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, Intel Corporation.
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 <obd_class.h>
47 #include <lustre_mds.h>
48 #include <obd_support.h>
49 #include <lu_object.h>
50 #include <lustre_param.h>
51 #include <lustre_fid.h>
53 #include "mdd_internal.h"
55 const struct md_device_operations mdd_ops;
56 static struct lu_device_type mdd_device_type;
58 static const char mdd_root_dir_name[] = "ROOT";
59 static const char mdd_obf_dir_name[] = "fid";
61 /* Slab for MDD object allocation */
62 cfs_mem_cache_t *mdd_object_kmem;
64 static struct lu_kmem_descr mdd_caches[] = {
66 .ckd_cache = &mdd_object_kmem,
67 .ckd_name = "mdd_obj",
68 .ckd_size = sizeof(struct mdd_object)
75 static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
78 struct obd_connect_data *data = NULL;
79 struct lu_device *lud = mdd2lu_dev(m);
80 struct obd_device *obd;
84 LASSERT(m->mdd_child_exp == NULL);
86 OBD_ALLOC(data, sizeof(*data));
88 GOTO(out, rc = -ENOMEM);
90 obd = class_name2obd(nextdev);
92 CERROR("can't locate next device: %s\n", nextdev);
93 GOTO(out, rc = -ENOTCONN);
96 data->ocd_connect_flags = OBD_CONNECT_VERSION;
97 data->ocd_version = LUSTRE_VERSION_CODE;
99 rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
101 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
105 lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
106 LASSERT(lud->ld_site);
107 m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
108 m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
109 lu_dev_add_linkage(lud->ld_site, lud);
113 OBD_FREE(data, sizeof(*data));
117 static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
118 struct lu_device_type *t, struct lustre_cfg *lcfg)
123 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
124 mdd->mdd_md_dev.md_ops = &mdd_ops;
126 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
130 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
131 /* sync permission changes */
132 mdd->mdd_sync_permission = 1;
134 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
136 /* we are using service name but not mdd obd name
137 * for compatibility reasons.
138 * It is passed from MDT in lustre_cfg[2] buffer */
139 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
141 obd_disconnect(mdd->mdd_child_exp);
146 static struct lu_device *mdd_device_fini(const struct lu_env *env,
149 struct mdd_device *mdd = lu2mdd_dev(d);
153 lu_dev_del_linkage(d->ld_site, d);
155 rc = mdd_procfs_fini(mdd);
157 CERROR("proc fini error %d \n", rc);
163 static void mdd_changelog_fini(const struct lu_env *env,
164 struct mdd_device *mdd);
166 static void mdd_device_shutdown(const struct lu_env *env,
167 struct mdd_device *m, struct lustre_cfg *cfg)
170 if (m->mdd_dot_lustre_objs.mdd_obf)
171 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
172 if (m->mdd_dot_lustre)
173 mdd_object_put(env, m->mdd_dot_lustre);
174 orph_index_fini(env, m);
175 if (m->mdd_capa != NULL) {
176 lu_object_put(env, &m->mdd_capa->do_lu);
179 lu_site_purge(env, m->mdd_md_dev.md_lu_dev.ld_site, -1);
180 /* remove upcall device*/
181 md_upcall_fini(&m->mdd_md_dev);
183 if (m->mdd_child_exp)
184 obd_disconnect(m->mdd_child_exp);
189 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
190 struct llog_rec_hdr *hdr, void *data)
192 struct mdd_device *mdd = (struct mdd_device *)data;
193 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
195 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
196 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
199 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
200 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
201 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
202 llh->lgh_id.lgl_oid);
204 mdd->mdd_cl.mc_index = rec->cr.cr_index;
205 return LLOG_PROC_BREAK;
208 static int changelog_user_init_cb(const struct lu_env *env,
209 struct llog_handle *llh,
210 struct llog_rec_hdr *hdr, void *data)
212 struct mdd_device *mdd = (struct mdd_device *)data;
213 struct llog_changelog_user_rec *rec =
214 (struct llog_changelog_user_rec *)hdr;
216 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
217 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
219 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
220 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
221 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
223 spin_lock(&mdd->mdd_cl.mc_user_lock);
224 mdd->mdd_cl.mc_lastuser = rec->cur_id;
225 spin_unlock(&mdd->mdd_cl.mc_user_lock);
227 return LLOG_PROC_BREAK;
230 static int llog_changelog_cancel_cb(const struct lu_env *env,
231 struct llog_handle *llh,
232 struct llog_rec_hdr *hdr, void *data)
234 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
235 struct llog_cookie cookie;
236 long long endrec = *(long long *)data;
241 /* This is always a (sub)log, not the catalog */
242 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
244 if (rec->cr.cr_index > endrec)
245 /* records are in order, so we're done */
246 RETURN(LLOG_PROC_BREAK);
248 cookie.lgc_lgl = llh->lgh_id;
249 cookie.lgc_index = hdr->lrh_index;
251 /* cancel them one at a time. I suppose we could store up the cookies
252 * and cancel them all at once; probably more efficient, but this is
253 * done as a user call, so who cares... */
254 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
256 RETURN(rc < 0 ? rc : 0);
259 static int llog_changelog_cancel(const struct lu_env *env,
260 struct llog_ctxt *ctxt,
261 struct lov_stripe_md *lsm, int count,
262 struct llog_cookie *cookies, int flags)
264 struct llog_handle *cathandle = ctxt->loc_handle;
269 /* This should only be called with the catalog handle */
270 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
272 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
273 (void *)cookies, 0, 0);
275 /* 0 or 1 means we're done */
278 CERROR("%s: cancel idx %u of catalog "LPX64" rc=%d\n",
279 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
280 cathandle->lgh_id.lgl_oid, rc);
285 static struct llog_operations changelog_orig_logops;
287 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
289 static int mdd_changelog_llog_init(const struct lu_env *env,
290 struct mdd_device *mdd)
292 struct obd_device *obd = mdd2obd_dev(mdd);
293 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
297 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
298 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
299 rc = dt_root_get(env, mdd->mdd_bottom, &rfid);
303 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
304 obd, &changelog_orig_logops);
306 CERROR("%s: changelog llog setup failed: rc = %d\n",
311 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
314 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
317 GOTO(out_cleanup, rc);
319 rc = llog_cat_init_and_process(env, ctxt->loc_handle);
323 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
324 changelog_init_cb, mdd);
327 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
332 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
333 mdd->mdd_cl.mc_index);
335 /* setup user changelog */
336 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
337 obd, &changelog_orig_logops);
339 CERROR("%s: changelog users llog setup failed: rc = %d\n",
344 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
347 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
350 GOTO(out_ucleanup, rc);
352 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
353 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
355 rc = llog_cat_init_and_process(env, uctxt->loc_handle);
357 GOTO(out_uclose, rc);
359 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
360 changelog_user_init_cb, mdd);
362 CERROR("%s: changelog user init failed: rc = %d\n",
364 GOTO(out_uclose, rc);
367 /* If we have registered users, assume we want changelogs on */
368 if (mdd->mdd_cl.mc_lastuser > 0) {
369 rc = mdd_changelog_on(env, mdd, 1);
371 GOTO(out_uclose, rc);
374 llog_ctxt_put(uctxt);
377 llog_cat_close(env, uctxt->loc_handle);
379 llog_cleanup(env, uctxt);
381 llog_cat_close(env, ctxt->loc_handle);
383 llog_cleanup(env, ctxt);
387 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
389 struct obd_device *obd = mdd2obd_dev(mdd);
392 mdd->mdd_cl.mc_index = 0;
393 spin_lock_init(&mdd->mdd_cl.mc_lock);
394 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
395 mdd->mdd_cl.mc_flags = 0; /* off by default */
396 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
397 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
398 mdd->mdd_cl.mc_lastuser = 0;
400 rc = mdd_changelog_llog_init(env, mdd);
402 CERROR("%s: changelog setup during init failed: rc = %d\n",
404 mdd->mdd_cl.mc_flags |= CLM_ERR;
410 static void mdd_changelog_fini(const struct lu_env *env,
411 struct mdd_device *mdd)
413 struct obd_device *obd = mdd2obd_dev(mdd);
414 struct llog_ctxt *ctxt;
416 mdd->mdd_cl.mc_flags = 0;
418 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
420 llog_cat_close(env, ctxt->loc_handle);
421 llog_cleanup(env, ctxt);
423 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
425 llog_cat_close(env, ctxt->loc_handle);
426 llog_cleanup(env, ctxt);
430 int mdd_changelog_write_header(const struct lu_env *env,
431 struct mdd_device *mdd, int markerflags);
433 /* Start / stop recording */
434 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
438 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
439 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
440 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
441 CERROR("Changelogs cannot be enabled due to error "
442 "condition (see %s log).\n",
443 mdd2obd_dev(mdd)->obd_name);
446 spin_lock(&mdd->mdd_cl.mc_lock);
447 mdd->mdd_cl.mc_flags |= CLM_ON;
448 spin_unlock(&mdd->mdd_cl.mc_lock);
449 rc = mdd_changelog_write_header(env, mdd, CLM_START);
451 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
452 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
453 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
454 spin_lock(&mdd->mdd_cl.mc_lock);
455 mdd->mdd_cl.mc_flags &= ~CLM_ON;
456 spin_unlock(&mdd->mdd_cl.mc_lock);
461 /** Remove entries with indicies up to and including \a endrec from the
467 int mdd_changelog_llog_cancel(const struct lu_env *env,
468 struct mdd_device *mdd, long long endrec)
470 struct obd_device *obd = mdd2obd_dev(mdd);
471 struct llog_ctxt *ctxt;
472 long long unsigned cur;
475 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
479 spin_lock(&mdd->mdd_cl.mc_lock);
480 cur = (long long)mdd->mdd_cl.mc_index;
481 spin_unlock(&mdd->mdd_cl.mc_lock);
485 /* purge to "0" is shorthand for everything */
489 /* If purging all records, write a header entry so we don't have an
490 empty catalog and we're sure to have a valid starting index next
491 time. In case of crash, we just restart with old log so we're
494 /* XXX: transaction is started by llog itself */
495 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
500 /* Some records were purged, so reset repeat-access time (so we
501 record new mtime update records, so users can see a file has been
502 changed since the last purge) */
503 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
505 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
511 /** Add a CL_MARK record to the changelog
513 * \param markerflags - CLM_*
516 int mdd_changelog_write_header(const struct lu_env *env,
517 struct mdd_device *mdd, int markerflags)
519 struct obd_device *obd = mdd2obd_dev(mdd);
520 struct llog_changelog_rec *rec;
522 struct llog_ctxt *ctxt;
524 int len = strlen(obd->obd_name);
529 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
530 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
534 reclen = llog_data_len(sizeof(*rec) + len);
535 buf = mdd_buf_alloc(env, reclen);
536 if (buf->lb_buf == NULL)
540 rec->cr.cr_flags = CLF_VERSION;
541 rec->cr.cr_type = CL_MARK;
542 rec->cr.cr_namelen = len;
543 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
544 /* Status and action flags */
545 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
546 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
547 rec->cr_hdr.lrh_type = CHANGELOG_REC;
548 rec->cr.cr_time = cl_time();
549 spin_lock(&mdd->mdd_cl.mc_lock);
550 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
551 spin_unlock(&mdd->mdd_cl.mc_lock);
553 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
556 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
561 /* assume on or off event; reset repeat-access time */
562 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
567 * Create ".lustre" directory.
569 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
571 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
572 struct md_object *mdo;
575 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
576 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
577 mdd_root_dir_name, dot_lustre_name,
578 fid, &dt_directory_features);
579 /* .lustre dir may be already present */
580 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
582 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
583 dot_lustre_name, PFID(fid), rc);
588 lu_object_put(env, &mdo->mo_lu);
594 static int dot_lustre_mdd_permission(const struct lu_env *env,
595 struct md_object *pobj,
596 struct md_object *cobj,
597 struct md_attr *attr, int mask)
599 if (mask & ~(MAY_READ | MAY_EXEC))
605 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
606 struct md_object *obj, struct lu_buf *buf,
612 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
613 struct md_object *obj, struct lu_buf *buf)
618 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
619 struct md_object *obj,
620 const struct lu_buf *buf, const char *name,
626 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
627 struct md_object *obj,
633 static int dot_lustre_mdd_swap_layouts(const struct lu_env *env,
634 struct md_object *obj1,
635 struct md_object *obj2,
641 static int dot_lustre_mdd_readlink(const struct lu_env *env,
642 struct md_object *obj, struct lu_buf *buf)
647 static int dot_lustre_mdd_object_create(const struct lu_env *env,
648 struct md_object *obj,
649 const struct md_op_spec *spec,
655 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
656 struct md_object *obj,
657 const struct md_attr *ma)
662 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
663 struct md_object *obj,
669 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
672 struct mdd_object *mdd_obj = md2mdd_obj(obj);
674 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
675 mdd_obj->mod_count++;
676 mdd_write_unlock(env, mdd_obj);
681 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
682 struct md_attr *ma, int mode)
684 struct mdd_object *mdd_obj = md2mdd_obj(obj);
686 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
687 mdd_obj->mod_count--;
688 mdd_write_unlock(env, mdd_obj);
693 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
694 struct md_object *obj)
699 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
700 char *path, int pathlen, __u64 *recno, int *linkno)
705 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
706 struct lov_mds_md *lmm, struct ldlm_extent *extent,
707 struct lustre_handle *lockh)
712 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
713 struct lov_mds_md *lmm, struct lustre_handle *lockh)
718 static struct md_object_operations mdd_dot_lustre_obj_ops = {
719 .moo_permission = dot_lustre_mdd_permission,
720 .moo_attr_get = mdd_attr_get,
721 .moo_attr_set = mdd_attr_set,
722 .moo_xattr_get = dot_lustre_mdd_xattr_get,
723 .moo_xattr_list = dot_lustre_mdd_xattr_list,
724 .moo_xattr_set = dot_lustre_mdd_xattr_set,
725 .moo_xattr_del = dot_lustre_mdd_xattr_del,
726 .moo_swap_layouts = dot_lustre_mdd_swap_layouts,
727 .moo_readpage = mdd_readpage,
728 .moo_readlink = dot_lustre_mdd_readlink,
729 .moo_object_create = dot_lustre_mdd_object_create,
730 .moo_ref_add = dot_lustre_mdd_ref_add,
731 .moo_ref_del = dot_lustre_mdd_ref_del,
732 .moo_open = dot_lustre_mdd_open,
733 .moo_close = dot_lustre_mdd_close,
734 .moo_capa_get = mdd_capa_get,
735 .moo_object_sync = dot_lustre_mdd_object_sync,
736 .moo_path = dot_lustre_mdd_path,
737 .moo_file_lock = dot_file_lock,
738 .moo_file_unlock = dot_file_unlock,
742 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
743 const struct lu_name *lname, struct lu_fid *f,
744 struct md_op_spec *spec)
746 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0) {
754 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
755 struct md_object *obj,
761 static int dot_lustre_mdd_create(const struct lu_env *env,
762 struct md_object *pobj,
763 const struct lu_name *lname,
764 struct md_object *child,
765 struct md_op_spec *spec,
771 static int dot_lustre_mdd_create_data(const struct lu_env *env,
774 const struct md_op_spec *spec,
780 static int dot_lustre_mdd_rename(const struct lu_env *env,
781 struct md_object *src_pobj,
782 struct md_object *tgt_pobj,
783 const struct lu_fid *lf,
784 const struct lu_name *lsname,
785 struct md_object *tobj,
786 const struct lu_name *ltname,
792 static int dot_lustre_mdd_link(const struct lu_env *env,
793 struct md_object *tgt_obj,
794 struct md_object *src_obj,
795 const struct lu_name *lname,
801 static int dot_lustre_mdd_unlink(const struct lu_env *env,
802 struct md_object *pobj,
803 struct md_object *cobj,
804 const struct lu_name *lname,
810 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
811 struct md_object *obj,
812 const struct lu_name *lname,
813 const struct lu_fid *fid,
814 const struct md_attr *ma)
819 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
820 struct md_object *obj,
821 const struct lu_name *lname,
822 const struct md_attr *ma)
827 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
828 struct md_object *pobj,
829 struct md_object *tobj,
830 const struct lu_fid *fid,
831 const struct lu_name *lname,
838 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
839 .mdo_is_subdir = mdd_is_subdir,
840 .mdo_lookup = dot_lustre_mdd_lookup,
841 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
842 .mdo_create = dot_lustre_mdd_create,
843 .mdo_create_data = dot_lustre_mdd_create_data,
844 .mdo_rename = dot_lustre_mdd_rename,
845 .mdo_link = dot_lustre_mdd_link,
846 .mdo_unlink = dot_lustre_mdd_unlink,
847 .mdo_name_insert = dot_lustre_mdd_name_insert,
848 .mdo_name_remove = dot_lustre_mdd_name_remove,
849 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
852 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
855 struct mdd_device *mdd = mdo2mdd(obj);
857 /* "fid" is a virtual object and hence does not have any "real"
858 * attributes. So we reuse attributes of .lustre for "fid" dir */
859 return mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
862 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
863 const struct md_attr *ma)
868 static int obf_xattr_list(const struct lu_env *env,
869 struct md_object *obj, struct lu_buf *buf)
874 static int obf_xattr_get(const struct lu_env *env,
875 struct md_object *obj, struct lu_buf *buf,
878 struct mdd_device *mdd = mdo2mdd(obj);
879 struct mdd_object *root;
880 struct lu_fid rootfid;
884 * .lustre returns default striping which is 'stored'
887 if (strcmp(name, XATTR_NAME_LOV) == 0) {
888 rc = dt_root_get(env, mdd->mdd_child, &rootfid);
891 root = mdd_object_find(env, mdd, &rootfid);
893 return PTR_ERR(root);
894 rc = mdo_xattr_get(env, root, buf, name,
895 mdd_object_capa(env, md2mdd_obj(obj)));
896 mdd_object_put(env, root);
902 static int obf_xattr_set(const struct lu_env *env,
903 struct md_object *obj,
904 const struct lu_buf *buf, const char *name,
910 static int obf_xattr_del(const struct lu_env *env,
911 struct md_object *obj,
917 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
920 struct mdd_object *mdd_obj = md2mdd_obj(obj);
922 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
923 mdd_obj->mod_count++;
924 mdd_write_unlock(env, mdd_obj);
929 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
930 struct md_attr *ma, int mode)
932 struct mdd_object *mdd_obj = md2mdd_obj(obj);
934 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
935 mdd_obj->mod_count--;
936 mdd_write_unlock(env, mdd_obj);
941 /** Nothing to list in "fid" directory */
942 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
943 const struct lu_rdpg *rdpg)
948 static int obf_path(const struct lu_env *env, struct md_object *obj,
949 char *path, int pathlen, __u64 *recno, int *linkno)
954 static struct md_object_operations mdd_obf_obj_ops = {
955 .moo_attr_get = obf_attr_get,
956 .moo_attr_set = obf_attr_set,
957 .moo_xattr_list = obf_xattr_list,
958 .moo_xattr_get = obf_xattr_get,
959 .moo_xattr_set = obf_xattr_set,
960 .moo_xattr_del = obf_xattr_del,
961 .moo_open = obf_mdd_open,
962 .moo_close = obf_mdd_close,
963 .moo_readpage = obf_mdd_readpage,
968 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
969 * are valid. We also check if object with passed fid exists or not.
971 static int obf_lookup(const struct lu_env *env, struct md_object *p,
972 const struct lu_name *lname, struct lu_fid *f,
973 struct md_op_spec *spec)
975 char *name = (char *)lname->ln_name;
976 struct mdd_device *mdd = mdo2mdd(p);
977 struct mdd_object *child;
983 sscanf(name, SFID, RFID(f));
984 if (!fid_is_sane(f)) {
985 CWARN("%s: bad FID format [%s], should be "DFID"\n",
986 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
987 (__u64)FID_SEQ_NORMAL, 1, 0);
988 GOTO(out, rc = -EINVAL);
991 if (!fid_is_norm(f)) {
992 CWARN("%s: "DFID" is invalid, sequence should be "
993 ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
994 (__u64)FID_SEQ_NORMAL);
995 GOTO(out, rc = -EINVAL);
998 /* Check if object with this fid exists */
999 child = mdd_object_find(env, mdd, f);
1003 GOTO(out, rc = PTR_ERR(child));
1005 if (mdd_object_exists(child) == 0)
1008 mdd_object_put(env, child);
1014 static int obf_create(const struct lu_env *env, struct md_object *pobj,
1015 const struct lu_name *lname, struct md_object *child,
1016 struct md_op_spec *spec, struct md_attr* ma)
1021 static int obf_rename(const struct lu_env *env,
1022 struct md_object *src_pobj, struct md_object *tgt_pobj,
1023 const struct lu_fid *lf, const struct lu_name *lsname,
1024 struct md_object *tobj, const struct lu_name *ltname,
1030 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
1031 struct md_object *src_obj, const struct lu_name *lname,
1037 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
1038 struct md_object *cobj, const struct lu_name *lname,
1044 static struct md_dir_operations mdd_obf_dir_ops = {
1045 .mdo_lookup = obf_lookup,
1046 .mdo_create = obf_create,
1047 .mdo_rename = obf_rename,
1048 .mdo_link = obf_link,
1049 .mdo_unlink = obf_unlink
1053 * Create special in-memory "fid" object for open-by-fid.
1055 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
1057 struct mdd_object *mdd_obf;
1058 struct lu_object *obf_lu_obj;
1061 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
1063 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
1064 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
1065 GOTO(out, rc = -ENOENT);
1067 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
1068 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1069 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
1070 /* Don't allow objects to be created in "fid" dir */
1071 mdd_obf->mod_flags |= IMMUTE_OBJ;
1073 obf_lu_obj = mdd2lu_obj(mdd_obf);
1074 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
1080 /** Setup ".lustre" directory object */
1081 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1083 struct dt_object *dt_dot_lustre;
1084 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
1088 rc = create_dot_lustre_dir(env, m);
1092 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
1093 dot_lustre_name, fid);
1094 if (IS_ERR(dt_dot_lustre)) {
1095 rc = PTR_ERR(dt_dot_lustre);
1099 /* references are released in mdd_device_shutdown() */
1100 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
1103 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
1104 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
1106 rc = mdd_obf_setup(env, m);
1108 CERROR("Error initializing \"fid\" object - %d.\n", rc);
1114 static int mdd_process_config(const struct lu_env *env,
1115 struct lu_device *d, struct lustre_cfg *cfg)
1117 struct mdd_device *m = lu2mdd_dev(d);
1118 struct dt_device *dt = m->mdd_child;
1119 struct lu_device *next = &dt->dd_lu_dev;
1123 switch (cfg->lcfg_command) {
1125 struct lprocfs_static_vars lvars;
1127 lprocfs_mdd_init_vars(&lvars);
1128 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1129 if (rc > 0 || rc == -ENOSYS)
1130 /* we don't understand; pass it on */
1131 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1135 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1138 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1141 mdd_lfsck_cleanup(env, m);
1142 mdd_changelog_fini(env, m);
1143 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1144 lu_dev_del_linkage(d->ld_site, d);
1145 mdd_device_shutdown(env, m, cfg);
1148 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1155 static int mdd_recovery_complete(const struct lu_env *env,
1156 struct lu_device *d)
1158 struct mdd_device *mdd = lu2mdd_dev(d);
1159 struct lu_device *next;
1163 LASSERT(mdd != NULL);
1164 next = &mdd->mdd_child->dd_lu_dev;
1166 /* XXX: orphans handling. */
1167 __mdd_orphan_cleanup(env, mdd);
1168 rc = next->ld_ops->ldo_recovery_complete(env, next);
1173 static int mdd_prepare(const struct lu_env *env,
1174 struct lu_device *pdev,
1175 struct lu_device *cdev)
1177 struct mdd_device *mdd = lu2mdd_dev(cdev);
1178 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1179 struct dt_object *root;
1184 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1188 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1189 &mdd->mdd_root_fid);
1190 if (!IS_ERR(root)) {
1191 LASSERT(root != NULL);
1192 lu_object_put(env, &root->do_lu);
1193 rc = orph_index_init(env, mdd);
1200 rc = mdd_dot_lustre_setup(env, mdd);
1202 CERROR("Error(%d) initializing .lustre objects\n", rc);
1206 /* we use capa file to declare llog changes,
1207 * will be fixed with new llog in 2.3 */
1208 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1210 GOTO(out, rc = PTR_ERR(root));
1212 mdd->mdd_capa = root;
1214 rc = mdd_changelog_init(env, mdd);
1218 rc = mdd_lfsck_setup(env, mdd);
1220 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1221 mdd2obd_dev(mdd)->obd_name, rc);
1229 const struct lu_device_operations mdd_lu_ops = {
1230 .ldo_object_alloc = mdd_object_alloc,
1231 .ldo_process_config = mdd_process_config,
1232 .ldo_recovery_complete = mdd_recovery_complete,
1233 .ldo_prepare = mdd_prepare,
1237 * No permission check is needed.
1239 static int mdd_root_get(const struct lu_env *env,
1240 struct md_device *m, struct lu_fid *f)
1242 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1245 *f = mdd->mdd_root_fid;
1250 * No permission check is needed.
1252 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1253 struct obd_statfs *sfs)
1255 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1260 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1266 * No permission check is needed.
1268 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1269 int mode, unsigned long timeout, __u32 alg,
1270 struct lustre_capa_key *keys)
1272 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1276 /* need barrier for mds_capa_keys access. */
1278 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1279 timeout, alg, keys);
1283 static int mdd_update_capa_key(const struct lu_env *env,
1284 struct md_device *m,
1285 struct lustre_capa_key *key)
1287 /* we do not support capabilities ... */
1291 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1294 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1296 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1297 return (*h == NULL ? -ENOENT : 0);
1300 static struct lu_device *mdd_device_free(const struct lu_env *env,
1301 struct lu_device *lu)
1303 struct mdd_device *m = lu2mdd_dev(lu);
1306 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1307 md_device_fini(&m->mdd_md_dev);
1312 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1313 struct lu_device_type *t,
1314 struct lustre_cfg *lcfg)
1316 struct lu_device *l;
1317 struct mdd_device *m;
1321 l = ERR_PTR(-ENOMEM);
1326 md_device_init(&m->mdd_md_dev, t);
1327 rc = mdd_init0(env, m, t, lcfg);
1329 mdd_device_free(env, l);
1338 * we use exports to track all mdd users
1340 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1341 struct obd_device *obd, struct obd_uuid *cluuid,
1342 struct obd_connect_data *data, void *localdata)
1344 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1345 struct lustre_handle conn;
1349 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1351 rc = class_connect(&conn, obd, cluuid);
1355 *exp = class_conn2export(&conn);
1357 /* Why should there ever be more than 1 connect? */
1358 LASSERT(mdd->mdd_connects == 0);
1359 mdd->mdd_connects++;
1365 * once last export (we don't count self-export) disappeared
1366 * mdd can be released
1368 static int mdd_obd_disconnect(struct obd_export *exp)
1370 struct obd_device *obd = exp->exp_obd;
1371 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1372 int rc, release = 0;
1375 mdd->mdd_connects--;
1376 if (mdd->mdd_connects == 0)
1379 rc = class_disconnect(exp);
1381 if (rc == 0 && release)
1382 class_manual_cleanup(obd);
1386 static int mdd_obd_health_check(const struct lu_env *env,
1387 struct obd_device *obd)
1389 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1394 rc = obd_health_check(env, mdd->mdd_child_exp->exp_obd);
1398 static struct obd_ops mdd_obd_device_ops = {
1399 .o_owner = THIS_MODULE,
1400 .o_connect = mdd_obd_connect,
1401 .o_disconnect = mdd_obd_disconnect,
1402 .o_health_check = mdd_obd_health_check
1406 * context key constructor/destructor:
1407 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1409 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1411 struct lu_context_key mdd_capainfo_key = {
1412 .lct_tags = LCT_SESSION,
1413 .lct_init = mdd_capainfo_key_init,
1414 .lct_fini = mdd_capainfo_key_fini
1417 struct md_capainfo *md_capainfo(const struct lu_env *env)
1419 /* NB, in mdt_init0 */
1420 if (env->le_ses == NULL)
1422 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1424 EXPORT_SYMBOL(md_capainfo);
1426 static int mdd_changelog_user_register(const struct lu_env *env,
1427 struct mdd_device *mdd, int *id)
1429 struct llog_ctxt *ctxt;
1430 struct llog_changelog_user_rec *rec;
1434 ctxt = llog_get_context(mdd2obd_dev(mdd),
1435 LLOG_CHANGELOG_USER_ORIG_CTXT);
1441 llog_ctxt_put(ctxt);
1445 /* Assume we want it on since somebody registered */
1446 rc = mdd_changelog_on(env, mdd, 1);
1450 rec->cur_hdr.lrh_len = sizeof(*rec);
1451 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1452 spin_lock(&mdd->mdd_cl.mc_user_lock);
1453 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1454 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1455 CERROR("Maximum number of changelog users exceeded!\n");
1456 GOTO(out, rc = -EOVERFLOW);
1458 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1459 rec->cur_endrec = mdd->mdd_cl.mc_index;
1460 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1462 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
1464 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1467 llog_ctxt_put(ctxt);
1471 struct mdd_changelog_user_data {
1472 __u64 mcud_endrec; /**< purge record for this user */
1473 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1475 __u32 mcud_minid; /**< user id with lowest rec reference */
1476 __u32 mcud_usercount;
1479 #define MCUD_UNREGISTER -1LL
1482 * 1. Find the smallest record everyone is willing to purge
1483 * 2. Update the last purgeable record for this user
1485 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1486 struct llog_handle *llh,
1487 struct llog_rec_hdr *hdr, void *data)
1489 struct llog_changelog_user_rec *rec;
1490 struct mdd_changelog_user_data *mcud = data;
1495 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1497 rec = (struct llog_changelog_user_rec *)hdr;
1499 mcud->mcud_usercount++;
1501 /* If we have a new endrec for this id, use it for the following
1502 min check instead of its old value */
1503 if (rec->cur_id == mcud->mcud_id)
1504 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1506 /* Track the minimum referenced record */
1507 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1508 mcud->mcud_minid = rec->cur_id;
1509 mcud->mcud_minrec = rec->cur_endrec;
1512 if (rec->cur_id != mcud->mcud_id)
1515 /* Update this user's record */
1516 mcud->mcud_found = 1;
1518 /* Special case: unregister this user */
1519 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1520 struct llog_cookie cookie;
1522 cookie.lgc_lgl = llh->lgh_id;
1523 cookie.lgc_index = hdr->lrh_index;
1525 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1528 mcud->mcud_usercount--;
1533 /* Update the endrec */
1534 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1535 mcud->mcud_id, rec->cur_endrec);
1537 /* hdr+1 is loc of data */
1538 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1539 rc = llog_write(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1545 static int mdd_changelog_user_purge(const struct lu_env *env,
1546 struct mdd_device *mdd, int id,
1549 struct mdd_changelog_user_data data;
1550 struct llog_ctxt *ctxt;
1554 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1557 data.mcud_minid = 0;
1558 data.mcud_minrec = 0;
1559 data.mcud_usercount = 0;
1560 data.mcud_endrec = endrec;
1561 spin_lock(&mdd->mdd_cl.mc_lock);
1562 endrec = mdd->mdd_cl.mc_index;
1563 spin_unlock(&mdd->mdd_cl.mc_lock);
1564 if ((data.mcud_endrec == 0) ||
1565 ((data.mcud_endrec > endrec) &&
1566 (data.mcud_endrec != MCUD_UNREGISTER)))
1567 data.mcud_endrec = endrec;
1569 ctxt = llog_get_context(mdd2obd_dev(mdd),
1570 LLOG_CHANGELOG_USER_ORIG_CTXT);
1574 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1576 rc = llog_cat_process(env, ctxt->loc_handle,
1577 mdd_changelog_user_purge_cb, (void *)&data,
1579 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1580 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1581 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1582 data.mcud_minrec, data.mcud_minid);
1583 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1585 CWARN("Could not determine changelog records to purge; rc=%d\n",
1589 llog_ctxt_put(ctxt);
1591 if (!data.mcud_found) {
1592 CWARN("No entry for user %d. Last changelog reference is "
1593 LPD64" by changelog user %d\n", data.mcud_id,
1594 data.mcud_minrec, data.mcud_minid);
1598 if (!rc && data.mcud_usercount == 0)
1599 /* No more users; turn changelogs off */
1600 rc = mdd_changelog_on(env, mdd, 0);
1606 * May be called remotely from mdt_iocontrol_handle or locally from
1607 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1608 * an obd_ioctl_data format (but local ioctl handler does).
1610 * \param len - data len
1611 * \param karg - ioctl data, in kernel space
1613 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1614 unsigned int cmd, int len, void *karg)
1616 struct mdd_device *mdd;
1617 struct obd_ioctl_data *data = karg;
1621 mdd = lu2mdd_dev(&m->md_lu_dev);
1623 /* Doesn't use obd_ioctl_data */
1625 case OBD_IOC_CHANGELOG_CLEAR: {
1626 struct changelog_setinfo *cs = karg;
1627 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1631 case OBD_IOC_GET_MNTOPT: {
1632 mntopt_t *mntopts = (mntopt_t *)karg;
1633 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1636 case OBD_IOC_START_LFSCK: {
1637 struct lfsck_start *start = karg;
1638 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1640 /* Return the kernel service version. */
1641 /* XXX: version can be used for compatibility in the future. */
1642 start->ls_version = lfsck->ml_version;
1643 rc = mdd_lfsck_start(env, lfsck, start);
1646 case OBD_IOC_STOP_LFSCK: {
1647 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1652 /* Below ioctls use obd_ioctl_data */
1653 if (len != sizeof(*data)) {
1654 CERROR("Bad ioctl size %d\n", len);
1657 if (data->ioc_version != OBD_IOCTL_VERSION) {
1658 CERROR("Bad magic %x != %x\n", data->ioc_version,
1664 case OBD_IOC_CHANGELOG_REG:
1665 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1667 case OBD_IOC_CHANGELOG_DEREG:
1668 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1678 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1679 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_capainfo_key);
1681 const struct md_device_operations mdd_ops = {
1682 .mdo_statfs = mdd_statfs,
1683 .mdo_root_get = mdd_root_get,
1684 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1685 .mdo_update_capa_key= mdd_update_capa_key,
1686 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1687 .mdo_iocontrol = mdd_iocontrol,
1690 static struct lu_device_type_operations mdd_device_type_ops = {
1691 .ldto_init = mdd_type_init,
1692 .ldto_fini = mdd_type_fini,
1694 .ldto_start = mdd_type_start,
1695 .ldto_stop = mdd_type_stop,
1697 .ldto_device_alloc = mdd_device_alloc,
1698 .ldto_device_free = mdd_device_free,
1700 .ldto_device_fini = mdd_device_fini
1703 static struct lu_device_type mdd_device_type = {
1704 .ldt_tags = LU_DEVICE_MD,
1705 .ldt_name = LUSTRE_MDD_NAME,
1706 .ldt_ops = &mdd_device_type_ops,
1707 .ldt_ctx_tags = LCT_MD_THREAD
1710 /* context key constructor: mdd_key_init */
1711 LU_KEY_INIT(mdd, struct mdd_thread_info);
1713 static void mdd_key_fini(const struct lu_context *ctx,
1714 struct lu_context_key *key, void *data)
1716 struct mdd_thread_info *info = data;
1717 if (info->mti_max_lmm != NULL)
1718 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1719 if (info->mti_max_cookie != NULL)
1720 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1721 mdd_buf_put(&info->mti_big_buf);
1726 /* context key: mdd_thread_key */
1727 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1729 static struct lu_local_obj_desc llod_capa_key = {
1730 .llod_name = CAPA_KEYS,
1731 .llod_oid = MDD_CAPA_KEYS_OID,
1735 static struct lu_local_obj_desc llod_mdd_orphan = {
1736 .llod_name = orph_index_name,
1737 .llod_oid = MDD_ORPHAN_OID,
1739 .llod_feat = &dt_directory_features,
1742 static struct lu_local_obj_desc llod_mdd_root = {
1743 .llod_name = mdd_root_dir_name,
1744 .llod_oid = MDD_ROOT_INDEX_OID,
1746 .llod_feat = &dt_directory_features,
1749 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1750 .llod_name = lfsck_bookmark_name,
1751 .llod_oid = LFSCK_BOOKMARK_OID,
1755 static int __init mdd_mod_init(void)
1757 struct lprocfs_static_vars lvars;
1760 lprocfs_mdd_init_vars(&lvars);
1762 rc = lu_kmem_init(mdd_caches);
1766 changelog_orig_logops = llog_osd_ops;
1767 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
1768 changelog_orig_logops.lop_add = llog_cat_add_rec;
1769 changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1771 llo_local_obj_register(&llod_capa_key);
1772 llo_local_obj_register(&llod_mdd_orphan);
1773 llo_local_obj_register(&llod_mdd_root);
1774 llo_local_obj_register(&llod_lfsck_bookmark);
1776 rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1777 LUSTRE_MDD_NAME, &mdd_device_type);
1779 lu_kmem_fini(mdd_caches);
1783 static void __exit mdd_mod_exit(void)
1785 llo_local_obj_unregister(&llod_capa_key);
1786 llo_local_obj_unregister(&llod_mdd_orphan);
1787 llo_local_obj_unregister(&llod_mdd_root);
1788 llo_local_obj_unregister(&llod_lfsck_bookmark);
1790 class_unregister_type(LUSTRE_MDD_NAME);
1791 lu_kmem_fini(mdd_caches);
1794 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1795 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1796 MODULE_LICENSE("GPL");
1798 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);