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 /* with a significant amount of idle time */
153 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
154 /* or a significant amount of late indexes */
155 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
156 /* with a reasonable interval between each check */
157 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
158 /* with a very few number of free catalog entries */
159 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
160 /* special default striping for files created with O_APPEND */
161 mdd->mdd_append_stripe_count = 1;
162 mdd->mdd_append_pool[0] = '\0';
164 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
166 /* we are using service name but not mdd obd name
167 * for compatibility reasons.
168 * It is passed from MDT in lustre_cfg[2] buffer */
169 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
171 obd_disconnect(mdd->mdd_child_exp);
176 static struct lu_device *mdd_device_fini(const struct lu_env *env,
179 struct mdd_device *mdd = lu2mdd_dev(d);
182 lu_dev_del_linkage(d->ld_site, d);
184 mdd_procfs_fini(mdd);
188 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
189 struct llog_rec_hdr *hdr, void *data)
191 struct mdd_device *mdd = (struct mdd_device *)data;
192 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
194 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
195 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
198 "seeing record at index %d/%d/%llu t=%x %.*s in log"
199 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
200 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
201 changelog_rec_name(&rec->cr), PFID(&llh->lgh_id.lgl_oi.oi_fid));
202 spin_lock(&mdd->mdd_cl.mc_lock);
203 mdd->mdd_cl.mc_index = rec->cr.cr_index;
204 spin_unlock(&mdd->mdd_cl.mc_lock);
205 return LLOG_PROC_BREAK;
208 char *mdd_chlg_username(struct llog_changelog_user_rec2 *rec, char *buf,
211 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
213 snprintf(buf, len, "%s%u-%s", CHANGELOG_USER_PREFIX,
214 rec->cur_id, rec->cur_name);
216 snprintf(buf, len, "%s%u", CHANGELOG_USER_PREFIX, rec->cur_id);
220 __u32 mdd_chlg_usermask(struct llog_changelog_user_rec2 *rec)
222 return rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 ?
226 static int changelog_user_init_cb(const struct lu_env *env,
227 struct llog_handle *llh,
228 struct llog_rec_hdr *hdr, void *data)
230 struct mdd_device *mdd = data;
231 struct llog_changelog_user_rec2 *rec;
232 char user_name[CHANGELOG_USER_NAMELEN_FULL];
234 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
236 rec = container_of(hdr, typeof(*rec), cur_hdr);
237 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
238 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
239 CWARN("%s: unknown user type %x at index %u in log "DFID"\n",
240 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
241 rec->cur_hdr.lrh_type, PFID(&llh->lgh_id.lgl_oi.oi_fid));
246 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
247 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
249 hdr->lrh_index, rec->cur_hdr.lrh_index, rec->cur_endrec,
250 PFID(&llh->lgh_id.lgl_oi.oi_fid));
252 spin_lock(&mdd->mdd_cl.mc_user_lock);
253 mdd->mdd_cl.mc_lastuser = rec->cur_id;
254 mdd->mdd_cl.mc_users++;
255 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
256 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
257 else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
258 mdd->mdd_cl.mc_current_mask |= CHANGELOG_DEFMASK;
259 spin_unlock(&mdd->mdd_cl.mc_user_lock);
260 spin_lock(&mdd->mdd_cl.mc_lock);
261 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
262 mdd->mdd_cl.mc_index = rec->cur_endrec;
263 spin_unlock(&mdd->mdd_cl.mc_lock);
265 return LLOG_PROC_BREAK;
268 struct changelog_orphan_data {
270 struct mdd_device *clod_mdd;
273 /* find oldest changelog record index */
274 static int changelog_detect_orphan_cb(const struct lu_env *env,
275 struct llog_handle *llh,
276 struct llog_rec_hdr *hdr, void *data)
278 struct changelog_orphan_data *clod = data;
279 struct mdd_device *mdd = clod->clod_mdd;
280 struct llog_changelog_rec *rec;
282 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
284 rec = container_of(hdr, typeof(*rec), cr_hdr);
285 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
286 CWARN("%s: invalid record at index %d in log "DFID"\n",
287 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
288 PFID(&llh->lgh_id.lgl_oi.oi_fid));
289 /* try to find some next valid record and thus allow to recover
290 * from a corrupted LLOG, instead to assert and force a crash
296 "%s: record at index %d/%d/%llu t=%x %.*s in log "DFID"\n",
297 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
298 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
299 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
300 PFID(&llh->lgh_id.lgl_oi.oi_fid));
302 clod->clod_index = rec->cr.cr_index;
304 return LLOG_PROC_BREAK;
307 /* find oldest changelog user index */
308 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
309 struct llog_handle *llh,
310 struct llog_rec_hdr *hdr, void *data)
312 struct changelog_orphan_data *clod = data;
313 struct mdd_device *mdd = clod->clod_mdd;
314 struct llog_changelog_user_rec2 *rec;
315 char user_name[CHANGELOG_USER_NAMELEN_FULL];
317 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
319 rec = container_of(hdr, typeof(*rec), cur_hdr);
320 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
321 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
322 CWARN("%s: unknown user type %u at index %u in log "DFID"\n",
323 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
324 rec->cur_hdr.lrh_type, PFID(&llh->lgh_id.lgl_oi.oi_fid));
325 /* try to find some next valid record and thus allow to recover
326 * from a corrupted LLOG, instead to assert and force a crash
331 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
332 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
334 hdr->lrh_index, rec->cur_hdr.lrh_index,
335 rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
337 clod->clod_index = min_t(__u64, clod->clod_index, rec->cur_endrec);
342 struct changelog_cancel_cookie {
344 struct mdd_device *mdd;
347 static int llog_changelog_cancel_cb(const struct lu_env *env,
348 struct llog_handle *llh,
349 struct llog_rec_hdr *hdr, void *data)
351 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
352 struct changelog_cancel_cookie *cl_cookie =
353 (struct changelog_cancel_cookie *)data;
357 /* This is always a (sub)log, not the catalog */
358 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
360 /* if current context is GC-thread allow it to stop upon umount
361 * remaining records cleanup will occur upon next mount
363 * also during testing, wait for GC-thread to be released
365 * XXX this requires the GC-thread to not fork a sub-thread via
366 * llog[_cat]_process_or_fork() and we may think to also implement
367 * this shutdown mechanism for manually started user unregister which
368 * can also take a long time if huge backlog of records
370 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
371 /* wait to be released */
372 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
375 if (kthread_should_stop())
376 RETURN(LLOG_PROC_BREAK);
379 if (rec->cr.cr_index > cl_cookie->endrec)
380 /* records are in order, so we're done */
381 RETURN(LLOG_PROC_BREAK);
383 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
384 if (cfs_fail_val == 0)
385 cfs_fail_val = hdr->lrh_index;
386 if (cfs_fail_val == hdr->lrh_index)
387 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
390 /* Records folow one by one, cr_index++. We could calculate the
391 * last cr_index at this plain llog. And if it less then cookie endrec
392 * cancel the whole file.
394 if ((LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr) - hdr->lrh_index +
395 rec->cr.cr_index) < cl_cookie->endrec) {
398 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
399 if (cfs_fail_val == 0) {
400 cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
401 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
404 rc = llog_destroy(env, llh);
406 CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
407 PFID(&llh->lgh_id.lgl_oi.oi_fid));
408 RETURN(LLOG_DEL_PLAIN);
412 /* cancel them one at a time. I suppose we could store up the cookies
413 * and cancel them all at once; probably more efficient, but this is
414 * done as a user call, so who cares... */
416 RETURN(LLOG_DEL_RECORD);
419 static int llog_changelog_cancel(const struct lu_env *env,
420 struct llog_ctxt *ctxt,
421 struct changelog_cancel_cookie *cookie)
423 struct llog_handle *cathandle = ctxt->loc_handle;
428 /* This should only be called with the catalog handle */
429 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
431 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
434 /* 0 or 1 means we're done */
437 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
438 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
439 PFID(&cathandle->lgh_id.lgl_oi.oi_fid), rc);
444 static struct llog_operations changelog_orig_logops;
447 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
451 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
455 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
458 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
459 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
460 CERROR("Changelogs cannot be enabled due to error "
461 "condition (see %s log).\n",
462 mdd2obd_dev(mdd)->obd_name);
465 spin_lock(&mdd->mdd_cl.mc_lock);
466 mdd->mdd_cl.mc_flags |= CLM_ON;
467 spin_unlock(&mdd->mdd_cl.mc_lock);
468 rc = mdd_changelog_write_header(env, mdd, CLM_START);
474 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
478 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
481 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
482 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
483 spin_lock(&mdd->mdd_cl.mc_lock);
484 mdd->mdd_cl.mc_flags &= ~CLM_ON;
485 spin_unlock(&mdd->mdd_cl.mc_lock);
490 static int mdd_changelog_llog_init(const struct lu_env *env,
491 struct mdd_device *mdd)
493 struct obd_device *obd = mdd2obd_dev(mdd);
494 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
495 struct changelog_orphan_data clod = {
506 /* LU-2844 mdd setup failure should not cause umount oops */
507 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
510 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
511 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
512 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
513 obd, &changelog_orig_logops);
515 CERROR("%s: changelog llog setup failed: rc = %d\n",
520 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
523 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
526 GOTO(out_cleanup, rc);
528 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
532 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
533 changelog_init_cb, mdd);
536 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
541 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
542 mdd->mdd_cl.mc_index);
544 /* setup user changelog */
545 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
546 obd, &changelog_orig_logops);
548 CERROR("%s: changelog users llog setup failed: rc = %d\n",
553 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
556 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
559 GOTO(out_ucleanup, rc);
561 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
563 GOTO(out_uclose, rc);
565 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
566 changelog_user_init_cb, mdd);
568 CERROR("%s: changelog user init failed: rc = %d\n",
570 GOTO(out_uclose, rc);
573 /* Finally apply per-server mask */
574 mdd->mdd_cl.mc_current_mask |= mdd->mdd_cl.mc_proc_mask;
576 /* If we have registered users, assume we want changelogs on */
577 if (mdd->mdd_cl.mc_lastuser > 0) {
578 rc = mdd_changelog_on(env, mdd);
580 GOTO(out_uclose, rc);
583 /* find and clear any orphan changelog records (1st record index <
584 * smallest of all users current index), likely to come from an
585 * interrupted manual or GC-thread purge, as its user record had
587 * XXX we may wait for a still registered user clear operation to
588 * do the job, but it may then take a long time to reach the user's
589 * real targetted records if a huge purge backlog is still to be
590 * processed as a long time idle user record could have been deleted
591 * XXX we may need to run end of purge as a separate thread
593 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
596 CERROR("%s: changelog detect orphan failed: rc = %d\n",
598 GOTO(out_uclose, rc);
600 rc = llog_cat_process(env, uctxt->loc_handle,
601 changelog_user_detect_orphan_cb,
604 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
606 GOTO(out_uclose, rc);
608 if (unlikely(clod.clod_index < user_orphan.clod_index)) {
609 struct changelog_cancel_cookie cl_cookie = {
610 .endrec = user_orphan.clod_index,
614 CWARN("%s : orphan changelog records found, starting from "
615 "index %llu to index %llu, being cleared now\n",
616 obd->obd_name, clod.clod_index, user_orphan.clod_index);
618 /* XXX we may need to run end of purge as a separate thread */
619 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
621 CERROR("%s: purge of changelog orphan records failed: "
622 "rc = %d\n", obd->obd_name, rc);
623 GOTO(out_uclose, rc);
628 llog_ctxt_put(uctxt);
631 llog_cat_close(env, uctxt->loc_handle);
633 llog_cleanup(env, uctxt);
635 llog_cat_close(env, ctxt->loc_handle);
637 llog_cleanup(env, ctxt);
641 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
643 struct obd_device *obd = mdd2obd_dev(mdd);
646 mdd->mdd_cl.mc_index = 0;
647 spin_lock_init(&mdd->mdd_cl.mc_lock);
648 mdd->mdd_cl.mc_starttime = ktime_get();
649 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
650 mdd->mdd_cl.mc_lastuser = 0;
652 /* ensure a GC check will, and a thread run may, occur upon start */
653 mdd->mdd_cl.mc_gc_time = 0;
654 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
656 rc = mdd_changelog_llog_init(env, mdd);
658 CERROR("%s: changelog setup during init failed: rc = %d\n",
660 mdd->mdd_cl.mc_flags |= CLM_ERR;
666 static void mdd_changelog_fini(const struct lu_env *env,
667 struct mdd_device *mdd)
669 struct obd_device *obd = mdd2obd_dev(mdd);
670 struct llog_ctxt *ctxt;
672 if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
674 mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
677 /* stop GC-thread if running */
678 spin_lock(&mdd->mdd_cl.mc_lock);
679 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
680 /* avoid any attempt to run a GC-thread */
681 mdd->mdd_cl.mc_gc_task = current;
682 spin_unlock(&mdd->mdd_cl.mc_lock);
684 struct task_struct *gc_task;
686 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
687 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
688 /* need to wait for birthing GC-thread to be started
689 * and to have set mc_gc_task to itself
691 spin_unlock(&mdd->mdd_cl.mc_lock);
692 /* Add a tiny sleep */
693 schedule_timeout_uninterruptible(1);
694 /* go back to fully check if GC-thread has started or
695 * even already exited or if a new one is starting...
699 /* take a reference on task_struct to avoid it to be freed
702 gc_task = mdd->mdd_cl.mc_gc_task;
703 get_task_struct(gc_task);
704 spin_unlock(&mdd->mdd_cl.mc_lock);
705 kthread_stop(gc_task);
706 put_task_struct(gc_task);
709 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
711 llog_cat_close(env, ctxt->loc_handle);
712 llog_cleanup(env, ctxt);
714 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
716 llog_cat_close(env, ctxt->loc_handle);
717 llog_cleanup(env, ctxt);
721 /** Remove entries with indicies up to and including \a endrec from the
728 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
731 struct obd_device *obd = mdd2obd_dev(mdd);
732 struct llog_ctxt *ctxt;
733 long long unsigned cur;
734 struct changelog_cancel_cookie cookie;
737 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
741 spin_lock(&mdd->mdd_cl.mc_lock);
742 cur = (long long)mdd->mdd_cl.mc_index;
743 spin_unlock(&mdd->mdd_cl.mc_lock);
747 /* purge to "0" is shorthand for everything */
751 /* If purging all records, write a header entry so we don't have an
752 empty catalog and we're sure to have a valid starting index next
753 time. In case of crash, we just restart with old log so we're
756 /* XXX: transaction is started by llog itself */
757 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
762 /* Some records were purged, so reset repeat-access time (so we
763 record new mtime update records, so users can see a file has been
764 changed since the last purge) */
765 mdd->mdd_cl.mc_starttime = ktime_get();
767 cookie.endrec = endrec;
769 rc = llog_changelog_cancel(env, ctxt, &cookie);
775 /** Add a CL_MARK record to the changelog
777 * \param markerflags - CLM_*
780 int mdd_changelog_write_header(const struct lu_env *env,
781 struct mdd_device *mdd, int markerflags)
783 struct obd_device *obd = mdd2obd_dev(mdd);
784 struct llog_changelog_rec *rec;
786 struct llog_ctxt *ctxt;
788 int len = strlen(obd->obd_name);
793 if (mdd->mdd_cl.mc_current_mask & BIT(CL_MARK)) {
794 mdd->mdd_cl.mc_starttime = ktime_get();
798 reclen = llog_data_len(sizeof(*rec) + len);
799 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
800 if (buf->lb_buf == NULL)
804 rec->cr.cr_flags = CLF_VERSION;
805 rec->cr.cr_type = CL_MARK;
806 rec->cr.cr_namelen = len;
807 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
808 /* Status and action flags */
809 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
810 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
812 rec->cr_hdr.lrh_type = CHANGELOG_REC;
813 rec->cr.cr_time = cl_time();
815 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
818 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
823 /* assume on or off event; reset repeat-access time */
824 mdd->mdd_cl.mc_starttime = ktime_get();
829 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
830 * are valid. We also check if object with passed fid exists or not.
832 static int obf_lookup(const struct lu_env *env, struct md_object *p,
833 const struct lu_name *lname, struct lu_fid *f,
834 struct md_op_spec *spec)
836 char *name = (char *)lname->ln_name;
837 struct mdd_device *mdd = mdo2mdd(p);
838 struct mdd_object *child;
844 sscanf(name, SFID, RFID(f));
846 GOTO(out, rc = -ENOENT);
848 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
849 !fid_seq_is_dot(f->f_seq))
850 GOTO(out, rc = -ENOENT);
852 /* Check if object with this fid exists */
853 child = mdd_object_find(env, mdd, f);
855 GOTO(out, rc = PTR_ERR(child));
857 if (mdd_object_exists(child) == 0)
860 mdd_object_put(env, child);
866 static int mdd_dummy_create(const struct lu_env *env,
867 struct md_object *pobj,
868 const struct lu_name *lname,
869 struct md_object *child,
870 struct md_op_spec *spec,
876 static int mdd_dummy_rename(const struct lu_env *env,
877 struct md_object *src_pobj,
878 struct md_object *tgt_pobj,
879 const struct lu_fid *lf,
880 const struct lu_name *lsname,
881 struct md_object *tobj,
882 const struct lu_name *ltname,
888 static int mdd_dummy_link(const struct lu_env *env,
889 struct md_object *tgt_obj,
890 struct md_object *src_obj,
891 const struct lu_name *lname,
897 static int mdd_dummy_unlink(const struct lu_env *env,
898 struct md_object *pobj,
899 struct md_object *cobj,
900 const struct lu_name *lname,
907 int mdd_create(const struct lu_env *env, struct md_object *pobj,
908 const struct lu_name *lname, struct md_object *child,
909 struct md_op_spec *spec, struct md_attr *ma);
910 static int mdd_obf_create(const struct lu_env *env, struct md_object *pobj,
911 const struct lu_name *lname, struct md_object *child,
912 struct md_op_spec *spec, struct md_attr *ma)
914 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
915 return mdd_create(env, pobj, lname, child, spec, ma);
919 static const struct md_dir_operations mdd_obf_dir_ops = {
920 .mdo_lookup = obf_lookup,
921 .mdo_create = mdd_obf_create,
922 .mdo_rename = mdd_dummy_rename,
923 .mdo_link = mdd_dummy_link,
924 .mdo_unlink = mdd_dummy_unlink
927 static const struct md_dir_operations mdd_lpf_dir_ops = {
928 .mdo_lookup = mdd_lookup,
929 .mdo_create = mdd_dummy_create,
930 .mdo_rename = mdd_dummy_rename,
931 .mdo_link = mdd_dummy_link,
932 .mdo_unlink = mdd_dummy_unlink
935 static struct md_object *mdo_locate(const struct lu_env *env,
936 struct md_device *md,
937 const struct lu_fid *fid)
939 struct lu_object *obj;
940 struct md_object *mdo;
942 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
944 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
945 LASSERT(obj != NULL);
953 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
955 struct md_object *mdo;
956 struct mdd_object *mdd_lpf;
957 struct lu_fid fid = LU_LPF_FID;
961 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
962 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
967 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
969 RETURN(PTR_ERR(mdo));
971 LASSERT(lu_object_exists(&mdo->mo_lu));
973 mdd_lpf = md2mdd_obj(mdo);
974 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
975 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
981 * Create special in-memory "fid" object for open-by-fid.
983 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
985 struct md_object *mdo;
986 struct mdd_object *mdd_obf;
987 struct lu_fid fid = LU_OBF_FID;
990 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
991 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
995 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
997 RETURN(PTR_ERR(mdo));
999 LASSERT(lu_object_exists(&mdo->mo_lu));
1001 mdd_obf = md2mdd_obj(mdo);
1002 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1003 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
1008 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
1009 struct mdd_device *m)
1011 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
1012 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
1013 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
1015 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
1016 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
1017 m->mdd_dot_lustre_objs.mdd_obf = NULL;
1019 if (m->mdd_dot_lustre != NULL) {
1020 mdd_object_put(env, m->mdd_dot_lustre);
1021 m->mdd_dot_lustre = NULL;
1025 /** Setup ".lustre" directory object */
1026 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1028 struct md_object *mdo;
1033 /* Create ".lustre" directory in ROOT. */
1034 fid = LU_DOT_LUSTRE_FID;
1035 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
1037 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
1041 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1043 RETURN(PTR_ERR(mdo));
1044 LASSERT(lu_object_exists(&mdo->mo_lu));
1046 m->mdd_dot_lustre = md2mdd_obj(mdo);
1048 rc = mdd_obf_setup(env, m);
1050 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
1051 mdd2obd_dev(m)->obd_name, rc);
1055 rc = mdd_lpf_setup(env, m);
1057 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
1058 mdd2obd_dev(m)->obd_name, rc);
1065 mdd_dot_lustre_cleanup(env, m);
1071 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
1072 * object in obd_device
1074 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
1075 struct mdd_device *m)
1077 struct obd_device *obd = mdd2obd_dev(m);
1078 struct llog_ctxt *ctxt = NULL;
1082 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1083 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1085 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1086 obd, &llog_common_cat_ops);
1088 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1093 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1096 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1099 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1101 GOTO(out_cleanup, rc);
1104 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1106 GOTO(out_close, rc);
1108 llog_ctxt_put(ctxt);
1112 llog_cat_close(env, ctxt->loc_handle);
1113 ctxt->loc_handle = NULL;
1115 llog_cleanup(env, ctxt);
1121 * cleanup the context created by llog_setup_named()
1123 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1124 struct mdd_device *m)
1126 struct obd_device *obd = mdd2obd_dev(m);
1127 struct llog_ctxt *lctxt;
1130 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1132 llog_cat_close(env, lctxt->loc_handle);
1133 lctxt->loc_handle = NULL;
1134 llog_cleanup(env, lctxt);
1140 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1141 struct lustre_cfg *cfg)
1143 barrier_deregister(m->mdd_bottom);
1144 lfsck_degister(env, m->mdd_bottom);
1145 mdd_hsm_actions_llog_fini(env, m);
1146 mdd_changelog_fini(env, m);
1147 mdd_orphan_index_fini(env, m);
1148 mdd_dot_lustre_cleanup(env, m);
1149 if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) {
1150 nm_config_file_deregister_tgt(env,
1151 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
1152 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL;
1154 if (m->mdd_los != NULL) {
1155 local_oid_storage_fini(env, m->mdd_los);
1158 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1160 if (m->mdd_child_exp)
1161 obd_disconnect(m->mdd_child_exp);
1164 static int mdd_process_config(const struct lu_env *env,
1165 struct lu_device *d, struct lustre_cfg *cfg)
1167 struct mdd_device *m = lu2mdd_dev(d);
1168 struct dt_device *dt = m->mdd_child;
1169 struct lu_device *next = &dt->dd_lu_dev;
1173 switch (cfg->lcfg_command) {
1177 count = class_modify_config(cfg, PARAM_MDD, &m->mdd_kobj);
1178 rc = count > 0 ? 0 : count;
1180 /* we don't understand; pass it on */
1181 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1185 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1188 dt_conf_get(env, dt, &m->mdd_dt_conf);
1190 case LCFG_PRE_CLEANUP:
1191 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1192 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1195 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1196 lu_dev_del_linkage(d->ld_site, d);
1197 mdd_device_shutdown(env, m, cfg);
1200 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1207 static int mdd_recovery_complete(const struct lu_env *env,
1208 struct lu_device *d)
1210 struct mdd_device *mdd = lu2mdd_dev(d);
1211 struct lu_device *next;
1215 LASSERT(mdd != NULL);
1216 next = &mdd->mdd_child->dd_lu_dev;
1218 if (!mdd->mdd_bottom->dd_rdonly)
1219 mdd_orphan_cleanup(env, mdd);
1220 rc = next->ld_ops->ldo_recovery_complete(env, next);
1225 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1226 const struct lu_fid *pfid, const char *name,
1227 __u32 mode, struct lu_fid *fid)
1229 struct dt_object *parent, *dto;
1234 LASSERT(!fid_is_zero(pfid));
1235 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1236 if (unlikely(IS_ERR(parent)))
1237 RETURN(PTR_ERR(parent));
1239 /* create local file/dir, if @fid is passed then try to use it */
1240 if (fid_is_zero(fid))
1241 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1244 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1248 GOTO(out_put, rc = PTR_ERR(dto));
1249 *fid = *lu_object_fid(&dto->do_lu);
1250 /* since stack is not fully set up the local_storage uses own stack
1251 * and we should drop its object from cache */
1252 dt_object_put_nocache(env, dto);
1255 dt_object_put(env, parent);
1259 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1260 enum lfsck_events event)
1265 static int mdd_prepare(const struct lu_env *env,
1266 struct lu_device *pdev,
1267 struct lu_device *cdev)
1269 struct mdd_device *mdd = lu2mdd_dev(cdev);
1270 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1271 struct nm_config_file *nodemap_config;
1272 struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
1273 struct dt_object *root = NULL;
1274 struct thandle *th = NULL;
1280 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1284 /* Setup local dirs */
1285 fid.f_seq = FID_SEQ_LOCAL_NAME;
1288 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1293 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1298 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1299 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1300 mdd_root_dir_name, S_IFDIR |
1301 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1303 CERROR("%s: create root fid failed: rc = %d\n",
1304 mdd2obd_dev(mdd)->obd_name, rc);
1308 /* store a default directory layout on the root directory if
1309 * it doesn't already exist to improve MDT space balance.
1311 root = dt_locate(env, mdd->mdd_bottom, &fid);
1312 if (unlikely(IS_ERR(root)))
1313 GOTO(out_los, rc = PTR_ERR(root));
1315 rc = dt_xattr_get(env, root, &LU_BUF_NULL,
1316 XATTR_NAME_DEFAULT_LMV);
1317 if (rc == -ENODATA) {
1319 struct lmv_user_md lmv_default = {
1320 .lum_magic = LMV_USER_MAGIC,
1321 .lum_stripe_count = 1,
1322 .lum_stripe_offset = LMV_OFFSET_DEFAULT,
1325 th = dt_trans_create(env, mdd->mdd_bottom);
1327 GOTO(out_root_put, rc = PTR_ERR(th));
1329 buf.lb_buf = &lmv_default;
1330 buf.lb_len = sizeof(lmv_default);
1331 rc = dt_declare_xattr_set(env, root, &buf,
1332 XATTR_NAME_DEFAULT_LMV, 0,
1335 GOTO(out_trans_stop, rc);
1337 rc = dt_trans_start_local(env, mdd->mdd_bottom, th);
1339 GOTO(out_trans_stop, rc);
1341 rc = dt_xattr_set(env, root, &buf,
1342 XATTR_NAME_DEFAULT_LMV, 0, th);
1344 GOTO(out_trans_stop, rc);
1346 dt_trans_stop(env, mdd->mdd_bottom, th);
1348 } else if (rc < 0 && rc != -ERANGE) {
1349 CERROR("%s: get default LMV of root failed: rc = %d\n",
1350 mdd2obd_dev(mdd)->obd_name, rc);
1352 GOTO(out_root_put, rc);
1355 dt_object_put(env, root);
1358 mdd->mdd_root_fid = fid;
1359 rc = mdd_dot_lustre_setup(env, mdd);
1361 CERROR("%s: initializing .lustre failed: rc = %d\n",
1362 mdd2obd_dev(mdd)->obd_name, rc);
1366 /* Normal client usually send root access to MDT0 directly,
1367 * the root FID on non-MDT0 will only be used by echo client. */
1368 mdd->mdd_root_fid = fid;
1371 rc = mdd_orphan_index_init(env, mdd);
1375 rc = mdd_changelog_init(env, mdd);
1377 CERROR("%s: failed to initialize changelog: rc = %d\n",
1378 mdd2obd_dev(mdd)->obd_name, rc);
1382 rc = mdd_hsm_actions_llog_init(env, mdd);
1384 GOTO(out_changelog, rc);
1386 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1388 if (IS_ERR(nodemap_config)) {
1389 rc = PTR_ERR(nodemap_config);
1393 obt->obt_nodemap_config_file = nodemap_config;
1396 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1397 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1400 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1401 mdd2obd_dev(mdd)->obd_name, rc);
1402 GOTO(out_nodemap, rc);
1405 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1407 CERROR("%s: failed to register to barrier: rc = %d\n",
1408 mdd2obd_dev(mdd)->obd_name, rc);
1409 GOTO(out_lfsck, rc);
1415 lfsck_degister(env, mdd->mdd_bottom);
1417 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1418 obt->obt_nodemap_config_file = NULL;
1420 mdd_hsm_actions_llog_fini(env, mdd);
1422 mdd_changelog_fini(env, mdd);
1424 mdd_orphan_index_fini(env, mdd);
1426 if (mdd_seq_site(mdd)->ss_node_id == 0)
1427 mdd_dot_lustre_cleanup(env, mdd);
1430 dt_trans_stop(env, mdd->mdd_bottom, th);
1433 dt_object_put(env, root);
1435 local_oid_storage_fini(env, mdd->mdd_los);
1436 mdd->mdd_los = NULL;
1442 * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
1444 * Find corresponding device by passed parent and name, and allocate FID from
1447 * see include/lu_object.h for the details.
1449 static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1450 struct lu_fid *fid, struct lu_object *parent,
1451 const struct lu_name *name)
1453 struct mdd_device *mdd = lu2mdd_dev(d);
1454 struct lu_object *o = lu_object_next(parent);
1456 return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
1459 const struct lu_device_operations mdd_lu_ops = {
1460 .ldo_object_alloc = mdd_object_alloc,
1461 .ldo_process_config = mdd_process_config,
1462 .ldo_recovery_complete = mdd_recovery_complete,
1463 .ldo_prepare = mdd_prepare,
1464 .ldo_fid_alloc = mdd_fid_alloc,
1467 static int mdd_root_get(const struct lu_env *env,
1468 struct md_device *m, struct lu_fid *f)
1470 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1473 *f = mdd->mdd_root_fid;
1478 * No permission check is needed.
1480 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1481 struct obd_statfs *sfs)
1483 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1488 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs, NULL);
1490 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1495 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1496 struct md_device *m)
1498 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1500 return &mdd->mdd_dt_conf;
1503 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1506 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1508 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1509 return (*h == NULL ? -ENOENT : 0);
1512 static struct lu_device *mdd_device_free(const struct lu_env *env,
1513 struct lu_device *lu)
1515 struct mdd_device *m = lu2mdd_dev(lu);
1518 LASSERT(atomic_read(&lu->ld_ref) == 0);
1519 md_device_fini(&m->mdd_md_dev);
1524 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1525 struct lu_device_type *t,
1526 struct lustre_cfg *lcfg)
1528 struct lu_device *l;
1529 struct mdd_device *m;
1533 l = ERR_PTR(-ENOMEM);
1538 md_device_init(&m->mdd_md_dev, t);
1539 rc = mdd_init0(env, m, t, lcfg);
1541 mdd_device_free(env, l);
1550 * we use exports to track all mdd users
1552 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1553 struct obd_device *obd, struct obd_uuid *cluuid,
1554 struct obd_connect_data *data, void *localdata)
1556 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1557 struct lustre_handle conn;
1561 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1563 rc = class_connect(&conn, obd, cluuid);
1567 *exp = class_conn2export(&conn);
1569 /* Why should there ever be more than 1 connect? */
1570 LASSERT(mdd->mdd_connects == 0);
1571 mdd->mdd_connects++;
1577 * once last export (we don't count self-export) disappeared
1578 * mdd can be released
1580 static int mdd_obd_disconnect(struct obd_export *exp)
1582 struct obd_device *obd = exp->exp_obd;
1583 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1584 int rc, release = 0;
1587 mdd->mdd_connects--;
1588 if (mdd->mdd_connects == 0)
1591 rc = class_disconnect(exp);
1593 if (rc == 0 && release)
1594 class_manual_cleanup(obd);
1598 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1599 __u32 keylen, void *key, __u32 *vallen, void *val)
1603 if (KEY_IS(KEY_OSP_CONNECTED)) {
1604 struct obd_device *obd = exp->exp_obd;
1605 struct mdd_device *mdd;
1607 if (!obd->obd_set_up || obd->obd_stopping)
1610 mdd = lu2mdd_dev(obd->obd_lu_dev);
1612 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1620 static int mdd_obd_set_info_async(const struct lu_env *env,
1621 struct obd_export *exp,
1622 __u32 keylen, void *key,
1623 __u32 vallen, void *val,
1624 struct ptlrpc_request_set *set)
1626 struct obd_device *obd = exp->exp_obd;
1627 struct mdd_device *mdd;
1630 if (!obd->obd_set_up || obd->obd_stopping)
1633 mdd = lu2mdd_dev(obd->obd_lu_dev);
1635 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1640 static const struct obd_ops mdd_obd_device_ops = {
1641 .o_owner = THIS_MODULE,
1642 .o_connect = mdd_obd_connect,
1643 .o_disconnect = mdd_obd_disconnect,
1644 .o_get_info = mdd_obd_get_info,
1645 .o_set_info_async = mdd_obd_set_info_async,
1648 struct mdd_changelog_name_check_data {
1649 const char *mcnc_name;
1654 * changelog_recalc_mask callback
1656 * Is is called per each registered user and calculates combined mask of
1657 * all registered users.
1659 static int mdd_changelog_name_check_cb(const struct lu_env *env,
1660 struct llog_handle *llh,
1661 struct llog_rec_hdr *hdr, void *data)
1663 struct llog_changelog_user_rec2 *rec;
1664 struct mdd_changelog_name_check_data *mcnc = data;
1666 rec = container_of(hdr, typeof(*rec), cur_hdr);
1667 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
1668 !strncmp(rec->cur_name, mcnc->mcnc_name, sizeof(rec->cur_name))) {
1669 mcnc->mcnc_id = rec->cur_id;
1675 static int mdd_changelog_name_check(const struct lu_env *env,
1676 struct llog_ctxt *ctxt,
1677 struct mdd_device *mdd, const char *name)
1679 struct mdd_changelog_name_check_data mcnc = { .mcnc_name = name, };
1685 /* first symbol is a letter */
1686 if (!isalpha(name[0])) {
1688 CERROR("%s: first char '%c' in '%s' is not letter: rc = %d\n",
1689 mdd2obd_dev(mdd)->obd_name, name[0], name, rc);
1693 /* name is valid: contains letters, numbers and '-', '_' only */
1694 while (name[++chr]) {
1695 if (!(isalnum(name[chr]) || name[chr] == '_' ||
1696 name[chr] == '-')) {
1698 CERROR("%s: wrong char '%c' in name '%s': rc = %d\n",
1699 mdd2obd_dev(mdd)->obd_name, name[chr], name, rc);
1704 if (chr > CHANGELOG_USER_NAMELEN) {
1706 CERROR("%s: name '%s' is over %d symbols limit: rc = %d\n",
1707 mdd2obd_dev(mdd)->obd_name, name,
1708 CHANGELOG_USER_NAMELEN, rc);
1712 rc = llog_cat_process(env, ctxt->loc_handle,
1713 mdd_changelog_name_check_cb, &mcnc, 0, 0);
1715 CWARN("%s: changelog name %s exists already: rc = %d\n",
1716 mdd2obd_dev(mdd)->obd_name, name, rc);
1718 CWARN("%s: failed user changelog processing: rc = %d\n",
1719 mdd2obd_dev(mdd)->obd_name, rc);
1723 static int mdd_changelog_user_register(const struct lu_env *env,
1724 struct mdd_device *mdd, int *id,
1725 const char *name, const char *mask)
1727 struct llog_ctxt *ctxt;
1728 struct llog_changelog_user_rec2 *rec;
1729 char user_name[CHANGELOG_USER_NAMELEN_FULL];
1734 ctxt = llog_get_context(mdd2obd_dev(mdd),
1735 LLOG_CHANGELOG_USER_ORIG_CTXT);
1741 llog_ctxt_put(ctxt);
1745 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1747 rec->cur_hdr.lrh_len = sizeof(*rec);
1748 /* keep old record type for users without mask/name for
1749 * compatibility needs
1751 if (mask || (name && name[0]))
1752 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
1754 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1755 spin_lock(&mdd->mdd_cl.mc_user_lock);
1756 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1757 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1759 CERROR("%s: registering %s user: max ID is exceeded: rc = %d\n",
1760 mdd2obd_dev(mdd)->obd_name,
1761 (name && name[0]) ? name : "new", rc);
1764 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1765 mdd->mdd_cl.mc_users++;
1766 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1768 rec->cur_time = (__u32)ktime_get_real_seconds();
1769 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1772 spin_lock(&mdd->mdd_cl.mc_lock);
1773 rec->cur_endrec = mdd->mdd_cl.mc_index;
1774 spin_unlock(&mdd->mdd_cl.mc_lock);
1777 /* if user will use relative mask apply it on default one */
1778 rec->cur_mask = CHANGELOG_DEFMASK;
1779 rc = cfs_str2mask(mask, changelog_type2str, &rec->cur_mask,
1780 CHANGELOG_MINMASK, CHANGELOG_ALLMASK);
1782 GOTO(out_users, rc);
1785 if (name && name[0]) {
1786 rc = mdd_changelog_name_check(env, ctxt, mdd, name);
1788 GOTO(out_users, rc);
1789 strlcpy(rec->cur_name, name, sizeof(rec->cur_name));
1791 mdd_chlg_username(rec, user_name, sizeof(user_name));
1793 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1795 CWARN("%s: failed to register changelog user %s: rc = %d\n",
1796 mdd2obd_dev(mdd)->obd_name, user_name, rc);
1797 GOTO(out_users, rc);
1800 /* apply user mask finally */
1801 spin_lock(&mdd->mdd_cl.mc_user_lock);
1802 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
1803 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1805 CDEBUG(D_IOCTL, "%s: registered changelog user '%s', mask %#x\n",
1806 mdd2obd_dev(mdd)->obd_name, user_name, rec->cur_mask);
1808 /* Assume we want it on since somebody registered */
1809 rc = mdd_changelog_on(env, mdd);
1811 /* record is added, so don't decrement users on error */
1815 spin_lock(&mdd->mdd_cl.mc_user_lock);
1816 mdd->mdd_cl.mc_users--;
1817 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1821 llog_ctxt_put(ctxt);
1825 struct mdd_changelog_recalc_mask_data {
1826 struct mdd_device *mcrm_mdd;
1831 * changelog_recalc_mask callback
1833 * Is is called per each registered user and calculates combined mask of
1834 * all registered users.
1836 static int mdd_changelog_recalc_mask_cb(const struct lu_env *env,
1837 struct llog_handle *llh,
1838 struct llog_rec_hdr *hdr, void *data)
1840 struct llog_changelog_user_rec2 *rec;
1841 struct mdd_changelog_recalc_mask_data *mcrm = data;
1843 rec = container_of(hdr, typeof(*rec), cur_hdr);
1844 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
1845 mcrm->mcrm_mask |= rec->cur_mask;
1846 else if (mcrm->mcrm_mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
1847 mcrm->mcrm_mask |= CHANGELOG_DEFMASK;
1852 int mdd_changelog_recalc_mask(const struct lu_env *env, struct mdd_device *mdd)
1854 struct llog_ctxt *ctxt;
1855 struct mdd_changelog_recalc_mask_data mcrm = {
1857 .mcrm_mask = mdd->mdd_cl.mc_proc_mask,
1863 ctxt = llog_get_context(mdd2obd_dev(mdd),
1864 LLOG_CHANGELOG_USER_ORIG_CTXT);
1868 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1869 GOTO(out, rc = -ENXIO);
1871 rc = llog_cat_process(env, ctxt->loc_handle,
1872 mdd_changelog_recalc_mask_cb, &mcrm, 0, 0);
1874 CWARN("%s: failed user changelog processing: rc = %d\n",
1875 mdd2obd_dev(mdd)->obd_name, rc);
1877 spin_lock(&mdd->mdd_cl.mc_user_lock);
1878 CDEBUG(D_INFO, "%s: recalc changelog mask: %#x -> %#x\n",
1879 mdd2obd_dev(mdd)->obd_name, mdd->mdd_cl.mc_current_mask,
1881 mdd->mdd_cl.mc_current_mask = mcrm.mcrm_mask;
1882 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1886 llog_ctxt_put(ctxt);
1891 struct mdd_changelog_user_purge {
1892 struct mdd_device *mcup_mdd;
1894 __u32 mcup_usercount;
1900 * changelog_user_purge callback
1902 * Is called once per user.
1904 * Check to see the user requested is available from rec.
1905 * Truncate the changelog.
1906 * Keep track of the total number of users (calls).
1908 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1909 struct llog_handle *llh,
1910 struct llog_rec_hdr *hdr, void *data)
1912 struct llog_changelog_user_rec2 *rec;
1913 struct mdd_changelog_user_purge *mcup = data;
1914 struct llog_cookie cookie;
1919 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1922 rec = container_of(hdr, typeof(*rec), cur_hdr);
1924 mcup->mcup_usercount++;
1926 if (rec->cur_id != mcup->mcup_id) {
1927 /* truncate to the lowest endrec that is not this user */
1928 mcup->mcup_minrec = min(mcup->mcup_minrec, rec->cur_endrec);
1932 /* Unregister this user */
1933 cookie.lgc_lgl = llh->lgh_id;
1934 cookie.lgc_index = hdr->lrh_index;
1936 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1939 mcup->mcup_found = true;
1940 mcup->mcup_usercount--;
1941 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1942 mcup->mcup_mdd->mdd_cl.mc_users--;
1943 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1949 int mdd_changelog_user_purge(const struct lu_env *env,
1950 struct mdd_device *mdd, __u32 id)
1952 struct mdd_changelog_user_purge mcup = {
1955 .mcup_found = false,
1956 .mcup_usercount = 0,
1957 .mcup_minrec = ULLONG_MAX,
1959 struct llog_ctxt *ctxt;
1964 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1965 mdd2obd_dev(mdd)->obd_name, id);
1967 ctxt = llog_get_context(mdd2obd_dev(mdd),
1968 LLOG_CHANGELOG_USER_ORIG_CTXT);
1970 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1971 GOTO(out, rc = -ENXIO);
1973 rc = llog_cat_process(env, ctxt->loc_handle,
1974 mdd_changelog_user_purge_cb, &mcup,
1977 OBD_FAIL_TIMEOUT(OBD_FAIL_LLOG_PURGE_DELAY, cfs_fail_val);
1979 if ((rc == 0) && (mcup.mcup_usercount == 0)) {
1980 spin_lock(&mdd->mdd_cl.mc_user_lock);
1981 if (mdd->mdd_cl.mc_users == 0) {
1982 /* No more users; turn changelogs off */
1983 CDEBUG(D_IOCTL, "turning off changelogs\n");
1984 rc = mdd_changelog_off(env, mdd);
1986 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1989 if ((rc == 0) && mcup.mcup_found) {
1990 CDEBUG(D_IOCTL, "%s: Purging changelog entries for user %d "
1992 mdd2obd_dev(mdd)->obd_name, id, mcup.mcup_minrec);
1993 /* Cancelling record 0 destroys the entire changelog, make sure
1994 we don't do that unless we mean it. */
1995 if (mcup.mcup_minrec != 0 || mcup.mcup_usercount == 0) {
1996 rc = mdd_changelog_llog_cancel(env, mdd,
2000 CWARN("%s: No changelog for user %u; rc=%d\n",
2001 mdd2obd_dev(mdd)->obd_name, id, rc);
2002 GOTO(out, rc = -ENOENT);
2005 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
2010 llog_ctxt_put(ctxt);
2015 struct mdd_changelog_user_clear {
2020 struct mdd_device *mcuc_mdd;
2024 * changelog_clear callback
2026 * Is called once per user.
2028 * Check to see the user requested is available from rec.
2029 * Check the oldest (smallest) record for boundary conditions.
2030 * Truncate the changelog.
2032 static int mdd_changelog_clear_cb(const struct lu_env *env,
2033 struct llog_handle *llh,
2034 struct llog_rec_hdr *hdr,
2037 struct llog_changelog_user_rec2 *rec;
2038 struct mdd_changelog_user_clear *mcuc = data;
2039 char user_name[CHANGELOG_USER_NAMELEN_FULL];
2040 struct mdd_device *mdd = mcuc->mcuc_mdd;
2045 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
2048 rec = container_of(hdr, typeof(*rec), cur_hdr);
2049 /* Does the changelog id match the requested id? */
2050 if (rec->cur_id != mcuc->mcuc_id) {
2051 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec,
2056 /* cur_endrec is the oldest purgeable record, make sure we're newer */
2057 if (rec->cur_endrec > mcuc->mcuc_endrec) {
2060 "%s: request %llu > endrec %llu for user %s: rc = %d\n",
2061 mdd2obd_dev(mdd)->obd_name,
2062 mcuc->mcuc_endrec, rec->cur_endrec,
2063 mdd_chlg_username(rec, user_name, sizeof(user_name)),
2068 /* Flag that we've met all the range and user checks.
2069 * We now know the record to flush.
2071 rec->cur_endrec = mcuc->mcuc_endrec;
2073 rec->cur_time = (__u32)ktime_get_real_seconds();
2074 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
2077 mcuc->mcuc_flush = true;
2079 CDEBUG(D_IOCTL, "%s: rewriting changelog user %s endrec = %llu\n",
2080 mdd2obd_dev(mdd)->obd_name,
2081 mdd_chlg_username(rec, user_name, sizeof(user_name)),
2084 /* Update the endrec */
2085 rc = llog_write(env, llh, hdr, hdr->lrh_index);
2091 * Clear a changelog up to entry specified by endrec for user id.
2093 static int mdd_changelog_clear(const struct lu_env *env,
2094 struct mdd_device *mdd, __u32 id,
2097 struct mdd_changelog_user_clear mcuc = {
2099 .mcuc_minrec = endrec,
2100 .mcuc_flush = false,
2103 struct llog_ctxt *ctxt;
2109 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
2110 mdd2obd_dev(mdd)->obd_name, id, endrec);
2111 /* start_rec is the newest (largest value) entry in the changelogs*/
2112 spin_lock(&mdd->mdd_cl.mc_lock);
2113 start_rec = mdd->mdd_cl.mc_index;
2114 spin_unlock(&mdd->mdd_cl.mc_lock);
2116 if (start_rec < endrec) {
2117 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
2118 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
2123 mcuc.mcuc_endrec = start_rec;
2124 mcuc.mcuc_minrec = ULLONG_MAX;
2126 mcuc.mcuc_endrec = endrec;
2129 ctxt = llog_get_context(mdd2obd_dev(mdd),
2130 LLOG_CHANGELOG_USER_ORIG_CTXT);
2132 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
2133 GOTO(out, rc = -ENXIO);
2135 rc = llog_cat_process(env, ctxt->loc_handle,
2136 mdd_changelog_clear_cb, (void *)&mcuc,
2139 if (rc == -EINVAL) {
2140 CDEBUG(D_IOCTL, "%s: No changelog recnum <= %llu to clear\n",
2141 mdd2obd_dev(mdd)->obd_name, (unsigned long long) endrec);
2143 } else if (rc < 0) {
2144 CWARN("%s: Failure to clear the changelog for user %d: %d\n",
2145 mdd2obd_dev(mdd)->obd_name, id, rc);
2146 } else if (mcuc.mcuc_flush) {
2147 /* Cancelling record 0 destroys the entire changelog, make sure
2148 we don't do that unless we mean it. */
2149 if (mcuc.mcuc_minrec != 0) {
2150 CDEBUG(D_IOCTL, "%s: Purging changelog entries up "\
2151 "to %llu\n", mdd2obd_dev(mdd)->obd_name,
2154 rc = mdd_changelog_llog_cancel(env, mdd,
2158 CDEBUG(D_IOCTL, "%s: No entry for user %d\n",
2159 mdd2obd_dev(mdd)->obd_name, id);
2166 llog_ctxt_put(ctxt);
2171 static int mdd_changelog_user_deregister(const struct lu_env *env,
2172 struct mdd_device *mdd, int *id,
2175 struct llog_ctxt *ctxt;
2176 struct mdd_changelog_name_check_data mcnc = {
2185 ctxt = llog_get_context(mdd2obd_dev(mdd),
2186 LLOG_CHANGELOG_USER_ORIG_CTXT);
2190 rc = llog_cat_process(env, ctxt->loc_handle,
2191 mdd_changelog_name_check_cb, &mcnc, 0, 0);
2192 llog_ctxt_put(ctxt);
2194 if (rc != -EEXIST) {
2195 CDEBUG(D_IOCTL, "%s: no entry for username %s\n",
2196 mdd2obd_dev(mdd)->obd_name, name);
2202 /* explicitly clear changelog first, to protect from crash in
2203 * the middle of purge that would lead to unregistered consumer
2204 * but pending changelog entries
2206 rc = mdd_changelog_clear(env, mdd, *id, 0);
2208 rc = mdd_changelog_user_purge(env, mdd, *id);
2210 /* recalc changelog current mask */
2211 mdd_changelog_recalc_mask(env, mdd);
2217 * May be called remotely from mdt_iocontrol_handle or locally from
2218 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
2219 * an obd_ioctl_data format (but local ioctl handler does).
2221 * \param len - data len
2222 * \param karg - ioctl data, in kernel space
2224 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
2225 unsigned int cmd, int len, void *karg)
2227 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
2228 struct obd_ioctl_data *data = karg;
2232 /* Doesn't use obd_ioctl_data */
2234 case OBD_IOC_CHANGELOG_CLEAR: {
2235 struct changelog_setinfo *cs = karg;
2237 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2238 RETURN(-EINPROGRESS);
2240 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
2242 barrier_exit(mdd->mdd_bottom);
2245 case OBD_IOC_START_LFSCK: {
2246 rc = lfsck_start(env, mdd->mdd_bottom,
2247 (struct lfsck_start_param *)karg);
2250 case OBD_IOC_STOP_LFSCK: {
2251 rc = lfsck_stop(env, mdd->mdd_bottom,
2252 (struct lfsck_stop *)karg);
2255 case OBD_IOC_QUERY_LFSCK: {
2256 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
2257 (struct lfsck_query *)karg);
2262 /* Below ioctls use obd_ioctl_data */
2263 if (data->ioc_version != OBD_IOCTL_VERSION) {
2264 CERROR("Bad magic %x != %x\n", data->ioc_version,
2270 case OBD_IOC_CHANGELOG_REG:
2271 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2272 RETURN(-EINPROGRESS);
2274 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1,
2277 barrier_exit(mdd->mdd_bottom);
2279 case OBD_IOC_CHANGELOG_DEREG:
2280 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2281 RETURN(-EINPROGRESS);
2283 rc = mdd_changelog_user_deregister(env, mdd, &data->ioc_u32_1,
2286 barrier_exit(mdd->mdd_bottom);
2295 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
2296 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
2298 static const struct md_device_operations mdd_ops = {
2299 .mdo_statfs = mdd_statfs,
2300 .mdo_root_get = mdd_root_get,
2301 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
2302 .mdo_iocontrol = mdd_iocontrol,
2303 .mdo_dtconf_get = mdd_dtconf_get,
2306 static const struct lu_device_type_operations mdd_device_type_ops = {
2307 .ldto_init = mdd_type_init,
2308 .ldto_fini = mdd_type_fini,
2310 .ldto_start = mdd_type_start,
2311 .ldto_stop = mdd_type_stop,
2313 .ldto_device_alloc = mdd_device_alloc,
2314 .ldto_device_free = mdd_device_free,
2316 .ldto_device_fini = mdd_device_fini
2319 static struct lu_device_type mdd_device_type = {
2320 .ldt_tags = LU_DEVICE_MD,
2321 .ldt_name = LUSTRE_MDD_NAME,
2322 .ldt_ops = &mdd_device_type_ops,
2323 .ldt_ctx_tags = LCT_MD_THREAD
2326 /* context key constructor: mdd_key_init */
2327 LU_KEY_INIT(mdd, struct mdd_thread_info);
2329 static void mdd_key_fini(const struct lu_context *ctx,
2330 struct lu_context_key *key, void *data)
2332 struct mdd_thread_info *info = data;
2334 lu_buf_free(&info->mdi_big_buf);
2335 lu_buf_free(&info->mdi_link_buf);
2336 lu_buf_free(&info->mdi_xattr_buf);
2337 lu_buf_free(&info->mdi_chlg_buf);
2342 /* context key: mdd_thread_key */
2343 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
2345 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
2346 int (*func)(void *), void *data, char *name)
2348 struct task_struct *task;
2350 LASSERT(thread->mgt_init == false);
2351 init_completion(&thread->mgt_started);
2352 init_completion(&thread->mgt_finished);
2353 thread->mgt_data = data;
2354 thread->mgt_abort = false;
2355 thread->mgt_init = true;
2357 task = kthread_run(func, thread, name);
2359 complete(&thread->mgt_finished);
2360 return PTR_ERR(task);
2362 wait_for_completion(&thread->mgt_started);
2366 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
2368 if (thread->mgt_init == true) {
2369 thread->mgt_abort = true;
2370 wait_for_completion(&thread->mgt_finished);
2374 static int __init mdd_init(void)
2378 rc = lu_kmem_init(mdd_caches);
2382 changelog_orig_logops = llog_common_cat_ops;
2383 changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
2385 rc = class_register_type(&mdd_obd_device_ops, NULL, false,
2386 LUSTRE_MDD_NAME, &mdd_device_type);
2388 lu_kmem_fini(mdd_caches);
2392 static void __exit mdd_exit(void)
2394 class_unregister_type(LUSTRE_MDD_NAME);
2395 lu_kmem_fini(mdd_caches);
2398 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2399 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2400 MODULE_VERSION(LUSTRE_VERSION_STRING);
2401 MODULE_LICENSE("GPL");
2403 module_init(mdd_init);
2404 module_exit(mdd_exit);