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 mdd->mdd_cl.mc_mintime = min(mdd->mdd_cl.mc_mintime, rec->cur_time);
260 mdd->mdd_cl.mc_minrec = min(mdd->mdd_cl.mc_minrec, rec->cur_endrec);
261 spin_unlock(&mdd->mdd_cl.mc_user_lock);
262 spin_lock(&mdd->mdd_cl.mc_lock);
263 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
264 mdd->mdd_cl.mc_index = rec->cur_endrec;
265 spin_unlock(&mdd->mdd_cl.mc_lock);
267 return LLOG_PROC_BREAK;
270 struct changelog_orphan_data {
272 struct mdd_device *clod_mdd;
275 /* find oldest changelog record index */
276 static int changelog_detect_orphan_cb(const struct lu_env *env,
277 struct llog_handle *llh,
278 struct llog_rec_hdr *hdr, void *data)
280 struct changelog_orphan_data *clod = data;
281 struct mdd_device *mdd = clod->clod_mdd;
282 struct llog_changelog_rec *rec;
284 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
286 rec = container_of(hdr, typeof(*rec), cr_hdr);
287 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
288 CWARN("%s: invalid record at index %d in log "DFID"\n",
289 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
290 PFID(&llh->lgh_id.lgl_oi.oi_fid));
291 /* try to find some next valid record and thus allow to recover
292 * from a corrupted LLOG, instead to assert and force a crash
298 "%s: record at index %d/%d/%llu t=%x %.*s in log "DFID"\n",
299 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
300 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
301 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
302 PFID(&llh->lgh_id.lgl_oi.oi_fid));
304 clod->clod_index = rec->cr.cr_index;
306 return LLOG_PROC_BREAK;
309 /* find oldest changelog user index */
310 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
311 struct llog_handle *llh,
312 struct llog_rec_hdr *hdr, void *data)
314 struct changelog_orphan_data *clod = data;
315 struct mdd_device *mdd = clod->clod_mdd;
316 struct llog_changelog_user_rec2 *rec;
317 char user_name[CHANGELOG_USER_NAMELEN_FULL];
319 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
321 rec = container_of(hdr, typeof(*rec), cur_hdr);
322 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
323 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
324 CWARN("%s: unknown user type %u at index %u in log "DFID"\n",
325 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
326 rec->cur_hdr.lrh_type, PFID(&llh->lgh_id.lgl_oi.oi_fid));
327 /* try to find some next valid record and thus allow to recover
328 * from a corrupted LLOG, instead to assert and force a crash
333 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
334 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
336 hdr->lrh_index, rec->cur_hdr.lrh_index,
337 rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
339 clod->clod_index = min_t(__u64, clod->clod_index, rec->cur_endrec);
344 struct changelog_cancel_cookie {
346 struct mdd_device *mdd;
349 static int llog_changelog_cancel_cb(const struct lu_env *env,
350 struct llog_handle *llh,
351 struct llog_rec_hdr *hdr, void *data)
353 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
354 struct changelog_cancel_cookie *cl_cookie =
355 (struct changelog_cancel_cookie *)data;
359 /* This is always a (sub)log, not the catalog */
360 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
362 /* if current context is GC-thread allow it to stop upon umount
363 * remaining records cleanup will occur upon next mount
365 * also during testing, wait for GC-thread to be released
367 * XXX this requires the GC-thread to not fork a sub-thread via
368 * llog[_cat]_process_or_fork() and we may think to also implement
369 * this shutdown mechanism for manually started user unregister which
370 * can also take a long time if huge backlog of records
372 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
373 /* wait to be released */
374 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
377 if (kthread_should_stop())
378 RETURN(LLOG_PROC_BREAK);
381 if (rec->cr.cr_index > cl_cookie->endrec)
382 /* records are in order, so we're done */
383 RETURN(LLOG_PROC_BREAK);
385 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
386 if (cfs_fail_val == 0)
387 cfs_fail_val = hdr->lrh_index;
388 if (cfs_fail_val == hdr->lrh_index)
389 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
392 /* Records folow one by one, cr_index++. We could calculate the
393 * last cr_index at this plain llog. And if it less then cookie endrec
394 * cancel the whole file.
396 if ((LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr) - hdr->lrh_index +
397 rec->cr.cr_index) < cl_cookie->endrec) {
400 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
401 if (cfs_fail_val == 0) {
402 cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
403 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
406 rc = llog_destroy(env, llh);
408 CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
409 PFID(&llh->lgh_id.lgl_oi.oi_fid));
410 RETURN(LLOG_DEL_PLAIN);
414 /* cancel them one at a time. I suppose we could store up the cookies
415 * and cancel them all at once; probably more efficient, but this is
416 * done as a user call, so who cares... */
418 RETURN(LLOG_DEL_RECORD);
421 static int llog_changelog_cancel(const struct lu_env *env,
422 struct llog_ctxt *ctxt,
423 struct changelog_cancel_cookie *cookie)
425 struct llog_handle *cathandle = ctxt->loc_handle;
430 /* This should only be called with the catalog handle */
431 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
433 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
436 /* 0 or 1 means we're done */
439 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
440 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
441 PFID(&cathandle->lgh_id.lgl_oi.oi_fid), rc);
446 static struct llog_operations changelog_orig_logops;
449 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
453 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
457 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
460 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
461 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
462 CERROR("Changelogs cannot be enabled due to error "
463 "condition (see %s log).\n",
464 mdd2obd_dev(mdd)->obd_name);
467 spin_lock(&mdd->mdd_cl.mc_lock);
468 mdd->mdd_cl.mc_flags |= CLM_ON;
469 spin_unlock(&mdd->mdd_cl.mc_lock);
470 rc = mdd_changelog_write_header(env, mdd, CLM_START);
476 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
480 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
483 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
484 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
485 spin_lock(&mdd->mdd_cl.mc_lock);
486 mdd->mdd_cl.mc_flags &= ~CLM_ON;
487 spin_unlock(&mdd->mdd_cl.mc_lock);
492 static int mdd_changelog_llog_init(const struct lu_env *env,
493 struct mdd_device *mdd)
495 struct obd_device *obd = mdd2obd_dev(mdd);
496 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
497 struct changelog_orphan_data clod = {
508 /* LU-2844 mdd setup failure should not cause umount oops */
509 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
512 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
513 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
514 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
515 obd, &changelog_orig_logops);
517 CERROR("%s: changelog llog setup failed: rc = %d\n",
522 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
525 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
528 GOTO(out_cleanup, rc);
530 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
534 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
535 changelog_init_cb, mdd);
538 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
543 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
544 mdd->mdd_cl.mc_index);
546 /* setup user changelog */
547 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
548 obd, &changelog_orig_logops);
550 CERROR("%s: changelog users llog setup failed: rc = %d\n",
555 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
558 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
561 GOTO(out_ucleanup, rc);
563 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
565 GOTO(out_uclose, rc);
567 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
568 changelog_user_init_cb, mdd);
570 CERROR("%s: changelog user init failed: rc = %d\n",
572 GOTO(out_uclose, rc);
575 /* Finally apply per-server mask */
576 mdd->mdd_cl.mc_current_mask |= mdd->mdd_cl.mc_proc_mask;
578 /* If we have registered users, assume we want changelogs on */
579 if (mdd->mdd_cl.mc_lastuser > 0) {
580 rc = mdd_changelog_on(env, mdd);
582 GOTO(out_uclose, rc);
585 /* find and clear any orphan changelog records (1st record index <
586 * smallest of all users current index), likely to come from an
587 * interrupted manual or GC-thread purge, as its user record had
589 * XXX we may wait for a still registered user clear operation to
590 * do the job, but it may then take a long time to reach the user's
591 * real targetted records if a huge purge backlog is still to be
592 * processed as a long time idle user record could have been deleted
593 * XXX we may need to run end of purge as a separate thread
595 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
598 CERROR("%s: changelog detect orphan failed: rc = %d\n",
600 GOTO(out_uclose, rc);
602 rc = llog_cat_process(env, uctxt->loc_handle,
603 changelog_user_detect_orphan_cb,
606 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
608 GOTO(out_uclose, rc);
610 if (unlikely(clod.clod_index < user_orphan.clod_index)) {
611 struct changelog_cancel_cookie cl_cookie = {
612 .endrec = user_orphan.clod_index,
616 CWARN("%s : orphan changelog records found, starting from "
617 "index %llu to index %llu, being cleared now\n",
618 obd->obd_name, clod.clod_index, user_orphan.clod_index);
620 /* XXX we may need to run end of purge as a separate thread */
621 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
623 CERROR("%s: purge of changelog orphan records failed: "
624 "rc = %d\n", obd->obd_name, rc);
625 GOTO(out_uclose, rc);
630 llog_ctxt_put(uctxt);
633 llog_cat_close(env, uctxt->loc_handle);
635 llog_cleanup(env, uctxt);
637 llog_cat_close(env, ctxt->loc_handle);
639 llog_cleanup(env, ctxt);
643 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
645 struct obd_device *obd = mdd2obd_dev(mdd);
648 mdd->mdd_cl.mc_index = 0;
649 spin_lock_init(&mdd->mdd_cl.mc_lock);
650 mdd->mdd_cl.mc_starttime = ktime_get();
651 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
652 mdd->mdd_cl.mc_lastuser = 0;
654 /* ensure a GC check will, and a thread run may, occur upon start */
655 mdd->mdd_cl.mc_gc_time = 0;
656 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
657 mdd->mdd_cl.mc_mintime = (__u32)ktime_get_real_seconds();
658 mdd->mdd_cl.mc_minrec = ULLONG_MAX;
660 rc = mdd_changelog_llog_init(env, mdd);
662 CERROR("%s: changelog setup during init failed: rc = %d\n",
664 mdd->mdd_cl.mc_flags |= CLM_ERR;
670 static void mdd_changelog_fini(const struct lu_env *env,
671 struct mdd_device *mdd)
673 struct obd_device *obd = mdd2obd_dev(mdd);
674 struct llog_ctxt *ctxt;
676 if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
678 mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
681 /* stop GC-thread if running */
682 spin_lock(&mdd->mdd_cl.mc_lock);
683 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
684 /* avoid any attempt to run a GC-thread */
685 mdd->mdd_cl.mc_gc_task = current;
686 spin_unlock(&mdd->mdd_cl.mc_lock);
688 struct task_struct *gc_task;
690 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
691 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
692 /* need to wait for birthing GC-thread to be started
693 * and to have set mc_gc_task to itself
695 spin_unlock(&mdd->mdd_cl.mc_lock);
696 /* Add a tiny sleep */
697 schedule_timeout_uninterruptible(1);
698 /* go back to fully check if GC-thread has started or
699 * even already exited or if a new one is starting...
703 /* take a reference on task_struct to avoid it to be freed
706 gc_task = mdd->mdd_cl.mc_gc_task;
707 get_task_struct(gc_task);
708 spin_unlock(&mdd->mdd_cl.mc_lock);
709 kthread_stop(gc_task);
710 put_task_struct(gc_task);
713 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
715 llog_cat_close(env, ctxt->loc_handle);
716 llog_cleanup(env, ctxt);
718 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
720 llog_cat_close(env, ctxt->loc_handle);
721 llog_cleanup(env, ctxt);
726 * Remove entries with indicies up to and including \a endrec from the
733 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
734 unsigned long long endrec)
736 struct obd_device *obd = mdd2obd_dev(mdd);
737 struct llog_ctxt *ctxt;
738 unsigned long long cur;
739 struct changelog_cancel_cookie cookie;
742 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
746 spin_lock(&mdd->mdd_cl.mc_lock);
747 cur = (long long)mdd->mdd_cl.mc_index;
748 spin_unlock(&mdd->mdd_cl.mc_lock);
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 a case of crash, we just restart with old log so we're
757 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
764 * Some records were purged, so reset repeat-access time (so we
765 * record new mtime update records, so users can see a file has been
766 * changed since the last purge)
768 mdd->mdd_cl.mc_starttime = ktime_get();
770 cookie.endrec = endrec;
772 rc = llog_changelog_cancel(env, ctxt, &cookie);
778 /** Add a CL_MARK record to the changelog
780 * \param markerflags - CLM_*
783 int mdd_changelog_write_header(const struct lu_env *env,
784 struct mdd_device *mdd, int markerflags)
786 struct obd_device *obd = mdd2obd_dev(mdd);
787 struct llog_changelog_rec *rec;
789 struct llog_ctxt *ctxt;
791 int len = strlen(obd->obd_name);
796 if (mdd->mdd_cl.mc_current_mask & BIT(CL_MARK)) {
797 mdd->mdd_cl.mc_starttime = ktime_get();
801 reclen = llog_data_len(sizeof(*rec) + len);
802 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
803 if (buf->lb_buf == NULL)
807 rec->cr.cr_flags = CLF_VERSION;
808 rec->cr.cr_type = CL_MARK;
809 rec->cr.cr_namelen = len;
810 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
811 /* Status and action flags */
812 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
813 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
815 rec->cr_hdr.lrh_type = CHANGELOG_REC;
816 rec->cr.cr_time = cl_time();
818 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
821 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
826 /* assume on or off event; reset repeat-access time */
827 mdd->mdd_cl.mc_starttime = ktime_get();
832 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
833 * are valid. We also check if object with passed fid exists or not.
835 static int obf_lookup(const struct lu_env *env, struct md_object *p,
836 const struct lu_name *lname, struct lu_fid *f,
837 struct md_op_spec *spec)
839 char *name = (char *)lname->ln_name;
840 struct mdd_device *mdd = mdo2mdd(p);
841 struct mdd_object *child;
847 sscanf(name, SFID, RFID(f));
849 GOTO(out, rc = -ENOENT);
851 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
852 !fid_seq_is_dot(f->f_seq))
853 GOTO(out, rc = -ENOENT);
855 /* Check if object with this fid exists */
856 child = mdd_object_find(env, mdd, f);
858 GOTO(out, rc = PTR_ERR(child));
860 if (mdd_object_exists(child) == 0)
863 mdd_object_put(env, child);
869 static int mdd_dummy_create(const struct lu_env *env,
870 struct md_object *pobj,
871 const struct lu_name *lname,
872 struct md_object *child,
873 struct md_op_spec *spec,
879 static int mdd_dummy_rename(const struct lu_env *env,
880 struct md_object *src_pobj,
881 struct md_object *tgt_pobj,
882 const struct lu_fid *lf,
883 const struct lu_name *lsname,
884 struct md_object *tobj,
885 const struct lu_name *ltname,
891 static int mdd_dummy_link(const struct lu_env *env,
892 struct md_object *tgt_obj,
893 struct md_object *src_obj,
894 const struct lu_name *lname,
900 static int mdd_dummy_unlink(const struct lu_env *env,
901 struct md_object *pobj,
902 struct md_object *cobj,
903 const struct lu_name *lname,
910 int mdd_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);
913 static int mdd_obf_create(const struct lu_env *env, struct md_object *pobj,
914 const struct lu_name *lname, struct md_object *child,
915 struct md_op_spec *spec, struct md_attr *ma)
917 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
918 return mdd_create(env, pobj, lname, child, spec, ma);
922 static const struct md_dir_operations mdd_obf_dir_ops = {
923 .mdo_lookup = obf_lookup,
924 .mdo_create = mdd_obf_create,
925 .mdo_rename = mdd_dummy_rename,
926 .mdo_link = mdd_dummy_link,
927 .mdo_unlink = mdd_dummy_unlink
930 static const struct md_dir_operations mdd_lpf_dir_ops = {
931 .mdo_lookup = mdd_lookup,
932 .mdo_create = mdd_dummy_create,
933 .mdo_rename = mdd_dummy_rename,
934 .mdo_link = mdd_dummy_link,
935 .mdo_unlink = mdd_dummy_unlink
938 static struct md_object *mdo_locate(const struct lu_env *env,
939 struct md_device *md,
940 const struct lu_fid *fid)
942 struct lu_object *obj;
943 struct md_object *mdo;
945 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
947 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
948 LASSERT(obj != NULL);
956 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
958 struct md_object *mdo;
959 struct mdd_object *mdd_lpf;
960 struct lu_fid fid = LU_LPF_FID;
964 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
965 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
970 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
972 RETURN(PTR_ERR(mdo));
974 LASSERT(lu_object_exists(&mdo->mo_lu));
976 mdd_lpf = md2mdd_obj(mdo);
977 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
978 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
984 * Create special in-memory "fid" object for open-by-fid.
986 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
988 struct md_object *mdo;
989 struct mdd_object *mdd_obf;
990 struct lu_fid fid = LU_OBF_FID;
993 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
994 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
998 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1000 RETURN(PTR_ERR(mdo));
1002 LASSERT(lu_object_exists(&mdo->mo_lu));
1004 mdd_obf = md2mdd_obj(mdo);
1005 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1006 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
1011 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
1012 struct mdd_device *m)
1014 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
1015 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
1016 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
1018 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
1019 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
1020 m->mdd_dot_lustre_objs.mdd_obf = NULL;
1022 if (m->mdd_dot_lustre != NULL) {
1023 mdd_object_put(env, m->mdd_dot_lustre);
1024 m->mdd_dot_lustre = NULL;
1028 /** Setup ".lustre" directory object */
1029 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1031 struct md_object *mdo;
1036 /* Create ".lustre" directory in ROOT. */
1037 fid = LU_DOT_LUSTRE_FID;
1038 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
1040 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
1044 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1046 RETURN(PTR_ERR(mdo));
1047 LASSERT(lu_object_exists(&mdo->mo_lu));
1049 m->mdd_dot_lustre = md2mdd_obj(mdo);
1051 rc = mdd_obf_setup(env, m);
1053 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
1054 mdd2obd_dev(m)->obd_name, rc);
1058 rc = mdd_lpf_setup(env, m);
1060 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
1061 mdd2obd_dev(m)->obd_name, rc);
1068 mdd_dot_lustre_cleanup(env, m);
1074 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
1075 * object in obd_device
1077 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
1078 struct mdd_device *m)
1080 struct obd_device *obd = mdd2obd_dev(m);
1081 struct llog_ctxt *ctxt = NULL;
1085 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1086 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1088 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1089 obd, &llog_common_cat_ops);
1091 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1096 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1099 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1102 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1104 GOTO(out_cleanup, rc);
1107 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1109 GOTO(out_close, rc);
1111 llog_ctxt_put(ctxt);
1115 llog_cat_close(env, ctxt->loc_handle);
1116 ctxt->loc_handle = NULL;
1118 llog_cleanup(env, ctxt);
1124 * cleanup the context created by llog_setup_named()
1126 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1127 struct mdd_device *m)
1129 struct obd_device *obd = mdd2obd_dev(m);
1130 struct llog_ctxt *lctxt;
1133 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1135 llog_cat_close(env, lctxt->loc_handle);
1136 lctxt->loc_handle = NULL;
1137 llog_cleanup(env, lctxt);
1143 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1144 struct lustre_cfg *cfg)
1146 barrier_deregister(m->mdd_bottom);
1147 lfsck_degister(env, m->mdd_bottom);
1148 mdd_hsm_actions_llog_fini(env, m);
1149 mdd_changelog_fini(env, m);
1150 mdd_orphan_index_fini(env, m);
1151 mdd_dot_lustre_cleanup(env, m);
1152 if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) {
1153 nm_config_file_deregister_tgt(env,
1154 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
1155 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL;
1157 if (m->mdd_los != NULL) {
1158 local_oid_storage_fini(env, m->mdd_los);
1161 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1163 if (m->mdd_child_exp)
1164 obd_disconnect(m->mdd_child_exp);
1167 static int mdd_process_config(const struct lu_env *env,
1168 struct lu_device *d, struct lustre_cfg *cfg)
1170 struct mdd_device *m = lu2mdd_dev(d);
1171 struct dt_device *dt = m->mdd_child;
1172 struct lu_device *next = &dt->dd_lu_dev;
1176 switch (cfg->lcfg_command) {
1180 count = class_modify_config(cfg, PARAM_MDD, &m->mdd_kobj);
1181 rc = count > 0 ? 0 : count;
1183 /* we don't understand; pass it on */
1184 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1188 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1191 dt_conf_get(env, dt, &m->mdd_dt_conf);
1193 case LCFG_PRE_CLEANUP:
1194 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1195 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1198 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1199 lu_dev_del_linkage(d->ld_site, d);
1200 mdd_device_shutdown(env, m, cfg);
1203 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1210 static int mdd_recovery_complete(const struct lu_env *env,
1211 struct lu_device *d)
1213 struct mdd_device *mdd = lu2mdd_dev(d);
1214 struct lu_device *next;
1218 LASSERT(mdd != NULL);
1219 next = &mdd->mdd_child->dd_lu_dev;
1221 if (!mdd->mdd_bottom->dd_rdonly)
1222 mdd_orphan_cleanup(env, mdd);
1223 rc = next->ld_ops->ldo_recovery_complete(env, next);
1228 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1229 const struct lu_fid *pfid, const char *name,
1230 __u32 mode, struct lu_fid *fid)
1232 struct dt_object *parent, *dto;
1237 LASSERT(!fid_is_zero(pfid));
1238 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1239 if (unlikely(IS_ERR(parent)))
1240 RETURN(PTR_ERR(parent));
1242 /* create local file/dir, if @fid is passed then try to use it */
1243 if (fid_is_zero(fid))
1244 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1247 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1251 GOTO(out_put, rc = PTR_ERR(dto));
1252 *fid = *lu_object_fid(&dto->do_lu);
1253 /* since stack is not fully set up the local_storage uses own stack
1254 * and we should drop its object from cache */
1255 dt_object_put_nocache(env, dto);
1258 dt_object_put(env, parent);
1262 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1263 enum lfsck_events event)
1268 static int mdd_prepare(const struct lu_env *env,
1269 struct lu_device *pdev,
1270 struct lu_device *cdev)
1272 struct mdd_device *mdd = lu2mdd_dev(cdev);
1273 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1274 struct nm_config_file *nodemap_config;
1275 struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
1276 struct dt_object *root = NULL;
1277 struct thandle *th = NULL;
1283 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1287 /* Setup local dirs */
1288 fid.f_seq = FID_SEQ_LOCAL_NAME;
1291 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1296 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1301 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1302 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1303 mdd_root_dir_name, S_IFDIR |
1304 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1306 CERROR("%s: create root fid failed: rc = %d\n",
1307 mdd2obd_dev(mdd)->obd_name, rc);
1311 /* store a default directory layout on the root directory if
1312 * it doesn't already exist to improve MDT space balance.
1314 root = dt_locate(env, mdd->mdd_bottom, &fid);
1315 if (unlikely(IS_ERR(root)))
1316 GOTO(out_los, rc = PTR_ERR(root));
1318 rc = dt_xattr_get(env, root, &LU_BUF_NULL,
1319 XATTR_NAME_DEFAULT_LMV);
1320 if (rc == -ENODATA) {
1322 struct lmv_user_md lmv_default = {
1323 .lum_magic = LMV_USER_MAGIC,
1324 .lum_stripe_count = 1,
1325 .lum_stripe_offset = LMV_OFFSET_DEFAULT,
1326 .lum_max_inherit = LMV_INHERIT_UNLIMITED,
1327 .lum_max_inherit_rr = LMV_INHERIT_RR_ROOT,
1330 th = dt_trans_create(env, mdd->mdd_bottom);
1332 GOTO(out_root_put, rc = PTR_ERR(th));
1334 buf.lb_buf = &lmv_default;
1335 buf.lb_len = sizeof(lmv_default);
1336 rc = dt_declare_xattr_set(env, root, &buf,
1337 XATTR_NAME_DEFAULT_LMV, 0,
1340 GOTO(out_trans_stop, rc);
1342 rc = dt_trans_start_local(env, mdd->mdd_bottom, th);
1344 GOTO(out_trans_stop, rc);
1346 rc = dt_xattr_set(env, root, &buf,
1347 XATTR_NAME_DEFAULT_LMV, 0, th);
1349 GOTO(out_trans_stop, rc);
1351 dt_trans_stop(env, mdd->mdd_bottom, th);
1353 } else if (rc < 0 && rc != -ERANGE) {
1354 CERROR("%s: get default LMV of root failed: rc = %d\n",
1355 mdd2obd_dev(mdd)->obd_name, rc);
1357 GOTO(out_root_put, rc);
1360 dt_object_put(env, root);
1363 mdd->mdd_root_fid = fid;
1364 rc = mdd_dot_lustre_setup(env, mdd);
1366 CERROR("%s: initializing .lustre failed: rc = %d\n",
1367 mdd2obd_dev(mdd)->obd_name, rc);
1371 /* Normal client usually send root access to MDT0 directly,
1372 * the root FID on non-MDT0 will only be used by echo client. */
1373 mdd->mdd_root_fid = fid;
1376 rc = mdd_orphan_index_init(env, mdd);
1380 rc = mdd_changelog_init(env, mdd);
1382 CERROR("%s: failed to initialize changelog: rc = %d\n",
1383 mdd2obd_dev(mdd)->obd_name, rc);
1387 rc = mdd_hsm_actions_llog_init(env, mdd);
1389 GOTO(out_changelog, rc);
1391 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1393 if (IS_ERR(nodemap_config)) {
1394 rc = PTR_ERR(nodemap_config);
1398 obt->obt_nodemap_config_file = nodemap_config;
1401 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1402 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1405 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1406 mdd2obd_dev(mdd)->obd_name, rc);
1407 GOTO(out_nodemap, rc);
1410 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1412 CERROR("%s: failed to register to barrier: rc = %d\n",
1413 mdd2obd_dev(mdd)->obd_name, rc);
1414 GOTO(out_lfsck, rc);
1420 lfsck_degister(env, mdd->mdd_bottom);
1422 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1423 obt->obt_nodemap_config_file = NULL;
1425 mdd_hsm_actions_llog_fini(env, mdd);
1427 mdd_changelog_fini(env, mdd);
1429 mdd_orphan_index_fini(env, mdd);
1431 if (mdd_seq_site(mdd)->ss_node_id == 0)
1432 mdd_dot_lustre_cleanup(env, mdd);
1435 dt_trans_stop(env, mdd->mdd_bottom, th);
1438 dt_object_put(env, root);
1440 local_oid_storage_fini(env, mdd->mdd_los);
1441 mdd->mdd_los = NULL;
1447 * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
1449 * Find corresponding device by passed parent and name, and allocate FID from
1452 * see include/lu_object.h for the details.
1454 static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1455 struct lu_fid *fid, struct lu_object *parent,
1456 const struct lu_name *name)
1458 struct mdd_device *mdd = lu2mdd_dev(d);
1459 struct lu_object *o = lu_object_next(parent);
1461 return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
1464 const struct lu_device_operations mdd_lu_ops = {
1465 .ldo_object_alloc = mdd_object_alloc,
1466 .ldo_process_config = mdd_process_config,
1467 .ldo_recovery_complete = mdd_recovery_complete,
1468 .ldo_prepare = mdd_prepare,
1469 .ldo_fid_alloc = mdd_fid_alloc,
1472 static int mdd_root_get(const struct lu_env *env,
1473 struct md_device *m, struct lu_fid *f)
1475 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1478 *f = mdd->mdd_root_fid;
1483 * No permission check is needed.
1485 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1486 struct obd_statfs *sfs)
1488 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1493 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs, NULL);
1495 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1500 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1501 struct md_device *m)
1503 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1505 return &mdd->mdd_dt_conf;
1508 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1511 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1513 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1514 return (*h == NULL ? -ENOENT : 0);
1517 static struct lu_device *mdd_device_free(const struct lu_env *env,
1518 struct lu_device *lu)
1520 struct mdd_device *m = lu2mdd_dev(lu);
1523 LASSERT(atomic_read(&lu->ld_ref) == 0);
1524 md_device_fini(&m->mdd_md_dev);
1529 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1530 struct lu_device_type *t,
1531 struct lustre_cfg *lcfg)
1533 struct lu_device *l;
1534 struct mdd_device *m;
1538 l = ERR_PTR(-ENOMEM);
1543 md_device_init(&m->mdd_md_dev, t);
1544 rc = mdd_init0(env, m, t, lcfg);
1546 mdd_device_free(env, l);
1555 * we use exports to track all mdd users
1557 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1558 struct obd_device *obd, struct obd_uuid *cluuid,
1559 struct obd_connect_data *data, void *localdata)
1561 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1562 struct lustre_handle conn;
1566 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1568 rc = class_connect(&conn, obd, cluuid);
1572 *exp = class_conn2export(&conn);
1574 /* Why should there ever be more than 1 connect? */
1575 LASSERT(mdd->mdd_connects == 0);
1576 mdd->mdd_connects++;
1582 * once last export (we don't count self-export) disappeared
1583 * mdd can be released
1585 static int mdd_obd_disconnect(struct obd_export *exp)
1587 struct obd_device *obd = exp->exp_obd;
1588 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1589 int rc, release = 0;
1592 mdd->mdd_connects--;
1593 if (mdd->mdd_connects == 0)
1596 rc = class_disconnect(exp);
1598 if (rc == 0 && release)
1599 class_manual_cleanup(obd);
1603 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1604 __u32 keylen, void *key, __u32 *vallen, void *val)
1608 if (KEY_IS(KEY_OSP_CONNECTED)) {
1609 struct obd_device *obd = exp->exp_obd;
1610 struct mdd_device *mdd;
1612 if (!obd->obd_set_up || obd->obd_stopping)
1615 mdd = lu2mdd_dev(obd->obd_lu_dev);
1617 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1625 static int mdd_obd_set_info_async(const struct lu_env *env,
1626 struct obd_export *exp,
1627 __u32 keylen, void *key,
1628 __u32 vallen, void *val,
1629 struct ptlrpc_request_set *set)
1631 struct obd_device *obd = exp->exp_obd;
1632 struct mdd_device *mdd;
1635 if (!obd->obd_set_up || obd->obd_stopping)
1638 mdd = lu2mdd_dev(obd->obd_lu_dev);
1640 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1645 static const struct obd_ops mdd_obd_device_ops = {
1646 .o_owner = THIS_MODULE,
1647 .o_connect = mdd_obd_connect,
1648 .o_disconnect = mdd_obd_disconnect,
1649 .o_get_info = mdd_obd_get_info,
1650 .o_set_info_async = mdd_obd_set_info_async,
1653 struct mdd_changelog_name_check_data {
1654 const char *mcnc_name;
1659 * changelog_recalc_mask callback
1661 * Is is called per each registered user and calculates combined mask of
1662 * all registered users.
1664 static int mdd_changelog_name_check_cb(const struct lu_env *env,
1665 struct llog_handle *llh,
1666 struct llog_rec_hdr *hdr, void *data)
1668 struct llog_changelog_user_rec2 *rec;
1669 struct mdd_changelog_name_check_data *mcnc = data;
1671 rec = container_of(hdr, typeof(*rec), cur_hdr);
1672 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
1673 !strncmp(rec->cur_name, mcnc->mcnc_name, sizeof(rec->cur_name))) {
1674 mcnc->mcnc_id = rec->cur_id;
1680 static int mdd_changelog_name_check(const struct lu_env *env,
1681 struct llog_ctxt *ctxt,
1682 struct mdd_device *mdd, const char *name)
1684 struct mdd_changelog_name_check_data mcnc = { .mcnc_name = name, };
1690 /* first symbol is a letter */
1691 if (!isalpha(name[0])) {
1693 CERROR("%s: first char '%c' in '%s' is not letter: rc = %d\n",
1694 mdd2obd_dev(mdd)->obd_name, name[0], name, rc);
1698 /* name is valid: contains letters, numbers and '-', '_' only */
1699 while (name[++chr]) {
1700 if (!(isalnum(name[chr]) || name[chr] == '_' ||
1701 name[chr] == '-')) {
1703 CERROR("%s: wrong char '%c' in name '%s': rc = %d\n",
1704 mdd2obd_dev(mdd)->obd_name, name[chr], name, rc);
1709 if (chr > CHANGELOG_USER_NAMELEN) {
1711 CERROR("%s: name '%s' is over %d symbols limit: rc = %d\n",
1712 mdd2obd_dev(mdd)->obd_name, name,
1713 CHANGELOG_USER_NAMELEN, rc);
1717 rc = llog_cat_process(env, ctxt->loc_handle,
1718 mdd_changelog_name_check_cb, &mcnc, 0, 0);
1720 CWARN("%s: changelog name %s exists already: rc = %d\n",
1721 mdd2obd_dev(mdd)->obd_name, name, rc);
1723 CWARN("%s: failed user changelog processing: rc = %d\n",
1724 mdd2obd_dev(mdd)->obd_name, rc);
1728 static int mdd_changelog_user_register(const struct lu_env *env,
1729 struct mdd_device *mdd, int *id,
1730 const char *name, const char *mask)
1732 struct llog_ctxt *ctxt;
1733 struct llog_changelog_user_rec2 *rec;
1734 char user_name[CHANGELOG_USER_NAMELEN_FULL];
1739 ctxt = llog_get_context(mdd2obd_dev(mdd),
1740 LLOG_CHANGELOG_USER_ORIG_CTXT);
1746 llog_ctxt_put(ctxt);
1750 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1752 rec->cur_hdr.lrh_len = sizeof(*rec);
1753 /* keep old record type for users without mask/name for
1754 * compatibility needs
1756 if (mask || (name && name[0]))
1757 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
1759 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1760 spin_lock(&mdd->mdd_cl.mc_user_lock);
1761 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1762 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1764 CERROR("%s: registering %s user: max ID is exceeded: rc = %d\n",
1765 mdd2obd_dev(mdd)->obd_name,
1766 (name && name[0]) ? name : "new", rc);
1769 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1770 mdd->mdd_cl.mc_users++;
1771 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1773 rec->cur_time = (__u32)ktime_get_real_seconds();
1774 if (OBD_FAIL_PRECHECK(OBD_FAIL_TIME_IN_CHLOG_USER)) {
1775 rec->cur_time -= min(cfs_fail_val, rec->cur_time);
1776 spin_lock(&mdd->mdd_cl.mc_user_lock);
1777 mdd->mdd_cl.mc_mintime = rec->cur_time;
1778 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1781 spin_lock(&mdd->mdd_cl.mc_lock);
1782 rec->cur_endrec = mdd->mdd_cl.mc_index;
1783 spin_unlock(&mdd->mdd_cl.mc_lock);
1786 /* if user will use relative mask apply it on default one */
1787 rec->cur_mask = CHANGELOG_DEFMASK;
1788 rc = cfs_str2mask(mask, changelog_type2str, &rec->cur_mask,
1789 CHANGELOG_MINMASK, CHANGELOG_ALLMASK,
1792 GOTO(out_users, rc);
1793 } else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK) {
1794 /* a maskless users means default mask but only if server has
1795 * no specific mask set
1797 rec->cur_mask = CHANGELOG_DEFMASK;
1800 if (name && name[0]) {
1801 rc = mdd_changelog_name_check(env, ctxt, mdd, name);
1803 GOTO(out_users, rc);
1804 strlcpy(rec->cur_name, name, sizeof(rec->cur_name));
1806 mdd_chlg_username(rec, user_name, sizeof(user_name));
1808 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1810 CWARN("%s: failed to register changelog user %s: rc = %d\n",
1811 mdd2obd_dev(mdd)->obd_name, user_name, rc);
1812 GOTO(out_users, rc);
1815 /* apply user mask finally */
1816 spin_lock(&mdd->mdd_cl.mc_user_lock);
1817 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
1818 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1820 CDEBUG(D_IOCTL, "%s: registered changelog user '%s', mask %#x\n",
1821 mdd2obd_dev(mdd)->obd_name, user_name, rec->cur_mask);
1823 /* Assume we want it on since somebody registered */
1824 rc = mdd_changelog_on(env, mdd);
1826 /* record is added, so don't decrement users on error */
1830 spin_lock(&mdd->mdd_cl.mc_user_lock);
1831 mdd->mdd_cl.mc_users--;
1832 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1836 llog_ctxt_put(ctxt);
1840 struct mdd_changelog_recalc_mask_data {
1841 struct mdd_device *mcrm_mdd;
1846 * changelog_recalc_mask callback
1848 * Is is called per each registered user and calculates combined mask of
1849 * all registered users.
1851 static int mdd_changelog_recalc_mask_cb(const struct lu_env *env,
1852 struct llog_handle *llh,
1853 struct llog_rec_hdr *hdr, void *data)
1855 struct llog_changelog_user_rec2 *rec;
1856 struct mdd_changelog_recalc_mask_data *mcrm = data;
1858 rec = container_of(hdr, typeof(*rec), cur_hdr);
1859 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
1860 mcrm->mcrm_mask |= rec->cur_mask;
1861 else if (mcrm->mcrm_mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
1862 mcrm->mcrm_mask |= CHANGELOG_DEFMASK;
1867 int mdd_changelog_recalc_mask(const struct lu_env *env, struct mdd_device *mdd)
1869 struct llog_ctxt *ctxt;
1870 struct mdd_changelog_recalc_mask_data mcrm = {
1872 .mcrm_mask = mdd->mdd_cl.mc_proc_mask,
1878 ctxt = llog_get_context(mdd2obd_dev(mdd),
1879 LLOG_CHANGELOG_USER_ORIG_CTXT);
1883 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1884 GOTO(out, rc = -ENXIO);
1886 rc = llog_cat_process(env, ctxt->loc_handle,
1887 mdd_changelog_recalc_mask_cb, &mcrm, 0, 0);
1889 CWARN("%s: failed user changelog processing: rc = %d\n",
1890 mdd2obd_dev(mdd)->obd_name, rc);
1892 spin_lock(&mdd->mdd_cl.mc_user_lock);
1893 CDEBUG(D_INFO, "%s: recalc changelog mask: %#x -> %#x\n",
1894 mdd2obd_dev(mdd)->obd_name, mdd->mdd_cl.mc_current_mask,
1896 mdd->mdd_cl.mc_current_mask = mcrm.mcrm_mask;
1897 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1901 llog_ctxt_put(ctxt);
1906 struct mdd_changelog_user_purge {
1907 struct mdd_device *mcup_mdd;
1909 __u32 mcup_usercount;
1912 char mcup_name[CHANGELOG_USER_NAMELEN_FULL];
1916 * changelog_user_purge callback
1918 * Is called once per user.
1920 * Check to see the user requested is available from rec.
1921 * Truncate the changelog.
1922 * Keep track of the total number of users (calls).
1924 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1925 struct llog_handle *llh,
1926 struct llog_rec_hdr *hdr, void *data)
1928 struct llog_changelog_user_rec2 *rec;
1929 struct mdd_changelog_user_purge *mcup = data;
1930 struct llog_cookie cookie;
1935 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1938 rec = container_of(hdr, typeof(*rec), cur_hdr);
1940 mcup->mcup_usercount++;
1942 if (rec->cur_id != mcup->mcup_id) {
1943 /* truncate to the lowest endrec that is not this user */
1944 mcup->mcup_minrec = min(mcup->mcup_minrec, rec->cur_endrec);
1948 mdd_chlg_username(rec, mcup->mcup_name, sizeof(mcup->mcup_name));
1950 /* Unregister this user */
1951 cookie.lgc_lgl = llh->lgh_id;
1952 cookie.lgc_index = hdr->lrh_index;
1953 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1956 mcup->mcup_found = true;
1957 mcup->mcup_usercount--;
1958 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1959 mcup->mcup_mdd->mdd_cl.mc_users--;
1960 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1966 int mdd_changelog_user_purge(const struct lu_env *env,
1967 struct mdd_device *mdd, __u32 id)
1969 struct mdd_changelog_user_purge mcup = {
1972 .mcup_found = false,
1973 .mcup_usercount = 0,
1974 .mcup_minrec = ULLONG_MAX,
1977 struct llog_ctxt *ctxt;
1982 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1983 mdd2obd_dev(mdd)->obd_name, id);
1985 ctxt = llog_get_context(mdd2obd_dev(mdd),
1986 LLOG_CHANGELOG_USER_ORIG_CTXT);
1989 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1990 GOTO(out, rc = -ENXIO);
1992 rc = llog_cat_process(env, ctxt->loc_handle,
1993 mdd_changelog_user_purge_cb, &mcup, 0, 0);
1995 CWARN("%s: failed to purge changelog for user %s: rc = %d\n",
1996 mdd2obd_dev(mdd)->obd_name, mcup.mcup_name, rc);
2000 OBD_FAIL_TIMEOUT(OBD_FAIL_LLOG_PURGE_DELAY, cfs_fail_val);
2001 if (mcup.mcup_usercount == 0) {
2002 spin_lock(&mdd->mdd_cl.mc_user_lock);
2003 if (mdd->mdd_cl.mc_users == 0) {
2004 /* No more users; turn changelogs off */
2005 CDEBUG(D_IOCTL, "turning off changelogs\n");
2006 rc = mdd_changelog_off(env, mdd);
2008 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2011 if (mcup.mcup_found) {
2013 "%s: Purge changelog entries for user %s record=%llu\n",
2014 mdd2obd_dev(mdd)->obd_name,
2015 mcup.mcup_name, mcup.mcup_minrec);
2016 rc = mdd_changelog_llog_cancel(env, mdd, mcup.mcup_minrec);
2018 CWARN("%s: No changelog for user id %u: rc = %d\n",
2019 mdd2obd_dev(mdd)->obd_name, id, rc);
2020 GOTO(out, rc = -ENOENT);
2023 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
2027 llog_ctxt_put(ctxt);
2032 struct mdd_changelog_user_clear {
2038 struct mdd_device *mcuc_mdd;
2039 char mcuc_name[CHANGELOG_USER_NAMELEN_FULL];
2043 * changelog_clear callback
2045 * Is called once per user.
2047 * Check to see the user requested is available from rec.
2048 * Check the oldest (smallest) record for boundary conditions.
2049 * Truncate the changelog.
2051 static int mdd_changelog_clear_cb(const struct lu_env *env,
2052 struct llog_handle *llh,
2053 struct llog_rec_hdr *hdr,
2056 struct llog_changelog_user_rec2 *rec;
2057 struct mdd_changelog_user_clear *mcuc = data;
2058 struct mdd_device *mdd = mcuc->mcuc_mdd;
2063 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
2066 rec = container_of(hdr, typeof(*rec), cur_hdr);
2067 /* Does the changelog id match the requested id? */
2068 if (rec->cur_id != mcuc->mcuc_id) {
2069 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec, rec->cur_endrec);
2070 mcuc->mcuc_mintime = min(mcuc->mcuc_mintime, rec->cur_time);
2074 mdd_chlg_username(rec, mcuc->mcuc_name, sizeof(mcuc->mcuc_name));
2075 /* cur_endrec is the oldest purgeable record, make sure we're newer */
2076 if (rec->cur_endrec > mcuc->mcuc_endrec) {
2079 "%s: request %llu < endrec %llu for user %s: rc = %d\n",
2080 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_endrec,
2081 rec->cur_endrec, mcuc->mcuc_name, rc);
2086 * Flag that we've met all the range and user checks.
2087 * We now know the record to flush.
2089 mcuc->mcuc_flush = true;
2091 rec->cur_endrec = mcuc->mcuc_endrec;
2092 rec->cur_time = (__u32)ktime_get_real_seconds();
2094 CDEBUG(D_IOCTL, "%s: update changelog user %s endrec = %llu\n",
2095 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_name, rec->cur_endrec);
2097 /* Update the endrec */
2098 rc = llog_write(env, llh, hdr, hdr->lrh_index);
2104 * Clear a changelog up to entry specified by endrec for user id.
2106 static int mdd_changelog_clear(const struct lu_env *env,
2107 struct mdd_device *mdd, __u32 id,
2110 struct mdd_changelog_user_clear mcuc = {
2112 .mcuc_minrec = endrec,
2113 .mcuc_flush = false,
2115 .mcuc_mintime = ktime_get_real_seconds(),
2118 struct llog_ctxt *ctxt;
2124 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
2125 mdd2obd_dev(mdd)->obd_name, id, endrec);
2126 /* start_rec is the newest (largest value) entry in the changelogs*/
2127 spin_lock(&mdd->mdd_cl.mc_lock);
2128 start_rec = mdd->mdd_cl.mc_index;
2129 spin_unlock(&mdd->mdd_cl.mc_lock);
2131 if (start_rec < endrec) {
2132 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
2133 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
2138 mcuc.mcuc_endrec = start_rec;
2139 mcuc.mcuc_minrec = ULLONG_MAX;
2141 mcuc.mcuc_endrec = endrec;
2144 ctxt = llog_get_context(mdd2obd_dev(mdd),
2145 LLOG_CHANGELOG_USER_ORIG_CTXT);
2149 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
2150 GOTO(out, rc = -ENXIO);
2152 rc = llog_cat_process(env, ctxt->loc_handle, mdd_changelog_clear_cb,
2154 if (rc == -EINVAL) {
2155 CDEBUG(D_IOCTL, "%s: no changelog recnum <= %llu to clear\n",
2156 mdd2obd_dev(mdd)->obd_name, (unsigned long long)endrec);
2161 CWARN("%s: can't clear the changelog for user %s: rc = %d\n",
2162 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, rc);
2166 if (!mcuc.mcuc_flush) {
2167 CDEBUG(D_IOCTL, "%s: no entry for user %d\n",
2168 mdd2obd_dev(mdd)->obd_name, id);
2169 GOTO(out, rc = -ENOENT);
2172 CDEBUG(D_IOCTL, "%s: purge changelog user %s entries up to %llu\n",
2173 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, mcuc.mcuc_minrec);
2175 rc = mdd_changelog_llog_cancel(env, mdd, mcuc.mcuc_minrec);
2179 spin_lock(&mdd->mdd_cl.mc_user_lock);
2180 mdd->mdd_cl.mc_minrec = mcuc.mcuc_minrec;
2181 mdd->mdd_cl.mc_mintime = mcuc.mcuc_mintime;
2182 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2186 llog_ctxt_put(ctxt);
2191 static int mdd_changelog_user_deregister(const struct lu_env *env,
2192 struct mdd_device *mdd, int *id,
2195 struct llog_ctxt *ctxt;
2196 struct mdd_changelog_name_check_data mcnc = {
2205 ctxt = llog_get_context(mdd2obd_dev(mdd),
2206 LLOG_CHANGELOG_USER_ORIG_CTXT);
2210 rc = llog_cat_process(env, ctxt->loc_handle,
2211 mdd_changelog_name_check_cb, &mcnc, 0, 0);
2212 llog_ctxt_put(ctxt);
2214 if (rc != -EEXIST) {
2215 CDEBUG(D_IOCTL, "%s: no entry for username %s\n",
2216 mdd2obd_dev(mdd)->obd_name, name);
2222 /* explicitly clear changelog first, to protect from crash in
2223 * the middle of purge that would lead to unregistered consumer
2224 * but pending changelog entries
2226 rc = mdd_changelog_clear(env, mdd, *id, 0);
2228 rc = mdd_changelog_user_purge(env, mdd, *id);
2230 /* recalc changelog current mask */
2231 mdd_changelog_recalc_mask(env, mdd);
2237 * May be called remotely from mdt_iocontrol_handle or locally from
2238 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
2239 * an obd_ioctl_data format (but local ioctl handler does).
2241 * \param len - data len
2242 * \param karg - ioctl data, in kernel space
2244 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
2245 unsigned int cmd, int len, void *karg)
2247 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
2248 struct obd_ioctl_data *data = karg;
2252 /* Doesn't use obd_ioctl_data */
2254 case OBD_IOC_CHANGELOG_CLEAR: {
2255 struct changelog_setinfo *cs = karg;
2257 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2258 RETURN(-EINPROGRESS);
2260 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
2262 barrier_exit(mdd->mdd_bottom);
2265 case OBD_IOC_START_LFSCK: {
2266 rc = lfsck_start(env, mdd->mdd_bottom,
2267 (struct lfsck_start_param *)karg);
2270 case OBD_IOC_STOP_LFSCK: {
2271 rc = lfsck_stop(env, mdd->mdd_bottom,
2272 (struct lfsck_stop *)karg);
2275 case OBD_IOC_QUERY_LFSCK: {
2276 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
2277 (struct lfsck_query *)karg);
2282 /* Below ioctls use obd_ioctl_data */
2283 if (data->ioc_version != OBD_IOCTL_VERSION) {
2284 CERROR("Bad magic %x != %x\n", data->ioc_version,
2290 case OBD_IOC_CHANGELOG_REG:
2291 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2292 RETURN(-EINPROGRESS);
2294 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1,
2297 barrier_exit(mdd->mdd_bottom);
2299 case OBD_IOC_CHANGELOG_DEREG:
2300 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2301 RETURN(-EINPROGRESS);
2303 rc = mdd_changelog_user_deregister(env, mdd, &data->ioc_u32_1,
2306 barrier_exit(mdd->mdd_bottom);
2315 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
2316 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
2318 static const struct md_device_operations mdd_ops = {
2319 .mdo_statfs = mdd_statfs,
2320 .mdo_root_get = mdd_root_get,
2321 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
2322 .mdo_iocontrol = mdd_iocontrol,
2323 .mdo_dtconf_get = mdd_dtconf_get,
2326 static const struct lu_device_type_operations mdd_device_type_ops = {
2327 .ldto_init = mdd_type_init,
2328 .ldto_fini = mdd_type_fini,
2330 .ldto_start = mdd_type_start,
2331 .ldto_stop = mdd_type_stop,
2333 .ldto_device_alloc = mdd_device_alloc,
2334 .ldto_device_free = mdd_device_free,
2336 .ldto_device_fini = mdd_device_fini
2339 static struct lu_device_type mdd_device_type = {
2340 .ldt_tags = LU_DEVICE_MD,
2341 .ldt_name = LUSTRE_MDD_NAME,
2342 .ldt_ops = &mdd_device_type_ops,
2343 .ldt_ctx_tags = LCT_MD_THREAD
2346 /* context key constructor: mdd_key_init */
2347 LU_KEY_INIT(mdd, struct mdd_thread_info);
2349 static void mdd_key_fini(const struct lu_context *ctx,
2350 struct lu_context_key *key, void *data)
2352 struct mdd_thread_info *info = data;
2354 lu_buf_free(&info->mdi_big_buf);
2355 lu_buf_free(&info->mdi_link_buf);
2356 lu_buf_free(&info->mdi_xattr_buf);
2357 lu_buf_free(&info->mdi_chlg_buf);
2362 /* context key: mdd_thread_key */
2363 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
2365 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
2366 int (*func)(void *), void *data, char *name)
2368 struct task_struct *task;
2370 LASSERT(thread->mgt_init == false);
2371 init_completion(&thread->mgt_started);
2372 init_completion(&thread->mgt_finished);
2373 thread->mgt_data = data;
2374 thread->mgt_abort = false;
2375 thread->mgt_init = true;
2377 task = kthread_run(func, thread, name);
2379 complete(&thread->mgt_finished);
2380 return PTR_ERR(task);
2382 wait_for_completion(&thread->mgt_started);
2386 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
2388 if (thread->mgt_init == true) {
2389 thread->mgt_abort = true;
2390 wait_for_completion(&thread->mgt_finished);
2394 static int __init mdd_init(void)
2398 rc = lu_kmem_init(mdd_caches);
2402 changelog_orig_logops = llog_common_cat_ops;
2403 changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
2405 rc = class_register_type(&mdd_obd_device_ops, NULL, false,
2406 LUSTRE_MDD_NAME, &mdd_device_type);
2408 lu_kmem_fini(mdd_caches);
2412 static void __exit mdd_exit(void)
2414 class_unregister_type(LUSTRE_MDD_NAME);
2415 lu_kmem_fini(mdd_caches);
2418 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2419 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2420 MODULE_VERSION(LUSTRE_VERSION_STRING);
2421 MODULE_LICENSE("GPL");
2423 module_init(mdd_init);
2424 module_exit(mdd_exit);