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;
104 LASSERT(m->mdd_child_exp == NULL);
106 OBD_ALLOC(data, sizeof(*data));
108 GOTO(out, rc = -ENOMEM);
110 obd = class_name2obd(nextdev);
112 CERROR("can't locate next device: %s\n", nextdev);
113 GOTO(out, rc = -ENOTCONN);
116 data->ocd_connect_flags = OBD_CONNECT_VERSION;
117 data->ocd_version = LUSTRE_VERSION_CODE;
119 rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data,
122 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
126 lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
127 LASSERT(lud->ld_site);
128 m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
129 m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
130 lu_dev_add_linkage(lud->ld_site, lud);
134 OBD_FREE(data, sizeof(*data));
138 static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
139 struct lu_device_type *t, struct lustre_cfg *lcfg)
146 /* LU-8040 Set defaults here, before values configs */
147 mdd->mdd_cl.mc_flags = 0; /* off by default */
148 /* per-server mask is set via parameters if needed */
149 mdd->mdd_cl.mc_proc_mask = CHANGELOG_MINMASK;
150 /* current mask is calculated from mask above and users masks */
151 mdd->mdd_cl.mc_current_mask = CHANGELOG_MINMASK;
152 mdd->mdd_cl.mc_deniednext = 60; /* 60 secs by default */
153 mdd->mdd_cl.mc_enable_shard_pfid = false; /* master pFID by default */
155 dev = lustre_cfg_string(lcfg, 0);
159 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
160 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
162 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
163 mdd->mdd_md_dev.md_ops = &mdd_ops;
165 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
169 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
170 /* sync permission changes */
171 mdd->mdd_sync_permission = 1;
172 /* enable changelog garbage collection */
173 mdd->mdd_changelog_gc = 1;
174 /* enable changelog cleanup due to lack of space */
175 mdd->mdd_changelog_free_space_gc = true;
176 /* set when emergency GC is started */
177 mdd->mdd_changelog_emrg_gc = false;
178 /* with a significant amount of idle time */
179 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
180 /* or a significant amount of late indexes */
181 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
182 /* with a reasonable interval between each check */
183 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
184 /* with a very few number of free catalog entries */
185 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
186 /* special default striping for files created with O_APPEND */
187 mdd->mdd_append_stripe_count = 1;
188 mdd->mdd_append_pool[0] = '\0';
190 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
192 /* we are using service name but not mdd obd name for compatibility
193 * reasons. It is passed from MDT in lustre_cfg[2] buffer
195 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
197 obd_disconnect(mdd->mdd_child_exp);
202 static struct lu_device *mdd_device_fini(const struct lu_env *env,
205 struct mdd_device *mdd = lu2mdd_dev(d);
208 lu_dev_del_linkage(d->ld_site, d);
210 mdd_procfs_fini(mdd);
214 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
215 struct llog_rec_hdr *hdr, void *data)
217 struct mdd_device *mdd = (struct mdd_device *)data;
218 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
220 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
221 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
224 "seeing record at index %d/%d/%llu t=%x %.*s in log"
225 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
226 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
227 changelog_rec_name(&rec->cr), PLOGID(&llh->lgh_id));
228 spin_lock(&mdd->mdd_cl.mc_lock);
229 mdd->mdd_cl.mc_index = rec->cr.cr_index;
230 spin_unlock(&mdd->mdd_cl.mc_lock);
231 return LLOG_PROC_BREAK;
234 char *mdd_chlg_username(struct llog_changelog_user_rec2 *rec, char *buf,
237 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
239 snprintf(buf, len, "%s%u-%s", CHANGELOG_USER_PREFIX,
240 rec->cur_id, rec->cur_name);
242 snprintf(buf, len, "%s%u", CHANGELOG_USER_PREFIX, rec->cur_id);
246 __u32 mdd_chlg_usermask(struct llog_changelog_user_rec2 *rec)
248 return rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 ?
252 static int changelog_user_init_cb(const struct lu_env *env,
253 struct llog_handle *llh,
254 struct llog_rec_hdr *hdr, void *data)
256 struct mdd_device *mdd = data;
257 struct llog_changelog_user_rec2 *rec;
258 char user_name[CHANGELOG_USER_NAMELEN_FULL];
260 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
262 rec = container_of(hdr, typeof(*rec), cur_hdr);
263 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
264 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
265 CWARN("%s: unknown user type %x at index %u in log "DFID"\n",
266 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
267 rec->cur_hdr.lrh_type, PLOGID(&llh->lgh_id));
272 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
273 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
275 hdr->lrh_index, rec->cur_hdr.lrh_index, rec->cur_endrec,
276 PLOGID(&llh->lgh_id));
278 spin_lock(&mdd->mdd_cl.mc_user_lock);
279 mdd->mdd_cl.mc_lastuser = rec->cur_id;
280 mdd->mdd_cl.mc_users++;
281 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
282 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
283 else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
284 mdd->mdd_cl.mc_current_mask |= CHANGELOG_DEFMASK;
285 mdd->mdd_cl.mc_mintime = min(mdd->mdd_cl.mc_mintime, rec->cur_time);
286 mdd->mdd_cl.mc_minrec = min(mdd->mdd_cl.mc_minrec, rec->cur_endrec);
287 spin_unlock(&mdd->mdd_cl.mc_user_lock);
288 spin_lock(&mdd->mdd_cl.mc_lock);
289 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
290 mdd->mdd_cl.mc_index = rec->cur_endrec;
291 spin_unlock(&mdd->mdd_cl.mc_lock);
293 return LLOG_PROC_BREAK;
296 struct changelog_orphan_data {
298 struct mdd_device *clod_mdd;
301 /* find oldest changelog record index */
302 static int changelog_detect_orphan_cb(const struct lu_env *env,
303 struct llog_handle *llh,
304 struct llog_rec_hdr *hdr, void *data)
306 struct changelog_orphan_data *clod = data;
307 struct mdd_device *mdd = clod->clod_mdd;
308 struct llog_changelog_rec *rec;
310 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
312 rec = container_of(hdr, typeof(*rec), cr_hdr);
313 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
314 CWARN("%s: invalid record at index %d in log "DFID"\n",
315 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
316 PLOGID(&llh->lgh_id));
317 /* try to find some next valid record and thus allow to recover
318 * from a corrupted LLOG, instead to assert and force a crash
324 "%s: record at index %d/%d/%llu t=%x %.*s in log "DFID"\n",
325 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
326 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
327 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
328 PLOGID(&llh->lgh_id));
330 clod->clod_index = rec->cr.cr_index;
332 return LLOG_PROC_BREAK;
335 /* find oldest changelog user index */
336 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
337 struct llog_handle *llh,
338 struct llog_rec_hdr *hdr, void *data)
340 struct changelog_orphan_data *clod = data;
341 struct mdd_device *mdd = clod->clod_mdd;
342 struct llog_changelog_user_rec2 *rec;
343 char user_name[CHANGELOG_USER_NAMELEN_FULL];
345 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
347 rec = container_of(hdr, typeof(*rec), cur_hdr);
348 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC &&
349 rec->cur_hdr.lrh_type != CHANGELOG_USER_REC2) {
350 CWARN("%s: unknown user type %u at index %u in log "DFID"\n",
351 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
352 rec->cur_hdr.lrh_type, PLOGID(&llh->lgh_id));
353 /* try to find some next valid record and thus allow to recover
354 * from a corrupted LLOG, instead to assert and force a crash
359 CDEBUG(D_INFO, "%s: user %s at index %u/%u endrec=%llu in log "DFID"\n",
360 mdd2obd_dev(mdd)->obd_name, mdd_chlg_username(rec, user_name,
362 hdr->lrh_index, rec->cur_hdr.lrh_index,
363 rec->cur_endrec, PLOGID(&llh->lgh_id));
365 clod->clod_index = min_t(__u64, clod->clod_index, rec->cur_endrec);
370 struct changelog_cancel_cookie {
372 struct mdd_device *mdd;
375 static int llog_changelog_cancel_cb(const struct lu_env *env,
376 struct llog_handle *llh,
377 struct llog_rec_hdr *hdr, void *data)
379 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
380 struct changelog_cancel_cookie *cl_cookie =
381 (struct changelog_cancel_cookie *)data;
385 /* This is always a (sub)log, not the catalog */
386 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
388 /* if current context is GC-thread allow it to stop upon umount
389 * remaining records cleanup will occur upon next mount
391 * also during testing, wait for GC-thread to be released
393 * XXX this requires the GC-thread to not fork a sub-thread via
394 * llog[_cat]_process_or_fork() and we may think to also implement
395 * this shutdown mechanism for manually started user unregister which
396 * can also take a long time if huge backlog of records
398 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
399 /* wait to be released */
400 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
403 if (kthread_should_stop())
404 RETURN(LLOG_PROC_BREAK);
407 if (rec->cr.cr_index > cl_cookie->endrec)
408 /* records are in order, so we're done */
409 RETURN(LLOG_PROC_BREAK);
411 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
412 if (cfs_fail_val == 0)
413 cfs_fail_val = hdr->lrh_index;
414 if (cfs_fail_val == hdr->lrh_index)
415 CFS_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
418 /* Records folow one by one, cr_index++. We could calculate the
419 * last cr_index at this plain llog. And if it less then cookie endrec
420 * cancel the whole file.
422 if (llog_is_plain_skipable(llh->lgh_hdr, hdr, rec->cr.cr_index,
423 cl_cookie->endrec)) {
426 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
427 if (cfs_fail_val == 0) {
428 cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
429 CFS_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
432 rc = llog_destroy(env, llh);
434 CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
435 PLOGID(&llh->lgh_id));
436 RETURN(LLOG_DEL_PLAIN);
440 /* cancel them one at a time. I suppose we could store up the cookies
441 * and cancel them all at once; probably more efficient, but this is
442 * done as a user call, so who cares...
445 RETURN(LLOG_DEL_RECORD);
448 static int llog_changelog_cancel(const struct lu_env *env,
449 struct llog_ctxt *ctxt,
450 struct changelog_cancel_cookie *cookie)
452 struct llog_handle *cathandle = ctxt->loc_handle;
457 /* This should only be called with the catalog handle */
458 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
460 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
463 rc = 0; /* 0 or 1 means we're done */
465 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
466 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
467 PLOGID(&cathandle->lgh_id), rc);
472 static struct llog_operations changelog_orig_logops;
475 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
479 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
483 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
486 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
487 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
488 CERROR("Changelogs cannot be enabled due to error condition (see %s log).\n",
489 mdd2obd_dev(mdd)->obd_name);
492 spin_lock(&mdd->mdd_cl.mc_lock);
493 mdd->mdd_cl.mc_flags |= CLM_ON;
494 spin_unlock(&mdd->mdd_cl.mc_lock);
495 rc = mdd_changelog_write_header(env, mdd, CLM_START);
501 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
505 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
508 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
509 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
510 spin_lock(&mdd->mdd_cl.mc_lock);
511 mdd->mdd_cl.mc_flags &= ~CLM_ON;
512 spin_unlock(&mdd->mdd_cl.mc_lock);
517 static int mdd_changelog_llog_init(const struct lu_env *env,
518 struct mdd_device *mdd)
520 struct obd_device *obd = mdd2obd_dev(mdd);
521 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
522 struct changelog_orphan_data clod = {
533 /* LU-2844 mdd setup failure should not cause umount oops */
534 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
537 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
538 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
539 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
540 obd, &changelog_orig_logops);
542 CERROR("%s: changelog llog setup failed: rc = %d\n",
547 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
550 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
553 GOTO(out_cleanup, rc);
555 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
559 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
560 changelog_init_cb, mdd);
563 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
568 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
569 mdd->mdd_cl.mc_index);
571 /* setup user changelog */
572 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
573 obd, &changelog_orig_logops);
575 CERROR("%s: changelog users llog setup failed: rc = %d\n",
580 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
583 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
586 GOTO(out_ucleanup, rc);
588 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
590 GOTO(out_uclose, rc);
592 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
593 changelog_user_init_cb, mdd);
595 CERROR("%s: changelog user init failed: rc = %d\n",
597 GOTO(out_uclose, rc);
600 /* Finally apply per-server mask */
601 mdd->mdd_cl.mc_current_mask |= mdd->mdd_cl.mc_proc_mask;
603 /* If we have registered users, assume we want changelogs on */
604 if (mdd->mdd_cl.mc_lastuser > 0) {
605 rc = mdd_changelog_on(env, mdd);
607 GOTO(out_uclose, rc);
610 /* find and clear any orphan changelog records (1st record index <
611 * smallest of all users current index), likely to come from an
612 * interrupted manual or GC-thread purge, as its user record had
614 * XXX we may wait for a still registered user clear operation to
615 * do the job, but it may then take a long time to reach the user's
616 * real targetted records if a huge purge backlog is still to be
617 * processed as a long time idle user record could have been deleted
618 * XXX we may need to run end of purge as a separate thread
620 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
623 CERROR("%s: changelog detect orphan failed: rc = %d\n",
625 GOTO(out_uclose, rc);
627 rc = llog_cat_process(env, uctxt->loc_handle,
628 changelog_user_detect_orphan_cb,
631 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
633 GOTO(out_uclose, rc);
635 if (unlikely(clod.clod_index < user_orphan.clod_index)) {
636 struct changelog_cancel_cookie cl_cookie = {
637 .endrec = user_orphan.clod_index,
641 CWARN("%s : orphan changelog records found, starting from index %llu to index %llu, being cleared now\n",
642 obd->obd_name, clod.clod_index, user_orphan.clod_index);
644 /* XXX we may need to run end of purge as a separate thread */
645 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
647 CERROR("%s: purge of changelog orphan records failed: rc = %d\n",
649 GOTO(out_uclose, rc);
654 llog_ctxt_put(uctxt);
657 llog_cat_close(env, uctxt->loc_handle);
659 llog_cleanup(env, uctxt);
661 llog_cat_close(env, ctxt->loc_handle);
663 llog_cleanup(env, ctxt);
667 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
669 struct obd_device *obd = mdd2obd_dev(mdd);
672 mdd->mdd_cl.mc_index = 0;
673 spin_lock_init(&mdd->mdd_cl.mc_lock);
674 mdd->mdd_cl.mc_starttime = ktime_get();
675 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
676 mdd->mdd_cl.mc_lastuser = 0;
678 /* ensure a GC check will, and a thread run may, occur upon start */
679 mdd->mdd_cl.mc_gc_time = 0;
680 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
681 mdd->mdd_cl.mc_mintime = (__u32)ktime_get_real_seconds();
682 mdd->mdd_cl.mc_minrec = ULLONG_MAX;
684 rc = mdd_changelog_llog_init(env, mdd);
686 CERROR("%s: changelog setup during init failed: rc = %d\n",
688 mdd->mdd_cl.mc_flags |= CLM_ERR;
694 static void mdd_changelog_fini(const struct lu_env *env,
695 struct mdd_device *mdd)
697 struct obd_device *obd = mdd2obd_dev(mdd);
698 struct llog_ctxt *ctxt;
700 if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
702 mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
705 /* stop GC-thread if running */
706 spin_lock(&mdd->mdd_cl.mc_lock);
707 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
708 /* avoid any attempt to run a GC-thread */
709 mdd->mdd_cl.mc_gc_task = current;
710 spin_unlock(&mdd->mdd_cl.mc_lock);
712 struct task_struct *gc_task;
714 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
715 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
716 /* need to wait for birthing GC-thread to be started
717 * and to have set mc_gc_task to itself
719 spin_unlock(&mdd->mdd_cl.mc_lock);
720 /* Add a tiny sleep */
721 schedule_timeout_uninterruptible(1);
722 /* go back to fully check if GC-thread has started or
723 * even already exited or if a new one is starting...
727 /* take a reference on task_struct to avoid it to be freed
730 gc_task = mdd->mdd_cl.mc_gc_task;
731 get_task_struct(gc_task);
732 spin_unlock(&mdd->mdd_cl.mc_lock);
733 kthread_stop(gc_task);
734 put_task_struct(gc_task);
737 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
739 llog_cat_close(env, ctxt->loc_handle);
740 llog_cleanup(env, ctxt);
742 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
744 llog_cat_close(env, ctxt->loc_handle);
745 llog_cleanup(env, ctxt);
749 /* Remove entries with indicies up to and including @endrec from changelog
750 * Returns: 0 on success
753 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
754 unsigned long long endrec)
756 struct obd_device *obd = mdd2obd_dev(mdd);
757 struct llog_ctxt *ctxt;
758 unsigned long long cur;
759 struct changelog_cancel_cookie cookie;
762 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
766 spin_lock(&mdd->mdd_cl.mc_lock);
767 cur = (long long)mdd->mdd_cl.mc_index;
768 spin_unlock(&mdd->mdd_cl.mc_lock);
771 * If purging all records, write a header entry so we don't have an
772 * empty catalog and we're sure to have a valid starting index next
773 * time. In a case of crash, we just restart with old log so we're
777 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
784 * Some records were purged, so reset repeat-access time (so we
785 * record new mtime update records, so users can see a file has been
786 * changed since the last purge)
788 mdd->mdd_cl.mc_starttime = ktime_get();
790 cookie.endrec = endrec;
792 rc = llog_changelog_cancel(env, ctxt, &cookie);
798 /** Add a CL_MARK record to the changelog
800 * \param markerflags - CLM_*
803 int mdd_changelog_write_header(const struct lu_env *env,
804 struct mdd_device *mdd, int markerflags)
806 struct obd_device *obd = mdd2obd_dev(mdd);
807 struct llog_changelog_rec *rec;
809 struct llog_ctxt *ctxt;
811 int len = strlen(obd->obd_name);
816 if (mdd->mdd_cl.mc_current_mask & BIT(CL_MARK)) {
817 mdd->mdd_cl.mc_starttime = ktime_get();
821 reclen = llog_data_len(sizeof(*rec) + len);
822 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
823 if (buf->lb_buf == NULL)
827 rec->cr.cr_flags = CLF_VERSION;
828 rec->cr.cr_type = CL_MARK;
829 rec->cr.cr_namelen = len;
830 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
831 /* Status and action flags */
832 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
833 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
835 rec->cr_hdr.lrh_type = CHANGELOG_REC;
836 rec->cr.cr_time = cl_time();
838 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
841 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
846 /* assume on or off event; reset repeat-access time */
847 mdd->mdd_cl.mc_starttime = ktime_get();
852 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
853 * are valid. We also check if object with passed fid exists or not.
855 static int obf_lookup(const struct lu_env *env, struct md_object *p,
856 const struct lu_name *lname, struct lu_fid *f,
857 struct md_op_spec *spec)
859 char *name = (char *)lname->ln_name;
860 struct mdd_device *mdd = mdo2mdd(p);
861 struct mdd_object *child;
867 sscanf(name, SFID, RFID(f));
869 GOTO(out, rc = -ENOENT);
871 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
872 !fid_seq_is_dot(f->f_seq))
873 GOTO(out, rc = -ENOENT);
875 /* Check if object with this fid exists */
876 child = mdd_object_find(env, mdd, f);
878 GOTO(out, rc = PTR_ERR(child));
880 if (mdd_object_exists(child) == 0)
883 mdd_object_put(env, child);
889 static int mdd_dummy_create(const struct lu_env *env,
890 struct md_object *pobj,
891 const struct lu_name *lname,
892 struct md_object *child,
893 struct md_op_spec *spec,
899 static int mdd_dummy_rename(const struct lu_env *env,
900 struct md_object *src_pobj,
901 struct md_object *tgt_pobj,
902 const struct lu_fid *lf,
903 const struct lu_name *lsname,
904 struct md_object *tobj,
905 const struct lu_name *ltname,
911 static int mdd_dummy_link(const struct lu_env *env,
912 struct md_object *tgt_obj,
913 struct md_object *src_obj,
914 const struct lu_name *lname,
920 static int mdd_dummy_unlink(const struct lu_env *env,
921 struct md_object *pobj,
922 struct md_object *cobj,
923 const struct lu_name *lname,
930 static int mdd_obf_create(const struct lu_env *env, struct md_object *pobj,
931 const struct lu_name *lname, struct md_object *child,
932 struct md_op_spec *spec, struct md_attr *ma)
934 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
935 return mdd_create(env, pobj, lname, child, spec, ma);
939 static const struct md_dir_operations mdd_obf_dir_ops = {
940 .mdo_lookup = obf_lookup,
941 .mdo_create = mdd_obf_create,
942 .mdo_rename = mdd_dummy_rename,
943 .mdo_link = mdd_dummy_link,
944 .mdo_unlink = mdd_dummy_unlink
947 static const struct md_dir_operations mdd_lpf_dir_ops = {
948 .mdo_lookup = mdd_lookup,
949 .mdo_create = mdd_dummy_create,
950 .mdo_rename = mdd_dummy_rename,
951 .mdo_link = mdd_dummy_link,
952 .mdo_unlink = mdd_dummy_unlink
955 static struct md_object *mdo_locate(const struct lu_env *env,
956 struct md_device *md,
957 const struct lu_fid *fid)
959 struct lu_object *obj;
960 struct md_object *mdo;
962 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
964 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
965 LASSERT(obj != NULL);
973 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
975 struct md_object *mdo;
976 struct mdd_object *mdd_lpf;
977 struct lu_fid fid = LU_LPF_FID;
982 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
983 mdd_lpf_dir_name, S_IFDIR | 0500,
988 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
990 RETURN(PTR_ERR(mdo));
992 LASSERT(lu_object_exists(&mdo->mo_lu));
994 mdd_lpf = md2mdd_obj(mdo);
995 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
996 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
1001 /* Create special in-memory "fid" object for open-by-fid. */
1002 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
1004 struct md_object *mdo;
1005 struct mdd_object *mdd_obf;
1006 struct lu_fid fid = LU_OBF_FID;
1009 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
1010 mdd_obf_dir_name, S_IFDIR | 0100, &fid);
1014 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1016 RETURN(PTR_ERR(mdo));
1018 LASSERT(lu_object_exists(&mdo->mo_lu));
1020 mdd_obf = md2mdd_obj(mdo);
1021 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
1022 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
1027 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
1028 struct mdd_device *m)
1030 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
1031 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
1032 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
1034 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
1035 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
1036 m->mdd_dot_lustre_objs.mdd_obf = NULL;
1038 if (m->mdd_dot_lustre != NULL) {
1039 mdd_object_put(env, m->mdd_dot_lustre);
1040 m->mdd_dot_lustre = NULL;
1044 /** Setup ".lustre" directory object */
1045 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
1047 struct md_object *mdo;
1052 /* Create ".lustre" directory in ROOT. */
1053 fid = LU_DOT_LUSTRE_FID;
1054 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
1060 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
1062 RETURN(PTR_ERR(mdo));
1063 LASSERT(lu_object_exists(&mdo->mo_lu));
1065 m->mdd_dot_lustre = md2mdd_obj(mdo);
1067 rc = mdd_obf_setup(env, m);
1069 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
1070 mdd2obd_dev(m)->obd_name, rc);
1074 rc = mdd_lpf_setup(env, m);
1076 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
1077 mdd2obd_dev(m)->obd_name, rc);
1084 mdd_dot_lustre_cleanup(env, m);
1089 /* set llog methods and create LLOG_AGENT_ORIG_CTXT llog object in obd_device */
1090 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
1091 struct mdd_device *m)
1093 struct obd_device *obd = mdd2obd_dev(m);
1094 struct llog_ctxt *ctxt = NULL;
1099 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1100 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1102 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1103 obd, &llog_common_cat_ops);
1105 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1110 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1113 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1116 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1118 GOTO(out_cleanup, rc);
1121 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1123 GOTO(out_close, rc);
1125 llog_ctxt_put(ctxt);
1129 llog_cat_close(env, ctxt->loc_handle);
1130 ctxt->loc_handle = NULL;
1132 llog_cleanup(env, ctxt);
1137 /* cleanup the context created by llog_setup_named() */
1138 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1139 struct mdd_device *m)
1141 struct obd_device *obd = mdd2obd_dev(m);
1142 struct llog_ctxt *lctxt;
1146 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1148 llog_cat_close(env, lctxt->loc_handle);
1149 lctxt->loc_handle = NULL;
1150 llog_cleanup(env, lctxt);
1156 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1157 struct lustre_cfg *cfg)
1159 barrier_deregister(m->mdd_bottom);
1160 lfsck_degister(env, m->mdd_bottom);
1161 mdd_hsm_actions_llog_fini(env, m);
1162 mdd_changelog_fini(env, m);
1163 mdd_orphan_index_fini(env, m);
1164 mdd_dot_lustre_cleanup(env, m);
1165 if (obd2obt(mdd2obd_dev(m))->obt_nodemap_config_file) {
1166 struct obd_device_target *obt = obd2obt(mdd2obd_dev(m));
1168 nm_config_file_deregister_tgt(env,
1169 obt->obt_nodemap_config_file);
1170 obt->obt_nodemap_config_file = NULL;
1172 if (m->mdd_los != NULL) {
1173 local_oid_storage_fini(env, m->mdd_los);
1176 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1178 if (m->mdd_child_exp)
1179 obd_disconnect(m->mdd_child_exp);
1182 static int mdd_process_config(const struct lu_env *env,
1183 struct lu_device *d, struct lustre_cfg *cfg)
1185 struct mdd_device *m = lu2mdd_dev(d);
1186 struct dt_device *dt = m->mdd_child;
1187 struct lu_device *next = &dt->dd_lu_dev;
1192 switch (cfg->lcfg_command) {
1196 count = class_modify_config(cfg, PARAM_MDD, &m->mdd_kobj);
1197 rc = count > 0 ? 0 : count;
1199 /* we don't understand; pass it on */
1200 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1204 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1207 dt_conf_get(env, dt, &m->mdd_dt_conf);
1209 case LCFG_PRE_CLEANUP:
1210 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1211 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1214 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1215 lu_dev_del_linkage(d->ld_site, d);
1216 mdd_device_shutdown(env, m, cfg);
1219 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1226 static int mdd_recovery_complete(const struct lu_env *env,
1227 struct lu_device *d)
1229 struct mdd_device *mdd = lu2mdd_dev(d);
1230 struct lu_device *next;
1235 LASSERT(mdd != NULL);
1236 next = &mdd->mdd_child->dd_lu_dev;
1238 if (!mdd->mdd_bottom->dd_rdonly)
1239 mdd_orphan_cleanup(env, mdd);
1240 rc = next->ld_ops->ldo_recovery_complete(env, next);
1245 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1246 const struct lu_fid *pfid, const char *name,
1247 __u32 mode, struct lu_fid *fid)
1249 struct dt_object *parent, *dto;
1254 LASSERT(!fid_is_zero(pfid));
1255 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1256 if (unlikely(IS_ERR(parent)))
1257 RETURN(PTR_ERR(parent));
1259 /* create local file/dir, if @fid is passed then try to use it */
1260 if (fid_is_zero(fid))
1261 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1264 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1268 GOTO(out_put, rc = PTR_ERR(dto));
1269 *fid = *lu_object_fid(&dto->do_lu);
1270 /* since stack is not fully set up the local_storage uses own stack
1271 * and we should drop its object from cache
1273 dt_object_put_nocache(env, dto);
1276 dt_object_put(env, parent);
1280 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1281 enum lfsck_events event)
1286 static int mdd_prepare(const struct lu_env *env,
1287 struct lu_device *pdev,
1288 struct lu_device *cdev)
1290 struct mdd_device *mdd = lu2mdd_dev(cdev);
1291 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1292 struct nm_config_file *nodemap_config;
1293 struct obd_device_target *obt = obd2obt(mdd2obd_dev(mdd));
1294 struct dt_object *root = NULL;
1295 struct thandle *th = NULL;
1301 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1305 /* Setup local dirs */
1306 fid.f_seq = FID_SEQ_LOCAL_NAME;
1309 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1314 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1319 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1320 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1321 mdd_root_dir_name, S_IFDIR |
1324 CERROR("%s: create root fid failed: rc = %d\n",
1325 mdd2obd_dev(mdd)->obd_name, rc);
1329 /* store a default directory layout on the root directory if
1330 * it doesn't already exist to improve MDT space balance.
1332 root = dt_locate(env, mdd->mdd_bottom, &fid);
1333 if (unlikely(IS_ERR(root)))
1334 GOTO(out_los, rc = PTR_ERR(root));
1336 rc = dt_xattr_get(env, root, &LU_BUF_NULL,
1337 XATTR_NAME_DEFAULT_LMV);
1338 if (rc == -ENODATA) {
1340 struct lmv_user_md lmv_default = {
1341 .lum_magic = LMV_USER_MAGIC,
1342 .lum_stripe_count = 1,
1343 .lum_stripe_offset = root_stripe_offset,
1344 .lum_max_inherit = root_max_inherit,
1345 .lum_max_inherit_rr = root_max_inherit_rr,
1348 th = dt_trans_create(env, mdd->mdd_bottom);
1350 GOTO(out_root_put, rc = PTR_ERR(th));
1352 buf.lb_buf = &lmv_default;
1353 buf.lb_len = sizeof(lmv_default);
1354 rc = dt_declare_xattr_set(env, root, &buf,
1355 XATTR_NAME_DEFAULT_LMV, 0,
1358 GOTO(out_trans_stop, rc);
1360 rc = dt_trans_start_local(env, mdd->mdd_bottom, th);
1362 GOTO(out_trans_stop, rc);
1364 rc = dt_xattr_set(env, root, &buf,
1365 XATTR_NAME_DEFAULT_LMV, 0, th);
1367 GOTO(out_trans_stop, rc);
1369 dt_trans_stop(env, mdd->mdd_bottom, th);
1371 } else if (rc < 0 && rc != -ERANGE) {
1372 CERROR("%s: get default LMV of root failed: rc = %d\n",
1373 mdd2obd_dev(mdd)->obd_name, rc);
1375 GOTO(out_root_put, rc);
1378 dt_object_put(env, root);
1381 mdd->mdd_root_fid = fid;
1382 rc = mdd_dot_lustre_setup(env, mdd);
1384 CERROR("%s: initializing .lustre failed: rc = %d\n",
1385 mdd2obd_dev(mdd)->obd_name, rc);
1389 /* Normal client usually send root access to MDT0 directly,
1390 * the root FID on non-MDT0 will only be used by echo client.
1392 mdd->mdd_root_fid = fid;
1395 rc = mdd_orphan_index_init(env, mdd);
1399 rc = mdd_changelog_init(env, mdd);
1401 CERROR("%s: failed to initialize changelog: rc = %d\n",
1402 mdd2obd_dev(mdd)->obd_name, rc);
1406 rc = mdd_hsm_actions_llog_init(env, mdd);
1408 GOTO(out_changelog, rc);
1410 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1412 if (IS_ERR(nodemap_config)) {
1413 rc = PTR_ERR(nodemap_config);
1417 obt->obt_nodemap_config_file = nodemap_config;
1420 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1421 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1424 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1425 mdd2obd_dev(mdd)->obd_name, rc);
1426 GOTO(out_nodemap, rc);
1429 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1431 CERROR("%s: failed to register to barrier: rc = %d\n",
1432 mdd2obd_dev(mdd)->obd_name, rc);
1433 GOTO(out_lfsck, rc);
1439 lfsck_degister(env, mdd->mdd_bottom);
1441 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1442 obt->obt_nodemap_config_file = NULL;
1444 mdd_hsm_actions_llog_fini(env, mdd);
1446 mdd_changelog_fini(env, mdd);
1448 mdd_orphan_index_fini(env, mdd);
1450 if (mdd_seq_site(mdd)->ss_node_id == 0)
1451 mdd_dot_lustre_cleanup(env, mdd);
1454 dt_trans_stop(env, mdd->mdd_bottom, th);
1457 dt_object_put(env, root);
1459 local_oid_storage_fini(env, mdd->mdd_los);
1460 mdd->mdd_los = NULL;
1466 * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
1468 * Find corresponding device by passed parent and name, and allocate FID from
1471 * see include/lu_object.h for the details.
1473 static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1474 struct lu_fid *fid, struct lu_object *parent,
1475 const struct lu_name *name)
1477 struct mdd_device *mdd = lu2mdd_dev(d);
1478 struct lu_object *o = lu_object_next(parent);
1480 return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
1483 const struct lu_device_operations mdd_lu_ops = {
1484 .ldo_object_alloc = mdd_object_alloc,
1485 .ldo_process_config = mdd_process_config,
1486 .ldo_recovery_complete = mdd_recovery_complete,
1487 .ldo_prepare = mdd_prepare,
1488 .ldo_fid_alloc = mdd_fid_alloc,
1491 static int mdd_root_get(const struct lu_env *env,
1492 struct md_device *m, struct lu_fid *f)
1494 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1497 *f = mdd->mdd_root_fid;
1501 /* No permission check is needed. */
1502 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1503 struct obd_statfs *sfs)
1505 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1510 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs, NULL);
1512 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1517 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1518 struct md_device *m)
1520 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1522 return &mdd->mdd_dt_conf;
1525 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1528 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1530 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1531 return (*h == NULL ? -ENOENT : 0);
1534 static struct lu_device *mdd_device_free(const struct lu_env *env,
1535 struct lu_device *lu)
1537 struct mdd_device *m = lu2mdd_dev(lu);
1541 LASSERT(atomic_read(&lu->ld_ref) == 0);
1542 md_device_fini(&m->mdd_md_dev);
1547 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1548 struct lu_device_type *t,
1549 struct lustre_cfg *lcfg)
1551 struct lu_device *l;
1552 struct mdd_device *m;
1556 l = ERR_PTR(-ENOMEM);
1561 md_device_init(&m->mdd_md_dev, t);
1562 rc = mdd_init0(env, m, t, lcfg);
1564 mdd_device_free(env, l);
1572 /* we use exports to track all mdd users */
1573 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1574 struct obd_device *obd, struct obd_uuid *cluuid,
1575 struct obd_connect_data *data, void *localdata)
1577 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1578 struct lustre_handle conn;
1583 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1585 rc = class_connect(&conn, obd, cluuid);
1589 *exp = class_conn2export(&conn);
1591 /* Why should there ever be more than 1 connect? */
1592 LASSERT(mdd->mdd_connects == 0);
1593 mdd->mdd_connects++;
1598 /* once last export (don't count self-export) disappeared mdd can be released */
1599 static int mdd_obd_disconnect(struct obd_export *exp)
1601 struct obd_device *obd = exp->exp_obd;
1602 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1603 int rc, release = 0;
1607 mdd->mdd_connects--;
1608 if (mdd->mdd_connects == 0)
1611 rc = class_disconnect(exp);
1613 if (rc == 0 && release)
1614 class_manual_cleanup(obd);
1618 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1619 __u32 keylen, void *key, __u32 *vallen, void *val)
1623 if (KEY_IS(KEY_OSP_CONNECTED)) {
1624 struct obd_device *obd = exp->exp_obd;
1625 struct mdd_device *mdd;
1627 if (!obd->obd_set_up || obd->obd_stopping)
1630 mdd = lu2mdd_dev(obd->obd_lu_dev);
1632 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1640 static int mdd_obd_set_info_async(const struct lu_env *env,
1641 struct obd_export *exp,
1642 __u32 keylen, void *key,
1643 __u32 vallen, void *val,
1644 struct ptlrpc_request_set *set)
1646 struct obd_device *obd = exp->exp_obd;
1647 struct mdd_device *mdd;
1650 if (!obd->obd_set_up || obd->obd_stopping)
1653 mdd = lu2mdd_dev(obd->obd_lu_dev);
1655 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1660 static const struct obd_ops mdd_obd_device_ops = {
1661 .o_owner = THIS_MODULE,
1662 .o_connect = mdd_obd_connect,
1663 .o_disconnect = mdd_obd_disconnect,
1664 .o_get_info = mdd_obd_get_info,
1665 .o_set_info_async = mdd_obd_set_info_async,
1668 struct mdd_changelog_name_check_data {
1669 const char *mcnc_name;
1674 * changelog_recalc_mask callback
1676 * Is is called per each registered user and calculates combined mask of
1677 * all registered users.
1679 static int mdd_changelog_name_check_cb(const struct lu_env *env,
1680 struct llog_handle *llh,
1681 struct llog_rec_hdr *hdr, void *data)
1683 struct llog_changelog_user_rec2 *rec;
1684 struct mdd_changelog_name_check_data *mcnc = data;
1686 rec = container_of(hdr, typeof(*rec), cur_hdr);
1687 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 &&
1688 !strncmp(rec->cur_name, mcnc->mcnc_name, sizeof(rec->cur_name))) {
1689 mcnc->mcnc_id = rec->cur_id;
1695 static int mdd_changelog_name_check(const struct lu_env *env,
1696 struct llog_ctxt *ctxt,
1697 struct mdd_device *mdd, const char *name)
1699 struct mdd_changelog_name_check_data mcnc = { .mcnc_name = name, };
1705 /* first symbol is a letter */
1706 if (!isalpha(name[0])) {
1708 CERROR("%s: first char '%c' in '%s' is not letter: rc = %d\n",
1709 mdd2obd_dev(mdd)->obd_name, name[0], name, rc);
1713 /* name is valid: contains letters, numbers and '-', '_' only */
1714 while (name[++chr]) {
1715 if (!(isalnum(name[chr]) || name[chr] == '_' ||
1716 name[chr] == '-')) {
1718 CERROR("%s: wrong char '%c' in name '%s': rc = %d\n",
1719 mdd2obd_dev(mdd)->obd_name, name[chr], name, rc);
1724 if (chr > CHANGELOG_USER_NAMELEN) {
1726 CERROR("%s: name '%s' is over %d symbols limit: rc = %d\n",
1727 mdd2obd_dev(mdd)->obd_name, name,
1728 CHANGELOG_USER_NAMELEN, rc);
1732 rc = llog_cat_process(env, ctxt->loc_handle,
1733 mdd_changelog_name_check_cb, &mcnc, 0, 0);
1735 CWARN("%s: changelog name %s exists already: rc = %d\n",
1736 mdd2obd_dev(mdd)->obd_name, name, rc);
1738 CWARN("%s: failed user changelog processing: rc = %d\n",
1739 mdd2obd_dev(mdd)->obd_name, rc);
1743 static int mdd_changelog_user_register(const struct lu_env *env,
1744 struct mdd_device *mdd, int *id,
1745 const char *name, const char *mask)
1747 struct llog_ctxt *ctxt;
1748 struct llog_changelog_user_rec2 *rec;
1749 char user_name[CHANGELOG_USER_NAMELEN_FULL];
1754 ctxt = llog_get_context(mdd2obd_dev(mdd),
1755 LLOG_CHANGELOG_USER_ORIG_CTXT);
1761 llog_ctxt_put(ctxt);
1765 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1767 rec->cur_hdr.lrh_len = sizeof(*rec);
1768 /* keep old record type for users without mask/name for
1769 * compatibility needs
1771 if (mask || (name && name[0]))
1772 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC2;
1774 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1775 spin_lock(&mdd->mdd_cl.mc_user_lock);
1776 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1777 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1779 CERROR("%s: registering %s user: max ID is exceeded: rc = %d\n",
1780 mdd2obd_dev(mdd)->obd_name,
1781 (name && name[0]) ? name : "new", rc);
1784 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1785 mdd->mdd_cl.mc_users++;
1786 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1788 rec->cur_time = (__u32)ktime_get_real_seconds();
1789 if (CFS_FAIL_PRECHECK(OBD_FAIL_TIME_IN_CHLOG_USER)) {
1790 rec->cur_time -= min(cfs_fail_val, rec->cur_time);
1791 spin_lock(&mdd->mdd_cl.mc_user_lock);
1792 mdd->mdd_cl.mc_mintime = rec->cur_time;
1793 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1796 spin_lock(&mdd->mdd_cl.mc_lock);
1797 rec->cur_endrec = mdd->mdd_cl.mc_index;
1798 spin_unlock(&mdd->mdd_cl.mc_lock);
1801 u64 newmask = CHANGELOG_DEFMASK;
1803 /* if user will use relative mask apply it on default one */
1804 rc = cfs_str2mask(mask, changelog_type2str, &newmask,
1805 CHANGELOG_MINMASK, CHANGELOG_ALLMASK,
1808 GOTO(out_users, rc);
1809 rec->cur_mask = newmask;
1810 } else if (mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK) {
1811 /* a maskless users means default mask but only if server has
1812 * no specific mask set
1814 rec->cur_mask = CHANGELOG_DEFMASK;
1817 if (name && name[0]) {
1818 rc = mdd_changelog_name_check(env, ctxt, mdd, name);
1820 GOTO(out_users, rc);
1821 strscpy(rec->cur_name, name, sizeof(rec->cur_name));
1823 mdd_chlg_username(rec, user_name, sizeof(user_name));
1825 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1827 CWARN("%s: failed to register changelog user %s: rc = %d\n",
1828 mdd2obd_dev(mdd)->obd_name, user_name, rc);
1829 GOTO(out_users, rc);
1832 /* apply user mask finally */
1833 spin_lock(&mdd->mdd_cl.mc_user_lock);
1834 mdd->mdd_cl.mc_current_mask |= rec->cur_mask;
1835 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1837 CDEBUG(D_IOCTL, "%s: registered changelog user '%s', mask %#x\n",
1838 mdd2obd_dev(mdd)->obd_name, user_name, rec->cur_mask);
1840 /* Assume we want it on since somebody registered */
1841 rc = mdd_changelog_on(env, mdd);
1843 /* record is added, so don't decrement users on error */
1847 spin_lock(&mdd->mdd_cl.mc_user_lock);
1848 mdd->mdd_cl.mc_users--;
1849 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1853 llog_ctxt_put(ctxt);
1857 struct mdd_changelog_recalc_mask_data {
1858 struct mdd_device *mcrm_mdd;
1863 * changelog_recalc_mask callback
1865 * Is is called per each registered user and calculates combined mask of
1866 * all registered users.
1868 static int mdd_changelog_recalc_mask_cb(const struct lu_env *env,
1869 struct llog_handle *llh,
1870 struct llog_rec_hdr *hdr, void *data)
1872 struct llog_changelog_user_rec2 *rec;
1873 struct mdd_changelog_recalc_mask_data *mcrm = data;
1875 rec = container_of(hdr, typeof(*rec), cur_hdr);
1876 if (rec->cur_hdr.lrh_type == CHANGELOG_USER_REC2 && rec->cur_mask)
1877 mcrm->mcrm_mask |= rec->cur_mask;
1878 else if (mcrm->mcrm_mdd->mdd_cl.mc_proc_mask == CHANGELOG_MINMASK)
1879 mcrm->mcrm_mask |= CHANGELOG_DEFMASK;
1884 int mdd_changelog_recalc_mask(const struct lu_env *env, struct mdd_device *mdd)
1886 struct llog_ctxt *ctxt;
1887 struct mdd_changelog_recalc_mask_data mcrm = {
1889 .mcrm_mask = mdd->mdd_cl.mc_proc_mask,
1895 ctxt = llog_get_context(mdd2obd_dev(mdd),
1896 LLOG_CHANGELOG_USER_ORIG_CTXT);
1900 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
1901 GOTO(out, rc = -ENXIO);
1903 rc = llog_cat_process(env, ctxt->loc_handle,
1904 mdd_changelog_recalc_mask_cb, &mcrm, 0, 0);
1906 CWARN("%s: failed user changelog processing: rc = %d\n",
1907 mdd2obd_dev(mdd)->obd_name, rc);
1909 spin_lock(&mdd->mdd_cl.mc_user_lock);
1910 CDEBUG(D_INFO, "%s: recalc changelog mask: %#x -> %#x\n",
1911 mdd2obd_dev(mdd)->obd_name, mdd->mdd_cl.mc_current_mask,
1913 mdd->mdd_cl.mc_current_mask = mcrm.mcrm_mask;
1914 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1918 llog_ctxt_put(ctxt);
1923 struct mdd_changelog_user_purge {
1924 struct mdd_device *mcup_mdd;
1926 __u32 mcup_usercount;
1929 char mcup_name[CHANGELOG_USER_NAMELEN_FULL];
1933 * changelog_user_purge callback
1935 * Is called once per user.
1937 * Check to see the user requested is available from rec.
1938 * Truncate the changelog.
1939 * Keep track of the total number of users (calls).
1941 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1942 struct llog_handle *llh,
1943 struct llog_rec_hdr *hdr, void *data)
1945 struct llog_changelog_user_rec2 *rec;
1946 struct mdd_changelog_user_purge *mcup = data;
1947 struct llog_cookie cookie;
1952 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1955 rec = container_of(hdr, typeof(*rec), cur_hdr);
1957 mcup->mcup_usercount++;
1959 if (rec->cur_id != mcup->mcup_id) {
1960 /* truncate to the lowest endrec that is not this user */
1961 mcup->mcup_minrec = min(mcup->mcup_minrec, rec->cur_endrec);
1965 mdd_chlg_username(rec, mcup->mcup_name, sizeof(mcup->mcup_name));
1967 /* Unregister this user */
1968 cookie.lgc_lgl = llh->lgh_id;
1969 cookie.lgc_index = hdr->lrh_index;
1970 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1973 mcup->mcup_found = true;
1974 mcup->mcup_usercount--;
1975 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1976 mcup->mcup_mdd->mdd_cl.mc_users--;
1977 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1983 int mdd_changelog_user_purge(const struct lu_env *env,
1984 struct mdd_device *mdd, __u32 id)
1986 struct mdd_changelog_user_purge mcup = {
1989 .mcup_found = false,
1990 .mcup_usercount = 0,
1991 .mcup_minrec = ULLONG_MAX,
1994 struct llog_ctxt *ctxt;
1999 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
2000 mdd2obd_dev(mdd)->obd_name, id);
2002 ctxt = llog_get_context(mdd2obd_dev(mdd),
2003 LLOG_CHANGELOG_USER_ORIG_CTXT);
2006 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
2007 GOTO(out, rc = -ENXIO);
2009 rc = llog_cat_process(env, ctxt->loc_handle,
2010 mdd_changelog_user_purge_cb, &mcup, 0, 0);
2012 CWARN("%s: failed to purge changelog for user %s: rc = %d\n",
2013 mdd2obd_dev(mdd)->obd_name, mcup.mcup_name, rc);
2017 CFS_FAIL_TIMEOUT(OBD_FAIL_LLOG_PURGE_DELAY, cfs_fail_val);
2018 if (mcup.mcup_usercount == 0) {
2019 spin_lock(&mdd->mdd_cl.mc_user_lock);
2020 if (mdd->mdd_cl.mc_users == 0) {
2021 /* No more users; turn changelogs off */
2022 CDEBUG(D_IOCTL, "turning off changelogs\n");
2023 rc = mdd_changelog_off(env, mdd);
2025 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2028 if (mcup.mcup_found) {
2030 "%s: Purge changelog entries for user %s record=%llu\n",
2031 mdd2obd_dev(mdd)->obd_name,
2032 mcup.mcup_name, mcup.mcup_minrec);
2033 rc = mdd_changelog_llog_cancel(env, mdd, mcup.mcup_minrec);
2035 CWARN("%s: No changelog for user id %u: rc = %d\n",
2036 mdd2obd_dev(mdd)->obd_name, id, rc);
2037 GOTO(out, rc = -ENOENT);
2040 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
2044 llog_ctxt_put(ctxt);
2049 struct mdd_changelog_user_clear {
2055 struct mdd_device *mcuc_mdd;
2056 char mcuc_name[CHANGELOG_USER_NAMELEN_FULL];
2060 * changelog_clear callback
2062 * Is called once per user.
2064 * Check to see the user requested is available from rec.
2065 * Check the oldest (smallest) record for boundary conditions.
2066 * Truncate the changelog.
2068 static int mdd_changelog_clear_cb(const struct lu_env *env,
2069 struct llog_handle *llh,
2070 struct llog_rec_hdr *hdr,
2073 struct llog_changelog_user_rec2 *rec;
2074 struct mdd_changelog_user_clear *mcuc = data;
2075 struct mdd_device *mdd = mcuc->mcuc_mdd;
2080 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
2083 rec = container_of(hdr, typeof(*rec), cur_hdr);
2084 /* Does the changelog id match the requested id? */
2085 if (rec->cur_id != mcuc->mcuc_id) {
2086 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec, rec->cur_endrec);
2087 mcuc->mcuc_mintime = min(mcuc->mcuc_mintime, rec->cur_time);
2091 mdd_chlg_username(rec, mcuc->mcuc_name, sizeof(mcuc->mcuc_name));
2092 /* cur_endrec is the oldest purgeable record, make sure we're newer */
2093 if (rec->cur_endrec > mcuc->mcuc_endrec) {
2096 "%s: request %llu < endrec %llu for user %s: rc = %d\n",
2097 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_endrec,
2098 rec->cur_endrec, mcuc->mcuc_name, rc);
2103 * Flag that we've met all the range and user checks.
2104 * We now know the record to flush.
2106 mcuc->mcuc_flush = true;
2108 rec->cur_endrec = mcuc->mcuc_endrec;
2109 rec->cur_time = (__u32)ktime_get_real_seconds();
2111 CDEBUG(D_IOCTL, "%s: update changelog user %s endrec = %llu\n",
2112 mdd2obd_dev(mdd)->obd_name, mcuc->mcuc_name, rec->cur_endrec);
2114 /* Update the endrec */
2115 rc = llog_write(env, llh, hdr, hdr->lrh_index);
2120 /* Clear a changelog up to entry specified by endrec for user id. */
2121 static int mdd_changelog_clear(const struct lu_env *env,
2122 struct mdd_device *mdd, __u32 id,
2125 struct mdd_changelog_user_clear mcuc = {
2127 .mcuc_minrec = endrec,
2128 .mcuc_flush = false,
2130 .mcuc_mintime = ktime_get_real_seconds(),
2133 struct llog_ctxt *ctxt;
2139 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
2140 mdd2obd_dev(mdd)->obd_name, id, endrec);
2141 /* start_rec is the newest (largest value) entry in the changelogs*/
2142 spin_lock(&mdd->mdd_cl.mc_lock);
2143 start_rec = mdd->mdd_cl.mc_index;
2144 spin_unlock(&mdd->mdd_cl.mc_lock);
2146 if (start_rec < endrec) {
2148 "%s: Could not clear changelog, requested address out of range\n",
2149 mdd2obd_dev(mdd)->obd_name);
2154 mcuc.mcuc_endrec = start_rec;
2155 mcuc.mcuc_minrec = ULLONG_MAX;
2157 mcuc.mcuc_endrec = endrec;
2160 ctxt = llog_get_context(mdd2obd_dev(mdd),
2161 LLOG_CHANGELOG_USER_ORIG_CTXT);
2165 if (!(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT))
2166 GOTO(out, rc = -ENXIO);
2168 rc = llog_cat_process(env, ctxt->loc_handle, mdd_changelog_clear_cb,
2170 if (rc == -EINVAL) {
2171 CDEBUG(D_IOCTL, "%s: no changelog recnum <= %llu to clear\n",
2172 mdd2obd_dev(mdd)->obd_name, (unsigned long long)endrec);
2177 CWARN("%s: can't clear the changelog for user %s: rc = %d\n",
2178 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, rc);
2182 if (!mcuc.mcuc_flush) {
2183 CDEBUG(D_IOCTL, "%s: no entry for user %d\n",
2184 mdd2obd_dev(mdd)->obd_name, id);
2185 GOTO(out, rc = -ENOENT);
2188 CDEBUG(D_IOCTL, "%s: purge changelog user %s entries up to %llu\n",
2189 mdd2obd_dev(mdd)->obd_name, mcuc.mcuc_name, mcuc.mcuc_minrec);
2191 rc = mdd_changelog_llog_cancel(env, mdd, mcuc.mcuc_minrec);
2195 spin_lock(&mdd->mdd_cl.mc_user_lock);
2196 mdd->mdd_cl.mc_minrec = mcuc.mcuc_minrec;
2197 mdd->mdd_cl.mc_mintime = mcuc.mcuc_mintime;
2198 spin_unlock(&mdd->mdd_cl.mc_user_lock);
2202 llog_ctxt_put(ctxt);
2207 static int mdd_changelog_user_deregister(const struct lu_env *env,
2208 struct mdd_device *mdd, int *id,
2211 struct llog_ctxt *ctxt;
2212 struct mdd_changelog_name_check_data mcnc = {
2221 ctxt = llog_get_context(mdd2obd_dev(mdd),
2222 LLOG_CHANGELOG_USER_ORIG_CTXT);
2226 rc = llog_cat_process(env, ctxt->loc_handle,
2227 mdd_changelog_name_check_cb, &mcnc, 0, 0);
2228 llog_ctxt_put(ctxt);
2230 if (rc != -EEXIST) {
2231 CDEBUG(D_IOCTL, "%s: no entry for username %s\n",
2232 mdd2obd_dev(mdd)->obd_name, name);
2238 /* explicitly clear changelog first, to protect from crash in
2239 * the middle of purge that would lead to unregistered consumer
2240 * but pending changelog entries
2242 rc = mdd_changelog_clear(env, mdd, *id, 0);
2244 rc = mdd_changelog_user_purge(env, mdd, *id);
2246 /* recalc changelog current mask */
2247 mdd_changelog_recalc_mask(env, mdd);
2253 * May be called remotely from mdt_iocontrol_handle or locally from
2254 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
2255 * an obd_ioctl_data format (but local ioctl handler does).
2257 * \param len - data len
2258 * \param karg - ioctl data, in kernel space
2260 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
2261 unsigned int cmd, int len, void *karg)
2263 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
2264 struct obd_device *obd = mdd2obd_dev(mdd);
2265 struct obd_ioctl_data *data;
2269 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK\n",
2270 obd->obd_name, cmd, len, karg);
2271 if (unlikely(karg == NULL))
2272 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc));
2275 /* Doesn't use obd_ioctl_data */
2277 case OBD_IOC_CHANGELOG_CLEAR: {
2278 struct changelog_setinfo *cs = karg;
2280 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2281 RETURN(-EINPROGRESS);
2283 rc = mdd_changelog_clear(env, mdd, cs->cs_id, cs->cs_recno);
2284 barrier_exit(mdd->mdd_bottom);
2287 case OBD_IOC_START_LFSCK: {
2288 rc = lfsck_start(env, mdd->mdd_bottom, karg);
2291 case OBD_IOC_STOP_LFSCK: {
2292 rc = lfsck_stop(env, mdd->mdd_bottom, karg);
2295 case OBD_IOC_QUERY_LFSCK: {
2296 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL, karg);
2299 case OBD_IOC_LLOG_PRINT:
2300 case OBD_IOC_LLOG_CANCEL:
2301 rc = obd_iocontrol(cmd, mdd->mdd_child_exp, len, karg, NULL);
2305 /* Below ioctls use obd_ioctl_data */
2306 if (data->ioc_version != OBD_IOCTL_VERSION) {
2307 CERROR("%s: iocontrol from '%s' bad magic %x != %x: rc = %d\n",
2308 obd->obd_name, current->comm,
2309 data->ioc_version, OBD_IOCTL_VERSION, rc);
2314 case OBD_IOC_CHANGELOG_REG:
2315 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2316 RETURN(-EINPROGRESS);
2318 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1,
2321 barrier_exit(mdd->mdd_bottom);
2323 case OBD_IOC_CHANGELOG_DEREG:
2324 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
2325 RETURN(-EINPROGRESS);
2327 rc = mdd_changelog_user_deregister(env, mdd, &data->ioc_u32_1,
2330 barrier_exit(mdd->mdd_bottom);
2333 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
2340 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
2341 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
2343 static const struct md_device_operations mdd_ops = {
2344 .mdo_statfs = mdd_statfs,
2345 .mdo_root_get = mdd_root_get,
2346 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
2347 .mdo_iocontrol = mdd_iocontrol,
2348 .mdo_dtconf_get = mdd_dtconf_get,
2351 static const struct lu_device_type_operations mdd_device_type_ops = {
2352 .ldto_init = mdd_type_init,
2353 .ldto_fini = mdd_type_fini,
2355 .ldto_start = mdd_type_start,
2356 .ldto_stop = mdd_type_stop,
2358 .ldto_device_alloc = mdd_device_alloc,
2359 .ldto_device_free = mdd_device_free,
2361 .ldto_device_fini = mdd_device_fini
2364 static struct lu_device_type mdd_device_type = {
2365 .ldt_tags = LU_DEVICE_MD,
2366 .ldt_name = LUSTRE_MDD_NAME,
2367 .ldt_ops = &mdd_device_type_ops,
2368 .ldt_ctx_tags = LCT_MD_THREAD
2371 /* context key constructor: mdd_key_init */
2372 LU_KEY_INIT(mdd, struct mdd_thread_info);
2374 static void mdd_key_fini(const struct lu_context *ctx,
2375 struct lu_context_key *key, void *data)
2377 struct mdd_thread_info *info = data;
2379 lu_buf_free(&info->mdi_big_buf);
2380 lu_buf_free(&info->mdi_link_buf);
2381 lu_buf_free(&info->mdi_xattr_buf);
2382 lu_buf_free(&info->mdi_chlg_buf);
2387 /* context key: mdd_thread_key */
2388 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
2390 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
2391 int (*func)(void *), void *data, char *name)
2393 struct task_struct *task;
2395 LASSERT(thread->mgt_init == false);
2396 init_completion(&thread->mgt_started);
2397 init_completion(&thread->mgt_finished);
2398 thread->mgt_data = data;
2399 thread->mgt_abort = false;
2400 thread->mgt_init = true;
2402 task = kthread_run(func, thread, "%s", name);
2404 complete(&thread->mgt_finished);
2405 return PTR_ERR(task);
2407 wait_for_completion(&thread->mgt_started);
2411 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
2413 if (thread->mgt_init == true) {
2414 thread->mgt_abort = true;
2415 wait_for_completion(&thread->mgt_finished);
2419 static int __init mdd_init(void)
2423 rc = libcfs_setup();
2427 rc = lu_kmem_init(mdd_caches);
2431 changelog_orig_logops = llog_common_cat_ops;
2432 changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
2434 rc = class_register_type(&mdd_obd_device_ops, NULL, false,
2435 LUSTRE_MDD_NAME, &mdd_device_type);
2437 lu_kmem_fini(mdd_caches);
2441 static void __exit mdd_exit(void)
2443 class_unregister_type(LUSTRE_MDD_NAME);
2444 lu_kmem_fini(mdd_caches);
2447 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2448 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2449 MODULE_VERSION(LUSTRE_VERSION_STRING);
2450 MODULE_LICENSE("GPL");
2452 module_init(mdd_init);
2453 module_exit(mdd_exit);