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(const struct lu_env *env, struct llog_handle *llh,
135 struct llog_rec_hdr *hdr, void *data)
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(const struct lu_env *env,
155 struct llog_handle *llh,
156 struct llog_rec_hdr *hdr, void *data)
158 struct mdd_device *mdd = (struct mdd_device *)data;
159 struct llog_changelog_user_rec *rec =
160 (struct llog_changelog_user_rec *)hdr;
163 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
164 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
166 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
167 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
168 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
170 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
171 mdd->mdd_cl.mc_lastuser = rec->cur_id;
172 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
174 RETURN(LLOG_PROC_BREAK);
178 static int mdd_changelog_llog_init(struct mdd_device *mdd)
180 struct obd_device *obd = mdd2obd_dev(mdd);
181 struct llog_ctxt *ctxt;
184 /* Find last changelog entry number */
185 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
187 CERROR("no changelog context\n");
190 if (!ctxt->loc_handle) {
195 rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
196 changelog_init_cb, mdd);
200 CERROR("changelog init failed: %d\n", rc);
203 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
204 mdd->mdd_cl.mc_index);
206 /* Find last changelog user id */
207 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
209 CERROR("no changelog user context\n");
212 if (!ctxt->loc_handle) {
217 rc = llog_cat_reverse_process(NULL, ctxt->loc_handle,
218 changelog_user_init_cb, mdd);
222 CERROR("changelog user init failed: %d\n", rc);
226 /* If we have registered users, assume we want changelogs on */
227 if (mdd->mdd_cl.mc_lastuser > 0)
228 rc = mdd_changelog_on(mdd, 1);
233 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
237 mdd->mdd_cl.mc_index = 0;
238 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
239 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
240 mdd->mdd_cl.mc_flags = 0; /* off by default */
241 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
242 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
243 mdd->mdd_cl.mc_lastuser = 0;
245 rc = mdd_changelog_llog_init(mdd);
247 CERROR("Changelog setup during init failed %d\n", rc);
248 mdd->mdd_cl.mc_flags |= CLM_ERR;
254 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
256 mdd->mdd_cl.mc_flags = 0;
259 /* Start / stop recording */
260 int mdd_changelog_on(struct mdd_device *mdd, int on)
264 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
265 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
266 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
267 CERROR("Changelogs cannot be enabled due to error "
268 "condition (see %s log).\n",
269 mdd2obd_dev(mdd)->obd_name);
272 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
273 mdd->mdd_cl.mc_flags |= CLM_ON;
274 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
275 rc = mdd_changelog_write_header(mdd, CLM_START);
277 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
278 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
279 rc = mdd_changelog_write_header(mdd, CLM_FINI);
280 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
281 mdd->mdd_cl.mc_flags &= ~CLM_ON;
282 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
287 static __u64 cl_time(void) {
290 cfs_fs_time_current(&time);
291 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
294 /** Add a changelog entry \a rec to the changelog llog
297 * \param handle - currently ignored since llogs start their own transaction;
298 * this will hopefully be fixed in llog rewrite
301 int mdd_changelog_llog_write(struct mdd_device *mdd,
302 struct llog_changelog_rec *rec,
303 struct thandle *handle)
305 struct obd_device *obd = mdd2obd_dev(mdd);
306 struct llog_ctxt *ctxt;
309 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
310 /* llog_lvfs_write_rec sets the llog tail len */
311 rec->cr_hdr.lrh_type = CHANGELOG_REC;
312 rec->cr.cr_time = cl_time();
313 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
314 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
315 but as long as the MDD transactions are ordered correctly for e.g.
316 rename conflicts, I don't think this should matter. */
317 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
318 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
319 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
323 /* nested journal transaction */
324 rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
330 /** Add a changelog_ext entry \a rec to the changelog llog
333 * \param handle - currently ignored since llogs start their own transaction;
334 * this will hopefully be fixed in llog rewrite
337 int mdd_changelog_ext_llog_write(struct mdd_device *mdd,
338 struct llog_changelog_ext_rec *rec,
339 struct thandle *handle)
341 struct obd_device *obd = mdd2obd_dev(mdd);
342 struct llog_ctxt *ctxt;
345 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
346 /* llog_lvfs_write_rec sets the llog tail len */
347 rec->cr_hdr.lrh_type = CHANGELOG_REC;
348 rec->cr.cr_time = cl_time();
349 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
350 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
351 * but as long as the MDD transactions are ordered correctly for e.g.
352 * rename conflicts, I don't think this should matter. */
353 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
354 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
355 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
359 /* nested journal transaction */
360 rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0);
366 /** Remove entries with indicies up to and including \a endrec from the
372 int mdd_changelog_llog_cancel(const struct lu_env *env,
373 struct mdd_device *mdd, long long endrec)
375 struct obd_device *obd = mdd2obd_dev(mdd);
376 struct llog_ctxt *ctxt;
377 long long unsigned cur;
380 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
384 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
385 cur = (long long)mdd->mdd_cl.mc_index;
386 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
390 /* purge to "0" is shorthand for everything */
394 /* If purging all records, write a header entry so we don't have an
395 empty catalog and we're sure to have a valid starting index next
396 time. In case of crash, we just restart with old log so we're
399 /* XXX: transaction is started by llog itself */
400 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
405 /* Some records were purged, so reset repeat-access time (so we
406 record new mtime update records, so users can see a file has been
407 changed since the last purge) */
408 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
410 /* XXX: transaction is started by llog itself */
411 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
417 /** Add a CL_MARK record to the changelog
419 * \param markerflags - CLM_*
422 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
424 struct obd_device *obd = mdd2obd_dev(mdd);
425 struct llog_changelog_rec *rec;
427 int len = strlen(obd->obd_name);
431 reclen = llog_data_len(sizeof(*rec) + len);
432 OBD_ALLOC(rec, reclen);
436 rec->cr.cr_flags = CLF_VERSION;
437 rec->cr.cr_type = CL_MARK;
438 rec->cr.cr_namelen = len;
439 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
440 /* Status and action flags */
441 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
443 /* XXX: transaction is started by llog itself */
444 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
445 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
447 /* assume on or off event; reset repeat-access time */
448 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
450 OBD_FREE(rec, reclen);
455 * Create ".lustre" directory.
457 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
459 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
460 struct md_object *mdo;
463 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
464 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
465 mdd_root_dir_name, dot_lustre_name,
466 fid, &dt_directory_features);
467 /* .lustre dir may be already present */
468 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
470 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
471 dot_lustre_name, PFID(fid), rc);
476 lu_object_put(env, &mdo->mo_lu);
482 static int dot_lustre_mdd_permission(const struct lu_env *env,
483 struct md_object *pobj,
484 struct md_object *cobj,
485 struct md_attr *attr, int mask)
487 if (mask & ~(MAY_READ | MAY_EXEC))
493 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
494 struct md_object *obj, struct lu_buf *buf,
500 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
501 struct md_object *obj, struct lu_buf *buf)
506 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
507 struct md_object *obj,
508 const struct lu_buf *buf, const char *name,
514 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
515 struct md_object *obj,
521 static int dot_lustre_mdd_readlink(const struct lu_env *env,
522 struct md_object *obj, struct lu_buf *buf)
527 static int dot_lustre_mdd_object_create(const struct lu_env *env,
528 struct md_object *obj,
529 const struct md_op_spec *spec,
535 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
536 struct md_object *obj,
537 const struct md_attr *ma)
542 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
543 struct md_object *obj,
549 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
552 struct mdd_object *mdd_obj = md2mdd_obj(obj);
554 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
555 mdd_obj->mod_count++;
556 mdd_write_unlock(env, mdd_obj);
561 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
562 struct md_attr *ma, int mode)
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_object_sync(const struct lu_env *env,
574 struct md_object *obj)
579 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
580 char *path, int pathlen, __u64 *recno, int *linkno)
585 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
586 struct lov_mds_md *lmm, struct ldlm_extent *extent,
587 struct lustre_handle *lockh)
592 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
593 struct lov_mds_md *lmm, struct lustre_handle *lockh)
598 static struct md_object_operations mdd_dot_lustre_obj_ops = {
599 .moo_permission = dot_lustre_mdd_permission,
600 .moo_attr_get = mdd_attr_get,
601 .moo_attr_set = mdd_attr_set,
602 .moo_xattr_get = dot_lustre_mdd_xattr_get,
603 .moo_xattr_list = dot_lustre_mdd_xattr_list,
604 .moo_xattr_set = dot_lustre_mdd_xattr_set,
605 .moo_xattr_del = dot_lustre_mdd_xattr_del,
606 .moo_readpage = mdd_readpage,
607 .moo_readlink = dot_lustre_mdd_readlink,
608 .moo_object_create = dot_lustre_mdd_object_create,
609 .moo_ref_add = dot_lustre_mdd_ref_add,
610 .moo_ref_del = dot_lustre_mdd_ref_del,
611 .moo_open = dot_lustre_mdd_open,
612 .moo_close = dot_lustre_mdd_close,
613 .moo_capa_get = mdd_capa_get,
614 .moo_object_sync = dot_lustre_mdd_object_sync,
615 .moo_path = dot_lustre_mdd_path,
616 .moo_file_lock = dot_file_lock,
617 .moo_file_unlock = dot_file_unlock,
621 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
622 const struct lu_name *lname, struct lu_fid *f,
623 struct md_op_spec *spec)
625 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
633 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
634 struct md_object *obj,
640 static int dot_lustre_mdd_create(const struct lu_env *env,
641 struct md_object *pobj,
642 const struct lu_name *lname,
643 struct md_object *child,
644 struct md_op_spec *spec,
650 static int dot_lustre_mdd_create_data(const struct lu_env *env,
653 const struct md_op_spec *spec,
659 static int dot_lustre_mdd_rename(const struct lu_env *env,
660 struct md_object *src_pobj,
661 struct md_object *tgt_pobj,
662 const struct lu_fid *lf,
663 const struct lu_name *lsname,
664 struct md_object *tobj,
665 const struct lu_name *ltname,
671 static int dot_lustre_mdd_link(const struct lu_env *env,
672 struct md_object *tgt_obj,
673 struct md_object *src_obj,
674 const struct lu_name *lname,
680 static int dot_lustre_mdd_unlink(const struct lu_env *env,
681 struct md_object *pobj,
682 struct md_object *cobj,
683 const struct lu_name *lname,
689 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
690 struct md_object *obj,
691 const struct lu_name *lname,
692 const struct lu_fid *fid,
693 const struct md_attr *ma)
698 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
699 struct md_object *obj,
700 const struct lu_name *lname,
701 const struct md_attr *ma)
706 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
707 struct md_object *pobj,
708 struct md_object *tobj,
709 const struct lu_fid *fid,
710 const struct lu_name *lname,
717 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
718 .mdo_is_subdir = mdd_is_subdir,
719 .mdo_lookup = dot_lustre_mdd_lookup,
720 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
721 .mdo_create = dot_lustre_mdd_create,
722 .mdo_create_data = dot_lustre_mdd_create_data,
723 .mdo_rename = dot_lustre_mdd_rename,
724 .mdo_link = dot_lustre_mdd_link,
725 .mdo_unlink = dot_lustre_mdd_unlink,
726 .mdo_name_insert = dot_lustre_mdd_name_insert,
727 .mdo_name_remove = dot_lustre_mdd_name_remove,
728 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
731 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
736 if (ma->ma_need & MA_INODE) {
737 struct mdd_device *mdd = mdo2mdd(obj);
739 /* "fid" is a virtual object and hence does not have any "real"
740 * attributes. So we reuse attributes of .lustre for "fid" dir */
741 ma->ma_need |= MA_INODE;
742 rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
745 ma->ma_valid |= MA_INODE;
748 /* "fid" directory does not have any striping information. */
749 if (ma->ma_need & MA_LOV) {
750 struct mdd_object *mdd_obj = md2mdd_obj(obj);
752 if (ma->ma_valid & MA_LOV)
755 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
756 S_ISDIR(mdd_object_type(mdd_obj))))
759 if (ma->ma_need & MA_LOV_DEF) {
760 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
762 ma->ma_lmm_size = rc;
763 ma->ma_valid |= MA_LOV;
772 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
773 const struct md_attr *ma)
778 static int obf_xattr_list(const struct lu_env *env,
779 struct md_object *obj, struct lu_buf *buf)
784 static int obf_xattr_get(const struct lu_env *env,
785 struct md_object *obj, struct lu_buf *buf,
791 static int obf_xattr_set(const struct lu_env *env,
792 struct md_object *obj,
793 const struct lu_buf *buf, const char *name,
799 static int obf_xattr_del(const struct lu_env *env,
800 struct md_object *obj,
806 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
809 struct mdd_object *mdd_obj = md2mdd_obj(obj);
811 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
812 mdd_obj->mod_count++;
813 mdd_write_unlock(env, mdd_obj);
818 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
819 struct md_attr *ma, int mode)
821 struct mdd_object *mdd_obj = md2mdd_obj(obj);
823 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
824 mdd_obj->mod_count--;
825 mdd_write_unlock(env, mdd_obj);
830 /** Nothing to list in "fid" directory */
831 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
832 const struct lu_rdpg *rdpg)
837 static int obf_path(const struct lu_env *env, struct md_object *obj,
838 char *path, int pathlen, __u64 *recno, int *linkno)
843 static struct md_object_operations mdd_obf_obj_ops = {
844 .moo_attr_get = obf_attr_get,
845 .moo_attr_set = obf_attr_set,
846 .moo_xattr_list = obf_xattr_list,
847 .moo_xattr_get = obf_xattr_get,
848 .moo_xattr_set = obf_xattr_set,
849 .moo_xattr_del = obf_xattr_del,
850 .moo_open = obf_mdd_open,
851 .moo_close = obf_mdd_close,
852 .moo_readpage = obf_mdd_readpage,
857 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
858 * are valid. We also check if object with passed fid exists or not.
860 static int obf_lookup(const struct lu_env *env, struct md_object *p,
861 const struct lu_name *lname, struct lu_fid *f,
862 struct md_op_spec *spec)
864 char *name = (char *)lname->ln_name;
865 struct mdd_device *mdd = mdo2mdd(p);
866 struct mdd_object *child;
872 sscanf(name, SFID, RFID(f));
873 if (!fid_is_sane(f)) {
874 CWARN("%s: bad FID format [%s], should be "DFID"\n",
875 mdd->mdd_obd_dev->obd_name, lname->ln_name,
876 (__u64)FID_SEQ_NORMAL, 1, 0);
877 GOTO(out, rc = -EINVAL);
880 if (!fid_is_norm(f)) {
881 CWARN("%s: "DFID" is invalid, sequence should be "
882 ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f),
883 (__u64)FID_SEQ_NORMAL);
884 GOTO(out, rc = -EINVAL);
887 /* Check if object with this fid exists */
888 child = mdd_object_find(env, mdd, f);
892 GOTO(out, rc = PTR_ERR(child));
894 if (mdd_object_exists(child) == 0)
897 mdd_object_put(env, child);
903 static int obf_create(const struct lu_env *env, struct md_object *pobj,
904 const struct lu_name *lname, struct md_object *child,
905 struct md_op_spec *spec, struct md_attr* ma)
910 static int obf_rename(const struct lu_env *env,
911 struct md_object *src_pobj, struct md_object *tgt_pobj,
912 const struct lu_fid *lf, const struct lu_name *lsname,
913 struct md_object *tobj, const struct lu_name *ltname,
919 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
920 struct md_object *src_obj, const struct lu_name *lname,
926 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
927 struct md_object *cobj, const struct lu_name *lname,
933 static struct md_dir_operations mdd_obf_dir_ops = {
934 .mdo_lookup = obf_lookup,
935 .mdo_create = obf_create,
936 .mdo_rename = obf_rename,
937 .mdo_link = obf_link,
938 .mdo_unlink = obf_unlink
942 * Create special in-memory "fid" object for open-by-fid.
944 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
946 struct mdd_object *mdd_obf;
947 struct lu_object *obf_lu_obj;
950 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
952 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
953 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
954 GOTO(out, rc = -ENOENT);
956 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
957 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
958 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
959 /* Don't allow objects to be created in "fid" dir */
960 mdd_obf->mod_flags |= IMMUTE_OBJ;
962 obf_lu_obj = mdd2lu_obj(mdd_obf);
963 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
969 /** Setup ".lustre" directory object */
970 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
972 struct dt_object *dt_dot_lustre;
973 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
976 rc = create_dot_lustre_dir(env, m);
980 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
981 dot_lustre_name, fid);
982 if (IS_ERR(dt_dot_lustre)) {
983 rc = PTR_ERR(dt_dot_lustre);
987 /* references are released in mdd_device_shutdown() */
988 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
991 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
992 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
994 rc = mdd_obf_setup(env, m);
996 CERROR("Error initializing \"fid\" object - %d.\n", rc);
1002 static int mdd_process_config(const struct lu_env *env,
1003 struct lu_device *d, struct lustre_cfg *cfg)
1005 struct mdd_device *m = lu2mdd_dev(d);
1006 struct dt_device *dt = m->mdd_child;
1007 struct lu_device *next = &dt->dd_lu_dev;
1011 switch (cfg->lcfg_command) {
1013 struct lprocfs_static_vars lvars;
1015 lprocfs_mdd_init_vars(&lvars);
1016 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1017 if (rc > 0 || rc == -ENOSYS)
1018 /* we don't understand; pass it on */
1019 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1023 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1026 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1028 rc = mdd_init_obd(env, m, cfg);
1030 CERROR("lov init error %d\n", rc);
1034 mdd_changelog_init(env, m);
1037 lu_dev_del_linkage(d->ld_site, d);
1038 mdd_device_shutdown(env, m, cfg);
1040 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1048 static int mdd_lov_set_nextid(const struct lu_env *env,
1049 struct mdd_device *mdd)
1051 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1055 LASSERT(mds->mds_lov_objids != NULL);
1056 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1057 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1058 mds->mds_lov_objids, NULL);
1063 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1064 struct mdd_device *mdd)
1066 /* XXX: to be implemented! */
1071 static int mdd_recovery_complete(const struct lu_env *env,
1072 struct lu_device *d)
1074 struct mdd_device *mdd = lu2mdd_dev(d);
1075 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1076 struct obd_device *obd = mdd2obd_dev(mdd);
1080 LASSERT(mdd != NULL);
1081 LASSERT(obd != NULL);
1083 /* XXX: Do we need this in new stack? */
1084 rc = mdd_lov_set_nextid(env, mdd);
1086 CERROR("mdd_lov_set_nextid() failed %d\n",
1091 /* XXX: cleanup unlink. */
1092 rc = mdd_cleanup_unlink_llog(env, mdd);
1094 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1099 /* Call that with obd_recovering = 1 just to update objids */
1100 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1101 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1103 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1104 cfs_spin_lock(&obd->obd_dev_lock);
1105 obd->obd_recovering = 0;
1106 cfs_spin_unlock(&obd->obd_dev_lock);
1107 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1109 /* XXX: orphans handling. */
1110 __mdd_orphan_cleanup(env, mdd);
1111 rc = next->ld_ops->ldo_recovery_complete(env, next);
1116 static int mdd_prepare(const struct lu_env *env,
1117 struct lu_device *pdev,
1118 struct lu_device *cdev)
1120 struct mdd_device *mdd = lu2mdd_dev(cdev);
1121 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1122 struct dt_object *root;
1127 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1131 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1132 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1133 &mdd->mdd_root_fid);
1134 if (!IS_ERR(root)) {
1135 LASSERT(root != NULL);
1136 lu_object_put(env, &root->do_lu);
1137 rc = orph_index_init(env, mdd);
1144 rc = mdd_dot_lustre_setup(env, mdd);
1146 CERROR("Error(%d) initializing .lustre objects\n", rc);
1150 /* we use capa file to declare llog changes,
1151 * will be fixed with new llog in 2.3 */
1152 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1154 GOTO(out, rc = PTR_ERR(root));
1156 mdd->mdd_capa = root;
1157 rc = mdd_lfsck_setup(env, mdd);
1165 const struct lu_device_operations mdd_lu_ops = {
1166 .ldo_object_alloc = mdd_object_alloc,
1167 .ldo_process_config = mdd_process_config,
1168 .ldo_recovery_complete = mdd_recovery_complete,
1169 .ldo_prepare = mdd_prepare,
1173 * No permission check is needed.
1175 static int mdd_root_get(const struct lu_env *env,
1176 struct md_device *m, struct lu_fid *f)
1178 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1181 *f = mdd->mdd_root_fid;
1186 * No permission check is needed.
1188 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1189 struct obd_statfs *sfs)
1191 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1196 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1202 * No permission check is needed.
1204 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1205 int *md_size, int *cookie_size)
1207 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1210 *md_size = mdd_lov_mdsize(env, mdd);
1211 *cookie_size = mdd_lov_cookiesize(env, mdd);
1216 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1217 int mode, unsigned long timeout, __u32 alg,
1218 struct lustre_capa_key *keys)
1220 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1221 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1225 /* need barrier for mds_capa_keys access. */
1226 cfs_down_write(&mds->mds_notify_lock);
1227 mds->mds_capa_keys = keys;
1228 cfs_up_write(&mds->mds_notify_lock);
1230 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1231 timeout, alg, keys);
1235 static int mdd_update_capa_key(const struct lu_env *env,
1236 struct md_device *m,
1237 struct lustre_capa_key *key)
1239 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1240 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1241 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1245 rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY),
1246 KEY_CAPA_KEY, sizeof(info), &info, NULL);
1250 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1253 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1255 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1256 return (*h == NULL ? -ENOENT : 0);
1259 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1260 struct lu_device_type *t,
1261 struct lustre_cfg *lcfg)
1263 struct lu_device *l;
1264 struct mdd_device *m;
1268 l = ERR_PTR(-ENOMEM);
1270 md_device_init(&m->mdd_md_dev, t);
1272 l->ld_ops = &mdd_lu_ops;
1273 m->mdd_md_dev.md_ops = &mdd_ops;
1274 md_upcall_init(&m->mdd_md_dev, NULL);
1280 static struct lu_device *mdd_device_free(const struct lu_env *env,
1281 struct lu_device *lu)
1283 struct mdd_device *m = lu2mdd_dev(lu);
1284 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1287 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1288 md_device_fini(&m->mdd_md_dev);
1293 static struct obd_ops mdd_obd_device_ops = {
1294 .o_owner = THIS_MODULE
1297 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1298 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1300 static struct lu_context_key mdd_ucred_key = {
1301 .lct_tags = LCT_SESSION,
1302 .lct_init = mdd_ucred_key_init,
1303 .lct_fini = mdd_ucred_key_fini
1306 struct md_ucred *md_ucred(const struct lu_env *env)
1308 LASSERT(env->le_ses != NULL);
1309 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1311 EXPORT_SYMBOL(md_ucred);
1314 * context key constructor/destructor:
1315 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1317 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1319 struct lu_context_key mdd_capainfo_key = {
1320 .lct_tags = LCT_SESSION,
1321 .lct_init = mdd_capainfo_key_init,
1322 .lct_fini = mdd_capainfo_key_fini
1325 struct md_capainfo *md_capainfo(const struct lu_env *env)
1327 /* NB, in mdt_init0 */
1328 if (env->le_ses == NULL)
1330 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1332 EXPORT_SYMBOL(md_capainfo);
1335 * context key constructor/destructor:
1336 * mdd_quota_key_init, mdd_quota_key_fini
1338 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1340 struct lu_context_key mdd_quota_key = {
1341 .lct_tags = LCT_SESSION,
1342 .lct_init = mdd_quota_key_init,
1343 .lct_fini = mdd_quota_key_fini
1346 struct md_quota *md_quota(const struct lu_env *env)
1348 LASSERT(env->le_ses != NULL);
1349 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1351 EXPORT_SYMBOL(md_quota);
1353 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1355 struct llog_ctxt *ctxt;
1356 struct llog_changelog_user_rec *rec;
1360 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1366 llog_ctxt_put(ctxt);
1370 /* Assume we want it on since somebody registered */
1371 rc = mdd_changelog_on(mdd, 1);
1375 rec->cur_hdr.lrh_len = sizeof(*rec);
1376 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1377 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1378 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1379 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1380 CERROR("Maximum number of changelog users exceeded!\n");
1381 GOTO(out, rc = -EOVERFLOW);
1383 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1384 rec->cur_endrec = mdd->mdd_cl.mc_index;
1385 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1387 rc = llog_add(NULL, ctxt, &rec->cur_hdr, NULL, NULL, 0);
1389 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1392 llog_ctxt_put(ctxt);
1396 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1397 struct thandle *handle)
1402 /* XXX: this is a temporary solution to declare llog changes
1403 * will be fixed in 2.3 with new llog implementation */
1405 LASSERT(mdd->mdd_capa);
1407 /* the llog record could be canceled either by modifying
1408 * the plain llog's header or by destroying the llog itself
1409 * when this record is the last one in it, it can't be known
1410 * here, but the catlog's header will also be modified for
1411 * the second case, then the first case can be covered and
1412 * is no need to declare it */
1414 /* destroy empty plain log */
1415 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1419 /* record the catlog's header if an empty plain log was destroyed */
1420 rc = dt_declare_record_write(env, mdd->mdd_capa,
1421 sizeof(struct llog_logid_rec), 0, handle);
1425 struct mdd_changelog_user_data {
1426 __u64 mcud_endrec; /**< purge record for this user */
1427 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1429 __u32 mcud_minid; /**< user id with lowest rec reference */
1430 __u32 mcud_usercount;
1432 struct mdd_device *mcud_mdd;
1434 #define MCUD_UNREGISTER -1LL
1437 * 1. Find the smallest record everyone is willing to purge
1438 * 2. Update the last purgeable record for this user
1440 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1441 struct llog_handle *llh,
1442 struct llog_rec_hdr *hdr, void *data)
1444 struct llog_changelog_user_rec *rec;
1445 struct mdd_changelog_user_data *mcud = data;
1450 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1452 rec = (struct llog_changelog_user_rec *)hdr;
1454 mcud->mcud_usercount++;
1456 /* If we have a new endrec for this id, use it for the following
1457 min check instead of its old value */
1458 if (rec->cur_id == mcud->mcud_id)
1459 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1461 /* Track the minimum referenced record */
1462 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1463 mcud->mcud_minid = rec->cur_id;
1464 mcud->mcud_minrec = rec->cur_endrec;
1467 if (rec->cur_id != mcud->mcud_id)
1470 /* Update this user's record */
1471 mcud->mcud_found = 1;
1473 /* Special case: unregister this user */
1474 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1475 struct llog_cookie cookie;
1477 struct mdd_device *mdd = mcud->mcud_mdd;
1479 cookie.lgc_lgl = llh->lgh_id;
1480 cookie.lgc_index = hdr->lrh_index;
1482 /* XXX This is a workaround for the deadlock of changelog
1483 * adding vs. changelog cancelling. LU-81. */
1484 th = mdd_trans_create(env, mdd);
1486 CERROR("Cannot get thandle\n");
1490 rc = mdd_declare_llog_cancel(env, mdd, th);
1494 rc = mdd_trans_start(env, mdd, th);
1498 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1501 mcud->mcud_usercount--;
1504 mdd_trans_stop(env, mdd, 0, th);
1508 /* Update the endrec */
1509 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1510 mcud->mcud_id, rec->cur_endrec);
1512 /* hdr+1 is loc of data */
1513 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1514 rc = llog_write_rec(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1520 static int mdd_changelog_user_purge(const struct lu_env *env,
1521 struct mdd_device *mdd, int id,
1524 struct mdd_changelog_user_data data;
1525 struct llog_ctxt *ctxt;
1529 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1532 data.mcud_minid = 0;
1533 data.mcud_minrec = 0;
1534 data.mcud_usercount = 0;
1535 data.mcud_endrec = endrec;
1536 data.mcud_mdd = mdd;
1537 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1538 endrec = mdd->mdd_cl.mc_index;
1539 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1540 if ((data.mcud_endrec == 0) ||
1541 ((data.mcud_endrec > endrec) &&
1542 (data.mcud_endrec != MCUD_UNREGISTER)))
1543 data.mcud_endrec = endrec;
1545 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1548 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1550 rc = llog_cat_process(env, ctxt->loc_handle,
1551 mdd_changelog_user_purge_cb, (void *)&data,
1553 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1554 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1555 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1556 data.mcud_minrec, data.mcud_minid);
1557 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1559 CWARN("Could not determine changelog records to purge; rc=%d\n",
1563 llog_ctxt_put(ctxt);
1565 if (!data.mcud_found) {
1566 CWARN("No entry for user %d. Last changelog reference is "
1567 LPD64" by changelog user %d\n", data.mcud_id,
1568 data.mcud_minrec, data.mcud_minid);
1572 if (!rc && data.mcud_usercount == 0)
1573 /* No more users; turn changelogs off */
1574 rc = mdd_changelog_on(mdd, 0);
1580 * May be called remotely from mdt_iocontrol_handle or locally from
1581 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1582 * an obd_ioctl_data format (but local ioctl handler does).
1584 * \param len - data len
1585 * \param karg - ioctl data, in kernel space
1587 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1588 unsigned int cmd, int len, void *karg)
1590 struct mdd_device *mdd;
1591 struct obd_ioctl_data *data = karg;
1595 mdd = lu2mdd_dev(&m->md_lu_dev);
1597 /* Doesn't use obd_ioctl_data */
1599 case OBD_IOC_CHANGELOG_CLEAR: {
1600 struct changelog_setinfo *cs = karg;
1601 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1605 case OBD_IOC_GET_MNTOPT: {
1606 mntopt_t *mntopts = (mntopt_t *)karg;
1607 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1610 case OBD_IOC_START_LFSCK: {
1611 struct lfsck_start *start = karg;
1612 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1614 /* Return the kernel service version. */
1615 /* XXX: version can be used for compatibility in the future. */
1616 start->ls_version = lfsck->ml_version;
1617 rc = mdd_lfsck_start(env, lfsck, start);
1620 case OBD_IOC_STOP_LFSCK: {
1621 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1626 /* Below ioctls use obd_ioctl_data */
1627 if (len != sizeof(*data)) {
1628 CERROR("Bad ioctl size %d\n", len);
1631 if (data->ioc_version != OBD_IOCTL_VERSION) {
1632 CERROR("Bad magic %x != %x\n", data->ioc_version,
1638 case OBD_IOC_CHANGELOG_REG:
1639 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1641 case OBD_IOC_CHANGELOG_DEREG:
1642 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1652 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1653 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1656 const struct md_device_operations mdd_ops = {
1657 .mdo_statfs = mdd_statfs,
1658 .mdo_root_get = mdd_root_get,
1659 .mdo_maxsize_get = mdd_maxsize_get,
1660 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1661 .mdo_update_capa_key= mdd_update_capa_key,
1662 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1663 .mdo_iocontrol = mdd_iocontrol,
1664 #ifdef HAVE_QUOTA_SUPPORT
1666 .mqo_notify = mdd_quota_notify,
1667 .mqo_setup = mdd_quota_setup,
1668 .mqo_cleanup = mdd_quota_cleanup,
1669 .mqo_recovery = mdd_quota_recovery,
1670 .mqo_check = mdd_quota_check,
1671 .mqo_on = mdd_quota_on,
1672 .mqo_off = mdd_quota_off,
1673 .mqo_setinfo = mdd_quota_setinfo,
1674 .mqo_getinfo = mdd_quota_getinfo,
1675 .mqo_setquota = mdd_quota_setquota,
1676 .mqo_getquota = mdd_quota_getquota,
1677 .mqo_getoinfo = mdd_quota_getoinfo,
1678 .mqo_getoquota = mdd_quota_getoquota,
1679 .mqo_invalidate = mdd_quota_invalidate,
1680 .mqo_finvalidate = mdd_quota_finvalidate
1685 static struct lu_device_type_operations mdd_device_type_ops = {
1686 .ldto_init = mdd_type_init,
1687 .ldto_fini = mdd_type_fini,
1689 .ldto_start = mdd_type_start,
1690 .ldto_stop = mdd_type_stop,
1692 .ldto_device_alloc = mdd_device_alloc,
1693 .ldto_device_free = mdd_device_free,
1695 .ldto_device_init = mdd_device_init,
1696 .ldto_device_fini = mdd_device_fini
1699 static struct lu_device_type mdd_device_type = {
1700 .ldt_tags = LU_DEVICE_MD,
1701 .ldt_name = LUSTRE_MDD_NAME,
1702 .ldt_ops = &mdd_device_type_ops,
1703 .ldt_ctx_tags = LCT_MD_THREAD
1706 /* context key constructor: mdd_key_init */
1707 LU_KEY_INIT(mdd, struct mdd_thread_info);
1709 static void mdd_key_fini(const struct lu_context *ctx,
1710 struct lu_context_key *key, void *data)
1712 struct mdd_thread_info *info = data;
1713 if (info->mti_max_lmm != NULL)
1714 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1715 if (info->mti_max_cookie != NULL)
1716 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1717 mdd_buf_put(&info->mti_big_buf);
1722 /* context key: mdd_thread_key */
1723 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1725 static struct lu_local_obj_desc llod_capa_key = {
1726 .llod_name = CAPA_KEYS,
1727 .llod_oid = MDD_CAPA_KEYS_OID,
1731 static struct lu_local_obj_desc llod_mdd_orphan = {
1732 .llod_name = orph_index_name,
1733 .llod_oid = MDD_ORPHAN_OID,
1735 .llod_feat = &dt_directory_features,
1738 static struct lu_local_obj_desc llod_mdd_root = {
1739 .llod_name = mdd_root_dir_name,
1740 .llod_oid = MDD_ROOT_INDEX_OID,
1742 .llod_feat = &dt_directory_features,
1745 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1746 .llod_name = lfsck_bookmark_name,
1747 .llod_oid = LFSCK_BOOKMARK_OID,
1751 static int __init mdd_mod_init(void)
1753 struct lprocfs_static_vars lvars;
1754 lprocfs_mdd_init_vars(&lvars);
1756 llo_local_obj_register(&llod_capa_key);
1757 llo_local_obj_register(&llod_mdd_orphan);
1758 llo_local_obj_register(&llod_mdd_root);
1759 llo_local_obj_register(&llod_lfsck_bookmark);
1761 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1762 LUSTRE_MDD_NAME, &mdd_device_type);
1765 static void __exit mdd_mod_exit(void)
1767 llo_local_obj_unregister(&llod_capa_key);
1768 llo_local_obj_unregister(&llod_mdd_orphan);
1769 llo_local_obj_unregister(&llod_mdd_root);
1770 llo_local_obj_unregister(&llod_lfsck_bookmark);
1772 class_unregister_type(LUSTRE_MDD_NAME);
1775 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1776 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1777 MODULE_LICENSE("GPL");
1779 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);