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;
298 /* LU-2844 mdd setup failure should not cause umount oops */
299 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
302 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
303 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
304 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
305 obd, &changelog_orig_logops);
307 CERROR("%s: changelog llog setup failed: rc = %d\n",
312 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
315 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
318 GOTO(out_cleanup, rc);
320 rc = llog_cat_init_and_process(env, ctxt->loc_handle);
324 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
325 changelog_init_cb, mdd);
328 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
333 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
334 mdd->mdd_cl.mc_index);
336 /* setup user changelog */
337 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
338 obd, &changelog_orig_logops);
340 CERROR("%s: changelog users llog setup failed: rc = %d\n",
345 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
348 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
351 GOTO(out_ucleanup, rc);
353 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
354 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
356 rc = llog_cat_init_and_process(env, uctxt->loc_handle);
358 GOTO(out_uclose, rc);
360 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
361 changelog_user_init_cb, mdd);
363 CERROR("%s: changelog user init failed: rc = %d\n",
365 GOTO(out_uclose, rc);
368 /* If we have registered users, assume we want changelogs on */
369 if (mdd->mdd_cl.mc_lastuser > 0) {
370 rc = mdd_changelog_on(env, mdd, 1);
372 GOTO(out_uclose, rc);
375 llog_ctxt_put(uctxt);
378 llog_cat_close(env, uctxt->loc_handle);
380 llog_cleanup(env, uctxt);
382 llog_cat_close(env, ctxt->loc_handle);
384 llog_cleanup(env, ctxt);
388 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
390 struct obd_device *obd = mdd2obd_dev(mdd);
393 mdd->mdd_cl.mc_index = 0;
394 spin_lock_init(&mdd->mdd_cl.mc_lock);
395 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
396 mdd->mdd_cl.mc_flags = 0; /* off by default */
397 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
398 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
399 mdd->mdd_cl.mc_lastuser = 0;
401 rc = mdd_changelog_llog_init(env, mdd);
403 CERROR("%s: changelog setup during init failed: rc = %d\n",
405 mdd->mdd_cl.mc_flags |= CLM_ERR;
411 static void mdd_changelog_fini(const struct lu_env *env,
412 struct mdd_device *mdd)
414 struct obd_device *obd = mdd2obd_dev(mdd);
415 struct llog_ctxt *ctxt;
417 mdd->mdd_cl.mc_flags = 0;
419 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
421 llog_cat_close(env, ctxt->loc_handle);
422 llog_cleanup(env, ctxt);
424 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
426 llog_cat_close(env, ctxt->loc_handle);
427 llog_cleanup(env, ctxt);
431 int mdd_changelog_write_header(const struct lu_env *env,
432 struct mdd_device *mdd, int markerflags);
434 /* Start / stop recording */
435 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
439 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
440 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
441 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
442 CERROR("Changelogs cannot be enabled due to error "
443 "condition (see %s log).\n",
444 mdd2obd_dev(mdd)->obd_name);
447 spin_lock(&mdd->mdd_cl.mc_lock);
448 mdd->mdd_cl.mc_flags |= CLM_ON;
449 spin_unlock(&mdd->mdd_cl.mc_lock);
450 rc = mdd_changelog_write_header(env, mdd, CLM_START);
452 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
453 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
454 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
455 spin_lock(&mdd->mdd_cl.mc_lock);
456 mdd->mdd_cl.mc_flags &= ~CLM_ON;
457 spin_unlock(&mdd->mdd_cl.mc_lock);
462 /** Remove entries with indicies up to and including \a endrec from the
468 int mdd_changelog_llog_cancel(const struct lu_env *env,
469 struct mdd_device *mdd, long long endrec)
471 struct obd_device *obd = mdd2obd_dev(mdd);
472 struct llog_ctxt *ctxt;
473 long long unsigned cur;
476 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
480 spin_lock(&mdd->mdd_cl.mc_lock);
481 cur = (long long)mdd->mdd_cl.mc_index;
482 spin_unlock(&mdd->mdd_cl.mc_lock);
486 /* purge to "0" is shorthand for everything */
490 /* If purging all records, write a header entry so we don't have an
491 empty catalog and we're sure to have a valid starting index next
492 time. In case of crash, we just restart with old log so we're
495 /* XXX: transaction is started by llog itself */
496 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
501 /* Some records were purged, so reset repeat-access time (so we
502 record new mtime update records, so users can see a file has been
503 changed since the last purge) */
504 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
506 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
512 /** Add a CL_MARK record to the changelog
514 * \param markerflags - CLM_*
517 int mdd_changelog_write_header(const struct lu_env *env,
518 struct mdd_device *mdd, int markerflags)
520 struct obd_device *obd = mdd2obd_dev(mdd);
521 struct llog_changelog_rec *rec;
523 struct llog_ctxt *ctxt;
525 int len = strlen(obd->obd_name);
530 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
531 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
535 reclen = llog_data_len(sizeof(*rec) + len);
536 buf = mdd_buf_alloc(env, reclen);
537 if (buf->lb_buf == NULL)
541 rec->cr.cr_flags = CLF_VERSION;
542 rec->cr.cr_type = CL_MARK;
543 rec->cr.cr_namelen = len;
544 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
545 /* Status and action flags */
546 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
547 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
548 rec->cr_hdr.lrh_type = CHANGELOG_REC;
549 rec->cr.cr_time = cl_time();
550 spin_lock(&mdd->mdd_cl.mc_lock);
551 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
552 spin_unlock(&mdd->mdd_cl.mc_lock);
554 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
557 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
562 /* assume on or off event; reset repeat-access time */
563 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
568 * Create ".lustre" directory.
570 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
572 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
573 struct md_object *mdo;
576 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
577 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
578 mdd_root_dir_name, dot_lustre_name,
579 fid, &dt_directory_features);
580 /* .lustre dir may be already present */
581 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
583 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
584 dot_lustre_name, PFID(fid), rc);
589 lu_object_put(env, &mdo->mo_lu);
595 static int dot_lustre_mdd_permission(const struct lu_env *env,
596 struct md_object *pobj,
597 struct md_object *cobj,
598 struct md_attr *attr, int mask)
600 if (mask & ~(MAY_READ | MAY_EXEC))
606 static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
607 struct md_object *obj, struct lu_buf *buf,
613 static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
614 struct md_object *obj, struct lu_buf *buf)
619 static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
620 struct md_object *obj,
621 const struct lu_buf *buf, const char *name,
627 static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
628 struct md_object *obj,
634 static int dot_lustre_mdd_swap_layouts(const struct lu_env *env,
635 struct md_object *obj1,
636 struct md_object *obj2,
642 static int dot_lustre_mdd_readlink(const struct lu_env *env,
643 struct md_object *obj, struct lu_buf *buf)
648 static int dot_lustre_mdd_object_create(const struct lu_env *env,
649 struct md_object *obj,
650 const struct md_op_spec *spec,
656 static int dot_lustre_mdd_ref_add(const struct lu_env *env,
657 struct md_object *obj,
658 const struct md_attr *ma)
663 static int dot_lustre_mdd_ref_del(const struct lu_env *env,
664 struct md_object *obj,
670 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
673 struct mdd_object *mdd_obj = md2mdd_obj(obj);
675 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
676 mdd_obj->mod_count++;
677 mdd_write_unlock(env, mdd_obj);
682 static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
683 struct md_attr *ma, int mode)
685 struct mdd_object *mdd_obj = md2mdd_obj(obj);
687 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
688 mdd_obj->mod_count--;
689 mdd_write_unlock(env, mdd_obj);
694 static int dot_lustre_mdd_object_sync(const struct lu_env *env,
695 struct md_object *obj)
700 static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
701 char *path, int pathlen, __u64 *recno, int *linkno)
706 static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
707 struct lov_mds_md *lmm, struct ldlm_extent *extent,
708 struct lustre_handle *lockh)
713 static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
714 struct lov_mds_md *lmm, struct lustre_handle *lockh)
719 static struct md_object_operations mdd_dot_lustre_obj_ops = {
720 .moo_permission = dot_lustre_mdd_permission,
721 .moo_attr_get = mdd_attr_get,
722 .moo_attr_set = mdd_attr_set,
723 .moo_xattr_get = dot_lustre_mdd_xattr_get,
724 .moo_xattr_list = dot_lustre_mdd_xattr_list,
725 .moo_xattr_set = dot_lustre_mdd_xattr_set,
726 .moo_xattr_del = dot_lustre_mdd_xattr_del,
727 .moo_swap_layouts = dot_lustre_mdd_swap_layouts,
728 .moo_readpage = mdd_readpage,
729 .moo_readlink = dot_lustre_mdd_readlink,
730 .moo_object_create = dot_lustre_mdd_object_create,
731 .moo_ref_add = dot_lustre_mdd_ref_add,
732 .moo_ref_del = dot_lustre_mdd_ref_del,
733 .moo_open = dot_lustre_mdd_open,
734 .moo_close = dot_lustre_mdd_close,
735 .moo_capa_get = mdd_capa_get,
736 .moo_object_sync = dot_lustre_mdd_object_sync,
737 .moo_path = dot_lustre_mdd_path,
738 .moo_file_lock = dot_file_lock,
739 .moo_file_unlock = dot_file_unlock,
743 static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
744 const struct lu_name *lname, struct lu_fid *f,
745 struct md_op_spec *spec)
747 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0) {
755 static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
756 struct md_object *obj,
762 static int dot_lustre_mdd_create(const struct lu_env *env,
763 struct md_object *pobj,
764 const struct lu_name *lname,
765 struct md_object *child,
766 struct md_op_spec *spec,
772 static int dot_lustre_mdd_create_data(const struct lu_env *env,
775 const struct md_op_spec *spec,
781 static int dot_lustre_mdd_rename(const struct lu_env *env,
782 struct md_object *src_pobj,
783 struct md_object *tgt_pobj,
784 const struct lu_fid *lf,
785 const struct lu_name *lsname,
786 struct md_object *tobj,
787 const struct lu_name *ltname,
793 static int dot_lustre_mdd_link(const struct lu_env *env,
794 struct md_object *tgt_obj,
795 struct md_object *src_obj,
796 const struct lu_name *lname,
802 static int dot_lustre_mdd_unlink(const struct lu_env *env,
803 struct md_object *pobj,
804 struct md_object *cobj,
805 const struct lu_name *lname,
811 static int dot_lustre_mdd_name_insert(const struct lu_env *env,
812 struct md_object *obj,
813 const struct lu_name *lname,
814 const struct lu_fid *fid,
815 const struct md_attr *ma)
820 static int dot_lustre_mdd_name_remove(const struct lu_env *env,
821 struct md_object *obj,
822 const struct lu_name *lname,
823 const struct md_attr *ma)
828 static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
829 struct md_object *pobj,
830 struct md_object *tobj,
831 const struct lu_fid *fid,
832 const struct lu_name *lname,
839 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
840 .mdo_is_subdir = mdd_is_subdir,
841 .mdo_lookup = dot_lustre_mdd_lookup,
842 .mdo_lock_mode = dot_lustre_mdd_lock_mode,
843 .mdo_create = dot_lustre_mdd_create,
844 .mdo_create_data = dot_lustre_mdd_create_data,
845 .mdo_rename = dot_lustre_mdd_rename,
846 .mdo_link = dot_lustre_mdd_link,
847 .mdo_unlink = dot_lustre_mdd_unlink,
848 .mdo_name_insert = dot_lustre_mdd_name_insert,
849 .mdo_name_remove = dot_lustre_mdd_name_remove,
850 .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
853 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
856 struct mdd_device *mdd = mdo2mdd(obj);
858 /* "fid" is a virtual object and hence does not have any "real"
859 * attributes. So we reuse attributes of .lustre for "fid" dir */
860 return mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
863 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
864 const struct md_attr *ma)
869 static int obf_xattr_list(const struct lu_env *env,
870 struct md_object *obj, struct lu_buf *buf)
875 static int obf_xattr_get(const struct lu_env *env,
876 struct md_object *obj, struct lu_buf *buf,
879 struct mdd_device *mdd = mdo2mdd(obj);
880 struct mdd_object *root;
884 * .lustre returns default striping which is 'stored'
887 if (strcmp(name, XATTR_NAME_LOV) == 0) {
888 root = mdd_object_find(env, mdd, &mdd->mdd_local_root_fid);
890 return PTR_ERR(root);
891 rc = mdo_xattr_get(env, root, buf, name,
892 mdd_object_capa(env, md2mdd_obj(obj)));
893 mdd_object_put(env, root);
899 static int obf_xattr_set(const struct lu_env *env,
900 struct md_object *obj,
901 const struct lu_buf *buf, const char *name,
907 static int obf_xattr_del(const struct lu_env *env,
908 struct md_object *obj,
914 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
917 struct mdd_object *mdd_obj = md2mdd_obj(obj);
919 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
920 mdd_obj->mod_count++;
921 mdd_write_unlock(env, mdd_obj);
926 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
927 struct md_attr *ma, int mode)
929 struct mdd_object *mdd_obj = md2mdd_obj(obj);
931 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
932 mdd_obj->mod_count--;
933 mdd_write_unlock(env, mdd_obj);
938 /** Nothing to list in "fid" directory */
939 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
940 const struct lu_rdpg *rdpg)
945 static int obf_path(const struct lu_env *env, struct md_object *obj,
946 char *path, int pathlen, __u64 *recno, int *linkno)
951 static struct md_object_operations mdd_obf_obj_ops = {
952 .moo_attr_get = obf_attr_get,
953 .moo_attr_set = obf_attr_set,
954 .moo_xattr_list = obf_xattr_list,
955 .moo_xattr_get = obf_xattr_get,
956 .moo_xattr_set = obf_xattr_set,
957 .moo_xattr_del = obf_xattr_del,
958 .moo_open = obf_mdd_open,
959 .moo_close = obf_mdd_close,
960 .moo_readpage = obf_mdd_readpage,
965 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
966 * are valid. We also check if object with passed fid exists or not.
968 static int obf_lookup(const struct lu_env *env, struct md_object *p,
969 const struct lu_name *lname, struct lu_fid *f,
970 struct md_op_spec *spec)
972 char *name = (char *)lname->ln_name;
973 struct mdd_device *mdd = mdo2mdd(p);
974 struct mdd_object *child;
980 sscanf(name, SFID, RFID(f));
981 if (!fid_is_sane(f)) {
982 CWARN("%s: bad FID format [%s], should be "DFID"\n",
983 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
984 (__u64)FID_SEQ_NORMAL, 1, 0);
985 GOTO(out, rc = -EINVAL);
988 if (!fid_is_norm(f)) {
989 CWARN("%s: "DFID" is invalid, sequence should be "
990 ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
991 (__u64)FID_SEQ_NORMAL);
992 GOTO(out, rc = -EINVAL);
995 /* Check if object with this fid exists */
996 child = mdd_object_find(env, mdd, f);
1000 GOTO(out, rc = PTR_ERR(child));
1002 if (mdd_object_exists(child) == 0)
1005 mdd_object_put(env, child);
1011 static int obf_create(const struct lu_env *env, struct md_object *pobj,
1012 const struct lu_name *lname, struct md_object *child,
1013 struct md_op_spec *spec, struct md_attr* ma)
1018 static int obf_rename(const struct lu_env *env,
1019 struct md_object *src_pobj, struct md_object *tgt_pobj,
1020 const struct lu_fid *lf, const struct lu_name *lsname,
1021 struct md_object *tobj, const struct lu_name *ltname,
1027 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
1028 struct md_object *src_obj, const struct lu_name *lname,
1034 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
1035 struct md_object *cobj, const struct lu_name *lname,
1041 static struct md_dir_operations mdd_obf_dir_ops = {
1042 .mdo_lookup = obf_lookup,
1043 .mdo_create = obf_create,
1044 .mdo_rename = obf_rename,
1045 .mdo_link = obf_link,
1046 .mdo_unlink = obf_unlink
1050 * Create special in-memory "fid" object for open-by-fid.
1052 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
1054 struct mdd_object *mdd_obf;
1055 struct lu_object *obf_lu_obj;
1058 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
1060 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
1061 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
1062 GOTO(out, rc = -ENOENT);
1064 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
1065 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1066 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
1067 /* Don't allow objects to be created in "fid" dir */
1068 mdd_obf->mod_flags |= IMMUTE_OBJ;
1070 obf_lu_obj = mdd2lu_obj(mdd_obf);
1071 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
1077 /** Setup ".lustre" directory object */
1078 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1080 struct dt_object *dt_dot_lustre;
1081 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
1085 rc = create_dot_lustre_dir(env, m);
1089 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
1090 dot_lustre_name, fid);
1091 if (IS_ERR(dt_dot_lustre)) {
1092 rc = PTR_ERR(dt_dot_lustre);
1096 /* references are released in mdd_device_shutdown() */
1097 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
1100 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
1101 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
1103 rc = mdd_obf_setup(env, m);
1105 CERROR("Error initializing \"fid\" object - %d.\n", rc);
1111 static int mdd_process_config(const struct lu_env *env,
1112 struct lu_device *d, struct lustre_cfg *cfg)
1114 struct mdd_device *m = lu2mdd_dev(d);
1115 struct dt_device *dt = m->mdd_child;
1116 struct lu_device *next = &dt->dd_lu_dev;
1120 switch (cfg->lcfg_command) {
1122 struct lprocfs_static_vars lvars;
1124 lprocfs_mdd_init_vars(&lvars);
1125 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
1126 if (rc > 0 || rc == -ENOSYS)
1127 /* we don't understand; pass it on */
1128 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1132 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1135 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
1138 mdd_lfsck_cleanup(env, m);
1139 mdd_changelog_fini(env, m);
1140 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1141 lu_dev_del_linkage(d->ld_site, d);
1142 mdd_device_shutdown(env, m, cfg);
1145 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1152 static int mdd_recovery_complete(const struct lu_env *env,
1153 struct lu_device *d)
1155 struct mdd_device *mdd = lu2mdd_dev(d);
1156 struct lu_device *next;
1160 LASSERT(mdd != NULL);
1161 next = &mdd->mdd_child->dd_lu_dev;
1163 /* XXX: orphans handling. */
1164 __mdd_orphan_cleanup(env, mdd);
1165 rc = next->ld_ops->ldo_recovery_complete(env, next);
1170 static int mdd_find_or_create_root(const struct lu_env *env,
1171 struct mdd_device *mdd)
1173 struct dt_object *root;
1174 struct md_object *mroot;
1180 /* Check if the "ROOT" entry exists already */
1181 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
1183 if (!IS_ERR(root)) {
1184 lu_object_put(env, &root->do_lu);
1189 /* New Filesystem, create /ROOT */
1190 mroot = llo_store_create_index(env, &mdd->mdd_md_dev, mdd->mdd_bottom,
1191 "", mdd_root_dir_name, &fid,
1192 &dt_directory_features);
1194 GOTO(out, rc = PTR_ERR(mroot));
1196 lu_object_put(env, &mroot->mo_lu);
1199 mdd->mdd_root_fid = fid;
1204 static int mdd_prepare(const struct lu_env *env,
1205 struct lu_device *pdev,
1206 struct lu_device *cdev)
1208 struct mdd_device *mdd = lu2mdd_dev(cdev);
1209 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1210 struct dt_object *root;
1216 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1220 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1224 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1225 rc = mdd_find_or_create_root(env, mdd);
1227 CERROR("%s: create root fid failed: rc = %d\n",
1228 mdd2obd_dev(mdd)->obd_name, rc);
1232 rc = mdd_dot_lustre_setup(env, mdd);
1234 CERROR("%s: initializing .lustre failed: rc = %d\n",
1235 mdd2obd_dev(mdd)->obd_name, rc);
1240 rc = orph_index_init(env, mdd);
1244 /* we use capa file to declare llog changes,
1245 * will be fixed with new llog in 2.3 */
1246 root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid);
1248 GOTO(out, rc = PTR_ERR(root));
1250 mdd->mdd_capa = root;
1252 rc = mdd_changelog_init(env, mdd);
1256 rc = mdd_lfsck_setup(env, mdd);
1258 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1259 mdd2obd_dev(mdd)->obd_name, rc);
1264 const struct lu_device_operations mdd_lu_ops = {
1265 .ldo_object_alloc = mdd_object_alloc,
1266 .ldo_process_config = mdd_process_config,
1267 .ldo_recovery_complete = mdd_recovery_complete,
1268 .ldo_prepare = mdd_prepare,
1272 * No permission check is needed.
1274 static int mdd_root_get(const struct lu_env *env,
1275 struct md_device *m, struct lu_fid *f)
1277 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1280 *f = mdd->mdd_root_fid;
1285 * No permission check is needed.
1287 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1288 struct obd_statfs *sfs)
1290 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1295 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1301 * No permission check is needed.
1303 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1304 int mode, unsigned long timeout, __u32 alg,
1305 struct lustre_capa_key *keys)
1307 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1311 /* need barrier for mds_capa_keys access. */
1313 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1314 timeout, alg, keys);
1318 static int mdd_update_capa_key(const struct lu_env *env,
1319 struct md_device *m,
1320 struct lustre_capa_key *key)
1322 /* we do not support capabilities ... */
1326 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1329 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1331 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1332 return (*h == NULL ? -ENOENT : 0);
1335 static struct lu_device *mdd_device_free(const struct lu_env *env,
1336 struct lu_device *lu)
1338 struct mdd_device *m = lu2mdd_dev(lu);
1341 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
1342 md_device_fini(&m->mdd_md_dev);
1347 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1348 struct lu_device_type *t,
1349 struct lustre_cfg *lcfg)
1351 struct lu_device *l;
1352 struct mdd_device *m;
1356 l = ERR_PTR(-ENOMEM);
1361 md_device_init(&m->mdd_md_dev, t);
1362 rc = mdd_init0(env, m, t, lcfg);
1364 mdd_device_free(env, l);
1373 * we use exports to track all mdd users
1375 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1376 struct obd_device *obd, struct obd_uuid *cluuid,
1377 struct obd_connect_data *data, void *localdata)
1379 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1380 struct lustre_handle conn;
1384 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1386 rc = class_connect(&conn, obd, cluuid);
1390 *exp = class_conn2export(&conn);
1392 /* Why should there ever be more than 1 connect? */
1393 LASSERT(mdd->mdd_connects == 0);
1394 mdd->mdd_connects++;
1400 * once last export (we don't count self-export) disappeared
1401 * mdd can be released
1403 static int mdd_obd_disconnect(struct obd_export *exp)
1405 struct obd_device *obd = exp->exp_obd;
1406 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1407 int rc, release = 0;
1410 mdd->mdd_connects--;
1411 if (mdd->mdd_connects == 0)
1414 rc = class_disconnect(exp);
1416 if (rc == 0 && release)
1417 class_manual_cleanup(obd);
1421 static int mdd_obd_health_check(const struct lu_env *env,
1422 struct obd_device *obd)
1424 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1429 rc = obd_health_check(env, mdd->mdd_child_exp->exp_obd);
1433 static struct obd_ops mdd_obd_device_ops = {
1434 .o_owner = THIS_MODULE,
1435 .o_connect = mdd_obd_connect,
1436 .o_disconnect = mdd_obd_disconnect,
1437 .o_health_check = mdd_obd_health_check
1441 * context key constructor/destructor:
1442 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1444 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1446 struct lu_context_key mdd_capainfo_key = {
1447 .lct_tags = LCT_SESSION,
1448 .lct_init = mdd_capainfo_key_init,
1449 .lct_fini = mdd_capainfo_key_fini
1452 struct md_capainfo *md_capainfo(const struct lu_env *env)
1454 /* NB, in mdt_init0 */
1455 if (env->le_ses == NULL)
1457 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1459 EXPORT_SYMBOL(md_capainfo);
1461 static int mdd_changelog_user_register(const struct lu_env *env,
1462 struct mdd_device *mdd, int *id)
1464 struct llog_ctxt *ctxt;
1465 struct llog_changelog_user_rec *rec;
1469 ctxt = llog_get_context(mdd2obd_dev(mdd),
1470 LLOG_CHANGELOG_USER_ORIG_CTXT);
1476 llog_ctxt_put(ctxt);
1480 /* Assume we want it on since somebody registered */
1481 rc = mdd_changelog_on(env, mdd, 1);
1485 rec->cur_hdr.lrh_len = sizeof(*rec);
1486 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1487 spin_lock(&mdd->mdd_cl.mc_user_lock);
1488 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1489 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1490 CERROR("Maximum number of changelog users exceeded!\n");
1491 GOTO(out, rc = -EOVERFLOW);
1493 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1494 rec->cur_endrec = mdd->mdd_cl.mc_index;
1495 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1497 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
1499 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1502 llog_ctxt_put(ctxt);
1506 struct mdd_changelog_user_data {
1507 __u64 mcud_endrec; /**< purge record for this user */
1508 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1510 __u32 mcud_minid; /**< user id with lowest rec reference */
1511 __u32 mcud_usercount;
1512 unsigned int mcud_found:1;
1514 #define MCUD_UNREGISTER -1LL
1517 * 1. Find the smallest record everyone is willing to purge
1518 * 2. Update the last purgeable record for this user
1520 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1521 struct llog_handle *llh,
1522 struct llog_rec_hdr *hdr, void *data)
1524 struct llog_changelog_user_rec *rec;
1525 struct mdd_changelog_user_data *mcud = data;
1530 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1532 rec = (struct llog_changelog_user_rec *)hdr;
1534 mcud->mcud_usercount++;
1536 /* If we have a new endrec for this id, use it for the following
1537 min check instead of its old value */
1538 if (rec->cur_id == mcud->mcud_id)
1539 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1541 /* Track the minimum referenced record */
1542 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1543 mcud->mcud_minid = rec->cur_id;
1544 mcud->mcud_minrec = rec->cur_endrec;
1547 if (rec->cur_id != mcud->mcud_id)
1550 /* Update this user's record */
1551 mcud->mcud_found = 1;
1553 /* Special case: unregister this user */
1554 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1555 struct llog_cookie cookie;
1557 cookie.lgc_lgl = llh->lgh_id;
1558 cookie.lgc_index = hdr->lrh_index;
1560 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1563 mcud->mcud_usercount--;
1568 /* Update the endrec */
1569 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1570 mcud->mcud_id, rec->cur_endrec);
1572 /* hdr+1 is loc of data */
1573 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1574 rc = llog_write(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1580 static int mdd_changelog_user_purge(const struct lu_env *env,
1581 struct mdd_device *mdd, int id,
1584 struct mdd_changelog_user_data data;
1585 struct llog_ctxt *ctxt;
1589 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1592 data.mcud_minid = 0;
1593 data.mcud_minrec = 0;
1594 data.mcud_usercount = 0;
1595 data.mcud_endrec = endrec;
1596 spin_lock(&mdd->mdd_cl.mc_lock);
1597 endrec = mdd->mdd_cl.mc_index;
1598 spin_unlock(&mdd->mdd_cl.mc_lock);
1599 if ((data.mcud_endrec == 0) ||
1600 ((data.mcud_endrec > endrec) &&
1601 (data.mcud_endrec != MCUD_UNREGISTER)))
1602 data.mcud_endrec = endrec;
1604 ctxt = llog_get_context(mdd2obd_dev(mdd),
1605 LLOG_CHANGELOG_USER_ORIG_CTXT);
1609 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1611 rc = llog_cat_process(env, ctxt->loc_handle,
1612 mdd_changelog_user_purge_cb, (void *)&data,
1614 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1615 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1616 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1617 data.mcud_minrec, data.mcud_minid);
1618 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1620 CWARN("Could not determine changelog records to purge; rc=%d\n",
1624 llog_ctxt_put(ctxt);
1626 if (!data.mcud_found) {
1627 CWARN("No entry for user %d. Last changelog reference is "
1628 LPD64" by changelog user %d\n", data.mcud_id,
1629 data.mcud_minrec, data.mcud_minid);
1633 if (!rc && data.mcud_usercount == 0)
1634 /* No more users; turn changelogs off */
1635 rc = mdd_changelog_on(env, mdd, 0);
1641 * May be called remotely from mdt_iocontrol_handle or locally from
1642 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1643 * an obd_ioctl_data format (but local ioctl handler does).
1645 * \param len - data len
1646 * \param karg - ioctl data, in kernel space
1648 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1649 unsigned int cmd, int len, void *karg)
1651 struct mdd_device *mdd;
1652 struct obd_ioctl_data *data = karg;
1656 mdd = lu2mdd_dev(&m->md_lu_dev);
1658 /* Doesn't use obd_ioctl_data */
1660 case OBD_IOC_CHANGELOG_CLEAR: {
1661 struct changelog_setinfo *cs = karg;
1662 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1666 case OBD_IOC_GET_MNTOPT: {
1667 mntopt_t *mntopts = (mntopt_t *)karg;
1668 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1671 case OBD_IOC_START_LFSCK: {
1672 rc = mdd_lfsck_start(env, &mdd->mdd_lfsck,
1673 (struct lfsck_start *)karg);
1676 case OBD_IOC_STOP_LFSCK: {
1677 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck, false);
1680 case OBD_IOC_PAUSE_LFSCK: {
1681 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck, true);
1686 /* Below ioctls use obd_ioctl_data */
1687 if (len != sizeof(*data)) {
1688 CERROR("Bad ioctl size %d\n", len);
1691 if (data->ioc_version != OBD_IOCTL_VERSION) {
1692 CERROR("Bad magic %x != %x\n", data->ioc_version,
1698 case OBD_IOC_CHANGELOG_REG:
1699 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1701 case OBD_IOC_CHANGELOG_DEREG:
1702 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1712 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1713 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_capainfo_key);
1715 const struct md_device_operations mdd_ops = {
1716 .mdo_statfs = mdd_statfs,
1717 .mdo_root_get = mdd_root_get,
1718 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1719 .mdo_update_capa_key= mdd_update_capa_key,
1720 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1721 .mdo_iocontrol = mdd_iocontrol,
1724 static struct lu_device_type_operations mdd_device_type_ops = {
1725 .ldto_init = mdd_type_init,
1726 .ldto_fini = mdd_type_fini,
1728 .ldto_start = mdd_type_start,
1729 .ldto_stop = mdd_type_stop,
1731 .ldto_device_alloc = mdd_device_alloc,
1732 .ldto_device_free = mdd_device_free,
1734 .ldto_device_fini = mdd_device_fini
1737 static struct lu_device_type mdd_device_type = {
1738 .ldt_tags = LU_DEVICE_MD,
1739 .ldt_name = LUSTRE_MDD_NAME,
1740 .ldt_ops = &mdd_device_type_ops,
1741 .ldt_ctx_tags = LCT_MD_THREAD
1744 /* context key constructor: mdd_key_init */
1745 LU_KEY_INIT(mdd, struct mdd_thread_info);
1747 static void mdd_key_fini(const struct lu_context *ctx,
1748 struct lu_context_key *key, void *data)
1750 struct mdd_thread_info *info = data;
1751 if (info->mti_max_lmm != NULL)
1752 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1753 if (info->mti_max_cookie != NULL)
1754 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1755 mdd_buf_put(&info->mti_big_buf);
1760 /* context key: mdd_thread_key */
1761 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1763 static struct lu_local_obj_desc llod_capa_key = {
1764 .llod_name = CAPA_KEYS,
1765 .llod_oid = MDD_CAPA_KEYS_OID,
1769 static struct lu_local_obj_desc llod_mdd_orphan = {
1770 .llod_name = orph_index_name,
1771 .llod_oid = MDD_ORPHAN_OID,
1773 .llod_feat = &dt_directory_features,
1776 static struct lu_local_obj_desc llod_lfsck_bookmark = {
1777 .llod_name = lfsck_bookmark_name,
1778 .llod_oid = LFSCK_BOOKMARK_OID,
1782 static struct lu_local_obj_desc llod_lfsck_namespace = {
1783 .llod_name = lfsck_namespace_name,
1784 .llod_oid = LFSCK_NAMESPACE_OID,
1786 .llod_feat = &dt_lfsck_features,
1789 static int __init mdd_mod_init(void)
1791 struct lprocfs_static_vars lvars;
1794 lprocfs_mdd_init_vars(&lvars);
1796 rc = lu_kmem_init(mdd_caches);
1800 changelog_orig_logops = llog_osd_ops;
1801 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
1802 changelog_orig_logops.lop_add = llog_cat_add_rec;
1803 changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1805 llo_local_obj_register(&llod_capa_key);
1806 llo_local_obj_register(&llod_mdd_orphan);
1807 llo_local_obj_register(&llod_lfsck_bookmark);
1808 llo_local_obj_register(&llod_lfsck_namespace);
1810 rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1811 LUSTRE_MDD_NAME, &mdd_device_type);
1813 lu_kmem_fini(mdd_caches);
1817 static void __exit mdd_mod_exit(void)
1819 llo_local_obj_unregister(&llod_capa_key);
1820 llo_local_obj_unregister(&llod_mdd_orphan);
1821 llo_local_obj_unregister(&llod_lfsck_bookmark);
1822 llo_local_obj_unregister(&llod_lfsck_namespace);
1824 class_unregister_type(LUSTRE_MDD_NAME);
1825 lu_kmem_fini(mdd_caches);
1828 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1829 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1830 MODULE_LICENSE("GPL");
1832 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);