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)
76 * Mod params for the fs-wide (stored in ROOT) default dir layout,
77 * to be set and stored in EA, if EA is absent yet:
79 * stripe_offset, max_inherit, max_inherit_rr, see lfs setdirstripe for more
82 static unsigned int root_stripe_offset = LMV_OFFSET_DEFAULT;
83 module_param(root_stripe_offset, uint, 0444);
84 MODULE_PARM_DESC(root_stripe_offset, "fs-wide default LMV stripe offset");
86 static unsigned int root_max_inherit = LMV_INHERIT_UNLIMITED;
87 module_param(root_max_inherit, uint, 0444);
88 MODULE_PARM_DESC(root_max_inherit, "fs-wide default LMV max_inherit");
90 static unsigned int root_max_inherit_rr = LMV_INHERIT_RR_ROOT;
91 module_param(root_max_inherit_rr, uint, 0444);
92 MODULE_PARM_DESC(root_max_inherit_rr, "fs-wide default LMV max_inherit_rr");
94 static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
97 struct obd_connect_data *data = NULL;
98 struct lu_device *lud = mdd2lu_dev(m);
99 struct obd_device *obd;
103 LASSERT(m->mdd_child_exp == NULL);
105 OBD_ALLOC(data, sizeof(*data));
107 GOTO(out, rc = -ENOMEM);
109 obd = class_name2obd(nextdev);
111 CERROR("can't locate next device: %s\n", nextdev);
112 GOTO(out, rc = -ENOTCONN);
115 data->ocd_connect_flags = OBD_CONNECT_VERSION;
116 data->ocd_version = LUSTRE_VERSION_CODE;
118 rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
120 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
124 lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
125 LASSERT(lud->ld_site);
126 m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
127 m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
128 lu_dev_add_linkage(lud->ld_site, lud);
132 OBD_FREE(data, sizeof(*data));
136 static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
137 struct lu_device_type *t, struct lustre_cfg *lcfg)
144 /* LU-8040 Set defaults here, before values configs */
145 mdd->mdd_cl.mc_flags = 0; /* off by default */
146 /* per-server mask is set via parameters if needed */
147 mdd->mdd_cl.mc_proc_mask = CHANGELOG_MINMASK;
148 /* current mask is calculated from mask above and users masks */
149 mdd->mdd_cl.mc_current_mask = CHANGELOG_MINMASK;
150 mdd->mdd_cl.mc_deniednext = 60; /* 60 secs by default */
152 dev = lustre_cfg_string(lcfg, 0);
156 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
157 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
159 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
160 mdd->mdd_md_dev.md_ops = &mdd_ops;
162 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
166 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
167 /* sync permission changes */
168 mdd->mdd_sync_permission = 1;
169 /* enable changelog garbage collection */
170 mdd->mdd_changelog_gc = 1;
171 /* enable changelog cleanup due to lack of space */
172 mdd->mdd_changelog_free_space_gc = true;
173 /* set when emergency GC is started */
174 mdd->mdd_changelog_emrg_gc = false;
175 /* with a significant amount of idle time */
176 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
177 /* or a significant amount of late indexes */
178 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
179 /* with a reasonable interval between each check */
180 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
181 /* with a very few number of free catalog entries */
182 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
183 /* special default striping for files created with O_APPEND */
184 mdd->mdd_append_stripe_count = 1;
185 mdd->mdd_append_pool[0] = '\0';
187 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
189 /* we are using service name but not mdd obd name
190 * for compatibility reasons.
191 * It is passed from MDT in lustre_cfg[2] buffer */
192 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
194 obd_disconnect(mdd->mdd_child_exp);
199 static struct lu_device *mdd_device_fini(const struct lu_env *env,
202 struct mdd_device *mdd = lu2mdd_dev(d);
205 lu_dev_del_linkage(d->ld_site, d);
207 mdd_procfs_fini(mdd);
211 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
212 struct llog_rec_hdr *hdr, void *data)
214 struct mdd_device *mdd = (struct mdd_device *)data;
215 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
217 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
218 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
221 "seeing record at index %d/%d/%llu t=%x %.*s in log"
222 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
223 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
224 changelog_rec_name(&rec->cr), PLOGID(&llh->lgh_id));
225 spin_lock(&mdd->mdd_cl.mc_lock);
226 mdd->mdd_cl.mc_index = rec->cr.cr_index;
227 spin_unlock(&mdd->mdd_cl.mc_lock);
228 return LLOG_PROC_BREAK;
231 char *mdd_chlg_username(struct llog_changelog_user_rec2 *rec, char *buf,
234 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
236 snprintf(buf, len, "%s%u-%s", CHANGELOG_USER_PREFIX,
237 rec->cur_id, rec->cur_name);
239 snprintf(buf, len, "%s%u", CHANGELOG_USER_PREFIX, rec->cur_id);
243 __u32 mdd_chlg_usermask(struct llog_changelog_user_rec2 *rec)
245 return rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 ?
249 static int changelog_user_init_cb(const struct lu_env *env,
250 struct llog_handle *llh,
251 struct llog_rec_hdr *hdr, void *data)
253 struct mdd_device *mdd = data;
254 struct llog_changelog_user_rec2 *rec;
255 char user_name[CHANGELOG_USER_NAMELEN_FULL];
257 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
259 rec = container_of(hdr, typeof(*rec), cur_hdr);
260 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
261 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
262 CWARN("%s: unknown user type %x at index %u in log "DFID"\n",
263 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
264 rec->cur_hdr.lrh_type, PLOGID(&llh->lgh_id));
269 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
270 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
272 hdr->lrh_index, rec->cur_hdr.lrh_index, rec->cur_endrec,
273 PLOGID(&llh->lgh_id));
275 spin_lock(&mdd->mdd_cl.mc_user_lock);
276 mdd->mdd_cl.mc_lastuser = rec->cur_id;
277 mdd->mdd_cl.mc_users++;
278 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
279 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
280 else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
281 mdd->mdd_cl.mc_current_mask |= CHANGELOG_DEFMASK;
282 mdd->mdd_cl.mc_mintime = min(mdd->mdd_cl.mc_mintime, rec->cur_time);
283 mdd->mdd_cl.mc_minrec = min(mdd->mdd_cl.mc_minrec, rec->cur_endrec);
284 spin_unlock(&mdd->mdd_cl.mc_user_lock);
285 spin_lock(&mdd->mdd_cl.mc_lock);
286 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
287 mdd->mdd_cl.mc_index = rec->cur_endrec;
288 spin_unlock(&mdd->mdd_cl.mc_lock);
290 return LLOG_PROC_BREAK;
293 struct changelog_orphan_data {
295 struct mdd_device *clod_mdd;
298 /* find oldest changelog record index */
299 static int changelog_detect_orphan_cb(const struct lu_env *env,
300 struct llog_handle *llh,
301 struct llog_rec_hdr *hdr, void *data)
303 struct changelog_orphan_data *clod = data;
304 struct mdd_device *mdd = clod->clod_mdd;
305 struct llog_changelog_rec *rec;
307 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
309 rec = container_of(hdr, typeof(*rec), cr_hdr);
310 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
311 CWARN("%s: invalid record at index %d in log "DFID"\n",
312 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
313 PLOGID(&llh->lgh_id));
314 /* try to find some next valid record and thus allow to recover
315 * from a corrupted LLOG, instead to assert and force a crash
321 "%s: record at index %d/%d/%llu t=%x %.*s in log "DFID"\n",
322 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
323 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
324 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
325 PLOGID(&llh->lgh_id));
327 clod->clod_index = rec->cr.cr_index;
329 return LLOG_PROC_BREAK;
332 /* find oldest changelog user index */
333 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
334 struct llog_handle *llh,
335 struct llog_rec_hdr *hdr, void *data)
337 struct changelog_orphan_data *clod = data;
338 struct mdd_device *mdd = clod->clod_mdd;
339 struct llog_changelog_user_rec2 *rec;
340 char user_name[CHANGELOG_USER_NAMELEN_FULL];
342 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
344 rec = container_of(hdr, typeof(*rec), cur_hdr);
345 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
346 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
347 CWARN("%s: unknown user type %u at index %u in log "DFID"\n",
348 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
349 rec->cur_hdr.lrh_type, PLOGID(&llh->lgh_id));
350 /* try to find some next valid record and thus allow to recover
351 * from a corrupted LLOG, instead to assert and force a crash
356 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
357 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
359 hdr->lrh_index, rec->cur_hdr.lrh_index,
360 rec->cur_endrec, PLOGID(&llh->lgh_id));
362 clod->clod_index = min_t(__u64, clod->clod_index, rec->cur_endrec);
367 struct changelog_cancel_cookie {
369 struct mdd_device *mdd;
372 static int llog_changelog_cancel_cb(const struct lu_env *env,
373 struct llog_handle *llh,
374 struct llog_rec_hdr *hdr, void *data)
376 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
377 struct changelog_cancel_cookie *cl_cookie =
378 (struct changelog_cancel_cookie *)data;
382 /* This is always a (sub)log, not the catalog */
383 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
385 /* if current context is GC-thread allow it to stop upon umount
386 * remaining records cleanup will occur upon next mount
388 * also during testing, wait for GC-thread to be released
390 * XXX this requires the GC-thread to not fork a sub-thread via
391 * llog[_cat]_process_or_fork() and we may think to also implement
392 * this shutdown mechanism for manually started user unregister which
393 * can also take a long time if huge backlog of records
395 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
396 /* wait to be released */
397 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
400 if (kthread_should_stop())
401 RETURN(LLOG_PROC_BREAK);
404 if (rec->cr.cr_index > cl_cookie->endrec)
405 /* records are in order, so we're done */
406 RETURN(LLOG_PROC_BREAK);
408 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
409 if (cfs_fail_val == 0)
410 cfs_fail_val = hdr->lrh_index;
411 if (cfs_fail_val == hdr->lrh_index)
412 CFS_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
415 /* Records folow one by one, cr_index++. We could calculate the
416 * last cr_index at this plain llog. And if it less then cookie endrec
417 * cancel the whole file.
419 if (llog_is_plain_skipable(llh->lgh_hdr, hdr, rec->cr.cr_index,
420 cl_cookie->endrec)) {
423 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
424 if (cfs_fail_val == 0) {
425 cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
426 CFS_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
429 rc = llog_destroy(env, llh);
431 CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
432 PLOGID(&llh->lgh_id));
433 RETURN(LLOG_DEL_PLAIN);
437 /* cancel them one at a time. I suppose we could store up the cookies
438 * and cancel them all at once; probably more efficient, but this is
439 * done as a user call, so who cares... */
441 RETURN(LLOG_DEL_RECORD);
444 static int llog_changelog_cancel(const struct lu_env *env,
445 struct llog_ctxt *ctxt,
446 struct changelog_cancel_cookie *cookie)
448 struct llog_handle *cathandle = ctxt->loc_handle;
453 /* This should only be called with the catalog handle */
454 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
456 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
459 /* 0 or 1 means we're done */
462 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
463 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
464 PLOGID(&cathandle->lgh_id), rc);
469 static struct llog_operations changelog_orig_logops;
472 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
476 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
480 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
483 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
484 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
485 CERROR("Changelogs cannot be enabled due to error "
486 "condition (see %s log).\n",
487 mdd2obd_dev(mdd)->obd_name);
490 spin_lock(&mdd->mdd_cl.mc_lock);
491 mdd->mdd_cl.mc_flags |= CLM_ON;
492 spin_unlock(&mdd->mdd_cl.mc_lock);
493 rc = mdd_changelog_write_header(env, mdd, CLM_START);
499 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
503 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
506 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
507 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
508 spin_lock(&mdd->mdd_cl.mc_lock);
509 mdd->mdd_cl.mc_flags &= ~CLM_ON;
510 spin_unlock(&mdd->mdd_cl.mc_lock);
515 static int mdd_changelog_llog_init(const struct lu_env *env,
516 struct mdd_device *mdd)
518 struct obd_device *obd = mdd2obd_dev(mdd);
519 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
520 struct changelog_orphan_data clod = {
531 /* LU-2844 mdd setup failure should not cause umount oops */
532 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
535 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
536 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
537 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
538 obd, &changelog_orig_logops);
540 CERROR("%s: changelog llog setup failed: rc = %d\n",
545 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
548 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
551 GOTO(out_cleanup, rc);
553 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
557 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
558 changelog_init_cb, mdd);
561 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
566 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
567 mdd->mdd_cl.mc_index);
569 /* setup user changelog */
570 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
571 obd, &changelog_orig_logops);
573 CERROR("%s: changelog users llog setup failed: rc = %d\n",
578 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
581 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
584 GOTO(out_ucleanup, rc);
586 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
588 GOTO(out_uclose, rc);
590 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
591 changelog_user_init_cb, mdd);
593 CERROR("%s: changelog user init failed: rc = %d\n",
595 GOTO(out_uclose, rc);
598 /* Finally apply per-server mask */
599 mdd->mdd_cl.mc_current_mask |= mdd->mdd_cl.mc_proc_mask;
601 /* If we have registered users, assume we want changelogs on */
602 if (mdd->mdd_cl.mc_lastuser > 0) {
603 rc = mdd_changelog_on(env, mdd);
605 GOTO(out_uclose, rc);
608 /* find and clear any orphan changelog records (1st record index <
609 * smallest of all users current index), likely to come from an
610 * interrupted manual or GC-thread purge, as its user record had
612 * XXX we may wait for a still registered user clear operation to
613 * do the job, but it may then take a long time to reach the user's
614 * real targetted records if a huge purge backlog is still to be
615 * processed as a long time idle user record could have been deleted
616 * XXX we may need to run end of purge as a separate thread
618 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
621 CERROR("%s: changelog detect orphan failed: rc = %d\n",
623 GOTO(out_uclose, rc);
625 rc = llog_cat_process(env, uctxt->loc_handle,
626 changelog_user_detect_orphan_cb,
629 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
631 GOTO(out_uclose, rc);
633 if (unlikely(clod.clod_index < user_orphan.clod_index)) {
634 struct changelog_cancel_cookie cl_cookie = {
635 .endrec = user_orphan.clod_index,
639 CWARN("%s : orphan changelog records found, starting from "
640 "index %llu to index %llu, being cleared now\n",
641 obd->obd_name, clod.clod_index, user_orphan.clod_index);
643 /* XXX we may need to run end of purge as a separate thread */
644 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
646 CERROR("%s: purge of changelog orphan records failed: "
647 "rc = %d\n", obd->obd_name, rc);
648 GOTO(out_uclose, rc);
653 llog_ctxt_put(uctxt);
656 llog_cat_close(env, uctxt->loc_handle);
658 llog_cleanup(env, uctxt);
660 llog_cat_close(env, ctxt->loc_handle);
662 llog_cleanup(env, ctxt);
666 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
668 struct obd_device *obd = mdd2obd_dev(mdd);
671 mdd->mdd_cl.mc_index = 0;
672 spin_lock_init(&mdd->mdd_cl.mc_lock);
673 mdd->mdd_cl.mc_starttime = ktime_get();
674 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
675 mdd->mdd_cl.mc_lastuser = 0;
677 /* ensure a GC check will, and a thread run may, occur upon start */
678 mdd->mdd_cl.mc_gc_time = 0;
679 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
680 mdd->mdd_cl.mc_mintime = (__u32)ktime_get_real_seconds();
681 mdd->mdd_cl.mc_minrec = ULLONG_MAX;
683 rc = mdd_changelog_llog_init(env, mdd);
685 CERROR("%s: changelog setup during init failed: rc = %d\n",
687 mdd->mdd_cl.mc_flags |= CLM_ERR;
693 static void mdd_changelog_fini(const struct lu_env *env,
694 struct mdd_device *mdd)
696 struct obd_device *obd = mdd2obd_dev(mdd);
697 struct llog_ctxt *ctxt;
699 if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
701 mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
704 /* stop GC-thread if running */
705 spin_lock(&mdd->mdd_cl.mc_lock);
706 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
707 /* avoid any attempt to run a GC-thread */
708 mdd->mdd_cl.mc_gc_task = current;
709 spin_unlock(&mdd->mdd_cl.mc_lock);
711 struct task_struct *gc_task;
713 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
714 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
715 /* need to wait for birthing GC-thread to be started
716 * and to have set mc_gc_task to itself
718 spin_unlock(&mdd->mdd_cl.mc_lock);
719 /* Add a tiny sleep */
720 schedule_timeout_uninterruptible(1);
721 /* go back to fully check if GC-thread has started or
722 * even already exited or if a new one is starting...
726 /* take a reference on task_struct to avoid it to be freed
729 gc_task = mdd->mdd_cl.mc_gc_task;
730 get_task_struct(gc_task);
731 spin_unlock(&mdd->mdd_cl.mc_lock);
732 kthread_stop(gc_task);
733 put_task_struct(gc_task);
736 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
738 llog_cat_close(env, ctxt->loc_handle);
739 llog_cleanup(env, ctxt);
741 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
743 llog_cat_close(env, ctxt->loc_handle);
744 llog_cleanup(env, ctxt);
749 * Remove entries with indicies up to and including \a endrec from the
756 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
757 unsigned long long endrec)
759 struct obd_device *obd = mdd2obd_dev(mdd);
760 struct llog_ctxt *ctxt;
761 unsigned long long cur;
762 struct changelog_cancel_cookie cookie;
765 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
769 spin_lock(&mdd->mdd_cl.mc_lock);
770 cur = (long long)mdd->mdd_cl.mc_index;
771 spin_unlock(&mdd->mdd_cl.mc_lock);
774 * If purging all records, write a header entry so we don't have an
775 * empty catalog and we're sure to have a valid starting index next
776 * time. In a case of crash, we just restart with old log so we're
780 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
787 * Some records were purged, so reset repeat-access time (so we
788 * record new mtime update records, so users can see a file has been
789 * changed since the last purge)
791 mdd->mdd_cl.mc_starttime = ktime_get();
793 cookie.endrec = endrec;
795 rc = llog_changelog_cancel(env, ctxt, &cookie);
801 /** Add a CL_MARK record to the changelog
803 * \param markerflags - CLM_*
806 int mdd_changelog_write_header(const struct lu_env *env,
807 struct mdd_device *mdd, int markerflags)
809 struct obd_device *obd = mdd2obd_dev(mdd);
810 struct llog_changelog_rec *rec;
812 struct llog_ctxt *ctxt;
814 int len = strlen(obd->obd_name);
819 if (mdd->mdd_cl.mc_current_mask & BIT(CL_MARK)) {
820 mdd->mdd_cl.mc_starttime = ktime_get();
824 reclen = llog_data_len(sizeof(*rec) + len);
825 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
826 if (buf->lb_buf == NULL)
830 rec->cr.cr_flags = CLF_VERSION;
831 rec->cr.cr_type = CL_MARK;
832 rec->cr.cr_namelen = len;
833 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
834 /* Status and action flags */
835 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
836 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
838 rec->cr_hdr.lrh_type = CHANGELOG_REC;
839 rec->cr.cr_time = cl_time();
841 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
844 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
849 /* assume on or off event; reset repeat-access time */
850 mdd->mdd_cl.mc_starttime = ktime_get();
855 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
856 * are valid. We also check if object with passed fid exists or not.
858 static int obf_lookup(const struct lu_env *env, struct md_object *p,
859 const struct lu_name *lname, struct lu_fid *f,
860 struct md_op_spec *spec)
862 char *name = (char *)lname->ln_name;
863 struct mdd_device *mdd = mdo2mdd(p);
864 struct mdd_object *child;
870 sscanf(name, SFID, RFID(f));
872 GOTO(out, rc = -ENOENT);
874 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
875 !fid_seq_is_dot(f->f_seq))
876 GOTO(out, rc = -ENOENT);
878 /* Check if object with this fid exists */
879 child = mdd_object_find(env, mdd, f);
881 GOTO(out, rc = PTR_ERR(child));
883 if (mdd_object_exists(child) == 0)
886 mdd_object_put(env, child);
892 static int mdd_dummy_create(const struct lu_env *env,
893 struct md_object *pobj,
894 const struct lu_name *lname,
895 struct md_object *child,
896 struct md_op_spec *spec,
902 static int mdd_dummy_rename(const struct lu_env *env,
903 struct md_object *src_pobj,
904 struct md_object *tgt_pobj,
905 const struct lu_fid *lf,
906 const struct lu_name *lsname,
907 struct md_object *tobj,
908 const struct lu_name *ltname,
914 static int mdd_dummy_link(const struct lu_env *env,
915 struct md_object *tgt_obj,
916 struct md_object *src_obj,
917 const struct lu_name *lname,
923 static int mdd_dummy_unlink(const struct lu_env *env,
924 struct md_object *pobj,
925 struct md_object *cobj,
926 const struct lu_name *lname,
933 int mdd_create(const struct lu_env *env, struct md_object *pobj,
934 const struct lu_name *lname, struct md_object *child,
935 struct md_op_spec *spec, struct md_attr *ma);
936 static int mdd_obf_create(const struct lu_env *env, struct md_object *pobj,
937 const struct lu_name *lname, struct md_object *child,
938 struct md_op_spec *spec, struct md_attr *ma)
940 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
941 return mdd_create(env, pobj, lname, child, spec, ma);
945 static const struct md_dir_operations mdd_obf_dir_ops = {
946 .mdo_lookup = obf_lookup,
947 .mdo_create = mdd_obf_create,
948 .mdo_rename = mdd_dummy_rename,
949 .mdo_link = mdd_dummy_link,
950 .mdo_unlink = mdd_dummy_unlink
953 static const struct md_dir_operations mdd_lpf_dir_ops = {
954 .mdo_lookup = mdd_lookup,
955 .mdo_create = mdd_dummy_create,
956 .mdo_rename = mdd_dummy_rename,
957 .mdo_link = mdd_dummy_link,
958 .mdo_unlink = mdd_dummy_unlink
961 static struct md_object *mdo_locate(const struct lu_env *env,
962 struct md_device *md,
963 const struct lu_fid *fid)
965 struct lu_object *obj;
966 struct md_object *mdo;
968 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
970 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
971 LASSERT(obj != NULL);
979 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
981 struct md_object *mdo;
982 struct mdd_object *mdd_lpf;
983 struct lu_fid fid = LU_LPF_FID;
987 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
988 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
993 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
995 RETURN(PTR_ERR(mdo));
997 LASSERT(lu_object_exists(&mdo->mo_lu));
999 mdd_lpf = md2mdd_obj(mdo);
1000 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
1001 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
1007 * Create special in-memory "fid" object for open-by-fid.
1009 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
1011 struct md_object *mdo;
1012 struct mdd_object *mdd_obf;
1013 struct lu_fid fid = LU_OBF_FID;
1016 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
1017 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
1021 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1023 RETURN(PTR_ERR(mdo));
1025 LASSERT(lu_object_exists(&mdo->mo_lu));
1027 mdd_obf = md2mdd_obj(mdo);
1028 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1029 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
1034 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
1035 struct mdd_device *m)
1037 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
1038 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
1039 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
1041 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
1042 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
1043 m->mdd_dot_lustre_objs.mdd_obf = NULL;
1045 if (m->mdd_dot_lustre != NULL) {
1046 mdd_object_put(env, m->mdd_dot_lustre);
1047 m->mdd_dot_lustre = NULL;
1051 /** Setup ".lustre" directory object */
1052 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1054 struct md_object *mdo;
1059 /* Create ".lustre" directory in ROOT. */
1060 fid = LU_DOT_LUSTRE_FID;
1061 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
1063 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
1067 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1069 RETURN(PTR_ERR(mdo));
1070 LASSERT(lu_object_exists(&mdo->mo_lu));
1072 m->mdd_dot_lustre = md2mdd_obj(mdo);
1074 rc = mdd_obf_setup(env, m);
1076 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
1077 mdd2obd_dev(m)->obd_name, rc);
1081 rc = mdd_lpf_setup(env, m);
1083 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
1084 mdd2obd_dev(m)->obd_name, rc);
1091 mdd_dot_lustre_cleanup(env, m);
1097 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
1098 * object in obd_device
1100 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
1101 struct mdd_device *m)
1103 struct obd_device *obd = mdd2obd_dev(m);
1104 struct llog_ctxt *ctxt = NULL;
1108 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1109 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1111 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1112 obd, &llog_common_cat_ops);
1114 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1119 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1122 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1125 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1127 GOTO(out_cleanup, rc);
1130 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1132 GOTO(out_close, rc);
1134 llog_ctxt_put(ctxt);
1138 llog_cat_close(env, ctxt->loc_handle);
1139 ctxt->loc_handle = NULL;
1141 llog_cleanup(env, ctxt);
1147 * cleanup the context created by llog_setup_named()
1149 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1150 struct mdd_device *m)
1152 struct obd_device *obd = mdd2obd_dev(m);
1153 struct llog_ctxt *lctxt;
1156 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1158 llog_cat_close(env, lctxt->loc_handle);
1159 lctxt->loc_handle = NULL;
1160 llog_cleanup(env, lctxt);
1166 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1167 struct lustre_cfg *cfg)
1169 barrier_deregister(m->mdd_bottom);
1170 lfsck_degister(env, m->mdd_bottom);
1171 mdd_hsm_actions_llog_fini(env, m);
1172 mdd_changelog_fini(env, m);
1173 mdd_orphan_index_fini(env, m);
1174 mdd_dot_lustre_cleanup(env, m);
1175 if (obd2obt(mdd2obd_dev(m))->obt_nodemap_config_file) {
1176 struct obd_device_target *obt = obd2obt(mdd2obd_dev(m));
1178 nm_config_file_deregister_tgt(env,
1179 obt->obt_nodemap_config_file);
1180 obt->obt_nodemap_config_file = NULL;
1182 if (m->mdd_los != NULL) {
1183 local_oid_storage_fini(env, m->mdd_los);
1186 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1188 if (m->mdd_child_exp)
1189 obd_disconnect(m->mdd_child_exp);
1192 static int mdd_process_config(const struct lu_env *env,
1193 struct lu_device *d, struct lustre_cfg *cfg)
1195 struct mdd_device *m = lu2mdd_dev(d);
1196 struct dt_device *dt = m->mdd_child;
1197 struct lu_device *next = &dt->dd_lu_dev;
1201 switch (cfg->lcfg_command) {
1205 count = class_modify_config(cfg, PARAM_MDD, &m->mdd_kobj);
1206 rc = count > 0 ? 0 : count;
1208 /* we don't understand; pass it on */
1209 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1213 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1216 dt_conf_get(env, dt, &m->mdd_dt_conf);
1218 case LCFG_PRE_CLEANUP:
1219 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1220 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1223 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1224 lu_dev_del_linkage(d->ld_site, d);
1225 mdd_device_shutdown(env, m, cfg);
1228 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1235 static int mdd_recovery_complete(const struct lu_env *env,
1236 struct lu_device *d)
1238 struct mdd_device *mdd = lu2mdd_dev(d);
1239 struct lu_device *next;
1243 LASSERT(mdd != NULL);
1244 next = &mdd->mdd_child->dd_lu_dev;
1246 if (!mdd->mdd_bottom->dd_rdonly)
1247 mdd_orphan_cleanup(env, mdd);
1248 rc = next->ld_ops->ldo_recovery_complete(env, next);
1253 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1254 const struct lu_fid *pfid, const char *name,
1255 __u32 mode, struct lu_fid *fid)
1257 struct dt_object *parent, *dto;
1262 LASSERT(!fid_is_zero(pfid));
1263 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1264 if (unlikely(IS_ERR(parent)))
1265 RETURN(PTR_ERR(parent));
1267 /* create local file/dir, if @fid is passed then try to use it */
1268 if (fid_is_zero(fid))
1269 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1272 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1276 GOTO(out_put, rc = PTR_ERR(dto));
1277 *fid = *lu_object_fid(&dto->do_lu);
1278 /* since stack is not fully set up the local_storage uses own stack
1279 * and we should drop its object from cache */
1280 dt_object_put_nocache(env, dto);
1283 dt_object_put(env, parent);
1287 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1288 enum lfsck_events event)
1293 static int mdd_prepare(const struct lu_env *env,
1294 struct lu_device *pdev,
1295 struct lu_device *cdev)
1297 struct mdd_device *mdd = lu2mdd_dev(cdev);
1298 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1299 struct nm_config_file *nodemap_config;
1300 struct obd_device_target *obt = obd2obt(mdd2obd_dev(mdd));
1301 struct dt_object *root = NULL;
1302 struct thandle *th = NULL;
1308 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1312 /* Setup local dirs */
1313 fid.f_seq = FID_SEQ_LOCAL_NAME;
1316 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1321 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1326 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1327 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1328 mdd_root_dir_name, S_IFDIR |
1329 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1331 CERROR("%s: create root fid failed: rc = %d\n",
1332 mdd2obd_dev(mdd)->obd_name, rc);
1336 /* store a default directory layout on the root directory if
1337 * it doesn't already exist to improve MDT space balance.
1339 root = dt_locate(env, mdd->mdd_bottom, &fid);
1340 if (unlikely(IS_ERR(root)))
1341 GOTO(out_los, rc = PTR_ERR(root));
1343 rc = dt_xattr_get(env, root, &LU_BUF_NULL,
1344 XATTR_NAME_DEFAULT_LMV);
1345 if (rc == -ENODATA) {
1347 struct lmv_user_md lmv_default = {
1348 .lum_magic = LMV_USER_MAGIC,
1349 .lum_stripe_count = 1,
1350 .lum_stripe_offset = root_stripe_offset,
1351 .lum_max_inherit = root_max_inherit,
1352 .lum_max_inherit_rr = root_max_inherit_rr,
1355 th = dt_trans_create(env, mdd->mdd_bottom);
1357 GOTO(out_root_put, rc = PTR_ERR(th));
1359 buf.lb_buf = &lmv_default;
1360 buf.lb_len = sizeof(lmv_default);
1361 rc = dt_declare_xattr_set(env, root, &buf,
1362 XATTR_NAME_DEFAULT_LMV, 0,
1365 GOTO(out_trans_stop, rc);
1367 rc = dt_trans_start_local(env, mdd->mdd_bottom, th);
1369 GOTO(out_trans_stop, rc);
1371 rc = dt_xattr_set(env, root, &buf,
1372 XATTR_NAME_DEFAULT_LMV, 0, th);
1374 GOTO(out_trans_stop, rc);
1376 dt_trans_stop(env, mdd->mdd_bottom, th);
1378 } else if (rc < 0 && rc != -ERANGE) {
1379 CERROR("%s: get default LMV of root failed: rc = %d\n",
1380 mdd2obd_dev(mdd)->obd_name, rc);
1382 GOTO(out_root_put, rc);
1385 dt_object_put(env, root);
1388 mdd->mdd_root_fid = fid;
1389 rc = mdd_dot_lustre_setup(env, mdd);
1391 CERROR("%s: initializing .lustre failed: rc = %d\n",
1392 mdd2obd_dev(mdd)->obd_name, rc);
1396 /* Normal client usually send root access to MDT0 directly,
1397 * the root FID on non-MDT0 will only be used by echo client. */
1398 mdd->mdd_root_fid = fid;
1401 rc = mdd_orphan_index_init(env, mdd);
1405 rc = mdd_changelog_init(env, mdd);
1407 CERROR("%s: failed to initialize changelog: rc = %d\n",
1408 mdd2obd_dev(mdd)->obd_name, rc);
1412 rc = mdd_hsm_actions_llog_init(env, mdd);
1414 GOTO(out_changelog, rc);
1416 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1418 if (IS_ERR(nodemap_config)) {
1419 rc = PTR_ERR(nodemap_config);
1423 obt->obt_nodemap_config_file = nodemap_config;
1426 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1427 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1430 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1431 mdd2obd_dev(mdd)->obd_name, rc);
1432 GOTO(out_nodemap, rc);
1435 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1437 CERROR("%s: failed to register to barrier: rc = %d\n",
1438 mdd2obd_dev(mdd)->obd_name, rc);
1439 GOTO(out_lfsck, rc);
1445 lfsck_degister(env, mdd->mdd_bottom);
1447 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1448 obt->obt_nodemap_config_file = NULL;
1450 mdd_hsm_actions_llog_fini(env, mdd);
1452 mdd_changelog_fini(env, mdd);
1454 mdd_orphan_index_fini(env, mdd);
1456 if (mdd_seq_site(mdd)->ss_node_id == 0)
1457 mdd_dot_lustre_cleanup(env, mdd);
1460 dt_trans_stop(env, mdd->mdd_bottom, th);
1463 dt_object_put(env, root);
1465 local_oid_storage_fini(env, mdd->mdd_los);
1466 mdd->mdd_los = NULL;
1472 * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
1474 * Find corresponding device by passed parent and name, and allocate FID from
1477 * see include/lu_object.h for the details.
1479 static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1480 struct lu_fid *fid, struct lu_object *parent,
1481 const struct lu_name *name)
1483 struct mdd_device *mdd = lu2mdd_dev(d);
1484 struct lu_object *o = lu_object_next(parent);
1486 return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
1489 const struct lu_device_operations mdd_lu_ops = {
1490 .ldo_object_alloc = mdd_object_alloc,
1491 .ldo_process_config = mdd_process_config,
1492 .ldo_recovery_complete = mdd_recovery_complete,
1493 .ldo_prepare = mdd_prepare,
1494 .ldo_fid_alloc = mdd_fid_alloc,
1497 static int mdd_root_get(const struct lu_env *env,
1498 struct md_device *m, struct lu_fid *f)
1500 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1503 *f = mdd->mdd_root_fid;
1508 * No permission check is needed.
1510 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1511 struct obd_statfs *sfs)
1513 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1518 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs, NULL);
1520 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1525 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1526 struct md_device *m)
1528 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1530 return &mdd->mdd_dt_conf;
1533 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1536 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1538 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1539 return (*h == NULL ? -ENOENT : 0);
1542 static struct lu_device *mdd_device_free(const struct lu_env *env,
1543 struct lu_device *lu)
1545 struct mdd_device *m = lu2mdd_dev(lu);
1548 LASSERT(atomic_read(&lu->ld_ref) == 0);
1549 md_device_fini(&m->mdd_md_dev);
1554 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1555 struct lu_device_type *t,
1556 struct lustre_cfg *lcfg)
1558 struct lu_device *l;
1559 struct mdd_device *m;
1563 l = ERR_PTR(-ENOMEM);
1568 md_device_init(&m->mdd_md_dev, t);
1569 rc = mdd_init0(env, m, t, lcfg);
1571 mdd_device_free(env, l);
1580 * we use exports to track all mdd users
1582 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1583 struct obd_device *obd, struct obd_uuid *cluuid,
1584 struct obd_connect_data *data, void *localdata)
1586 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1587 struct lustre_handle conn;
1591 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1593 rc = class_connect(&conn, obd, cluuid);
1597 *exp = class_conn2export(&conn);
1599 /* Why should there ever be more than 1 connect? */
1600 LASSERT(mdd->mdd_connects == 0);
1601 mdd->mdd_connects++;
1607 * once last export (we don't count self-export) disappeared
1608 * mdd can be released
1610 static int mdd_obd_disconnect(struct obd_export *exp)
1612 struct obd_device *obd = exp->exp_obd;
1613 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1614 int rc, release = 0;
1617 mdd->mdd_connects--;
1618 if (mdd->mdd_connects == 0)
1621 rc = class_disconnect(exp);
1623 if (rc == 0 && release)
1624 class_manual_cleanup(obd);
1628 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1629 __u32 keylen, void *key, __u32 *vallen, void *val)
1633 if (KEY_IS(KEY_OSP_CONNECTED)) {
1634 struct obd_device *obd = exp->exp_obd;
1635 struct mdd_device *mdd;
1637 if (!obd->obd_set_up || obd->obd_stopping)
1640 mdd = lu2mdd_dev(obd->obd_lu_dev);
1642 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1650 static int mdd_obd_set_info_async(const struct lu_env *env,
1651 struct obd_export *exp,
1652 __u32 keylen, void *key,
1653 __u32 vallen, void *val,
1654 struct ptlrpc_request_set *set)
1656 struct obd_device *obd = exp->exp_obd;
1657 struct mdd_device *mdd;
1660 if (!obd->obd_set_up || obd->obd_stopping)
1663 mdd = lu2mdd_dev(obd->obd_lu_dev);
1665 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1670 static const struct obd_ops mdd_obd_device_ops = {
1671 .o_owner = THIS_MODULE,
1672 .o_connect = mdd_obd_connect,
1673 .o_disconnect = mdd_obd_disconnect,
1674 .o_get_info = mdd_obd_get_info,
1675 .o_set_info_async = mdd_obd_set_info_async,
1678 struct mdd_changelog_name_check_data {
1679 const char *mcnc_name;
1684 * changelog_recalc_mask callback
1686 * Is is called per each registered user and calculates combined mask of
1687 * all registered users.
1689 static int mdd_changelog_name_check_cb(const struct lu_env *env,
1690 struct llog_handle *llh,
1691 struct llog_rec_hdr *hdr, void *data)
1693 struct llog_changelog_user_rec2 *rec;
1694 struct mdd_changelog_name_check_data *mcnc = data;
1696 rec = container_of(hdr, typeof(*rec), cur_hdr);
1697 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
1698 !strncmp(rec->cur_name, mcnc->mcnc_name, sizeof(rec->cur_name))) {
1699 mcnc->mcnc_id = rec->cur_id;
1705 static int mdd_changelog_name_check(const struct lu_env *env,
1706 struct llog_ctxt *ctxt,
1707 struct mdd_device *mdd, const char *name)
1709 struct mdd_changelog_name_check_data mcnc = { .mcnc_name = name, };
1715 /* first symbol is a letter */
1716 if (!isalpha(name[0])) {
1718 CERROR("%s: first char '%c' in '%s' is not letter: rc = %d\n",
1719 mdd2obd_dev(mdd)->obd_name, name[0], name, rc);
1723 /* name is valid: contains letters, numbers and '-', '_' only */
1724 while (name[++chr]) {
1725 if (!(isalnum(name[chr]) || name[chr] == '_' ||
1726 name[chr] == '-')) {
1728 CERROR("%s: wrong char '%c' in name '%s': rc = %d\n",
1729 mdd2obd_dev(mdd)->obd_name, name[chr], name, rc);
1734 if (chr > CHANGELOG_USER_NAMELEN) {
1736 CERROR("%s: name '%s' is over %d symbols limit: rc = %d\n",
1737 mdd2obd_dev(mdd)->obd_name, name,
1738 CHANGELOG_USER_NAMELEN, rc);
1742 rc = llog_cat_process(env, ctxt->loc_handle,
1743 mdd_changelog_name_check_cb, &mcnc, 0, 0);
1745 CWARN("%s: changelog name %s exists already: rc = %d\n",
1746 mdd2obd_dev(mdd)->obd_name, name, rc);
1748 CWARN("%s: failed user changelog processing: rc = %d\n",
1749 mdd2obd_dev(mdd)->obd_name, rc);
1753 static int mdd_changelog_user_register(const struct lu_env *env,
1754 struct mdd_device *mdd, int *id,
1755 const char *name, const char *mask)
1757 struct llog_ctxt *ctxt;
1758 struct llog_changelog_user_rec2 *rec;
1759 char user_name[CHANGELOG_USER_NAMELEN_FULL];
1764 ctxt = llog_get_context(mdd2obd_dev(mdd),
1765 LLOG_CHANGELOG_USER_ORIG_CTXT);
1771 llog_ctxt_put(ctxt);
1775 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1777 rec->cur_hdr.lrh_len = sizeof(*rec);
1778 /* keep old record type for users without mask/name for
1779 * compatibility needs
1781 if (mask || (name && name[0]))
1782 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
1784 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1785 spin_lock(&mdd->mdd_cl.mc_user_lock);
1786 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1787 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1789 CERROR("%s: registering %s user: max ID is exceeded: rc = %d\n",
1790 mdd2obd_dev(mdd)->obd_name,
1791 (name && name[0]) ? name : "new", rc);
1794 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1795 mdd->mdd_cl.mc_users++;
1796 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1798 rec->cur_time = (__u32)ktime_get_real_seconds();
1799 if (CFS_FAIL_PRECHECK(OBD_FAIL_TIME_IN_CHLOG_USER)) {
1800 rec->cur_time -= min(cfs_fail_val, rec->cur_time);
1801 spin_lock(&mdd->mdd_cl.mc_user_lock);
1802 mdd->mdd_cl.mc_mintime = rec->cur_time;
1803 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1806 spin_lock(&mdd->mdd_cl.mc_lock);
1807 rec->cur_endrec = mdd->mdd_cl.mc_index;
1808 spin_unlock(&mdd->mdd_cl.mc_lock);
1811 /* if user will use relative mask apply it on default one */
1812 rec->cur_mask = CHANGELOG_DEFMASK;
1813 rc = cfs_str2mask(mask, changelog_type2str, &rec->cur_mask,
1814 CHANGELOG_MINMASK, CHANGELOG_ALLMASK,
1817 GOTO(out_users, rc);
1818 } else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK) {
1819 /* a maskless users means default mask but only if server has
1820 * no specific mask set
1822 rec->cur_mask = CHANGELOG_DEFMASK;
1825 if (name && name[0]) {
1826 rc = mdd_changelog_name_check(env, ctxt, mdd, name);
1828 GOTO(out_users, rc);
1829 strlcpy(rec->cur_name, name, sizeof(rec->cur_name));
1831 mdd_chlg_username(rec, user_name, sizeof(user_name));
1833 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1835 CWARN("%s: failed to register changelog user %s: rc = %d\n",
1836 mdd2obd_dev(mdd)->obd_name, user_name, rc);
1837 GOTO(out_users, rc);
1840 /* apply user mask finally */
1841 spin_lock(&mdd->mdd_cl.mc_user_lock);
1842 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
1843 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1845 CDEBUG(D_IOCTL, "%s: registered changelog user '%s', mask %#x\n",
1846 mdd2obd_dev(mdd)->obd_name, user_name, rec->cur_mask);
1848 /* Assume we want it on since somebody registered */
1849 rc = mdd_changelog_on(env, mdd);
1851 /* record is added, so don't decrement users on error */
1855 spin_lock(&mdd->mdd_cl.mc_user_lock);
1856 mdd->mdd_cl.mc_users--;
1857 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1861 llog_ctxt_put(ctxt);
1865 struct mdd_changelog_recalc_mask_data {
1866 struct mdd_device *mcrm_mdd;
1871 * changelog_recalc_mask callback
1873 * Is is called per each registered user and calculates combined mask of
1874 * all registered users.
1876 static int mdd_changelog_recalc_mask_cb(const struct lu_env *env,
1877 struct llog_handle *llh,
1878 struct llog_rec_hdr *hdr, void *data)
1880 struct llog_changelog_user_rec2 *rec;
1881 struct mdd_changelog_recalc_mask_data *mcrm = data;
1883 rec = container_of(hdr, typeof(*rec), cur_hdr);
1884 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
1885 mcrm->mcrm_mask |= rec->cur_mask;
1886 else if (mcrm->mcrm_mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
1887 mcrm->mcrm_mask |= CHANGELOG_DEFMASK;
1892 int mdd_changelog_recalc_mask(const struct lu_env *env, struct mdd_device *mdd)
1894 struct llog_ctxt *ctxt;
1895 struct mdd_changelog_recalc_mask_data mcrm = {
1897 .mcrm_mask = mdd->mdd_cl.mc_proc_mask,
1903 ctxt = llog_get_context(mdd2obd_dev(mdd),
1904 LLOG_CHANGELOG_USER_ORIG_CTXT);
1908 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1909 GOTO(out, rc = -ENXIO);
1911 rc = llog_cat_process(env, ctxt->loc_handle,
1912 mdd_changelog_recalc_mask_cb, &mcrm, 0, 0);
1914 CWARN("%s: failed user changelog processing: rc = %d\n",
1915 mdd2obd_dev(mdd)->obd_name, rc);
1917 spin_lock(&mdd->mdd_cl.mc_user_lock);
1918 CDEBUG(D_INFO, "%s: recalc changelog mask: %#x -> %#x\n",
1919 mdd2obd_dev(mdd)->obd_name, mdd->mdd_cl.mc_current_mask,
1921 mdd->mdd_cl.mc_current_mask = mcrm.mcrm_mask;
1922 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1926 llog_ctxt_put(ctxt);
1931 struct mdd_changelog_user_purge {
1932 struct mdd_device *mcup_mdd;
1934 __u32 mcup_usercount;
1937 char mcup_name[CHANGELOG_USER_NAMELEN_FULL];
1941 * changelog_user_purge callback
1943 * Is called once per user.
1945 * Check to see the user requested is available from rec.
1946 * Truncate the changelog.
1947 * Keep track of the total number of users (calls).
1949 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1950 struct llog_handle *llh,
1951 struct llog_rec_hdr *hdr, void *data)
1953 struct llog_changelog_user_rec2 *rec;
1954 struct mdd_changelog_user_purge *mcup = data;
1955 struct llog_cookie cookie;
1960 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1963 rec = container_of(hdr, typeof(*rec), cur_hdr);
1965 mcup->mcup_usercount++;
1967 if (rec->cur_id != mcup->mcup_id) {
1968 /* truncate to the lowest endrec that is not this user */
1969 mcup->mcup_minrec = min(mcup->mcup_minrec, rec->cur_endrec);
1973 mdd_chlg_username(rec, mcup->mcup_name, sizeof(mcup->mcup_name));
1975 /* Unregister this user */
1976 cookie.lgc_lgl = llh->lgh_id;
1977 cookie.lgc_index = hdr->lrh_index;
1978 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1981 mcup->mcup_found = true;
1982 mcup->mcup_usercount--;
1983 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1984 mcup->mcup_mdd->mdd_cl.mc_users--;
1985 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1991 int mdd_changelog_user_purge(const struct lu_env *env,
1992 struct mdd_device *mdd, __u32 id)
1994 struct mdd_changelog_user_purge mcup = {
1997 .mcup_found = false,
1998 .mcup_usercount = 0,
1999 .mcup_minrec = ULLONG_MAX,
2002 struct llog_ctxt *ctxt;
2007 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
2008 mdd2obd_dev(mdd)->obd_name, id);
2010 ctxt = llog_get_context(mdd2obd_dev(mdd),
2011 LLOG_CHANGELOG_USER_ORIG_CTXT);
2014 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
2015 GOTO(out, rc = -ENXIO);
2017 rc = llog_cat_process(env, ctxt->loc_handle,
2018 mdd_changelog_user_purge_cb, &mcup, 0, 0);
2020 CWARN("%s: failed to purge changelog for user %s: rc = %d\n",
2021 mdd2obd_dev(mdd)->obd_name, mcup.mcup_name, rc);
2025 CFS_FAIL_TIMEOUT(OBD_FAIL_LLOG_PURGE_DELAY, cfs_fail_val);
2026 if (mcup.mcup_usercount == 0) {
2027 spin_lock(&mdd->mdd_cl.mc_user_lock);
2028 if (mdd->mdd_cl.mc_users == 0) {
2029 /* No more users; turn changelogs off */
2030 CDEBUG(D_IOCTL, "turning off changelogs\n");
2031 rc = mdd_changelog_off(env, mdd);
2033 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2036 if (mcup.mcup_found) {
2038 "%s: Purge changelog entries for user %s record=%llu\n",
2039 mdd2obd_dev(mdd)->obd_name,
2040 mcup.mcup_name, mcup.mcup_minrec);
2041 rc = mdd_changelog_llog_cancel(env, mdd, mcup.mcup_minrec);
2043 CWARN("%s: No changelog for user id %u: rc = %d\n",
2044 mdd2obd_dev(mdd)->obd_name, id, rc);
2045 GOTO(out, rc = -ENOENT);
2048 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
2052 llog_ctxt_put(ctxt);
2057 struct mdd_changelog_user_clear {
2063 struct mdd_device *mcuc_mdd;
2064 char mcuc_name[CHANGELOG_USER_NAMELEN_FULL];
2068 * changelog_clear callback
2070 * Is called once per user.
2072 * Check to see the user requested is available from rec.
2073 * Check the oldest (smallest) record for boundary conditions.
2074 * Truncate the changelog.
2076 static int mdd_changelog_clear_cb(const struct lu_env *env,
2077 struct llog_handle *llh,
2078 struct llog_rec_hdr *hdr,
2081 struct llog_changelog_user_rec2 *rec;
2082 struct mdd_changelog_user_clear *mcuc = data;
2083 struct mdd_device *mdd = mcuc->mcuc_mdd;
2088 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
2091 rec = container_of(hdr, typeof(*rec), cur_hdr);
2092 /* Does the changelog id match the requested id? */
2093 if (rec->cur_id != mcuc->mcuc_id) {
2094 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec, rec->cur_endrec);
2095 mcuc->mcuc_mintime = min(mcuc->mcuc_mintime, rec->cur_time);
2099 mdd_chlg_username(rec, mcuc->mcuc_name, sizeof(mcuc->mcuc_name));
2100 /* cur_endrec is the oldest purgeable record, make sure we're newer */
2101 if (rec->cur_endrec > mcuc->mcuc_endrec) {
2104 "%s: request %llu < endrec %llu for user %s: rc = %d\n",
2105 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_endrec,
2106 rec->cur_endrec, mcuc->mcuc_name, rc);
2111 * Flag that we've met all the range and user checks.
2112 * We now know the record to flush.
2114 mcuc->mcuc_flush = true;
2116 rec->cur_endrec = mcuc->mcuc_endrec;
2117 rec->cur_time = (__u32)ktime_get_real_seconds();
2119 CDEBUG(D_IOCTL, "%s: update changelog user %s endrec = %llu\n",
2120 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_name, rec->cur_endrec);
2122 /* Update the endrec */
2123 rc = llog_write(env, llh, hdr, hdr->lrh_index);
2129 * Clear a changelog up to entry specified by endrec for user id.
2131 static int mdd_changelog_clear(const struct lu_env *env,
2132 struct mdd_device *mdd, __u32 id,
2135 struct mdd_changelog_user_clear mcuc = {
2137 .mcuc_minrec = endrec,
2138 .mcuc_flush = false,
2140 .mcuc_mintime = ktime_get_real_seconds(),
2143 struct llog_ctxt *ctxt;
2149 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
2150 mdd2obd_dev(mdd)->obd_name, id, endrec);
2151 /* start_rec is the newest (largest value) entry in the changelogs*/
2152 spin_lock(&mdd->mdd_cl.mc_lock);
2153 start_rec = mdd->mdd_cl.mc_index;
2154 spin_unlock(&mdd->mdd_cl.mc_lock);
2156 if (start_rec < endrec) {
2157 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
2158 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
2163 mcuc.mcuc_endrec = start_rec;
2164 mcuc.mcuc_minrec = ULLONG_MAX;
2166 mcuc.mcuc_endrec = endrec;
2169 ctxt = llog_get_context(mdd2obd_dev(mdd),
2170 LLOG_CHANGELOG_USER_ORIG_CTXT);
2174 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
2175 GOTO(out, rc = -ENXIO);
2177 rc = llog_cat_process(env, ctxt->loc_handle, mdd_changelog_clear_cb,
2179 if (rc == -EINVAL) {
2180 CDEBUG(D_IOCTL, "%s: no changelog recnum <= %llu to clear\n",
2181 mdd2obd_dev(mdd)->obd_name, (unsigned long long)endrec);
2186 CWARN("%s: can't clear the changelog for user %s: rc = %d\n",
2187 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, rc);
2191 if (!mcuc.mcuc_flush) {
2192 CDEBUG(D_IOCTL, "%s: no entry for user %d\n",
2193 mdd2obd_dev(mdd)->obd_name, id);
2194 GOTO(out, rc = -ENOENT);
2197 CDEBUG(D_IOCTL, "%s: purge changelog user %s entries up to %llu\n",
2198 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, mcuc.mcuc_minrec);
2200 rc = mdd_changelog_llog_cancel(env, mdd, mcuc.mcuc_minrec);
2204 spin_lock(&mdd->mdd_cl.mc_user_lock);
2205 mdd->mdd_cl.mc_minrec = mcuc.mcuc_minrec;
2206 mdd->mdd_cl.mc_mintime = mcuc.mcuc_mintime;
2207 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2211 llog_ctxt_put(ctxt);
2216 static int mdd_changelog_user_deregister(const struct lu_env *env,
2217 struct mdd_device *mdd, int *id,
2220 struct llog_ctxt *ctxt;
2221 struct mdd_changelog_name_check_data mcnc = {
2230 ctxt = llog_get_context(mdd2obd_dev(mdd),
2231 LLOG_CHANGELOG_USER_ORIG_CTXT);
2235 rc = llog_cat_process(env, ctxt->loc_handle,
2236 mdd_changelog_name_check_cb, &mcnc, 0, 0);
2237 llog_ctxt_put(ctxt);
2239 if (rc != -EEXIST) {
2240 CDEBUG(D_IOCTL, "%s: no entry for username %s\n",
2241 mdd2obd_dev(mdd)->obd_name, name);
2247 /* explicitly clear changelog first, to protect from crash in
2248 * the middle of purge that would lead to unregistered consumer
2249 * but pending changelog entries
2251 rc = mdd_changelog_clear(env, mdd, *id, 0);
2253 rc = mdd_changelog_user_purge(env, mdd, *id);
2255 /* recalc changelog current mask */
2256 mdd_changelog_recalc_mask(env, mdd);
2262 * May be called remotely from mdt_iocontrol_handle or locally from
2263 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
2264 * an obd_ioctl_data format (but local ioctl handler does).
2266 * \param len - data len
2267 * \param karg - ioctl data, in kernel space
2269 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
2270 unsigned int cmd, int len, void *karg)
2272 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
2273 struct obd_device *obd = mdd2obd_dev(mdd);
2274 struct obd_ioctl_data *data;
2278 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK\n",
2279 obd->obd_name, cmd, len, karg);
2280 if (unlikely(karg == NULL))
2281 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc));
2284 /* Doesn't use obd_ioctl_data */
2286 case OBD_IOC_CHANGELOG_CLEAR: {
2287 struct changelog_setinfo *cs = karg;
2289 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2290 RETURN(-EINPROGRESS);
2292 rc = mdd_changelog_clear(env, mdd, cs->cs_id, cs->cs_recno);
2293 barrier_exit(mdd->mdd_bottom);
2296 case OBD_IOC_START_LFSCK: {
2297 rc = lfsck_start(env, mdd->mdd_bottom, karg);
2300 case OBD_IOC_STOP_LFSCK: {
2301 rc = lfsck_stop(env, mdd->mdd_bottom, karg);
2304 case OBD_IOC_QUERY_LFSCK: {
2305 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL, karg);
2308 case OBD_IOC_LLOG_PRINT:
2309 case OBD_IOC_LLOG_CANCEL:
2310 rc = obd_iocontrol(cmd, mdd->mdd_child_exp, len, karg, NULL);
2314 /* Below ioctls use obd_ioctl_data */
2315 if (data->ioc_version != OBD_IOCTL_VERSION) {
2316 CERROR("%s: iocontrol from '%s' bad magic %x != %x: rc = %d\n",
2317 obd->obd_name, current->comm,
2318 data->ioc_version, OBD_IOCTL_VERSION, rc);
2323 case OBD_IOC_CHANGELOG_REG:
2324 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2325 RETURN(-EINPROGRESS);
2327 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1,
2330 barrier_exit(mdd->mdd_bottom);
2332 case OBD_IOC_CHANGELOG_DEREG:
2333 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2334 RETURN(-EINPROGRESS);
2336 rc = mdd_changelog_user_deregister(env, mdd, &data->ioc_u32_1,
2339 barrier_exit(mdd->mdd_bottom);
2342 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
2349 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
2350 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
2352 static const struct md_device_operations mdd_ops = {
2353 .mdo_statfs = mdd_statfs,
2354 .mdo_root_get = mdd_root_get,
2355 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
2356 .mdo_iocontrol = mdd_iocontrol,
2357 .mdo_dtconf_get = mdd_dtconf_get,
2360 static const struct lu_device_type_operations mdd_device_type_ops = {
2361 .ldto_init = mdd_type_init,
2362 .ldto_fini = mdd_type_fini,
2364 .ldto_start = mdd_type_start,
2365 .ldto_stop = mdd_type_stop,
2367 .ldto_device_alloc = mdd_device_alloc,
2368 .ldto_device_free = mdd_device_free,
2370 .ldto_device_fini = mdd_device_fini
2373 static struct lu_device_type mdd_device_type = {
2374 .ldt_tags = LU_DEVICE_MD,
2375 .ldt_name = LUSTRE_MDD_NAME,
2376 .ldt_ops = &mdd_device_type_ops,
2377 .ldt_ctx_tags = LCT_MD_THREAD
2380 /* context key constructor: mdd_key_init */
2381 LU_KEY_INIT(mdd, struct mdd_thread_info);
2383 static void mdd_key_fini(const struct lu_context *ctx,
2384 struct lu_context_key *key, void *data)
2386 struct mdd_thread_info *info = data;
2388 lu_buf_free(&info->mdi_big_buf);
2389 lu_buf_free(&info->mdi_link_buf);
2390 lu_buf_free(&info->mdi_xattr_buf);
2391 lu_buf_free(&info->mdi_chlg_buf);
2396 /* context key: mdd_thread_key */
2397 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
2399 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
2400 int (*func)(void *), void *data, char *name)
2402 struct task_struct *task;
2404 LASSERT(thread->mgt_init == false);
2405 init_completion(&thread->mgt_started);
2406 init_completion(&thread->mgt_finished);
2407 thread->mgt_data = data;
2408 thread->mgt_abort = false;
2409 thread->mgt_init = true;
2411 task = kthread_run(func, thread, name);
2413 complete(&thread->mgt_finished);
2414 return PTR_ERR(task);
2416 wait_for_completion(&thread->mgt_started);
2420 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
2422 if (thread->mgt_init == true) {
2423 thread->mgt_abort = true;
2424 wait_for_completion(&thread->mgt_finished);
2428 static int __init mdd_init(void)
2432 rc = lu_kmem_init(mdd_caches);
2436 changelog_orig_logops = llog_common_cat_ops;
2437 changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
2439 rc = class_register_type(&mdd_obd_device_ops, NULL, false,
2440 LUSTRE_MDD_NAME, &mdd_device_type);
2442 lu_kmem_fini(mdd_caches);
2446 static void __exit mdd_exit(void)
2448 class_unregister_type(LUSTRE_MDD_NAME);
2449 lu_kmem_fini(mdd_caches);
2452 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2453 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2454 MODULE_VERSION(LUSTRE_VERSION_STRING);
2455 MODULE_LICENSE("GPL");
2457 module_init(mdd_init);
2458 module_exit(mdd_exit);