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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_device.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <linux/module.h>
42 #include <linux/kthread.h>
43 #include <obd_class.h>
44 #include <lustre_ioctl.h>
45 #include <lustre_mds.h>
46 #include <obd_support.h>
47 #include <lu_object.h>
48 #include <lustre_param.h>
49 #include <lustre_fid.h>
50 #include <lustre_nodemap.h>
52 #include "mdd_internal.h"
54 static const struct md_device_operations mdd_ops;
55 static struct lu_device_type mdd_device_type;
57 static const char mdd_root_dir_name[] = "ROOT";
58 static const char mdd_obf_dir_name[] = "fid";
59 static const char mdd_lpf_dir_name[] = "lost+found";
61 /* Slab for MDD object allocation */
62 struct kmem_cache *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)
124 /* LU-8040 Set defaults here, before values configs */
125 mdd->mdd_cl.mc_flags = 0; /* off by default */
126 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
128 dev = lustre_cfg_string(lcfg, 0);
132 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
133 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
135 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
136 mdd->mdd_md_dev.md_ops = &mdd_ops;
138 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
142 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
143 /* sync permission changes */
144 mdd->mdd_sync_permission = 1;
146 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
148 /* we are using service name but not mdd obd name
149 * for compatibility reasons.
150 * It is passed from MDT in lustre_cfg[2] buffer */
151 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
153 obd_disconnect(mdd->mdd_child_exp);
158 static struct lu_device *mdd_device_fini(const struct lu_env *env,
161 struct mdd_device *mdd = lu2mdd_dev(d);
164 lu_dev_del_linkage(d->ld_site, d);
166 mdd_procfs_fini(mdd);
170 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
171 struct llog_rec_hdr *hdr, void *data)
173 struct mdd_device *mdd = (struct mdd_device *)data;
174 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
176 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
177 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
180 "seeing record at index %d/%d/%llu t=%x %.*s in log"
181 DOSTID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
182 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
183 changelog_rec_name(&rec->cr), POSTID(&llh->lgh_id.lgl_oi));
185 mdd->mdd_cl.mc_index = rec->cr.cr_index;
186 return LLOG_PROC_BREAK;
189 static int changelog_user_init_cb(const struct lu_env *env,
190 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_user_rec *rec =
195 (struct llog_changelog_user_rec *)hdr;
197 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
198 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
200 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec=%llu"
201 " in log "DOSTID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
202 rec->cur_id, rec->cur_endrec, POSTID(&llh->lgh_id.lgl_oi));
204 spin_lock(&mdd->mdd_cl.mc_user_lock);
205 mdd->mdd_cl.mc_lastuser = rec->cur_id;
206 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
207 mdd->mdd_cl.mc_index = rec->cur_endrec;
208 spin_unlock(&mdd->mdd_cl.mc_user_lock);
210 return LLOG_PROC_BREAK;
213 static int llog_changelog_cancel_cb(const struct lu_env *env,
214 struct llog_handle *llh,
215 struct llog_rec_hdr *hdr, void *data)
217 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
218 struct llog_cookie cookie;
219 long long endrec = *(long long *)data;
224 /* This is always a (sub)log, not the catalog */
225 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
227 if (rec->cr.cr_index > endrec)
228 /* records are in order, so we're done */
229 RETURN(LLOG_PROC_BREAK);
231 cookie.lgc_lgl = llh->lgh_id;
232 cookie.lgc_index = hdr->lrh_index;
234 /* cancel them one at a time. I suppose we could store up the cookies
235 * and cancel them all at once; probably more efficient, but this is
236 * done as a user call, so who cares... */
237 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
239 RETURN(rc < 0 ? rc : 0);
242 static int llog_changelog_cancel(const struct lu_env *env,
243 struct llog_ctxt *ctxt,
244 struct llog_cookie *cookies, int flags)
246 struct llog_handle *cathandle = ctxt->loc_handle;
251 /* This should only be called with the catalog handle */
252 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
254 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
255 (void *)cookies, 0, 0);
257 /* 0 or 1 means we're done */
260 CERROR("%s: cancel idx %u of catalog "DOSTID" rc=%d\n",
261 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
262 POSTID(&cathandle->lgh_id.lgl_oi), rc);
267 static struct llog_operations changelog_orig_logops;
270 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
274 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
278 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
281 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
282 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
283 CERROR("Changelogs cannot be enabled due to error "
284 "condition (see %s log).\n",
285 mdd2obd_dev(mdd)->obd_name);
288 spin_lock(&mdd->mdd_cl.mc_lock);
289 mdd->mdd_cl.mc_flags |= CLM_ON;
290 spin_unlock(&mdd->mdd_cl.mc_lock);
291 rc = mdd_changelog_write_header(env, mdd, CLM_START);
297 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
301 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
304 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
305 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
306 spin_lock(&mdd->mdd_cl.mc_lock);
307 mdd->mdd_cl.mc_flags &= ~CLM_ON;
308 spin_unlock(&mdd->mdd_cl.mc_lock);
313 static int mdd_changelog_llog_init(const struct lu_env *env,
314 struct mdd_device *mdd)
316 struct obd_device *obd = mdd2obd_dev(mdd);
317 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
322 /* LU-2844 mdd setup failure should not cause umount oops */
323 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
326 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
327 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
328 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
329 obd, &changelog_orig_logops);
331 CERROR("%s: changelog llog setup failed: rc = %d\n",
336 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
339 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
342 GOTO(out_cleanup, rc);
344 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
348 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
349 changelog_init_cb, mdd);
352 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
357 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
358 mdd->mdd_cl.mc_index);
360 /* setup user changelog */
361 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
362 obd, &changelog_orig_logops);
364 CERROR("%s: changelog users llog setup failed: rc = %d\n",
369 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
372 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
375 GOTO(out_ucleanup, rc);
377 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
378 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
380 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
382 GOTO(out_uclose, rc);
384 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
385 changelog_user_init_cb, mdd);
387 CERROR("%s: changelog user init failed: rc = %d\n",
389 GOTO(out_uclose, rc);
392 /* If we have registered users, assume we want changelogs on */
393 if (mdd->mdd_cl.mc_lastuser > 0) {
394 rc = mdd_changelog_on(env, mdd);
396 GOTO(out_uclose, rc);
399 llog_ctxt_put(uctxt);
402 llog_cat_close(env, uctxt->loc_handle);
404 llog_cleanup(env, uctxt);
406 llog_cat_close(env, ctxt->loc_handle);
408 llog_cleanup(env, ctxt);
412 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
414 struct obd_device *obd = mdd2obd_dev(mdd);
417 mdd->mdd_cl.mc_index = 0;
418 spin_lock_init(&mdd->mdd_cl.mc_lock);
419 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
420 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
421 mdd->mdd_cl.mc_lastuser = 0;
423 rc = mdd_changelog_llog_init(env, mdd);
425 CERROR("%s: changelog setup during init failed: rc = %d\n",
427 mdd->mdd_cl.mc_flags |= CLM_ERR;
433 static void mdd_changelog_fini(const struct lu_env *env,
434 struct mdd_device *mdd)
436 struct obd_device *obd = mdd2obd_dev(mdd);
437 struct llog_ctxt *ctxt;
439 mdd->mdd_cl.mc_flags = 0;
441 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
443 llog_cat_close(env, ctxt->loc_handle);
444 llog_cleanup(env, ctxt);
446 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
448 llog_cat_close(env, ctxt->loc_handle);
449 llog_cleanup(env, ctxt);
453 /** Remove entries with indicies up to and including \a endrec from the
460 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
463 struct obd_device *obd = mdd2obd_dev(mdd);
464 struct llog_ctxt *ctxt;
465 long long unsigned cur;
468 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
472 spin_lock(&mdd->mdd_cl.mc_lock);
473 cur = (long long)mdd->mdd_cl.mc_index;
474 spin_unlock(&mdd->mdd_cl.mc_lock);
478 /* purge to "0" is shorthand for everything */
482 /* If purging all records, write a header entry so we don't have an
483 empty catalog and we're sure to have a valid starting index next
484 time. In case of crash, we just restart with old log so we're
487 /* XXX: transaction is started by llog itself */
488 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
493 /* Some records were purged, so reset repeat-access time (so we
494 record new mtime update records, so users can see a file has been
495 changed since the last purge) */
496 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
498 rc = llog_cancel(env, ctxt, (struct llog_cookie *)&endrec, 0);
504 /** Add a CL_MARK record to the changelog
506 * \param markerflags - CLM_*
509 int mdd_changelog_write_header(const struct lu_env *env,
510 struct mdd_device *mdd, int markerflags)
512 struct obd_device *obd = mdd2obd_dev(mdd);
513 struct llog_changelog_rec *rec;
515 struct llog_ctxt *ctxt;
517 int len = strlen(obd->obd_name);
522 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
523 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
527 reclen = llog_data_len(sizeof(*rec) + len);
528 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
529 if (buf->lb_buf == NULL)
533 rec->cr.cr_flags = CLF_VERSION;
534 rec->cr.cr_type = CL_MARK;
535 rec->cr.cr_namelen = len;
536 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
537 /* Status and action flags */
538 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
539 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
541 rec->cr_hdr.lrh_type = CHANGELOG_REC;
542 rec->cr.cr_time = cl_time();
543 spin_lock(&mdd->mdd_cl.mc_lock);
544 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
545 spin_unlock(&mdd->mdd_cl.mc_lock);
547 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
550 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
555 /* assume on or off event; reset repeat-access time */
556 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
561 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
562 * are valid. We also check if object with passed fid exists or not.
564 static int obf_lookup(const struct lu_env *env, struct md_object *p,
565 const struct lu_name *lname, struct lu_fid *f,
566 struct md_op_spec *spec)
568 char *name = (char *)lname->ln_name;
569 struct mdd_device *mdd = mdo2mdd(p);
570 struct mdd_object *child;
576 sscanf(name, SFID, RFID(f));
577 if (!fid_is_sane(f)) {
578 CWARN("%s: Trying to lookup invalid FID [%s] in %s/%s, FID "
579 "format should be "DFID"\n", mdd2obd_dev(mdd)->obd_name,
580 lname->ln_name, dot_lustre_name, mdd_obf_dir_name,
581 (__u64)FID_SEQ_NORMAL, 1, 0);
582 GOTO(out, rc = -EINVAL);
585 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
586 !fid_seq_is_dot(f->f_seq)) {
587 CWARN("%s: Trying to lookup invalid FID "DFID" in %s/%s, "
588 "sequence should be >= %#llx or within [%#llx,"
589 "%#llx].\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
590 dot_lustre_name, mdd_obf_dir_name, (__u64)FID_SEQ_NORMAL,
591 (__u64)FID_SEQ_IGIF, (__u64)FID_SEQ_IGIF_MAX);
592 GOTO(out, rc = -EINVAL);
595 /* Check if object with this fid exists */
596 child = mdd_object_find(env, mdd, f);
598 GOTO(out, rc = PTR_ERR(child));
600 if (mdd_object_exists(child) == 0)
603 mdd_object_put(env, child);
609 static int mdd_dummy_create(const struct lu_env *env,
610 struct md_object *pobj,
611 const struct lu_name *lname,
612 struct md_object *child,
613 struct md_op_spec *spec,
619 static int mdd_dummy_rename(const struct lu_env *env,
620 struct md_object *src_pobj,
621 struct md_object *tgt_pobj,
622 const struct lu_fid *lf,
623 const struct lu_name *lsname,
624 struct md_object *tobj,
625 const struct lu_name *ltname,
631 static int mdd_dummy_link(const struct lu_env *env,
632 struct md_object *tgt_obj,
633 struct md_object *src_obj,
634 const struct lu_name *lname,
640 static int mdd_dummy_unlink(const struct lu_env *env,
641 struct md_object *pobj,
642 struct md_object *cobj,
643 const struct lu_name *lname,
650 static struct md_dir_operations mdd_obf_dir_ops = {
651 .mdo_lookup = obf_lookup,
652 .mdo_create = mdd_dummy_create,
653 .mdo_rename = mdd_dummy_rename,
654 .mdo_link = mdd_dummy_link,
655 .mdo_unlink = mdd_dummy_unlink
658 static struct md_dir_operations mdd_lpf_dir_ops = {
659 .mdo_lookup = mdd_lookup,
660 .mdo_create = mdd_dummy_create,
661 .mdo_rename = mdd_dummy_rename,
662 .mdo_link = mdd_dummy_link,
663 .mdo_unlink = mdd_dummy_unlink
666 static struct md_object *mdo_locate(const struct lu_env *env,
667 struct md_device *md,
668 const struct lu_fid *fid)
670 struct lu_object *obj;
671 struct md_object *mdo;
673 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
675 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
676 LASSERT(obj != NULL);
679 mdo = ERR_PTR(PTR_ERR(obj));
684 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
686 struct md_object *mdo;
687 struct mdd_object *mdd_lpf;
688 struct lu_fid fid = LU_LPF_FID;
692 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
693 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
698 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
700 RETURN(PTR_ERR(mdo));
702 LASSERT(lu_object_exists(&mdo->mo_lu));
704 mdd_lpf = md2mdd_obj(mdo);
705 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
706 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
712 * Create special in-memory "fid" object for open-by-fid.
714 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
716 struct md_object *mdo;
717 struct mdd_object *mdd_obf;
718 struct lu_fid fid = LU_OBF_FID;
721 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
722 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
726 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
728 RETURN(PTR_ERR(mdo));
730 LASSERT(lu_object_exists(&mdo->mo_lu));
732 mdd_obf = md2mdd_obj(mdo);
733 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
734 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
739 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
740 struct mdd_device *m)
742 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
743 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
744 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
746 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
747 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
748 m->mdd_dot_lustre_objs.mdd_obf = NULL;
750 if (m->mdd_dot_lustre != NULL) {
751 mdd_object_put(env, m->mdd_dot_lustre);
752 m->mdd_dot_lustre = NULL;
756 /** Setup ".lustre" directory object */
757 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
759 struct md_object *mdo;
764 /* Create ".lustre" directory in ROOT. */
765 fid = LU_DOT_LUSTRE_FID;
766 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
768 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
772 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
774 RETURN(PTR_ERR(mdo));
775 LASSERT(lu_object_exists(&mdo->mo_lu));
777 m->mdd_dot_lustre = md2mdd_obj(mdo);
779 rc = mdd_obf_setup(env, m);
781 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
782 mdd2obd_dev(m)->obd_name, rc);
786 rc = mdd_lpf_setup(env, m);
788 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
789 mdd2obd_dev(m)->obd_name, rc);
796 mdd_dot_lustre_cleanup(env, m);
802 static struct llog_operations hsm_actions_logops;
805 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
806 * object in obd_device
808 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
809 struct mdd_device *m)
811 struct obd_device *obd = mdd2obd_dev(m);
812 struct llog_ctxt *ctxt = NULL;
816 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
817 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
819 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
820 obd, &hsm_actions_logops);
822 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
827 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
830 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
833 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
835 GOTO(out_cleanup, rc);
838 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
846 llog_cat_close(env, ctxt->loc_handle);
847 ctxt->loc_handle = NULL;
849 llog_cleanup(env, ctxt);
855 * cleanup the context created by llog_setup_named()
857 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
858 struct mdd_device *m)
860 struct obd_device *obd = mdd2obd_dev(m);
861 struct llog_ctxt *lctxt;
864 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
866 llog_cat_close(env, lctxt->loc_handle);
867 lctxt->loc_handle = NULL;
868 llog_cleanup(env, lctxt);
874 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
875 struct lustre_cfg *cfg)
877 lfsck_degister(env, m->mdd_bottom);
878 mdd_hsm_actions_llog_fini(env, m);
879 mdd_changelog_fini(env, m);
880 orph_index_fini(env, m);
881 mdd_dot_lustre_cleanup(env, m);
882 nm_config_file_deregister_tgt(env, mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
883 if (m->mdd_los != NULL) {
884 local_oid_storage_fini(env, m->mdd_los);
887 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
889 if (m->mdd_child_exp)
890 obd_disconnect(m->mdd_child_exp);
893 static int mdd_process_config(const struct lu_env *env,
894 struct lu_device *d, struct lustre_cfg *cfg)
896 struct mdd_device *m = lu2mdd_dev(d);
897 struct dt_device *dt = m->mdd_child;
898 struct lu_device *next = &dt->dd_lu_dev;
902 switch (cfg->lcfg_command) {
904 struct obd_device *obd = mdd2obd_dev(m);
906 rc = class_process_proc_param(PARAM_MDD, obd->obd_vars, cfg, m);
907 if (rc > 0 || rc == -ENOSYS)
908 /* we don't understand; pass it on */
909 rc = next->ld_ops->ldo_process_config(env, next, cfg);
913 rc = next->ld_ops->ldo_process_config(env, next, cfg);
916 dt_conf_get(env, dt, &m->mdd_dt_conf);
918 case LCFG_PRE_CLEANUP:
919 rc = next->ld_ops->ldo_process_config(env, next, cfg);
920 mdd_generic_thread_stop(&m->mdd_orph_cleanup_thread);
923 rc = next->ld_ops->ldo_process_config(env, next, cfg);
924 lu_dev_del_linkage(d->ld_site, d);
925 mdd_device_shutdown(env, m, cfg);
928 rc = next->ld_ops->ldo_process_config(env, next, cfg);
935 static int mdd_recovery_complete(const struct lu_env *env,
938 struct mdd_device *mdd = lu2mdd_dev(d);
939 struct lu_device *next;
943 LASSERT(mdd != NULL);
944 next = &mdd->mdd_child->dd_lu_dev;
946 /* XXX: orphans handling. */
947 mdd_orphan_cleanup(env, mdd);
948 rc = next->ld_ops->ldo_recovery_complete(env, next);
953 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
954 const struct lu_fid *pfid, const char *name,
955 __u32 mode, struct lu_fid *fid)
957 struct dt_object *parent, *dto;
962 LASSERT(!fid_is_zero(pfid));
963 parent = dt_locate(env, mdd->mdd_bottom, pfid);
964 if (unlikely(IS_ERR(parent)))
965 RETURN(PTR_ERR(parent));
967 /* create local file/dir, if @fid is passed then try to use it */
968 if (fid_is_zero(fid))
969 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
972 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
976 GOTO(out_put, rc = PTR_ERR(dto));
977 *fid = *lu_object_fid(&dto->do_lu);
978 /* since stack is not fully set up the local_storage uses own stack
979 * and we should drop its object from cache */
980 lu_object_put_nocache(env, &dto->do_lu);
983 lu_object_put(env, &parent->do_lu);
987 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
988 enum lfsck_events event)
993 static int mdd_prepare(const struct lu_env *env,
994 struct lu_device *pdev,
995 struct lu_device *cdev)
997 struct mdd_device *mdd = lu2mdd_dev(cdev);
998 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
999 struct nm_config_file *nodemap_config;
1005 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1009 /* Setup local dirs */
1010 fid.f_seq = FID_SEQ_LOCAL_NAME;
1013 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1018 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1023 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1024 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1025 mdd_root_dir_name, S_IFDIR |
1026 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1028 CERROR("%s: create root fid failed: rc = %d\n",
1029 mdd2obd_dev(mdd)->obd_name, rc);
1033 mdd->mdd_root_fid = fid;
1034 rc = mdd_dot_lustre_setup(env, mdd);
1036 CERROR("%s: initializing .lustre failed: rc = %d\n",
1037 mdd2obd_dev(mdd)->obd_name, rc);
1041 /* Normal client usually send root access to MDT0 directly,
1042 * the root FID on non-MDT0 will only be used by echo client. */
1043 mdd->mdd_root_fid = fid;
1046 rc = orph_index_init(env, mdd);
1050 rc = mdd_changelog_init(env, mdd);
1052 CERROR("%s: failed to initialize changelog: rc = %d\n",
1053 mdd2obd_dev(mdd)->obd_name, rc);
1057 rc = mdd_hsm_actions_llog_init(env, mdd);
1059 GOTO(out_changelog, rc);
1061 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1063 if (IS_ERR(nodemap_config))
1064 GOTO(out_hsm, rc = PTR_ERR(nodemap_config));
1066 mdd2obd_dev(mdd)->u.obt.obt_nodemap_config_file = nodemap_config;
1068 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1069 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1072 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1073 mdd2obd_dev(mdd)->obd_name, rc);
1074 GOTO(out_nodemap, rc);
1080 nm_config_file_deregister_tgt(env, mdd2obd_dev(mdd)->u.obt.obt_nodemap_config_file);
1081 mdd2obd_dev(mdd)->u.obt.obt_nodemap_config_file = NULL;
1083 mdd_hsm_actions_llog_fini(env, mdd);
1085 mdd_changelog_fini(env, mdd);
1087 orph_index_fini(env, mdd);
1089 if (mdd_seq_site(mdd)->ss_node_id == 0)
1090 mdd_dot_lustre_cleanup(env, mdd);
1092 local_oid_storage_fini(env, mdd->mdd_los);
1093 mdd->mdd_los = NULL;
1098 const struct lu_device_operations mdd_lu_ops = {
1099 .ldo_object_alloc = mdd_object_alloc,
1100 .ldo_process_config = mdd_process_config,
1101 .ldo_recovery_complete = mdd_recovery_complete,
1102 .ldo_prepare = mdd_prepare,
1105 static int mdd_root_get(const struct lu_env *env,
1106 struct md_device *m, struct lu_fid *f)
1108 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1111 *f = mdd->mdd_root_fid;
1116 * No permission check is needed.
1118 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1119 struct obd_statfs *sfs)
1121 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1126 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1128 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1133 static int mdd_maxeasize_get(const struct lu_env *env, struct md_device *m,
1136 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1139 *easize = mdd->mdd_dt_conf.ddp_max_ea_size;
1144 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1147 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1149 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1150 return (*h == NULL ? -ENOENT : 0);
1153 static struct lu_device *mdd_device_free(const struct lu_env *env,
1154 struct lu_device *lu)
1156 struct mdd_device *m = lu2mdd_dev(lu);
1159 LASSERT(atomic_read(&lu->ld_ref) == 0);
1160 md_device_fini(&m->mdd_md_dev);
1165 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1166 struct lu_device_type *t,
1167 struct lustre_cfg *lcfg)
1169 struct lu_device *l;
1170 struct mdd_device *m;
1174 l = ERR_PTR(-ENOMEM);
1179 md_device_init(&m->mdd_md_dev, t);
1180 rc = mdd_init0(env, m, t, lcfg);
1182 mdd_device_free(env, l);
1191 * we use exports to track all mdd users
1193 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1194 struct obd_device *obd, struct obd_uuid *cluuid,
1195 struct obd_connect_data *data, void *localdata)
1197 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1198 struct lustre_handle conn;
1202 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1204 rc = class_connect(&conn, obd, cluuid);
1208 *exp = class_conn2export(&conn);
1210 /* Why should there ever be more than 1 connect? */
1211 LASSERT(mdd->mdd_connects == 0);
1212 mdd->mdd_connects++;
1218 * once last export (we don't count self-export) disappeared
1219 * mdd can be released
1221 static int mdd_obd_disconnect(struct obd_export *exp)
1223 struct obd_device *obd = exp->exp_obd;
1224 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1225 int rc, release = 0;
1228 mdd->mdd_connects--;
1229 if (mdd->mdd_connects == 0)
1232 rc = class_disconnect(exp);
1234 if (rc == 0 && release)
1235 class_manual_cleanup(obd);
1239 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1240 __u32 keylen, void *key, __u32 *vallen, void *val)
1244 if (KEY_IS(KEY_OSP_CONNECTED)) {
1245 struct obd_device *obd = exp->exp_obd;
1246 struct mdd_device *mdd;
1248 if (!obd->obd_set_up || obd->obd_stopping)
1251 mdd = lu2mdd_dev(obd->obd_lu_dev);
1253 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1261 static int mdd_obd_set_info_async(const struct lu_env *env,
1262 struct obd_export *exp,
1263 __u32 keylen, void *key,
1264 __u32 vallen, void *val,
1265 struct ptlrpc_request_set *set)
1267 struct obd_device *obd = exp->exp_obd;
1268 struct mdd_device *mdd;
1271 if (!obd->obd_set_up || obd->obd_stopping)
1274 mdd = lu2mdd_dev(obd->obd_lu_dev);
1276 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1281 static struct obd_ops mdd_obd_device_ops = {
1282 .o_owner = THIS_MODULE,
1283 .o_connect = mdd_obd_connect,
1284 .o_disconnect = mdd_obd_disconnect,
1285 .o_get_info = mdd_obd_get_info,
1286 .o_set_info_async = mdd_obd_set_info_async,
1289 static int mdd_changelog_user_register(const struct lu_env *env,
1290 struct mdd_device *mdd, int *id)
1292 struct llog_ctxt *ctxt;
1293 struct llog_changelog_user_rec *rec;
1297 ctxt = llog_get_context(mdd2obd_dev(mdd),
1298 LLOG_CHANGELOG_USER_ORIG_CTXT);
1304 llog_ctxt_put(ctxt);
1308 /* Assume we want it on since somebody registered */
1309 rc = mdd_changelog_on(env, mdd);
1313 rec->cur_hdr.lrh_len = sizeof(*rec);
1314 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1315 spin_lock(&mdd->mdd_cl.mc_user_lock);
1316 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1317 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1318 CERROR("Maximum number of changelog users exceeded!\n");
1319 GOTO(out, rc = -EOVERFLOW);
1321 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1322 rec->cur_endrec = mdd->mdd_cl.mc_index;
1323 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1325 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1327 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1330 llog_ctxt_put(ctxt);
1334 struct mdd_changelog_user_purge {
1336 __u32 mcup_usercount;
1342 * changelog_user_purge callback
1344 * Is called once per user.
1346 * Check to see the user requested is available from rec.
1347 * Truncate the changelog.
1348 * Keep track of the total number of users (calls).
1350 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1351 struct llog_handle *llh,
1352 struct llog_rec_hdr *hdr, void *data)
1354 struct llog_changelog_user_rec *rec;
1355 struct mdd_changelog_user_purge *mcup = data;
1356 struct llog_cookie cookie;
1361 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1364 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1366 mcup->mcup_usercount++;
1368 if (rec->cur_id != mcup->mcup_id) {
1369 /* truncate to the lowest endrec that is not this user */
1370 mcup->mcup_minrec = min(mcup->mcup_minrec,
1375 /* Unregister this user */
1376 cookie.lgc_lgl = llh->lgh_id;
1377 cookie.lgc_index = hdr->lrh_index;
1379 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1382 mcup->mcup_found = true;
1383 mcup->mcup_usercount--;
1389 static int mdd_changelog_user_purge(const struct lu_env *env,
1390 struct mdd_device *mdd, __u32 id)
1392 struct mdd_changelog_user_purge mcup = {
1394 .mcup_found = false,
1395 .mcup_usercount = 0,
1396 .mcup_minrec = ULLONG_MAX,
1398 struct llog_ctxt *ctxt;
1403 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1404 mdd2obd_dev(mdd)->obd_name, id);
1406 ctxt = llog_get_context(mdd2obd_dev(mdd),
1407 LLOG_CHANGELOG_USER_ORIG_CTXT);
1409 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1410 GOTO(out, rc = -ENXIO);
1412 rc = llog_cat_process(env, ctxt->loc_handle,
1413 mdd_changelog_user_purge_cb, &mcup,
1416 if ((rc == 0) && mcup.mcup_found) {
1417 CDEBUG(D_IOCTL, "%s: Purging changelog entries for user %d "
1419 mdd2obd_dev(mdd)->obd_name, id, mcup.mcup_minrec);
1420 /* Cancelling record 0 destroys the entire changelog, make sure
1421 we don't do that unless we mean it. */
1422 if (mcup.mcup_minrec != 0 || mcup.mcup_usercount == 0) {
1423 rc = mdd_changelog_llog_cancel(env, mdd,
1427 CWARN("%s: No changelog for user %u; rc=%d\n",
1428 mdd2obd_dev(mdd)->obd_name, id, rc);
1429 GOTO(out, rc = -ENOENT);
1432 if ((rc == 0) && (mcup.mcup_usercount == 0)) {
1433 /* No more users; turn changelogs off */
1434 CDEBUG(D_IOCTL, "turning off changelogs\n");
1435 rc = mdd_changelog_off(env, mdd);
1441 llog_ctxt_put(ctxt);
1446 struct mdd_changelog_user_clear {
1454 * changelog_clear callback
1456 * Is called once per user.
1458 * Check to see the user requested is available from rec.
1459 * Check the oldest (smallest) record for boundary conditions.
1460 * Truncate the changelog.
1462 static int mdd_changelog_clear_cb(const struct lu_env *env,
1463 struct llog_handle *llh,
1464 struct llog_rec_hdr *hdr,
1467 struct llog_changelog_user_rec *rec;
1468 struct mdd_changelog_user_clear *mcuc = data;
1473 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1476 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1478 /* Does the changelog id match the requested id? */
1479 if (rec->cur_id != mcuc->mcuc_id) {
1480 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec,
1485 /* cur_endrec is the oldest purgeable record, make sure we're newer */
1486 if (rec->cur_endrec > mcuc->mcuc_endrec) {
1487 CDEBUG(D_IOCTL, "Request %llu out of range: %llu\n",
1488 mcuc->mcuc_endrec, rec->cur_endrec);
1492 /* Flag that we've met all the range and user checks.
1493 * We now know the record to flush.
1495 rec->cur_endrec = mcuc->mcuc_endrec;
1496 mcuc->mcuc_flush = true;
1498 CDEBUG(D_IOCTL, "Rewriting changelog user %u endrec to %llu\n",
1499 mcuc->mcuc_id, rec->cur_endrec);
1501 /* Update the endrec */
1502 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1508 * Clear a changelog up to entry specified by endrec for user id.
1510 static int mdd_changelog_clear(const struct lu_env *env,
1511 struct mdd_device *mdd, __u32 id,
1514 struct mdd_changelog_user_clear mcuc = {
1516 .mcuc_minrec = endrec,
1517 .mcuc_flush = false,
1519 struct llog_ctxt *ctxt;
1525 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
1526 mdd2obd_dev(mdd)->obd_name, id, endrec);
1527 /* start_rec is the newest (largest value) entry in the changelogs*/
1528 spin_lock(&mdd->mdd_cl.mc_lock);
1529 start_rec = mdd->mdd_cl.mc_index;
1530 spin_unlock(&mdd->mdd_cl.mc_lock);
1532 if (start_rec < endrec) {
1533 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
1534 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
1539 mcuc.mcuc_endrec = start_rec;
1540 mcuc.mcuc_minrec = ULLONG_MAX;
1542 mcuc.mcuc_endrec = endrec;
1545 ctxt = llog_get_context(mdd2obd_dev(mdd),
1546 LLOG_CHANGELOG_USER_ORIG_CTXT);
1548 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1549 GOTO(out, rc = -ENXIO);
1551 rc = llog_cat_process(env, ctxt->loc_handle,
1552 mdd_changelog_clear_cb, (void *)&mcuc,
1556 CWARN("%s: Failure to clear the changelog for user %d: %d\n",
1557 mdd2obd_dev(mdd)->obd_name, id, rc);
1558 } else if (mcuc.mcuc_flush) {
1559 /* Cancelling record 0 destroys the entire changelog, make sure
1560 we don't do that unless we mean it. */
1561 if (mcuc.mcuc_minrec != 0) {
1562 CDEBUG(D_IOCTL, "%s: Purging changelog entries up "\
1563 "to %llu\n", mdd2obd_dev(mdd)->obd_name,
1566 rc = mdd_changelog_llog_cancel(env, mdd,
1570 CWARN("%s: No entry for user %d\n",
1571 mdd2obd_dev(mdd)->obd_name, id);
1578 llog_ctxt_put(ctxt);
1584 * May be called remotely from mdt_iocontrol_handle or locally from
1585 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1586 * an obd_ioctl_data format (but local ioctl handler does).
1588 * \param len - data len
1589 * \param karg - ioctl data, in kernel space
1591 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1592 unsigned int cmd, int len, void *karg)
1594 struct mdd_device *mdd;
1595 struct obd_ioctl_data *data = karg;
1600 mdd = lu2mdd_dev(&m->md_lu_dev);
1602 /* Doesn't use obd_ioctl_data */
1604 case OBD_IOC_CHANGELOG_CLEAR: {
1605 struct changelog_setinfo *cs = karg;
1606 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
1610 case OBD_IOC_GET_MNTOPT: {
1611 mntopt_t *mntopts = (mntopt_t *)karg;
1612 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1615 case OBD_IOC_START_LFSCK: {
1616 rc = lfsck_start(env, mdd->mdd_bottom,
1617 (struct lfsck_start_param *)karg);
1620 case OBD_IOC_STOP_LFSCK: {
1621 rc = lfsck_stop(env, mdd->mdd_bottom,
1622 (struct lfsck_stop *)karg);
1625 case OBD_IOC_QUERY_LFSCK: {
1626 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
1627 (struct lfsck_query *)karg);
1632 /* Below ioctls use obd_ioctl_data */
1633 if (len != sizeof(*data)) {
1634 CERROR("Bad ioctl size %d\n", len);
1637 if (data->ioc_version != OBD_IOCTL_VERSION) {
1638 CERROR("Bad magic %x != %x\n", data->ioc_version,
1644 case OBD_IOC_CHANGELOG_REG:
1645 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1647 case OBD_IOC_CHANGELOG_DEREG:
1648 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1);
1657 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1658 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
1660 static const struct md_device_operations mdd_ops = {
1661 .mdo_statfs = mdd_statfs,
1662 .mdo_root_get = mdd_root_get,
1663 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1664 .mdo_iocontrol = mdd_iocontrol,
1665 .mdo_maxeasize_get = mdd_maxeasize_get,
1668 static struct lu_device_type_operations mdd_device_type_ops = {
1669 .ldto_init = mdd_type_init,
1670 .ldto_fini = mdd_type_fini,
1672 .ldto_start = mdd_type_start,
1673 .ldto_stop = mdd_type_stop,
1675 .ldto_device_alloc = mdd_device_alloc,
1676 .ldto_device_free = mdd_device_free,
1678 .ldto_device_fini = mdd_device_fini
1681 static struct lu_device_type mdd_device_type = {
1682 .ldt_tags = LU_DEVICE_MD,
1683 .ldt_name = LUSTRE_MDD_NAME,
1684 .ldt_ops = &mdd_device_type_ops,
1685 .ldt_ctx_tags = LCT_MD_THREAD
1688 /* context key constructor: mdd_key_init */
1689 LU_KEY_INIT(mdd, struct mdd_thread_info);
1691 static void mdd_key_fini(const struct lu_context *ctx,
1692 struct lu_context_key *key, void *data)
1694 struct mdd_thread_info *info = data;
1696 lu_buf_free(&info->mti_big_buf);
1697 lu_buf_free(&info->mti_link_buf);
1702 /* context key: mdd_thread_key */
1703 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1705 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
1706 int (*func)(void *), void *data, char *name)
1708 struct task_struct *task;
1710 LASSERT(thread->mgt_init == false);
1711 init_completion(&thread->mgt_started);
1712 init_completion(&thread->mgt_finished);
1713 thread->mgt_data = data;
1714 thread->mgt_abort = false;
1715 thread->mgt_init = true;
1717 task = kthread_run(func, thread, name);
1719 complete(&thread->mgt_finished);
1720 return PTR_ERR(task);
1722 wait_for_completion(&thread->mgt_started);
1726 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
1728 if (thread->mgt_init == true) {
1729 thread->mgt_abort = true;
1730 wait_for_completion(&thread->mgt_finished);
1734 static int __init mdd_init(void)
1738 rc = lu_kmem_init(mdd_caches);
1742 changelog_orig_logops = llog_osd_ops;
1743 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
1744 changelog_orig_logops.lop_add = llog_cat_add_rec;
1745 changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1747 hsm_actions_logops = llog_osd_ops;
1748 hsm_actions_logops.lop_add = llog_cat_add_rec;
1749 hsm_actions_logops.lop_declare_add = llog_cat_declare_add_rec;
1751 rc = class_register_type(&mdd_obd_device_ops, NULL, true, NULL,
1752 LUSTRE_MDD_NAME, &mdd_device_type);
1754 lu_kmem_fini(mdd_caches);
1758 static void __exit mdd_exit(void)
1760 class_unregister_type(LUSTRE_MDD_NAME);
1761 lu_kmem_fini(mdd_caches);
1764 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1765 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
1766 MODULE_VERSION(LUSTRE_VERSION_STRING);
1767 MODULE_LICENSE("GPL");
1769 module_init(mdd_init);
1770 module_exit(mdd_exit);