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, 2017, 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 <uapi/linux/lustre/lustre_ioctl.h>
45 #include <lustre_mds.h>
46 #include <obd_support.h>
47 #include <lu_object.h>
48 #include <uapi/linux/lustre/lustre_param.h>
49 #include <lustre_fid.h>
50 #include <lustre_nodemap.h>
51 #include <lustre_barrier.h>
53 #include "mdd_internal.h"
55 static 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";
60 static const char mdd_lpf_dir_name[] = "lost+found";
62 /* Slab for MDD object allocation */
63 struct kmem_cache *mdd_object_kmem;
65 static struct lu_kmem_descr mdd_caches[] = {
67 .ckd_cache = &mdd_object_kmem,
68 .ckd_name = "mdd_obj",
69 .ckd_size = sizeof(struct mdd_object)
76 static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
79 struct obd_connect_data *data = NULL;
80 struct lu_device *lud = mdd2lu_dev(m);
81 struct obd_device *obd;
85 LASSERT(m->mdd_child_exp == NULL);
87 OBD_ALLOC(data, sizeof(*data));
89 GOTO(out, rc = -ENOMEM);
91 obd = class_name2obd(nextdev);
93 CERROR("can't locate next device: %s\n", nextdev);
94 GOTO(out, rc = -ENOTCONN);
97 data->ocd_connect_flags = OBD_CONNECT_VERSION;
98 data->ocd_version = LUSTRE_VERSION_CODE;
100 rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
102 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
106 lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
107 LASSERT(lud->ld_site);
108 m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
109 m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
110 lu_dev_add_linkage(lud->ld_site, lud);
114 OBD_FREE(data, sizeof(*data));
118 static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
119 struct lu_device_type *t, struct lustre_cfg *lcfg)
125 /* LU-8040 Set defaults here, before values configs */
126 mdd->mdd_cl.mc_flags = 0; /* off by default */
127 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
129 dev = lustre_cfg_string(lcfg, 0);
133 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
134 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
136 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
137 mdd->mdd_md_dev.md_ops = &mdd_ops;
139 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
143 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
144 /* sync permission changes */
145 mdd->mdd_sync_permission = 1;
146 /* enable changelog garbage collection */
147 mdd->mdd_changelog_gc = 1;
148 /* with a significant amount of idle time */
149 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
150 /* or a significant amount of late indexes */
151 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
152 /* with a reasonable interval between each check */
153 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
154 /* with a very few number of free entries */
155 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
157 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
159 /* we are using service name but not mdd obd name
160 * for compatibility reasons.
161 * It is passed from MDT in lustre_cfg[2] buffer */
162 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
164 obd_disconnect(mdd->mdd_child_exp);
169 static struct lu_device *mdd_device_fini(const struct lu_env *env,
172 struct mdd_device *mdd = lu2mdd_dev(d);
175 lu_dev_del_linkage(d->ld_site, d);
177 mdd_procfs_fini(mdd);
181 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
182 struct llog_rec_hdr *hdr, void *data)
184 struct mdd_device *mdd = (struct mdd_device *)data;
185 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
187 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
188 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
191 "seeing record at index %d/%d/%llu t=%x %.*s in log"
192 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
193 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
194 changelog_rec_name(&rec->cr), PFID(&llh->lgh_id.lgl_oi.oi_fid));
196 mdd->mdd_cl.mc_index = rec->cr.cr_index;
197 return LLOG_PROC_BREAK;
200 static int changelog_user_init_cb(const struct lu_env *env,
201 struct llog_handle *llh,
202 struct llog_rec_hdr *hdr, void *data)
204 struct mdd_device *mdd = (struct mdd_device *)data;
205 struct llog_changelog_user_rec *rec =
206 (struct llog_changelog_user_rec *)hdr;
208 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
209 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
211 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec=%llu"
212 " in log "DFID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
213 rec->cur_id, rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
215 spin_lock(&mdd->mdd_cl.mc_user_lock);
216 mdd->mdd_cl.mc_lastuser = rec->cur_id;
217 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
218 mdd->mdd_cl.mc_index = rec->cur_endrec;
219 spin_unlock(&mdd->mdd_cl.mc_user_lock);
221 return LLOG_PROC_BREAK;
224 static int llog_changelog_cancel_cb(const struct lu_env *env,
225 struct llog_handle *llh,
226 struct llog_rec_hdr *hdr, void *data)
228 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
229 struct llog_cookie cookie;
230 long long endrec = *(long long *)data;
235 /* This is always a (sub)log, not the catalog */
236 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
238 if (rec->cr.cr_index > endrec)
239 /* records are in order, so we're done */
240 RETURN(LLOG_PROC_BREAK);
242 cookie.lgc_lgl = llh->lgh_id;
243 cookie.lgc_index = hdr->lrh_index;
245 /* cancel them one at a time. I suppose we could store up the cookies
246 * and cancel them all at once; probably more efficient, but this is
247 * done as a user call, so who cares... */
248 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
250 RETURN(rc < 0 ? rc : 0);
253 static int llog_changelog_cancel(const struct lu_env *env,
254 struct llog_ctxt *ctxt,
255 struct llog_cookie *cookies, int flags)
257 struct llog_handle *cathandle = ctxt->loc_handle;
262 /* This should only be called with the catalog handle */
263 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
265 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
266 (void *)cookies, 0, 0);
268 /* 0 or 1 means we're done */
271 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
272 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
273 PFID(&cathandle->lgh_id.lgl_oi.oi_fid), rc);
278 static struct llog_operations changelog_orig_logops;
281 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
285 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
289 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
292 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
293 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
294 CERROR("Changelogs cannot be enabled due to error "
295 "condition (see %s log).\n",
296 mdd2obd_dev(mdd)->obd_name);
299 spin_lock(&mdd->mdd_cl.mc_lock);
300 mdd->mdd_cl.mc_flags |= CLM_ON;
301 spin_unlock(&mdd->mdd_cl.mc_lock);
302 rc = mdd_changelog_write_header(env, mdd, CLM_START);
308 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
312 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
315 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
316 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
317 spin_lock(&mdd->mdd_cl.mc_lock);
318 mdd->mdd_cl.mc_flags &= ~CLM_ON;
319 spin_unlock(&mdd->mdd_cl.mc_lock);
324 static int mdd_changelog_llog_init(const struct lu_env *env,
325 struct mdd_device *mdd)
327 struct obd_device *obd = mdd2obd_dev(mdd);
328 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
333 /* LU-2844 mdd setup failure should not cause umount oops */
334 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
337 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
338 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
339 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
340 obd, &changelog_orig_logops);
342 CERROR("%s: changelog llog setup failed: rc = %d\n",
347 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
350 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
353 GOTO(out_cleanup, rc);
355 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
359 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
360 changelog_init_cb, mdd);
363 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
368 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
369 mdd->mdd_cl.mc_index);
371 /* setup user changelog */
372 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
373 obd, &changelog_orig_logops);
375 CERROR("%s: changelog users llog setup failed: rc = %d\n",
380 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
383 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
386 GOTO(out_ucleanup, rc);
388 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
389 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
391 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
393 GOTO(out_uclose, rc);
395 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
396 changelog_user_init_cb, mdd);
398 CERROR("%s: changelog user init failed: rc = %d\n",
400 GOTO(out_uclose, rc);
403 /* If we have registered users, assume we want changelogs on */
404 if (mdd->mdd_cl.mc_lastuser > 0) {
405 rc = mdd_changelog_on(env, mdd);
407 GOTO(out_uclose, rc);
410 llog_ctxt_put(uctxt);
413 llog_cat_close(env, uctxt->loc_handle);
415 llog_cleanup(env, uctxt);
417 llog_cat_close(env, ctxt->loc_handle);
419 llog_cleanup(env, ctxt);
423 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
425 struct obd_device *obd = mdd2obd_dev(mdd);
428 mdd->mdd_cl.mc_index = 0;
429 spin_lock_init(&mdd->mdd_cl.mc_lock);
430 mdd->mdd_cl.mc_starttime = ktime_get();
431 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
432 mdd->mdd_cl.mc_lastuser = 0;
434 rc = mdd_changelog_llog_init(env, mdd);
436 CERROR("%s: changelog setup during init failed: rc = %d\n",
438 mdd->mdd_cl.mc_flags |= CLM_ERR;
444 static void mdd_changelog_fini(const struct lu_env *env,
445 struct mdd_device *mdd)
447 struct obd_device *obd = mdd2obd_dev(mdd);
448 struct llog_ctxt *ctxt;
450 mdd->mdd_cl.mc_flags = 0;
452 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
454 llog_cat_close(env, ctxt->loc_handle);
455 llog_cleanup(env, ctxt);
457 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
459 llog_cat_close(env, ctxt->loc_handle);
460 llog_cleanup(env, ctxt);
464 /** Remove entries with indicies up to and including \a endrec from the
471 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
474 struct obd_device *obd = mdd2obd_dev(mdd);
475 struct llog_ctxt *ctxt;
476 long long unsigned cur;
479 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
483 spin_lock(&mdd->mdd_cl.mc_lock);
484 cur = (long long)mdd->mdd_cl.mc_index;
485 spin_unlock(&mdd->mdd_cl.mc_lock);
489 /* purge to "0" is shorthand for everything */
493 /* If purging all records, write a header entry so we don't have an
494 empty catalog and we're sure to have a valid starting index next
495 time. In case of crash, we just restart with old log so we're
498 /* XXX: transaction is started by llog itself */
499 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
504 /* Some records were purged, so reset repeat-access time (so we
505 record new mtime update records, so users can see a file has been
506 changed since the last purge) */
507 mdd->mdd_cl.mc_starttime = ktime_get();
509 rc = llog_cancel(env, ctxt, (struct llog_cookie *)&endrec, 0);
515 /** Add a CL_MARK record to the changelog
517 * \param markerflags - CLM_*
520 int mdd_changelog_write_header(const struct lu_env *env,
521 struct mdd_device *mdd, int markerflags)
523 struct obd_device *obd = mdd2obd_dev(mdd);
524 struct llog_changelog_rec *rec;
526 struct llog_ctxt *ctxt;
528 int len = strlen(obd->obd_name);
533 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
534 mdd->mdd_cl.mc_starttime = ktime_get();
538 reclen = llog_data_len(sizeof(*rec) + len);
539 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
540 if (buf->lb_buf == NULL)
544 rec->cr.cr_flags = CLF_VERSION;
545 rec->cr.cr_type = CL_MARK;
546 rec->cr.cr_namelen = len;
547 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
548 /* Status and action flags */
549 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
550 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
552 rec->cr_hdr.lrh_type = CHANGELOG_REC;
553 rec->cr.cr_time = cl_time();
554 spin_lock(&mdd->mdd_cl.mc_lock);
555 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
556 spin_unlock(&mdd->mdd_cl.mc_lock);
558 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
561 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
566 /* assume on or off event; reset repeat-access time */
567 mdd->mdd_cl.mc_starttime = ktime_get();
572 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
573 * are valid. We also check if object with passed fid exists or not.
575 static int obf_lookup(const struct lu_env *env, struct md_object *p,
576 const struct lu_name *lname, struct lu_fid *f,
577 struct md_op_spec *spec)
579 char *name = (char *)lname->ln_name;
580 struct mdd_device *mdd = mdo2mdd(p);
581 struct mdd_object *child;
587 sscanf(name, SFID, RFID(f));
588 if (!fid_is_sane(f)) {
589 CWARN("%s: Trying to lookup invalid FID [%s] in %s/%s, FID "
590 "format should be "DFID"\n", mdd2obd_dev(mdd)->obd_name,
591 lname->ln_name, dot_lustre_name, mdd_obf_dir_name,
592 (__u64)FID_SEQ_NORMAL, 1, 0);
593 GOTO(out, rc = -EINVAL);
596 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
597 !fid_seq_is_dot(f->f_seq)) {
598 CWARN("%s: Trying to lookup invalid FID "DFID" in %s/%s, "
599 "sequence should be >= %#llx or within [%#llx,"
600 "%#llx].\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
601 dot_lustre_name, mdd_obf_dir_name, (__u64)FID_SEQ_NORMAL,
602 (__u64)FID_SEQ_IGIF, (__u64)FID_SEQ_IGIF_MAX);
603 GOTO(out, rc = -EINVAL);
606 /* Check if object with this fid exists */
607 child = mdd_object_find(env, mdd, f);
609 GOTO(out, rc = PTR_ERR(child));
611 if (mdd_object_exists(child) == 0)
614 mdd_object_put(env, child);
620 static int mdd_dummy_create(const struct lu_env *env,
621 struct md_object *pobj,
622 const struct lu_name *lname,
623 struct md_object *child,
624 struct md_op_spec *spec,
630 static int mdd_dummy_rename(const struct lu_env *env,
631 struct md_object *src_pobj,
632 struct md_object *tgt_pobj,
633 const struct lu_fid *lf,
634 const struct lu_name *lsname,
635 struct md_object *tobj,
636 const struct lu_name *ltname,
642 static int mdd_dummy_link(const struct lu_env *env,
643 struct md_object *tgt_obj,
644 struct md_object *src_obj,
645 const struct lu_name *lname,
651 static int mdd_dummy_unlink(const struct lu_env *env,
652 struct md_object *pobj,
653 struct md_object *cobj,
654 const struct lu_name *lname,
661 static struct md_dir_operations mdd_obf_dir_ops = {
662 .mdo_lookup = obf_lookup,
663 .mdo_create = mdd_dummy_create,
664 .mdo_rename = mdd_dummy_rename,
665 .mdo_link = mdd_dummy_link,
666 .mdo_unlink = mdd_dummy_unlink
669 static struct md_dir_operations mdd_lpf_dir_ops = {
670 .mdo_lookup = mdd_lookup,
671 .mdo_create = mdd_dummy_create,
672 .mdo_rename = mdd_dummy_rename,
673 .mdo_link = mdd_dummy_link,
674 .mdo_unlink = mdd_dummy_unlink
677 static struct md_object *mdo_locate(const struct lu_env *env,
678 struct md_device *md,
679 const struct lu_fid *fid)
681 struct lu_object *obj;
682 struct md_object *mdo;
684 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
686 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
687 LASSERT(obj != NULL);
690 mdo = ERR_PTR(PTR_ERR(obj));
695 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
697 struct md_object *mdo;
698 struct mdd_object *mdd_lpf;
699 struct lu_fid fid = LU_LPF_FID;
703 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
704 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
709 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
711 RETURN(PTR_ERR(mdo));
713 LASSERT(lu_object_exists(&mdo->mo_lu));
715 mdd_lpf = md2mdd_obj(mdo);
716 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
717 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
723 * Create special in-memory "fid" object for open-by-fid.
725 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
727 struct md_object *mdo;
728 struct mdd_object *mdd_obf;
729 struct lu_fid fid = LU_OBF_FID;
732 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
733 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
737 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
739 RETURN(PTR_ERR(mdo));
741 LASSERT(lu_object_exists(&mdo->mo_lu));
743 mdd_obf = md2mdd_obj(mdo);
744 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
745 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
750 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
751 struct mdd_device *m)
753 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
754 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
755 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
757 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
758 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
759 m->mdd_dot_lustre_objs.mdd_obf = NULL;
761 if (m->mdd_dot_lustre != NULL) {
762 mdd_object_put(env, m->mdd_dot_lustre);
763 m->mdd_dot_lustre = NULL;
767 /** Setup ".lustre" directory object */
768 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
770 struct md_object *mdo;
775 /* Create ".lustre" directory in ROOT. */
776 fid = LU_DOT_LUSTRE_FID;
777 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
779 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
783 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
785 RETURN(PTR_ERR(mdo));
786 LASSERT(lu_object_exists(&mdo->mo_lu));
788 m->mdd_dot_lustre = md2mdd_obj(mdo);
790 rc = mdd_obf_setup(env, m);
792 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
793 mdd2obd_dev(m)->obd_name, rc);
797 rc = mdd_lpf_setup(env, m);
799 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
800 mdd2obd_dev(m)->obd_name, rc);
807 mdd_dot_lustre_cleanup(env, m);
813 static struct llog_operations hsm_actions_logops;
816 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
817 * object in obd_device
819 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
820 struct mdd_device *m)
822 struct obd_device *obd = mdd2obd_dev(m);
823 struct llog_ctxt *ctxt = NULL;
827 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
828 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
830 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
831 obd, &hsm_actions_logops);
833 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
838 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
841 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
844 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
846 GOTO(out_cleanup, rc);
849 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
857 llog_cat_close(env, ctxt->loc_handle);
858 ctxt->loc_handle = NULL;
860 llog_cleanup(env, ctxt);
866 * cleanup the context created by llog_setup_named()
868 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
869 struct mdd_device *m)
871 struct obd_device *obd = mdd2obd_dev(m);
872 struct llog_ctxt *lctxt;
875 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
877 llog_cat_close(env, lctxt->loc_handle);
878 lctxt->loc_handle = NULL;
879 llog_cleanup(env, lctxt);
885 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
886 struct lustre_cfg *cfg)
888 barrier_deregister(m->mdd_bottom);
889 lfsck_degister(env, m->mdd_bottom);
890 mdd_hsm_actions_llog_fini(env, m);
891 mdd_changelog_fini(env, m);
892 orph_index_fini(env, m);
893 mdd_dot_lustre_cleanup(env, m);
894 if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) {
895 nm_config_file_deregister_tgt(env,
896 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
897 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL;
899 if (m->mdd_los != NULL) {
900 local_oid_storage_fini(env, m->mdd_los);
903 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
905 if (m->mdd_child_exp)
906 obd_disconnect(m->mdd_child_exp);
909 static int mdd_process_config(const struct lu_env *env,
910 struct lu_device *d, struct lustre_cfg *cfg)
912 struct mdd_device *m = lu2mdd_dev(d);
913 struct dt_device *dt = m->mdd_child;
914 struct lu_device *next = &dt->dd_lu_dev;
918 switch (cfg->lcfg_command) {
920 struct obd_device *obd = mdd2obd_dev(m);
922 rc = class_process_proc_param(PARAM_MDD, obd->obd_vars, cfg, m);
923 if (rc > 0 || rc == -ENOSYS)
924 /* we don't understand; pass it on */
925 rc = next->ld_ops->ldo_process_config(env, next, cfg);
929 rc = next->ld_ops->ldo_process_config(env, next, cfg);
932 dt_conf_get(env, dt, &m->mdd_dt_conf);
934 case LCFG_PRE_CLEANUP:
935 rc = next->ld_ops->ldo_process_config(env, next, cfg);
936 mdd_generic_thread_stop(&m->mdd_orph_cleanup_thread);
939 rc = next->ld_ops->ldo_process_config(env, next, cfg);
940 lu_dev_del_linkage(d->ld_site, d);
941 mdd_device_shutdown(env, m, cfg);
944 rc = next->ld_ops->ldo_process_config(env, next, cfg);
951 static int mdd_recovery_complete(const struct lu_env *env,
954 struct mdd_device *mdd = lu2mdd_dev(d);
955 struct lu_device *next;
959 LASSERT(mdd != NULL);
960 next = &mdd->mdd_child->dd_lu_dev;
962 /* XXX: orphans handling. */
963 if (!mdd->mdd_bottom->dd_rdonly)
964 mdd_orphan_cleanup(env, mdd);
965 rc = next->ld_ops->ldo_recovery_complete(env, next);
970 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
971 const struct lu_fid *pfid, const char *name,
972 __u32 mode, struct lu_fid *fid)
974 struct dt_object *parent, *dto;
979 LASSERT(!fid_is_zero(pfid));
980 parent = dt_locate(env, mdd->mdd_bottom, pfid);
981 if (unlikely(IS_ERR(parent)))
982 RETURN(PTR_ERR(parent));
984 /* create local file/dir, if @fid is passed then try to use it */
985 if (fid_is_zero(fid))
986 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
989 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
993 GOTO(out_put, rc = PTR_ERR(dto));
994 *fid = *lu_object_fid(&dto->do_lu);
995 /* since stack is not fully set up the local_storage uses own stack
996 * and we should drop its object from cache */
997 dt_object_put_nocache(env, dto);
1000 dt_object_put(env, parent);
1004 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1005 enum lfsck_events event)
1010 static int mdd_prepare(const struct lu_env *env,
1011 struct lu_device *pdev,
1012 struct lu_device *cdev)
1014 struct mdd_device *mdd = lu2mdd_dev(cdev);
1015 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1016 struct nm_config_file *nodemap_config;
1017 struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
1023 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1027 /* Setup local dirs */
1028 fid.f_seq = FID_SEQ_LOCAL_NAME;
1031 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1036 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1041 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1042 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1043 mdd_root_dir_name, S_IFDIR |
1044 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1046 CERROR("%s: create root fid failed: rc = %d\n",
1047 mdd2obd_dev(mdd)->obd_name, rc);
1051 mdd->mdd_root_fid = fid;
1052 rc = mdd_dot_lustre_setup(env, mdd);
1054 CERROR("%s: initializing .lustre failed: rc = %d\n",
1055 mdd2obd_dev(mdd)->obd_name, rc);
1059 /* Normal client usually send root access to MDT0 directly,
1060 * the root FID on non-MDT0 will only be used by echo client. */
1061 mdd->mdd_root_fid = fid;
1064 rc = orph_index_init(env, mdd);
1068 rc = mdd_changelog_init(env, mdd);
1070 CERROR("%s: failed to initialize changelog: rc = %d\n",
1071 mdd2obd_dev(mdd)->obd_name, rc);
1075 rc = mdd_hsm_actions_llog_init(env, mdd);
1077 GOTO(out_changelog, rc);
1079 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1081 if (IS_ERR(nodemap_config)) {
1082 rc = PTR_ERR(nodemap_config);
1086 obt->obt_nodemap_config_file = nodemap_config;
1089 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1090 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1093 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1094 mdd2obd_dev(mdd)->obd_name, rc);
1095 GOTO(out_nodemap, rc);
1098 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1100 CERROR("%s: failed to register to barrier: rc = %d\n",
1101 mdd2obd_dev(mdd)->obd_name, rc);
1102 GOTO(out_lfsck, rc);
1108 lfsck_degister(env, mdd->mdd_bottom);
1110 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1111 obt->obt_nodemap_config_file = NULL;
1113 mdd_hsm_actions_llog_fini(env, mdd);
1115 mdd_changelog_fini(env, mdd);
1117 orph_index_fini(env, mdd);
1119 if (mdd_seq_site(mdd)->ss_node_id == 0)
1120 mdd_dot_lustre_cleanup(env, mdd);
1122 local_oid_storage_fini(env, mdd->mdd_los);
1123 mdd->mdd_los = NULL;
1128 const struct lu_device_operations mdd_lu_ops = {
1129 .ldo_object_alloc = mdd_object_alloc,
1130 .ldo_process_config = mdd_process_config,
1131 .ldo_recovery_complete = mdd_recovery_complete,
1132 .ldo_prepare = mdd_prepare,
1135 static int mdd_root_get(const struct lu_env *env,
1136 struct md_device *m, struct lu_fid *f)
1138 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1141 *f = mdd->mdd_root_fid;
1146 * No permission check is needed.
1148 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1149 struct obd_statfs *sfs)
1151 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1156 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1158 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1163 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1164 struct md_device *m)
1166 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1168 return &mdd->mdd_dt_conf;
1171 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1174 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1176 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1177 return (*h == NULL ? -ENOENT : 0);
1180 static struct lu_device *mdd_device_free(const struct lu_env *env,
1181 struct lu_device *lu)
1183 struct mdd_device *m = lu2mdd_dev(lu);
1186 LASSERT(atomic_read(&lu->ld_ref) == 0);
1187 md_device_fini(&m->mdd_md_dev);
1192 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1193 struct lu_device_type *t,
1194 struct lustre_cfg *lcfg)
1196 struct lu_device *l;
1197 struct mdd_device *m;
1201 l = ERR_PTR(-ENOMEM);
1206 md_device_init(&m->mdd_md_dev, t);
1207 rc = mdd_init0(env, m, t, lcfg);
1209 mdd_device_free(env, l);
1218 * we use exports to track all mdd users
1220 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1221 struct obd_device *obd, struct obd_uuid *cluuid,
1222 struct obd_connect_data *data, void *localdata)
1224 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1225 struct lustre_handle conn;
1229 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1231 rc = class_connect(&conn, obd, cluuid);
1235 *exp = class_conn2export(&conn);
1237 /* Why should there ever be more than 1 connect? */
1238 LASSERT(mdd->mdd_connects == 0);
1239 mdd->mdd_connects++;
1245 * once last export (we don't count self-export) disappeared
1246 * mdd can be released
1248 static int mdd_obd_disconnect(struct obd_export *exp)
1250 struct obd_device *obd = exp->exp_obd;
1251 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1252 int rc, release = 0;
1255 mdd->mdd_connects--;
1256 if (mdd->mdd_connects == 0)
1259 rc = class_disconnect(exp);
1261 if (rc == 0 && release)
1262 class_manual_cleanup(obd);
1266 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1267 __u32 keylen, void *key, __u32 *vallen, void *val)
1271 if (KEY_IS(KEY_OSP_CONNECTED)) {
1272 struct obd_device *obd = exp->exp_obd;
1273 struct mdd_device *mdd;
1275 if (!obd->obd_set_up || obd->obd_stopping)
1278 mdd = lu2mdd_dev(obd->obd_lu_dev);
1280 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1288 static int mdd_obd_set_info_async(const struct lu_env *env,
1289 struct obd_export *exp,
1290 __u32 keylen, void *key,
1291 __u32 vallen, void *val,
1292 struct ptlrpc_request_set *set)
1294 struct obd_device *obd = exp->exp_obd;
1295 struct mdd_device *mdd;
1298 if (!obd->obd_set_up || obd->obd_stopping)
1301 mdd = lu2mdd_dev(obd->obd_lu_dev);
1303 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1308 static struct obd_ops mdd_obd_device_ops = {
1309 .o_owner = THIS_MODULE,
1310 .o_connect = mdd_obd_connect,
1311 .o_disconnect = mdd_obd_disconnect,
1312 .o_get_info = mdd_obd_get_info,
1313 .o_set_info_async = mdd_obd_set_info_async,
1316 static int mdd_changelog_user_register(const struct lu_env *env,
1317 struct mdd_device *mdd, int *id)
1319 struct llog_ctxt *ctxt;
1320 struct llog_changelog_user_rec *rec;
1324 ctxt = llog_get_context(mdd2obd_dev(mdd),
1325 LLOG_CHANGELOG_USER_ORIG_CTXT);
1331 llog_ctxt_put(ctxt);
1335 /* Assume we want it on since somebody registered */
1336 rc = mdd_changelog_on(env, mdd);
1340 rec->cur_hdr.lrh_len = sizeof(*rec);
1341 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1342 spin_lock(&mdd->mdd_cl.mc_user_lock);
1343 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1344 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1345 CERROR("Maximum number of changelog users exceeded!\n");
1346 GOTO(out, rc = -EOVERFLOW);
1348 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1349 rec->cur_endrec = mdd->mdd_cl.mc_index;
1351 rec->cur_time = (__u32)get_seconds();
1352 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1355 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1357 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1359 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1362 llog_ctxt_put(ctxt);
1366 struct mdd_changelog_user_purge {
1368 __u32 mcup_usercount;
1374 * changelog_user_purge callback
1376 * Is called once per user.
1378 * Check to see the user requested is available from rec.
1379 * Truncate the changelog.
1380 * Keep track of the total number of users (calls).
1382 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1383 struct llog_handle *llh,
1384 struct llog_rec_hdr *hdr, void *data)
1386 struct llog_changelog_user_rec *rec;
1387 struct mdd_changelog_user_purge *mcup = data;
1388 struct llog_cookie cookie;
1393 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1396 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1398 mcup->mcup_usercount++;
1400 if (rec->cur_id != mcup->mcup_id) {
1401 /* truncate to the lowest endrec that is not this user */
1402 mcup->mcup_minrec = min(mcup->mcup_minrec,
1407 /* Unregister this user */
1408 cookie.lgc_lgl = llh->lgh_id;
1409 cookie.lgc_index = hdr->lrh_index;
1411 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1414 mcup->mcup_found = true;
1415 mcup->mcup_usercount--;
1421 int mdd_changelog_user_purge(const struct lu_env *env,
1422 struct mdd_device *mdd, __u32 id)
1424 struct mdd_changelog_user_purge mcup = {
1426 .mcup_found = false,
1427 .mcup_usercount = 0,
1428 .mcup_minrec = ULLONG_MAX,
1430 struct llog_ctxt *ctxt;
1435 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1436 mdd2obd_dev(mdd)->obd_name, id);
1438 ctxt = llog_get_context(mdd2obd_dev(mdd),
1439 LLOG_CHANGELOG_USER_ORIG_CTXT);
1441 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1442 GOTO(out, rc = -ENXIO);
1444 rc = llog_cat_process(env, ctxt->loc_handle,
1445 mdd_changelog_user_purge_cb, &mcup,
1448 if ((rc == 0) && mcup.mcup_found) {
1449 CDEBUG(D_IOCTL, "%s: Purging changelog entries for user %d "
1451 mdd2obd_dev(mdd)->obd_name, id, mcup.mcup_minrec);
1452 /* Cancelling record 0 destroys the entire changelog, make sure
1453 we don't do that unless we mean it. */
1454 if (mcup.mcup_minrec != 0 || mcup.mcup_usercount == 0) {
1455 rc = mdd_changelog_llog_cancel(env, mdd,
1459 CWARN("%s: No changelog for user %u; rc=%d\n",
1460 mdd2obd_dev(mdd)->obd_name, id, rc);
1461 GOTO(out, rc = -ENOENT);
1464 if ((rc == 0) && (mcup.mcup_usercount == 0)) {
1465 /* No more users; turn changelogs off */
1466 CDEBUG(D_IOCTL, "turning off changelogs\n");
1467 rc = mdd_changelog_off(env, mdd);
1473 llog_ctxt_put(ctxt);
1478 struct mdd_changelog_user_clear {
1486 * changelog_clear callback
1488 * Is called once per user.
1490 * Check to see the user requested is available from rec.
1491 * Check the oldest (smallest) record for boundary conditions.
1492 * Truncate the changelog.
1494 static int mdd_changelog_clear_cb(const struct lu_env *env,
1495 struct llog_handle *llh,
1496 struct llog_rec_hdr *hdr,
1499 struct llog_changelog_user_rec *rec;
1500 struct mdd_changelog_user_clear *mcuc = data;
1505 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1508 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1510 /* Does the changelog id match the requested id? */
1511 if (rec->cur_id != mcuc->mcuc_id) {
1512 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec,
1517 /* cur_endrec is the oldest purgeable record, make sure we're newer */
1518 if (rec->cur_endrec > mcuc->mcuc_endrec) {
1519 CDEBUG(D_IOCTL, "Request %llu out of range: %llu\n",
1520 mcuc->mcuc_endrec, rec->cur_endrec);
1524 /* Flag that we've met all the range and user checks.
1525 * We now know the record to flush.
1527 rec->cur_endrec = mcuc->mcuc_endrec;
1529 rec->cur_time = (__u32)get_seconds();
1530 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1533 mcuc->mcuc_flush = true;
1535 CDEBUG(D_IOCTL, "Rewriting changelog user %u endrec to %llu\n",
1536 mcuc->mcuc_id, rec->cur_endrec);
1538 /* Update the endrec */
1539 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1545 * Clear a changelog up to entry specified by endrec for user id.
1547 static int mdd_changelog_clear(const struct lu_env *env,
1548 struct mdd_device *mdd, __u32 id,
1551 struct mdd_changelog_user_clear mcuc = {
1553 .mcuc_minrec = endrec,
1554 .mcuc_flush = false,
1556 struct llog_ctxt *ctxt;
1562 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
1563 mdd2obd_dev(mdd)->obd_name, id, endrec);
1564 /* start_rec is the newest (largest value) entry in the changelogs*/
1565 spin_lock(&mdd->mdd_cl.mc_lock);
1566 start_rec = mdd->mdd_cl.mc_index;
1567 spin_unlock(&mdd->mdd_cl.mc_lock);
1569 if (start_rec < endrec) {
1570 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
1571 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
1576 mcuc.mcuc_endrec = start_rec;
1577 mcuc.mcuc_minrec = ULLONG_MAX;
1579 mcuc.mcuc_endrec = endrec;
1582 ctxt = llog_get_context(mdd2obd_dev(mdd),
1583 LLOG_CHANGELOG_USER_ORIG_CTXT);
1585 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1586 GOTO(out, rc = -ENXIO);
1588 rc = llog_cat_process(env, ctxt->loc_handle,
1589 mdd_changelog_clear_cb, (void *)&mcuc,
1593 CWARN("%s: Failure to clear the changelog for user %d: %d\n",
1594 mdd2obd_dev(mdd)->obd_name, id, rc);
1595 } else if (mcuc.mcuc_flush) {
1596 /* Cancelling record 0 destroys the entire changelog, make sure
1597 we don't do that unless we mean it. */
1598 if (mcuc.mcuc_minrec != 0) {
1599 CDEBUG(D_IOCTL, "%s: Purging changelog entries up "\
1600 "to %llu\n", mdd2obd_dev(mdd)->obd_name,
1603 rc = mdd_changelog_llog_cancel(env, mdd,
1607 CWARN("%s: No entry for user %d\n",
1608 mdd2obd_dev(mdd)->obd_name, id);
1615 llog_ctxt_put(ctxt);
1621 * May be called remotely from mdt_iocontrol_handle or locally from
1622 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1623 * an obd_ioctl_data format (but local ioctl handler does).
1625 * \param len - data len
1626 * \param karg - ioctl data, in kernel space
1628 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1629 unsigned int cmd, int len, void *karg)
1631 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1632 struct obd_ioctl_data *data = karg;
1636 /* Doesn't use obd_ioctl_data */
1638 case OBD_IOC_CHANGELOG_CLEAR: {
1639 struct changelog_setinfo *cs = karg;
1641 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1642 RETURN(-EINPROGRESS);
1644 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
1646 barrier_exit(mdd->mdd_bottom);
1649 case OBD_IOC_START_LFSCK: {
1650 rc = lfsck_start(env, mdd->mdd_bottom,
1651 (struct lfsck_start_param *)karg);
1654 case OBD_IOC_STOP_LFSCK: {
1655 rc = lfsck_stop(env, mdd->mdd_bottom,
1656 (struct lfsck_stop *)karg);
1659 case OBD_IOC_QUERY_LFSCK: {
1660 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
1661 (struct lfsck_query *)karg);
1666 /* Below ioctls use obd_ioctl_data */
1667 if (len != sizeof(*data)) {
1668 CERROR("Bad ioctl size %d\n", len);
1671 if (data->ioc_version != OBD_IOCTL_VERSION) {
1672 CERROR("Bad magic %x != %x\n", data->ioc_version,
1678 case OBD_IOC_CHANGELOG_REG:
1679 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1680 RETURN(-EINPROGRESS);
1682 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1683 barrier_exit(mdd->mdd_bottom);
1685 case OBD_IOC_CHANGELOG_DEREG:
1686 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1687 RETURN(-EINPROGRESS);
1689 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1);
1690 barrier_exit(mdd->mdd_bottom);
1699 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1700 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
1702 static const struct md_device_operations mdd_ops = {
1703 .mdo_statfs = mdd_statfs,
1704 .mdo_root_get = mdd_root_get,
1705 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1706 .mdo_iocontrol = mdd_iocontrol,
1707 .mdo_dtconf_get = mdd_dtconf_get,
1710 static struct lu_device_type_operations mdd_device_type_ops = {
1711 .ldto_init = mdd_type_init,
1712 .ldto_fini = mdd_type_fini,
1714 .ldto_start = mdd_type_start,
1715 .ldto_stop = mdd_type_stop,
1717 .ldto_device_alloc = mdd_device_alloc,
1718 .ldto_device_free = mdd_device_free,
1720 .ldto_device_fini = mdd_device_fini
1723 static struct lu_device_type mdd_device_type = {
1724 .ldt_tags = LU_DEVICE_MD,
1725 .ldt_name = LUSTRE_MDD_NAME,
1726 .ldt_ops = &mdd_device_type_ops,
1727 .ldt_ctx_tags = LCT_MD_THREAD
1730 /* context key constructor: mdd_key_init */
1731 LU_KEY_INIT(mdd, struct mdd_thread_info);
1733 static void mdd_key_fini(const struct lu_context *ctx,
1734 struct lu_context_key *key, void *data)
1736 struct mdd_thread_info *info = data;
1738 lu_buf_free(&info->mti_big_buf);
1739 lu_buf_free(&info->mti_link_buf);
1740 lu_buf_free(&info->mti_xattr_buf);
1745 /* context key: mdd_thread_key */
1746 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1748 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
1749 int (*func)(void *), void *data, char *name)
1751 struct task_struct *task;
1753 LASSERT(thread->mgt_init == false);
1754 init_completion(&thread->mgt_started);
1755 init_completion(&thread->mgt_finished);
1756 thread->mgt_data = data;
1757 thread->mgt_abort = false;
1758 thread->mgt_init = true;
1760 task = kthread_run(func, thread, name);
1762 complete(&thread->mgt_finished);
1763 return PTR_ERR(task);
1765 wait_for_completion(&thread->mgt_started);
1769 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
1771 if (thread->mgt_init == true) {
1772 thread->mgt_abort = true;
1773 wait_for_completion(&thread->mgt_finished);
1777 static int __init mdd_init(void)
1781 rc = lu_kmem_init(mdd_caches);
1785 changelog_orig_logops = llog_osd_ops;
1786 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
1787 changelog_orig_logops.lop_add = llog_cat_add_rec;
1788 changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1790 hsm_actions_logops = llog_osd_ops;
1791 hsm_actions_logops.lop_add = llog_cat_add_rec;
1792 hsm_actions_logops.lop_declare_add = llog_cat_declare_add_rec;
1794 rc = class_register_type(&mdd_obd_device_ops, NULL, true, NULL,
1795 LUSTRE_MDD_NAME, &mdd_device_type);
1797 lu_kmem_fini(mdd_caches);
1801 static void __exit mdd_exit(void)
1803 class_unregister_type(LUSTRE_MDD_NAME);
1804 lu_kmem_fini(mdd_caches);
1807 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1808 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
1809 MODULE_VERSION(LUSTRE_VERSION_STRING);
1810 MODULE_LICENSE("GPL");
1812 module_init(mdd_init);
1813 module_exit(mdd_exit);