1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/mdd/mdd_device.c
40 * Lustre Metadata Server (mdd) routines
42 * Author: Wang Di <wangdi@clusterfs.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_MDS
50 #include <linux/module.h>
51 #ifdef HAVE_EXT4_LDISKFS
52 #include <ldiskfs/ldiskfs_jbd2.h>
54 #include <linux/jbd.h>
57 #include <obd_class.h>
58 #include <lustre_ver.h>
59 #include <obd_support.h>
60 #include <lprocfs_status.h>
62 #include <lustre_disk.h>
63 #include <lustre_fid.h>
64 #ifdef HAVE_EXT4_LDISKFS
65 #include <ldiskfs/ldiskfs.h>
67 #include <linux/ldiskfs_fs.h>
69 #include <lustre_mds.h>
70 #include <lustre/lustre_idl.h>
71 #include <lustre_disk.h> /* for changelogs */
72 #include <lustre_param.h>
73 #include <lustre_fid.h>
75 #include "mdd_internal.h"
77 const struct md_device_operations mdd_ops;
78 static struct lu_device_type mdd_device_type;
80 static const char mdd_root_dir_name[] = "ROOT";
81 static const char mdd_obf_dir_name[] = "fid";
83 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
84 const char *name, struct lu_device *next)
86 struct mdd_device *mdd = lu2mdd_dev(d);
90 mdd->mdd_child = lu2dt_dev(next);
92 /* Prepare transactions callbacks. */
93 mdd->mdd_txn_cb.dtc_txn_start = NULL;
94 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
95 mdd->mdd_txn_cb.dtc_txn_commit = NULL;
96 mdd->mdd_txn_cb.dtc_cookie = mdd;
97 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
98 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
99 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
100 /* sync permission changes */
101 mdd->mdd_sync_permission = 1;
103 rc = mdd_procfs_init(mdd, name);
107 static struct lu_device *mdd_device_fini(const struct lu_env *env,
110 struct mdd_device *mdd = lu2mdd_dev(d);
111 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
114 rc = mdd_procfs_fini(mdd);
116 CERROR("proc fini error %d \n", rc);
122 static void mdd_changelog_fini(const struct lu_env *env,
123 struct mdd_device *mdd);
125 static void mdd_device_shutdown(const struct lu_env *env,
126 struct mdd_device *m, struct lustre_cfg *cfg)
129 mdd_changelog_fini(env, m);
130 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
131 if (m->mdd_dot_lustre_objs.mdd_obf)
132 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
133 if (m->mdd_dot_lustre)
134 mdd_object_put(env, m->mdd_dot_lustre);
136 mdd_fini_obd(env, m, cfg);
137 orph_index_fini(env, m);
138 if (m->mdd_capa != NULL) {
139 lu_object_put(env, &m->mdd_capa->do_lu);
142 /* remove upcall device*/
143 md_upcall_fini(&m->mdd_md_dev);
147 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
150 struct mdd_device *mdd = (struct mdd_device *)data;
151 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
154 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
155 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
158 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
159 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
160 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
161 llh->lgh_id.lgl_oid);
163 mdd->mdd_cl.mc_index = rec->cr.cr_index;
164 RETURN(LLOG_PROC_BREAK);
167 static int changelog_user_init_cb(struct llog_handle *llh,
168 struct llog_rec_hdr *hdr, void *data)
170 struct mdd_device *mdd = (struct mdd_device *)data;
171 struct llog_changelog_user_rec *rec =
172 (struct llog_changelog_user_rec *)hdr;
175 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
176 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
178 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
179 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
180 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
182 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
183 mdd->mdd_cl.mc_lastuser = rec->cur_id;
184 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
186 RETURN(LLOG_PROC_BREAK);
190 static int mdd_changelog_llog_init(struct mdd_device *mdd)
192 struct obd_device *obd = mdd2obd_dev(mdd);
193 struct llog_ctxt *ctxt;
196 /* Find last changelog entry number */
197 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
199 CERROR("no changelog context\n");
202 if (!ctxt->loc_handle) {
207 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
211 CERROR("changelog init failed: %d\n", rc);
214 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
215 mdd->mdd_cl.mc_index);
217 /* Find last changelog user id */
218 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
220 CERROR("no changelog user context\n");
223 if (!ctxt->loc_handle) {
228 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
233 CERROR("changelog user init failed: %d\n", rc);
237 /* If we have registered users, assume we want changelogs on */
238 if (mdd->mdd_cl.mc_lastuser > 0)
239 rc = mdd_changelog_on(mdd, 1);
244 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
248 mdd->mdd_cl.mc_index = 0;
249 cfs_spin_lock_init(&mdd->mdd_cl.mc_lock);
250 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
251 mdd->mdd_cl.mc_flags = 0; /* off by default */
252 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
253 cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock);
254 mdd->mdd_cl.mc_lastuser = 0;
256 rc = mdd_changelog_llog_init(mdd);
258 CERROR("Changelog setup during init failed %d\n", rc);
259 mdd->mdd_cl.mc_flags |= CLM_ERR;
265 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
267 mdd->mdd_cl.mc_flags = 0;
270 /* Start / stop recording */
271 int mdd_changelog_on(struct mdd_device *mdd, int on)
275 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
276 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
277 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
278 CERROR("Changelogs cannot be enabled due to error "
279 "condition (see %s log).\n",
280 mdd2obd_dev(mdd)->obd_name);
283 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
284 mdd->mdd_cl.mc_flags |= CLM_ON;
285 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
286 rc = mdd_changelog_write_header(mdd, CLM_START);
288 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
289 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
290 rc = mdd_changelog_write_header(mdd, CLM_FINI);
291 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
292 mdd->mdd_cl.mc_flags &= ~CLM_ON;
293 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
298 static __u64 cl_time(void) {
301 cfs_fs_time_current(&time);
302 return (((__u64)time.tv_sec) << 30) + time.tv_nsec;
305 /** Add a changelog entry \a rec to the changelog llog
308 * \param handle - currently ignored since llogs start their own transaction;
309 * this will hopefully be fixed in llog rewrite
312 int mdd_changelog_llog_write(struct mdd_device *mdd,
313 struct llog_changelog_rec *rec,
314 struct thandle *handle)
316 struct obd_device *obd = mdd2obd_dev(mdd);
317 struct llog_ctxt *ctxt;
320 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
321 /* llog_lvfs_write_rec sets the llog tail len */
322 rec->cr_hdr.lrh_type = CHANGELOG_REC;
323 rec->cr.cr_time = cl_time();
324 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
325 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
326 but as long as the MDD transactions are ordered correctly for e.g.
327 rename conflicts, I don't think this should matter. */
328 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
329 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
330 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
334 /* nested journal transaction */
335 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
341 /** Remove entries with indicies up to and including \a endrec from the
347 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
349 struct obd_device *obd = mdd2obd_dev(mdd);
350 struct llog_ctxt *ctxt;
351 long long unsigned cur;
354 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
358 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
359 cur = (long long)mdd->mdd_cl.mc_index;
360 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
364 /* purge to "0" is shorthand for everything */
368 /* If purging all records, write a header entry so we don't have an
369 empty catalog and we're sure to have a valid starting index next
370 time. In case of crash, we just restart with old log so we're
373 /* XXX: transaction is started by llog itself */
374 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
379 /* Some records were purged, so reset repeat-access time (so we
380 record new mtime update records, so users can see a file has been
381 changed since the last purge) */
382 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
384 /* XXX: transaction is started by llog itself */
385 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
391 /** Add a CL_MARK record to the changelog
393 * \param markerflags - CLM_*
396 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
398 struct obd_device *obd = mdd2obd_dev(mdd);
399 struct llog_changelog_rec *rec;
401 int len = strlen(obd->obd_name);
405 reclen = llog_data_len(sizeof(*rec) + len);
406 OBD_ALLOC(rec, reclen);
410 rec->cr.cr_flags = CLF_VERSION;
411 rec->cr.cr_type = CL_MARK;
412 rec->cr.cr_namelen = len;
413 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
414 /* Status and action flags */
415 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
417 /* XXX: transaction is started by llog itself */
418 rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ?
419 mdd_changelog_llog_write(mdd, rec, NULL) : 0;
421 /* assume on or off event; reset repeat-access time */
422 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
424 OBD_FREE(rec, reclen);
429 * Create ".lustre" directory.
431 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
433 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
434 struct md_object *mdo;
437 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
438 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
439 mdd_root_dir_name, dot_lustre_name,
440 fid, &dt_directory_features);
441 /* .lustre dir may be already present */
442 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
444 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
445 dot_lustre_name, PFID(fid), rc);
450 lu_object_put(env, &mdo->mo_lu);
456 static int dot_lustre_mdd_permission(const struct lu_env *env,
457 struct md_object *pobj,
458 struct md_object *cobj,
459 struct md_attr *attr, int mask)
461 if (mask & ~(MAY_READ | MAY_EXEC))
467 static int dot_lustre_mdd_attr_get(const struct lu_env *env,
468 struct md_object *obj, struct md_attr *ma)
470 struct mdd_object *mdd_obj = md2mdd_obj(obj);
472 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
475 static int dot_lustre_mdd_attr_set(const struct lu_env *env,
476 struct md_object *obj,
477 const struct md_attr *ma)
482 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
483 struct md_object *obj, struct lu_buf *buf,
489 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
490 struct md_object *obj, struct lu_buf *buf)
495 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
496 struct md_object *obj,
497 const struct lu_buf *buf, const char *name,
503 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
504 struct md_object *obj,
510 static int dot_lustre_mdd_readlink(const struct lu_env *env,
511 struct md_object *obj, struct lu_buf *buf)
516 static int dot_lustre_mdd_object_create(const struct lu_env *env,
517 struct md_object *obj,
518 const struct md_op_spec *spec,
524 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
525 struct md_object *obj,
526 const struct md_attr *ma)
531 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
532 struct md_object *obj,
538 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
541 struct mdd_object *mdd_obj = md2mdd_obj(obj);
543 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
544 mdd_obj->mod_count++;
545 mdd_write_unlock(env, mdd_obj);
550 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
551 struct md_attr *ma, int mode)
553 struct mdd_object *mdd_obj = md2mdd_obj(obj);
555 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
556 mdd_obj->mod_count--;
557 mdd_write_unlock(env, mdd_obj);
562 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
563 struct md_object *obj)
568 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
569 char *path, int pathlen, __u64 *recno, int *linkno)
574 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
575 struct lov_mds_md *lmm, struct ldlm_extent *extent,
576 struct lustre_handle *lockh)
581 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
582 struct lov_mds_md *lmm, struct lustre_handle *lockh)
587 static struct md_object_operations mdd_dot_lustre_obj_ops = {
588 .moo_permission = dot_lustre_mdd_permission,
589 .moo_attr_get = dot_lustre_mdd_attr_get,
590 .moo_attr_set = dot_lustre_mdd_attr_set,
591 .moo_xattr_get = dot_lustre_mdd_xattr_get,
592 .moo_xattr_list = dot_lustre_mdd_xattr_list,
593 .moo_xattr_set = dot_lustre_mdd_xattr_set,
594 .moo_xattr_del = dot_lustre_mdd_xattr_del,
595 .moo_readpage = mdd_readpage,
596 .moo_readlink = dot_lustre_mdd_readlink,
597 .moo_object_create = dot_lustre_mdd_object_create,
598 .moo_ref_add = dot_lustre_mdd_ref_add,
599 .moo_ref_del = dot_lustre_mdd_ref_del,
600 .moo_open = dot_lustre_mdd_open,
601 .moo_close = dot_lustre_mdd_close,
602 .moo_capa_get = mdd_capa_get,
603 .moo_object_sync = dot_lustre_mdd_object_sync,
604 .moo_path = dot_lustre_mdd_path,
605 .moo_file_lock = dot_file_lock,
606 .moo_file_unlock = dot_file_unlock,
610 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
611 const struct lu_name *lname, struct lu_fid *f,
612 struct md_op_spec *spec)
614 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
622 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
623 struct md_object *obj,
629 static int dot_lustre_mdd_create(const struct lu_env *env,
630 struct md_object *pobj,
631 const struct lu_name *lname,
632 struct md_object *child,
633 struct md_op_spec *spec,
639 static int dot_lustre_mdd_create_data(const struct lu_env *env,
642 const struct md_op_spec *spec,
648 static int dot_lustre_mdd_rename(const struct lu_env *env,
649 struct md_object *src_pobj,
650 struct md_object *tgt_pobj,
651 const struct lu_fid *lf,
652 const struct lu_name *lsname,
653 struct md_object *tobj,
654 const struct lu_name *ltname,
660 static int dot_lustre_mdd_link(const struct lu_env *env,
661 struct md_object *tgt_obj,
662 struct md_object *src_obj,
663 const struct lu_name *lname,
669 static int dot_lustre_mdd_unlink(const struct lu_env *env,
670 struct md_object *pobj,
671 struct md_object *cobj,
672 const struct lu_name *lname,
678 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
679 struct md_object *obj,
680 const struct lu_name *lname,
681 const struct lu_fid *fid,
682 const struct md_attr *ma)
687 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
688 struct md_object *obj,
689 const struct lu_name *lname,
690 const struct md_attr *ma)
695 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
696 struct md_object *pobj,
697 struct md_object *tobj,
698 const struct lu_fid *fid,
699 const struct lu_name *lname,
706 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
707 .mdo_is_subdir = mdd_is_subdir,
708 .mdo_lookup = dot_lustre_mdd_lookup,
709 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
710 .mdo_create = dot_lustre_mdd_create,
711 .mdo_create_data = dot_lustre_mdd_create_data,
712 .mdo_rename = dot_lustre_mdd_rename,
713 .mdo_link = dot_lustre_mdd_link,
714 .mdo_unlink = dot_lustre_mdd_unlink,
715 .mdo_name_insert = dot_lustre_mdd_name_insert,
716 .mdo_name_remove = dot_lustre_mdd_name_remove,
717 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
720 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
725 if (ma->ma_need & MA_INODE) {
726 struct mdd_device *mdd = mdo2mdd(obj);
728 /* "fid" is a virtual object and hence does not have any "real"
729 * attributes. So we reuse attributes of .lustre for "fid" dir */
730 ma->ma_need |= MA_INODE;
731 rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj,
735 ma->ma_valid |= MA_INODE;
738 /* "fid" directory does not have any striping information. */
739 if (ma->ma_need & MA_LOV) {
740 struct mdd_object *mdd_obj = md2mdd_obj(obj);
742 if (ma->ma_valid & MA_LOV)
745 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
746 S_ISDIR(mdd_object_type(mdd_obj))))
749 if (ma->ma_need & MA_LOV_DEF) {
750 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
752 ma->ma_lmm_size = rc;
753 ma->ma_valid |= MA_LOV;
762 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
763 const struct md_attr *ma)
768 static int obf_xattr_get(const struct lu_env *env,
769 struct md_object *obj, struct lu_buf *buf,
775 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
778 struct mdd_object *mdd_obj = md2mdd_obj(obj);
780 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
781 mdd_obj->mod_count++;
782 mdd_write_unlock(env, mdd_obj);
787 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
788 struct md_attr *ma, int mode)
790 struct mdd_object *mdd_obj = md2mdd_obj(obj);
792 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
793 mdd_obj->mod_count--;
794 mdd_write_unlock(env, mdd_obj);
799 /** Nothing to list in "fid" directory */
800 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
801 const struct lu_rdpg *rdpg)
806 static int obf_path(const struct lu_env *env, struct md_object *obj,
807 char *path, int pathlen, __u64 *recno, int *linkno)
812 static struct md_object_operations mdd_obf_obj_ops = {
813 .moo_attr_get = obf_attr_get,
814 .moo_attr_set = obf_attr_set,
815 .moo_xattr_get = obf_xattr_get,
816 .moo_open = obf_mdd_open,
817 .moo_close = obf_mdd_close,
818 .moo_readpage = obf_mdd_readpage,
823 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
824 * are valid. We also check if object with passed fid exists or not.
826 static int obf_lookup(const struct lu_env *env, struct md_object *p,
827 const struct lu_name *lname, struct lu_fid *f,
828 struct md_op_spec *spec)
830 char *name = (char *)lname->ln_name;
831 struct mdd_device *mdd = mdo2mdd(p);
832 struct mdd_object *child;
838 sscanf(name, SFID, RFID(f));
839 if (!fid_is_sane(f)) {
840 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
842 GOTO(out, rc = -EINVAL);
845 /* Check if object with this fid exists */
846 child = mdd_object_find(env, mdd, f);
850 GOTO(out, rc = PTR_ERR(child));
852 if (mdd_object_exists(child) == 0)
855 mdd_object_put(env, child);
861 static int obf_create(const struct lu_env *env, struct md_object *pobj,
862 const struct lu_name *lname, struct md_object *child,
863 struct md_op_spec *spec, struct md_attr* ma)
868 static int obf_rename(const struct lu_env *env,
869 struct md_object *src_pobj, struct md_object *tgt_pobj,
870 const struct lu_fid *lf, const struct lu_name *lsname,
871 struct md_object *tobj, const struct lu_name *ltname,
877 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
878 struct md_object *src_obj, const struct lu_name *lname,
884 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
885 struct md_object *cobj, const struct lu_name *lname,
891 static struct md_dir_operations mdd_obf_dir_ops = {
892 .mdo_lookup = obf_lookup,
893 .mdo_create = obf_create,
894 .mdo_rename = obf_rename,
895 .mdo_link = obf_link,
896 .mdo_unlink = obf_unlink
900 * Create special in-memory "fid" object for open-by-fid.
902 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
904 struct mdd_object *mdd_obf;
905 struct lu_object *obf_lu_obj;
908 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
910 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
911 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
912 GOTO(out, rc = -ENOENT);
914 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
915 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
916 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
917 /* Don't allow objects to be created in "fid" dir */
918 mdd_obf->mod_flags |= IMMUTE_OBJ;
920 obf_lu_obj = mdd2lu_obj(mdd_obf);
921 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
927 /** Setup ".lustre" directory object */
928 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
930 struct dt_object *dt_dot_lustre;
931 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
934 rc = create_dot_lustre_dir(env, m);
938 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
939 dot_lustre_name, fid);
940 if (IS_ERR(dt_dot_lustre)) {
941 rc = PTR_ERR(dt_dot_lustre);
945 /* references are released in mdd_device_shutdown() */
946 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
949 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
950 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
952 rc = mdd_obf_setup(env, m);
954 CERROR("Error initializing \"fid\" object - %d.\n", rc);
960 static int mdd_process_config(const struct lu_env *env,
961 struct lu_device *d, struct lustre_cfg *cfg)
963 struct mdd_device *m = lu2mdd_dev(d);
964 struct dt_device *dt = m->mdd_child;
965 struct lu_device *next = &dt->dd_lu_dev;
969 switch (cfg->lcfg_command) {
971 struct lprocfs_static_vars lvars;
973 lprocfs_mdd_init_vars(&lvars);
974 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
975 if (rc > 0 || rc == -ENOSYS)
976 /* we don't understand; pass it on */
977 rc = next->ld_ops->ldo_process_config(env, next, cfg);
981 rc = next->ld_ops->ldo_process_config(env, next, cfg);
984 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
986 rc = mdd_init_obd(env, m, cfg);
988 CERROR("lov init error %d\n", rc);
992 mdd_changelog_init(env, m);
995 mdd_device_shutdown(env, m, cfg);
997 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1005 static int mdd_lov_set_nextid(const struct lu_env *env,
1006 struct mdd_device *mdd)
1008 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
1012 LASSERT(mds->mds_lov_objids != NULL);
1013 rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID),
1014 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
1015 mds->mds_lov_objids, NULL);
1020 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
1021 struct mdd_device *mdd)
1023 /* XXX: to be implemented! */
1028 static int mdd_recovery_complete(const struct lu_env *env,
1029 struct lu_device *d)
1031 struct mdd_device *mdd = lu2mdd_dev(d);
1032 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1033 struct obd_device *obd = mdd2obd_dev(mdd);
1037 LASSERT(mdd != NULL);
1038 LASSERT(obd != NULL);
1040 /* XXX: Do we need this in new stack? */
1041 rc = mdd_lov_set_nextid(env, mdd);
1043 CERROR("mdd_lov_set_nextid() failed %d\n",
1048 /* XXX: cleanup unlink. */
1049 rc = mdd_cleanup_unlink_llog(env, mdd);
1051 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
1056 /* Call that with obd_recovering = 1 just to update objids */
1057 obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ?
1058 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
1060 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
1061 cfs_spin_lock(&obd->obd_dev_lock);
1062 obd->obd_recovering = 0;
1063 cfs_spin_unlock(&obd->obd_dev_lock);
1064 obd->obd_type->typ_dt_ops->o_postrecov(obd);
1066 /* XXX: orphans handling. */
1067 __mdd_orphan_cleanup(env, mdd);
1068 rc = next->ld_ops->ldo_recovery_complete(env, next);
1073 static int mdd_prepare(const struct lu_env *env,
1074 struct lu_device *pdev,
1075 struct lu_device *cdev)
1077 struct mdd_device *mdd = lu2mdd_dev(cdev);
1078 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1079 struct dt_object *root;
1084 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1088 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
1089 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1090 &mdd->mdd_root_fid);
1091 if (!IS_ERR(root)) {
1092 LASSERT(root != NULL);
1093 lu_object_put(env, &root->do_lu);
1094 rc = orph_index_init(env, mdd);
1101 rc = mdd_dot_lustre_setup(env, mdd);
1103 CERROR("Error(%d) initializing .lustre objects\n", rc);
1107 /* we use capa file to declare llog changes,
1108 * will be fixed with new llog in 2.3 */
1109 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1111 mdd->mdd_capa = root;
1119 const struct lu_device_operations mdd_lu_ops = {
1120 .ldo_object_alloc = mdd_object_alloc,
1121 .ldo_process_config = mdd_process_config,
1122 .ldo_recovery_complete = mdd_recovery_complete,
1123 .ldo_prepare = mdd_prepare,
1127 * No permission check is needed.
1129 static int mdd_root_get(const struct lu_env *env,
1130 struct md_device *m, struct lu_fid *f)
1132 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1135 *f = mdd->mdd_root_fid;
1140 * No permission check is needed.
1142 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1143 struct obd_statfs *sfs)
1145 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1150 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1156 * No permission check is needed.
1158 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1159 int *md_size, int *cookie_size)
1161 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1164 *md_size = mdd_lov_mdsize(env, mdd);
1165 *cookie_size = mdd_lov_cookiesize(env, mdd);
1170 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1171 int mode, unsigned long timeout, __u32 alg,
1172 struct lustre_capa_key *keys)
1174 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1175 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1179 /* need barrier for mds_capa_keys access. */
1180 cfs_down_write(&mds->mds_notify_lock);
1181 mds->mds_capa_keys = keys;
1182 cfs_up_write(&mds->mds_notify_lock);
1184 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1185 timeout, alg, keys);
1189 static int mdd_update_capa_key(const struct lu_env *env,
1190 struct md_device *m,
1191 struct lustre_capa_key *key)
1193 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1194 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1195 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp;
1199 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1200 sizeof(info), &info, NULL);
1204 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1207 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1209 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1210 return (*h == NULL ? -ENOENT : 0);
1213 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1214 struct lu_device_type *t,
1215 struct lustre_cfg *lcfg)
1217 struct lu_device *l;
1218 struct mdd_device *m;
1222 l = ERR_PTR(-ENOMEM);
1224 md_device_init(&m->mdd_md_dev, t);
1226 l->ld_ops = &mdd_lu_ops;
1227 m->mdd_md_dev.md_ops = &mdd_ops;
1228 md_upcall_init(&m->mdd_md_dev, NULL);
1234 static struct lu_device *mdd_device_free(const struct lu_env *env,
1235 struct lu_device *lu)
1237 struct mdd_device *m = lu2mdd_dev(lu);
1238 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1241 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1242 md_device_fini(&m->mdd_md_dev);
1247 static struct obd_ops mdd_obd_device_ops = {
1248 .o_owner = THIS_MODULE
1251 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1252 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1254 static struct lu_context_key mdd_ucred_key = {
1255 .lct_tags = LCT_SESSION,
1256 .lct_init = mdd_ucred_key_init,
1257 .lct_fini = mdd_ucred_key_fini
1260 struct md_ucred *md_ucred(const struct lu_env *env)
1262 LASSERT(env->le_ses != NULL);
1263 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1265 EXPORT_SYMBOL(md_ucred);
1268 * context key constructor/destructor:
1269 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1271 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1273 struct lu_context_key mdd_capainfo_key = {
1274 .lct_tags = LCT_SESSION,
1275 .lct_init = mdd_capainfo_key_init,
1276 .lct_fini = mdd_capainfo_key_fini
1279 struct md_capainfo *md_capainfo(const struct lu_env *env)
1281 /* NB, in mdt_init0 */
1282 if (env->le_ses == NULL)
1284 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1286 EXPORT_SYMBOL(md_capainfo);
1289 * context key constructor/destructor:
1290 * mdd_quota_key_init, mdd_quota_key_fini
1292 LU_KEY_INIT_FINI(mdd_quota, struct md_quota);
1294 struct lu_context_key mdd_quota_key = {
1295 .lct_tags = LCT_SESSION,
1296 .lct_init = mdd_quota_key_init,
1297 .lct_fini = mdd_quota_key_fini
1300 struct md_quota *md_quota(const struct lu_env *env)
1302 LASSERT(env->le_ses != NULL);
1303 return lu_context_key_get(env->le_ses, &mdd_quota_key);
1305 EXPORT_SYMBOL(md_quota);
1307 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1309 struct llog_ctxt *ctxt;
1310 struct llog_changelog_user_rec *rec;
1314 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1320 llog_ctxt_put(ctxt);
1324 /* Assume we want it on since somebody registered */
1325 rc = mdd_changelog_on(mdd, 1);
1329 rec->cur_hdr.lrh_len = sizeof(*rec);
1330 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1331 cfs_spin_lock(&mdd->mdd_cl.mc_user_lock);
1332 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1333 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1334 CERROR("Maximum number of changelog users exceeded!\n");
1335 GOTO(out, rc = -EOVERFLOW);
1337 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1338 rec->cur_endrec = mdd->mdd_cl.mc_index;
1339 cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock);
1341 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1343 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1346 llog_ctxt_put(ctxt);
1350 int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
1351 struct thandle *handle)
1356 /* XXX: this is a temporary solution to declare llog changes
1357 * will be fixed in 2.3 with new llog implementation */
1359 LASSERT(mdd->mdd_capa);
1361 /* the llog record could be canceled either by modifying
1362 * the plain llog's header or by destroying the llog itself
1363 * when this record is the last one in it, it can't be known
1364 * here, but the catlog's header will also be modified for
1365 * the second case, then the first case can be covered and
1366 * is no need to declare it */
1368 /* destroy empty plain log */
1369 rc = dt_declare_destroy(env, mdd->mdd_capa, handle);
1373 /* record the catlog's header if an empty plain log was destroyed */
1374 rc = dt_declare_record_write(env, mdd->mdd_capa,
1375 sizeof(struct llog_logid_rec), 0, handle);
1379 struct mdd_changelog_user_data {
1380 __u64 mcud_endrec; /**< purge record for this user */
1381 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1383 __u32 mcud_minid; /**< user id with lowest rec reference */
1384 __u32 mcud_usercount;
1386 struct mdd_device *mcud_mdd;
1387 const struct lu_env *mcud_env;
1389 #define MCUD_UNREGISTER -1LL
1392 * 1. Find the smallest record everyone is willing to purge
1393 * 2. Update the last purgeable record for this user
1395 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1396 struct llog_rec_hdr *hdr, void *data)
1398 struct llog_changelog_user_rec *rec;
1399 struct mdd_changelog_user_data *mcud =
1400 (struct mdd_changelog_user_data *)data;
1404 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1406 rec = (struct llog_changelog_user_rec *)hdr;
1408 mcud->mcud_usercount++;
1410 /* If we have a new endrec for this id, use it for the following
1411 min check instead of its old value */
1412 if (rec->cur_id == mcud->mcud_id)
1413 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1415 /* Track the minimum referenced record */
1416 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1417 mcud->mcud_minid = rec->cur_id;
1418 mcud->mcud_minrec = rec->cur_endrec;
1421 if (rec->cur_id != mcud->mcud_id)
1424 /* Update this user's record */
1425 mcud->mcud_found = 1;
1427 /* Special case: unregister this user */
1428 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1429 struct llog_cookie cookie;
1431 struct mdd_device *mdd = mcud->mcud_mdd;
1433 cookie.lgc_lgl = llh->lgh_id;
1434 cookie.lgc_index = hdr->lrh_index;
1436 /* XXX This is a workaround for the deadlock of changelog
1437 * adding vs. changelog cancelling. LU-81. */
1438 th = mdd_trans_create(mcud->mcud_env, mdd);
1440 CERROR("Cannot get thandle\n");
1444 rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th);
1448 rc = mdd_trans_start(mcud->mcud_env, mdd, th);
1452 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1455 mcud->mcud_usercount--;
1458 mdd_trans_stop(mcud->mcud_env, mdd, 0, th);
1462 /* Update the endrec */
1463 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1464 mcud->mcud_id, rec->cur_endrec);
1466 /* hdr+1 is loc of data */
1467 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1468 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1474 static int mdd_changelog_user_purge(const struct lu_env *env,
1475 struct mdd_device *mdd, int id,
1478 struct mdd_changelog_user_data data;
1479 struct llog_ctxt *ctxt;
1483 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1486 data.mcud_minid = 0;
1487 data.mcud_minrec = 0;
1488 data.mcud_usercount = 0;
1489 data.mcud_endrec = endrec;
1490 data.mcud_mdd = mdd;
1491 data.mcud_env = env;
1492 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
1493 endrec = mdd->mdd_cl.mc_index;
1494 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
1495 if ((data.mcud_endrec == 0) ||
1496 ((data.mcud_endrec > endrec) &&
1497 (data.mcud_endrec != MCUD_UNREGISTER)))
1498 data.mcud_endrec = endrec;
1500 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1503 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1505 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1506 (void *)&data, 0, 0);
1507 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1508 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1509 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1510 data.mcud_minrec, data.mcud_minid);
1511 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1513 CWARN("Could not determine changelog records to purge; rc=%d\n",
1517 llog_ctxt_put(ctxt);
1519 if (!data.mcud_found) {
1520 CWARN("No entry for user %d. Last changelog reference is "
1521 LPD64" by changelog user %d\n", data.mcud_id,
1522 data.mcud_minrec, data.mcud_minid);
1526 if (!rc && data.mcud_usercount == 0)
1527 /* No more users; turn changelogs off */
1528 rc = mdd_changelog_on(mdd, 0);
1534 * May be called remotely from mdt_iocontrol_handle or locally from
1535 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1536 * an obd_ioctl_data format (but local ioctl handler does).
1538 * \param len - data len
1539 * \param karg - ioctl data, in kernel space
1541 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1542 unsigned int cmd, int len, void *karg)
1544 struct mdd_device *mdd;
1545 struct obd_ioctl_data *data = karg;
1549 mdd = lu2mdd_dev(&m->md_lu_dev);
1551 /* Doesn't use obd_ioctl_data */
1553 case OBD_IOC_CHANGELOG_CLEAR: {
1554 struct changelog_setinfo *cs = karg;
1555 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1559 case OBD_IOC_GET_MNTOPT: {
1560 mntopt_t *mntopts = (mntopt_t *)karg;
1561 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1566 /* Below ioctls use obd_ioctl_data */
1567 if (len != sizeof(*data)) {
1568 CERROR("Bad ioctl size %d\n", len);
1571 if (data->ioc_version != OBD_IOCTL_VERSION) {
1572 CERROR("Bad magic %x != %x\n", data->ioc_version,
1578 case OBD_IOC_CHANGELOG_REG:
1579 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1581 case OBD_IOC_CHANGELOG_DEREG:
1582 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1592 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1593 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key,
1596 const struct md_device_operations mdd_ops = {
1597 .mdo_statfs = mdd_statfs,
1598 .mdo_root_get = mdd_root_get,
1599 .mdo_maxsize_get = mdd_maxsize_get,
1600 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1601 .mdo_update_capa_key= mdd_update_capa_key,
1602 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1603 .mdo_iocontrol = mdd_iocontrol,
1604 #ifdef HAVE_QUOTA_SUPPORT
1606 .mqo_notify = mdd_quota_notify,
1607 .mqo_setup = mdd_quota_setup,
1608 .mqo_cleanup = mdd_quota_cleanup,
1609 .mqo_recovery = mdd_quota_recovery,
1610 .mqo_check = mdd_quota_check,
1611 .mqo_on = mdd_quota_on,
1612 .mqo_off = mdd_quota_off,
1613 .mqo_setinfo = mdd_quota_setinfo,
1614 .mqo_getinfo = mdd_quota_getinfo,
1615 .mqo_setquota = mdd_quota_setquota,
1616 .mqo_getquota = mdd_quota_getquota,
1617 .mqo_getoinfo = mdd_quota_getoinfo,
1618 .mqo_getoquota = mdd_quota_getoquota,
1619 .mqo_invalidate = mdd_quota_invalidate,
1620 .mqo_finvalidate = mdd_quota_finvalidate
1625 static struct lu_device_type_operations mdd_device_type_ops = {
1626 .ldto_init = mdd_type_init,
1627 .ldto_fini = mdd_type_fini,
1629 .ldto_start = mdd_type_start,
1630 .ldto_stop = mdd_type_stop,
1632 .ldto_device_alloc = mdd_device_alloc,
1633 .ldto_device_free = mdd_device_free,
1635 .ldto_device_init = mdd_device_init,
1636 .ldto_device_fini = mdd_device_fini
1639 static struct lu_device_type mdd_device_type = {
1640 .ldt_tags = LU_DEVICE_MD,
1641 .ldt_name = LUSTRE_MDD_NAME,
1642 .ldt_ops = &mdd_device_type_ops,
1643 .ldt_ctx_tags = LCT_MD_THREAD
1646 /* context key constructor: mdd_key_init */
1647 LU_KEY_INIT(mdd, struct mdd_thread_info);
1649 static void mdd_key_fini(const struct lu_context *ctx,
1650 struct lu_context_key *key, void *data)
1652 struct mdd_thread_info *info = data;
1653 if (info->mti_max_lmm != NULL)
1654 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1655 if (info->mti_max_cookie != NULL)
1656 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1657 mdd_buf_put(&info->mti_big_buf);
1662 /* context key: mdd_thread_key */
1663 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1665 static struct lu_local_obj_desc llod_capa_key = {
1666 .llod_name = CAPA_KEYS,
1667 .llod_oid = MDD_CAPA_KEYS_OID,
1671 static struct lu_local_obj_desc llod_mdd_orphan = {
1672 .llod_name = orph_index_name,
1673 .llod_oid = MDD_ORPHAN_OID,
1675 .llod_feat = &dt_directory_features,
1678 static struct lu_local_obj_desc llod_mdd_root = {
1679 .llod_name = mdd_root_dir_name,
1680 .llod_oid = MDD_ROOT_INDEX_OID,
1682 .llod_feat = &dt_directory_features,
1685 static int __init mdd_mod_init(void)
1687 struct lprocfs_static_vars lvars;
1688 lprocfs_mdd_init_vars(&lvars);
1690 llo_local_obj_register(&llod_capa_key);
1691 llo_local_obj_register(&llod_mdd_orphan);
1692 llo_local_obj_register(&llod_mdd_root);
1694 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1695 LUSTRE_MDD_NAME, &mdd_device_type);
1698 static void __exit mdd_mod_exit(void)
1700 llo_local_obj_unregister(&llod_capa_key);
1701 llo_local_obj_unregister(&llod_mdd_orphan);
1702 llo_local_obj_unregister(&llod_mdd_root);
1704 class_unregister_type(LUSTRE_MDD_NAME);
1707 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1708 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1709 MODULE_LICENSE("GPL");
1711 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);