4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_device.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #include <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 mdd_changelog_fini(env, m);
171 if (m->mdd_dot_lustre_objs.mdd_obf)
172 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
173 if (m->mdd_dot_lustre)
174 mdd_object_put(env, m->mdd_dot_lustre);
175 orph_index_fini(env, m);
176 if (m->mdd_capa != NULL) {
177 lu_object_put(env, &m->mdd_capa->do_lu);
180 lu_site_purge(env, m->mdd_md_dev.md_lu_dev.ld_site, -1);
181 /* remove upcall device*/
182 md_upcall_fini(&m->mdd_md_dev);
184 if (m->mdd_child_exp)
185 obd_disconnect(m->mdd_child_exp);
190 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
191 struct llog_rec_hdr *hdr, void *data)
193 struct mdd_device *mdd = (struct mdd_device *)data;
194 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
196 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
197 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
200 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
201 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
202 rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
203 llh->lgh_id.lgl_oid);
205 mdd->mdd_cl.mc_index = rec->cr.cr_index;
206 return LLOG_PROC_BREAK;
209 static int changelog_user_init_cb(const struct lu_env *env,
210 struct llog_handle *llh,
211 struct llog_rec_hdr *hdr, void *data)
213 struct mdd_device *mdd = (struct mdd_device *)data;
214 struct llog_changelog_user_rec *rec =
215 (struct llog_changelog_user_rec *)hdr;
217 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
218 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
220 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
221 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
222 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
224 spin_lock(&mdd->mdd_cl.mc_user_lock);
225 mdd->mdd_cl.mc_lastuser = rec->cur_id;
226 spin_unlock(&mdd->mdd_cl.mc_user_lock);
228 return LLOG_PROC_BREAK;
231 static int llog_changelog_cancel_cb(const struct lu_env *env,
232 struct llog_handle *llh,
233 struct llog_rec_hdr *hdr, void *data)
235 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
236 struct llog_cookie cookie;
237 long long endrec = *(long long *)data;
242 /* This is always a (sub)log, not the catalog */
243 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
245 if (rec->cr.cr_index > endrec)
246 /* records are in order, so we're done */
247 RETURN(LLOG_PROC_BREAK);
249 cookie.lgc_lgl = llh->lgh_id;
250 cookie.lgc_index = hdr->lrh_index;
252 /* cancel them one at a time. I suppose we could store up the cookies
253 * and cancel them all at once; probably more efficient, but this is
254 * done as a user call, so who cares... */
255 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
257 RETURN(rc < 0 ? rc : 0);
260 static int llog_changelog_cancel(const struct lu_env *env,
261 struct llog_ctxt *ctxt,
262 struct lov_stripe_md *lsm, int count,
263 struct llog_cookie *cookies, int flags)
265 struct llog_handle *cathandle = ctxt->loc_handle;
270 /* This should only be called with the catalog handle */
271 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
273 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
274 (void *)cookies, 0, 0);
276 /* 0 or 1 means we're done */
279 CERROR("%s: cancel idx %u of catalog "LPX64" rc=%d\n",
280 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
281 cathandle->lgh_id.lgl_oid, rc);
286 static struct llog_operations changelog_orig_logops;
288 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
290 static int mdd_changelog_llog_init(const struct lu_env *env,
291 struct mdd_device *mdd)
293 struct obd_device *obd = mdd2obd_dev(mdd);
294 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
298 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
299 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
300 rc = dt_root_get(env, mdd->mdd_bottom, &rfid);
304 changelog_orig_logops = llog_osd_ops;
305 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
306 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
307 obd, &changelog_orig_logops);
309 CERROR("%s: changelog llog setup failed: rc = %d\n",
314 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
317 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
320 GOTO(out_cleanup, rc);
322 ctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
323 ctxt->loc_handle->lgh_logops->lop_declare_add =
324 llog_cat_declare_add_rec;
326 rc = llog_cat_init_and_process(env, ctxt->loc_handle);
330 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
331 changelog_init_cb, mdd);
334 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
339 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
340 mdd->mdd_cl.mc_index);
342 /* setup user changelog */
343 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
344 obd, &changelog_orig_logops);
346 CERROR("%s: changelog users llog setup failed: rc = %d\n",
351 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
354 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
357 GOTO(out_ucleanup, rc);
359 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
360 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
362 rc = llog_cat_init_and_process(env, uctxt->loc_handle);
364 GOTO(out_uclose, rc);
366 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
367 changelog_user_init_cb, mdd);
369 CERROR("%s: changelog user init failed: rc = %d\n",
371 GOTO(out_uclose, rc);
374 /* If we have registered users, assume we want changelogs on */
375 if (mdd->mdd_cl.mc_lastuser > 0) {
376 rc = mdd_changelog_on(env, mdd, 1);
378 GOTO(out_uclose, rc);
381 llog_ctxt_put(uctxt);
384 llog_cat_close(env, uctxt->loc_handle);
386 llog_cleanup(env, uctxt);
388 llog_cat_close(env, ctxt->loc_handle);
390 llog_cleanup(env, ctxt);
394 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
396 struct obd_device *obd = mdd2obd_dev(mdd);
399 mdd->mdd_cl.mc_index = 0;
400 spin_lock_init(&mdd->mdd_cl.mc_lock);
401 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
402 mdd->mdd_cl.mc_flags = 0; /* off by default */
403 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
404 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
405 mdd->mdd_cl.mc_lastuser = 0;
407 rc = mdd_changelog_llog_init(env, mdd);
409 CERROR("%s: changelog setup during init failed: rc = %d\n",
411 mdd->mdd_cl.mc_flags |= CLM_ERR;
417 static void mdd_changelog_fini(const struct lu_env *env,
418 struct mdd_device *mdd)
420 struct obd_device *obd = mdd2obd_dev(mdd);
421 struct llog_ctxt *ctxt;
423 mdd->mdd_cl.mc_flags = 0;
425 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
427 llog_cat_close(env, ctxt->loc_handle);
428 llog_cleanup(env, ctxt);
430 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
432 llog_cat_close(env, ctxt->loc_handle);
433 llog_cleanup(env, ctxt);
437 int mdd_changelog_write_header(const struct lu_env *env,
438 struct mdd_device *mdd, int markerflags);
440 /* Start / stop recording */
441 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
445 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
446 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
447 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
448 CERROR("Changelogs cannot be enabled due to error "
449 "condition (see %s log).\n",
450 mdd2obd_dev(mdd)->obd_name);
453 spin_lock(&mdd->mdd_cl.mc_lock);
454 mdd->mdd_cl.mc_flags |= CLM_ON;
455 spin_unlock(&mdd->mdd_cl.mc_lock);
456 rc = mdd_changelog_write_header(env, mdd, CLM_START);
458 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
459 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
460 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
461 spin_lock(&mdd->mdd_cl.mc_lock);
462 mdd->mdd_cl.mc_flags &= ~CLM_ON;
463 spin_unlock(&mdd->mdd_cl.mc_lock);
468 /** Remove entries with indicies up to and including \a endrec from the
474 int mdd_changelog_llog_cancel(const struct lu_env *env,
475 struct mdd_device *mdd, long long endrec)
477 struct obd_device *obd = mdd2obd_dev(mdd);
478 struct llog_ctxt *ctxt;
479 long long unsigned cur;
482 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
486 spin_lock(&mdd->mdd_cl.mc_lock);
487 cur = (long long)mdd->mdd_cl.mc_index;
488 spin_unlock(&mdd->mdd_cl.mc_lock);
492 /* purge to "0" is shorthand for everything */
496 /* If purging all records, write a header entry so we don't have an
497 empty catalog and we're sure to have a valid starting index next
498 time. In case of crash, we just restart with old log so we're
501 /* XXX: transaction is started by llog itself */
502 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
507 /* Some records were purged, so reset repeat-access time (so we
508 record new mtime update records, so users can see a file has been
509 changed since the last purge) */
510 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
512 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
518 /** Add a CL_MARK record to the changelog
520 * \param markerflags - CLM_*
523 int mdd_changelog_write_header(const struct lu_env *env,
524 struct mdd_device *mdd, int markerflags)
526 struct obd_device *obd = mdd2obd_dev(mdd);
527 struct llog_changelog_rec *rec;
529 struct llog_ctxt *ctxt;
531 int len = strlen(obd->obd_name);
536 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
537 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
541 reclen = llog_data_len(sizeof(*rec) + len);
542 buf = mdd_buf_alloc(env, reclen);
543 if (buf->lb_buf == NULL)
547 rec->cr.cr_flags = CLF_VERSION;
548 rec->cr.cr_type = CL_MARK;
549 rec->cr.cr_namelen = len;
550 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
551 /* Status and action flags */
552 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
553 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
554 rec->cr_hdr.lrh_type = CHANGELOG_REC;
555 rec->cr.cr_time = cl_time();
556 spin_lock(&mdd->mdd_cl.mc_lock);
557 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
558 spin_unlock(&mdd->mdd_cl.mc_lock);
560 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
563 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
568 /* assume on or off event; reset repeat-access time */
569 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
574 * Create ".lustre" directory.
576 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
578 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
579 struct md_object *mdo;
582 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
583 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
584 mdd_root_dir_name, dot_lustre_name,
585 fid, &dt_directory_features);
586 /* .lustre dir may be already present */
587 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
589 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
590 dot_lustre_name, PFID(fid), rc);
595 lu_object_put(env, &mdo->mo_lu);
601 static int dot_lustre_mdd_permission(const struct lu_env *env,
602 struct md_object *pobj,
603 struct md_object *cobj,
604 struct md_attr *attr, int mask)
606 if (mask & ~(MAY_READ | MAY_EXEC))
612 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
613 struct md_object *obj, struct lu_buf *buf,
619 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
620 struct md_object *obj, struct lu_buf *buf)
625 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
626 struct md_object *obj,
627 const struct lu_buf *buf, const char *name,
633 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
634 struct md_object *obj,
640 static int dot_lustre_mdd_swap_layouts(const struct lu_env *env,
641 struct md_object *obj1,
642 struct md_object *obj2,
648 static int dot_lustre_mdd_readlink(const struct lu_env *env,
649 struct md_object *obj, struct lu_buf *buf)
654 static int dot_lustre_mdd_object_create(const struct lu_env *env,
655 struct md_object *obj,
656 const struct md_op_spec *spec,
662 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
663 struct md_object *obj,
664 const struct md_attr *ma)
669 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
670 struct md_object *obj,
676 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
679 struct mdd_object *mdd_obj = md2mdd_obj(obj);
681 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
682 mdd_obj->mod_count++;
683 mdd_write_unlock(env, mdd_obj);
688 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
689 struct md_attr *ma, int mode)
691 struct mdd_object *mdd_obj = md2mdd_obj(obj);
693 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
694 mdd_obj->mod_count--;
695 mdd_write_unlock(env, mdd_obj);
700 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
701 struct md_object *obj)
706 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
707 char *path, int pathlen, __u64 *recno, int *linkno)
712 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
713 struct lov_mds_md *lmm, struct ldlm_extent *extent,
714 struct lustre_handle *lockh)
719 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
720 struct lov_mds_md *lmm, struct lustre_handle *lockh)
725 static struct md_object_operations mdd_dot_lustre_obj_ops = {
726 .moo_permission = dot_lustre_mdd_permission,
727 .moo_attr_get = mdd_attr_get,
728 .moo_attr_set = mdd_attr_set,
729 .moo_xattr_get = dot_lustre_mdd_xattr_get,
730 .moo_xattr_list = dot_lustre_mdd_xattr_list,
731 .moo_xattr_set = dot_lustre_mdd_xattr_set,
732 .moo_xattr_del = dot_lustre_mdd_xattr_del,
733 .moo_swap_layouts = dot_lustre_mdd_swap_layouts,
734 .moo_readpage = mdd_readpage,
735 .moo_readlink = dot_lustre_mdd_readlink,
736 .moo_object_create = dot_lustre_mdd_object_create,
737 .moo_ref_add = dot_lustre_mdd_ref_add,
738 .moo_ref_del = dot_lustre_mdd_ref_del,
739 .moo_open = dot_lustre_mdd_open,
740 .moo_close = dot_lustre_mdd_close,
741 .moo_capa_get = mdd_capa_get,
742 .moo_object_sync = dot_lustre_mdd_object_sync,
743 .moo_path = dot_lustre_mdd_path,
744 .moo_file_lock = dot_file_lock,
745 .moo_file_unlock = dot_file_unlock,
749 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
750 const struct lu_name *lname, struct lu_fid *f,
751 struct md_op_spec *spec)
753 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
761 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
762 struct md_object *obj,
768 static int dot_lustre_mdd_create(const struct lu_env *env,
769 struct md_object *pobj,
770 const struct lu_name *lname,
771 struct md_object *child,
772 struct md_op_spec *spec,
778 static int dot_lustre_mdd_create_data(const struct lu_env *env,
781 const struct md_op_spec *spec,
787 static int dot_lustre_mdd_rename(const struct lu_env *env,
788 struct md_object *src_pobj,
789 struct md_object *tgt_pobj,
790 const struct lu_fid *lf,
791 const struct lu_name *lsname,
792 struct md_object *tobj,
793 const struct lu_name *ltname,
799 static int dot_lustre_mdd_link(const struct lu_env *env,
800 struct md_object *tgt_obj,
801 struct md_object *src_obj,
802 const struct lu_name *lname,
808 static int dot_lustre_mdd_unlink(const struct lu_env *env,
809 struct md_object *pobj,
810 struct md_object *cobj,
811 const struct lu_name *lname,
817 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
818 struct md_object *obj,
819 const struct lu_name *lname,
820 const struct lu_fid *fid,
821 const struct md_attr *ma)
826 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
827 struct md_object *obj,
828 const struct lu_name *lname,
829 const struct md_attr *ma)
834 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
835 struct md_object *pobj,
836 struct md_object *tobj,
837 const struct lu_fid *fid,
838 const struct lu_name *lname,
845 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
846 .mdo_is_subdir = mdd_is_subdir,
847 .mdo_lookup = dot_lustre_mdd_lookup,
848 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
849 .mdo_create = dot_lustre_mdd_create,
850 .mdo_create_data = dot_lustre_mdd_create_data,
851 .mdo_rename = dot_lustre_mdd_rename,
852 .mdo_link = dot_lustre_mdd_link,
853 .mdo_unlink = dot_lustre_mdd_unlink,
854 .mdo_name_insert = dot_lustre_mdd_name_insert,
855 .mdo_name_remove = dot_lustre_mdd_name_remove,
856 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
859 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
862 struct mdd_device *mdd = mdo2mdd(obj);
864 /* "fid" is a virtual object and hence does not have any "real"
865 * attributes. So we reuse attributes of .lustre for "fid" dir */
866 return mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
869 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
870 const struct md_attr *ma)
875 static int obf_xattr_list(const struct lu_env *env,
876 struct md_object *obj, struct lu_buf *buf)
881 static int obf_xattr_get(const struct lu_env *env,
882 struct md_object *obj, struct lu_buf *buf,
885 struct mdd_device *mdd = mdo2mdd(obj);
886 struct mdd_object *root;
887 struct lu_fid rootfid;
891 * .lustre returns default striping which is 'stored'
894 if (strcmp(name, XATTR_NAME_LOV) == 0) {
895 rc = dt_root_get(env, mdd->mdd_child, &rootfid);
898 root = mdd_object_find(env, mdd, &rootfid);
900 return PTR_ERR(root);
901 rc = mdo_xattr_get(env, root, buf, name,
902 mdd_object_capa(env, md2mdd_obj(obj)));
903 mdd_object_put(env, root);
909 static int obf_xattr_set(const struct lu_env *env,
910 struct md_object *obj,
911 const struct lu_buf *buf, const char *name,
917 static int obf_xattr_del(const struct lu_env *env,
918 struct md_object *obj,
924 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
927 struct mdd_object *mdd_obj = md2mdd_obj(obj);
929 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
930 mdd_obj->mod_count++;
931 mdd_write_unlock(env, mdd_obj);
936 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
937 struct md_attr *ma, int mode)
939 struct mdd_object *mdd_obj = md2mdd_obj(obj);
941 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
942 mdd_obj->mod_count--;
943 mdd_write_unlock(env, mdd_obj);
948 /** Nothing to list in "fid" directory */
949 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
950 const struct lu_rdpg *rdpg)
955 static int obf_path(const struct lu_env *env, struct md_object *obj,
956 char *path, int pathlen, __u64 *recno, int *linkno)
961 static struct md_object_operations mdd_obf_obj_ops = {
962 .moo_attr_get = obf_attr_get,
963 .moo_attr_set = obf_attr_set,
964 .moo_xattr_list = obf_xattr_list,
965 .moo_xattr_get = obf_xattr_get,
966 .moo_xattr_set = obf_xattr_set,
967 .moo_xattr_del = obf_xattr_del,
968 .moo_open = obf_mdd_open,
969 .moo_close = obf_mdd_close,
970 .moo_readpage = obf_mdd_readpage,
975 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
976 * are valid. We also check if object with passed fid exists or not.
978 static int obf_lookup(const struct lu_env *env, struct md_object *p,
979 const struct lu_name *lname, struct lu_fid *f,
980 struct md_op_spec *spec)
982 char *name = (char *)lname->ln_name;
983 struct mdd_device *mdd = mdo2mdd(p);
984 struct mdd_object *child;
990 sscanf(name, SFID, RFID(f));
991 if (!fid_is_sane(f)) {
992 CWARN("%s: bad FID format [%s], should be "DFID"\n",
993 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
994 (__u64)FID_SEQ_NORMAL, 1, 0);
995 GOTO(out, rc = -EINVAL);
998 if (!fid_is_norm(f)) {
999 CWARN("%s: "DFID" is invalid, sequence should be "
1000 ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
1001 (__u64)FID_SEQ_NORMAL);
1002 GOTO(out, rc = -EINVAL);
1005 /* Check if object with this fid exists */
1006 child = mdd_object_find(env, mdd, f);
1010 GOTO(out, rc = PTR_ERR(child));
1012 if (mdd_object_exists(child) == 0)
1015 mdd_object_put(env, child);
1021 static int obf_create(const struct lu_env *env, struct md_object *pobj,
1022 const struct lu_name *lname, struct md_object *child,
1023 struct md_op_spec *spec, struct md_attr* ma)
1028 static int obf_rename(const struct lu_env *env,
1029 struct md_object *src_pobj, struct md_object *tgt_pobj,
1030 const struct lu_fid *lf, const struct lu_name *lsname,
1031 struct md_object *tobj, const struct lu_name *ltname,
1037 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
1038 struct md_object *src_obj, const struct lu_name *lname,
1044 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
1045 struct md_object *cobj, const struct lu_name *lname,
1051 static struct md_dir_operations mdd_obf_dir_ops = {
1052 .mdo_lookup = obf_lookup,
1053 .mdo_create = obf_create,
1054 .mdo_rename = obf_rename,
1055 .mdo_link = obf_link,
1056 .mdo_unlink = obf_unlink
1060 * Create special in-memory "fid" object for open-by-fid.
1062 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
1064 struct mdd_object *mdd_obf;
1065 struct lu_object *obf_lu_obj;
1068 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
1070 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
1071 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
1072 GOTO(out, rc = -ENOENT);
1074 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
1075 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1076 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
1077 /* Don't allow objects to be created in "fid" dir */
1078 mdd_obf->mod_flags |= IMMUTE_OBJ;
1080 obf_lu_obj = mdd2lu_obj(mdd_obf);
1081 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
1087 /** Setup ".lustre" directory object */
1088 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1090 struct dt_object *dt_dot_lustre;
1091 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
1094 rc = create_dot_lustre_dir(env, m);
1098 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
1099 dot_lustre_name, fid);
1100 if (IS_ERR(dt_dot_lustre)) {
1101 rc = PTR_ERR(dt_dot_lustre);
1105 /* references are released in mdd_device_shutdown() */
1106 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
1109 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
1110 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
1112 rc = mdd_obf_setup(env, m);
1114 CERROR("Error initializing \"fid\" object - %d.\n", rc);
1120 static int mdd_process_config(const struct lu_env *env,
1121 struct lu_device *d, struct lustre_cfg *cfg)
1123 struct mdd_device *m = lu2mdd_dev(d);
1124 struct dt_device *dt = m->mdd_child;
1125 struct lu_device *next = &dt->dd_lu_dev;
1129 switch (cfg->lcfg_command) {
1131 struct lprocfs_static_vars lvars;
1133 lprocfs_mdd_init_vars(&lvars);
1134 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1135 if (rc > 0 || rc == -ENOSYS)
1136 /* we don't understand; pass it on */
1137 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1141 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1144 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1147 mdd_lfsck_cleanup(env, m);
1148 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1149 lu_dev_del_linkage(d->ld_site, d);
1150 mdd_device_shutdown(env, m, cfg);
1153 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1160 static int mdd_recovery_complete(const struct lu_env *env,
1161 struct lu_device *d)
1163 struct mdd_device *mdd = lu2mdd_dev(d);
1164 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1168 LASSERT(mdd != NULL);
1170 /* XXX: orphans handling. */
1171 __mdd_orphan_cleanup(env, mdd);
1172 rc = next->ld_ops->ldo_recovery_complete(env, next);
1177 static int mdd_prepare(const struct lu_env *env,
1178 struct lu_device *pdev,
1179 struct lu_device *cdev)
1181 struct mdd_device *mdd = lu2mdd_dev(cdev);
1182 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1183 struct dt_object *root;
1188 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1192 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1193 &mdd->mdd_root_fid);
1194 if (!IS_ERR(root)) {
1195 LASSERT(root != NULL);
1196 lu_object_put(env, &root->do_lu);
1197 rc = orph_index_init(env, mdd);
1204 rc = mdd_dot_lustre_setup(env, mdd);
1206 CERROR("Error(%d) initializing .lustre objects\n", rc);
1210 /* we use capa file to declare llog changes,
1211 * will be fixed with new llog in 2.3 */
1212 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1214 GOTO(out, rc = PTR_ERR(root));
1216 mdd->mdd_capa = root;
1218 rc = mdd_lfsck_setup(env, mdd);
1220 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1221 mdd2obd_dev(mdd)->obd_name, rc);
1224 rc = mdd_changelog_init(env, mdd);
1232 const struct lu_device_operations mdd_lu_ops = {
1233 .ldo_object_alloc = mdd_object_alloc,
1234 .ldo_process_config = mdd_process_config,
1235 .ldo_recovery_complete = mdd_recovery_complete,
1236 .ldo_prepare = mdd_prepare,
1240 * No permission check is needed.
1242 static int mdd_root_get(const struct lu_env *env,
1243 struct md_device *m, struct lu_fid *f)
1245 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1248 *f = mdd->mdd_root_fid;
1253 * No permission check is needed.
1255 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1256 struct obd_statfs *sfs)
1258 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1263 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1269 * No permission check is needed.
1271 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1272 int mode, unsigned long timeout, __u32 alg,
1273 struct lustre_capa_key *keys)
1275 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1279 /* need barrier for mds_capa_keys access. */
1281 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1282 timeout, alg, keys);
1286 static int mdd_update_capa_key(const struct lu_env *env,
1287 struct md_device *m,
1288 struct lustre_capa_key *key)
1290 /* we do not support capabilities ... */
1294 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1297 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1299 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1300 return (*h == NULL ? -ENOENT : 0);
1303 static struct lu_device *mdd_device_free(const struct lu_env *env,
1304 struct lu_device *lu)
1306 struct mdd_device *m = lu2mdd_dev(lu);
1309 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1310 md_device_fini(&m->mdd_md_dev);
1315 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1316 struct lu_device_type *t,
1317 struct lustre_cfg *lcfg)
1319 struct lu_device *l;
1320 struct mdd_device *m;
1324 l = ERR_PTR(-ENOMEM);
1329 md_device_init(&m->mdd_md_dev, t);
1330 rc = mdd_init0(env, m, t, lcfg);
1332 mdd_device_free(env, l);
1341 * we use exports to track all mdd users
1343 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1344 struct obd_device *obd, struct obd_uuid *cluuid,
1345 struct obd_connect_data *data, void *localdata)
1347 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1348 struct lustre_handle conn;
1352 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1354 rc = class_connect(&conn, obd, cluuid);
1358 *exp = class_conn2export(&conn);
1360 /* Why should there ever be more than 1 connect? */
1361 LASSERT(mdd->mdd_connects == 0);
1362 mdd->mdd_connects++;
1368 * once last export (we don't count self-export) disappeared
1369 * mdd can be released
1371 static int mdd_obd_disconnect(struct obd_export *exp)
1373 struct obd_device *obd = exp->exp_obd;
1374 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1375 int rc, release = 0;
1378 mdd->mdd_connects--;
1379 if (mdd->mdd_connects == 0)
1382 rc = class_disconnect(exp);
1384 if (rc == 0 && release)
1385 class_manual_cleanup(obd);
1389 static int mdd_obd_health_check(const struct lu_env *env,
1390 struct obd_device *obd)
1392 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1397 rc = obd_health_check(env, mdd->mdd_child_exp->exp_obd);
1401 static struct obd_ops mdd_obd_device_ops = {
1402 .o_owner = THIS_MODULE,
1403 .o_connect = mdd_obd_connect,
1404 .o_disconnect = mdd_obd_disconnect,
1405 .o_health_check = mdd_obd_health_check
1409 * context key constructor/destructor:
1410 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1412 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1414 struct lu_context_key mdd_capainfo_key = {
1415 .lct_tags = LCT_SESSION,
1416 .lct_init = mdd_capainfo_key_init,
1417 .lct_fini = mdd_capainfo_key_fini
1420 struct md_capainfo *md_capainfo(const struct lu_env *env)
1422 /* NB, in mdt_init0 */
1423 if (env->le_ses == NULL)
1425 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1427 EXPORT_SYMBOL(md_capainfo);
1429 static int mdd_changelog_user_register(const struct lu_env *env,
1430 struct mdd_device *mdd, int *id)
1432 struct llog_ctxt *ctxt;
1433 struct llog_changelog_user_rec *rec;
1437 ctxt = llog_get_context(mdd2obd_dev(mdd),
1438 LLOG_CHANGELOG_USER_ORIG_CTXT);
1444 llog_ctxt_put(ctxt);
1448 /* Assume we want it on since somebody registered */
1449 rc = mdd_changelog_on(env, mdd, 1);
1453 rec->cur_hdr.lrh_len = sizeof(*rec);
1454 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1455 spin_lock(&mdd->mdd_cl.mc_user_lock);
1456 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1457 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1458 CERROR("Maximum number of changelog users exceeded!\n");
1459 GOTO(out, rc = -EOVERFLOW);
1461 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1462 rec->cur_endrec = mdd->mdd_cl.mc_index;
1463 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1465 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
1467 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1470 llog_ctxt_put(ctxt);
1474 struct mdd_changelog_user_data {
1475 __u64 mcud_endrec; /**< purge record for this user */
1476 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1478 __u32 mcud_minid; /**< user id with lowest rec reference */
1479 __u32 mcud_usercount;
1482 #define MCUD_UNREGISTER -1LL
1485 * 1. Find the smallest record everyone is willing to purge
1486 * 2. Update the last purgeable record for this user
1488 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1489 struct llog_handle *llh,
1490 struct llog_rec_hdr *hdr, void *data)
1492 struct llog_changelog_user_rec *rec;
1493 struct mdd_changelog_user_data *mcud = data;
1498 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1500 rec = (struct llog_changelog_user_rec *)hdr;
1502 mcud->mcud_usercount++;
1504 /* If we have a new endrec for this id, use it for the following
1505 min check instead of its old value */
1506 if (rec->cur_id == mcud->mcud_id)
1507 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1509 /* Track the minimum referenced record */
1510 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1511 mcud->mcud_minid = rec->cur_id;
1512 mcud->mcud_minrec = rec->cur_endrec;
1515 if (rec->cur_id != mcud->mcud_id)
1518 /* Update this user's record */
1519 mcud->mcud_found = 1;
1521 /* Special case: unregister this user */
1522 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1523 struct llog_cookie cookie;
1525 cookie.lgc_lgl = llh->lgh_id;
1526 cookie.lgc_index = hdr->lrh_index;
1528 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1531 mcud->mcud_usercount--;
1536 /* Update the endrec */
1537 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1538 mcud->mcud_id, rec->cur_endrec);
1540 /* hdr+1 is loc of data */
1541 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1542 rc = llog_write(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1548 static int mdd_changelog_user_purge(const struct lu_env *env,
1549 struct mdd_device *mdd, int id,
1552 struct mdd_changelog_user_data data;
1553 struct llog_ctxt *ctxt;
1557 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1560 data.mcud_minid = 0;
1561 data.mcud_minrec = 0;
1562 data.mcud_usercount = 0;
1563 data.mcud_endrec = endrec;
1564 spin_lock(&mdd->mdd_cl.mc_lock);
1565 endrec = mdd->mdd_cl.mc_index;
1566 spin_unlock(&mdd->mdd_cl.mc_lock);
1567 if ((data.mcud_endrec == 0) ||
1568 ((data.mcud_endrec > endrec) &&
1569 (data.mcud_endrec != MCUD_UNREGISTER)))
1570 data.mcud_endrec = endrec;
1572 ctxt = llog_get_context(mdd2obd_dev(mdd),
1573 LLOG_CHANGELOG_USER_ORIG_CTXT);
1577 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1579 rc = llog_cat_process(env, ctxt->loc_handle,
1580 mdd_changelog_user_purge_cb, (void *)&data,
1582 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1583 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1584 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1585 data.mcud_minrec, data.mcud_minid);
1586 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1588 CWARN("Could not determine changelog records to purge; rc=%d\n",
1592 llog_ctxt_put(ctxt);
1594 if (!data.mcud_found) {
1595 CWARN("No entry for user %d. Last changelog reference is "
1596 LPD64" by changelog user %d\n", data.mcud_id,
1597 data.mcud_minrec, data.mcud_minid);
1601 if (!rc && data.mcud_usercount == 0)
1602 /* No more users; turn changelogs off */
1603 rc = mdd_changelog_on(env, mdd, 0);
1609 * May be called remotely from mdt_iocontrol_handle or locally from
1610 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1611 * an obd_ioctl_data format (but local ioctl handler does).
1613 * \param len - data len
1614 * \param karg - ioctl data, in kernel space
1616 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1617 unsigned int cmd, int len, void *karg)
1619 struct mdd_device *mdd;
1620 struct obd_ioctl_data *data = karg;
1624 mdd = lu2mdd_dev(&m->md_lu_dev);
1626 /* Doesn't use obd_ioctl_data */
1628 case OBD_IOC_CHANGELOG_CLEAR: {
1629 struct changelog_setinfo *cs = karg;
1630 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1634 case OBD_IOC_GET_MNTOPT: {
1635 mntopt_t *mntopts = (mntopt_t *)karg;
1636 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1639 case OBD_IOC_START_LFSCK: {
1640 struct lfsck_start *start = karg;
1641 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1643 /* Return the kernel service version. */
1644 /* XXX: version can be used for compatibility in the future. */
1645 start->ls_version = lfsck->ml_version;
1646 rc = mdd_lfsck_start(env, lfsck, start);
1649 case OBD_IOC_STOP_LFSCK: {
1650 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck);
1655 /* Below ioctls use obd_ioctl_data */
1656 if (len != sizeof(*data)) {
1657 CERROR("Bad ioctl size %d\n", len);
1660 if (data->ioc_version != OBD_IOCTL_VERSION) {
1661 CERROR("Bad magic %x != %x\n", data->ioc_version,
1667 case OBD_IOC_CHANGELOG_REG:
1668 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1670 case OBD_IOC_CHANGELOG_DEREG:
1671 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1681 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1682 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_capainfo_key);
1684 const struct md_device_operations mdd_ops = {
1685 .mdo_statfs = mdd_statfs,
1686 .mdo_root_get = mdd_root_get,
1687 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1688 .mdo_update_capa_key= mdd_update_capa_key,
1689 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1690 .mdo_iocontrol = mdd_iocontrol,
1693 static struct lu_device_type_operations mdd_device_type_ops = {
1694 .ldto_init = mdd_type_init,
1695 .ldto_fini = mdd_type_fini,
1697 .ldto_start = mdd_type_start,
1698 .ldto_stop = mdd_type_stop,
1700 .ldto_device_alloc = mdd_device_alloc,
1701 .ldto_device_free = mdd_device_free,
1703 .ldto_device_fini = mdd_device_fini
1706 static struct lu_device_type mdd_device_type = {
1707 .ldt_tags = LU_DEVICE_MD,
1708 .ldt_name = LUSTRE_MDD_NAME,
1709 .ldt_ops = &mdd_device_type_ops,
1710 .ldt_ctx_tags = LCT_MD_THREAD
1713 /* context key constructor: mdd_key_init */
1714 LU_KEY_INIT(mdd, struct mdd_thread_info);
1716 static void mdd_key_fini(const struct lu_context *ctx,
1717 struct lu_context_key *key, void *data)
1719 struct mdd_thread_info *info = data;
1720 if (info->mti_max_lmm != NULL)
1721 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1722 if (info->mti_max_cookie != NULL)
1723 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1724 mdd_buf_put(&info->mti_big_buf);
1729 /* context key: mdd_thread_key */
1730 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1732 static struct lu_local_obj_desc llod_capa_key = {
1733 .llod_name = CAPA_KEYS,
1734 .llod_oid = MDD_CAPA_KEYS_OID,
1738 static struct lu_local_obj_desc llod_mdd_orphan = {
1739 .llod_name = orph_index_name,
1740 .llod_oid = MDD_ORPHAN_OID,
1742 .llod_feat = &dt_directory_features,
1745 static struct lu_local_obj_desc llod_mdd_root = {
1746 .llod_name = mdd_root_dir_name,
1747 .llod_oid = MDD_ROOT_INDEX_OID,
1749 .llod_feat = &dt_directory_features,
1752 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1753 .llod_name = lfsck_bookmark_name,
1754 .llod_oid = LFSCK_BOOKMARK_OID,
1758 static int __init mdd_mod_init(void)
1760 struct lprocfs_static_vars lvars;
1763 lprocfs_mdd_init_vars(&lvars);
1765 rc = lu_kmem_init(mdd_caches);
1769 llo_local_obj_register(&llod_capa_key);
1770 llo_local_obj_register(&llod_mdd_orphan);
1771 llo_local_obj_register(&llod_mdd_root);
1772 llo_local_obj_register(&llod_lfsck_bookmark);
1774 rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1775 LUSTRE_MDD_NAME, &mdd_device_type);
1777 lu_kmem_fini(mdd_caches);
1781 static void __exit mdd_mod_exit(void)
1783 llo_local_obj_unregister(&llod_capa_key);
1784 llo_local_obj_unregister(&llod_mdd_orphan);
1785 llo_local_obj_unregister(&llod_mdd_root);
1786 llo_local_obj_unregister(&llod_lfsck_bookmark);
1788 class_unregister_type(LUSTRE_MDD_NAME);
1789 lu_kmem_fini(mdd_caches);
1792 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1793 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1794 MODULE_LICENSE("GPL");
1796 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);