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;
128 mdd->mdd_cl.mc_deniednext = 60; /* 60 secs by default */
130 dev = lustre_cfg_string(lcfg, 0);
134 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
135 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
137 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
138 mdd->mdd_md_dev.md_ops = &mdd_ops;
140 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
144 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
145 /* sync permission changes */
146 mdd->mdd_sync_permission = 1;
147 /* enable changelog garbage collection */
148 mdd->mdd_changelog_gc = 0;
149 /* with a significant amount of idle time */
150 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
151 /* or a significant amount of late indexes */
152 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
153 /* with a reasonable interval between each check */
154 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
155 /* with a very few number of free catalog entries */
156 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
158 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
160 /* we are using service name but not mdd obd name
161 * for compatibility reasons.
162 * It is passed from MDT in lustre_cfg[2] buffer */
163 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
165 obd_disconnect(mdd->mdd_child_exp);
170 static struct lu_device *mdd_device_fini(const struct lu_env *env,
173 struct mdd_device *mdd = lu2mdd_dev(d);
176 lu_dev_del_linkage(d->ld_site, d);
178 mdd_procfs_fini(mdd);
182 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
183 struct llog_rec_hdr *hdr, void *data)
185 struct mdd_device *mdd = (struct mdd_device *)data;
186 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
188 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
189 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
192 "seeing record at index %d/%d/%llu t=%x %.*s in log"
193 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
194 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
195 changelog_rec_name(&rec->cr), PFID(&llh->lgh_id.lgl_oi.oi_fid));
197 mdd->mdd_cl.mc_index = rec->cr.cr_index;
198 return LLOG_PROC_BREAK;
201 static int changelog_user_init_cb(const struct lu_env *env,
202 struct llog_handle *llh,
203 struct llog_rec_hdr *hdr, void *data)
205 struct mdd_device *mdd = (struct mdd_device *)data;
206 struct llog_changelog_user_rec *rec =
207 (struct llog_changelog_user_rec *)hdr;
209 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
210 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
212 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec=%llu"
213 " in log "DFID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
214 rec->cur_id, rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
216 spin_lock(&mdd->mdd_cl.mc_user_lock);
217 mdd->mdd_cl.mc_lastuser = rec->cur_id;
218 mdd->mdd_cl.mc_users++;
219 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
220 mdd->mdd_cl.mc_index = rec->cur_endrec;
221 spin_unlock(&mdd->mdd_cl.mc_user_lock);
223 return LLOG_PROC_BREAK;
226 struct changelog_orphan_data {
228 struct mdd_device *mdd;
231 /* find oldest changelog record index */
232 static int changelog_detect_orphan_cb(const struct lu_env *env,
233 struct llog_handle *llh,
234 struct llog_rec_hdr *hdr, void *data)
236 struct mdd_device *mdd = ((struct changelog_orphan_data *)data)->mdd;
237 struct llog_changelog_rec *rec = container_of(hdr,
238 struct llog_changelog_rec,
241 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
243 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
244 CWARN("%s: invalid record at index %d in log "DFID"\n",
245 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
246 PFID(&llh->lgh_id.lgl_oi.oi_fid));
247 /* try to find some next valid record and thus allow to recover
248 * from a corrupted LLOG, instead to assert and force a crash
253 CDEBUG(D_INFO, "%s: seeing record at index %d/%d/%llu t=%x %.*s in log "
254 DFID"\n", mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
255 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
256 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
257 PFID(&llh->lgh_id.lgl_oi.oi_fid));
259 ((struct changelog_orphan_data *)data)->index = rec->cr.cr_index;
260 return LLOG_PROC_BREAK;
263 /* find oldest changelog user index */
264 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
265 struct llog_handle *llh,
266 struct llog_rec_hdr *hdr, void *data)
268 struct mdd_device *mdd = ((struct changelog_orphan_data *)data)->mdd;
269 struct llog_changelog_user_rec *rec = container_of(hdr,
270 struct llog_changelog_user_rec,
273 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
275 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC) {
276 CWARN("%s: invalid user at index %d in log "DFID"\n",
277 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
278 PFID(&llh->lgh_id.lgl_oi.oi_fid));
279 /* try to find some next valid record and thus allow to recover
280 * from a corrupted LLOG, instead to assert and force a crash
285 CDEBUG(D_INFO, "%s: seeing user at index %d/%d id=%d endrec=%llu in "
286 "log "DFID"\n", mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
287 rec->cur_hdr.lrh_index, rec->cur_id, rec->cur_endrec,
288 PFID(&llh->lgh_id.lgl_oi.oi_fid));
290 if (((struct changelog_orphan_data *)data)->index == 0 ||
291 rec->cur_endrec < ((struct changelog_orphan_data *)data)->index)
292 ((struct changelog_orphan_data *)data)->index = rec->cur_endrec;
297 struct changelog_cancel_cookie {
299 struct mdd_device *mdd;
302 static int llog_changelog_cancel_cb(const struct lu_env *env,
303 struct llog_handle *llh,
304 struct llog_rec_hdr *hdr, void *data)
306 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
307 struct llog_cookie cookie;
308 struct changelog_cancel_cookie *cl_cookie =
309 (struct changelog_cancel_cookie *)data;
314 /* This is always a (sub)log, not the catalog */
315 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
317 /* if current context is GC-thread allow it to stop upon umount
318 * remaining records cleanup will occur upon next mount
320 * also during testing, wait for GC-thread to be released
322 * XXX this requires the GC-thread to not fork a sub-thread via
323 * llog[_cat]_process_or_fork() and we may think to also implement
324 * this shutdown mechanism for manually started user unregister which
325 * can also take a long time if huge backlog of records
327 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
328 /* wait to be released */
329 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
332 if (kthread_should_stop())
333 RETURN(LLOG_PROC_BREAK);
336 if (rec->cr.cr_index > cl_cookie->endrec)
337 /* records are in order, so we're done */
338 RETURN(LLOG_PROC_BREAK);
340 cookie.lgc_lgl = llh->lgh_id;
341 cookie.lgc_index = hdr->lrh_index;
343 /* cancel them one at a time. I suppose we could store up the cookies
344 * and cancel them all at once; probably more efficient, but this is
345 * done as a user call, so who cares... */
346 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
348 RETURN(rc < 0 ? rc : 0);
351 static int llog_changelog_cancel(const struct lu_env *env,
352 struct llog_ctxt *ctxt,
353 struct changelog_cancel_cookie *cookie)
355 struct llog_handle *cathandle = ctxt->loc_handle;
360 /* This should only be called with the catalog handle */
361 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
363 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
366 /* 0 or 1 means we're done */
369 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
370 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
371 PFID(&cathandle->lgh_id.lgl_oi.oi_fid), rc);
377 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
381 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
385 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
388 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
389 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
390 CERROR("Changelogs cannot be enabled due to error "
391 "condition (see %s log).\n",
392 mdd2obd_dev(mdd)->obd_name);
395 spin_lock(&mdd->mdd_cl.mc_lock);
396 mdd->mdd_cl.mc_flags |= CLM_ON;
397 spin_unlock(&mdd->mdd_cl.mc_lock);
398 rc = mdd_changelog_write_header(env, mdd, CLM_START);
404 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
408 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
411 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
412 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
413 spin_lock(&mdd->mdd_cl.mc_lock);
414 mdd->mdd_cl.mc_flags &= ~CLM_ON;
415 spin_unlock(&mdd->mdd_cl.mc_lock);
420 static int mdd_changelog_llog_init(const struct lu_env *env,
421 struct mdd_device *mdd)
423 struct obd_device *obd = mdd2obd_dev(mdd);
424 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
425 struct changelog_orphan_data changelog_orphan = { .index = 0,
427 user_orphan = { .index = 0,
433 /* LU-2844 mdd setup failure should not cause umount oops */
434 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
437 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
438 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
439 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
440 obd, &llog_common_cat_ops);
442 CERROR("%s: changelog llog setup failed: rc = %d\n",
447 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
450 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
453 GOTO(out_cleanup, rc);
455 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
459 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
460 changelog_init_cb, mdd);
463 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
468 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
469 mdd->mdd_cl.mc_index);
471 /* setup user changelog */
472 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
473 obd, &llog_common_cat_ops);
475 CERROR("%s: changelog users llog setup failed: rc = %d\n",
480 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
483 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
486 GOTO(out_ucleanup, rc);
488 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
490 GOTO(out_uclose, rc);
492 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
493 changelog_user_init_cb, mdd);
495 CERROR("%s: changelog user init failed: rc = %d\n",
497 GOTO(out_uclose, rc);
500 /* If we have registered users, assume we want changelogs on */
501 if (mdd->mdd_cl.mc_lastuser > 0) {
502 rc = mdd_changelog_on(env, mdd);
504 GOTO(out_uclose, rc);
507 /* find and clear any orphan changelog records (1st record index <
508 * smallest of all users current index), likely to come from an
509 * interrupted manual or GC-thread purge, as its user record had
511 * XXX we may wait for a still registered user clear operation to
512 * do the job, but it may then take a long time to reach the user's
513 * real targetted records if a huge purge backlog is still to be
514 * processed as a long time idle user record could have been deleted
515 * XXX we may need to run end of purge as a separate thread
517 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
518 &changelog_orphan, 0, 0);
520 CERROR("%s: changelog detect orphan failed: rc = %d\n",
522 GOTO(out_uclose, rc);
524 rc = llog_cat_process(env, uctxt->loc_handle,
525 changelog_user_detect_orphan_cb,
528 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
530 GOTO(out_uclose, rc);
532 if (unlikely(changelog_orphan.index < user_orphan.index)) {
533 struct changelog_cancel_cookie cl_cookie = {
534 .endrec = user_orphan.index,
538 CWARN("%s : orphan changelog records found, starting from "
539 "index %llu to index %llu, being cleared now\n",
540 obd->obd_name, changelog_orphan.index, user_orphan.index);
542 /* XXX we may need to run end of purge as a separate thread */
543 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
545 CERROR("%s: purge of changelog orphan records failed: "
546 "rc = %d\n", obd->obd_name, rc);
547 GOTO(out_uclose, rc);
552 llog_ctxt_put(uctxt);
555 llog_cat_close(env, uctxt->loc_handle);
557 llog_cleanup(env, uctxt);
559 llog_cat_close(env, ctxt->loc_handle);
561 llog_cleanup(env, ctxt);
565 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
567 struct obd_device *obd = mdd2obd_dev(mdd);
570 mdd->mdd_cl.mc_index = 0;
571 spin_lock_init(&mdd->mdd_cl.mc_lock);
572 mdd->mdd_cl.mc_starttime = ktime_get();
573 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
574 mdd->mdd_cl.mc_lastuser = 0;
576 /* ensure a GC check will, and a thread run may, occur upon start */
577 mdd->mdd_cl.mc_gc_time = 0;
578 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
580 rc = mdd_changelog_llog_init(env, mdd);
582 CERROR("%s: changelog setup during init failed: rc = %d\n",
584 mdd->mdd_cl.mc_flags |= CLM_ERR;
590 static void mdd_changelog_fini(const struct lu_env *env,
591 struct mdd_device *mdd)
593 struct obd_device *obd = mdd2obd_dev(mdd);
594 struct llog_ctxt *ctxt;
596 mdd->mdd_cl.mc_flags = 0;
599 /* stop GC-thread if running */
600 spin_lock(&mdd->mdd_cl.mc_lock);
601 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
602 /* avoid any attempt to run a GC-thread */
603 mdd->mdd_cl.mc_gc_task = current;
604 spin_unlock(&mdd->mdd_cl.mc_lock);
606 struct task_struct *gc_task;
608 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
609 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
610 /* need to wait for birthing GC-thread to be started
611 * and to have set mc_gc_task to itself
613 spin_unlock(&mdd->mdd_cl.mc_lock);
614 schedule_timeout(usecs_to_jiffies(10));
615 /* go back to fully check if GC-thread has started or
616 * even already exited or if a new one is starting...
620 /* take a reference on task_struct to avoid it to be freed
623 gc_task = mdd->mdd_cl.mc_gc_task;
624 get_task_struct(gc_task);
625 spin_unlock(&mdd->mdd_cl.mc_lock);
626 kthread_stop(gc_task);
627 put_task_struct(gc_task);
630 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
632 llog_cat_close(env, ctxt->loc_handle);
633 llog_cleanup(env, ctxt);
635 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
637 llog_cat_close(env, ctxt->loc_handle);
638 llog_cleanup(env, ctxt);
642 /** Remove entries with indicies up to and including \a endrec from the
649 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
652 struct obd_device *obd = mdd2obd_dev(mdd);
653 struct llog_ctxt *ctxt;
654 long long unsigned cur;
655 struct changelog_cancel_cookie cookie;
658 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
662 spin_lock(&mdd->mdd_cl.mc_lock);
663 cur = (long long)mdd->mdd_cl.mc_index;
664 spin_unlock(&mdd->mdd_cl.mc_lock);
668 /* purge to "0" is shorthand for everything */
672 /* If purging all records, write a header entry so we don't have an
673 empty catalog and we're sure to have a valid starting index next
674 time. In case of crash, we just restart with old log so we're
677 /* XXX: transaction is started by llog itself */
678 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
683 /* Some records were purged, so reset repeat-access time (so we
684 record new mtime update records, so users can see a file has been
685 changed since the last purge) */
686 mdd->mdd_cl.mc_starttime = ktime_get();
688 cookie.endrec = endrec;
690 rc = llog_changelog_cancel(env, ctxt, &cookie);
696 /** Add a CL_MARK record to the changelog
698 * \param markerflags - CLM_*
701 int mdd_changelog_write_header(const struct lu_env *env,
702 struct mdd_device *mdd, int markerflags)
704 struct obd_device *obd = mdd2obd_dev(mdd);
705 struct llog_changelog_rec *rec;
707 struct llog_ctxt *ctxt;
709 int len = strlen(obd->obd_name);
714 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
715 mdd->mdd_cl.mc_starttime = ktime_get();
719 reclen = llog_data_len(sizeof(*rec) + len);
720 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
721 if (buf->lb_buf == NULL)
725 rec->cr.cr_flags = CLF_VERSION;
726 rec->cr.cr_type = CL_MARK;
727 rec->cr.cr_namelen = len;
728 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
729 /* Status and action flags */
730 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
731 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
733 rec->cr_hdr.lrh_type = CHANGELOG_REC;
734 rec->cr.cr_time = cl_time();
735 spin_lock(&mdd->mdd_cl.mc_lock);
736 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
737 spin_unlock(&mdd->mdd_cl.mc_lock);
739 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
742 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
747 /* assume on or off event; reset repeat-access time */
748 mdd->mdd_cl.mc_starttime = ktime_get();
753 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
754 * are valid. We also check if object with passed fid exists or not.
756 static int obf_lookup(const struct lu_env *env, struct md_object *p,
757 const struct lu_name *lname, struct lu_fid *f,
758 struct md_op_spec *spec)
760 char *name = (char *)lname->ln_name;
761 struct mdd_device *mdd = mdo2mdd(p);
762 struct mdd_object *child;
768 sscanf(name, SFID, RFID(f));
769 if (!fid_is_sane(f)) {
770 CWARN("%s: Trying to lookup invalid FID [%s] in %s/%s, FID "
771 "format should be "DFID"\n", mdd2obd_dev(mdd)->obd_name,
772 lname->ln_name, dot_lustre_name, mdd_obf_dir_name,
773 (__u64)FID_SEQ_NORMAL, 1, 0);
774 GOTO(out, rc = -EINVAL);
777 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
778 !fid_seq_is_dot(f->f_seq)) {
779 CWARN("%s: Trying to lookup invalid FID "DFID" in %s/%s, "
780 "sequence should be >= %#llx or within [%#llx,"
781 "%#llx].\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
782 dot_lustre_name, mdd_obf_dir_name, (__u64)FID_SEQ_NORMAL,
783 (__u64)FID_SEQ_IGIF, (__u64)FID_SEQ_IGIF_MAX);
784 GOTO(out, rc = -EINVAL);
787 /* Check if object with this fid exists */
788 child = mdd_object_find(env, mdd, f);
790 GOTO(out, rc = PTR_ERR(child));
792 if (mdd_object_exists(child) == 0)
795 mdd_object_put(env, child);
801 static int mdd_dummy_create(const struct lu_env *env,
802 struct md_object *pobj,
803 const struct lu_name *lname,
804 struct md_object *child,
805 struct md_op_spec *spec,
811 static int mdd_dummy_rename(const struct lu_env *env,
812 struct md_object *src_pobj,
813 struct md_object *tgt_pobj,
814 const struct lu_fid *lf,
815 const struct lu_name *lsname,
816 struct md_object *tobj,
817 const struct lu_name *ltname,
823 static int mdd_dummy_link(const struct lu_env *env,
824 struct md_object *tgt_obj,
825 struct md_object *src_obj,
826 const struct lu_name *lname,
832 static int mdd_dummy_unlink(const struct lu_env *env,
833 struct md_object *pobj,
834 struct md_object *cobj,
835 const struct lu_name *lname,
842 static struct md_dir_operations mdd_obf_dir_ops = {
843 .mdo_lookup = obf_lookup,
844 .mdo_create = mdd_dummy_create,
845 .mdo_rename = mdd_dummy_rename,
846 .mdo_link = mdd_dummy_link,
847 .mdo_unlink = mdd_dummy_unlink
850 static struct md_dir_operations mdd_lpf_dir_ops = {
851 .mdo_lookup = mdd_lookup,
852 .mdo_create = mdd_dummy_create,
853 .mdo_rename = mdd_dummy_rename,
854 .mdo_link = mdd_dummy_link,
855 .mdo_unlink = mdd_dummy_unlink
858 static struct md_object *mdo_locate(const struct lu_env *env,
859 struct md_device *md,
860 const struct lu_fid *fid)
862 struct lu_object *obj;
863 struct md_object *mdo;
865 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
867 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
868 LASSERT(obj != NULL);
871 mdo = ERR_PTR(PTR_ERR(obj));
876 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
878 struct md_object *mdo;
879 struct mdd_object *mdd_lpf;
880 struct lu_fid fid = LU_LPF_FID;
884 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
885 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
890 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
892 RETURN(PTR_ERR(mdo));
894 LASSERT(lu_object_exists(&mdo->mo_lu));
896 mdd_lpf = md2mdd_obj(mdo);
897 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
898 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
904 * Create special in-memory "fid" object for open-by-fid.
906 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
908 struct md_object *mdo;
909 struct mdd_object *mdd_obf;
910 struct lu_fid fid = LU_OBF_FID;
913 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
914 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
918 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
920 RETURN(PTR_ERR(mdo));
922 LASSERT(lu_object_exists(&mdo->mo_lu));
924 mdd_obf = md2mdd_obj(mdo);
925 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
926 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
931 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
932 struct mdd_device *m)
934 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
935 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
936 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
938 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
939 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
940 m->mdd_dot_lustre_objs.mdd_obf = NULL;
942 if (m->mdd_dot_lustre != NULL) {
943 mdd_object_put(env, m->mdd_dot_lustre);
944 m->mdd_dot_lustre = NULL;
948 /** Setup ".lustre" directory object */
949 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
951 struct md_object *mdo;
956 /* Create ".lustre" directory in ROOT. */
957 fid = LU_DOT_LUSTRE_FID;
958 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
960 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
964 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
966 RETURN(PTR_ERR(mdo));
967 LASSERT(lu_object_exists(&mdo->mo_lu));
969 m->mdd_dot_lustre = md2mdd_obj(mdo);
971 rc = mdd_obf_setup(env, m);
973 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
974 mdd2obd_dev(m)->obd_name, rc);
978 rc = mdd_lpf_setup(env, m);
980 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
981 mdd2obd_dev(m)->obd_name, rc);
988 mdd_dot_lustre_cleanup(env, m);
994 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
995 * object in obd_device
997 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
998 struct mdd_device *m)
1000 struct obd_device *obd = mdd2obd_dev(m);
1001 struct llog_ctxt *ctxt = NULL;
1005 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1006 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1008 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1009 obd, &llog_common_cat_ops);
1011 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1016 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1019 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1022 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1024 GOTO(out_cleanup, rc);
1027 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1029 GOTO(out_close, rc);
1031 llog_ctxt_put(ctxt);
1035 llog_cat_close(env, ctxt->loc_handle);
1036 ctxt->loc_handle = NULL;
1038 llog_cleanup(env, ctxt);
1044 * cleanup the context created by llog_setup_named()
1046 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1047 struct mdd_device *m)
1049 struct obd_device *obd = mdd2obd_dev(m);
1050 struct llog_ctxt *lctxt;
1053 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1055 llog_cat_close(env, lctxt->loc_handle);
1056 lctxt->loc_handle = NULL;
1057 llog_cleanup(env, lctxt);
1063 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1064 struct lustre_cfg *cfg)
1066 barrier_deregister(m->mdd_bottom);
1067 lfsck_degister(env, m->mdd_bottom);
1068 mdd_hsm_actions_llog_fini(env, m);
1069 mdd_changelog_fini(env, m);
1070 mdd_orphan_index_fini(env, m);
1071 mdd_dot_lustre_cleanup(env, m);
1072 if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) {
1073 nm_config_file_deregister_tgt(env,
1074 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
1075 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL;
1077 if (m->mdd_los != NULL) {
1078 local_oid_storage_fini(env, m->mdd_los);
1081 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1083 if (m->mdd_child_exp)
1084 obd_disconnect(m->mdd_child_exp);
1087 static int mdd_process_config(const struct lu_env *env,
1088 struct lu_device *d, struct lustre_cfg *cfg)
1090 struct mdd_device *m = lu2mdd_dev(d);
1091 struct dt_device *dt = m->mdd_child;
1092 struct lu_device *next = &dt->dd_lu_dev;
1096 switch (cfg->lcfg_command) {
1098 struct obd_device *obd = mdd2obd_dev(m);
1100 rc = class_process_proc_param(PARAM_MDD, obd->obd_vars, cfg, m);
1101 if (rc > 0 || rc == -ENOSYS)
1102 /* we don't understand; pass it on */
1103 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1107 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1110 dt_conf_get(env, dt, &m->mdd_dt_conf);
1112 case LCFG_PRE_CLEANUP:
1113 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1114 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1117 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1118 lu_dev_del_linkage(d->ld_site, d);
1119 mdd_device_shutdown(env, m, cfg);
1122 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1129 static int mdd_recovery_complete(const struct lu_env *env,
1130 struct lu_device *d)
1132 struct mdd_device *mdd = lu2mdd_dev(d);
1133 struct lu_device *next;
1137 LASSERT(mdd != NULL);
1138 next = &mdd->mdd_child->dd_lu_dev;
1140 if (!mdd->mdd_bottom->dd_rdonly)
1141 mdd_orphan_cleanup(env, mdd);
1142 rc = next->ld_ops->ldo_recovery_complete(env, next);
1147 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1148 const struct lu_fid *pfid, const char *name,
1149 __u32 mode, struct lu_fid *fid)
1151 struct dt_object *parent, *dto;
1156 LASSERT(!fid_is_zero(pfid));
1157 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1158 if (unlikely(IS_ERR(parent)))
1159 RETURN(PTR_ERR(parent));
1161 /* create local file/dir, if @fid is passed then try to use it */
1162 if (fid_is_zero(fid))
1163 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1166 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1170 GOTO(out_put, rc = PTR_ERR(dto));
1171 *fid = *lu_object_fid(&dto->do_lu);
1172 /* since stack is not fully set up the local_storage uses own stack
1173 * and we should drop its object from cache */
1174 dt_object_put_nocache(env, dto);
1177 dt_object_put(env, parent);
1181 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1182 enum lfsck_events event)
1187 static int mdd_prepare(const struct lu_env *env,
1188 struct lu_device *pdev,
1189 struct lu_device *cdev)
1191 struct mdd_device *mdd = lu2mdd_dev(cdev);
1192 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1193 struct nm_config_file *nodemap_config;
1194 struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
1200 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1204 /* Setup local dirs */
1205 fid.f_seq = FID_SEQ_LOCAL_NAME;
1208 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1213 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1218 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1219 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1220 mdd_root_dir_name, S_IFDIR |
1221 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1223 CERROR("%s: create root fid failed: rc = %d\n",
1224 mdd2obd_dev(mdd)->obd_name, rc);
1228 mdd->mdd_root_fid = fid;
1229 rc = mdd_dot_lustre_setup(env, mdd);
1231 CERROR("%s: initializing .lustre failed: rc = %d\n",
1232 mdd2obd_dev(mdd)->obd_name, rc);
1236 /* Normal client usually send root access to MDT0 directly,
1237 * the root FID on non-MDT0 will only be used by echo client. */
1238 mdd->mdd_root_fid = fid;
1241 rc = mdd_orphan_index_init(env, mdd);
1245 rc = mdd_changelog_init(env, mdd);
1247 CERROR("%s: failed to initialize changelog: rc = %d\n",
1248 mdd2obd_dev(mdd)->obd_name, rc);
1252 rc = mdd_hsm_actions_llog_init(env, mdd);
1254 GOTO(out_changelog, rc);
1256 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1258 if (IS_ERR(nodemap_config)) {
1259 rc = PTR_ERR(nodemap_config);
1263 obt->obt_nodemap_config_file = nodemap_config;
1266 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1267 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1270 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1271 mdd2obd_dev(mdd)->obd_name, rc);
1272 GOTO(out_nodemap, rc);
1275 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1277 CERROR("%s: failed to register to barrier: rc = %d\n",
1278 mdd2obd_dev(mdd)->obd_name, rc);
1279 GOTO(out_lfsck, rc);
1285 lfsck_degister(env, mdd->mdd_bottom);
1287 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1288 obt->obt_nodemap_config_file = NULL;
1290 mdd_hsm_actions_llog_fini(env, mdd);
1292 mdd_changelog_fini(env, mdd);
1294 mdd_orphan_index_fini(env, mdd);
1296 if (mdd_seq_site(mdd)->ss_node_id == 0)
1297 mdd_dot_lustre_cleanup(env, mdd);
1299 local_oid_storage_fini(env, mdd->mdd_los);
1300 mdd->mdd_los = NULL;
1305 const struct lu_device_operations mdd_lu_ops = {
1306 .ldo_object_alloc = mdd_object_alloc,
1307 .ldo_process_config = mdd_process_config,
1308 .ldo_recovery_complete = mdd_recovery_complete,
1309 .ldo_prepare = mdd_prepare,
1312 static int mdd_root_get(const struct lu_env *env,
1313 struct md_device *m, struct lu_fid *f)
1315 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1318 *f = mdd->mdd_root_fid;
1323 * No permission check is needed.
1325 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1326 struct obd_statfs *sfs)
1328 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1333 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1335 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1340 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1341 struct md_device *m)
1343 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1345 return &mdd->mdd_dt_conf;
1348 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1351 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1353 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1354 return (*h == NULL ? -ENOENT : 0);
1357 static struct lu_device *mdd_device_free(const struct lu_env *env,
1358 struct lu_device *lu)
1360 struct mdd_device *m = lu2mdd_dev(lu);
1363 LASSERT(atomic_read(&lu->ld_ref) == 0);
1364 md_device_fini(&m->mdd_md_dev);
1369 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1370 struct lu_device_type *t,
1371 struct lustre_cfg *lcfg)
1373 struct lu_device *l;
1374 struct mdd_device *m;
1378 l = ERR_PTR(-ENOMEM);
1383 md_device_init(&m->mdd_md_dev, t);
1384 rc = mdd_init0(env, m, t, lcfg);
1386 mdd_device_free(env, l);
1395 * we use exports to track all mdd users
1397 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1398 struct obd_device *obd, struct obd_uuid *cluuid,
1399 struct obd_connect_data *data, void *localdata)
1401 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1402 struct lustre_handle conn;
1406 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1408 rc = class_connect(&conn, obd, cluuid);
1412 *exp = class_conn2export(&conn);
1414 /* Why should there ever be more than 1 connect? */
1415 LASSERT(mdd->mdd_connects == 0);
1416 mdd->mdd_connects++;
1422 * once last export (we don't count self-export) disappeared
1423 * mdd can be released
1425 static int mdd_obd_disconnect(struct obd_export *exp)
1427 struct obd_device *obd = exp->exp_obd;
1428 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1429 int rc, release = 0;
1432 mdd->mdd_connects--;
1433 if (mdd->mdd_connects == 0)
1436 rc = class_disconnect(exp);
1438 if (rc == 0 && release)
1439 class_manual_cleanup(obd);
1443 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1444 __u32 keylen, void *key, __u32 *vallen, void *val)
1448 if (KEY_IS(KEY_OSP_CONNECTED)) {
1449 struct obd_device *obd = exp->exp_obd;
1450 struct mdd_device *mdd;
1452 if (!obd->obd_set_up || obd->obd_stopping)
1455 mdd = lu2mdd_dev(obd->obd_lu_dev);
1457 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1465 static int mdd_obd_set_info_async(const struct lu_env *env,
1466 struct obd_export *exp,
1467 __u32 keylen, void *key,
1468 __u32 vallen, void *val,
1469 struct ptlrpc_request_set *set)
1471 struct obd_device *obd = exp->exp_obd;
1472 struct mdd_device *mdd;
1475 if (!obd->obd_set_up || obd->obd_stopping)
1478 mdd = lu2mdd_dev(obd->obd_lu_dev);
1480 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1485 static struct obd_ops mdd_obd_device_ops = {
1486 .o_owner = THIS_MODULE,
1487 .o_connect = mdd_obd_connect,
1488 .o_disconnect = mdd_obd_disconnect,
1489 .o_get_info = mdd_obd_get_info,
1490 .o_set_info_async = mdd_obd_set_info_async,
1493 static int mdd_changelog_user_register(const struct lu_env *env,
1494 struct mdd_device *mdd, int *id)
1496 struct llog_ctxt *ctxt;
1497 struct llog_changelog_user_rec *rec;
1501 ctxt = llog_get_context(mdd2obd_dev(mdd),
1502 LLOG_CHANGELOG_USER_ORIG_CTXT);
1508 llog_ctxt_put(ctxt);
1512 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1514 rec->cur_hdr.lrh_len = sizeof(*rec);
1515 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1516 spin_lock(&mdd->mdd_cl.mc_user_lock);
1517 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1518 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1519 CERROR("Maximum number of changelog users exceeded!\n");
1520 GOTO(out, rc = -EOVERFLOW);
1522 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1523 mdd->mdd_cl.mc_users++;
1524 rec->cur_endrec = mdd->mdd_cl.mc_index;
1526 rec->cur_time = (__u32)ktime_get_real_seconds();
1527 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1530 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1532 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1534 CWARN("%s: Failed to register changelog user %d: rc=%d\n",
1535 mdd2obd_dev(mdd)->obd_name, *id, rc);
1536 spin_lock(&mdd->mdd_cl.mc_user_lock);
1537 mdd->mdd_cl.mc_users--;
1538 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1542 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1544 /* Assume we want it on since somebody registered */
1545 rc = mdd_changelog_on(env, mdd);
1551 llog_ctxt_put(ctxt);
1555 struct mdd_changelog_user_purge {
1556 struct mdd_device *mcup_mdd;
1558 __u32 mcup_usercount;
1564 * changelog_user_purge callback
1566 * Is called once per user.
1568 * Check to see the user requested is available from rec.
1569 * Truncate the changelog.
1570 * Keep track of the total number of users (calls).
1572 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1573 struct llog_handle *llh,
1574 struct llog_rec_hdr *hdr, void *data)
1576 struct llog_changelog_user_rec *rec;
1577 struct mdd_changelog_user_purge *mcup = data;
1578 struct llog_cookie cookie;
1583 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1586 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1588 mcup->mcup_usercount++;
1590 if (rec->cur_id != mcup->mcup_id) {
1591 /* truncate to the lowest endrec that is not this user */
1592 mcup->mcup_minrec = min(mcup->mcup_minrec,
1597 /* Unregister this user */
1598 cookie.lgc_lgl = llh->lgh_id;
1599 cookie.lgc_index = hdr->lrh_index;
1601 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1604 mcup->mcup_found = true;
1605 mcup->mcup_usercount--;
1606 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1607 mcup->mcup_mdd->mdd_cl.mc_users--;
1608 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1614 int mdd_changelog_user_purge(const struct lu_env *env,
1615 struct mdd_device *mdd, __u32 id)
1617 struct mdd_changelog_user_purge mcup = {
1620 .mcup_found = false,
1621 .mcup_usercount = 0,
1622 .mcup_minrec = ULLONG_MAX,
1624 struct llog_ctxt *ctxt;
1629 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1630 mdd2obd_dev(mdd)->obd_name, id);
1632 ctxt = llog_get_context(mdd2obd_dev(mdd),
1633 LLOG_CHANGELOG_USER_ORIG_CTXT);
1635 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1636 GOTO(out, rc = -ENXIO);
1638 rc = llog_cat_process(env, ctxt->loc_handle,
1639 mdd_changelog_user_purge_cb, &mcup,
1642 if ((rc == 0) && (mcup.mcup_usercount == 0)) {
1643 spin_lock(&mdd->mdd_cl.mc_user_lock);
1644 if (mdd->mdd_cl.mc_users == 0) {
1645 /* No more users; turn changelogs off */
1646 CDEBUG(D_IOCTL, "turning off changelogs\n");
1647 rc = mdd_changelog_off(env, mdd);
1649 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1652 if ((rc == 0) && mcup.mcup_found) {
1653 CDEBUG(D_IOCTL, "%s: Purging changelog entries for user %d "
1655 mdd2obd_dev(mdd)->obd_name, id, mcup.mcup_minrec);
1656 /* Cancelling record 0 destroys the entire changelog, make sure
1657 we don't do that unless we mean it. */
1658 if (mcup.mcup_minrec != 0 || mcup.mcup_usercount == 0) {
1659 rc = mdd_changelog_llog_cancel(env, mdd,
1663 CWARN("%s: No changelog for user %u; rc=%d\n",
1664 mdd2obd_dev(mdd)->obd_name, id, rc);
1665 GOTO(out, rc = -ENOENT);
1668 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1673 llog_ctxt_put(ctxt);
1678 struct mdd_changelog_user_clear {
1686 * changelog_clear callback
1688 * Is called once per user.
1690 * Check to see the user requested is available from rec.
1691 * Check the oldest (smallest) record for boundary conditions.
1692 * Truncate the changelog.
1694 static int mdd_changelog_clear_cb(const struct lu_env *env,
1695 struct llog_handle *llh,
1696 struct llog_rec_hdr *hdr,
1699 struct llog_changelog_user_rec *rec;
1700 struct mdd_changelog_user_clear *mcuc = data;
1705 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1708 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1710 /* Does the changelog id match the requested id? */
1711 if (rec->cur_id != mcuc->mcuc_id) {
1712 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec,
1717 /* cur_endrec is the oldest purgeable record, make sure we're newer */
1718 if (rec->cur_endrec > mcuc->mcuc_endrec) {
1719 CDEBUG(D_IOCTL, "Request %llu out of range: %llu\n",
1720 mcuc->mcuc_endrec, rec->cur_endrec);
1724 /* Flag that we've met all the range and user checks.
1725 * We now know the record to flush.
1727 rec->cur_endrec = mcuc->mcuc_endrec;
1729 rec->cur_time = (__u32)ktime_get_real_seconds();
1730 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1733 mcuc->mcuc_flush = true;
1735 CDEBUG(D_IOCTL, "Rewriting changelog user %u endrec to %llu\n",
1736 mcuc->mcuc_id, rec->cur_endrec);
1738 /* Update the endrec */
1739 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1745 * Clear a changelog up to entry specified by endrec for user id.
1747 static int mdd_changelog_clear(const struct lu_env *env,
1748 struct mdd_device *mdd, __u32 id,
1751 struct mdd_changelog_user_clear mcuc = {
1753 .mcuc_minrec = endrec,
1754 .mcuc_flush = false,
1756 struct llog_ctxt *ctxt;
1762 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
1763 mdd2obd_dev(mdd)->obd_name, id, endrec);
1764 /* start_rec is the newest (largest value) entry in the changelogs*/
1765 spin_lock(&mdd->mdd_cl.mc_lock);
1766 start_rec = mdd->mdd_cl.mc_index;
1767 spin_unlock(&mdd->mdd_cl.mc_lock);
1769 if (start_rec < endrec) {
1770 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
1771 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
1776 mcuc.mcuc_endrec = start_rec;
1777 mcuc.mcuc_minrec = ULLONG_MAX;
1779 mcuc.mcuc_endrec = endrec;
1782 ctxt = llog_get_context(mdd2obd_dev(mdd),
1783 LLOG_CHANGELOG_USER_ORIG_CTXT);
1785 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1786 GOTO(out, rc = -ENXIO);
1788 rc = llog_cat_process(env, ctxt->loc_handle,
1789 mdd_changelog_clear_cb, (void *)&mcuc,
1793 CWARN("%s: Failure to clear the changelog for user %d: %d\n",
1794 mdd2obd_dev(mdd)->obd_name, id, rc);
1795 } else if (mcuc.mcuc_flush) {
1796 /* Cancelling record 0 destroys the entire changelog, make sure
1797 we don't do that unless we mean it. */
1798 if (mcuc.mcuc_minrec != 0) {
1799 CDEBUG(D_IOCTL, "%s: Purging changelog entries up "\
1800 "to %llu\n", mdd2obd_dev(mdd)->obd_name,
1803 rc = mdd_changelog_llog_cancel(env, mdd,
1807 CWARN("%s: No entry for user %d\n",
1808 mdd2obd_dev(mdd)->obd_name, id);
1815 llog_ctxt_put(ctxt);
1821 * May be called remotely from mdt_iocontrol_handle or locally from
1822 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1823 * an obd_ioctl_data format (but local ioctl handler does).
1825 * \param len - data len
1826 * \param karg - ioctl data, in kernel space
1828 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1829 unsigned int cmd, int len, void *karg)
1831 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1832 struct obd_ioctl_data *data = karg;
1836 /* Doesn't use obd_ioctl_data */
1838 case OBD_IOC_CHANGELOG_CLEAR: {
1839 struct changelog_setinfo *cs = karg;
1841 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1842 RETURN(-EINPROGRESS);
1844 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
1846 barrier_exit(mdd->mdd_bottom);
1849 case OBD_IOC_START_LFSCK: {
1850 rc = lfsck_start(env, mdd->mdd_bottom,
1851 (struct lfsck_start_param *)karg);
1854 case OBD_IOC_STOP_LFSCK: {
1855 rc = lfsck_stop(env, mdd->mdd_bottom,
1856 (struct lfsck_stop *)karg);
1859 case OBD_IOC_QUERY_LFSCK: {
1860 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
1861 (struct lfsck_query *)karg);
1866 /* Below ioctls use obd_ioctl_data */
1867 if (len != sizeof(*data)) {
1868 CERROR("Bad ioctl size %d\n", len);
1871 if (data->ioc_version != OBD_IOCTL_VERSION) {
1872 CERROR("Bad magic %x != %x\n", data->ioc_version,
1878 case OBD_IOC_CHANGELOG_REG:
1879 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1880 RETURN(-EINPROGRESS);
1882 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1883 barrier_exit(mdd->mdd_bottom);
1885 case OBD_IOC_CHANGELOG_DEREG:
1886 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1887 RETURN(-EINPROGRESS);
1889 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1);
1890 barrier_exit(mdd->mdd_bottom);
1899 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1900 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
1902 static const struct md_device_operations mdd_ops = {
1903 .mdo_statfs = mdd_statfs,
1904 .mdo_root_get = mdd_root_get,
1905 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1906 .mdo_iocontrol = mdd_iocontrol,
1907 .mdo_dtconf_get = mdd_dtconf_get,
1910 static struct lu_device_type_operations mdd_device_type_ops = {
1911 .ldto_init = mdd_type_init,
1912 .ldto_fini = mdd_type_fini,
1914 .ldto_start = mdd_type_start,
1915 .ldto_stop = mdd_type_stop,
1917 .ldto_device_alloc = mdd_device_alloc,
1918 .ldto_device_free = mdd_device_free,
1920 .ldto_device_fini = mdd_device_fini
1923 static struct lu_device_type mdd_device_type = {
1924 .ldt_tags = LU_DEVICE_MD,
1925 .ldt_name = LUSTRE_MDD_NAME,
1926 .ldt_ops = &mdd_device_type_ops,
1927 .ldt_ctx_tags = LCT_MD_THREAD
1930 /* context key constructor: mdd_key_init */
1931 LU_KEY_INIT(mdd, struct mdd_thread_info);
1933 static void mdd_key_fini(const struct lu_context *ctx,
1934 struct lu_context_key *key, void *data)
1936 struct mdd_thread_info *info = data;
1938 lu_buf_free(&info->mti_big_buf);
1939 lu_buf_free(&info->mti_link_buf);
1940 lu_buf_free(&info->mti_xattr_buf);
1945 /* context key: mdd_thread_key */
1946 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1948 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
1949 int (*func)(void *), void *data, char *name)
1951 struct task_struct *task;
1953 LASSERT(thread->mgt_init == false);
1954 init_completion(&thread->mgt_started);
1955 init_completion(&thread->mgt_finished);
1956 thread->mgt_data = data;
1957 thread->mgt_abort = false;
1958 thread->mgt_init = true;
1960 task = kthread_run(func, thread, name);
1962 complete(&thread->mgt_finished);
1963 return PTR_ERR(task);
1965 wait_for_completion(&thread->mgt_started);
1969 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
1971 if (thread->mgt_init == true) {
1972 thread->mgt_abort = true;
1973 wait_for_completion(&thread->mgt_finished);
1977 static int __init mdd_init(void)
1981 rc = lu_kmem_init(mdd_caches);
1985 rc = class_register_type(&mdd_obd_device_ops, NULL, true, NULL,
1986 LUSTRE_MDD_NAME, &mdd_device_type);
1988 lu_kmem_fini(mdd_caches);
1992 static void __exit mdd_exit(void)
1994 class_unregister_type(LUSTRE_MDD_NAME);
1995 lu_kmem_fini(mdd_caches);
1998 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1999 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2000 MODULE_VERSION(LUSTRE_VERSION_STRING);
2001 MODULE_LICENSE("GPL");
2003 module_init(mdd_init);
2004 module_exit(mdd_exit);