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 /* Slab for MDD object allocation */
70 cfs_mem_cache_t *mdd_object_kmem;
72 static struct lu_kmem_descr mdd_caches[] = {
74 .ckd_cache = &mdd_object_kmem,
75 .ckd_name = "mdd_obj",
76 .ckd_size = sizeof(struct mdd_object)
83 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
84 const char *name, struct lu_device *next)
86 struct mdd_device *mdd = lu2mdd_dev(d);
90 mdd->mdd_child = lu2dt_dev(next);
92 /* Prepare transactions callbacks. */
93 mdd->mdd_txn_cb.dtc_txn_start = NULL;
94 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
95 mdd->mdd_txn_cb.dtc_txn_commit = NULL;
96 mdd->mdd_txn_cb.dtc_cookie = mdd;
97 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
98 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
99 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
100 /* sync permission changes */
101 mdd->mdd_sync_permission = 1;
103 rc = mdd_procfs_init(mdd, name);
107 static struct lu_device *mdd_device_fini(const struct lu_env *env,
110 struct mdd_device *mdd = lu2mdd_dev(d);
113 rc = mdd_procfs_fini(mdd);
115 CERROR("proc fini error %d \n", rc);
121 static void mdd_changelog_fini(const struct lu_env *env,
122 struct mdd_device *mdd);
124 static void mdd_device_shutdown(const struct lu_env *env,
125 struct mdd_device *m, struct lustre_cfg *cfg)
128 mdd_lfsck_cleanup(env, m);
129 mdd_changelog_fini(env, m);
130 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
131 if (m->mdd_dot_lustre_objs.mdd_obf)
132 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
133 if (m->mdd_dot_lustre)
134 mdd_object_put(env, m->mdd_dot_lustre);
136 mdd_fini_obd(env, m, cfg);
137 orph_index_fini(env, m);
138 if (m->mdd_capa != NULL) {
139 lu_object_put(env, &m->mdd_capa->do_lu);
142 /* remove upcall device*/
143 md_upcall_fini(&m->mdd_md_dev);
147 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
148 struct llog_rec_hdr *hdr, void *data)
150 struct mdd_device *mdd = (struct mdd_device *)data;
151 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
154 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
155 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
158 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
159 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
160 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
161 llh->lgh_id.lgl_oid);
163 mdd->mdd_cl.mc_index = rec->cr.cr_index;
164 RETURN(LLOG_PROC_BREAK);
167 static int changelog_user_init_cb(const struct lu_env *env,
168 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(NULL, ctxt->loc_handle,
209 changelog_init_cb, mdd);
213 CERROR("changelog init failed: %d\n", rc);
216 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
217 mdd->mdd_cl.mc_index);
219 /* Find last changelog user id */
220 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
222 CERROR("no changelog user context\n");
225 if (!ctxt->loc_handle) {
230 rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
231 changelog_user_init_cb, mdd);
235 CERROR("changelog user init failed: %d\n", rc);
239 /* If we have registered users, assume we want changelogs on */
240 if (mdd->mdd_cl.mc_lastuser > 0)
241 rc = mdd_changelog_on(mdd, 1);
246 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
250 mdd->mdd_cl.mc_index = 0;
251 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
252 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
253 mdd->mdd_cl.mc_flags = 0; /* off by default */
254 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
255 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
256 mdd->mdd_cl.mc_lastuser = 0;
258 rc = mdd_changelog_llog_init(mdd);
260 CERROR("Changelog setup during init failed %d\n", rc);
261 mdd->mdd_cl.mc_flags |= CLM_ERR;
267 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
269 mdd->mdd_cl.mc_flags = 0;
272 /* Start / stop recording */
273 int mdd_changelog_on(struct mdd_device *mdd, int on)
277 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
278 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
279 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
280 CERROR("Changelogs cannot be enabled due to error "
281 "condition (see %s log).\n",
282 mdd2obd_dev(mdd)->obd_name);
285 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
286 mdd->mdd_cl.mc_flags |= CLM_ON;
287 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
288 rc = mdd_changelog_write_header(mdd, CLM_START);
290 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
291 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
292 rc = mdd_changelog_write_header(mdd, CLM_FINI);
293 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
294 mdd->mdd_cl.mc_flags &= ~CLM_ON;
295 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
300 static __u64 cl_time(void) {
303 cfs_fs_time_current(&time);
304 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
307 /** Add a changelog entry \a rec to the changelog llog
310 * \param handle - currently ignored since llogs start their own transaction;
311 * this will hopefully be fixed in llog rewrite
314 int mdd_changelog_llog_write(struct mdd_device *mdd,
315 struct llog_changelog_rec *rec,
316 struct thandle *handle)
318 struct obd_device *obd = mdd2obd_dev(mdd);
319 struct llog_ctxt *ctxt;
322 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
323 /* llog_lvfs_write_rec sets the llog tail len */
324 rec->cr_hdr.lrh_type = CHANGELOG_REC;
325 rec->cr.cr_time = cl_time();
326 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
327 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
328 but as long as the MDD transactions are ordered correctly for e.g.
329 rename conflicts, I don't think this should matter. */
330 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
331 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
332 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
336 /* nested journal transaction */
337 rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
343 /** Add a changelog_ext entry \a rec to the changelog llog
346 * \param handle - currently ignored since llogs start their own transaction;
347 * this will hopefully be fixed in llog rewrite
350 int mdd_changelog_ext_llog_write(struct mdd_device *mdd,
351 struct llog_changelog_ext_rec *rec,
352 struct thandle *handle)
354 struct obd_device *obd = mdd2obd_dev(mdd);
355 struct llog_ctxt *ctxt;
358 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
359 /* llog_lvfs_write_rec sets the llog tail len */
360 rec->cr_hdr.lrh_type = CHANGELOG_REC;
361 rec->cr.cr_time = cl_time();
362 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
363 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
364 * but as long as the MDD transactions are ordered correctly for e.g.
365 * rename conflicts, I don't think this should matter. */
366 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
367 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
368 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
372 /* nested journal transaction */
373 rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
379 /** Remove entries with indicies up to and including \a endrec from the
385 int mdd_changelog_llog_cancel(const struct lu_env *env,
386 struct mdd_device *mdd, long long endrec)
388 struct obd_device *obd = mdd2obd_dev(mdd);
389 struct llog_ctxt *ctxt;
390 long long unsigned cur;
393 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
397 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
398 cur = (long long)mdd->mdd_cl.mc_index;
399 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
403 /* purge to "0" is shorthand for everything */
407 /* If purging all records, write a header entry so we don't have an
408 empty catalog and we're sure to have a valid starting index next
409 time. In case of crash, we just restart with old log so we're
412 /* XXX: transaction is started by llog itself */
413 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
418 /* Some records were purged, so reset repeat-access time (so we
419 record new mtime update records, so users can see a file has been
420 changed since the last purge) */
421 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
423 /* XXX: transaction is started by llog itself */
424 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
430 /** Add a CL_MARK record to the changelog
432 * \param markerflags - CLM_*
435 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
437 struct obd_device *obd = mdd2obd_dev(mdd);
438 struct llog_changelog_rec *rec;
440 int len = strlen(obd->obd_name);
444 reclen = llog_data_len(sizeof(*rec) + len);
445 OBD_ALLOC(rec, reclen);
449 rec->cr.cr_flags = CLF_VERSION;
450 rec->cr.cr_type = CL_MARK;
451 rec->cr.cr_namelen = len;
452 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
453 /* Status and action flags */
454 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
456 /* XXX: transaction is started by llog itself */
457 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
458 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
460 /* assume on or off event; reset repeat-access time */
461 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
463 OBD_FREE(rec, reclen);
468 * Create ".lustre" directory.
470 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
472 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
473 struct md_object *mdo;
476 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
477 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
478 mdd_root_dir_name, dot_lustre_name,
479 fid, &dt_directory_features);
480 /* .lustre dir may be already present */
481 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
483 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
484 dot_lustre_name, PFID(fid), rc);
489 lu_object_put(env, &mdo->mo_lu);
495 static int dot_lustre_mdd_permission(const struct lu_env *env,
496 struct md_object *pobj,
497 struct md_object *cobj,
498 struct md_attr *attr, int mask)
500 if (mask & ~(MAY_READ | MAY_EXEC))
506 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
507 struct md_object *obj, struct lu_buf *buf,
513 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
514 struct md_object *obj, struct lu_buf *buf)
519 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
520 struct md_object *obj,
521 const struct lu_buf *buf, const char *name,
527 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
528 struct md_object *obj,
534 static int dot_lustre_mdd_readlink(const struct lu_env *env,
535 struct md_object *obj, struct lu_buf *buf)
540 static int dot_lustre_mdd_object_create(const struct lu_env *env,
541 struct md_object *obj,
542 const struct md_op_spec *spec,
548 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
549 struct md_object *obj,
550 const struct md_attr *ma)
555 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
556 struct md_object *obj,
562 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
565 struct mdd_object *mdd_obj = md2mdd_obj(obj);
567 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
568 mdd_obj->mod_count++;
569 mdd_write_unlock(env, mdd_obj);
574 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
575 struct md_attr *ma, int mode)
577 struct mdd_object *mdd_obj = md2mdd_obj(obj);
579 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
580 mdd_obj->mod_count--;
581 mdd_write_unlock(env, mdd_obj);
586 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
587 struct md_object *obj)
592 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
593 char *path, int pathlen, __u64 *recno, int *linkno)
598 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
599 struct lov_mds_md *lmm, struct ldlm_extent *extent,
600 struct lustre_handle *lockh)
605 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
606 struct lov_mds_md *lmm, struct lustre_handle *lockh)
611 static struct md_object_operations mdd_dot_lustre_obj_ops = {
612 .moo_permission = dot_lustre_mdd_permission,
613 .moo_attr_get = mdd_attr_get,
614 .moo_attr_set = mdd_attr_set,
615 .moo_xattr_get = dot_lustre_mdd_xattr_get,
616 .moo_xattr_list = dot_lustre_mdd_xattr_list,
617 .moo_xattr_set = dot_lustre_mdd_xattr_set,
618 .moo_xattr_del = dot_lustre_mdd_xattr_del,
619 .moo_readpage = mdd_readpage,
620 .moo_readlink = dot_lustre_mdd_readlink,
621 .moo_object_create = dot_lustre_mdd_object_create,
622 .moo_ref_add = dot_lustre_mdd_ref_add,
623 .moo_ref_del = dot_lustre_mdd_ref_del,
624 .moo_open = dot_lustre_mdd_open,
625 .moo_close = dot_lustre_mdd_close,
626 .moo_capa_get = mdd_capa_get,
627 .moo_object_sync = dot_lustre_mdd_object_sync,
628 .moo_path = dot_lustre_mdd_path,
629 .moo_file_lock = dot_file_lock,
630 .moo_file_unlock = dot_file_unlock,
634 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
635 const struct lu_name *lname, struct lu_fid *f,
636 struct md_op_spec *spec)
638 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
646 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
647 struct md_object *obj,
653 static int dot_lustre_mdd_create(const struct lu_env *env,
654 struct md_object *pobj,
655 const struct lu_name *lname,
656 struct md_object *child,
657 struct md_op_spec *spec,
663 static int dot_lustre_mdd_create_data(const struct lu_env *env,
666 const struct md_op_spec *spec,
672 static int dot_lustre_mdd_rename(const struct lu_env *env,
673 struct md_object *src_pobj,
674 struct md_object *tgt_pobj,
675 const struct lu_fid *lf,
676 const struct lu_name *lsname,
677 struct md_object *tobj,
678 const struct lu_name *ltname,
684 static int dot_lustre_mdd_link(const struct lu_env *env,
685 struct md_object *tgt_obj,
686 struct md_object *src_obj,
687 const struct lu_name *lname,
693 static int dot_lustre_mdd_unlink(const struct lu_env *env,
694 struct md_object *pobj,
695 struct md_object *cobj,
696 const struct lu_name *lname,
702 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
703 struct md_object *obj,
704 const struct lu_name *lname,
705 const struct lu_fid *fid,
706 const struct md_attr *ma)
711 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
712 struct md_object *obj,
713 const struct lu_name *lname,
714 const struct md_attr *ma)
719 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
720 struct md_object *pobj,
721 struct md_object *tobj,
722 const struct lu_fid *fid,
723 const struct lu_name *lname,
730 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
731 .mdo_is_subdir = mdd_is_subdir,
732 .mdo_lookup = dot_lustre_mdd_lookup,
733 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
734 .mdo_create = dot_lustre_mdd_create,
735 .mdo_create_data = dot_lustre_mdd_create_data,
736 .mdo_rename = dot_lustre_mdd_rename,
737 .mdo_link = dot_lustre_mdd_link,
738 .mdo_unlink = dot_lustre_mdd_unlink,
739 .mdo_name_insert = dot_lustre_mdd_name_insert,
740 .mdo_name_remove = dot_lustre_mdd_name_remove,
741 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
744 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
749 if (ma->ma_need & MA_INODE) {
750 struct mdd_device *mdd = mdo2mdd(obj);
752 /* "fid" is a virtual object and hence does not have any "real"
753 * attributes. So we reuse attributes of .lustre for "fid" dir */
754 ma->ma_need |= MA_INODE;
755 rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
758 ma->ma_valid |= MA_INODE;
761 /* "fid" directory does not have any striping information. */
762 if (ma->ma_need & MA_LOV) {
763 struct mdd_object *mdd_obj = md2mdd_obj(obj);
765 if (ma->ma_valid & MA_LOV)
768 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
769 S_ISDIR(mdd_object_type(mdd_obj))))
772 if (ma->ma_need & MA_LOV_DEF) {
773 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
775 ma->ma_lmm_size = rc;
776 ma->ma_valid |= MA_LOV;
785 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
786 const struct md_attr *ma)
791 static int obf_xattr_list(const struct lu_env *env,
792 struct md_object *obj, struct lu_buf *buf)
797 static int obf_xattr_get(const struct lu_env *env,
798 struct md_object *obj, struct lu_buf *buf,
804 static int obf_xattr_set(const struct lu_env *env,
805 struct md_object *obj,
806 const struct lu_buf *buf, const char *name,
812 static int obf_xattr_del(const struct lu_env *env,
813 struct md_object *obj,
819 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
822 struct mdd_object *mdd_obj = md2mdd_obj(obj);
824 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
825 mdd_obj->mod_count++;
826 mdd_write_unlock(env, mdd_obj);
831 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
832 struct md_attr *ma, int mode)
834 struct mdd_object *mdd_obj = md2mdd_obj(obj);
836 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
837 mdd_obj->mod_count--;
838 mdd_write_unlock(env, mdd_obj);
843 /** Nothing to list in "fid" directory */
844 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
845 const struct lu_rdpg *rdpg)
850 static int obf_path(const struct lu_env *env, struct md_object *obj,
851 char *path, int pathlen, __u64 *recno, int *linkno)
856 static struct md_object_operations mdd_obf_obj_ops = {
857 .moo_attr_get = obf_attr_get,
858 .moo_attr_set = obf_attr_set,
859 .moo_xattr_list = obf_xattr_list,
860 .moo_xattr_get = obf_xattr_get,
861 .moo_xattr_set = obf_xattr_set,
862 .moo_xattr_del = obf_xattr_del,
863 .moo_open = obf_mdd_open,
864 .moo_close = obf_mdd_close,
865 .moo_readpage = obf_mdd_readpage,
870 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
871 * are valid. We also check if object with passed fid exists or not.
873 static int obf_lookup(const struct lu_env *env, struct md_object *p,
874 const struct lu_name *lname, struct lu_fid *f,
875 struct md_op_spec *spec)
877 char *name = (char *)lname->ln_name;
878 struct mdd_device *mdd = mdo2mdd(p);
879 struct mdd_object *child;
885 sscanf(name, SFID, RFID(f));
886 if (!fid_is_sane(f)) {
887 CWARN("%s: bad FID format [%s], should be "DFID"\n",
888 mdd->mdd_obd_dev->obd_name, lname->ln_name,
889 (__u64)FID_SEQ_NORMAL, 1, 0);
890 GOTO(out, rc = -EINVAL);
893 if (!fid_is_norm(f)) {
894 CWARN("%s: "DFID" is invalid, sequence should be "
895 ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f),
896 (__u64)FID_SEQ_NORMAL);
897 GOTO(out, rc = -EINVAL);
900 /* Check if object with this fid exists */
901 child = mdd_object_find(env, mdd, f);
905 GOTO(out, rc = PTR_ERR(child));
907 if (mdd_object_exists(child) == 0)
910 mdd_object_put(env, child);
916 static int obf_create(const struct lu_env *env, struct md_object *pobj,
917 const struct lu_name *lname, struct md_object *child,
918 struct md_op_spec *spec, struct md_attr* ma)
923 static int obf_rename(const struct lu_env *env,
924 struct md_object *src_pobj, struct md_object *tgt_pobj,
925 const struct lu_fid *lf, const struct lu_name *lsname,
926 struct md_object *tobj, const struct lu_name *ltname,
932 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
933 struct md_object *src_obj, const struct lu_name *lname,
939 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
940 struct md_object *cobj, const struct lu_name *lname,
946 static struct md_dir_operations mdd_obf_dir_ops = {
947 .mdo_lookup = obf_lookup,
948 .mdo_create = obf_create,
949 .mdo_rename = obf_rename,
950 .mdo_link = obf_link,
951 .mdo_unlink = obf_unlink
955 * Create special in-memory "fid" object for open-by-fid.
957 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
959 struct mdd_object *mdd_obf;
960 struct lu_object *obf_lu_obj;
963 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
965 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
966 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
967 GOTO(out, rc = -ENOENT);
969 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
970 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
971 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
972 /* Don't allow objects to be created in "fid" dir */
973 mdd_obf->mod_flags |= IMMUTE_OBJ;
975 obf_lu_obj = mdd2lu_obj(mdd_obf);
976 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
982 /** Setup ".lustre" directory object */
983 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
985 struct dt_object *dt_dot_lustre;
986 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
989 rc = create_dot_lustre_dir(env, m);
993 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
994 dot_lustre_name, fid);
995 if (IS_ERR(dt_dot_lustre)) {
996 rc = PTR_ERR(dt_dot_lustre);
1000 /* references are released in mdd_device_shutdown() */
1001 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
1004 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
1005 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
1007 rc = mdd_obf_setup(env, m);
1009 CERROR("Error initializing \"fid\" object - %d.\n", rc);
1015 static int mdd_process_config(const struct lu_env *env,
1016 struct lu_device *d, struct lustre_cfg *cfg)
1018 struct mdd_device *m = lu2mdd_dev(d);
1019 struct dt_device *dt = m->mdd_child;
1020 struct lu_device *next = &dt->dd_lu_dev;
1024 switch (cfg->lcfg_command) {
1026 struct lprocfs_static_vars lvars;
1028 lprocfs_mdd_init_vars(&lvars);
1029 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1030 if (rc > 0 || rc == -ENOSYS)
1031 /* we don't understand; pass it on */
1032 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1036 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1039 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1041 rc = mdd_init_obd(env, m, cfg);
1043 CERROR("lov init error %d\n", rc);
1047 mdd_changelog_init(env, m);
1050 lu_dev_del_linkage(d->ld_site, d);
1051 mdd_device_shutdown(env, m, cfg);
1053 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1061 static int mdd_lov_set_nextid(const struct lu_env *env,
1062 struct mdd_device *mdd)
1064 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1068 LASSERT(mds->mds_lov_objids != NULL);
1069 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1070 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1071 mds->mds_lov_objids, NULL);
1076 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1077 struct mdd_device *mdd)
1079 /* XXX: to be implemented! */
1084 static int mdd_recovery_complete(const struct lu_env *env,
1085 struct lu_device *d)
1087 struct mdd_device *mdd = lu2mdd_dev(d);
1088 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1089 struct obd_device *obd = mdd2obd_dev(mdd);
1093 LASSERT(mdd != NULL);
1094 LASSERT(obd != NULL);
1096 /* XXX: Do we need this in new stack? */
1097 rc = mdd_lov_set_nextid(env, mdd);
1099 CERROR("mdd_lov_set_nextid() failed %d\n",
1104 /* XXX: cleanup unlink. */
1105 rc = mdd_cleanup_unlink_llog(env, mdd);
1107 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1112 /* Call that with obd_recovering = 1 just to update objids */
1113 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1114 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1116 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1117 cfs_spin_lock(&obd->obd_dev_lock);
1118 obd->obd_recovering = 0;
1119 cfs_spin_unlock(&obd->obd_dev_lock);
1120 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1122 /* XXX: orphans handling. */
1123 __mdd_orphan_cleanup(env, mdd);
1124 rc = next->ld_ops->ldo_recovery_complete(env, next);
1129 static int mdd_prepare(const struct lu_env *env,
1130 struct lu_device *pdev,
1131 struct lu_device *cdev)
1133 struct mdd_device *mdd = lu2mdd_dev(cdev);
1134 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1135 struct dt_object *root;
1140 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1144 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1145 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1146 &mdd->mdd_root_fid);
1147 if (!IS_ERR(root)) {
1148 LASSERT(root != NULL);
1149 lu_object_put(env, &root->do_lu);
1150 rc = orph_index_init(env, mdd);
1157 rc = mdd_dot_lustre_setup(env, mdd);
1159 CERROR("Error(%d) initializing .lustre objects\n", rc);
1163 /* we use capa file to declare llog changes,
1164 * will be fixed with new llog in 2.3 */
1165 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1167 GOTO(out, rc = PTR_ERR(root));
1169 mdd->mdd_capa = root;
1170 rc = mdd_lfsck_setup(env, mdd);
1178 const struct lu_device_operations mdd_lu_ops = {
1179 .ldo_object_alloc = mdd_object_alloc,
1180 .ldo_process_config = mdd_process_config,
1181 .ldo_recovery_complete = mdd_recovery_complete,
1182 .ldo_prepare = mdd_prepare,
1186 * No permission check is needed.
1188 static int mdd_root_get(const struct lu_env *env,
1189 struct md_device *m, struct lu_fid *f)
1191 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1194 *f = mdd->mdd_root_fid;
1199 * No permission check is needed.
1201 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1202 struct obd_statfs *sfs)
1204 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1209 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1215 * No permission check is needed.
1217 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1218 int *md_size, int *cookie_size)
1220 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1223 *md_size = mdd_lov_mdsize(env, mdd);
1224 *cookie_size = mdd_lov_cookiesize(env, mdd);
1229 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1230 int mode, unsigned long timeout, __u32 alg,
1231 struct lustre_capa_key *keys)
1233 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1234 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1238 /* need barrier for mds_capa_keys access. */
1239 cfs_down_write(&mds->mds_notify_lock);
1240 mds->mds_capa_keys = keys;
1241 cfs_up_write(&mds->mds_notify_lock);
1243 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1244 timeout, alg, keys);
1248 static int mdd_update_capa_key(const struct lu_env *env,
1249 struct md_device *m,
1250 struct lustre_capa_key *key)
1252 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1253 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1254 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1258 rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
1259 KEY_CAPA_KEY, sizeof(info), &info, NULL);
1263 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1266 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1268 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1269 return (*h == NULL ? -ENOENT : 0);
1272 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1273 struct lu_device_type *t,
1274 struct lustre_cfg *lcfg)
1276 struct lu_device *l;
1277 struct mdd_device *m;
1281 l = ERR_PTR(-ENOMEM);
1283 md_device_init(&m->mdd_md_dev, t);
1285 l->ld_ops = &mdd_lu_ops;
1286 m->mdd_md_dev.md_ops = &mdd_ops;
1287 md_upcall_init(&m->mdd_md_dev, NULL);
1293 static struct lu_device *mdd_device_free(const struct lu_env *env,
1294 struct lu_device *lu)
1296 struct mdd_device *m = lu2mdd_dev(lu);
1299 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1300 md_device_fini(&m->mdd_md_dev);
1305 static struct obd_ops mdd_obd_device_ops = {
1306 .o_owner = THIS_MODULE
1309 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1310 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1312 static struct lu_context_key mdd_ucred_key = {
1313 .lct_tags = LCT_SESSION,
1314 .lct_init = mdd_ucred_key_init,
1315 .lct_fini = mdd_ucred_key_fini
1318 struct md_ucred *md_ucred(const struct lu_env *env)
1320 LASSERT(env->le_ses != NULL);
1321 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1323 EXPORT_SYMBOL(md_ucred);
1326 * context key constructor/destructor:
1327 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1329 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1331 struct lu_context_key mdd_capainfo_key = {
1332 .lct_tags = LCT_SESSION,
1333 .lct_init = mdd_capainfo_key_init,
1334 .lct_fini = mdd_capainfo_key_fini
1337 struct md_capainfo *md_capainfo(const struct lu_env *env)
1339 /* NB, in mdt_init0 */
1340 if (env->le_ses == NULL)
1342 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1344 EXPORT_SYMBOL(md_capainfo);
1347 * context key constructor/destructor:
1348 * mdd_quota_key_init, mdd_quota_key_fini
1350 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1352 struct lu_context_key mdd_quota_key = {
1353 .lct_tags = LCT_SESSION,
1354 .lct_init = mdd_quota_key_init,
1355 .lct_fini = mdd_quota_key_fini
1358 struct md_quota *md_quota(const struct lu_env *env)
1360 LASSERT(env->le_ses != NULL);
1361 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1363 EXPORT_SYMBOL(md_quota);
1365 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1367 struct llog_ctxt *ctxt;
1368 struct llog_changelog_user_rec *rec;
1372 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1378 llog_ctxt_put(ctxt);
1382 /* Assume we want it on since somebody registered */
1383 rc = mdd_changelog_on(mdd, 1);
1387 rec->cur_hdr.lrh_len = sizeof(*rec);
1388 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1389 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1390 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1391 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1392 CERROR("Maximum number of changelog users exceeded!\n");
1393 GOTO(out, rc = -EOVERFLOW);
1395 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1396 rec->cur_endrec = mdd->mdd_cl.mc_index;
1397 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1399 rc = llog_add(NULL, ctxt, &rec->cur_hdr, NULL, NULL, 0);
1401 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1404 llog_ctxt_put(ctxt);
1408 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1409 struct thandle *handle)
1414 /* XXX: this is a temporary solution to declare llog changes
1415 * will be fixed in 2.3 with new llog implementation */
1417 LASSERT(mdd->mdd_capa);
1419 /* the llog record could be canceled either by modifying
1420 * the plain llog's header or by destroying the llog itself
1421 * when this record is the last one in it, it can't be known
1422 * here, but the catlog's header will also be modified for
1423 * the second case, then the first case can be covered and
1424 * is no need to declare it */
1426 /* destroy empty plain log */
1427 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1431 /* record the catlog's header if an empty plain log was destroyed */
1432 rc = dt_declare_record_write(env, mdd->mdd_capa,
1433 sizeof(struct llog_logid_rec), 0, handle);
1437 struct mdd_changelog_user_data {
1438 __u64 mcud_endrec; /**< purge record for this user */
1439 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1441 __u32 mcud_minid; /**< user id with lowest rec reference */
1442 __u32 mcud_usercount;
1444 struct mdd_device *mcud_mdd;
1446 #define MCUD_UNREGISTER -1LL
1449 * 1. Find the smallest record everyone is willing to purge
1450 * 2. Update the last purgeable record for this user
1452 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1453 struct llog_handle *llh,
1454 struct llog_rec_hdr *hdr, void *data)
1456 struct llog_changelog_user_rec *rec;
1457 struct mdd_changelog_user_data *mcud = data;
1462 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1464 rec = (struct llog_changelog_user_rec *)hdr;
1466 mcud->mcud_usercount++;
1468 /* If we have a new endrec for this id, use it for the following
1469 min check instead of its old value */
1470 if (rec->cur_id == mcud->mcud_id)
1471 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1473 /* Track the minimum referenced record */
1474 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1475 mcud->mcud_minid = rec->cur_id;
1476 mcud->mcud_minrec = rec->cur_endrec;
1479 if (rec->cur_id != mcud->mcud_id)
1482 /* Update this user's record */
1483 mcud->mcud_found = 1;
1485 /* Special case: unregister this user */
1486 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1487 struct llog_cookie cookie;
1489 struct mdd_device *mdd = mcud->mcud_mdd;
1491 cookie.lgc_lgl = llh->lgh_id;
1492 cookie.lgc_index = hdr->lrh_index;
1494 /* XXX This is a workaround for the deadlock of changelog
1495 * adding vs. changelog cancelling. LU-81. */
1496 th = mdd_trans_create(env, mdd);
1498 CERROR("Cannot get thandle\n");
1502 rc = mdd_declare_llog_cancel(env, mdd, th);
1506 rc = mdd_trans_start(env, mdd, th);
1510 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1513 mcud->mcud_usercount--;
1516 mdd_trans_stop(env, mdd, 0, th);
1520 /* Update the endrec */
1521 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1522 mcud->mcud_id, rec->cur_endrec);
1524 /* hdr+1 is loc of data */
1525 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1526 rc = llog_write_rec(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1532 static int mdd_changelog_user_purge(const struct lu_env *env,
1533 struct mdd_device *mdd, int id,
1536 struct mdd_changelog_user_data data;
1537 struct llog_ctxt *ctxt;
1541 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1544 data.mcud_minid = 0;
1545 data.mcud_minrec = 0;
1546 data.mcud_usercount = 0;
1547 data.mcud_endrec = endrec;
1548 data.mcud_mdd = mdd;
1549 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1550 endrec = mdd->mdd_cl.mc_index;
1551 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1552 if ((data.mcud_endrec == 0) ||
1553 ((data.mcud_endrec > endrec) &&
1554 (data.mcud_endrec != MCUD_UNREGISTER)))
1555 data.mcud_endrec = endrec;
1557 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1560 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1562 rc = llog_cat_process(env, ctxt->loc_handle,
1563 mdd_changelog_user_purge_cb, (void *)&data,
1565 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1566 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1567 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1568 data.mcud_minrec, data.mcud_minid);
1569 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1571 CWARN("Could not determine changelog records to purge; rc=%d\n",
1575 llog_ctxt_put(ctxt);
1577 if (!data.mcud_found) {
1578 CWARN("No entry for user %d. Last changelog reference is "
1579 LPD64" by changelog user %d\n", data.mcud_id,
1580 data.mcud_minrec, data.mcud_minid);
1584 if (!rc && data.mcud_usercount == 0)
1585 /* No more users; turn changelogs off */
1586 rc = mdd_changelog_on(mdd, 0);
1592 * May be called remotely from mdt_iocontrol_handle or locally from
1593 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1594 * an obd_ioctl_data format (but local ioctl handler does).
1596 * \param len - data len
1597 * \param karg - ioctl data, in kernel space
1599 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1600 unsigned int cmd, int len, void *karg)
1602 struct mdd_device *mdd;
1603 struct obd_ioctl_data *data = karg;
1607 mdd = lu2mdd_dev(&m->md_lu_dev);
1609 /* Doesn't use obd_ioctl_data */
1611 case OBD_IOC_CHANGELOG_CLEAR: {
1612 struct changelog_setinfo *cs = karg;
1613 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1617 case OBD_IOC_GET_MNTOPT: {
1618 mntopt_t *mntopts = (mntopt_t *)karg;
1619 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1622 case OBD_IOC_START_LFSCK: {
1623 struct lfsck_start *start = karg;
1624 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1626 /* Return the kernel service version. */
1627 /* XXX: version can be used for compatibility in the future. */
1628 start->ls_version = lfsck->ml_version;
1629 rc = mdd_lfsck_start(env, lfsck, start);
1632 case OBD_IOC_STOP_LFSCK: {
1633 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1638 /* Below ioctls use obd_ioctl_data */
1639 if (len != sizeof(*data)) {
1640 CERROR("Bad ioctl size %d\n", len);
1643 if (data->ioc_version != OBD_IOCTL_VERSION) {
1644 CERROR("Bad magic %x != %x\n", data->ioc_version,
1650 case OBD_IOC_CHANGELOG_REG:
1651 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1653 case OBD_IOC_CHANGELOG_DEREG:
1654 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1664 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1665 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1668 const struct md_device_operations mdd_ops = {
1669 .mdo_statfs = mdd_statfs,
1670 .mdo_root_get = mdd_root_get,
1671 .mdo_maxsize_get = mdd_maxsize_get,
1672 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1673 .mdo_update_capa_key= mdd_update_capa_key,
1674 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1675 .mdo_iocontrol = mdd_iocontrol,
1676 #ifdef HAVE_QUOTA_SUPPORT
1678 .mqo_notify = mdd_quota_notify,
1679 .mqo_setup = mdd_quota_setup,
1680 .mqo_cleanup = mdd_quota_cleanup,
1681 .mqo_recovery = mdd_quota_recovery,
1682 .mqo_check = mdd_quota_check,
1683 .mqo_on = mdd_quota_on,
1684 .mqo_off = mdd_quota_off,
1685 .mqo_setinfo = mdd_quota_setinfo,
1686 .mqo_getinfo = mdd_quota_getinfo,
1687 .mqo_setquota = mdd_quota_setquota,
1688 .mqo_getquota = mdd_quota_getquota,
1689 .mqo_getoinfo = mdd_quota_getoinfo,
1690 .mqo_getoquota = mdd_quota_getoquota,
1691 .mqo_invalidate = mdd_quota_invalidate,
1692 .mqo_finvalidate = mdd_quota_finvalidate
1697 static struct lu_device_type_operations mdd_device_type_ops = {
1698 .ldto_init = mdd_type_init,
1699 .ldto_fini = mdd_type_fini,
1701 .ldto_start = mdd_type_start,
1702 .ldto_stop = mdd_type_stop,
1704 .ldto_device_alloc = mdd_device_alloc,
1705 .ldto_device_free = mdd_device_free,
1707 .ldto_device_init = mdd_device_init,
1708 .ldto_device_fini = mdd_device_fini
1711 static struct lu_device_type mdd_device_type = {
1712 .ldt_tags = LU_DEVICE_MD,
1713 .ldt_name = LUSTRE_MDD_NAME,
1714 .ldt_ops = &mdd_device_type_ops,
1715 .ldt_ctx_tags = LCT_MD_THREAD
1718 /* context key constructor: mdd_key_init */
1719 LU_KEY_INIT(mdd, struct mdd_thread_info);
1721 static void mdd_key_fini(const struct lu_context *ctx,
1722 struct lu_context_key *key, void *data)
1724 struct mdd_thread_info *info = data;
1725 if (info->mti_max_lmm != NULL)
1726 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1727 if (info->mti_max_cookie != NULL)
1728 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1729 mdd_buf_put(&info->mti_big_buf);
1734 /* context key: mdd_thread_key */
1735 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1737 static struct lu_local_obj_desc llod_capa_key = {
1738 .llod_name = CAPA_KEYS,
1739 .llod_oid = MDD_CAPA_KEYS_OID,
1743 static struct lu_local_obj_desc llod_mdd_orphan = {
1744 .llod_name = orph_index_name,
1745 .llod_oid = MDD_ORPHAN_OID,
1747 .llod_feat = &dt_directory_features,
1750 static struct lu_local_obj_desc llod_mdd_root = {
1751 .llod_name = mdd_root_dir_name,
1752 .llod_oid = MDD_ROOT_INDEX_OID,
1754 .llod_feat = &dt_directory_features,
1757 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1758 .llod_name = lfsck_bookmark_name,
1759 .llod_oid = LFSCK_BOOKMARK_OID,
1763 static int __init mdd_mod_init(void)
1765 struct lprocfs_static_vars lvars;
1768 lprocfs_mdd_init_vars(&lvars);
1770 rc = lu_kmem_init(mdd_caches);
1774 llo_local_obj_register(&llod_capa_key);
1775 llo_local_obj_register(&llod_mdd_orphan);
1776 llo_local_obj_register(&llod_mdd_root);
1777 llo_local_obj_register(&llod_lfsck_bookmark);
1779 rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1780 LUSTRE_MDD_NAME, &mdd_device_type);
1782 lu_kmem_fini(mdd_caches);
1786 static void __exit mdd_mod_exit(void)
1788 llo_local_obj_unregister(&llod_capa_key);
1789 llo_local_obj_unregister(&llod_mdd_orphan);
1790 llo_local_obj_unregister(&llod_mdd_root);
1791 llo_local_obj_unregister(&llod_lfsck_bookmark);
1793 class_unregister_type(LUSTRE_MDD_NAME);
1794 lu_kmem_fini(mdd_caches);
1797 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1798 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1799 MODULE_LICENSE("GPL");
1801 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);