPA_KEYS_OID,
1638 static struct lu_local_obj_desc llod_mdd_orphan = {
1639 .llod_name = orph_index_name,
1640 .llod_oid = MDD_ORPHAN_OID,
1642 .llod_feat = &dt_directory_features,
1645 static struct lu_local_obj_desc llod_mdd_root = {
1646 .llod_name = mdd_root_dir_name,
1647 .llod_oid = MDD_ROOT_INDEX_OID,
1649 .llod_feat = &dt_directory_features,
1652 static int __init mdd_mod_init(void)
1654 struct lprocfs_static_vars lvars;
1655 lprocfs_mdd_init_vars(&lvars);
1657 llo_local_obj_register(&llod_capa_key);
1658 llo_local_obj_register(&llod_mdd_orphan);
1659 llo_local_obj_register(&llod_mdd_root);
1661 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1662 LUSTRE_MDD_NAME, &mdd_device_type);
1665 static void __exit mdd_mod_exit(void)
1667 llo_local_obj_unregister(&llod_capa_key);
1668 llo_local_obj_unregister(&llod_mdd_orphan);
1669 llo_local_obj_unregister(&llod_mdd_root);
1671 class_unregister_type(LUSTRE_MDD_NAME);
1674 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1675 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1676 MODULE_LICENSE("GPL");
1678 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);