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);
99 rc = mdd_procfs_fini(mdd);
101 CERROR("proc fini error %d \n", rc);
107 static void mdd_changelog_fini(const struct lu_env *env,
108 struct mdd_device *mdd);
110 static void mdd_device_shutdown(const struct lu_env *env,
111 struct mdd_device *m, struct lustre_cfg *cfg)
114 mdd_lfsck_cleanup(env, m);
115 mdd_changelog_fini(env, m);
116 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
117 if (m->mdd_dot_lustre_objs.mdd_obf)
118 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
119 if (m->mdd_dot_lustre)
120 mdd_object_put(env, m->mdd_dot_lustre);
122 mdd_fini_obd(env, m, cfg);
123 orph_index_fini(env, m);
124 if (m->mdd_capa != NULL) {
125 lu_object_put(env, &m->mdd_capa->do_lu);
128 /* remove upcall device*/
129 md_upcall_fini(&m->mdd_md_dev);
133 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
134 struct llog_rec_hdr *hdr, void *data)
136 struct mdd_device *mdd = (struct mdd_device *)data;
137 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
140 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
141 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
144 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
145 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
146 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
147 llh->lgh_id.lgl_oid);
149 mdd->mdd_cl.mc_index = rec->cr.cr_index;
150 RETURN(LLOG_PROC_BREAK);
153 static int changelog_user_init_cb(const struct lu_env *env,
154 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(NULL, ctxt->loc_handle,
195 changelog_init_cb, mdd);
199 CERROR("changelog init failed: %d\n", rc);
202 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
203 mdd->mdd_cl.mc_index);
205 /* Find last changelog user id */
206 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
208 CERROR("no changelog user context\n");
211 if (!ctxt->loc_handle) {
216 rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
217 changelog_user_init_cb, mdd);
221 CERROR("changelog user init failed: %d\n", rc);
225 /* If we have registered users, assume we want changelogs on */
226 if (mdd->mdd_cl.mc_lastuser > 0)
227 rc = mdd_changelog_on(mdd, 1);
232 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
236 mdd->mdd_cl.mc_index = 0;
237 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
238 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
239 mdd->mdd_cl.mc_flags = 0; /* off by default */
240 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
241 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
242 mdd->mdd_cl.mc_lastuser = 0;
244 rc = mdd_changelog_llog_init(mdd);
246 CERROR("Changelog setup during init failed %d\n", rc);
247 mdd->mdd_cl.mc_flags |= CLM_ERR;
253 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
255 mdd->mdd_cl.mc_flags = 0;
258 /* Start / stop recording */
259 int mdd_changelog_on(struct mdd_device *mdd, int on)
263 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
264 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
265 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
266 CERROR("Changelogs cannot be enabled due to error "
267 "condition (see %s log).\n",
268 mdd2obd_dev(mdd)->obd_name);
271 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
272 mdd->mdd_cl.mc_flags |= CLM_ON;
273 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
274 rc = mdd_changelog_write_header(mdd, CLM_START);
276 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
277 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
278 rc = mdd_changelog_write_header(mdd, CLM_FINI);
279 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
280 mdd->mdd_cl.mc_flags &= ~CLM_ON;
281 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
286 static __u64 cl_time(void) {
289 cfs_fs_time_current(&time);
290 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
293 /** Add a changelog entry \a rec to the changelog llog
296 * \param handle - currently ignored since llogs start their own transaction;
297 * this will hopefully be fixed in llog rewrite
300 int mdd_changelog_llog_write(struct mdd_device *mdd,
301 struct llog_changelog_rec *rec,
302 struct thandle *handle)
304 struct obd_device *obd = mdd2obd_dev(mdd);
305 struct llog_ctxt *ctxt;
308 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
309 /* llog_lvfs_write_rec sets the llog tail len */
310 rec->cr_hdr.lrh_type = CHANGELOG_REC;
311 rec->cr.cr_time = cl_time();
312 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
313 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
314 but as long as the MDD transactions are ordered correctly for e.g.
315 rename conflicts, I don't think this should matter. */
316 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
317 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
318 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
322 /* nested journal transaction */
323 rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
329 /** Add a changelog_ext entry \a rec to the changelog llog
332 * \param handle - currently ignored since llogs start their own transaction;
333 * this will hopefully be fixed in llog rewrite
336 int mdd_changelog_ext_llog_write(struct mdd_device *mdd,
337 struct llog_changelog_ext_rec *rec,
338 struct thandle *handle)
340 struct obd_device *obd = mdd2obd_dev(mdd);
341 struct llog_ctxt *ctxt;
344 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
345 /* llog_lvfs_write_rec sets the llog tail len */
346 rec->cr_hdr.lrh_type = CHANGELOG_REC;
347 rec->cr.cr_time = cl_time();
348 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
349 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
350 * but as long as the MDD transactions are ordered correctly for e.g.
351 * rename conflicts, I don't think this should matter. */
352 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
353 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
354 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
358 /* nested journal transaction */
359 rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
365 /** Remove entries with indicies up to and including \a endrec from the
371 int mdd_changelog_llog_cancel(const struct lu_env *env,
372 struct mdd_device *mdd, long long endrec)
374 struct obd_device *obd = mdd2obd_dev(mdd);
375 struct llog_ctxt *ctxt;
376 long long unsigned cur;
379 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
383 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
384 cur = (long long)mdd->mdd_cl.mc_index;
385 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
389 /* purge to "0" is shorthand for everything */
393 /* If purging all records, write a header entry so we don't have an
394 empty catalog and we're sure to have a valid starting index next
395 time. In case of crash, we just restart with old log so we're
398 /* XXX: transaction is started by llog itself */
399 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
404 /* Some records were purged, so reset repeat-access time (so we
405 record new mtime update records, so users can see a file has been
406 changed since the last purge) */
407 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
409 /* XXX: transaction is started by llog itself */
410 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
416 /** Add a CL_MARK record to the changelog
418 * \param markerflags - CLM_*
421 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
423 struct obd_device *obd = mdd2obd_dev(mdd);
424 struct llog_changelog_rec *rec;
426 int len = strlen(obd->obd_name);
430 reclen = llog_data_len(sizeof(*rec) + len);
431 OBD_ALLOC(rec, reclen);
435 rec->cr.cr_flags = CLF_VERSION;
436 rec->cr.cr_type = CL_MARK;
437 rec->cr.cr_namelen = len;
438 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
439 /* Status and action flags */
440 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
442 /* XXX: transaction is started by llog itself */
443 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
444 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
446 /* assume on or off event; reset repeat-access time */
447 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
449 OBD_FREE(rec, reclen);
454 * Create ".lustre" directory.
456 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
458 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
459 struct md_object *mdo;
462 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
463 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
464 mdd_root_dir_name, dot_lustre_name,
465 fid, &dt_directory_features);
466 /* .lustre dir may be already present */
467 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
469 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
470 dot_lustre_name, PFID(fid), rc);
475 lu_object_put(env, &mdo->mo_lu);
481 static int dot_lustre_mdd_permission(const struct lu_env *env,
482 struct md_object *pobj,
483 struct md_object *cobj,
484 struct md_attr *attr, int mask)
486 if (mask & ~(MAY_READ | MAY_EXEC))
492 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
493 struct md_object *obj, struct lu_buf *buf,
499 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
500 struct md_object *obj, struct lu_buf *buf)
505 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
506 struct md_object *obj,
507 const struct lu_buf *buf, const char *name,
513 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
514 struct md_object *obj,
520 static int dot_lustre_mdd_readlink(const struct lu_env *env,
521 struct md_object *obj, struct lu_buf *buf)
526 static int dot_lustre_mdd_object_create(const struct lu_env *env,
527 struct md_object *obj,
528 const struct md_op_spec *spec,
534 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
535 struct md_object *obj,
536 const struct md_attr *ma)
541 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
542 struct md_object *obj,
548 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
551 struct mdd_object *mdd_obj = md2mdd_obj(obj);
553 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
554 mdd_obj->mod_count++;
555 mdd_write_unlock(env, mdd_obj);
560 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
561 struct md_attr *ma, int mode)
563 struct mdd_object *mdd_obj = md2mdd_obj(obj);
565 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
566 mdd_obj->mod_count--;
567 mdd_write_unlock(env, mdd_obj);
572 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
573 struct md_object *obj)
578 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
579 char *path, int pathlen, __u64 *recno, int *linkno)
584 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
585 struct lov_mds_md *lmm, struct ldlm_extent *extent,
586 struct lustre_handle *lockh)
591 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
592 struct lov_mds_md *lmm, struct lustre_handle *lockh)
597 static struct md_object_operations mdd_dot_lustre_obj_ops = {
598 .moo_permission = dot_lustre_mdd_permission,
599 .moo_attr_get = mdd_attr_get,
600 .moo_attr_set = mdd_attr_set,
601 .moo_xattr_get = dot_lustre_mdd_xattr_get,
602 .moo_xattr_list = dot_lustre_mdd_xattr_list,
603 .moo_xattr_set = dot_lustre_mdd_xattr_set,
604 .moo_xattr_del = dot_lustre_mdd_xattr_del,
605 .moo_readpage = mdd_readpage,
606 .moo_readlink = dot_lustre_mdd_readlink,
607 .moo_object_create = dot_lustre_mdd_object_create,
608 .moo_ref_add = dot_lustre_mdd_ref_add,
609 .moo_ref_del = dot_lustre_mdd_ref_del,
610 .moo_open = dot_lustre_mdd_open,
611 .moo_close = dot_lustre_mdd_close,
612 .moo_capa_get = mdd_capa_get,
613 .moo_object_sync = dot_lustre_mdd_object_sync,
614 .moo_path = dot_lustre_mdd_path,
615 .moo_file_lock = dot_file_lock,
616 .moo_file_unlock = dot_file_unlock,
620 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
621 const struct lu_name *lname, struct lu_fid *f,
622 struct md_op_spec *spec)
624 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
632 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
633 struct md_object *obj,
639 static int dot_lustre_mdd_create(const struct lu_env *env,
640 struct md_object *pobj,
641 const struct lu_name *lname,
642 struct md_object *child,
643 struct md_op_spec *spec,
649 static int dot_lustre_mdd_create_data(const struct lu_env *env,
652 const struct md_op_spec *spec,
658 static int dot_lustre_mdd_rename(const struct lu_env *env,
659 struct md_object *src_pobj,
660 struct md_object *tgt_pobj,
661 const struct lu_fid *lf,
662 const struct lu_name *lsname,
663 struct md_object *tobj,
664 const struct lu_name *ltname,
670 static int dot_lustre_mdd_link(const struct lu_env *env,
671 struct md_object *tgt_obj,
672 struct md_object *src_obj,
673 const struct lu_name *lname,
679 static int dot_lustre_mdd_unlink(const struct lu_env *env,
680 struct md_object *pobj,
681 struct md_object *cobj,
682 const struct lu_name *lname,
688 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
689 struct md_object *obj,
690 const struct lu_name *lname,
691 const struct lu_fid *fid,
692 const struct md_attr *ma)
697 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
698 struct md_object *obj,
699 const struct lu_name *lname,
700 const struct md_attr *ma)
705 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
706 struct md_object *pobj,
707 struct md_object *tobj,
708 const struct lu_fid *fid,
709 const struct lu_name *lname,
716 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
717 .mdo_is_subdir = mdd_is_subdir,
718 .mdo_lookup = dot_lustre_mdd_lookup,
719 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
720 .mdo_create = dot_lustre_mdd_create,
721 .mdo_create_data = dot_lustre_mdd_create_data,
722 .mdo_rename = dot_lustre_mdd_rename,
723 .mdo_link = dot_lustre_mdd_link,
724 .mdo_unlink = dot_lustre_mdd_unlink,
725 .mdo_name_insert = dot_lustre_mdd_name_insert,
726 .mdo_name_remove = dot_lustre_mdd_name_remove,
727 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
730 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
735 if (ma->ma_need & MA_INODE) {
736 struct mdd_device *mdd = mdo2mdd(obj);
738 /* "fid" is a virtual object and hence does not have any "real"
739 * attributes. So we reuse attributes of .lustre for "fid" dir */
740 ma->ma_need |= MA_INODE;
741 rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
744 ma->ma_valid |= MA_INODE;
747 /* "fid" directory does not have any striping information. */
748 if (ma->ma_need & MA_LOV) {
749 struct mdd_object *mdd_obj = md2mdd_obj(obj);
751 if (ma->ma_valid & MA_LOV)
754 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
755 S_ISDIR(mdd_object_type(mdd_obj))))
758 if (ma->ma_need & MA_LOV_DEF) {
759 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
761 ma->ma_lmm_size = rc;
762 ma->ma_valid |= MA_LOV;
771 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
772 const struct md_attr *ma)
777 static int obf_xattr_list(const struct lu_env *env,
778 struct md_object *obj, struct lu_buf *buf)
783 static int obf_xattr_get(const struct lu_env *env,
784 struct md_object *obj, struct lu_buf *buf,
790 static int obf_xattr_set(const struct lu_env *env,
791 struct md_object *obj,
792 const struct lu_buf *buf, const char *name,
798 static int obf_xattr_del(const struct lu_env *env,
799 struct md_object *obj,
805 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
808 struct mdd_object *mdd_obj = md2mdd_obj(obj);
810 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
811 mdd_obj->mod_count++;
812 mdd_write_unlock(env, mdd_obj);
817 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
818 struct md_attr *ma, int mode)
820 struct mdd_object *mdd_obj = md2mdd_obj(obj);
822 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
823 mdd_obj->mod_count--;
824 mdd_write_unlock(env, mdd_obj);
829 /** Nothing to list in "fid" directory */
830 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
831 const struct lu_rdpg *rdpg)
836 static int obf_path(const struct lu_env *env, struct md_object *obj,
837 char *path, int pathlen, __u64 *recno, int *linkno)
842 static struct md_object_operations mdd_obf_obj_ops = {
843 .moo_attr_get = obf_attr_get,
844 .moo_attr_set = obf_attr_set,
845 .moo_xattr_list = obf_xattr_list,
846 .moo_xattr_get = obf_xattr_get,
847 .moo_xattr_set = obf_xattr_set,
848 .moo_xattr_del = obf_xattr_del,
849 .moo_open = obf_mdd_open,
850 .moo_close = obf_mdd_close,
851 .moo_readpage = obf_mdd_readpage,
856 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
857 * are valid. We also check if object with passed fid exists or not.
859 static int obf_lookup(const struct lu_env *env, struct md_object *p,
860 const struct lu_name *lname, struct lu_fid *f,
861 struct md_op_spec *spec)
863 char *name = (char *)lname->ln_name;
864 struct mdd_device *mdd = mdo2mdd(p);
865 struct mdd_object *child;
871 sscanf(name, SFID, RFID(f));
872 if (!fid_is_sane(f)) {
873 CWARN("%s: bad FID format [%s], should be "DFID"\n",
874 mdd->mdd_obd_dev->obd_name, lname->ln_name,
875 (__u64)FID_SEQ_NORMAL, 1, 0);
876 GOTO(out, rc = -EINVAL);
879 if (!fid_is_norm(f)) {
880 CWARN("%s: "DFID" is invalid, sequence should be "
881 ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f),
882 (__u64)FID_SEQ_NORMAL);
883 GOTO(out, rc = -EINVAL);
886 /* Check if object with this fid exists */
887 child = mdd_object_find(env, mdd, f);
891 GOTO(out, rc = PTR_ERR(child));
893 if (mdd_object_exists(child) == 0)
896 mdd_object_put(env, child);
902 static int obf_create(const struct lu_env *env, struct md_object *pobj,
903 const struct lu_name *lname, struct md_object *child,
904 struct md_op_spec *spec, struct md_attr* ma)
909 static int obf_rename(const struct lu_env *env,
910 struct md_object *src_pobj, struct md_object *tgt_pobj,
911 const struct lu_fid *lf, const struct lu_name *lsname,
912 struct md_object *tobj, const struct lu_name *ltname,
918 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
919 struct md_object *src_obj, const struct lu_name *lname,
925 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
926 struct md_object *cobj, const struct lu_name *lname,
932 static struct md_dir_operations mdd_obf_dir_ops = {
933 .mdo_lookup = obf_lookup,
934 .mdo_create = obf_create,
935 .mdo_rename = obf_rename,
936 .mdo_link = obf_link,
937 .mdo_unlink = obf_unlink
941 * Create special in-memory "fid" object for open-by-fid.
943 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
945 struct mdd_object *mdd_obf;
946 struct lu_object *obf_lu_obj;
949 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
951 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
952 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
953 GOTO(out, rc = -ENOENT);
955 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
956 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
957 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
958 /* Don't allow objects to be created in "fid" dir */
959 mdd_obf->mod_flags |= IMMUTE_OBJ;
961 obf_lu_obj = mdd2lu_obj(mdd_obf);
962 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
968 /** Setup ".lustre" directory object */
969 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
971 struct dt_object *dt_dot_lustre;
972 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
975 rc = create_dot_lustre_dir(env, m);
979 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
980 dot_lustre_name, fid);
981 if (IS_ERR(dt_dot_lustre)) {
982 rc = PTR_ERR(dt_dot_lustre);
986 /* references are released in mdd_device_shutdown() */
987 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
990 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
991 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
993 rc = mdd_obf_setup(env, m);
995 CERROR("Error initializing \"fid\" object - %d.\n", rc);
1001 static int mdd_process_config(const struct lu_env *env,
1002 struct lu_device *d, struct lustre_cfg *cfg)
1004 struct mdd_device *m = lu2mdd_dev(d);
1005 struct dt_device *dt = m->mdd_child;
1006 struct lu_device *next = &dt->dd_lu_dev;
1010 switch (cfg->lcfg_command) {
1012 struct lprocfs_static_vars lvars;
1014 lprocfs_mdd_init_vars(&lvars);
1015 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1016 if (rc > 0 || rc == -ENOSYS)
1017 /* we don't understand; pass it on */
1018 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1022 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1025 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1027 rc = mdd_init_obd(env, m, cfg);
1029 CERROR("lov init error %d\n", rc);
1033 mdd_changelog_init(env, m);
1036 lu_dev_del_linkage(d->ld_site, d);
1037 mdd_device_shutdown(env, m, cfg);
1039 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1047 static int mdd_lov_set_nextid(const struct lu_env *env,
1048 struct mdd_device *mdd)
1050 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1054 LASSERT(mds->mds_lov_objids != NULL);
1055 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1056 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1057 mds->mds_lov_objids, NULL);
1062 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1063 struct mdd_device *mdd)
1065 /* XXX: to be implemented! */
1070 static int mdd_recovery_complete(const struct lu_env *env,
1071 struct lu_device *d)
1073 struct mdd_device *mdd = lu2mdd_dev(d);
1074 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1075 struct obd_device *obd = mdd2obd_dev(mdd);
1079 LASSERT(mdd != NULL);
1080 LASSERT(obd != NULL);
1082 /* XXX: Do we need this in new stack? */
1083 rc = mdd_lov_set_nextid(env, mdd);
1085 CERROR("mdd_lov_set_nextid() failed %d\n",
1090 /* XXX: cleanup unlink. */
1091 rc = mdd_cleanup_unlink_llog(env, mdd);
1093 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1098 /* Call that with obd_recovering = 1 just to update objids */
1099 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1100 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1102 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1103 cfs_spin_lock(&obd->obd_dev_lock);
1104 obd->obd_recovering = 0;
1105 cfs_spin_unlock(&obd->obd_dev_lock);
1106 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1108 /* XXX: orphans handling. */
1109 __mdd_orphan_cleanup(env, mdd);
1110 rc = next->ld_ops->ldo_recovery_complete(env, next);
1115 static int mdd_prepare(const struct lu_env *env,
1116 struct lu_device *pdev,
1117 struct lu_device *cdev)
1119 struct mdd_device *mdd = lu2mdd_dev(cdev);
1120 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1121 struct dt_object *root;
1126 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1130 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1131 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1132 &mdd->mdd_root_fid);
1133 if (!IS_ERR(root)) {
1134 LASSERT(root != NULL);
1135 lu_object_put(env, &root->do_lu);
1136 rc = orph_index_init(env, mdd);
1143 rc = mdd_dot_lustre_setup(env, mdd);
1145 CERROR("Error(%d) initializing .lustre objects\n", rc);
1149 /* we use capa file to declare llog changes,
1150 * will be fixed with new llog in 2.3 */
1151 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1153 GOTO(out, rc = PTR_ERR(root));
1155 mdd->mdd_capa = root;
1156 rc = mdd_lfsck_setup(env, mdd);
1164 const struct lu_device_operations mdd_lu_ops = {
1165 .ldo_object_alloc = mdd_object_alloc,
1166 .ldo_process_config = mdd_process_config,
1167 .ldo_recovery_complete = mdd_recovery_complete,
1168 .ldo_prepare = mdd_prepare,
1172 * No permission check is needed.
1174 static int mdd_root_get(const struct lu_env *env,
1175 struct md_device *m, struct lu_fid *f)
1177 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1180 *f = mdd->mdd_root_fid;
1185 * No permission check is needed.
1187 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1188 struct obd_statfs *sfs)
1190 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1195 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1201 * No permission check is needed.
1203 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1204 int *md_size, int *cookie_size)
1206 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1209 *md_size = mdd_lov_mdsize(env, mdd);
1210 *cookie_size = mdd_lov_cookiesize(env, mdd);
1215 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1216 int mode, unsigned long timeout, __u32 alg,
1217 struct lustre_capa_key *keys)
1219 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1220 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1224 /* need barrier for mds_capa_keys access. */
1225 cfs_down_write(&mds->mds_notify_lock);
1226 mds->mds_capa_keys = keys;
1227 cfs_up_write(&mds->mds_notify_lock);
1229 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1230 timeout, alg, keys);
1234 static int mdd_update_capa_key(const struct lu_env *env,
1235 struct md_device *m,
1236 struct lustre_capa_key *key)
1238 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1239 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1240 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1244 rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
1245 KEY_CAPA_KEY, sizeof(info), &info, NULL);
1249 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1252 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1254 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1255 return (*h == NULL ? -ENOENT : 0);
1258 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1259 struct lu_device_type *t,
1260 struct lustre_cfg *lcfg)
1262 struct lu_device *l;
1263 struct mdd_device *m;
1267 l = ERR_PTR(-ENOMEM);
1269 md_device_init(&m->mdd_md_dev, t);
1271 l->ld_ops = &mdd_lu_ops;
1272 m->mdd_md_dev.md_ops = &mdd_ops;
1273 md_upcall_init(&m->mdd_md_dev, NULL);
1279 static struct lu_device *mdd_device_free(const struct lu_env *env,
1280 struct lu_device *lu)
1282 struct mdd_device *m = lu2mdd_dev(lu);
1285 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1286 md_device_fini(&m->mdd_md_dev);
1291 static struct obd_ops mdd_obd_device_ops = {
1292 .o_owner = THIS_MODULE
1295 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1296 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1298 static struct lu_context_key mdd_ucred_key = {
1299 .lct_tags = LCT_SESSION,
1300 .lct_init = mdd_ucred_key_init,
1301 .lct_fini = mdd_ucred_key_fini
1304 struct md_ucred *md_ucred(const struct lu_env *env)
1306 LASSERT(env->le_ses != NULL);
1307 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1309 EXPORT_SYMBOL(md_ucred);
1312 * context key constructor/destructor:
1313 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1315 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1317 struct lu_context_key mdd_capainfo_key = {
1318 .lct_tags = LCT_SESSION,
1319 .lct_init = mdd_capainfo_key_init,
1320 .lct_fini = mdd_capainfo_key_fini
1323 struct md_capainfo *md_capainfo(const struct lu_env *env)
1325 /* NB, in mdt_init0 */
1326 if (env->le_ses == NULL)
1328 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1330 EXPORT_SYMBOL(md_capainfo);
1333 * context key constructor/destructor:
1334 * mdd_quota_key_init, mdd_quota_key_fini
1336 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1338 struct lu_context_key mdd_quota_key = {
1339 .lct_tags = LCT_SESSION,
1340 .lct_init = mdd_quota_key_init,
1341 .lct_fini = mdd_quota_key_fini
1344 struct md_quota *md_quota(const struct lu_env *env)
1346 LASSERT(env->le_ses != NULL);
1347 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1349 EXPORT_SYMBOL(md_quota);
1351 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1353 struct llog_ctxt *ctxt;
1354 struct llog_changelog_user_rec *rec;
1358 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1364 llog_ctxt_put(ctxt);
1368 /* Assume we want it on since somebody registered */
1369 rc = mdd_changelog_on(mdd, 1);
1373 rec->cur_hdr.lrh_len = sizeof(*rec);
1374 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1375 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1376 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1377 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1378 CERROR("Maximum number of changelog users exceeded!\n");
1379 GOTO(out, rc = -EOVERFLOW);
1381 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1382 rec->cur_endrec = mdd->mdd_cl.mc_index;
1383 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1385 rc = llog_add(NULL, ctxt, &rec->cur_hdr, NULL, NULL, 0);
1387 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1390 llog_ctxt_put(ctxt);
1394 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1395 struct thandle *handle)
1400 /* XXX: this is a temporary solution to declare llog changes
1401 * will be fixed in 2.3 with new llog implementation */
1403 LASSERT(mdd->mdd_capa);
1405 /* the llog record could be canceled either by modifying
1406 * the plain llog's header or by destroying the llog itself
1407 * when this record is the last one in it, it can't be known
1408 * here, but the catlog's header will also be modified for
1409 * the second case, then the first case can be covered and
1410 * is no need to declare it */
1412 /* destroy empty plain log */
1413 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1417 /* record the catlog's header if an empty plain log was destroyed */
1418 rc = dt_declare_record_write(env, mdd->mdd_capa,
1419 sizeof(struct llog_logid_rec), 0, handle);
1423 struct mdd_changelog_user_data {
1424 __u64 mcud_endrec; /**< purge record for this user */
1425 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1427 __u32 mcud_minid; /**< user id with lowest rec reference */
1428 __u32 mcud_usercount;
1430 struct mdd_device *mcud_mdd;
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(const struct lu_env *env,
1439 struct llog_handle *llh,
1440 struct llog_rec_hdr *hdr, void *data)
1442 struct llog_changelog_user_rec *rec;
1443 struct mdd_changelog_user_data *mcud = data;
1448 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1450 rec = (struct llog_changelog_user_rec *)hdr;
1452 mcud->mcud_usercount++;
1454 /* If we have a new endrec for this id, use it for the following
1455 min check instead of its old value */
1456 if (rec->cur_id == mcud->mcud_id)
1457 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1459 /* Track the minimum referenced record */
1460 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1461 mcud->mcud_minid = rec->cur_id;
1462 mcud->mcud_minrec = rec->cur_endrec;
1465 if (rec->cur_id != mcud->mcud_id)
1468 /* Update this user's record */
1469 mcud->mcud_found = 1;
1471 /* Special case: unregister this user */
1472 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1473 struct llog_cookie cookie;
1475 struct mdd_device *mdd = mcud->mcud_mdd;
1477 cookie.lgc_lgl = llh->lgh_id;
1478 cookie.lgc_index = hdr->lrh_index;
1480 /* XXX This is a workaround for the deadlock of changelog
1481 * adding vs. changelog cancelling. LU-81. */
1482 th = mdd_trans_create(env, mdd);
1484 CERROR("Cannot get thandle\n");
1488 rc = mdd_declare_llog_cancel(env, mdd, th);
1492 rc = mdd_trans_start(env, mdd, th);
1496 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1499 mcud->mcud_usercount--;
1502 mdd_trans_stop(env, mdd, 0, th);
1506 /* Update the endrec */
1507 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1508 mcud->mcud_id, rec->cur_endrec);
1510 /* hdr+1 is loc of data */
1511 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1512 rc = llog_write_rec(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1518 static int mdd_changelog_user_purge(const struct lu_env *env,
1519 struct mdd_device *mdd, int id,
1522 struct mdd_changelog_user_data data;
1523 struct llog_ctxt *ctxt;
1527 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1530 data.mcud_minid = 0;
1531 data.mcud_minrec = 0;
1532 data.mcud_usercount = 0;
1533 data.mcud_endrec = endrec;
1534 data.mcud_mdd = mdd;
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(env, ctxt->loc_handle,
1549 mdd_changelog_user_purge_cb, (void *)&data,
1551 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1552 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1553 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1554 data.mcud_minrec, data.mcud_minid);
1555 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1557 CWARN("Could not determine changelog records to purge; rc=%d\n",
1561 llog_ctxt_put(ctxt);
1563 if (!data.mcud_found) {
1564 CWARN("No entry for user %d. Last changelog reference is "
1565 LPD64" by changelog user %d\n", data.mcud_id,
1566 data.mcud_minrec, data.mcud_minid);
1570 if (!rc && data.mcud_usercount == 0)
1571 /* No more users; turn changelogs off */
1572 rc = mdd_changelog_on(mdd, 0);
1578 * May be called remotely from mdt_iocontrol_handle or locally from
1579 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1580 * an obd_ioctl_data format (but local ioctl handler does).
1582 * \param len - data len
1583 * \param karg - ioctl data, in kernel space
1585 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1586 unsigned int cmd, int len, void *karg)
1588 struct mdd_device *mdd;
1589 struct obd_ioctl_data *data = karg;
1593 mdd = lu2mdd_dev(&m->md_lu_dev);
1595 /* Doesn't use obd_ioctl_data */
1597 case OBD_IOC_CHANGELOG_CLEAR: {
1598 struct changelog_setinfo *cs = karg;
1599 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1603 case OBD_IOC_GET_MNTOPT: {
1604 mntopt_t *mntopts = (mntopt_t *)karg;
1605 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1608 case OBD_IOC_START_LFSCK: {
1609 struct lfsck_start *start = karg;
1610 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1612 /* Return the kernel service version. */
1613 /* XXX: version can be used for compatibility in the future. */
1614 start->ls_version = lfsck->ml_version;
1615 rc = mdd_lfsck_start(env, lfsck, start);
1618 case OBD_IOC_STOP_LFSCK: {
1619 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1624 /* Below ioctls use obd_ioctl_data */
1625 if (len != sizeof(*data)) {
1626 CERROR("Bad ioctl size %d\n", len);
1629 if (data->ioc_version != OBD_IOCTL_VERSION) {
1630 CERROR("Bad magic %x != %x\n", data->ioc_version,
1636 case OBD_IOC_CHANGELOG_REG:
1637 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1639 case OBD_IOC_CHANGELOG_DEREG:
1640 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1650 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1651 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1654 const struct md_device_operations mdd_ops = {
1655 .mdo_statfs = mdd_statfs,
1656 .mdo_root_get = mdd_root_get,
1657 .mdo_maxsize_get = mdd_maxsize_get,
1658 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1659 .mdo_update_capa_key= mdd_update_capa_key,
1660 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1661 .mdo_iocontrol = mdd_iocontrol,
1662 #ifdef HAVE_QUOTA_SUPPORT
1664 .mqo_notify = mdd_quota_notify,
1665 .mqo_setup = mdd_quota_setup,
1666 .mqo_cleanup = mdd_quota_cleanup,
1667 .mqo_recovery = mdd_quota_recovery,
1668 .mqo_check = mdd_quota_check,
1669 .mqo_on = mdd_quota_on,
1670 .mqo_off = mdd_quota_off,
1671 .mqo_setinfo = mdd_quota_setinfo,
1672 .mqo_getinfo = mdd_quota_getinfo,
1673 .mqo_setquota = mdd_quota_setquota,
1674 .mqo_getquota = mdd_quota_getquota,
1675 .mqo_getoinfo = mdd_quota_getoinfo,
1676 .mqo_getoquota = mdd_quota_getoquota,
1677 .mqo_invalidate = mdd_quota_invalidate,
1678 .mqo_finvalidate = mdd_quota_finvalidate
1683 static struct lu_device_type_operations mdd_device_type_ops = {
1684 .ldto_init = mdd_type_init,
1685 .ldto_fini = mdd_type_fini,
1687 .ldto_start = mdd_type_start,
1688 .ldto_stop = mdd_type_stop,
1690 .ldto_device_alloc = mdd_device_alloc,
1691 .ldto_device_free = mdd_device_free,
1693 .ldto_device_init = mdd_device_init,
1694 .ldto_device_fini = mdd_device_fini
1697 static struct lu_device_type mdd_device_type = {
1698 .ldt_tags = LU_DEVICE_MD,
1699 .ldt_name = LUSTRE_MDD_NAME,
1700 .ldt_ops = &mdd_device_type_ops,
1701 .ldt_ctx_tags = LCT_MD_THREAD
1704 /* context key constructor: mdd_key_init */
1705 LU_KEY_INIT(mdd, struct mdd_thread_info);
1707 static void mdd_key_fini(const struct lu_context *ctx,
1708 struct lu_context_key *key, void *data)
1710 struct mdd_thread_info *info = data;
1711 if (info->mti_max_lmm != NULL)
1712 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1713 if (info->mti_max_cookie != NULL)
1714 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1715 mdd_buf_put(&info->mti_big_buf);
1720 /* context key: mdd_thread_key */
1721 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1723 static struct lu_local_obj_desc llod_capa_key = {
1724 .llod_name = CAPA_KEYS,
1725 .llod_oid = MDD_CAPA_KEYS_OID,
1729 static struct lu_local_obj_desc llod_mdd_orphan = {
1730 .llod_name = orph_index_name,
1731 .llod_oid = MDD_ORPHAN_OID,
1733 .llod_feat = &dt_directory_features,
1736 static struct lu_local_obj_desc llod_mdd_root = {
1737 .llod_name = mdd_root_dir_name,
1738 .llod_oid = MDD_ROOT_INDEX_OID,
1740 .llod_feat = &dt_directory_features,
1743 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1744 .llod_name = lfsck_bookmark_name,
1745 .llod_oid = LFSCK_BOOKMARK_OID,
1749 static int __init mdd_mod_init(void)
1751 struct lprocfs_static_vars lvars;
1752 lprocfs_mdd_init_vars(&lvars);
1754 llo_local_obj_register(&llod_capa_key);
1755 llo_local_obj_register(&llod_mdd_orphan);
1756 llo_local_obj_register(&llod_mdd_root);
1757 llo_local_obj_register(&llod_lfsck_bookmark);
1759 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1760 LUSTRE_MDD_NAME, &mdd_device_type);
1763 static void __exit mdd_mod_exit(void)
1765 llo_local_obj_unregister(&llod_capa_key);
1766 llo_local_obj_unregister(&llod_mdd_orphan);
1767 llo_local_obj_unregister(&llod_mdd_root);
1768 llo_local_obj_unregister(&llod_lfsck_bookmark);
1770 class_unregister_type(LUSTRE_MDD_NAME);
1773 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1774 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1775 MODULE_LICENSE("GPL");
1777 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);