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>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #ifdef HAVE_EXT4_LDISKFS
50 #include <ldiskfs/ldiskfs_jbd2.h>
52 #include <linux/jbd.h>
55 #include <obd_class.h>
56 #include <lustre_ver.h>
57 #include <obd_support.h>
58 #include <lprocfs_status.h>
60 #include <lustre_disk.h>
61 #include <lustre_fid.h>
62 #ifdef HAVE_EXT4_LDISKFS
63 #include <ldiskfs/ldiskfs.h>
65 #include <linux/ldiskfs_fs.h>
67 #include <lustre_mds.h>
68 #include <lustre/lustre_idl.h>
69 #include <lustre_disk.h> /* for changelogs */
70 #include <lustre_param.h>
71 #include <lustre_fid.h>
73 #include "mdd_internal.h"
75 const struct md_device_operations mdd_ops;
76 static struct lu_device_type mdd_device_type;
78 static const char mdd_root_dir_name[] = "ROOT";
79 static const char mdd_obf_dir_name[] = "fid";
81 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
82 const char *name, struct lu_device *next)
84 struct mdd_device *mdd = lu2mdd_dev(d);
88 mdd->mdd_child = lu2dt_dev(next);
90 /* Prepare transactions callbacks. */
91 mdd->mdd_txn_cb.dtc_txn_start = NULL;
92 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
93 mdd->mdd_txn_cb.dtc_txn_commit = NULL;
94 mdd->mdd_txn_cb.dtc_cookie = mdd;
95 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
96 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
97 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
98 /* sync permission changes */
99 mdd->mdd_sync_permission = 1;
101 rc = mdd_procfs_init(mdd, name);
105 static struct lu_device *mdd_device_fini(const struct lu_env *env,
108 struct mdd_device *mdd = lu2mdd_dev(d);
109 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
112 rc = mdd_procfs_fini(mdd);
114 CERROR("proc fini error %d \n", rc);
120 static void mdd_changelog_fini(const struct lu_env *env,
121 struct mdd_device *mdd);
123 static void mdd_device_shutdown(const struct lu_env *env,
124 struct mdd_device *m, struct lustre_cfg *cfg)
127 mdd_changelog_fini(env, m);
128 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
129 if (m->mdd_dot_lustre_objs.mdd_obf)
130 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
131 if (m->mdd_dot_lustre)
132 mdd_object_put(env, m->mdd_dot_lustre);
134 mdd_fini_obd(env, m, cfg);
135 orph_index_fini(env, m);
136 if (m->mdd_capa != NULL) {
137 lu_object_put(env, &m->mdd_capa->do_lu);
140 /* remove upcall device*/
141 md_upcall_fini(&m->mdd_md_dev);
145 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
148 struct mdd_device *mdd = (struct mdd_device *)data;
149 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
152 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
153 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
156 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
157 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
158 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
159 llh->lgh_id.lgl_oid);
161 mdd->mdd_cl.mc_index = rec->cr.cr_index;
162 RETURN(LLOG_PROC_BREAK);
165 static int changelog_user_init_cb(struct llog_handle *llh,
166 struct llog_rec_hdr *hdr, void *data)
168 struct mdd_device *mdd = (struct mdd_device *)data;
169 struct llog_changelog_user_rec *rec =
170 (struct llog_changelog_user_rec *)hdr;
173 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
174 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
176 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
177 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
178 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
180 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
181 mdd->mdd_cl.mc_lastuser = rec->cur_id;
182 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
184 RETURN(LLOG_PROC_BREAK);
188 static int mdd_changelog_llog_init(struct mdd_device *mdd)
190 struct obd_device *obd = mdd2obd_dev(mdd);
191 struct llog_ctxt *ctxt;
194 /* Find last changelog entry number */
195 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
197 CERROR("no changelog context\n");
200 if (!ctxt->loc_handle) {
205 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
209 CERROR("changelog init failed: %d\n", rc);
212 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
213 mdd->mdd_cl.mc_index);
215 /* Find last changelog user id */
216 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
218 CERROR("no changelog user context\n");
221 if (!ctxt->loc_handle) {
226 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
231 CERROR("changelog user init failed: %d\n", rc);
235 /* If we have registered users, assume we want changelogs on */
236 if (mdd->mdd_cl.mc_lastuser > 0)
237 rc = mdd_changelog_on(mdd, 1);
242 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
246 mdd->mdd_cl.mc_index = 0;
247 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
248 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
249 mdd->mdd_cl.mc_flags = 0; /* off by default */
250 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
251 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
252 mdd->mdd_cl.mc_lastuser = 0;
254 rc = mdd_changelog_llog_init(mdd);
256 CERROR("Changelog setup during init failed %d\n", rc);
257 mdd->mdd_cl.mc_flags |= CLM_ERR;
263 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
265 mdd->mdd_cl.mc_flags = 0;
268 /* Start / stop recording */
269 int mdd_changelog_on(struct mdd_device *mdd, int on)
273 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
274 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
275 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
276 CERROR("Changelogs cannot be enabled due to error "
277 "condition (see %s log).\n",
278 mdd2obd_dev(mdd)->obd_name);
281 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
282 mdd->mdd_cl.mc_flags |= CLM_ON;
283 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
284 rc = mdd_changelog_write_header(mdd, CLM_START);
286 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
287 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
288 rc = mdd_changelog_write_header(mdd, CLM_FINI);
289 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
290 mdd->mdd_cl.mc_flags &= ~CLM_ON;
291 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
296 static __u64 cl_time(void) {
299 cfs_fs_time_current(&time);
300 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
303 /** Add a changelog entry \a rec to the changelog llog
306 * \param handle - currently ignored since llogs start their own transaction;
307 * this will hopefully be fixed in llog rewrite
310 int mdd_changelog_llog_write(struct mdd_device *mdd,
311 struct llog_changelog_rec *rec,
312 struct thandle *handle)
314 struct obd_device *obd = mdd2obd_dev(mdd);
315 struct llog_ctxt *ctxt;
318 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
319 /* llog_lvfs_write_rec sets the llog tail len */
320 rec->cr_hdr.lrh_type = CHANGELOG_REC;
321 rec->cr.cr_time = cl_time();
322 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
323 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
324 but as long as the MDD transactions are ordered correctly for e.g.
325 rename conflicts, I don't think this should matter. */
326 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
327 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
328 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
332 /* nested journal transaction */
333 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
339 /** Remove entries with indicies up to and including \a endrec from the
345 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
347 struct obd_device *obd = mdd2obd_dev(mdd);
348 struct llog_ctxt *ctxt;
349 long long unsigned cur;
352 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
356 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
357 cur = (long long)mdd->mdd_cl.mc_index;
358 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
362 /* purge to "0" is shorthand for everything */
366 /* If purging all records, write a header entry so we don't have an
367 empty catalog and we're sure to have a valid starting index next
368 time. In case of crash, we just restart with old log so we're
371 /* XXX: transaction is started by llog itself */
372 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
377 /* Some records were purged, so reset repeat-access time (so we
378 record new mtime update records, so users can see a file has been
379 changed since the last purge) */
380 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
382 /* XXX: transaction is started by llog itself */
383 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
389 /** Add a CL_MARK record to the changelog
391 * \param markerflags - CLM_*
394 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
396 struct obd_device *obd = mdd2obd_dev(mdd);
397 struct llog_changelog_rec *rec;
399 int len = strlen(obd->obd_name);
403 reclen = llog_data_len(sizeof(*rec) + len);
404 OBD_ALLOC(rec, reclen);
408 rec->cr.cr_flags = CLF_VERSION;
409 rec->cr.cr_type = CL_MARK;
410 rec->cr.cr_namelen = len;
411 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
412 /* Status and action flags */
413 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
415 /* XXX: transaction is started by llog itself */
416 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
417 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
419 /* assume on or off event; reset repeat-access time */
420 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
422 OBD_FREE(rec, reclen);
427 * Create ".lustre" directory.
429 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
431 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
432 struct md_object *mdo;
435 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
436 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
437 mdd_root_dir_name, dot_lustre_name,
438 fid, &dt_directory_features);
439 /* .lustre dir may be already present */
440 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
442 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
443 dot_lustre_name, PFID(fid), rc);
448 lu_object_put(env, &mdo->mo_lu);
454 static int dot_lustre_mdd_permission(const struct lu_env *env,
455 struct md_object *pobj,
456 struct md_object *cobj,
457 struct md_attr *attr, int mask)
459 if (mask & ~(MAY_READ | MAY_EXEC))
465 static int dot_lustre_mdd_attr_get(const struct lu_env *env,
466 struct md_object *obj, struct md_attr *ma)
468 struct mdd_object *mdd_obj = md2mdd_obj(obj);
470 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
473 static int dot_lustre_mdd_attr_set(const struct lu_env *env,
474 struct md_object *obj,
475 const struct md_attr *ma)
480 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
481 struct md_object *obj, struct lu_buf *buf,
487 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
488 struct md_object *obj, struct lu_buf *buf)
493 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
494 struct md_object *obj,
495 const struct lu_buf *buf, const char *name,
501 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
502 struct md_object *obj,
508 static int dot_lustre_mdd_readlink(const struct lu_env *env,
509 struct md_object *obj, struct lu_buf *buf)
514 static int dot_lustre_mdd_object_create(const struct lu_env *env,
515 struct md_object *obj,
516 const struct md_op_spec *spec,
522 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
523 struct md_object *obj,
524 const struct md_attr *ma)
529 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
530 struct md_object *obj,
536 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
539 struct mdd_object *mdd_obj = md2mdd_obj(obj);
541 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
542 mdd_obj->mod_count++;
543 mdd_write_unlock(env, mdd_obj);
548 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
549 struct md_attr *ma, int mode)
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_object_sync(const struct lu_env *env,
561 struct md_object *obj)
566 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
567 char *path, int pathlen, __u64 *recno, int *linkno)
572 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
573 struct lov_mds_md *lmm, struct ldlm_extent *extent,
574 struct lustre_handle *lockh)
579 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
580 struct lov_mds_md *lmm, struct lustre_handle *lockh)
585 static struct md_object_operations mdd_dot_lustre_obj_ops = {
586 .moo_permission = dot_lustre_mdd_permission,
587 .moo_attr_get = dot_lustre_mdd_attr_get,
588 .moo_attr_set = dot_lustre_mdd_attr_set,
589 .moo_xattr_get = dot_lustre_mdd_xattr_get,
590 .moo_xattr_list = dot_lustre_mdd_xattr_list,
591 .moo_xattr_set = dot_lustre_mdd_xattr_set,
592 .moo_xattr_del = dot_lustre_mdd_xattr_del,
593 .moo_readpage = mdd_readpage,
594 .moo_readlink = dot_lustre_mdd_readlink,
595 .moo_object_create = dot_lustre_mdd_object_create,
596 .moo_ref_add = dot_lustre_mdd_ref_add,
597 .moo_ref_del = dot_lustre_mdd_ref_del,
598 .moo_open = dot_lustre_mdd_open,
599 .moo_close = dot_lustre_mdd_close,
600 .moo_capa_get = mdd_capa_get,
601 .moo_object_sync = dot_lustre_mdd_object_sync,
602 .moo_path = dot_lustre_mdd_path,
603 .moo_file_lock = dot_file_lock,
604 .moo_file_unlock = dot_file_unlock,
608 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
609 const struct lu_name *lname, struct lu_fid *f,
610 struct md_op_spec *spec)
612 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
620 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
621 struct md_object *obj,
627 static int dot_lustre_mdd_create(const struct lu_env *env,
628 struct md_object *pobj,
629 const struct lu_name *lname,
630 struct md_object *child,
631 struct md_op_spec *spec,
637 static int dot_lustre_mdd_create_data(const struct lu_env *env,
640 const struct md_op_spec *spec,
646 static int dot_lustre_mdd_rename(const struct lu_env *env,
647 struct md_object *src_pobj,
648 struct md_object *tgt_pobj,
649 const struct lu_fid *lf,
650 const struct lu_name *lsname,
651 struct md_object *tobj,
652 const struct lu_name *ltname,
658 static int dot_lustre_mdd_link(const struct lu_env *env,
659 struct md_object *tgt_obj,
660 struct md_object *src_obj,
661 const struct lu_name *lname,
667 static int dot_lustre_mdd_unlink(const struct lu_env *env,
668 struct md_object *pobj,
669 struct md_object *cobj,
670 const struct lu_name *lname,
676 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
677 struct md_object *obj,
678 const struct lu_name *lname,
679 const struct lu_fid *fid,
680 const struct md_attr *ma)
685 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
686 struct md_object *obj,
687 const struct lu_name *lname,
688 const struct md_attr *ma)
693 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
694 struct md_object *pobj,
695 struct md_object *tobj,
696 const struct lu_fid *fid,
697 const struct lu_name *lname,
704 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
705 .mdo_is_subdir = mdd_is_subdir,
706 .mdo_lookup = dot_lustre_mdd_lookup,
707 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
708 .mdo_create = dot_lustre_mdd_create,
709 .mdo_create_data = dot_lustre_mdd_create_data,
710 .mdo_rename = dot_lustre_mdd_rename,
711 .mdo_link = dot_lustre_mdd_link,
712 .mdo_unlink = dot_lustre_mdd_unlink,
713 .mdo_name_insert = dot_lustre_mdd_name_insert,
714 .mdo_name_remove = dot_lustre_mdd_name_remove,
715 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
718 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
723 if (ma->ma_need & MA_INODE) {
724 struct mdd_device *mdd = mdo2mdd(obj);
726 /* "fid" is a virtual object and hence does not have any "real"
727 * attributes. So we reuse attributes of .lustre for "fid" dir */
728 ma->ma_need |= MA_INODE;
729 rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
733 ma->ma_valid |= MA_INODE;
736 /* "fid" directory does not have any striping information. */
737 if (ma->ma_need & MA_LOV) {
738 struct mdd_object *mdd_obj = md2mdd_obj(obj);
740 if (ma->ma_valid & MA_LOV)
743 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
744 S_ISDIR(mdd_object_type(mdd_obj))))
747 if (ma->ma_need & MA_LOV_DEF) {
748 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
750 ma->ma_lmm_size = rc;
751 ma->ma_valid |= MA_LOV;
760 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
761 const struct md_attr *ma)
766 static int obf_xattr_get(const struct lu_env *env,
767 struct md_object *obj, struct lu_buf *buf,
773 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
776 struct mdd_object *mdd_obj = md2mdd_obj(obj);
778 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
779 mdd_obj->mod_count++;
780 mdd_write_unlock(env, mdd_obj);
785 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
786 struct md_attr *ma, int mode)
788 struct mdd_object *mdd_obj = md2mdd_obj(obj);
790 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
791 mdd_obj->mod_count--;
792 mdd_write_unlock(env, mdd_obj);
797 /** Nothing to list in "fid" directory */
798 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
799 const struct lu_rdpg *rdpg)
804 static int obf_path(const struct lu_env *env, struct md_object *obj,
805 char *path, int pathlen, __u64 *recno, int *linkno)
810 static struct md_object_operations mdd_obf_obj_ops = {
811 .moo_attr_get = obf_attr_get,
812 .moo_attr_set = obf_attr_set,
813 .moo_xattr_get = obf_xattr_get,
814 .moo_open = obf_mdd_open,
815 .moo_close = obf_mdd_close,
816 .moo_readpage = obf_mdd_readpage,
821 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
822 * are valid. We also check if object with passed fid exists or not.
824 static int obf_lookup(const struct lu_env *env, struct md_object *p,
825 const struct lu_name *lname, struct lu_fid *f,
826 struct md_op_spec *spec)
828 char *name = (char *)lname->ln_name;
829 struct mdd_device *mdd = mdo2mdd(p);
830 struct mdd_object *child;
836 sscanf(name, SFID, RFID(f));
837 if (!fid_is_sane(f)) {
838 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
840 GOTO(out, rc = -EINVAL);
843 /* Check if object with this fid exists */
844 child = mdd_object_find(env, mdd, f);
848 GOTO(out, rc = PTR_ERR(child));
850 if (mdd_object_exists(child) == 0)
853 mdd_object_put(env, child);
859 static int obf_create(const struct lu_env *env, struct md_object *pobj,
860 const struct lu_name *lname, struct md_object *child,
861 struct md_op_spec *spec, struct md_attr* ma)
866 static int obf_rename(const struct lu_env *env,
867 struct md_object *src_pobj, struct md_object *tgt_pobj,
868 const struct lu_fid *lf, const struct lu_name *lsname,
869 struct md_object *tobj, const struct lu_name *ltname,
875 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
876 struct md_object *src_obj, const struct lu_name *lname,
882 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
883 struct md_object *cobj, const struct lu_name *lname,
889 static struct md_dir_operations mdd_obf_dir_ops = {
890 .mdo_lookup = obf_lookup,
891 .mdo_create = obf_create,
892 .mdo_rename = obf_rename,
893 .mdo_link = obf_link,
894 .mdo_unlink = obf_unlink
898 * Create special in-memory "fid" object for open-by-fid.
900 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
902 struct mdd_object *mdd_obf;
903 struct lu_object *obf_lu_obj;
906 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
908 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
909 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
910 GOTO(out, rc = -ENOENT);
912 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
913 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
914 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
915 /* Don't allow objects to be created in "fid" dir */
916 mdd_obf->mod_flags |= IMMUTE_OBJ;
918 obf_lu_obj = mdd2lu_obj(mdd_obf);
919 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
925 /** Setup ".lustre" directory object */
926 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
928 struct dt_object *dt_dot_lustre;
929 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
932 rc = create_dot_lustre_dir(env, m);
936 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
937 dot_lustre_name, fid);
938 if (IS_ERR(dt_dot_lustre)) {
939 rc = PTR_ERR(dt_dot_lustre);
943 /* references are released in mdd_device_shutdown() */
944 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
947 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
948 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
950 rc = mdd_obf_setup(env, m);
952 CERROR("Error initializing \"fid\" object - %d.\n", rc);
958 static int mdd_process_config(const struct lu_env *env,
959 struct lu_device *d, struct lustre_cfg *cfg)
961 struct mdd_device *m = lu2mdd_dev(d);
962 struct dt_device *dt = m->mdd_child;
963 struct lu_device *next = &dt->dd_lu_dev;
967 switch (cfg->lcfg_command) {
969 struct lprocfs_static_vars lvars;
971 lprocfs_mdd_init_vars(&lvars);
972 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
973 if (rc > 0 || rc == -ENOSYS)
974 /* we don't understand; pass it on */
975 rc = next->ld_ops->ldo_process_config(env, next, cfg);
979 rc = next->ld_ops->ldo_process_config(env, next, cfg);
982 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
984 rc = mdd_init_obd(env, m, cfg);
986 CERROR("lov init error %d\n", rc);
990 mdd_changelog_init(env, m);
993 mdd_device_shutdown(env, m, cfg);
995 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1003 static int mdd_lov_set_nextid(const struct lu_env *env,
1004 struct mdd_device *mdd)
1006 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1010 LASSERT(mds->mds_lov_objids != NULL);
1011 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1012 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1013 mds->mds_lov_objids, NULL);
1018 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1019 struct mdd_device *mdd)
1021 /* XXX: to be implemented! */
1026 static int mdd_recovery_complete(const struct lu_env *env,
1027 struct lu_device *d)
1029 struct mdd_device *mdd = lu2mdd_dev(d);
1030 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1031 struct obd_device *obd = mdd2obd_dev(mdd);
1035 LASSERT(mdd != NULL);
1036 LASSERT(obd != NULL);
1038 /* XXX: Do we need this in new stack? */
1039 rc = mdd_lov_set_nextid(env, mdd);
1041 CERROR("mdd_lov_set_nextid() failed %d\n",
1046 /* XXX: cleanup unlink. */
1047 rc = mdd_cleanup_unlink_llog(env, mdd);
1049 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1054 /* Call that with obd_recovering = 1 just to update objids */
1055 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1056 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1058 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1059 cfs_spin_lock(&obd->obd_dev_lock);
1060 obd->obd_recovering = 0;
1061 cfs_spin_unlock(&obd->obd_dev_lock);
1062 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1064 /* XXX: orphans handling. */
1065 __mdd_orphan_cleanup(env, mdd);
1066 rc = next->ld_ops->ldo_recovery_complete(env, next);
1071 static int mdd_prepare(const struct lu_env *env,
1072 struct lu_device *pdev,
1073 struct lu_device *cdev)
1075 struct mdd_device *mdd = lu2mdd_dev(cdev);
1076 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1077 struct dt_object *root;
1082 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1086 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1087 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1088 &mdd->mdd_root_fid);
1089 if (!IS_ERR(root)) {
1090 LASSERT(root != NULL);
1091 lu_object_put(env, &root->do_lu);
1092 rc = orph_index_init(env, mdd);
1099 rc = mdd_dot_lustre_setup(env, mdd);
1101 CERROR("Error(%d) initializing .lustre objects\n", rc);
1105 /* we use capa file to declare llog changes,
1106 * will be fixed with new llog in 2.3 */
1107 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1109 mdd->mdd_capa = root;
1117 const struct lu_device_operations mdd_lu_ops = {
1118 .ldo_object_alloc = mdd_object_alloc,
1119 .ldo_process_config = mdd_process_config,
1120 .ldo_recovery_complete = mdd_recovery_complete,
1121 .ldo_prepare = mdd_prepare,
1125 * No permission check is needed.
1127 static int mdd_root_get(const struct lu_env *env,
1128 struct md_device *m, struct lu_fid *f)
1130 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1133 *f = mdd->mdd_root_fid;
1138 * No permission check is needed.
1140 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1141 struct obd_statfs *sfs)
1143 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1148 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1154 * No permission check is needed.
1156 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1157 int *md_size, int *cookie_size)
1159 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1162 *md_size = mdd_lov_mdsize(env, mdd);
1163 *cookie_size = mdd_lov_cookiesize(env, mdd);
1168 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1169 int mode, unsigned long timeout, __u32 alg,
1170 struct lustre_capa_key *keys)
1172 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1173 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1177 /* need barrier for mds_capa_keys access. */
1178 cfs_down_write(&mds->mds_notify_lock);
1179 mds->mds_capa_keys = keys;
1180 cfs_up_write(&mds->mds_notify_lock);
1182 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1183 timeout, alg, keys);
1187 static int mdd_update_capa_key(const struct lu_env *env,
1188 struct md_device *m,
1189 struct lustre_capa_key *key)
1191 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1192 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1193 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1197 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1198 sizeof(info), &info, NULL);
1202 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1205 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1207 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1208 return (*h == NULL ? -ENOENT : 0);
1211 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1212 struct lu_device_type *t,
1213 struct lustre_cfg *lcfg)
1215 struct lu_device *l;
1216 struct mdd_device *m;
1220 l = ERR_PTR(-ENOMEM);
1222 md_device_init(&m->mdd_md_dev, t);
1224 l->ld_ops = &mdd_lu_ops;
1225 m->mdd_md_dev.md_ops = &mdd_ops;
1226 md_upcall_init(&m->mdd_md_dev, NULL);
1232 static struct lu_device *mdd_device_free(const struct lu_env *env,
1233 struct lu_device *lu)
1235 struct mdd_device *m = lu2mdd_dev(lu);
1236 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1239 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1240 md_device_fini(&m->mdd_md_dev);
1245 static struct obd_ops mdd_obd_device_ops = {
1246 .o_owner = THIS_MODULE
1249 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1250 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1252 static struct lu_context_key mdd_ucred_key = {
1253 .lct_tags = LCT_SESSION,
1254 .lct_init = mdd_ucred_key_init,
1255 .lct_fini = mdd_ucred_key_fini
1258 struct md_ucred *md_ucred(const struct lu_env *env)
1260 LASSERT(env->le_ses != NULL);
1261 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1263 EXPORT_SYMBOL(md_ucred);
1266 * context key constructor/destructor:
1267 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1269 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1271 struct lu_context_key mdd_capainfo_key = {
1272 .lct_tags = LCT_SESSION,
1273 .lct_init = mdd_capainfo_key_init,
1274 .lct_fini = mdd_capainfo_key_fini
1277 struct md_capainfo *md_capainfo(const struct lu_env *env)
1279 /* NB, in mdt_init0 */
1280 if (env->le_ses == NULL)
1282 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1284 EXPORT_SYMBOL(md_capainfo);
1287 * context key constructor/destructor:
1288 * mdd_quota_key_init, mdd_quota_key_fini
1290 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1292 struct lu_context_key mdd_quota_key = {
1293 .lct_tags = LCT_SESSION,
1294 .lct_init = mdd_quota_key_init,
1295 .lct_fini = mdd_quota_key_fini
1298 struct md_quota *md_quota(const struct lu_env *env)
1300 LASSERT(env->le_ses != NULL);
1301 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1303 EXPORT_SYMBOL(md_quota);
1305 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1307 struct llog_ctxt *ctxt;
1308 struct llog_changelog_user_rec *rec;
1312 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1318 llog_ctxt_put(ctxt);
1322 /* Assume we want it on since somebody registered */
1323 rc = mdd_changelog_on(mdd, 1);
1327 rec->cur_hdr.lrh_len = sizeof(*rec);
1328 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1329 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1330 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1331 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1332 CERROR("Maximum number of changelog users exceeded!\n");
1333 GOTO(out, rc = -EOVERFLOW);
1335 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1336 rec->cur_endrec = mdd->mdd_cl.mc_index;
1337 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1339 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1341 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1344 llog_ctxt_put(ctxt);
1348 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1349 struct thandle *handle)
1354 /* XXX: this is a temporary solution to declare llog changes
1355 * will be fixed in 2.3 with new llog implementation */
1357 LASSERT(mdd->mdd_capa);
1359 /* the llog record could be canceled either by modifying
1360 * the plain llog's header or by destroying the llog itself
1361 * when this record is the last one in it, it can't be known
1362 * here, but the catlog's header will also be modified for
1363 * the second case, then the first case can be covered and
1364 * is no need to declare it */
1366 /* destroy empty plain log */
1367 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1371 /* record the catlog's header if an empty plain log was destroyed */
1372 rc = dt_declare_record_write(env, mdd->mdd_capa,
1373 sizeof(struct llog_logid_rec), 0, handle);
1377 struct mdd_changelog_user_data {
1378 __u64 mcud_endrec; /**< purge record for this user */
1379 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1381 __u32 mcud_minid; /**< user id with lowest rec reference */
1382 __u32 mcud_usercount;
1384 struct mdd_device *mcud_mdd;
1385 const struct lu_env *mcud_env;
1387 #define MCUD_UNREGISTER -1LL
1390 * 1. Find the smallest record everyone is willing to purge
1391 * 2. Update the last purgeable record for this user
1393 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1394 struct llog_rec_hdr *hdr, void *data)
1396 struct llog_changelog_user_rec *rec;
1397 struct mdd_changelog_user_data *mcud =
1398 (struct mdd_changelog_user_data *)data;
1402 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1404 rec = (struct llog_changelog_user_rec *)hdr;
1406 mcud->mcud_usercount++;
1408 /* If we have a new endrec for this id, use it for the following
1409 min check instead of its old value */
1410 if (rec->cur_id == mcud->mcud_id)
1411 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1413 /* Track the minimum referenced record */
1414 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1415 mcud->mcud_minid = rec->cur_id;
1416 mcud->mcud_minrec = rec->cur_endrec;
1419 if (rec->cur_id != mcud->mcud_id)
1422 /* Update this user's record */
1423 mcud->mcud_found = 1;
1425 /* Special case: unregister this user */
1426 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1427 struct llog_cookie cookie;
1429 struct mdd_device *mdd = mcud->mcud_mdd;
1431 cookie.lgc_lgl = llh->lgh_id;
1432 cookie.lgc_index = hdr->lrh_index;
1434 /* XXX This is a workaround for the deadlock of changelog
1435 * adding vs. changelog cancelling. LU-81. */
1436 th = mdd_trans_create(mcud->mcud_env, mdd);
1438 CERROR("Cannot get thandle\n");
1442 rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
1446 rc = mdd_trans_start(mcud->mcud_env, mdd, th);
1450 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1453 mcud->mcud_usercount--;
1456 mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
1460 /* Update the endrec */
1461 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1462 mcud->mcud_id, rec->cur_endrec);
1464 /* hdr+1 is loc of data */
1465 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1466 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1472 static int mdd_changelog_user_purge(const struct lu_env *env,
1473 struct mdd_device *mdd, int id,
1476 struct mdd_changelog_user_data data;
1477 struct llog_ctxt *ctxt;
1481 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1484 data.mcud_minid = 0;
1485 data.mcud_minrec = 0;
1486 data.mcud_usercount = 0;
1487 data.mcud_endrec = endrec;
1488 data.mcud_mdd = mdd;
1489 data.mcud_env = env;
1490 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1491 endrec = mdd->mdd_cl.mc_index;
1492 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1493 if ((data.mcud_endrec == 0) ||
1494 ((data.mcud_endrec > endrec) &&
1495 (data.mcud_endrec != MCUD_UNREGISTER)))
1496 data.mcud_endrec = endrec;
1498 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1501 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1503 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1504 (void *)&data, 0, 0);
1505 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1506 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1507 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1508 data.mcud_minrec, data.mcud_minid);
1509 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1511 CWARN("Could not determine changelog records to purge; rc=%d\n",
1515 llog_ctxt_put(ctxt);
1517 if (!data.mcud_found) {
1518 CWARN("No entry for user %d. Last changelog reference is "
1519 LPD64" by changelog user %d\n", data.mcud_id,
1520 data.mcud_minrec, data.mcud_minid);
1524 if (!rc && data.mcud_usercount == 0)
1525 /* No more users; turn changelogs off */
1526 rc = mdd_changelog_on(mdd, 0);
1532 * May be called remotely from mdt_iocontrol_handle or locally from
1533 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1534 * an obd_ioctl_data format (but local ioctl handler does).
1536 * \param len - data len
1537 * \param karg - ioctl data, in kernel space
1539 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1540 unsigned int cmd, int len, void *karg)
1542 struct mdd_device *mdd;
1543 struct obd_ioctl_data *data = karg;
1547 mdd = lu2mdd_dev(&m->md_lu_dev);
1549 /* Doesn't use obd_ioctl_data */
1551 case OBD_IOC_CHANGELOG_CLEAR: {
1552 struct changelog_setinfo *cs = karg;
1553 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1557 case OBD_IOC_GET_MNTOPT: {
1558 mntopt_t *mntopts = (mntopt_t *)karg;
1559 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1564 /* Below ioctls use obd_ioctl_data */
1565 if (len != sizeof(*data)) {
1566 CERROR("Bad ioctl size %d\n", len);
1569 if (data->ioc_version != OBD_IOCTL_VERSION) {
1570 CERROR("Bad magic %x != %x\n", data->ioc_version,
1576 case OBD_IOC_CHANGELOG_REG:
1577 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1579 case OBD_IOC_CHANGELOG_DEREG:
1580 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1590 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1591 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1594 const struct md_device_operations mdd_ops = {
1595 .mdo_statfs = mdd_statfs,
1596 .mdo_root_get = mdd_root_get,
1597 .mdo_maxsize_get = mdd_maxsize_get,
1598 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1599 .mdo_update_capa_key= mdd_update_capa_key,
1600 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1601 .mdo_iocontrol = mdd_iocontrol,
1602 #ifdef HAVE_QUOTA_SUPPORT
1604 .mqo_notify = mdd_quota_notify,
1605 .mqo_setup = mdd_quota_setup,
1606 .mqo_cleanup = mdd_quota_cleanup,
1607 .mqo_recovery = mdd_quota_recovery,
1608 .mqo_check = mdd_quota_check,
1609 .mqo_on = mdd_quota_on,
1610 .mqo_off = mdd_quota_off,
1611 .mqo_setinfo = mdd_quota_setinfo,
1612 .mqo_getinfo = mdd_quota_getinfo,
1613 .mqo_setquota = mdd_quota_setquota,
1614 .mqo_getquota = mdd_quota_getquota,
1615 .mqo_getoinfo = mdd_quota_getoinfo,
1616 .mqo_getoquota = mdd_quota_getoquota,
1617 .mqo_invalidate = mdd_quota_invalidate,
1618 .mqo_finvalidate = mdd_quota_finvalidate
1623 static struct lu_device_type_operations mdd_device_type_ops = {
1624 .ldto_init = mdd_type_init,
1625 .ldto_fini = mdd_type_fini,
1627 .ldto_start = mdd_type_start,
1628 .ldto_stop = mdd_type_stop,
1630 .ldto_device_alloc = mdd_device_alloc,
1631 .ldto_device_free = mdd_device_free,
1633 .ldto_device_init = mdd_device_init,
1634 .ldto_device_fini = mdd_device_fini
1637 static struct lu_device_type mdd_device_type = {
1638 .ldt_tags = LU_DEVICE_MD,
1639 .ldt_name = LUSTRE_MDD_NAME,
1640 .ldt_ops = &mdd_device_type_ops,
1641 .ldt_ctx_tags = LCT_MD_THREAD
1644 /* context key constructor: mdd_key_init */
1645 LU_KEY_INIT(mdd, struct mdd_thread_info);
1647 static void mdd_key_fini(const struct lu_context *ctx,
1648 struct lu_context_key *key, void *data)
1650 struct mdd_thread_info *info = data;
1651 if (info->mti_max_lmm != NULL)
1652 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1653 if (info->mti_max_cookie != NULL)
1654 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1655 mdd_buf_put(&info->mti_big_buf);
1660 /* context key: mdd_thread_key */
1661 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1663 static struct lu_local_obj_desc llod_capa_key = {
1664 .llod_name = CAPA_KEYS,
1665 .llod_oid = MDD_CAPA_KEYS_OID,
1669 static struct lu_local_obj_desc llod_mdd_orphan = {
1670 .llod_name = orph_index_name,
1671 .llod_oid = MDD_ORPHAN_OID,
1673 .llod_feat = &dt_directory_features,
1676 static struct lu_local_obj_desc llod_mdd_root = {
1677 .llod_name = mdd_root_dir_name,
1678 .llod_oid = MDD_ROOT_INDEX_OID,
1680 .llod_feat = &dt_directory_features,
1683 static int __init mdd_mod_init(void)
1685 struct lprocfs_static_vars lvars;
1686 lprocfs_mdd_init_vars(&lvars);
1688 llo_local_obj_register(&llod_capa_key);
1689 llo_local_obj_register(&llod_mdd_orphan);
1690 llo_local_obj_register(&llod_mdd_root);
1692 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1693 LUSTRE_MDD_NAME, &mdd_device_type);
1696 static void __exit mdd_mod_exit(void)
1698 llo_local_obj_unregister(&llod_capa_key);
1699 llo_local_obj_unregister(&llod_mdd_orphan);
1700 llo_local_obj_unregister(&llod_mdd_root);
1702 class_unregister_type(LUSTRE_MDD_NAME);
1705 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1706 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1707 MODULE_LICENSE("GPL");
1709 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);