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 = NULL;
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 if (m->mdd_capa != NULL) {
140 lu_object_put(env, &m->mdd_capa->do_lu);
143 /* remove upcall device*/
144 md_upcall_fini(&m->mdd_md_dev);
148 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
151 struct mdd_device *mdd = (struct mdd_device *)data;
152 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
155 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
156 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
159 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
160 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
161 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
162 llh->lgh_id.lgl_oid);
164 mdd->mdd_cl.mc_index = rec->cr.cr_index;
165 RETURN(LLOG_PROC_BREAK);
168 static int changelog_user_init_cb(struct llog_handle *llh,
169 struct llog_rec_hdr *hdr, void *data)
171 struct mdd_device *mdd = (struct mdd_device *)data;
172 struct llog_changelog_user_rec *rec =
173 (struct llog_changelog_user_rec *)hdr;
176 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
177 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
179 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
180 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
181 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
183 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
184 mdd->mdd_cl.mc_lastuser = rec->cur_id;
185 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
187 RETURN(LLOG_PROC_BREAK);
191 static int mdd_changelog_llog_init(struct mdd_device *mdd)
193 struct obd_device *obd = mdd2obd_dev(mdd);
194 struct llog_ctxt *ctxt;
197 /* Find last changelog entry number */
198 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
200 CERROR("no changelog context\n");
203 if (!ctxt->loc_handle) {
208 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
212 CERROR("changelog init failed: %d\n", rc);
215 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
216 mdd->mdd_cl.mc_index);
218 /* Find last changelog user id */
219 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
221 CERROR("no changelog user context\n");
224 if (!ctxt->loc_handle) {
229 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
234 CERROR("changelog user init failed: %d\n", rc);
238 /* If we have registered users, assume we want changelogs on */
239 if (mdd->mdd_cl.mc_lastuser > 0)
240 rc = mdd_changelog_on(mdd, 1);
245 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
249 mdd->mdd_cl.mc_index = 0;
250 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
251 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
252 mdd->mdd_cl.mc_flags = 0; /* off by default */
253 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
254 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
255 mdd->mdd_cl.mc_lastuser = 0;
257 rc = mdd_changelog_llog_init(mdd);
259 CERROR("Changelog setup during init failed %d\n", rc);
260 mdd->mdd_cl.mc_flags |= CLM_ERR;
266 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
268 mdd->mdd_cl.mc_flags = 0;
271 /* Start / stop recording */
272 int mdd_changelog_on(struct mdd_device *mdd, int on)
276 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
277 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
278 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
279 CERROR("Changelogs cannot be enabled due to error "
280 "condition (see %s log).\n",
281 mdd2obd_dev(mdd)->obd_name);
284 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
285 mdd->mdd_cl.mc_flags |= CLM_ON;
286 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
287 rc = mdd_changelog_write_header(mdd, CLM_START);
289 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
290 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
291 rc = mdd_changelog_write_header(mdd, CLM_FINI);
292 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
293 mdd->mdd_cl.mc_flags &= ~CLM_ON;
294 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
299 static __u64 cl_time(void) {
302 cfs_fs_time_current(&time);
303 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
306 /** Add a changelog entry \a rec to the changelog llog
309 * \param handle - currently ignored since llogs start their own transaction;
310 * this will hopefully be fixed in llog rewrite
313 int mdd_changelog_llog_write(struct mdd_device *mdd,
314 struct llog_changelog_rec *rec,
315 struct thandle *handle)
317 struct obd_device *obd = mdd2obd_dev(mdd);
318 struct llog_ctxt *ctxt;
321 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
322 /* llog_lvfs_write_rec sets the llog tail len */
323 rec->cr_hdr.lrh_type = CHANGELOG_REC;
324 rec->cr.cr_time = cl_time();
325 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
326 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
327 but as long as the MDD transactions are ordered correctly for e.g.
328 rename conflicts, I don't think this should matter. */
329 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
330 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
331 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
335 /* nested journal transaction */
336 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
342 /** Remove entries with indicies up to and including \a endrec from the
348 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
350 struct obd_device *obd = mdd2obd_dev(mdd);
351 struct llog_ctxt *ctxt;
352 long long unsigned cur;
355 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
359 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
360 cur = (long long)mdd->mdd_cl.mc_index;
361 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
365 /* purge to "0" is shorthand for everything */
369 /* If purging all records, write a header entry so we don't have an
370 empty catalog and we're sure to have a valid starting index next
371 time. In case of crash, we just restart with old log so we're
374 /* XXX: transaction is started by llog itself */
375 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
380 /* Some records were purged, so reset repeat-access time (so we
381 record new mtime update records, so users can see a file has been
382 changed since the last purge) */
383 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
385 /* XXX: transaction is started by llog itself */
386 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
392 /** Add a CL_MARK record to the changelog
394 * \param markerflags - CLM_*
397 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
399 struct obd_device *obd = mdd2obd_dev(mdd);
400 struct llog_changelog_rec *rec;
402 int len = strlen(obd->obd_name);
406 reclen = llog_data_len(sizeof(*rec) + len);
407 OBD_ALLOC(rec, reclen);
411 rec->cr.cr_flags = CLF_VERSION;
412 rec->cr.cr_type = CL_MARK;
413 rec->cr.cr_namelen = len;
414 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
415 /* Status and action flags */
416 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
418 /* XXX: transaction is started by llog itself */
419 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
420 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
422 /* assume on or off event; reset repeat-access time */
423 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
425 OBD_FREE(rec, reclen);
430 * Create ".lustre" directory.
432 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
434 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
435 struct md_object *mdo;
438 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
439 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
440 mdd_root_dir_name, dot_lustre_name,
441 fid, &dt_directory_features);
442 /* .lustre dir may be already present */
443 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
445 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
446 dot_lustre_name, PFID(fid), rc);
451 lu_object_put(env, &mdo->mo_lu);
457 static int dot_lustre_mdd_permission(const struct lu_env *env,
458 struct md_object *pobj,
459 struct md_object *cobj,
460 struct md_attr *attr, int mask)
462 if (mask & ~(MAY_READ | MAY_EXEC))
468 static int dot_lustre_mdd_attr_get(const struct lu_env *env,
469 struct md_object *obj, struct md_attr *ma)
471 struct mdd_object *mdd_obj = md2mdd_obj(obj);
473 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
476 static int dot_lustre_mdd_attr_set(const struct lu_env *env,
477 struct md_object *obj,
478 const struct md_attr *ma)
483 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
484 struct md_object *obj, struct lu_buf *buf,
490 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
491 struct md_object *obj, struct lu_buf *buf)
496 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
497 struct md_object *obj,
498 const struct lu_buf *buf, const char *name,
504 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
505 struct md_object *obj,
511 static int dot_lustre_mdd_readlink(const struct lu_env *env,
512 struct md_object *obj, struct lu_buf *buf)
517 static int dot_lustre_mdd_object_create(const struct lu_env *env,
518 struct md_object *obj,
519 const struct md_op_spec *spec,
525 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
526 struct md_object *obj,
527 const struct md_attr *ma)
532 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
533 struct md_object *obj,
539 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
542 struct mdd_object *mdd_obj = md2mdd_obj(obj);
544 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
545 mdd_obj->mod_count++;
546 mdd_write_unlock(env, mdd_obj);
551 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
552 struct md_attr *ma, int mode)
554 struct mdd_object *mdd_obj = md2mdd_obj(obj);
556 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
557 mdd_obj->mod_count--;
558 mdd_write_unlock(env, mdd_obj);
563 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
564 struct md_object *obj)
569 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
570 char *path, int pathlen, __u64 *recno, int *linkno)
575 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
576 struct lov_mds_md *lmm, struct ldlm_extent *extent,
577 struct lustre_handle *lockh)
582 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
583 struct lov_mds_md *lmm, struct lustre_handle *lockh)
588 static struct md_object_operations mdd_dot_lustre_obj_ops = {
589 .moo_permission = dot_lustre_mdd_permission,
590 .moo_attr_get = dot_lustre_mdd_attr_get,
591 .moo_attr_set = dot_lustre_mdd_attr_set,
592 .moo_xattr_get = dot_lustre_mdd_xattr_get,
593 .moo_xattr_list = dot_lustre_mdd_xattr_list,
594 .moo_xattr_set = dot_lustre_mdd_xattr_set,
595 .moo_xattr_del = dot_lustre_mdd_xattr_del,
596 .moo_readpage = mdd_readpage,
597 .moo_readlink = dot_lustre_mdd_readlink,
598 .moo_object_create = dot_lustre_mdd_object_create,
599 .moo_ref_add = dot_lustre_mdd_ref_add,
600 .moo_ref_del = dot_lustre_mdd_ref_del,
601 .moo_open = dot_lustre_mdd_open,
602 .moo_close = dot_lustre_mdd_close,
603 .moo_capa_get = mdd_capa_get,
604 .moo_object_sync = dot_lustre_mdd_object_sync,
605 .moo_path = dot_lustre_mdd_path,
606 .moo_file_lock = dot_file_lock,
607 .moo_file_unlock = dot_file_unlock,
611 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
612 const struct lu_name *lname, struct lu_fid *f,
613 struct md_op_spec *spec)
615 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
623 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
624 struct md_object *obj,
630 static int dot_lustre_mdd_create(const struct lu_env *env,
631 struct md_object *pobj,
632 const struct lu_name *lname,
633 struct md_object *child,
634 struct md_op_spec *spec,
640 static int dot_lustre_mdd_create_data(const struct lu_env *env,
643 const struct md_op_spec *spec,
649 static int dot_lustre_mdd_rename(const struct lu_env *env,
650 struct md_object *src_pobj,
651 struct md_object *tgt_pobj,
652 const struct lu_fid *lf,
653 const struct lu_name *lsname,
654 struct md_object *tobj,
655 const struct lu_name *ltname,
661 static int dot_lustre_mdd_link(const struct lu_env *env,
662 struct md_object *tgt_obj,
663 struct md_object *src_obj,
664 const struct lu_name *lname,
670 static int dot_lustre_mdd_unlink(const struct lu_env *env,
671 struct md_object *pobj,
672 struct md_object *cobj,
673 const struct lu_name *lname,
679 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
680 struct md_object *obj,
681 const struct lu_name *lname,
682 const struct lu_fid *fid,
683 const struct md_attr *ma)
688 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
689 struct md_object *obj,
690 const struct lu_name *lname,
691 const struct md_attr *ma)
696 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
697 struct md_object *pobj,
698 struct md_object *tobj,
699 const struct lu_fid *fid,
700 const struct lu_name *lname,
707 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
708 .mdo_is_subdir = mdd_is_subdir,
709 .mdo_lookup = dot_lustre_mdd_lookup,
710 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
711 .mdo_create = dot_lustre_mdd_create,
712 .mdo_create_data = dot_lustre_mdd_create_data,
713 .mdo_rename = dot_lustre_mdd_rename,
714 .mdo_link = dot_lustre_mdd_link,
715 .mdo_unlink = dot_lustre_mdd_unlink,
716 .mdo_name_insert = dot_lustre_mdd_name_insert,
717 .mdo_name_remove = dot_lustre_mdd_name_remove,
718 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
721 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
726 if (ma->ma_need & MA_INODE) {
727 struct mdd_device *mdd = mdo2mdd(obj);
729 /* "fid" is a virtual object and hence does not have any "real"
730 * attributes. So we reuse attributes of .lustre for "fid" dir */
731 ma->ma_need |= MA_INODE;
732 rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
736 ma->ma_valid |= MA_INODE;
739 /* "fid" directory does not have any striping information. */
740 if (ma->ma_need & MA_LOV) {
741 struct mdd_object *mdd_obj = md2mdd_obj(obj);
743 if (ma->ma_valid & MA_LOV)
746 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
747 S_ISDIR(mdd_object_type(mdd_obj))))
750 if (ma->ma_need & MA_LOV_DEF) {
751 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
753 ma->ma_lmm_size = rc;
754 ma->ma_valid |= MA_LOV;
763 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
764 const struct md_attr *ma)
769 static int obf_xattr_get(const struct lu_env *env,
770 struct md_object *obj, struct lu_buf *buf,
776 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
779 struct mdd_object *mdd_obj = md2mdd_obj(obj);
781 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
782 mdd_obj->mod_count++;
783 mdd_write_unlock(env, mdd_obj);
788 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
789 struct md_attr *ma, int mode)
791 struct mdd_object *mdd_obj = md2mdd_obj(obj);
793 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
794 mdd_obj->mod_count--;
795 mdd_write_unlock(env, mdd_obj);
800 /** Nothing to list in "fid" directory */
801 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
802 const struct lu_rdpg *rdpg)
807 static int obf_path(const struct lu_env *env, struct md_object *obj,
808 char *path, int pathlen, __u64 *recno, int *linkno)
813 static struct md_object_operations mdd_obf_obj_ops = {
814 .moo_attr_get = obf_attr_get,
815 .moo_attr_set = obf_attr_set,
816 .moo_xattr_get = obf_xattr_get,
817 .moo_open = obf_mdd_open,
818 .moo_close = obf_mdd_close,
819 .moo_readpage = obf_mdd_readpage,
824 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
825 * are valid. We also check if object with passed fid exists or not.
827 static int obf_lookup(const struct lu_env *env, struct md_object *p,
828 const struct lu_name *lname, struct lu_fid *f,
829 struct md_op_spec *spec)
831 char *name = (char *)lname->ln_name;
832 struct mdd_device *mdd = mdo2mdd(p);
833 struct mdd_object *child;
839 sscanf(name, SFID, RFID(f));
840 if (!fid_is_sane(f)) {
841 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
843 GOTO(out, rc = -EINVAL);
846 /* Check if object with this fid exists */
847 child = mdd_object_find(env, mdd, f);
851 GOTO(out, rc = PTR_ERR(child));
853 if (mdd_object_exists(child) == 0)
856 mdd_object_put(env, child);
862 static int obf_create(const struct lu_env *env, struct md_object *pobj,
863 const struct lu_name *lname, struct md_object *child,
864 struct md_op_spec *spec, struct md_attr* ma)
869 static int obf_rename(const struct lu_env *env,
870 struct md_object *src_pobj, struct md_object *tgt_pobj,
871 const struct lu_fid *lf, const struct lu_name *lsname,
872 struct md_object *tobj, const struct lu_name *ltname,
878 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
879 struct md_object *src_obj, const struct lu_name *lname,
885 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
886 struct md_object *cobj, const struct lu_name *lname,
892 static struct md_dir_operations mdd_obf_dir_ops = {
893 .mdo_lookup = obf_lookup,
894 .mdo_create = obf_create,
895 .mdo_rename = obf_rename,
896 .mdo_link = obf_link,
897 .mdo_unlink = obf_unlink
901 * Create special in-memory "fid" object for open-by-fid.
903 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
905 struct mdd_object *mdd_obf;
906 struct lu_object *obf_lu_obj;
909 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
911 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
912 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
913 GOTO(out, rc = -ENOENT);
915 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
916 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
917 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
918 /* Don't allow objects to be created in "fid" dir */
919 mdd_obf->mod_flags |= IMMUTE_OBJ;
921 obf_lu_obj = mdd2lu_obj(mdd_obf);
922 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
928 /** Setup ".lustre" directory object */
929 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
931 struct dt_object *dt_dot_lustre;
932 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
935 rc = create_dot_lustre_dir(env, m);
939 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
940 dot_lustre_name, fid);
941 if (IS_ERR(dt_dot_lustre)) {
942 rc = PTR_ERR(dt_dot_lustre);
946 /* references are released in mdd_device_shutdown() */
947 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
950 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
951 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
953 rc = mdd_obf_setup(env, m);
955 CERROR("Error initializing \"fid\" object - %d.\n", rc);
961 static int mdd_process_config(const struct lu_env *env,
962 struct lu_device *d, struct lustre_cfg *cfg)
964 struct mdd_device *m = lu2mdd_dev(d);
965 struct dt_device *dt = m->mdd_child;
966 struct lu_device *next = &dt->dd_lu_dev;
970 switch (cfg->lcfg_command) {
972 struct lprocfs_static_vars lvars;
974 lprocfs_mdd_init_vars(&lvars);
975 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
976 if (rc > 0 || rc == -ENOSYS)
977 /* we don't understand; pass it on */
978 rc = next->ld_ops->ldo_process_config(env, next, cfg);
982 rc = next->ld_ops->ldo_process_config(env, next, cfg);
985 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
987 rc = mdd_init_obd(env, m, cfg);
989 CERROR("lov init error %d\n", rc);
993 mdd_changelog_init(env, m);
996 mdd_device_shutdown(env, m, cfg);
998 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1006 static int mdd_lov_set_nextid(const struct lu_env *env,
1007 struct mdd_device *mdd)
1009 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1013 LASSERT(mds->mds_lov_objids != NULL);
1014 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1015 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1016 mds->mds_lov_objids, NULL);
1021 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1022 struct mdd_device *mdd)
1024 /* XXX: to be implemented! */
1029 static int mdd_recovery_complete(const struct lu_env *env,
1030 struct lu_device *d)
1032 struct mdd_device *mdd = lu2mdd_dev(d);
1033 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1034 struct obd_device *obd = mdd2obd_dev(mdd);
1038 LASSERT(mdd != NULL);
1039 LASSERT(obd != NULL);
1041 /* XXX: Do we need this in new stack? */
1042 rc = mdd_lov_set_nextid(env, mdd);
1044 CERROR("mdd_lov_set_nextid() failed %d\n",
1049 /* XXX: cleanup unlink. */
1050 rc = mdd_cleanup_unlink_llog(env, mdd);
1052 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1057 /* Call that with obd_recovering = 1 just to update objids */
1058 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1059 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1061 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1062 cfs_spin_lock(&obd->obd_dev_lock);
1063 obd->obd_recovering = 0;
1064 cfs_spin_unlock(&obd->obd_dev_lock);
1065 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1067 /* XXX: orphans handling. */
1068 __mdd_orphan_cleanup(env, mdd);
1069 rc = next->ld_ops->ldo_recovery_complete(env, next);
1074 static int mdd_prepare(const struct lu_env *env,
1075 struct lu_device *pdev,
1076 struct lu_device *cdev)
1078 struct mdd_device *mdd = lu2mdd_dev(cdev);
1079 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1080 struct dt_object *root;
1085 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1089 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1090 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1091 &mdd->mdd_root_fid);
1092 if (!IS_ERR(root)) {
1093 LASSERT(root != NULL);
1094 lu_object_put(env, &root->do_lu);
1095 rc = orph_index_init(env, mdd);
1102 rc = mdd_dot_lustre_setup(env, mdd);
1104 CERROR("Error(%d) initializing .lustre objects\n", rc);
1108 /* we use capa file to declare llog changes,
1109 * will be fixed with new llog in 2.3 */
1110 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1112 mdd->mdd_capa = root;
1120 const struct lu_device_operations mdd_lu_ops = {
1121 .ldo_object_alloc = mdd_object_alloc,
1122 .ldo_process_config = mdd_process_config,
1123 .ldo_recovery_complete = mdd_recovery_complete,
1124 .ldo_prepare = mdd_prepare,
1128 * No permission check is needed.
1130 static int mdd_root_get(const struct lu_env *env,
1131 struct md_device *m, struct lu_fid *f)
1133 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1136 *f = mdd->mdd_root_fid;
1141 * No permission check is needed.
1143 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1146 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1151 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1157 * No permission check is needed.
1159 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1160 int *md_size, int *cookie_size)
1162 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1165 *md_size = mdd_lov_mdsize(env, mdd);
1166 *cookie_size = mdd_lov_cookiesize(env, mdd);
1171 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1172 int mode, unsigned long timeout, __u32 alg,
1173 struct lustre_capa_key *keys)
1175 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1176 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1180 /* need barrier for mds_capa_keys access. */
1181 cfs_down_write(&mds->mds_notify_lock);
1182 mds->mds_capa_keys = keys;
1183 cfs_up_write(&mds->mds_notify_lock);
1185 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1186 timeout, alg, keys);
1190 static int mdd_update_capa_key(const struct lu_env *env,
1191 struct md_device *m,
1192 struct lustre_capa_key *key)
1194 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1195 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1196 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1200 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1201 sizeof(info), &info, NULL);
1205 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1208 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1210 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1211 return (*h == NULL ? -ENOENT : 0);
1214 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1215 struct lu_device_type *t,
1216 struct lustre_cfg *lcfg)
1218 struct lu_device *l;
1219 struct mdd_device *m;
1223 l = ERR_PTR(-ENOMEM);
1225 md_device_init(&m->mdd_md_dev, t);
1227 l->ld_ops = &mdd_lu_ops;
1228 m->mdd_md_dev.md_ops = &mdd_ops;
1229 md_upcall_init(&m->mdd_md_dev, NULL);
1235 static struct lu_device *mdd_device_free(const struct lu_env *env,
1236 struct lu_device *lu)
1238 struct mdd_device *m = lu2mdd_dev(lu);
1239 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1242 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1243 md_device_fini(&m->mdd_md_dev);
1248 static struct obd_ops mdd_obd_device_ops = {
1249 .o_owner = THIS_MODULE
1252 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1253 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1255 static struct lu_context_key mdd_ucred_key = {
1256 .lct_tags = LCT_SESSION,
1257 .lct_init = mdd_ucred_key_init,
1258 .lct_fini = mdd_ucred_key_fini
1261 struct md_ucred *md_ucred(const struct lu_env *env)
1263 LASSERT(env->le_ses != NULL);
1264 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1266 EXPORT_SYMBOL(md_ucred);
1269 * context key constructor/destructor:
1270 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1272 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1274 struct lu_context_key mdd_capainfo_key = {
1275 .lct_tags = LCT_SESSION,
1276 .lct_init = mdd_capainfo_key_init,
1277 .lct_fini = mdd_capainfo_key_fini
1280 struct md_capainfo *md_capainfo(const struct lu_env *env)
1282 /* NB, in mdt_init0 */
1283 if (env->le_ses == NULL)
1285 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1287 EXPORT_SYMBOL(md_capainfo);
1290 * context key constructor/destructor:
1291 * mdd_quota_key_init, mdd_quota_key_fini
1293 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1295 struct lu_context_key mdd_quota_key = {
1296 .lct_tags = LCT_SESSION,
1297 .lct_init = mdd_quota_key_init,
1298 .lct_fini = mdd_quota_key_fini
1301 struct md_quota *md_quota(const struct lu_env *env)
1303 LASSERT(env->le_ses != NULL);
1304 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1306 EXPORT_SYMBOL(md_quota);
1308 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1310 struct llog_ctxt *ctxt;
1311 struct llog_changelog_user_rec *rec;
1315 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1321 llog_ctxt_put(ctxt);
1325 /* Assume we want it on since somebody registered */
1326 rc = mdd_changelog_on(mdd, 1);
1330 rec->cur_hdr.lrh_len = sizeof(*rec);
1331 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1332 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1333 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1334 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1335 CERROR("Maximum number of changelog users exceeded!\n");
1336 GOTO(out, rc = -EOVERFLOW);
1338 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1339 rec->cur_endrec = mdd->mdd_cl.mc_index;
1340 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1342 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1344 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1347 llog_ctxt_put(ctxt);
1351 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1352 struct thandle *handle)
1357 /* XXX: this is a temporary solution to declare llog changes
1358 * will be fixed in 2.3 with new llog implementation */
1360 LASSERT(mdd->mdd_capa);
1362 /* the llog record could be canceled either by modifying
1363 * the plain llog's header or by destroying the llog itself
1364 * when this record is the last one in it, it can't be known
1365 * here, but the catlog's header will also be modified for
1366 * the second case, then the first case can be covered and
1367 * is no need to declare it */
1369 /* destroy empty plain log */
1370 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1374 /* record the catlog's header if an empty plain log was destroyed */
1375 rc = dt_declare_record_write(env, mdd->mdd_capa,
1376 sizeof(struct llog_logid_rec), 0, handle);
1380 struct mdd_changelog_user_data {
1381 __u64 mcud_endrec; /**< purge record for this user */
1382 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1384 __u32 mcud_minid; /**< user id with lowest rec reference */
1385 __u32 mcud_usercount;
1387 struct mdd_device *mcud_mdd;
1388 const struct lu_env *mcud_env;
1390 #define MCUD_UNREGISTER -1LL
1393 * 1. Find the smallest record everyone is willing to purge
1394 * 2. Update the last purgeable record for this user
1396 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1397 struct llog_rec_hdr *hdr, void *data)
1399 struct llog_changelog_user_rec *rec;
1400 struct mdd_changelog_user_data *mcud =
1401 (struct mdd_changelog_user_data *)data;
1405 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1407 rec = (struct llog_changelog_user_rec *)hdr;
1409 mcud->mcud_usercount++;
1411 /* If we have a new endrec for this id, use it for the following
1412 min check instead of its old value */
1413 if (rec->cur_id == mcud->mcud_id)
1414 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1416 /* Track the minimum referenced record */
1417 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1418 mcud->mcud_minid = rec->cur_id;
1419 mcud->mcud_minrec = rec->cur_endrec;
1422 if (rec->cur_id != mcud->mcud_id)
1425 /* Update this user's record */
1426 mcud->mcud_found = 1;
1428 /* Special case: unregister this user */
1429 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1430 struct llog_cookie cookie;
1432 struct mdd_device *mdd = mcud->mcud_mdd;
1434 cookie.lgc_lgl = llh->lgh_id;
1435 cookie.lgc_index = hdr->lrh_index;
1437 /* XXX This is a workaround for the deadlock of changelog
1438 * adding vs. changelog cancelling. LU-81. */
1439 th = mdd_trans_create(mcud->mcud_env, mdd);
1441 CERROR("Cannot get thandle\n");
1445 rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
1449 rc = mdd_trans_start(mcud->mcud_env, mdd, th);
1453 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1456 mcud->mcud_usercount--;
1459 mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
1463 /* Update the endrec */
1464 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1465 mcud->mcud_id, rec->cur_endrec);
1467 /* hdr+1 is loc of data */
1468 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1469 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1475 static int mdd_changelog_user_purge(const struct lu_env *env,
1476 struct mdd_device *mdd, int id,
1479 struct mdd_changelog_user_data data;
1480 struct llog_ctxt *ctxt;
1484 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1487 data.mcud_minid = 0;
1488 data.mcud_minrec = 0;
1489 data.mcud_usercount = 0;
1490 data.mcud_endrec = endrec;
1491 data.mcud_mdd = mdd;
1492 data.mcud_env = env;
1493 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1494 endrec = mdd->mdd_cl.mc_index;
1495 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1496 if ((data.mcud_endrec == 0) ||
1497 ((data.mcud_endrec > endrec) &&
1498 (data.mcud_endrec != MCUD_UNREGISTER)))
1499 data.mcud_endrec = endrec;
1501 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1504 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1506 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1507 (void *)&data, 0, 0);
1508 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1509 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1510 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1511 data.mcud_minrec, data.mcud_minid);
1512 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1514 CWARN("Could not determine changelog records to purge; rc=%d\n",
1518 llog_ctxt_put(ctxt);
1520 if (!data.mcud_found) {
1521 CWARN("No entry for user %d. Last changelog reference is "
1522 LPD64" by changelog user %d\n", data.mcud_id,
1523 data.mcud_minrec, data.mcud_minid);
1527 if (!rc && data.mcud_usercount == 0)
1528 /* No more users; turn changelogs off */
1529 rc = mdd_changelog_on(mdd, 0);
1535 * May be called remotely from mdt_iocontrol_handle or locally from
1536 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1537 * an obd_ioctl_data format (but local ioctl handler does).
1539 * \param len - data len
1540 * \param karg - ioctl data, in kernel space
1542 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1543 unsigned int cmd, int len, void *karg)
1545 struct mdd_device *mdd;
1546 struct obd_ioctl_data *data = karg;
1550 mdd = lu2mdd_dev(&m->md_lu_dev);
1552 /* Doesn't use obd_ioctl_data */
1554 case OBD_IOC_CHANGELOG_CLEAR: {
1555 struct changelog_setinfo *cs = karg;
1556 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1560 case OBD_IOC_GET_MNTOPT: {
1561 mntopt_t *mntopts = (mntopt_t *)karg;
1562 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1567 /* Below ioctls use obd_ioctl_data */
1568 if (len != sizeof(*data)) {
1569 CERROR("Bad ioctl size %d\n", len);
1572 if (data->ioc_version != OBD_IOCTL_VERSION) {
1573 CERROR("Bad magic %x != %x\n", data->ioc_version,
1579 case OBD_IOC_CHANGELOG_REG:
1580 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1582 case OBD_IOC_CHANGELOG_DEREG:
1583 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1593 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1594 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1597 const struct md_device_operations mdd_ops = {
1598 .mdo_statfs = mdd_statfs,
1599 .mdo_root_get = mdd_root_get,
1600 .mdo_maxsize_get = mdd_maxsize_get,
1601 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1602 .mdo_update_capa_key= mdd_update_capa_key,
1603 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1604 .mdo_iocontrol = mdd_iocontrol,
1605 #ifdef HAVE_QUOTA_SUPPORT
1607 .mqo_notify = mdd_quota_notify,
1608 .mqo_setup = mdd_quota_setup,
1609 .mqo_cleanup = mdd_quota_cleanup,
1610 .mqo_recovery = mdd_quota_recovery,
1611 .mqo_check = mdd_quota_check,
1612 .mqo_on = mdd_quota_on,
1613 .mqo_off = mdd_quota_off,
1614 .mqo_setinfo = mdd_quota_setinfo,
1615 .mqo_getinfo = mdd_quota_getinfo,
1616 .mqo_setquota = mdd_quota_setquota,
1617 .mqo_getquota = mdd_quota_getquota,
1618 .mqo_getoinfo = mdd_quota_getoinfo,
1619 .mqo_getoquota = mdd_quota_getoquota,
1620 .mqo_invalidate = mdd_quota_invalidate,
1621 .mqo_finvalidate = mdd_quota_finvalidate
1626 static struct lu_device_type_operations mdd_device_type_ops = {
1627 .ldto_init = mdd_type_init,
1628 .ldto_fini = mdd_type_fini,
1630 .ldto_start = mdd_type_start,
1631 .ldto_stop = mdd_type_stop,
1633 .ldto_device_alloc = mdd_device_alloc,
1634 .ldto_device_free = mdd_device_free,
1636 .ldto_device_init = mdd_device_init,
1637 .ldto_device_fini = mdd_device_fini
1640 static struct lu_device_type mdd_device_type = {
1641 .ldt_tags = LU_DEVICE_MD,
1642 .ldt_name = LUSTRE_MDD_NAME,
1643 .ldt_ops = &mdd_device_type_ops,
1644 .ldt_ctx_tags = LCT_MD_THREAD
1647 /* context key constructor: mdd_key_init */
1648 LU_KEY_INIT(mdd, struct mdd_thread_info);
1650 static void mdd_key_fini(const struct lu_context *ctx,
1651 struct lu_context_key *key, void *data)
1653 struct mdd_thread_info *info = data;
1654 if (info->mti_max_lmm != NULL)
1655 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1656 if (info->mti_max_cookie != NULL)
1657 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1658 mdd_buf_put(&info->mti_big_buf);
1663 /* context key: mdd_thread_key */
1664 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1666 static struct lu_local_obj_desc llod_capa_key = {
1667 .llod_name = CAPA_KEYS,
1668 .llod_oid = MDD_CAPA_KEYS_OID,
1672 static struct lu_local_obj_desc llod_mdd_orphan = {
1673 .llod_name = orph_index_name,
1674 .llod_oid = MDD_ORPHAN_OID,
1676 .llod_feat = &dt_directory_features,
1679 static struct lu_local_obj_desc llod_mdd_root = {
1680 .llod_name = mdd_root_dir_name,
1681 .llod_oid = MDD_ROOT_INDEX_OID,
1683 .llod_feat = &dt_directory_features,
1686 static int __init mdd_mod_init(void)
1688 struct lprocfs_static_vars lvars;
1689 lprocfs_mdd_init_vars(&lvars);
1691 llo_local_obj_register(&llod_capa_key);
1692 llo_local_obj_register(&llod_mdd_orphan);
1693 llo_local_obj_register(&llod_mdd_root);
1695 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1696 LUSTRE_MDD_NAME, &mdd_device_type);
1699 static void __exit mdd_mod_exit(void)
1701 llo_local_obj_unregister(&llod_capa_key);
1702 llo_local_obj_unregister(&llod_mdd_orphan);
1703 llo_local_obj_unregister(&llod_mdd_root);
1705 class_unregister_type(LUSTRE_MDD_NAME);
1708 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1709 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1710 MODULE_LICENSE("GPL");
1712 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);