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_attr_get(const struct lu_env *env,
491 struct md_object *obj, struct md_attr *ma)
493 struct mdd_object *mdd_obj = md2mdd_obj(obj);
495 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
498 static int dot_lustre_mdd_attr_set(const struct lu_env *env,
499 struct md_object *obj,
500 const struct md_attr *ma)
505 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
506 struct md_object *obj, struct lu_buf *buf,
512 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
513 struct md_object *obj, struct lu_buf *buf)
518 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
519 struct md_object *obj,
520 const struct lu_buf *buf, const char *name,
526 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
527 struct md_object *obj,
533 static int dot_lustre_mdd_readlink(const struct lu_env *env,
534 struct md_object *obj, struct lu_buf *buf)
539 static int dot_lustre_mdd_object_create(const struct lu_env *env,
540 struct md_object *obj,
541 const struct md_op_spec *spec,
547 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
548 struct md_object *obj,
549 const struct md_attr *ma)
554 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
555 struct md_object *obj,
561 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
564 struct mdd_object *mdd_obj = md2mdd_obj(obj);
566 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
567 mdd_obj->mod_count++;
568 mdd_write_unlock(env, mdd_obj);
573 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
574 struct md_attr *ma, int mode)
576 struct mdd_object *mdd_obj = md2mdd_obj(obj);
578 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
579 mdd_obj->mod_count--;
580 mdd_write_unlock(env, mdd_obj);
585 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
586 struct md_object *obj)
591 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
592 char *path, int pathlen, __u64 *recno, int *linkno)
597 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
598 struct lov_mds_md *lmm, struct ldlm_extent *extent,
599 struct lustre_handle *lockh)
604 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
605 struct lov_mds_md *lmm, struct lustre_handle *lockh)
610 static struct md_object_operations mdd_dot_lustre_obj_ops = {
611 .moo_permission = dot_lustre_mdd_permission,
612 .moo_attr_get = dot_lustre_mdd_attr_get,
613 .moo_attr_set = dot_lustre_mdd_attr_set,
614 .moo_xattr_get = dot_lustre_mdd_xattr_get,
615 .moo_xattr_list = dot_lustre_mdd_xattr_list,
616 .moo_xattr_set = dot_lustre_mdd_xattr_set,
617 .moo_xattr_del = dot_lustre_mdd_xattr_del,
618 .moo_readpage = mdd_readpage,
619 .moo_readlink = dot_lustre_mdd_readlink,
620 .moo_object_create = dot_lustre_mdd_object_create,
621 .moo_ref_add = dot_lustre_mdd_ref_add,
622 .moo_ref_del = dot_lustre_mdd_ref_del,
623 .moo_open = dot_lustre_mdd_open,
624 .moo_close = dot_lustre_mdd_close,
625 .moo_capa_get = mdd_capa_get,
626 .moo_object_sync = dot_lustre_mdd_object_sync,
627 .moo_path = dot_lustre_mdd_path,
628 .moo_file_lock = dot_file_lock,
629 .moo_file_unlock = dot_file_unlock,
633 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
634 const struct lu_name *lname, struct lu_fid *f,
635 struct md_op_spec *spec)
637 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
645 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
646 struct md_object *obj,
652 static int dot_lustre_mdd_create(const struct lu_env *env,
653 struct md_object *pobj,
654 const struct lu_name *lname,
655 struct md_object *child,
656 struct md_op_spec *spec,
662 static int dot_lustre_mdd_create_data(const struct lu_env *env,
665 const struct md_op_spec *spec,
671 static int dot_lustre_mdd_rename(const struct lu_env *env,
672 struct md_object *src_pobj,
673 struct md_object *tgt_pobj,
674 const struct lu_fid *lf,
675 const struct lu_name *lsname,
676 struct md_object *tobj,
677 const struct lu_name *ltname,
683 static int dot_lustre_mdd_link(const struct lu_env *env,
684 struct md_object *tgt_obj,
685 struct md_object *src_obj,
686 const struct lu_name *lname,
692 static int dot_lustre_mdd_unlink(const struct lu_env *env,
693 struct md_object *pobj,
694 struct md_object *cobj,
695 const struct lu_name *lname,
701 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
702 struct md_object *obj,
703 const struct lu_name *lname,
704 const struct lu_fid *fid,
705 const struct md_attr *ma)
710 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
711 struct md_object *obj,
712 const struct lu_name *lname,
713 const struct md_attr *ma)
718 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
719 struct md_object *pobj,
720 struct md_object *tobj,
721 const struct lu_fid *fid,
722 const struct lu_name *lname,
729 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
730 .mdo_is_subdir = mdd_is_subdir,
731 .mdo_lookup = dot_lustre_mdd_lookup,
732 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
733 .mdo_create = dot_lustre_mdd_create,
734 .mdo_create_data = dot_lustre_mdd_create_data,
735 .mdo_rename = dot_lustre_mdd_rename,
736 .mdo_link = dot_lustre_mdd_link,
737 .mdo_unlink = dot_lustre_mdd_unlink,
738 .mdo_name_insert = dot_lustre_mdd_name_insert,
739 .mdo_name_remove = dot_lustre_mdd_name_remove,
740 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
743 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
748 if (ma->ma_need & MA_INODE) {
749 struct mdd_device *mdd = mdo2mdd(obj);
751 /* "fid" is a virtual object and hence does not have any "real"
752 * attributes. So we reuse attributes of .lustre for "fid" dir */
753 ma->ma_need |= MA_INODE;
754 rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
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_get(const struct lu_env *env,
792 struct md_object *obj, struct lu_buf *buf,
798 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
801 struct mdd_object *mdd_obj = md2mdd_obj(obj);
803 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
804 mdd_obj->mod_count++;
805 mdd_write_unlock(env, mdd_obj);
810 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
811 struct md_attr *ma, int mode)
813 struct mdd_object *mdd_obj = md2mdd_obj(obj);
815 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
816 mdd_obj->mod_count--;
817 mdd_write_unlock(env, mdd_obj);
822 /** Nothing to list in "fid" directory */
823 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
824 const struct lu_rdpg *rdpg)
829 static int obf_path(const struct lu_env *env, struct md_object *obj,
830 char *path, int pathlen, __u64 *recno, int *linkno)
835 static struct md_object_operations mdd_obf_obj_ops = {
836 .moo_attr_get = obf_attr_get,
837 .moo_attr_set = obf_attr_set,
838 .moo_xattr_get = obf_xattr_get,
839 .moo_open = obf_mdd_open,
840 .moo_close = obf_mdd_close,
841 .moo_readpage = obf_mdd_readpage,
846 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
847 * are valid. We also check if object with passed fid exists or not.
849 static int obf_lookup(const struct lu_env *env, struct md_object *p,
850 const struct lu_name *lname, struct lu_fid *f,
851 struct md_op_spec *spec)
853 char *name = (char *)lname->ln_name;
854 struct mdd_device *mdd = mdo2mdd(p);
855 struct mdd_object *child;
861 sscanf(name, SFID, RFID(f));
862 if (!fid_is_sane(f)) {
863 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
865 GOTO(out, rc = -EINVAL);
868 /* Check if object with this fid exists */
869 child = mdd_object_find(env, mdd, f);
873 GOTO(out, rc = PTR_ERR(child));
875 if (mdd_object_exists(child) == 0)
878 mdd_object_put(env, child);
884 static int obf_create(const struct lu_env *env, struct md_object *pobj,
885 const struct lu_name *lname, struct md_object *child,
886 struct md_op_spec *spec, struct md_attr* ma)
891 static int obf_rename(const struct lu_env *env,
892 struct md_object *src_pobj, struct md_object *tgt_pobj,
893 const struct lu_fid *lf, const struct lu_name *lsname,
894 struct md_object *tobj, const struct lu_name *ltname,
900 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
901 struct md_object *src_obj, const struct lu_name *lname,
907 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
908 struct md_object *cobj, const struct lu_name *lname,
914 static struct md_dir_operations mdd_obf_dir_ops = {
915 .mdo_lookup = obf_lookup,
916 .mdo_create = obf_create,
917 .mdo_rename = obf_rename,
918 .mdo_link = obf_link,
919 .mdo_unlink = obf_unlink
923 * Create special in-memory "fid" object for open-by-fid.
925 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
927 struct mdd_object *mdd_obf;
928 struct lu_object *obf_lu_obj;
931 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
933 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
934 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
935 GOTO(out, rc = -ENOENT);
937 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
938 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
939 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
940 /* Don't allow objects to be created in "fid" dir */
941 mdd_obf->mod_flags |= IMMUTE_OBJ;
943 obf_lu_obj = mdd2lu_obj(mdd_obf);
944 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
950 /** Setup ".lustre" directory object */
951 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
953 struct dt_object *dt_dot_lustre;
954 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
957 rc = create_dot_lustre_dir(env, m);
961 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
962 dot_lustre_name, fid);
963 if (IS_ERR(dt_dot_lustre)) {
964 rc = PTR_ERR(dt_dot_lustre);
968 /* references are released in mdd_device_shutdown() */
969 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
972 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
973 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
975 rc = mdd_obf_setup(env, m);
977 CERROR("Error initializing \"fid\" object - %d.\n", rc);
983 static int mdd_process_config(const struct lu_env *env,
984 struct lu_device *d, struct lustre_cfg *cfg)
986 struct mdd_device *m = lu2mdd_dev(d);
987 struct dt_device *dt = m->mdd_child;
988 struct lu_device *next = &dt->dd_lu_dev;
992 switch (cfg->lcfg_command) {
994 struct lprocfs_static_vars lvars;
996 lprocfs_mdd_init_vars(&lvars);
997 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
998 if (rc > 0 || rc == -ENOSYS)
999 /* we don't understand; pass it on */
1000 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1004 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1007 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1009 rc = mdd_init_obd(env, m, cfg);
1011 CERROR("lov init error %d\n", rc);
1015 mdd_changelog_init(env, m);
1018 lu_dev_del_linkage(d->ld_site, d);
1019 mdd_device_shutdown(env, m, cfg);
1021 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1029 static int mdd_lov_set_nextid(const struct lu_env *env,
1030 struct mdd_device *mdd)
1032 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1036 LASSERT(mds->mds_lov_objids != NULL);
1037 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1038 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1039 mds->mds_lov_objids, NULL);
1044 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1045 struct mdd_device *mdd)
1047 /* XXX: to be implemented! */
1052 static int mdd_recovery_complete(const struct lu_env *env,
1053 struct lu_device *d)
1055 struct mdd_device *mdd = lu2mdd_dev(d);
1056 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1057 struct obd_device *obd = mdd2obd_dev(mdd);
1061 LASSERT(mdd != NULL);
1062 LASSERT(obd != NULL);
1064 /* XXX: Do we need this in new stack? */
1065 rc = mdd_lov_set_nextid(env, mdd);
1067 CERROR("mdd_lov_set_nextid() failed %d\n",
1072 /* XXX: cleanup unlink. */
1073 rc = mdd_cleanup_unlink_llog(env, mdd);
1075 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1080 /* Call that with obd_recovering = 1 just to update objids */
1081 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1082 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1084 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1085 cfs_spin_lock(&obd->obd_dev_lock);
1086 obd->obd_recovering = 0;
1087 cfs_spin_unlock(&obd->obd_dev_lock);
1088 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1090 /* XXX: orphans handling. */
1091 __mdd_orphan_cleanup(env, mdd);
1092 rc = next->ld_ops->ldo_recovery_complete(env, next);
1097 static int mdd_prepare(const struct lu_env *env,
1098 struct lu_device *pdev,
1099 struct lu_device *cdev)
1101 struct mdd_device *mdd = lu2mdd_dev(cdev);
1102 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1103 struct dt_object *root;
1108 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1112 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1113 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1114 &mdd->mdd_root_fid);
1115 if (!IS_ERR(root)) {
1116 LASSERT(root != NULL);
1117 lu_object_put(env, &root->do_lu);
1118 rc = orph_index_init(env, mdd);
1125 rc = mdd_dot_lustre_setup(env, mdd);
1127 CERROR("Error(%d) initializing .lustre objects\n", rc);
1131 /* we use capa file to declare llog changes,
1132 * will be fixed with new llog in 2.3 */
1133 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1135 GOTO(out, rc = PTR_ERR(root));
1137 mdd->mdd_capa = root;
1138 rc = mdd_lfsck_setup(env, mdd);
1146 const struct lu_device_operations mdd_lu_ops = {
1147 .ldo_object_alloc = mdd_object_alloc,
1148 .ldo_process_config = mdd_process_config,
1149 .ldo_recovery_complete = mdd_recovery_complete,
1150 .ldo_prepare = mdd_prepare,
1154 * No permission check is needed.
1156 static int mdd_root_get(const struct lu_env *env,
1157 struct md_device *m, struct lu_fid *f)
1159 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1162 *f = mdd->mdd_root_fid;
1167 * No permission check is needed.
1169 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1170 struct obd_statfs *sfs)
1172 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1177 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1183 * No permission check is needed.
1185 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1186 int *md_size, int *cookie_size)
1188 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1191 *md_size = mdd_lov_mdsize(env, mdd);
1192 *cookie_size = mdd_lov_cookiesize(env, mdd);
1197 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1198 int mode, unsigned long timeout, __u32 alg,
1199 struct lustre_capa_key *keys)
1201 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1202 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1206 /* need barrier for mds_capa_keys access. */
1207 cfs_down_write(&mds->mds_notify_lock);
1208 mds->mds_capa_keys = keys;
1209 cfs_up_write(&mds->mds_notify_lock);
1211 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1212 timeout, alg, keys);
1216 static int mdd_update_capa_key(const struct lu_env *env,
1217 struct md_device *m,
1218 struct lustre_capa_key *key)
1220 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1221 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1222 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1226 rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
1227 KEY_CAPA_KEY, sizeof(info), &info, NULL);
1231 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1234 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1236 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1237 return (*h == NULL ? -ENOENT : 0);
1240 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1241 struct lu_device_type *t,
1242 struct lustre_cfg *lcfg)
1244 struct lu_device *l;
1245 struct mdd_device *m;
1249 l = ERR_PTR(-ENOMEM);
1251 md_device_init(&m->mdd_md_dev, t);
1253 l->ld_ops = &mdd_lu_ops;
1254 m->mdd_md_dev.md_ops = &mdd_ops;
1255 md_upcall_init(&m->mdd_md_dev, NULL);
1261 static struct lu_device *mdd_device_free(const struct lu_env *env,
1262 struct lu_device *lu)
1264 struct mdd_device *m = lu2mdd_dev(lu);
1265 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1268 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1269 md_device_fini(&m->mdd_md_dev);
1274 static struct obd_ops mdd_obd_device_ops = {
1275 .o_owner = THIS_MODULE
1278 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1279 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1281 static struct lu_context_key mdd_ucred_key = {
1282 .lct_tags = LCT_SESSION,
1283 .lct_init = mdd_ucred_key_init,
1284 .lct_fini = mdd_ucred_key_fini
1287 struct md_ucred *md_ucred(const struct lu_env *env)
1289 LASSERT(env->le_ses != NULL);
1290 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1292 EXPORT_SYMBOL(md_ucred);
1295 * context key constructor/destructor:
1296 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1298 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1300 struct lu_context_key mdd_capainfo_key = {
1301 .lct_tags = LCT_SESSION,
1302 .lct_init = mdd_capainfo_key_init,
1303 .lct_fini = mdd_capainfo_key_fini
1306 struct md_capainfo *md_capainfo(const struct lu_env *env)
1308 /* NB, in mdt_init0 */
1309 if (env->le_ses == NULL)
1311 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1313 EXPORT_SYMBOL(md_capainfo);
1316 * context key constructor/destructor:
1317 * mdd_quota_key_init, mdd_quota_key_fini
1319 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1321 struct lu_context_key mdd_quota_key = {
1322 .lct_tags = LCT_SESSION,
1323 .lct_init = mdd_quota_key_init,
1324 .lct_fini = mdd_quota_key_fini
1327 struct md_quota *md_quota(const struct lu_env *env)
1329 LASSERT(env->le_ses != NULL);
1330 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1332 EXPORT_SYMBOL(md_quota);
1334 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1336 struct llog_ctxt *ctxt;
1337 struct llog_changelog_user_rec *rec;
1341 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1347 llog_ctxt_put(ctxt);
1351 /* Assume we want it on since somebody registered */
1352 rc = mdd_changelog_on(mdd, 1);
1356 rec->cur_hdr.lrh_len = sizeof(*rec);
1357 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1358 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1359 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1360 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1361 CERROR("Maximum number of changelog users exceeded!\n");
1362 GOTO(out, rc = -EOVERFLOW);
1364 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1365 rec->cur_endrec = mdd->mdd_cl.mc_index;
1366 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1368 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1370 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1373 llog_ctxt_put(ctxt);
1377 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1378 struct thandle *handle)
1383 /* XXX: this is a temporary solution to declare llog changes
1384 * will be fixed in 2.3 with new llog implementation */
1386 LASSERT(mdd->mdd_capa);
1388 /* the llog record could be canceled either by modifying
1389 * the plain llog's header or by destroying the llog itself
1390 * when this record is the last one in it, it can't be known
1391 * here, but the catlog's header will also be modified for
1392 * the second case, then the first case can be covered and
1393 * is no need to declare it */
1395 /* destroy empty plain log */
1396 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1400 /* record the catlog's header if an empty plain log was destroyed */
1401 rc = dt_declare_record_write(env, mdd->mdd_capa,
1402 sizeof(struct llog_logid_rec), 0, handle);
1406 struct mdd_changelog_user_data {
1407 __u64 mcud_endrec; /**< purge record for this user */
1408 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1410 __u32 mcud_minid; /**< user id with lowest rec reference */
1411 __u32 mcud_usercount;
1413 struct mdd_device *mcud_mdd;
1414 const struct lu_env *mcud_env;
1416 #define MCUD_UNREGISTER -1LL
1419 * 1. Find the smallest record everyone is willing to purge
1420 * 2. Update the last purgeable record for this user
1422 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1423 struct llog_rec_hdr *hdr, void *data)
1425 struct llog_changelog_user_rec *rec;
1426 struct mdd_changelog_user_data *mcud =
1427 (struct mdd_changelog_user_data *)data;
1431 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1433 rec = (struct llog_changelog_user_rec *)hdr;
1435 mcud->mcud_usercount++;
1437 /* If we have a new endrec for this id, use it for the following
1438 min check instead of its old value */
1439 if (rec->cur_id == mcud->mcud_id)
1440 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1442 /* Track the minimum referenced record */
1443 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1444 mcud->mcud_minid = rec->cur_id;
1445 mcud->mcud_minrec = rec->cur_endrec;
1448 if (rec->cur_id != mcud->mcud_id)
1451 /* Update this user's record */
1452 mcud->mcud_found = 1;
1454 /* Special case: unregister this user */
1455 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1456 struct llog_cookie cookie;
1458 struct mdd_device *mdd = mcud->mcud_mdd;
1460 cookie.lgc_lgl = llh->lgh_id;
1461 cookie.lgc_index = hdr->lrh_index;
1463 /* XXX This is a workaround for the deadlock of changelog
1464 * adding vs. changelog cancelling. LU-81. */
1465 th = mdd_trans_create(mcud->mcud_env, mdd);
1467 CERROR("Cannot get thandle\n");
1471 rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
1475 rc = mdd_trans_start(mcud->mcud_env, mdd, th);
1479 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1482 mcud->mcud_usercount--;
1485 mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
1489 /* Update the endrec */
1490 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1491 mcud->mcud_id, rec->cur_endrec);
1493 /* hdr+1 is loc of data */
1494 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1495 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1501 static int mdd_changelog_user_purge(const struct lu_env *env,
1502 struct mdd_device *mdd, int id,
1505 struct mdd_changelog_user_data data;
1506 struct llog_ctxt *ctxt;
1510 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1513 data.mcud_minid = 0;
1514 data.mcud_minrec = 0;
1515 data.mcud_usercount = 0;
1516 data.mcud_endrec = endrec;
1517 data.mcud_mdd = mdd;
1518 data.mcud_env = env;
1519 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1520 endrec = mdd->mdd_cl.mc_index;
1521 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1522 if ((data.mcud_endrec == 0) ||
1523 ((data.mcud_endrec > endrec) &&
1524 (data.mcud_endrec != MCUD_UNREGISTER)))
1525 data.mcud_endrec = endrec;
1527 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1530 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1532 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1533 (void *)&data, 0, 0);
1534 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1535 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1536 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1537 data.mcud_minrec, data.mcud_minid);
1538 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1540 CWARN("Could not determine changelog records to purge; rc=%d\n",
1544 llog_ctxt_put(ctxt);
1546 if (!data.mcud_found) {
1547 CWARN("No entry for user %d. Last changelog reference is "
1548 LPD64" by changelog user %d\n", data.mcud_id,
1549 data.mcud_minrec, data.mcud_minid);
1553 if (!rc && data.mcud_usercount == 0)
1554 /* No more users; turn changelogs off */
1555 rc = mdd_changelog_on(mdd, 0);
1561 * May be called remotely from mdt_iocontrol_handle or locally from
1562 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1563 * an obd_ioctl_data format (but local ioctl handler does).
1565 * \param len - data len
1566 * \param karg - ioctl data, in kernel space
1568 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1569 unsigned int cmd, int len, void *karg)
1571 struct mdd_device *mdd;
1572 struct obd_ioctl_data *data = karg;
1576 mdd = lu2mdd_dev(&m->md_lu_dev);
1578 /* Doesn't use obd_ioctl_data */
1580 case OBD_IOC_CHANGELOG_CLEAR: {
1581 struct changelog_setinfo *cs = karg;
1582 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1586 case OBD_IOC_GET_MNTOPT: {
1587 mntopt_t *mntopts = (mntopt_t *)karg;
1588 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1591 case OBD_IOC_START_LFSCK: {
1592 struct lfsck_start *start = karg;
1593 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1595 /* Return the kernel service version. */
1596 /* XXX: version can be used for compatibility in the future. */
1597 start->ls_version = lfsck->ml_version;
1598 rc = mdd_lfsck_start(env, lfsck, start);
1601 case OBD_IOC_STOP_LFSCK: {
1602 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1607 /* Below ioctls use obd_ioctl_data */
1608 if (len != sizeof(*data)) {
1609 CERROR("Bad ioctl size %d\n", len);
1612 if (data->ioc_version != OBD_IOCTL_VERSION) {
1613 CERROR("Bad magic %x != %x\n", data->ioc_version,
1619 case OBD_IOC_CHANGELOG_REG:
1620 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1622 case OBD_IOC_CHANGELOG_DEREG:
1623 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1633 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1634 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1637 const struct md_device_operations mdd_ops = {
1638 .mdo_statfs = mdd_statfs,
1639 .mdo_root_get = mdd_root_get,
1640 .mdo_maxsize_get = mdd_maxsize_get,
1641 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1642 .mdo_update_capa_key= mdd_update_capa_key,
1643 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1644 .mdo_iocontrol = mdd_iocontrol,
1645 #ifdef HAVE_QUOTA_SUPPORT
1647 .mqo_notify = mdd_quota_notify,
1648 .mqo_setup = mdd_quota_setup,
1649 .mqo_cleanup = mdd_quota_cleanup,
1650 .mqo_recovery = mdd_quota_recovery,
1651 .mqo_check = mdd_quota_check,
1652 .mqo_on = mdd_quota_on,
1653 .mqo_off = mdd_quota_off,
1654 .mqo_setinfo = mdd_quota_setinfo,
1655 .mqo_getinfo = mdd_quota_getinfo,
1656 .mqo_setquota = mdd_quota_setquota,
1657 .mqo_getquota = mdd_quota_getquota,
1658 .mqo_getoinfo = mdd_quota_getoinfo,
1659 .mqo_getoquota = mdd_quota_getoquota,
1660 .mqo_invalidate = mdd_quota_invalidate,
1661 .mqo_finvalidate = mdd_quota_finvalidate
1666 static struct lu_device_type_operations mdd_device_type_ops = {
1667 .ldto_init = mdd_type_init,
1668 .ldto_fini = mdd_type_fini,
1670 .ldto_start = mdd_type_start,
1671 .ldto_stop = mdd_type_stop,
1673 .ldto_device_alloc = mdd_device_alloc,
1674 .ldto_device_free = mdd_device_free,
1676 .ldto_device_init = mdd_device_init,
1677 .ldto_device_fini = mdd_device_fini
1680 static struct lu_device_type mdd_device_type = {
1681 .ldt_tags = LU_DEVICE_MD,
1682 .ldt_name = LUSTRE_MDD_NAME,
1683 .ldt_ops = &mdd_device_type_ops,
1684 .ldt_ctx_tags = LCT_MD_THREAD
1687 /* context key constructor: mdd_key_init */
1688 LU_KEY_INIT(mdd, struct mdd_thread_info);
1690 static void mdd_key_fini(const struct lu_context *ctx,
1691 struct lu_context_key *key, void *data)
1693 struct mdd_thread_info *info = data;
1694 if (info->mti_max_lmm != NULL)
1695 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1696 if (info->mti_max_cookie != NULL)
1697 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1698 mdd_buf_put(&info->mti_big_buf);
1703 /* context key: mdd_thread_key */
1704 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1706 static struct lu_local_obj_desc llod_capa_key = {
1707 .llod_name = CAPA_KEYS,
1708 .llod_oid = MDD_CAPA_KEYS_OID,
1712 static struct lu_local_obj_desc llod_mdd_orphan = {
1713 .llod_name = orph_index_name,
1714 .llod_oid = MDD_ORPHAN_OID,
1716 .llod_feat = &dt_directory_features,
1719 static struct lu_local_obj_desc llod_mdd_root = {
1720 .llod_name = mdd_root_dir_name,
1721 .llod_oid = MDD_ROOT_INDEX_OID,
1723 .llod_feat = &dt_directory_features,
1726 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1727 .llod_name = lfsck_bookmark_name,
1728 .llod_oid = LFSCK_BOOKMARK_OID,
1732 static int __init mdd_mod_init(void)
1734 struct lprocfs_static_vars lvars;
1735 lprocfs_mdd_init_vars(&lvars);
1737 llo_local_obj_register(&llod_capa_key);
1738 llo_local_obj_register(&llod_mdd_orphan);
1739 llo_local_obj_register(&llod_mdd_root);
1740 llo_local_obj_register(&llod_lfsck_bookmark);
1742 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1743 LUSTRE_MDD_NAME, &mdd_device_type);
1746 static void __exit mdd_mod_exit(void)
1748 llo_local_obj_unregister(&llod_capa_key);
1749 llo_local_obj_unregister(&llod_mdd_orphan);
1750 llo_local_obj_unregister(&llod_mdd_root);
1751 llo_local_obj_unregister(&llod_lfsck_bookmark);
1753 class_unregister_type(LUSTRE_MDD_NAME);
1756 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1757 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1758 MODULE_LICENSE("GPL");
1760 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);