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 #include <ldiskfs/ldiskfs_jbd2.h>
48 #include <obd_class.h>
49 #include <lustre_ver.h>
50 #include <obd_support.h>
51 #include <lprocfs_status.h>
53 #include <lustre_fid.h>
54 #include <ldiskfs/ldiskfs.h>
55 #include <lustre_mds.h>
56 #include <lustre/lustre_idl.h>
57 #include <lustre_disk.h> /* for changelogs */
58 #include <lustre_param.h>
59 #include <lustre_fid.h>
61 #include "mdd_internal.h"
63 const struct md_device_operations mdd_ops;
64 static struct lu_device_type mdd_device_type;
66 static const char mdd_root_dir_name[] = "ROOT";
67 static const char mdd_obf_dir_name[] = "fid";
69 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
70 const char *name, struct lu_device *next)
72 struct mdd_device *mdd = lu2mdd_dev(d);
76 mdd->mdd_child = lu2dt_dev(next);
78 /* Prepare transactions callbacks. */
79 mdd->mdd_txn_cb.dtc_txn_start = NULL;
80 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
81 mdd->mdd_txn_cb.dtc_txn_commit = NULL;
82 mdd->mdd_txn_cb.dtc_cookie = mdd;
83 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
84 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
85 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
86 /* sync permission changes */
87 mdd->mdd_sync_permission = 1;
89 rc = mdd_procfs_init(mdd, name);
93 static struct lu_device *mdd_device_fini(const struct lu_env *env,
96 struct mdd_device *mdd = lu2mdd_dev(d);
97 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
100 rc = mdd_procfs_fini(mdd);
102 CERROR("proc fini error %d \n", rc);
108 static void mdd_changelog_fini(const struct lu_env *env,
109 struct mdd_device *mdd);
111 static void mdd_device_shutdown(const struct lu_env *env,
112 struct mdd_device *m, struct lustre_cfg *cfg)
115 mdd_lfsck_cleanup(env, m);
116 mdd_changelog_fini(env, m);
117 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
118 if (m->mdd_dot_lustre_objs.mdd_obf)
119 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
120 if (m->mdd_dot_lustre)
121 mdd_object_put(env, m->mdd_dot_lustre);
123 mdd_fini_obd(env, m, cfg);
124 orph_index_fini(env, m);
125 if (m->mdd_capa != NULL) {
126 lu_object_put(env, &m->mdd_capa->do_lu);
129 /* remove upcall device*/
130 md_upcall_fini(&m->mdd_md_dev);
134 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
137 struct mdd_device *mdd = (struct mdd_device *)data;
138 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
141 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
142 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
145 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
146 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
147 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
148 llh->lgh_id.lgl_oid);
150 mdd->mdd_cl.mc_index = rec->cr.cr_index;
151 RETURN(LLOG_PROC_BREAK);
154 static int changelog_user_init_cb(struct llog_handle *llh,
155 struct llog_rec_hdr *hdr, void *data)
157 struct mdd_device *mdd = (struct mdd_device *)data;
158 struct llog_changelog_user_rec *rec =
159 (struct llog_changelog_user_rec *)hdr;
162 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
163 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
165 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
166 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
167 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
169 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
170 mdd->mdd_cl.mc_lastuser = rec->cur_id;
171 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
173 RETURN(LLOG_PROC_BREAK);
177 static int mdd_changelog_llog_init(struct mdd_device *mdd)
179 struct obd_device *obd = mdd2obd_dev(mdd);
180 struct llog_ctxt *ctxt;
183 /* Find last changelog entry number */
184 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
186 CERROR("no changelog context\n");
189 if (!ctxt->loc_handle) {
194 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
198 CERROR("changelog init failed: %d\n", rc);
201 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
202 mdd->mdd_cl.mc_index);
204 /* Find last changelog user id */
205 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
207 CERROR("no changelog user context\n");
210 if (!ctxt->loc_handle) {
215 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
220 CERROR("changelog user init failed: %d\n", rc);
224 /* If we have registered users, assume we want changelogs on */
225 if (mdd->mdd_cl.mc_lastuser > 0)
226 rc = mdd_changelog_on(mdd, 1);
231 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
235 mdd->mdd_cl.mc_index = 0;
236 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
237 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
238 mdd->mdd_cl.mc_flags = 0; /* off by default */
239 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
240 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
241 mdd->mdd_cl.mc_lastuser = 0;
243 rc = mdd_changelog_llog_init(mdd);
245 CERROR("Changelog setup during init failed %d\n", rc);
246 mdd->mdd_cl.mc_flags |= CLM_ERR;
252 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
254 mdd->mdd_cl.mc_flags = 0;
257 /* Start / stop recording */
258 int mdd_changelog_on(struct mdd_device *mdd, int on)
262 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
263 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
264 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
265 CERROR("Changelogs cannot be enabled due to error "
266 "condition (see %s log).\n",
267 mdd2obd_dev(mdd)->obd_name);
270 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
271 mdd->mdd_cl.mc_flags |= CLM_ON;
272 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
273 rc = mdd_changelog_write_header(mdd, CLM_START);
275 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
276 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
277 rc = mdd_changelog_write_header(mdd, CLM_FINI);
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);
285 static __u64 cl_time(void) {
288 cfs_fs_time_current(&time);
289 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
292 /** Add a changelog entry \a rec to the changelog llog
295 * \param handle - currently ignored since llogs start their own transaction;
296 * this will hopefully be fixed in llog rewrite
299 int mdd_changelog_llog_write(struct mdd_device *mdd,
300 struct llog_changelog_rec *rec,
301 struct thandle *handle)
303 struct obd_device *obd = mdd2obd_dev(mdd);
304 struct llog_ctxt *ctxt;
307 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
308 /* llog_lvfs_write_rec sets the llog tail len */
309 rec->cr_hdr.lrh_type = CHANGELOG_REC;
310 rec->cr.cr_time = cl_time();
311 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
312 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
313 but as long as the MDD transactions are ordered correctly for e.g.
314 rename conflicts, I don't think this should matter. */
315 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
316 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
317 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
321 /* nested journal transaction */
322 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
328 /** Add a changelog_ext entry \a rec to the changelog llog
331 * \param handle - currently ignored since llogs start their own transaction;
332 * this will hopefully be fixed in llog rewrite
335 int mdd_changelog_ext_llog_write(struct mdd_device *mdd,
336 struct llog_changelog_ext_rec *rec,
337 struct thandle *handle)
339 struct obd_device *obd = mdd2obd_dev(mdd);
340 struct llog_ctxt *ctxt;
343 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
344 /* llog_lvfs_write_rec sets the llog tail len */
345 rec->cr_hdr.lrh_type = CHANGELOG_REC;
346 rec->cr.cr_time = cl_time();
347 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
348 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
349 * but as long as the MDD transactions are ordered correctly for e.g.
350 * rename conflicts, I don't think this should matter. */
351 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
352 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
353 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
357 /* nested journal transaction */
358 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
364 /** Remove entries with indicies up to and including \a endrec from the
370 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
372 struct obd_device *obd = mdd2obd_dev(mdd);
373 struct llog_ctxt *ctxt;
374 long long unsigned cur;
377 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
381 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
382 cur = (long long)mdd->mdd_cl.mc_index;
383 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
387 /* purge to "0" is shorthand for everything */
391 /* If purging all records, write a header entry so we don't have an
392 empty catalog and we're sure to have a valid starting index next
393 time. In case of crash, we just restart with old log so we're
396 /* XXX: transaction is started by llog itself */
397 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
402 /* Some records were purged, so reset repeat-access time (so we
403 record new mtime update records, so users can see a file has been
404 changed since the last purge) */
405 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
407 /* XXX: transaction is started by llog itself */
408 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
414 /** Add a CL_MARK record to the changelog
416 * \param markerflags - CLM_*
419 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
421 struct obd_device *obd = mdd2obd_dev(mdd);
422 struct llog_changelog_rec *rec;
424 int len = strlen(obd->obd_name);
428 reclen = llog_data_len(sizeof(*rec) + len);
429 OBD_ALLOC(rec, reclen);
433 rec->cr.cr_flags = CLF_VERSION;
434 rec->cr.cr_type = CL_MARK;
435 rec->cr.cr_namelen = len;
436 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
437 /* Status and action flags */
438 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
440 /* XXX: transaction is started by llog itself */
441 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
442 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
444 /* assume on or off event; reset repeat-access time */
445 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
447 OBD_FREE(rec, reclen);
452 * Create ".lustre" directory.
454 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
456 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
457 struct md_object *mdo;
460 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
461 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
462 mdd_root_dir_name, dot_lustre_name,
463 fid, &dt_directory_features);
464 /* .lustre dir may be already present */
465 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
467 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
468 dot_lustre_name, PFID(fid), rc);
473 lu_object_put(env, &mdo->mo_lu);
479 static int dot_lustre_mdd_permission(const struct lu_env *env,
480 struct md_object *pobj,
481 struct md_object *cobj,
482 struct md_attr *attr, int mask)
484 if (mask & ~(MAY_READ | MAY_EXEC))
490 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
491 struct md_object *obj, struct lu_buf *buf,
497 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
498 struct md_object *obj, struct lu_buf *buf)
503 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
504 struct md_object *obj,
505 const struct lu_buf *buf, const char *name,
511 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
512 struct md_object *obj,
518 static int dot_lustre_mdd_readlink(const struct lu_env *env,
519 struct md_object *obj, struct lu_buf *buf)
524 static int dot_lustre_mdd_object_create(const struct lu_env *env,
525 struct md_object *obj,
526 const struct md_op_spec *spec,
532 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
533 struct md_object *obj,
534 const struct md_attr *ma)
539 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
540 struct md_object *obj,
546 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
549 struct mdd_object *mdd_obj = md2mdd_obj(obj);
551 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
552 mdd_obj->mod_count++;
553 mdd_write_unlock(env, mdd_obj);
558 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
559 struct md_attr *ma, int mode)
561 struct mdd_object *mdd_obj = md2mdd_obj(obj);
563 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
564 mdd_obj->mod_count--;
565 mdd_write_unlock(env, mdd_obj);
570 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
571 struct md_object *obj)
576 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
577 char *path, int pathlen, __u64 *recno, int *linkno)
582 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
583 struct lov_mds_md *lmm, struct ldlm_extent *extent,
584 struct lustre_handle *lockh)
589 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
590 struct lov_mds_md *lmm, struct lustre_handle *lockh)
595 static struct md_object_operations mdd_dot_lustre_obj_ops = {
596 .moo_permission = dot_lustre_mdd_permission,
597 .moo_attr_get = mdd_attr_get,
598 .moo_attr_set = mdd_attr_set,
599 .moo_xattr_get = dot_lustre_mdd_xattr_get,
600 .moo_xattr_list = dot_lustre_mdd_xattr_list,
601 .moo_xattr_set = dot_lustre_mdd_xattr_set,
602 .moo_xattr_del = dot_lustre_mdd_xattr_del,
603 .moo_readpage = mdd_readpage,
604 .moo_readlink = dot_lustre_mdd_readlink,
605 .moo_object_create = dot_lustre_mdd_object_create,
606 .moo_ref_add = dot_lustre_mdd_ref_add,
607 .moo_ref_del = dot_lustre_mdd_ref_del,
608 .moo_open = dot_lustre_mdd_open,
609 .moo_close = dot_lustre_mdd_close,
610 .moo_capa_get = mdd_capa_get,
611 .moo_object_sync = dot_lustre_mdd_object_sync,
612 .moo_path = dot_lustre_mdd_path,
613 .moo_file_lock = dot_file_lock,
614 .moo_file_unlock = dot_file_unlock,
618 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
619 const struct lu_name *lname, struct lu_fid *f,
620 struct md_op_spec *spec)
622 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
630 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
631 struct md_object *obj,
637 static int dot_lustre_mdd_create(const struct lu_env *env,
638 struct md_object *pobj,
639 const struct lu_name *lname,
640 struct md_object *child,
641 struct md_op_spec *spec,
647 static int dot_lustre_mdd_create_data(const struct lu_env *env,
650 const struct md_op_spec *spec,
656 static int dot_lustre_mdd_rename(const struct lu_env *env,
657 struct md_object *src_pobj,
658 struct md_object *tgt_pobj,
659 const struct lu_fid *lf,
660 const struct lu_name *lsname,
661 struct md_object *tobj,
662 const struct lu_name *ltname,
668 static int dot_lustre_mdd_link(const struct lu_env *env,
669 struct md_object *tgt_obj,
670 struct md_object *src_obj,
671 const struct lu_name *lname,
677 static int dot_lustre_mdd_unlink(const struct lu_env *env,
678 struct md_object *pobj,
679 struct md_object *cobj,
680 const struct lu_name *lname,
686 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
687 struct md_object *obj,
688 const struct lu_name *lname,
689 const struct lu_fid *fid,
690 const struct md_attr *ma)
695 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
696 struct md_object *obj,
697 const struct lu_name *lname,
698 const struct md_attr *ma)
703 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
704 struct md_object *pobj,
705 struct md_object *tobj,
706 const struct lu_fid *fid,
707 const struct lu_name *lname,
714 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
715 .mdo_is_subdir = mdd_is_subdir,
716 .mdo_lookup = dot_lustre_mdd_lookup,
717 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
718 .mdo_create = dot_lustre_mdd_create,
719 .mdo_create_data = dot_lustre_mdd_create_data,
720 .mdo_rename = dot_lustre_mdd_rename,
721 .mdo_link = dot_lustre_mdd_link,
722 .mdo_unlink = dot_lustre_mdd_unlink,
723 .mdo_name_insert = dot_lustre_mdd_name_insert,
724 .mdo_name_remove = dot_lustre_mdd_name_remove,
725 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
728 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
733 if (ma->ma_need & MA_INODE) {
734 struct mdd_device *mdd = mdo2mdd(obj);
736 /* "fid" is a virtual object and hence does not have any "real"
737 * attributes. So we reuse attributes of .lustre for "fid" dir */
738 ma->ma_need |= MA_INODE;
739 rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
742 ma->ma_valid |= MA_INODE;
745 /* "fid" directory does not have any striping information. */
746 if (ma->ma_need & MA_LOV) {
747 struct mdd_object *mdd_obj = md2mdd_obj(obj);
749 if (ma->ma_valid & MA_LOV)
752 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
753 S_ISDIR(mdd_object_type(mdd_obj))))
756 if (ma->ma_need & MA_LOV_DEF) {
757 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
759 ma->ma_lmm_size = rc;
760 ma->ma_valid |= MA_LOV;
769 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
770 const struct md_attr *ma)
775 static int obf_xattr_list(const struct lu_env *env,
776 struct md_object *obj, struct lu_buf *buf)
781 static int obf_xattr_get(const struct lu_env *env,
782 struct md_object *obj, struct lu_buf *buf,
788 static int obf_xattr_set(const struct lu_env *env,
789 struct md_object *obj,
790 const struct lu_buf *buf, const char *name,
796 static int obf_xattr_del(const struct lu_env *env,
797 struct md_object *obj,
803 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
806 struct mdd_object *mdd_obj = md2mdd_obj(obj);
808 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
809 mdd_obj->mod_count++;
810 mdd_write_unlock(env, mdd_obj);
815 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
816 struct md_attr *ma, int mode)
818 struct mdd_object *mdd_obj = md2mdd_obj(obj);
820 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
821 mdd_obj->mod_count--;
822 mdd_write_unlock(env, mdd_obj);
827 /** Nothing to list in "fid" directory */
828 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
829 const struct lu_rdpg *rdpg)
834 static int obf_path(const struct lu_env *env, struct md_object *obj,
835 char *path, int pathlen, __u64 *recno, int *linkno)
840 static struct md_object_operations mdd_obf_obj_ops = {
841 .moo_attr_get = obf_attr_get,
842 .moo_attr_set = obf_attr_set,
843 .moo_xattr_list = obf_xattr_list,
844 .moo_xattr_get = obf_xattr_get,
845 .moo_xattr_set = obf_xattr_set,
846 .moo_xattr_del = obf_xattr_del,
847 .moo_open = obf_mdd_open,
848 .moo_close = obf_mdd_close,
849 .moo_readpage = obf_mdd_readpage,
854 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
855 * are valid. We also check if object with passed fid exists or not.
857 static int obf_lookup(const struct lu_env *env, struct md_object *p,
858 const struct lu_name *lname, struct lu_fid *f,
859 struct md_op_spec *spec)
861 char *name = (char *)lname->ln_name;
862 struct mdd_device *mdd = mdo2mdd(p);
863 struct mdd_object *child;
869 sscanf(name, SFID, RFID(f));
870 if (!fid_is_sane(f)) {
871 CWARN("%s: bad FID format [%s], should be "DFID"\n",
872 mdd->mdd_obd_dev->obd_name, lname->ln_name,
873 (__u64)FID_SEQ_NORMAL, 1, 0);
874 GOTO(out, rc = -EINVAL);
877 if (!fid_is_norm(f)) {
878 CWARN("%s: "DFID" is invalid, sequence should be "
879 ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f),
880 (__u64)FID_SEQ_NORMAL);
881 GOTO(out, rc = -EINVAL);
884 /* Check if object with this fid exists */
885 child = mdd_object_find(env, mdd, f);
889 GOTO(out, rc = PTR_ERR(child));
891 if (mdd_object_exists(child) == 0)
894 mdd_object_put(env, child);
900 static int obf_create(const struct lu_env *env, struct md_object *pobj,
901 const struct lu_name *lname, struct md_object *child,
902 struct md_op_spec *spec, struct md_attr* ma)
907 static int obf_rename(const struct lu_env *env,
908 struct md_object *src_pobj, struct md_object *tgt_pobj,
909 const struct lu_fid *lf, const struct lu_name *lsname,
910 struct md_object *tobj, const struct lu_name *ltname,
916 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
917 struct md_object *src_obj, const struct lu_name *lname,
923 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
924 struct md_object *cobj, const struct lu_name *lname,
930 static struct md_dir_operations mdd_obf_dir_ops = {
931 .mdo_lookup = obf_lookup,
932 .mdo_create = obf_create,
933 .mdo_rename = obf_rename,
934 .mdo_link = obf_link,
935 .mdo_unlink = obf_unlink
939 * Create special in-memory "fid" object for open-by-fid.
941 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
943 struct mdd_object *mdd_obf;
944 struct lu_object *obf_lu_obj;
947 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
949 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
950 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
951 GOTO(out, rc = -ENOENT);
953 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
954 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
955 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
956 /* Don't allow objects to be created in "fid" dir */
957 mdd_obf->mod_flags |= IMMUTE_OBJ;
959 obf_lu_obj = mdd2lu_obj(mdd_obf);
960 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
966 /** Setup ".lustre" directory object */
967 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
969 struct dt_object *dt_dot_lustre;
970 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
973 rc = create_dot_lustre_dir(env, m);
977 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
978 dot_lustre_name, fid);
979 if (IS_ERR(dt_dot_lustre)) {
980 rc = PTR_ERR(dt_dot_lustre);
984 /* references are released in mdd_device_shutdown() */
985 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
988 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
989 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
991 rc = mdd_obf_setup(env, m);
993 CERROR("Error initializing \"fid\" object - %d.\n", rc);
999 static int mdd_process_config(const struct lu_env *env,
1000 struct lu_device *d, struct lustre_cfg *cfg)
1002 struct mdd_device *m = lu2mdd_dev(d);
1003 struct dt_device *dt = m->mdd_child;
1004 struct lu_device *next = &dt->dd_lu_dev;
1008 switch (cfg->lcfg_command) {
1010 struct lprocfs_static_vars lvars;
1012 lprocfs_mdd_init_vars(&lvars);
1013 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1014 if (rc > 0 || rc == -ENOSYS)
1015 /* we don't understand; pass it on */
1016 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1020 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1023 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1025 rc = mdd_init_obd(env, m, cfg);
1027 CERROR("lov init error %d\n", rc);
1031 mdd_changelog_init(env, m);
1034 lu_dev_del_linkage(d->ld_site, d);
1035 mdd_device_shutdown(env, m, cfg);
1037 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1045 static int mdd_lov_set_nextid(const struct lu_env *env,
1046 struct mdd_device *mdd)
1048 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1052 LASSERT(mds->mds_lov_objids != NULL);
1053 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1054 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1055 mds->mds_lov_objids, NULL);
1060 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1061 struct mdd_device *mdd)
1063 /* XXX: to be implemented! */
1068 static int mdd_recovery_complete(const struct lu_env *env,
1069 struct lu_device *d)
1071 struct mdd_device *mdd = lu2mdd_dev(d);
1072 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1073 struct obd_device *obd = mdd2obd_dev(mdd);
1077 LASSERT(mdd != NULL);
1078 LASSERT(obd != NULL);
1080 /* XXX: Do we need this in new stack? */
1081 rc = mdd_lov_set_nextid(env, mdd);
1083 CERROR("mdd_lov_set_nextid() failed %d\n",
1088 /* XXX: cleanup unlink. */
1089 rc = mdd_cleanup_unlink_llog(env, mdd);
1091 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1096 /* Call that with obd_recovering = 1 just to update objids */
1097 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1098 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1100 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1101 cfs_spin_lock(&obd->obd_dev_lock);
1102 obd->obd_recovering = 0;
1103 cfs_spin_unlock(&obd->obd_dev_lock);
1104 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1106 /* XXX: orphans handling. */
1107 __mdd_orphan_cleanup(env, mdd);
1108 rc = next->ld_ops->ldo_recovery_complete(env, next);
1113 static int mdd_prepare(const struct lu_env *env,
1114 struct lu_device *pdev,
1115 struct lu_device *cdev)
1117 struct mdd_device *mdd = lu2mdd_dev(cdev);
1118 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1119 struct dt_object *root;
1124 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1128 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1129 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1130 &mdd->mdd_root_fid);
1131 if (!IS_ERR(root)) {
1132 LASSERT(root != NULL);
1133 lu_object_put(env, &root->do_lu);
1134 rc = orph_index_init(env, mdd);
1141 rc = mdd_dot_lustre_setup(env, mdd);
1143 CERROR("Error(%d) initializing .lustre objects\n", rc);
1147 /* we use capa file to declare llog changes,
1148 * will be fixed with new llog in 2.3 */
1149 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1151 GOTO(out, rc = PTR_ERR(root));
1153 mdd->mdd_capa = root;
1154 rc = mdd_lfsck_setup(env, mdd);
1162 const struct lu_device_operations mdd_lu_ops = {
1163 .ldo_object_alloc = mdd_object_alloc,
1164 .ldo_process_config = mdd_process_config,
1165 .ldo_recovery_complete = mdd_recovery_complete,
1166 .ldo_prepare = mdd_prepare,
1170 * No permission check is needed.
1172 static int mdd_root_get(const struct lu_env *env,
1173 struct md_device *m, struct lu_fid *f)
1175 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1178 *f = mdd->mdd_root_fid;
1183 * No permission check is needed.
1185 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1186 struct obd_statfs *sfs)
1188 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1193 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1199 * No permission check is needed.
1201 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1202 int *md_size, int *cookie_size)
1204 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1207 *md_size = mdd_lov_mdsize(env, mdd);
1208 *cookie_size = mdd_lov_cookiesize(env, mdd);
1213 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1214 int mode, unsigned long timeout, __u32 alg,
1215 struct lustre_capa_key *keys)
1217 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1218 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1222 /* need barrier for mds_capa_keys access. */
1223 cfs_down_write(&mds->mds_notify_lock);
1224 mds->mds_capa_keys = keys;
1225 cfs_up_write(&mds->mds_notify_lock);
1227 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1228 timeout, alg, keys);
1232 static int mdd_update_capa_key(const struct lu_env *env,
1233 struct md_device *m,
1234 struct lustre_capa_key *key)
1236 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1237 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1238 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1242 rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
1243 KEY_CAPA_KEY, sizeof(info), &info, NULL);
1247 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1250 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1252 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1253 return (*h == NULL ? -ENOENT : 0);
1256 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1257 struct lu_device_type *t,
1258 struct lustre_cfg *lcfg)
1260 struct lu_device *l;
1261 struct mdd_device *m;
1265 l = ERR_PTR(-ENOMEM);
1267 md_device_init(&m->mdd_md_dev, t);
1269 l->ld_ops = &mdd_lu_ops;
1270 m->mdd_md_dev.md_ops = &mdd_ops;
1271 md_upcall_init(&m->mdd_md_dev, NULL);
1277 static struct lu_device *mdd_device_free(const struct lu_env *env,
1278 struct lu_device *lu)
1280 struct mdd_device *m = lu2mdd_dev(lu);
1281 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1284 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1285 md_device_fini(&m->mdd_md_dev);
1290 static struct obd_ops mdd_obd_device_ops = {
1291 .o_owner = THIS_MODULE
1294 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1295 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1297 static struct lu_context_key mdd_ucred_key = {
1298 .lct_tags = LCT_SESSION,
1299 .lct_init = mdd_ucred_key_init,
1300 .lct_fini = mdd_ucred_key_fini
1303 struct md_ucred *md_ucred(const struct lu_env *env)
1305 LASSERT(env->le_ses != NULL);
1306 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1308 EXPORT_SYMBOL(md_ucred);
1311 * context key constructor/destructor:
1312 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1314 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1316 struct lu_context_key mdd_capainfo_key = {
1317 .lct_tags = LCT_SESSION,
1318 .lct_init = mdd_capainfo_key_init,
1319 .lct_fini = mdd_capainfo_key_fini
1322 struct md_capainfo *md_capainfo(const struct lu_env *env)
1324 /* NB, in mdt_init0 */
1325 if (env->le_ses == NULL)
1327 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1329 EXPORT_SYMBOL(md_capainfo);
1332 * context key constructor/destructor:
1333 * mdd_quota_key_init, mdd_quota_key_fini
1335 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1337 struct lu_context_key mdd_quota_key = {
1338 .lct_tags = LCT_SESSION,
1339 .lct_init = mdd_quota_key_init,
1340 .lct_fini = mdd_quota_key_fini
1343 struct md_quota *md_quota(const struct lu_env *env)
1345 LASSERT(env->le_ses != NULL);
1346 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1348 EXPORT_SYMBOL(md_quota);
1350 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1352 struct llog_ctxt *ctxt;
1353 struct llog_changelog_user_rec *rec;
1357 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1363 llog_ctxt_put(ctxt);
1367 /* Assume we want it on since somebody registered */
1368 rc = mdd_changelog_on(mdd, 1);
1372 rec->cur_hdr.lrh_len = sizeof(*rec);
1373 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1374 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1375 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1376 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1377 CERROR("Maximum number of changelog users exceeded!\n");
1378 GOTO(out, rc = -EOVERFLOW);
1380 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1381 rec->cur_endrec = mdd->mdd_cl.mc_index;
1382 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1384 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1386 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1389 llog_ctxt_put(ctxt);
1393 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1394 struct thandle *handle)
1399 /* XXX: this is a temporary solution to declare llog changes
1400 * will be fixed in 2.3 with new llog implementation */
1402 LASSERT(mdd->mdd_capa);
1404 /* the llog record could be canceled either by modifying
1405 * the plain llog's header or by destroying the llog itself
1406 * when this record is the last one in it, it can't be known
1407 * here, but the catlog's header will also be modified for
1408 * the second case, then the first case can be covered and
1409 * is no need to declare it */
1411 /* destroy empty plain log */
1412 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1416 /* record the catlog's header if an empty plain log was destroyed */
1417 rc = dt_declare_record_write(env, mdd->mdd_capa,
1418 sizeof(struct llog_logid_rec), 0, handle);
1422 struct mdd_changelog_user_data {
1423 __u64 mcud_endrec; /**< purge record for this user */
1424 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1426 __u32 mcud_minid; /**< user id with lowest rec reference */
1427 __u32 mcud_usercount;
1429 struct mdd_device *mcud_mdd;
1430 const struct lu_env *mcud_env;
1432 #define MCUD_UNREGISTER -1LL
1435 * 1. Find the smallest record everyone is willing to purge
1436 * 2. Update the last purgeable record for this user
1438 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1439 struct llog_rec_hdr *hdr, void *data)
1441 struct llog_changelog_user_rec *rec;
1442 struct mdd_changelog_user_data *mcud =
1443 (struct mdd_changelog_user_data *)data;
1447 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1449 rec = (struct llog_changelog_user_rec *)hdr;
1451 mcud->mcud_usercount++;
1453 /* If we have a new endrec for this id, use it for the following
1454 min check instead of its old value */
1455 if (rec->cur_id == mcud->mcud_id)
1456 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1458 /* Track the minimum referenced record */
1459 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1460 mcud->mcud_minid = rec->cur_id;
1461 mcud->mcud_minrec = rec->cur_endrec;
1464 if (rec->cur_id != mcud->mcud_id)
1467 /* Update this user's record */
1468 mcud->mcud_found = 1;
1470 /* Special case: unregister this user */
1471 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1472 struct llog_cookie cookie;
1474 struct mdd_device *mdd = mcud->mcud_mdd;
1476 cookie.lgc_lgl = llh->lgh_id;
1477 cookie.lgc_index = hdr->lrh_index;
1479 /* XXX This is a workaround for the deadlock of changelog
1480 * adding vs. changelog cancelling. LU-81. */
1481 th = mdd_trans_create(mcud->mcud_env, mdd);
1483 CERROR("Cannot get thandle\n");
1487 rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
1491 rc = mdd_trans_start(mcud->mcud_env, mdd, th);
1495 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1498 mcud->mcud_usercount--;
1501 mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
1505 /* Update the endrec */
1506 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1507 mcud->mcud_id, rec->cur_endrec);
1509 /* hdr+1 is loc of data */
1510 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1511 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1517 static int mdd_changelog_user_purge(const struct lu_env *env,
1518 struct mdd_device *mdd, int id,
1521 struct mdd_changelog_user_data data;
1522 struct llog_ctxt *ctxt;
1526 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1529 data.mcud_minid = 0;
1530 data.mcud_minrec = 0;
1531 data.mcud_usercount = 0;
1532 data.mcud_endrec = endrec;
1533 data.mcud_mdd = mdd;
1534 data.mcud_env = env;
1535 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1536 endrec = mdd->mdd_cl.mc_index;
1537 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1538 if ((data.mcud_endrec == 0) ||
1539 ((data.mcud_endrec > endrec) &&
1540 (data.mcud_endrec != MCUD_UNREGISTER)))
1541 data.mcud_endrec = endrec;
1543 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1546 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1548 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1549 (void *)&data, 0, 0);
1550 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1551 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1552 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1553 data.mcud_minrec, data.mcud_minid);
1554 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1556 CWARN("Could not determine changelog records to purge; rc=%d\n",
1560 llog_ctxt_put(ctxt);
1562 if (!data.mcud_found) {
1563 CWARN("No entry for user %d. Last changelog reference is "
1564 LPD64" by changelog user %d\n", data.mcud_id,
1565 data.mcud_minrec, data.mcud_minid);
1569 if (!rc && data.mcud_usercount == 0)
1570 /* No more users; turn changelogs off */
1571 rc = mdd_changelog_on(mdd, 0);
1577 * May be called remotely from mdt_iocontrol_handle or locally from
1578 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1579 * an obd_ioctl_data format (but local ioctl handler does).
1581 * \param len - data len
1582 * \param karg - ioctl data, in kernel space
1584 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1585 unsigned int cmd, int len, void *karg)
1587 struct mdd_device *mdd;
1588 struct obd_ioctl_data *data = karg;
1592 mdd = lu2mdd_dev(&m->md_lu_dev);
1594 /* Doesn't use obd_ioctl_data */
1596 case OBD_IOC_CHANGELOG_CLEAR: {
1597 struct changelog_setinfo *cs = karg;
1598 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1602 case OBD_IOC_GET_MNTOPT: {
1603 mntopt_t *mntopts = (mntopt_t *)karg;
1604 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1607 case OBD_IOC_START_LFSCK: {
1608 struct lfsck_start *start = karg;
1609 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1611 /* Return the kernel service version. */
1612 /* XXX: version can be used for compatibility in the future. */
1613 start->ls_version = lfsck->ml_version;
1614 rc = mdd_lfsck_start(env, lfsck, start);
1617 case OBD_IOC_STOP_LFSCK: {
1618 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1623 /* Below ioctls use obd_ioctl_data */
1624 if (len != sizeof(*data)) {
1625 CERROR("Bad ioctl size %d\n", len);
1628 if (data->ioc_version != OBD_IOCTL_VERSION) {
1629 CERROR("Bad magic %x != %x\n", data->ioc_version,
1635 case OBD_IOC_CHANGELOG_REG:
1636 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1638 case OBD_IOC_CHANGELOG_DEREG:
1639 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1649 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1650 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1653 const struct md_device_operations mdd_ops = {
1654 .mdo_statfs = mdd_statfs,
1655 .mdo_root_get = mdd_root_get,
1656 .mdo_maxsize_get = mdd_maxsize_get,
1657 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1658 .mdo_update_capa_key= mdd_update_capa_key,
1659 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1660 .mdo_iocontrol = mdd_iocontrol,
1661 #ifdef HAVE_QUOTA_SUPPORT
1663 .mqo_notify = mdd_quota_notify,
1664 .mqo_setup = mdd_quota_setup,
1665 .mqo_cleanup = mdd_quota_cleanup,
1666 .mqo_recovery = mdd_quota_recovery,
1667 .mqo_check = mdd_quota_check,
1668 .mqo_on = mdd_quota_on,
1669 .mqo_off = mdd_quota_off,
1670 .mqo_setinfo = mdd_quota_setinfo,
1671 .mqo_getinfo = mdd_quota_getinfo,
1672 .mqo_setquota = mdd_quota_setquota,
1673 .mqo_getquota = mdd_quota_getquota,
1674 .mqo_getoinfo = mdd_quota_getoinfo,
1675 .mqo_getoquota = mdd_quota_getoquota,
1676 .mqo_invalidate = mdd_quota_invalidate,
1677 .mqo_finvalidate = mdd_quota_finvalidate
1682 static struct lu_device_type_operations mdd_device_type_ops = {
1683 .ldto_init = mdd_type_init,
1684 .ldto_fini = mdd_type_fini,
1686 .ldto_start = mdd_type_start,
1687 .ldto_stop = mdd_type_stop,
1689 .ldto_device_alloc = mdd_device_alloc,
1690 .ldto_device_free = mdd_device_free,
1692 .ldto_device_init = mdd_device_init,
1693 .ldto_device_fini = mdd_device_fini
1696 static struct lu_device_type mdd_device_type = {
1697 .ldt_tags = LU_DEVICE_MD,
1698 .ldt_name = LUSTRE_MDD_NAME,
1699 .ldt_ops = &mdd_device_type_ops,
1700 .ldt_ctx_tags = LCT_MD_THREAD
1703 /* context key constructor: mdd_key_init */
1704 LU_KEY_INIT(mdd, struct mdd_thread_info);
1706 static void mdd_key_fini(const struct lu_context *ctx,
1707 struct lu_context_key *key, void *data)
1709 struct mdd_thread_info *info = data;
1710 if (info->mti_max_lmm != NULL)
1711 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1712 if (info->mti_max_cookie != NULL)
1713 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1714 mdd_buf_put(&info->mti_big_buf);
1719 /* context key: mdd_thread_key */
1720 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1722 static struct lu_local_obj_desc llod_capa_key = {
1723 .llod_name = CAPA_KEYS,
1724 .llod_oid = MDD_CAPA_KEYS_OID,
1728 static struct lu_local_obj_desc llod_mdd_orphan = {
1729 .llod_name = orph_index_name,
1730 .llod_oid = MDD_ORPHAN_OID,
1732 .llod_feat = &dt_directory_features,
1735 static struct lu_local_obj_desc llod_mdd_root = {
1736 .llod_name = mdd_root_dir_name,
1737 .llod_oid = MDD_ROOT_INDEX_OID,
1739 .llod_feat = &dt_directory_features,
1742 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1743 .llod_name = lfsck_bookmark_name,
1744 .llod_oid = LFSCK_BOOKMARK_OID,
1748 static int __init mdd_mod_init(void)
1750 struct lprocfs_static_vars lvars;
1751 lprocfs_mdd_init_vars(&lvars);
1753 llo_local_obj_register(&llod_capa_key);
1754 llo_local_obj_register(&llod_mdd_orphan);
1755 llo_local_obj_register(&llod_mdd_root);
1756 llo_local_obj_register(&llod_lfsck_bookmark);
1758 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1759 LUSTRE_MDD_NAME, &mdd_device_type);
1762 static void __exit mdd_mod_exit(void)
1764 llo_local_obj_unregister(&llod_capa_key);
1765 llo_local_obj_unregister(&llod_mdd_orphan);
1766 llo_local_obj_unregister(&llod_mdd_root);
1767 llo_local_obj_unregister(&llod_lfsck_bookmark);
1769 class_unregister_type(LUSTRE_MDD_NAME);
1772 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1773 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1774 MODULE_LICENSE("GPL");
1776 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);