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/
31 * lustre/mdd/mdd_device.c
33 * Lustre Metadata Server (mdd) routines
35 * Author: Wang Di <wangdi@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <linux/module.h>
41 #include <linux/kthread.h>
42 #include <obd_class.h>
43 #include <uapi/linux/lustre/lustre_ioctl.h>
44 #include <lustre_mds.h>
45 #include <obd_support.h>
46 #include <lu_object.h>
47 #include <uapi/linux/lustre/lustre_param.h>
48 #include <lustre_fid.h>
49 #include <lustre_nodemap.h>
50 #include <lustre_barrier.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)
125 /* LU-8040 Set defaults here, before values configs */
126 mdd->mdd_cl.mc_flags = 0; /* off by default */
127 /* per-server mask is set via parameters if needed */
128 mdd->mdd_cl.mc_proc_mask = CHANGELOG_MINMASK;
129 /* current mask is calculated from mask above and users masks */
130 mdd->mdd_cl.mc_current_mask = CHANGELOG_MINMASK;
131 mdd->mdd_cl.mc_deniednext = 60; /* 60 secs by default */
133 dev = lustre_cfg_string(lcfg, 0);
137 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
138 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
140 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
141 mdd->mdd_md_dev.md_ops = &mdd_ops;
143 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
147 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
148 /* sync permission changes */
149 mdd->mdd_sync_permission = 1;
150 /* enable changelog garbage collection */
151 mdd->mdd_changelog_gc = 1;
152 /* enable changelog cleanup due to lack of space */
153 mdd->mdd_changelog_free_space_gc = true;
154 /* set when emergency GC is started */
155 mdd->mdd_changelog_emrg_gc = false;
156 /* with a significant amount of idle time */
157 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
158 /* or a significant amount of late indexes */
159 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
160 /* with a reasonable interval between each check */
161 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
162 /* with a very few number of free catalog entries */
163 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
164 /* special default striping for files created with O_APPEND */
165 mdd->mdd_append_stripe_count = 1;
166 mdd->mdd_append_pool[0] = '\0';
168 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
170 /* we are using service name but not mdd obd name
171 * for compatibility reasons.
172 * It is passed from MDT in lustre_cfg[2] buffer */
173 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
175 obd_disconnect(mdd->mdd_child_exp);
180 static struct lu_device *mdd_device_fini(const struct lu_env *env,
183 struct mdd_device *mdd = lu2mdd_dev(d);
186 lu_dev_del_linkage(d->ld_site, d);
188 mdd_procfs_fini(mdd);
192 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
193 struct llog_rec_hdr *hdr, void *data)
195 struct mdd_device *mdd = (struct mdd_device *)data;
196 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
198 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
199 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
202 "seeing record at index %d/%d/%llu t=%x %.*s in log"
203 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
204 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
205 changelog_rec_name(&rec->cr), PLOGID(&llh->lgh_id));
206 spin_lock(&mdd->mdd_cl.mc_lock);
207 mdd->mdd_cl.mc_index = rec->cr.cr_index;
208 spin_unlock(&mdd->mdd_cl.mc_lock);
209 return LLOG_PROC_BREAK;
212 char *mdd_chlg_username(struct llog_changelog_user_rec2 *rec, char *buf,
215 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
217 snprintf(buf, len, "%s%u-%s", CHANGELOG_USER_PREFIX,
218 rec->cur_id, rec->cur_name);
220 snprintf(buf, len, "%s%u", CHANGELOG_USER_PREFIX, rec->cur_id);
224 __u32 mdd_chlg_usermask(struct llog_changelog_user_rec2 *rec)
226 return rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 ?
230 static int changelog_user_init_cb(const struct lu_env *env,
231 struct llog_handle *llh,
232 struct llog_rec_hdr *hdr, void *data)
234 struct mdd_device *mdd = data;
235 struct llog_changelog_user_rec2 *rec;
236 char user_name[CHANGELOG_USER_NAMELEN_FULL];
238 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
240 rec = container_of(hdr, typeof(*rec), cur_hdr);
241 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
242 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
243 CWARN("%s: unknown user type %x at index %u in log "DFID"\n",
244 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
245 rec->cur_hdr.lrh_type, PLOGID(&llh->lgh_id));
250 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
251 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
253 hdr->lrh_index, rec->cur_hdr.lrh_index, rec->cur_endrec,
254 PLOGID(&llh->lgh_id));
256 spin_lock(&mdd->mdd_cl.mc_user_lock);
257 mdd->mdd_cl.mc_lastuser = rec->cur_id;
258 mdd->mdd_cl.mc_users++;
259 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
260 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
261 else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
262 mdd->mdd_cl.mc_current_mask |= CHANGELOG_DEFMASK;
263 mdd->mdd_cl.mc_mintime = min(mdd->mdd_cl.mc_mintime, rec->cur_time);
264 mdd->mdd_cl.mc_minrec = min(mdd->mdd_cl.mc_minrec, rec->cur_endrec);
265 spin_unlock(&mdd->mdd_cl.mc_user_lock);
266 spin_lock(&mdd->mdd_cl.mc_lock);
267 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
268 mdd->mdd_cl.mc_index = rec->cur_endrec;
269 spin_unlock(&mdd->mdd_cl.mc_lock);
271 return LLOG_PROC_BREAK;
274 struct changelog_orphan_data {
276 struct mdd_device *clod_mdd;
279 /* find oldest changelog record index */
280 static int changelog_detect_orphan_cb(const struct lu_env *env,
281 struct llog_handle *llh,
282 struct llog_rec_hdr *hdr, void *data)
284 struct changelog_orphan_data *clod = data;
285 struct mdd_device *mdd = clod->clod_mdd;
286 struct llog_changelog_rec *rec;
288 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
290 rec = container_of(hdr, typeof(*rec), cr_hdr);
291 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
292 CWARN("%s: invalid record at index %d in log "DFID"\n",
293 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
294 PLOGID(&llh->lgh_id));
295 /* try to find some next valid record and thus allow to recover
296 * from a corrupted LLOG, instead to assert and force a crash
302 "%s: record at index %d/%d/%llu t=%x %.*s in log "DFID"\n",
303 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
304 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
305 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
306 PLOGID(&llh->lgh_id));
308 clod->clod_index = rec->cr.cr_index;
310 return LLOG_PROC_BREAK;
313 /* find oldest changelog user index */
314 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
315 struct llog_handle *llh,
316 struct llog_rec_hdr *hdr, void *data)
318 struct changelog_orphan_data *clod = data;
319 struct mdd_device *mdd = clod->clod_mdd;
320 struct llog_changelog_user_rec2 *rec;
321 char user_name[CHANGELOG_USER_NAMELEN_FULL];
323 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
325 rec = container_of(hdr, typeof(*rec), cur_hdr);
326 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
327 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
328 CWARN("%s: unknown user type %u at index %u in log "DFID"\n",
329 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
330 rec->cur_hdr.lrh_type, PLOGID(&llh->lgh_id));
331 /* try to find some next valid record and thus allow to recover
332 * from a corrupted LLOG, instead to assert and force a crash
337 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
338 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
340 hdr->lrh_index, rec->cur_hdr.lrh_index,
341 rec->cur_endrec, PLOGID(&llh->lgh_id));
343 clod->clod_index = min_t(__u64, clod->clod_index, rec->cur_endrec);
348 struct changelog_cancel_cookie {
350 struct mdd_device *mdd;
353 static int llog_changelog_cancel_cb(const struct lu_env *env,
354 struct llog_handle *llh,
355 struct llog_rec_hdr *hdr, void *data)
357 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
358 struct changelog_cancel_cookie *cl_cookie =
359 (struct changelog_cancel_cookie *)data;
363 /* This is always a (sub)log, not the catalog */
364 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
366 /* if current context is GC-thread allow it to stop upon umount
367 * remaining records cleanup will occur upon next mount
369 * also during testing, wait for GC-thread to be released
371 * XXX this requires the GC-thread to not fork a sub-thread via
372 * llog[_cat]_process_or_fork() and we may think to also implement
373 * this shutdown mechanism for manually started user unregister which
374 * can also take a long time if huge backlog of records
376 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
377 /* wait to be released */
378 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
381 if (kthread_should_stop())
382 RETURN(LLOG_PROC_BREAK);
385 if (rec->cr.cr_index > cl_cookie->endrec)
386 /* records are in order, so we're done */
387 RETURN(LLOG_PROC_BREAK);
389 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
390 if (cfs_fail_val == 0)
391 cfs_fail_val = hdr->lrh_index;
392 if (cfs_fail_val == hdr->lrh_index)
393 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
396 /* Records folow one by one, cr_index++. We could calculate the
397 * last cr_index at this plain llog. And if it less then cookie endrec
398 * cancel the whole file.
400 if (llog_is_plain_skipable(llh->lgh_hdr, hdr, rec->cr.cr_index,
401 cl_cookie->endrec)) {
404 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
405 if (cfs_fail_val == 0) {
406 cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
407 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
410 rc = llog_destroy(env, llh);
412 CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
413 PLOGID(&llh->lgh_id));
414 RETURN(LLOG_DEL_PLAIN);
418 /* cancel them one at a time. I suppose we could store up the cookies
419 * and cancel them all at once; probably more efficient, but this is
420 * done as a user call, so who cares... */
422 RETURN(LLOG_DEL_RECORD);
425 static int llog_changelog_cancel(const struct lu_env *env,
426 struct llog_ctxt *ctxt,
427 struct changelog_cancel_cookie *cookie)
429 struct llog_handle *cathandle = ctxt->loc_handle;
434 /* This should only be called with the catalog handle */
435 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
437 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
440 /* 0 or 1 means we're done */
443 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
444 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
445 PLOGID(&cathandle->lgh_id), rc);
450 static struct llog_operations changelog_orig_logops;
453 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
457 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
461 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
464 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
465 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
466 CERROR("Changelogs cannot be enabled due to error "
467 "condition (see %s log).\n",
468 mdd2obd_dev(mdd)->obd_name);
471 spin_lock(&mdd->mdd_cl.mc_lock);
472 mdd->mdd_cl.mc_flags |= CLM_ON;
473 spin_unlock(&mdd->mdd_cl.mc_lock);
474 rc = mdd_changelog_write_header(env, mdd, CLM_START);
480 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
484 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
487 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
488 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
489 spin_lock(&mdd->mdd_cl.mc_lock);
490 mdd->mdd_cl.mc_flags &= ~CLM_ON;
491 spin_unlock(&mdd->mdd_cl.mc_lock);
496 static int mdd_changelog_llog_init(const struct lu_env *env,
497 struct mdd_device *mdd)
499 struct obd_device *obd = mdd2obd_dev(mdd);
500 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
501 struct changelog_orphan_data clod = {
512 /* LU-2844 mdd setup failure should not cause umount oops */
513 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
516 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
517 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
518 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
519 obd, &changelog_orig_logops);
521 CERROR("%s: changelog llog setup failed: rc = %d\n",
526 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
529 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
532 GOTO(out_cleanup, rc);
534 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
538 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
539 changelog_init_cb, mdd);
542 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
547 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
548 mdd->mdd_cl.mc_index);
550 /* setup user changelog */
551 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
552 obd, &changelog_orig_logops);
554 CERROR("%s: changelog users llog setup failed: rc = %d\n",
559 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
562 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
565 GOTO(out_ucleanup, rc);
567 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
569 GOTO(out_uclose, rc);
571 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
572 changelog_user_init_cb, mdd);
574 CERROR("%s: changelog user init failed: rc = %d\n",
576 GOTO(out_uclose, rc);
579 /* Finally apply per-server mask */
580 mdd->mdd_cl.mc_current_mask |= mdd->mdd_cl.mc_proc_mask;
582 /* If we have registered users, assume we want changelogs on */
583 if (mdd->mdd_cl.mc_lastuser > 0) {
584 rc = mdd_changelog_on(env, mdd);
586 GOTO(out_uclose, rc);
589 /* find and clear any orphan changelog records (1st record index <
590 * smallest of all users current index), likely to come from an
591 * interrupted manual or GC-thread purge, as its user record had
593 * XXX we may wait for a still registered user clear operation to
594 * do the job, but it may then take a long time to reach the user's
595 * real targetted records if a huge purge backlog is still to be
596 * processed as a long time idle user record could have been deleted
597 * XXX we may need to run end of purge as a separate thread
599 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
602 CERROR("%s: changelog detect orphan failed: rc = %d\n",
604 GOTO(out_uclose, rc);
606 rc = llog_cat_process(env, uctxt->loc_handle,
607 changelog_user_detect_orphan_cb,
610 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
612 GOTO(out_uclose, rc);
614 if (unlikely(clod.clod_index < user_orphan.clod_index)) {
615 struct changelog_cancel_cookie cl_cookie = {
616 .endrec = user_orphan.clod_index,
620 CWARN("%s : orphan changelog records found, starting from "
621 "index %llu to index %llu, being cleared now\n",
622 obd->obd_name, clod.clod_index, user_orphan.clod_index);
624 /* XXX we may need to run end of purge as a separate thread */
625 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
627 CERROR("%s: purge of changelog orphan records failed: "
628 "rc = %d\n", obd->obd_name, rc);
629 GOTO(out_uclose, rc);
634 llog_ctxt_put(uctxt);
637 llog_cat_close(env, uctxt->loc_handle);
639 llog_cleanup(env, uctxt);
641 llog_cat_close(env, ctxt->loc_handle);
643 llog_cleanup(env, ctxt);
647 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
649 struct obd_device *obd = mdd2obd_dev(mdd);
652 mdd->mdd_cl.mc_index = 0;
653 spin_lock_init(&mdd->mdd_cl.mc_lock);
654 mdd->mdd_cl.mc_starttime = ktime_get();
655 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
656 mdd->mdd_cl.mc_lastuser = 0;
658 /* ensure a GC check will, and a thread run may, occur upon start */
659 mdd->mdd_cl.mc_gc_time = 0;
660 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
661 mdd->mdd_cl.mc_mintime = (__u32)ktime_get_real_seconds();
662 mdd->mdd_cl.mc_minrec = ULLONG_MAX;
664 rc = mdd_changelog_llog_init(env, mdd);
666 CERROR("%s: changelog setup during init failed: rc = %d\n",
668 mdd->mdd_cl.mc_flags |= CLM_ERR;
674 static void mdd_changelog_fini(const struct lu_env *env,
675 struct mdd_device *mdd)
677 struct obd_device *obd = mdd2obd_dev(mdd);
678 struct llog_ctxt *ctxt;
680 if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
682 mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
685 /* stop GC-thread if running */
686 spin_lock(&mdd->mdd_cl.mc_lock);
687 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
688 /* avoid any attempt to run a GC-thread */
689 mdd->mdd_cl.mc_gc_task = current;
690 spin_unlock(&mdd->mdd_cl.mc_lock);
692 struct task_struct *gc_task;
694 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
695 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
696 /* need to wait for birthing GC-thread to be started
697 * and to have set mc_gc_task to itself
699 spin_unlock(&mdd->mdd_cl.mc_lock);
700 /* Add a tiny sleep */
701 schedule_timeout_uninterruptible(1);
702 /* go back to fully check if GC-thread has started or
703 * even already exited or if a new one is starting...
707 /* take a reference on task_struct to avoid it to be freed
710 gc_task = mdd->mdd_cl.mc_gc_task;
711 get_task_struct(gc_task);
712 spin_unlock(&mdd->mdd_cl.mc_lock);
713 kthread_stop(gc_task);
714 put_task_struct(gc_task);
717 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
719 llog_cat_close(env, ctxt->loc_handle);
720 llog_cleanup(env, ctxt);
722 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
724 llog_cat_close(env, ctxt->loc_handle);
725 llog_cleanup(env, ctxt);
730 * Remove entries with indicies up to and including \a endrec from the
737 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
738 unsigned long long endrec)
740 struct obd_device *obd = mdd2obd_dev(mdd);
741 struct llog_ctxt *ctxt;
742 unsigned long long cur;
743 struct changelog_cancel_cookie cookie;
746 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
750 spin_lock(&mdd->mdd_cl.mc_lock);
751 cur = (long long)mdd->mdd_cl.mc_index;
752 spin_unlock(&mdd->mdd_cl.mc_lock);
755 * If purging all records, write a header entry so we don't have an
756 * empty catalog and we're sure to have a valid starting index next
757 * time. In a case of crash, we just restart with old log so we're
761 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
768 * Some records were purged, so reset repeat-access time (so we
769 * record new mtime update records, so users can see a file has been
770 * changed since the last purge)
772 mdd->mdd_cl.mc_starttime = ktime_get();
774 cookie.endrec = endrec;
776 rc = llog_changelog_cancel(env, ctxt, &cookie);
782 /** Add a CL_MARK record to the changelog
784 * \param markerflags - CLM_*
787 int mdd_changelog_write_header(const struct lu_env *env,
788 struct mdd_device *mdd, int markerflags)
790 struct obd_device *obd = mdd2obd_dev(mdd);
791 struct llog_changelog_rec *rec;
793 struct llog_ctxt *ctxt;
795 int len = strlen(obd->obd_name);
800 if (mdd->mdd_cl.mc_current_mask & BIT(CL_MARK)) {
801 mdd->mdd_cl.mc_starttime = ktime_get();
805 reclen = llog_data_len(sizeof(*rec) + len);
806 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
807 if (buf->lb_buf == NULL)
811 rec->cr.cr_flags = CLF_VERSION;
812 rec->cr.cr_type = CL_MARK;
813 rec->cr.cr_namelen = len;
814 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
815 /* Status and action flags */
816 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
817 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
819 rec->cr_hdr.lrh_type = CHANGELOG_REC;
820 rec->cr.cr_time = cl_time();
822 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
825 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
830 /* assume on or off event; reset repeat-access time */
831 mdd->mdd_cl.mc_starttime = ktime_get();
836 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
837 * are valid. We also check if object with passed fid exists or not.
839 static int obf_lookup(const struct lu_env *env, struct md_object *p,
840 const struct lu_name *lname, struct lu_fid *f,
841 struct md_op_spec *spec)
843 char *name = (char *)lname->ln_name;
844 struct mdd_device *mdd = mdo2mdd(p);
845 struct mdd_object *child;
851 sscanf(name, SFID, RFID(f));
853 GOTO(out, rc = -ENOENT);
855 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
856 !fid_seq_is_dot(f->f_seq))
857 GOTO(out, rc = -ENOENT);
859 /* Check if object with this fid exists */
860 child = mdd_object_find(env, mdd, f);
862 GOTO(out, rc = PTR_ERR(child));
864 if (mdd_object_exists(child) == 0)
867 mdd_object_put(env, child);
873 static int mdd_dummy_create(const struct lu_env *env,
874 struct md_object *pobj,
875 const struct lu_name *lname,
876 struct md_object *child,
877 struct md_op_spec *spec,
883 static int mdd_dummy_rename(const struct lu_env *env,
884 struct md_object *src_pobj,
885 struct md_object *tgt_pobj,
886 const struct lu_fid *lf,
887 const struct lu_name *lsname,
888 struct md_object *tobj,
889 const struct lu_name *ltname,
895 static int mdd_dummy_link(const struct lu_env *env,
896 struct md_object *tgt_obj,
897 struct md_object *src_obj,
898 const struct lu_name *lname,
904 static int mdd_dummy_unlink(const struct lu_env *env,
905 struct md_object *pobj,
906 struct md_object *cobj,
907 const struct lu_name *lname,
914 int mdd_create(const struct lu_env *env, struct md_object *pobj,
915 const struct lu_name *lname, struct md_object *child,
916 struct md_op_spec *spec, struct md_attr *ma);
917 static int mdd_obf_create(const struct lu_env *env, struct md_object *pobj,
918 const struct lu_name *lname, struct md_object *child,
919 struct md_op_spec *spec, struct md_attr *ma)
921 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
922 return mdd_create(env, pobj, lname, child, spec, ma);
926 static const struct md_dir_operations mdd_obf_dir_ops = {
927 .mdo_lookup = obf_lookup,
928 .mdo_create = mdd_obf_create,
929 .mdo_rename = mdd_dummy_rename,
930 .mdo_link = mdd_dummy_link,
931 .mdo_unlink = mdd_dummy_unlink
934 static const struct md_dir_operations mdd_lpf_dir_ops = {
935 .mdo_lookup = mdd_lookup,
936 .mdo_create = mdd_dummy_create,
937 .mdo_rename = mdd_dummy_rename,
938 .mdo_link = mdd_dummy_link,
939 .mdo_unlink = mdd_dummy_unlink
942 static struct md_object *mdo_locate(const struct lu_env *env,
943 struct md_device *md,
944 const struct lu_fid *fid)
946 struct lu_object *obj;
947 struct md_object *mdo;
949 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
951 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
952 LASSERT(obj != NULL);
960 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
962 struct md_object *mdo;
963 struct mdd_object *mdd_lpf;
964 struct lu_fid fid = LU_LPF_FID;
968 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
969 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
974 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
976 RETURN(PTR_ERR(mdo));
978 LASSERT(lu_object_exists(&mdo->mo_lu));
980 mdd_lpf = md2mdd_obj(mdo);
981 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
982 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
988 * Create special in-memory "fid" object for open-by-fid.
990 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
992 struct md_object *mdo;
993 struct mdd_object *mdd_obf;
994 struct lu_fid fid = LU_OBF_FID;
997 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
998 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
1002 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1004 RETURN(PTR_ERR(mdo));
1006 LASSERT(lu_object_exists(&mdo->mo_lu));
1008 mdd_obf = md2mdd_obj(mdo);
1009 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1010 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
1015 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
1016 struct mdd_device *m)
1018 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
1019 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
1020 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
1022 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
1023 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
1024 m->mdd_dot_lustre_objs.mdd_obf = NULL;
1026 if (m->mdd_dot_lustre != NULL) {
1027 mdd_object_put(env, m->mdd_dot_lustre);
1028 m->mdd_dot_lustre = NULL;
1032 /** Setup ".lustre" directory object */
1033 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1035 struct md_object *mdo;
1040 /* Create ".lustre" directory in ROOT. */
1041 fid = LU_DOT_LUSTRE_FID;
1042 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
1044 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
1048 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1050 RETURN(PTR_ERR(mdo));
1051 LASSERT(lu_object_exists(&mdo->mo_lu));
1053 m->mdd_dot_lustre = md2mdd_obj(mdo);
1055 rc = mdd_obf_setup(env, m);
1057 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
1058 mdd2obd_dev(m)->obd_name, rc);
1062 rc = mdd_lpf_setup(env, m);
1064 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
1065 mdd2obd_dev(m)->obd_name, rc);
1072 mdd_dot_lustre_cleanup(env, m);
1078 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
1079 * object in obd_device
1081 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
1082 struct mdd_device *m)
1084 struct obd_device *obd = mdd2obd_dev(m);
1085 struct llog_ctxt *ctxt = NULL;
1089 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1090 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1092 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1093 obd, &llog_common_cat_ops);
1095 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1100 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1103 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1106 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1108 GOTO(out_cleanup, rc);
1111 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1113 GOTO(out_close, rc);
1115 llog_ctxt_put(ctxt);
1119 llog_cat_close(env, ctxt->loc_handle);
1120 ctxt->loc_handle = NULL;
1122 llog_cleanup(env, ctxt);
1128 * cleanup the context created by llog_setup_named()
1130 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1131 struct mdd_device *m)
1133 struct obd_device *obd = mdd2obd_dev(m);
1134 struct llog_ctxt *lctxt;
1137 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1139 llog_cat_close(env, lctxt->loc_handle);
1140 lctxt->loc_handle = NULL;
1141 llog_cleanup(env, lctxt);
1147 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1148 struct lustre_cfg *cfg)
1150 barrier_deregister(m->mdd_bottom);
1151 lfsck_degister(env, m->mdd_bottom);
1152 mdd_hsm_actions_llog_fini(env, m);
1153 mdd_changelog_fini(env, m);
1154 mdd_orphan_index_fini(env, m);
1155 mdd_dot_lustre_cleanup(env, m);
1156 if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) {
1157 nm_config_file_deregister_tgt(env,
1158 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
1159 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL;
1161 if (m->mdd_los != NULL) {
1162 local_oid_storage_fini(env, m->mdd_los);
1165 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1167 if (m->mdd_child_exp)
1168 obd_disconnect(m->mdd_child_exp);
1171 static int mdd_process_config(const struct lu_env *env,
1172 struct lu_device *d, struct lustre_cfg *cfg)
1174 struct mdd_device *m = lu2mdd_dev(d);
1175 struct dt_device *dt = m->mdd_child;
1176 struct lu_device *next = &dt->dd_lu_dev;
1180 switch (cfg->lcfg_command) {
1184 count = class_modify_config(cfg, PARAM_MDD, &m->mdd_kobj);
1185 rc = count > 0 ? 0 : count;
1187 /* we don't understand; pass it on */
1188 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1192 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1195 dt_conf_get(env, dt, &m->mdd_dt_conf);
1197 case LCFG_PRE_CLEANUP:
1198 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1199 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1202 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1203 lu_dev_del_linkage(d->ld_site, d);
1204 mdd_device_shutdown(env, m, cfg);
1207 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1214 static int mdd_recovery_complete(const struct lu_env *env,
1215 struct lu_device *d)
1217 struct mdd_device *mdd = lu2mdd_dev(d);
1218 struct lu_device *next;
1222 LASSERT(mdd != NULL);
1223 next = &mdd->mdd_child->dd_lu_dev;
1225 if (!mdd->mdd_bottom->dd_rdonly)
1226 mdd_orphan_cleanup(env, mdd);
1227 rc = next->ld_ops->ldo_recovery_complete(env, next);
1232 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1233 const struct lu_fid *pfid, const char *name,
1234 __u32 mode, struct lu_fid *fid)
1236 struct dt_object *parent, *dto;
1241 LASSERT(!fid_is_zero(pfid));
1242 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1243 if (unlikely(IS_ERR(parent)))
1244 RETURN(PTR_ERR(parent));
1246 /* create local file/dir, if @fid is passed then try to use it */
1247 if (fid_is_zero(fid))
1248 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1251 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1255 GOTO(out_put, rc = PTR_ERR(dto));
1256 *fid = *lu_object_fid(&dto->do_lu);
1257 /* since stack is not fully set up the local_storage uses own stack
1258 * and we should drop its object from cache */
1259 dt_object_put_nocache(env, dto);
1262 dt_object_put(env, parent);
1266 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1267 enum lfsck_events event)
1272 static int mdd_prepare(const struct lu_env *env,
1273 struct lu_device *pdev,
1274 struct lu_device *cdev)
1276 struct mdd_device *mdd = lu2mdd_dev(cdev);
1277 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1278 struct nm_config_file *nodemap_config;
1279 struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
1280 struct dt_object *root = NULL;
1281 struct thandle *th = NULL;
1287 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1291 /* Setup local dirs */
1292 fid.f_seq = FID_SEQ_LOCAL_NAME;
1295 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1300 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1305 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1306 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1307 mdd_root_dir_name, S_IFDIR |
1308 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1310 CERROR("%s: create root fid failed: rc = %d\n",
1311 mdd2obd_dev(mdd)->obd_name, rc);
1315 /* store a default directory layout on the root directory if
1316 * it doesn't already exist to improve MDT space balance.
1318 root = dt_locate(env, mdd->mdd_bottom, &fid);
1319 if (unlikely(IS_ERR(root)))
1320 GOTO(out_los, rc = PTR_ERR(root));
1322 rc = dt_xattr_get(env, root, &LU_BUF_NULL,
1323 XATTR_NAME_DEFAULT_LMV);
1324 if (rc == -ENODATA) {
1326 struct lmv_user_md lmv_default = {
1327 .lum_magic = LMV_USER_MAGIC,
1328 .lum_stripe_count = 1,
1329 .lum_stripe_offset = LMV_OFFSET_DEFAULT,
1330 .lum_max_inherit = LMV_INHERIT_UNLIMITED,
1331 .lum_max_inherit_rr = LMV_INHERIT_RR_ROOT,
1334 th = dt_trans_create(env, mdd->mdd_bottom);
1336 GOTO(out_root_put, rc = PTR_ERR(th));
1338 buf.lb_buf = &lmv_default;
1339 buf.lb_len = sizeof(lmv_default);
1340 rc = dt_declare_xattr_set(env, root, &buf,
1341 XATTR_NAME_DEFAULT_LMV, 0,
1344 GOTO(out_trans_stop, rc);
1346 rc = dt_trans_start_local(env, mdd->mdd_bottom, th);
1348 GOTO(out_trans_stop, rc);
1350 rc = dt_xattr_set(env, root, &buf,
1351 XATTR_NAME_DEFAULT_LMV, 0, th);
1353 GOTO(out_trans_stop, rc);
1355 dt_trans_stop(env, mdd->mdd_bottom, th);
1357 } else if (rc < 0 && rc != -ERANGE) {
1358 CERROR("%s: get default LMV of root failed: rc = %d\n",
1359 mdd2obd_dev(mdd)->obd_name, rc);
1361 GOTO(out_root_put, rc);
1364 dt_object_put(env, root);
1367 mdd->mdd_root_fid = fid;
1368 rc = mdd_dot_lustre_setup(env, mdd);
1370 CERROR("%s: initializing .lustre failed: rc = %d\n",
1371 mdd2obd_dev(mdd)->obd_name, rc);
1375 /* Normal client usually send root access to MDT0 directly,
1376 * the root FID on non-MDT0 will only be used by echo client. */
1377 mdd->mdd_root_fid = fid;
1380 rc = mdd_orphan_index_init(env, mdd);
1384 rc = mdd_changelog_init(env, mdd);
1386 CERROR("%s: failed to initialize changelog: rc = %d\n",
1387 mdd2obd_dev(mdd)->obd_name, rc);
1391 rc = mdd_hsm_actions_llog_init(env, mdd);
1393 GOTO(out_changelog, rc);
1395 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1397 if (IS_ERR(nodemap_config)) {
1398 rc = PTR_ERR(nodemap_config);
1402 obt->obt_nodemap_config_file = nodemap_config;
1405 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1406 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1409 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1410 mdd2obd_dev(mdd)->obd_name, rc);
1411 GOTO(out_nodemap, rc);
1414 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1416 CERROR("%s: failed to register to barrier: rc = %d\n",
1417 mdd2obd_dev(mdd)->obd_name, rc);
1418 GOTO(out_lfsck, rc);
1424 lfsck_degister(env, mdd->mdd_bottom);
1426 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1427 obt->obt_nodemap_config_file = NULL;
1429 mdd_hsm_actions_llog_fini(env, mdd);
1431 mdd_changelog_fini(env, mdd);
1433 mdd_orphan_index_fini(env, mdd);
1435 if (mdd_seq_site(mdd)->ss_node_id == 0)
1436 mdd_dot_lustre_cleanup(env, mdd);
1439 dt_trans_stop(env, mdd->mdd_bottom, th);
1442 dt_object_put(env, root);
1444 local_oid_storage_fini(env, mdd->mdd_los);
1445 mdd->mdd_los = NULL;
1451 * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
1453 * Find corresponding device by passed parent and name, and allocate FID from
1456 * see include/lu_object.h for the details.
1458 static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1459 struct lu_fid *fid, struct lu_object *parent,
1460 const struct lu_name *name)
1462 struct mdd_device *mdd = lu2mdd_dev(d);
1463 struct lu_object *o = lu_object_next(parent);
1465 return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
1468 const struct lu_device_operations mdd_lu_ops = {
1469 .ldo_object_alloc = mdd_object_alloc,
1470 .ldo_process_config = mdd_process_config,
1471 .ldo_recovery_complete = mdd_recovery_complete,
1472 .ldo_prepare = mdd_prepare,
1473 .ldo_fid_alloc = mdd_fid_alloc,
1476 static int mdd_root_get(const struct lu_env *env,
1477 struct md_device *m, struct lu_fid *f)
1479 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1482 *f = mdd->mdd_root_fid;
1487 * No permission check is needed.
1489 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1490 struct obd_statfs *sfs)
1492 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1497 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs, NULL);
1499 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1504 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1505 struct md_device *m)
1507 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1509 return &mdd->mdd_dt_conf;
1512 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1515 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1517 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1518 return (*h == NULL ? -ENOENT : 0);
1521 static struct lu_device *mdd_device_free(const struct lu_env *env,
1522 struct lu_device *lu)
1524 struct mdd_device *m = lu2mdd_dev(lu);
1527 LASSERT(atomic_read(&lu->ld_ref) == 0);
1528 md_device_fini(&m->mdd_md_dev);
1533 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1534 struct lu_device_type *t,
1535 struct lustre_cfg *lcfg)
1537 struct lu_device *l;
1538 struct mdd_device *m;
1542 l = ERR_PTR(-ENOMEM);
1547 md_device_init(&m->mdd_md_dev, t);
1548 rc = mdd_init0(env, m, t, lcfg);
1550 mdd_device_free(env, l);
1559 * we use exports to track all mdd users
1561 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1562 struct obd_device *obd, struct obd_uuid *cluuid,
1563 struct obd_connect_data *data, void *localdata)
1565 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1566 struct lustre_handle conn;
1570 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1572 rc = class_connect(&conn, obd, cluuid);
1576 *exp = class_conn2export(&conn);
1578 /* Why should there ever be more than 1 connect? */
1579 LASSERT(mdd->mdd_connects == 0);
1580 mdd->mdd_connects++;
1586 * once last export (we don't count self-export) disappeared
1587 * mdd can be released
1589 static int mdd_obd_disconnect(struct obd_export *exp)
1591 struct obd_device *obd = exp->exp_obd;
1592 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1593 int rc, release = 0;
1596 mdd->mdd_connects--;
1597 if (mdd->mdd_connects == 0)
1600 rc = class_disconnect(exp);
1602 if (rc == 0 && release)
1603 class_manual_cleanup(obd);
1607 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1608 __u32 keylen, void *key, __u32 *vallen, void *val)
1612 if (KEY_IS(KEY_OSP_CONNECTED)) {
1613 struct obd_device *obd = exp->exp_obd;
1614 struct mdd_device *mdd;
1616 if (!obd->obd_set_up || obd->obd_stopping)
1619 mdd = lu2mdd_dev(obd->obd_lu_dev);
1621 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1629 static int mdd_obd_set_info_async(const struct lu_env *env,
1630 struct obd_export *exp,
1631 __u32 keylen, void *key,
1632 __u32 vallen, void *val,
1633 struct ptlrpc_request_set *set)
1635 struct obd_device *obd = exp->exp_obd;
1636 struct mdd_device *mdd;
1639 if (!obd->obd_set_up || obd->obd_stopping)
1642 mdd = lu2mdd_dev(obd->obd_lu_dev);
1644 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1649 static const struct obd_ops mdd_obd_device_ops = {
1650 .o_owner = THIS_MODULE,
1651 .o_connect = mdd_obd_connect,
1652 .o_disconnect = mdd_obd_disconnect,
1653 .o_get_info = mdd_obd_get_info,
1654 .o_set_info_async = mdd_obd_set_info_async,
1657 struct mdd_changelog_name_check_data {
1658 const char *mcnc_name;
1663 * changelog_recalc_mask callback
1665 * Is is called per each registered user and calculates combined mask of
1666 * all registered users.
1668 static int mdd_changelog_name_check_cb(const struct lu_env *env,
1669 struct llog_handle *llh,
1670 struct llog_rec_hdr *hdr, void *data)
1672 struct llog_changelog_user_rec2 *rec;
1673 struct mdd_changelog_name_check_data *mcnc = data;
1675 rec = container_of(hdr, typeof(*rec), cur_hdr);
1676 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
1677 !strncmp(rec->cur_name, mcnc->mcnc_name, sizeof(rec->cur_name))) {
1678 mcnc->mcnc_id = rec->cur_id;
1684 static int mdd_changelog_name_check(const struct lu_env *env,
1685 struct llog_ctxt *ctxt,
1686 struct mdd_device *mdd, const char *name)
1688 struct mdd_changelog_name_check_data mcnc = { .mcnc_name = name, };
1694 /* first symbol is a letter */
1695 if (!isalpha(name[0])) {
1697 CERROR("%s: first char '%c' in '%s' is not letter: rc = %d\n",
1698 mdd2obd_dev(mdd)->obd_name, name[0], name, rc);
1702 /* name is valid: contains letters, numbers and '-', '_' only */
1703 while (name[++chr]) {
1704 if (!(isalnum(name[chr]) || name[chr] == '_' ||
1705 name[chr] == '-')) {
1707 CERROR("%s: wrong char '%c' in name '%s': rc = %d\n",
1708 mdd2obd_dev(mdd)->obd_name, name[chr], name, rc);
1713 if (chr > CHANGELOG_USER_NAMELEN) {
1715 CERROR("%s: name '%s' is over %d symbols limit: rc = %d\n",
1716 mdd2obd_dev(mdd)->obd_name, name,
1717 CHANGELOG_USER_NAMELEN, rc);
1721 rc = llog_cat_process(env, ctxt->loc_handle,
1722 mdd_changelog_name_check_cb, &mcnc, 0, 0);
1724 CWARN("%s: changelog name %s exists already: rc = %d\n",
1725 mdd2obd_dev(mdd)->obd_name, name, rc);
1727 CWARN("%s: failed user changelog processing: rc = %d\n",
1728 mdd2obd_dev(mdd)->obd_name, rc);
1732 static int mdd_changelog_user_register(const struct lu_env *env,
1733 struct mdd_device *mdd, int *id,
1734 const char *name, const char *mask)
1736 struct llog_ctxt *ctxt;
1737 struct llog_changelog_user_rec2 *rec;
1738 char user_name[CHANGELOG_USER_NAMELEN_FULL];
1743 ctxt = llog_get_context(mdd2obd_dev(mdd),
1744 LLOG_CHANGELOG_USER_ORIG_CTXT);
1750 llog_ctxt_put(ctxt);
1754 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1756 rec->cur_hdr.lrh_len = sizeof(*rec);
1757 /* keep old record type for users without mask/name for
1758 * compatibility needs
1760 if (mask || (name && name[0]))
1761 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
1763 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1764 spin_lock(&mdd->mdd_cl.mc_user_lock);
1765 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1766 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1768 CERROR("%s: registering %s user: max ID is exceeded: rc = %d\n",
1769 mdd2obd_dev(mdd)->obd_name,
1770 (name && name[0]) ? name : "new", rc);
1773 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1774 mdd->mdd_cl.mc_users++;
1775 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1777 rec->cur_time = (__u32)ktime_get_real_seconds();
1778 if (OBD_FAIL_PRECHECK(OBD_FAIL_TIME_IN_CHLOG_USER)) {
1779 rec->cur_time -= min(cfs_fail_val, rec->cur_time);
1780 spin_lock(&mdd->mdd_cl.mc_user_lock);
1781 mdd->mdd_cl.mc_mintime = rec->cur_time;
1782 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1785 spin_lock(&mdd->mdd_cl.mc_lock);
1786 rec->cur_endrec = mdd->mdd_cl.mc_index;
1787 spin_unlock(&mdd->mdd_cl.mc_lock);
1790 /* if user will use relative mask apply it on default one */
1791 rec->cur_mask = CHANGELOG_DEFMASK;
1792 rc = cfs_str2mask(mask, changelog_type2str, &rec->cur_mask,
1793 CHANGELOG_MINMASK, CHANGELOG_ALLMASK,
1796 GOTO(out_users, rc);
1797 } else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK) {
1798 /* a maskless users means default mask but only if server has
1799 * no specific mask set
1801 rec->cur_mask = CHANGELOG_DEFMASK;
1804 if (name && name[0]) {
1805 rc = mdd_changelog_name_check(env, ctxt, mdd, name);
1807 GOTO(out_users, rc);
1808 strlcpy(rec->cur_name, name, sizeof(rec->cur_name));
1810 mdd_chlg_username(rec, user_name, sizeof(user_name));
1812 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1814 CWARN("%s: failed to register changelog user %s: rc = %d\n",
1815 mdd2obd_dev(mdd)->obd_name, user_name, rc);
1816 GOTO(out_users, rc);
1819 /* apply user mask finally */
1820 spin_lock(&mdd->mdd_cl.mc_user_lock);
1821 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
1822 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1824 CDEBUG(D_IOCTL, "%s: registered changelog user '%s', mask %#x\n",
1825 mdd2obd_dev(mdd)->obd_name, user_name, rec->cur_mask);
1827 /* Assume we want it on since somebody registered */
1828 rc = mdd_changelog_on(env, mdd);
1830 /* record is added, so don't decrement users on error */
1834 spin_lock(&mdd->mdd_cl.mc_user_lock);
1835 mdd->mdd_cl.mc_users--;
1836 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1840 llog_ctxt_put(ctxt);
1844 struct mdd_changelog_recalc_mask_data {
1845 struct mdd_device *mcrm_mdd;
1850 * changelog_recalc_mask callback
1852 * Is is called per each registered user and calculates combined mask of
1853 * all registered users.
1855 static int mdd_changelog_recalc_mask_cb(const struct lu_env *env,
1856 struct llog_handle *llh,
1857 struct llog_rec_hdr *hdr, void *data)
1859 struct llog_changelog_user_rec2 *rec;
1860 struct mdd_changelog_recalc_mask_data *mcrm = data;
1862 rec = container_of(hdr, typeof(*rec), cur_hdr);
1863 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
1864 mcrm->mcrm_mask |= rec->cur_mask;
1865 else if (mcrm->mcrm_mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
1866 mcrm->mcrm_mask |= CHANGELOG_DEFMASK;
1871 int mdd_changelog_recalc_mask(const struct lu_env *env, struct mdd_device *mdd)
1873 struct llog_ctxt *ctxt;
1874 struct mdd_changelog_recalc_mask_data mcrm = {
1876 .mcrm_mask = mdd->mdd_cl.mc_proc_mask,
1882 ctxt = llog_get_context(mdd2obd_dev(mdd),
1883 LLOG_CHANGELOG_USER_ORIG_CTXT);
1887 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1888 GOTO(out, rc = -ENXIO);
1890 rc = llog_cat_process(env, ctxt->loc_handle,
1891 mdd_changelog_recalc_mask_cb, &mcrm, 0, 0);
1893 CWARN("%s: failed user changelog processing: rc = %d\n",
1894 mdd2obd_dev(mdd)->obd_name, rc);
1896 spin_lock(&mdd->mdd_cl.mc_user_lock);
1897 CDEBUG(D_INFO, "%s: recalc changelog mask: %#x -> %#x\n",
1898 mdd2obd_dev(mdd)->obd_name, mdd->mdd_cl.mc_current_mask,
1900 mdd->mdd_cl.mc_current_mask = mcrm.mcrm_mask;
1901 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1905 llog_ctxt_put(ctxt);
1910 struct mdd_changelog_user_purge {
1911 struct mdd_device *mcup_mdd;
1913 __u32 mcup_usercount;
1916 char mcup_name[CHANGELOG_USER_NAMELEN_FULL];
1920 * changelog_user_purge callback
1922 * Is called once per user.
1924 * Check to see the user requested is available from rec.
1925 * Truncate the changelog.
1926 * Keep track of the total number of users (calls).
1928 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1929 struct llog_handle *llh,
1930 struct llog_rec_hdr *hdr, void *data)
1932 struct llog_changelog_user_rec2 *rec;
1933 struct mdd_changelog_user_purge *mcup = data;
1934 struct llog_cookie cookie;
1939 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1942 rec = container_of(hdr, typeof(*rec), cur_hdr);
1944 mcup->mcup_usercount++;
1946 if (rec->cur_id != mcup->mcup_id) {
1947 /* truncate to the lowest endrec that is not this user */
1948 mcup->mcup_minrec = min(mcup->mcup_minrec, rec->cur_endrec);
1952 mdd_chlg_username(rec, mcup->mcup_name, sizeof(mcup->mcup_name));
1954 /* Unregister this user */
1955 cookie.lgc_lgl = llh->lgh_id;
1956 cookie.lgc_index = hdr->lrh_index;
1957 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1960 mcup->mcup_found = true;
1961 mcup->mcup_usercount--;
1962 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1963 mcup->mcup_mdd->mdd_cl.mc_users--;
1964 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1970 int mdd_changelog_user_purge(const struct lu_env *env,
1971 struct mdd_device *mdd, __u32 id)
1973 struct mdd_changelog_user_purge mcup = {
1976 .mcup_found = false,
1977 .mcup_usercount = 0,
1978 .mcup_minrec = ULLONG_MAX,
1981 struct llog_ctxt *ctxt;
1986 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1987 mdd2obd_dev(mdd)->obd_name, id);
1989 ctxt = llog_get_context(mdd2obd_dev(mdd),
1990 LLOG_CHANGELOG_USER_ORIG_CTXT);
1993 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1994 GOTO(out, rc = -ENXIO);
1996 rc = llog_cat_process(env, ctxt->loc_handle,
1997 mdd_changelog_user_purge_cb, &mcup, 0, 0);
1999 CWARN("%s: failed to purge changelog for user %s: rc = %d\n",
2000 mdd2obd_dev(mdd)->obd_name, mcup.mcup_name, rc);
2004 OBD_FAIL_TIMEOUT(OBD_FAIL_LLOG_PURGE_DELAY, cfs_fail_val);
2005 if (mcup.mcup_usercount == 0) {
2006 spin_lock(&mdd->mdd_cl.mc_user_lock);
2007 if (mdd->mdd_cl.mc_users == 0) {
2008 /* No more users; turn changelogs off */
2009 CDEBUG(D_IOCTL, "turning off changelogs\n");
2010 rc = mdd_changelog_off(env, mdd);
2012 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2015 if (mcup.mcup_found) {
2017 "%s: Purge changelog entries for user %s record=%llu\n",
2018 mdd2obd_dev(mdd)->obd_name,
2019 mcup.mcup_name, mcup.mcup_minrec);
2020 rc = mdd_changelog_llog_cancel(env, mdd, mcup.mcup_minrec);
2022 CWARN("%s: No changelog for user id %u: rc = %d\n",
2023 mdd2obd_dev(mdd)->obd_name, id, rc);
2024 GOTO(out, rc = -ENOENT);
2027 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
2031 llog_ctxt_put(ctxt);
2036 struct mdd_changelog_user_clear {
2042 struct mdd_device *mcuc_mdd;
2043 char mcuc_name[CHANGELOG_USER_NAMELEN_FULL];
2047 * changelog_clear callback
2049 * Is called once per user.
2051 * Check to see the user requested is available from rec.
2052 * Check the oldest (smallest) record for boundary conditions.
2053 * Truncate the changelog.
2055 static int mdd_changelog_clear_cb(const struct lu_env *env,
2056 struct llog_handle *llh,
2057 struct llog_rec_hdr *hdr,
2060 struct llog_changelog_user_rec2 *rec;
2061 struct mdd_changelog_user_clear *mcuc = data;
2062 struct mdd_device *mdd = mcuc->mcuc_mdd;
2067 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
2070 rec = container_of(hdr, typeof(*rec), cur_hdr);
2071 /* Does the changelog id match the requested id? */
2072 if (rec->cur_id != mcuc->mcuc_id) {
2073 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec, rec->cur_endrec);
2074 mcuc->mcuc_mintime = min(mcuc->mcuc_mintime, rec->cur_time);
2078 mdd_chlg_username(rec, mcuc->mcuc_name, sizeof(mcuc->mcuc_name));
2079 /* cur_endrec is the oldest purgeable record, make sure we're newer */
2080 if (rec->cur_endrec > mcuc->mcuc_endrec) {
2083 "%s: request %llu < endrec %llu for user %s: rc = %d\n",
2084 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_endrec,
2085 rec->cur_endrec, mcuc->mcuc_name, rc);
2090 * Flag that we've met all the range and user checks.
2091 * We now know the record to flush.
2093 mcuc->mcuc_flush = true;
2095 rec->cur_endrec = mcuc->mcuc_endrec;
2096 rec->cur_time = (__u32)ktime_get_real_seconds();
2098 CDEBUG(D_IOCTL, "%s: update changelog user %s endrec = %llu\n",
2099 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_name, rec->cur_endrec);
2101 /* Update the endrec */
2102 rc = llog_write(env, llh, hdr, hdr->lrh_index);
2108 * Clear a changelog up to entry specified by endrec for user id.
2110 static int mdd_changelog_clear(const struct lu_env *env,
2111 struct mdd_device *mdd, __u32 id,
2114 struct mdd_changelog_user_clear mcuc = {
2116 .mcuc_minrec = endrec,
2117 .mcuc_flush = false,
2119 .mcuc_mintime = ktime_get_real_seconds(),
2122 struct llog_ctxt *ctxt;
2128 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
2129 mdd2obd_dev(mdd)->obd_name, id, endrec);
2130 /* start_rec is the newest (largest value) entry in the changelogs*/
2131 spin_lock(&mdd->mdd_cl.mc_lock);
2132 start_rec = mdd->mdd_cl.mc_index;
2133 spin_unlock(&mdd->mdd_cl.mc_lock);
2135 if (start_rec < endrec) {
2136 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
2137 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
2142 mcuc.mcuc_endrec = start_rec;
2143 mcuc.mcuc_minrec = ULLONG_MAX;
2145 mcuc.mcuc_endrec = endrec;
2148 ctxt = llog_get_context(mdd2obd_dev(mdd),
2149 LLOG_CHANGELOG_USER_ORIG_CTXT);
2153 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
2154 GOTO(out, rc = -ENXIO);
2156 rc = llog_cat_process(env, ctxt->loc_handle, mdd_changelog_clear_cb,
2158 if (rc == -EINVAL) {
2159 CDEBUG(D_IOCTL, "%s: no changelog recnum <= %llu to clear\n",
2160 mdd2obd_dev(mdd)->obd_name, (unsigned long long)endrec);
2165 CWARN("%s: can't clear the changelog for user %s: rc = %d\n",
2166 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, rc);
2170 if (!mcuc.mcuc_flush) {
2171 CDEBUG(D_IOCTL, "%s: no entry for user %d\n",
2172 mdd2obd_dev(mdd)->obd_name, id);
2173 GOTO(out, rc = -ENOENT);
2176 CDEBUG(D_IOCTL, "%s: purge changelog user %s entries up to %llu\n",
2177 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, mcuc.mcuc_minrec);
2179 rc = mdd_changelog_llog_cancel(env, mdd, mcuc.mcuc_minrec);
2183 spin_lock(&mdd->mdd_cl.mc_user_lock);
2184 mdd->mdd_cl.mc_minrec = mcuc.mcuc_minrec;
2185 mdd->mdd_cl.mc_mintime = mcuc.mcuc_mintime;
2186 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2190 llog_ctxt_put(ctxt);
2195 static int mdd_changelog_user_deregister(const struct lu_env *env,
2196 struct mdd_device *mdd, int *id,
2199 struct llog_ctxt *ctxt;
2200 struct mdd_changelog_name_check_data mcnc = {
2209 ctxt = llog_get_context(mdd2obd_dev(mdd),
2210 LLOG_CHANGELOG_USER_ORIG_CTXT);
2214 rc = llog_cat_process(env, ctxt->loc_handle,
2215 mdd_changelog_name_check_cb, &mcnc, 0, 0);
2216 llog_ctxt_put(ctxt);
2218 if (rc != -EEXIST) {
2219 CDEBUG(D_IOCTL, "%s: no entry for username %s\n",
2220 mdd2obd_dev(mdd)->obd_name, name);
2226 /* explicitly clear changelog first, to protect from crash in
2227 * the middle of purge that would lead to unregistered consumer
2228 * but pending changelog entries
2230 rc = mdd_changelog_clear(env, mdd, *id, 0);
2232 rc = mdd_changelog_user_purge(env, mdd, *id);
2234 /* recalc changelog current mask */
2235 mdd_changelog_recalc_mask(env, mdd);
2241 * May be called remotely from mdt_iocontrol_handle or locally from
2242 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
2243 * an obd_ioctl_data format (but local ioctl handler does).
2245 * \param len - data len
2246 * \param karg - ioctl data, in kernel space
2248 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
2249 unsigned int cmd, int len, void *karg)
2251 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
2252 struct obd_ioctl_data *data = karg;
2256 /* Doesn't use obd_ioctl_data */
2258 case OBD_IOC_CHANGELOG_CLEAR: {
2259 struct changelog_setinfo *cs = karg;
2261 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2262 RETURN(-EINPROGRESS);
2264 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
2266 barrier_exit(mdd->mdd_bottom);
2269 case OBD_IOC_START_LFSCK: {
2270 rc = lfsck_start(env, mdd->mdd_bottom,
2271 (struct lfsck_start_param *)karg);
2274 case OBD_IOC_STOP_LFSCK: {
2275 rc = lfsck_stop(env, mdd->mdd_bottom,
2276 (struct lfsck_stop *)karg);
2279 case OBD_IOC_QUERY_LFSCK: {
2280 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
2281 (struct lfsck_query *)karg);
2286 /* Below ioctls use obd_ioctl_data */
2287 if (data->ioc_version != OBD_IOCTL_VERSION) {
2288 CERROR("Bad magic %x != %x\n", data->ioc_version,
2294 case OBD_IOC_CHANGELOG_REG:
2295 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2296 RETURN(-EINPROGRESS);
2298 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1,
2301 barrier_exit(mdd->mdd_bottom);
2303 case OBD_IOC_CHANGELOG_DEREG:
2304 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2305 RETURN(-EINPROGRESS);
2307 rc = mdd_changelog_user_deregister(env, mdd, &data->ioc_u32_1,
2310 barrier_exit(mdd->mdd_bottom);
2319 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
2320 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
2322 static const struct md_device_operations mdd_ops = {
2323 .mdo_statfs = mdd_statfs,
2324 .mdo_root_get = mdd_root_get,
2325 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
2326 .mdo_iocontrol = mdd_iocontrol,
2327 .mdo_dtconf_get = mdd_dtconf_get,
2330 static const struct lu_device_type_operations mdd_device_type_ops = {
2331 .ldto_init = mdd_type_init,
2332 .ldto_fini = mdd_type_fini,
2334 .ldto_start = mdd_type_start,
2335 .ldto_stop = mdd_type_stop,
2337 .ldto_device_alloc = mdd_device_alloc,
2338 .ldto_device_free = mdd_device_free,
2340 .ldto_device_fini = mdd_device_fini
2343 static struct lu_device_type mdd_device_type = {
2344 .ldt_tags = LU_DEVICE_MD,
2345 .ldt_name = LUSTRE_MDD_NAME,
2346 .ldt_ops = &mdd_device_type_ops,
2347 .ldt_ctx_tags = LCT_MD_THREAD
2350 /* context key constructor: mdd_key_init */
2351 LU_KEY_INIT(mdd, struct mdd_thread_info);
2353 static void mdd_key_fini(const struct lu_context *ctx,
2354 struct lu_context_key *key, void *data)
2356 struct mdd_thread_info *info = data;
2358 lu_buf_free(&info->mdi_big_buf);
2359 lu_buf_free(&info->mdi_link_buf);
2360 lu_buf_free(&info->mdi_xattr_buf);
2361 lu_buf_free(&info->mdi_chlg_buf);
2366 /* context key: mdd_thread_key */
2367 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
2369 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
2370 int (*func)(void *), void *data, char *name)
2372 struct task_struct *task;
2374 LASSERT(thread->mgt_init == false);
2375 init_completion(&thread->mgt_started);
2376 init_completion(&thread->mgt_finished);
2377 thread->mgt_data = data;
2378 thread->mgt_abort = false;
2379 thread->mgt_init = true;
2381 task = kthread_run(func, thread, name);
2383 complete(&thread->mgt_finished);
2384 return PTR_ERR(task);
2386 wait_for_completion(&thread->mgt_started);
2390 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
2392 if (thread->mgt_init == true) {
2393 thread->mgt_abort = true;
2394 wait_for_completion(&thread->mgt_finished);
2398 static int __init mdd_init(void)
2402 rc = lu_kmem_init(mdd_caches);
2406 changelog_orig_logops = llog_common_cat_ops;
2407 changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
2409 rc = class_register_type(&mdd_obd_device_ops, NULL, false,
2410 LUSTRE_MDD_NAME, &mdd_device_type);
2412 lu_kmem_fini(mdd_caches);
2416 static void __exit mdd_exit(void)
2418 class_unregister_type(LUSTRE_MDD_NAME);
2419 lu_kmem_fini(mdd_caches);
2422 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2423 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2424 MODULE_VERSION(LUSTRE_VERSION_STRING);
2425 MODULE_LICENSE("GPL");
2427 module_init(mdd_init);
2428 module_exit(mdd_exit);