4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/mdd/mdd_device.c
33 * Lustre Metadata Server (mdd) routines
35 * Author: Wang Di <wangdi@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <linux/module.h>
41 #include <linux/kthread.h>
42 #include <obd_class.h>
43 #include <uapi/linux/lustre/lustre_ioctl.h>
44 #include <lustre_mds.h>
45 #include <obd_support.h>
46 #include <lu_object.h>
47 #include <uapi/linux/lustre/lustre_param.h>
48 #include <lustre_fid.h>
49 #include <lustre_nodemap.h>
50 #include <lustre_barrier.h>
52 #include "mdd_internal.h"
54 static const struct md_device_operations mdd_ops;
55 static struct lu_device_type mdd_device_type;
57 static const char mdd_root_dir_name[] = "ROOT";
58 static const char mdd_obf_dir_name[] = "fid";
59 static const char mdd_lpf_dir_name[] = "lost+found";
61 /* Slab for MDD object allocation */
62 struct kmem_cache *mdd_object_kmem;
64 static struct lu_kmem_descr mdd_caches[] = {
66 .ckd_cache = &mdd_object_kmem,
67 .ckd_name = "mdd_obj",
68 .ckd_size = sizeof(struct mdd_object)
75 static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
78 struct obd_connect_data *data = NULL;
79 struct lu_device *lud = mdd2lu_dev(m);
80 struct obd_device *obd;
84 LASSERT(m->mdd_child_exp == NULL);
86 OBD_ALLOC(data, sizeof(*data));
88 GOTO(out, rc = -ENOMEM);
90 obd = class_name2obd(nextdev);
92 CERROR("can't locate next device: %s\n", nextdev);
93 GOTO(out, rc = -ENOTCONN);
96 data->ocd_connect_flags = OBD_CONNECT_VERSION;
97 data->ocd_version = LUSTRE_VERSION_CODE;
99 rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
101 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
105 lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
106 LASSERT(lud->ld_site);
107 m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
108 m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
109 lu_dev_add_linkage(lud->ld_site, lud);
113 OBD_FREE(data, sizeof(*data));
117 static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
118 struct lu_device_type *t, struct lustre_cfg *lcfg)
124 /* LU-8040 Set defaults here, before values configs */
125 mdd->mdd_cl.mc_flags = 0; /* off by default */
126 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
127 mdd->mdd_cl.mc_deniednext = 60; /* 60 secs by default */
129 dev = lustre_cfg_string(lcfg, 0);
133 mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev);
134 if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL)
136 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
137 mdd->mdd_md_dev.md_ops = &mdd_ops;
139 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
143 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
144 /* sync permission changes */
145 mdd->mdd_sync_permission = 1;
146 /* enable changelog garbage collection */
147 mdd->mdd_changelog_gc = 1;
148 /* with a significant amount of idle time */
149 mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME;
150 /* or a significant amount of late indexes */
151 mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES;
152 /* with a reasonable interval between each check */
153 mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL;
154 /* with a very few number of free catalog entries */
155 mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES;
156 /* special default striping for files created with O_APPEND */
157 mdd->mdd_append_stripe_count = 1;
158 mdd->mdd_append_pool[0] = '\0';
160 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
162 /* we are using service name but not mdd obd name
163 * for compatibility reasons.
164 * It is passed from MDT in lustre_cfg[2] buffer */
165 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
167 obd_disconnect(mdd->mdd_child_exp);
172 static struct lu_device *mdd_device_fini(const struct lu_env *env,
175 struct mdd_device *mdd = lu2mdd_dev(d);
178 lu_dev_del_linkage(d->ld_site, d);
180 mdd_procfs_fini(mdd);
184 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
185 struct llog_rec_hdr *hdr, void *data)
187 struct mdd_device *mdd = (struct mdd_device *)data;
188 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
190 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
191 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
194 "seeing record at index %d/%d/%llu t=%x %.*s in log"
195 DFID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
196 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
197 changelog_rec_name(&rec->cr), PFID(&llh->lgh_id.lgl_oi.oi_fid));
199 mdd->mdd_cl.mc_index = rec->cr.cr_index;
200 return LLOG_PROC_BREAK;
203 static int changelog_user_init_cb(const struct lu_env *env,
204 struct llog_handle *llh,
205 struct llog_rec_hdr *hdr, void *data)
207 struct mdd_device *mdd = (struct mdd_device *)data;
208 struct llog_changelog_user_rec *rec =
209 (struct llog_changelog_user_rec *)hdr;
211 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
212 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
214 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec=%llu"
215 " in log "DFID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
216 rec->cur_id, rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid));
218 spin_lock(&mdd->mdd_cl.mc_user_lock);
219 mdd->mdd_cl.mc_lastuser = rec->cur_id;
220 mdd->mdd_cl.mc_users++;
221 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
222 mdd->mdd_cl.mc_index = rec->cur_endrec;
223 spin_unlock(&mdd->mdd_cl.mc_user_lock);
225 return LLOG_PROC_BREAK;
228 struct changelog_orphan_data {
230 struct mdd_device *mdd;
233 /* find oldest changelog record index */
234 static int changelog_detect_orphan_cb(const struct lu_env *env,
235 struct llog_handle *llh,
236 struct llog_rec_hdr *hdr, void *data)
238 struct mdd_device *mdd = ((struct changelog_orphan_data *)data)->mdd;
239 struct llog_changelog_rec *rec = container_of(hdr,
240 struct llog_changelog_rec,
243 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
245 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
246 CWARN("%s: invalid record at index %d in log "DFID"\n",
247 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
248 PFID(&llh->lgh_id.lgl_oi.oi_fid));
249 /* try to find some next valid record and thus allow to recover
250 * from a corrupted LLOG, instead to assert and force a crash
255 CDEBUG(D_INFO, "%s: seeing record at index %d/%d/%llu t=%x %.*s in log "
256 DFID"\n", mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
257 rec->cr_hdr.lrh_index, rec->cr.cr_index, rec->cr.cr_type,
258 rec->cr.cr_namelen, changelog_rec_name(&rec->cr),
259 PFID(&llh->lgh_id.lgl_oi.oi_fid));
261 ((struct changelog_orphan_data *)data)->index = rec->cr.cr_index;
262 return LLOG_PROC_BREAK;
265 /* find oldest changelog user index */
266 static int changelog_user_detect_orphan_cb(const struct lu_env *env,
267 struct llog_handle *llh,
268 struct llog_rec_hdr *hdr, void *data)
270 struct mdd_device *mdd = ((struct changelog_orphan_data *)data)->mdd;
271 struct llog_changelog_user_rec *rec = container_of(hdr,
272 struct llog_changelog_user_rec,
275 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
277 if (rec->cur_hdr.lrh_type != CHANGELOG_USER_REC) {
278 CWARN("%s: invalid user at index %d in log "DFID"\n",
279 mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
280 PFID(&llh->lgh_id.lgl_oi.oi_fid));
281 /* try to find some next valid record and thus allow to recover
282 * from a corrupted LLOG, instead to assert and force a crash
287 CDEBUG(D_INFO, "%s: seeing user at index %d/%d id=%d endrec=%llu in "
288 "log "DFID"\n", mdd2obd_dev(mdd)->obd_name, hdr->lrh_index,
289 rec->cur_hdr.lrh_index, rec->cur_id, rec->cur_endrec,
290 PFID(&llh->lgh_id.lgl_oi.oi_fid));
292 if (((struct changelog_orphan_data *)data)->index == 0 ||
293 rec->cur_endrec < ((struct changelog_orphan_data *)data)->index)
294 ((struct changelog_orphan_data *)data)->index = rec->cur_endrec;
299 struct changelog_cancel_cookie {
301 struct mdd_device *mdd;
304 static int llog_changelog_cancel_cb(const struct lu_env *env,
305 struct llog_handle *llh,
306 struct llog_rec_hdr *hdr, void *data)
308 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
309 struct changelog_cancel_cookie *cl_cookie =
310 (struct changelog_cancel_cookie *)data;
314 /* This is always a (sub)log, not the catalog */
315 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
317 /* if current context is GC-thread allow it to stop upon umount
318 * remaining records cleanup will occur upon next mount
320 * also during testing, wait for GC-thread to be released
322 * XXX this requires the GC-thread to not fork a sub-thread via
323 * llog[_cat]_process_or_fork() and we may think to also implement
324 * this shutdown mechanism for manually started user unregister which
325 * can also take a long time if huge backlog of records
327 if (unlikely(cl_cookie->mdd->mdd_cl.mc_gc_task == current)) {
328 /* wait to be released */
329 while (CFS_FAIL_CHECK_QUIET(OBD_FAIL_FORCE_GC_THREAD))
332 if (kthread_should_stop())
333 RETURN(LLOG_PROC_BREAK);
336 if (rec->cr.cr_index > cl_cookie->endrec)
337 /* records are in order, so we're done */
338 RETURN(LLOG_PROC_BREAK);
340 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_RACE))) {
341 if (cfs_fail_val == 0)
342 cfs_fail_val = hdr->lrh_index;
343 if (cfs_fail_val == hdr->lrh_index)
344 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_RACE);
347 /* Records folow one by one, cr_index++. We could calculate the
348 * last cr_index at this plain llog. And if it less then cookie endrec
349 * cancel the whole file.
351 if ((LLOG_HDR_BITMAP_SIZE(llh->lgh_hdr) - hdr->lrh_index +
352 rec->cr.cr_index) < cl_cookie->endrec) {
355 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_DEL))) {
356 if (cfs_fail_val == 0) {
357 cfs_fail_val = (unsigned long)llh & 0xFFFFFFFF;
358 OBD_RACE(OBD_FAIL_MDS_CHANGELOG_DEL);
361 rc = llog_destroy(env, llh);
363 CDEBUG(D_HA, "Changelog destroyed plain "DFID"\n",
364 PFID(&llh->lgh_id.lgl_oi.oi_fid));
365 RETURN(LLOG_DEL_PLAIN);
369 /* cancel them one at a time. I suppose we could store up the cookies
370 * and cancel them all at once; probably more efficient, but this is
371 * done as a user call, so who cares... */
373 RETURN(LLOG_DEL_RECORD);
376 static int llog_changelog_cancel(const struct lu_env *env,
377 struct llog_ctxt *ctxt,
378 struct changelog_cancel_cookie *cookie)
380 struct llog_handle *cathandle = ctxt->loc_handle;
385 /* This should only be called with the catalog handle */
386 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
388 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
391 /* 0 or 1 means we're done */
394 CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n",
395 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
396 PFID(&cathandle->lgh_id.lgl_oi.oi_fid), rc);
401 static struct llog_operations changelog_orig_logops;
404 mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd,
408 mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd)
412 if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0)
415 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
416 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
417 CERROR("Changelogs cannot be enabled due to error "
418 "condition (see %s log).\n",
419 mdd2obd_dev(mdd)->obd_name);
422 spin_lock(&mdd->mdd_cl.mc_lock);
423 mdd->mdd_cl.mc_flags |= CLM_ON;
424 spin_unlock(&mdd->mdd_cl.mc_lock);
425 rc = mdd_changelog_write_header(env, mdd, CLM_START);
431 mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd)
435 if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON)
438 LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name);
439 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
440 spin_lock(&mdd->mdd_cl.mc_lock);
441 mdd->mdd_cl.mc_flags &= ~CLM_ON;
442 spin_unlock(&mdd->mdd_cl.mc_lock);
447 static int mdd_changelog_llog_init(const struct lu_env *env,
448 struct mdd_device *mdd)
450 struct obd_device *obd = mdd2obd_dev(mdd);
451 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
452 struct changelog_orphan_data changelog_orphan = { .index = 0,
454 user_orphan = { .index = 0,
460 /* LU-2844 mdd setup failure should not cause umount oops */
461 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
464 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
465 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
466 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
467 obd, &changelog_orig_logops);
469 CERROR("%s: changelog llog setup failed: rc = %d\n",
474 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
477 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
480 GOTO(out_cleanup, rc);
482 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
486 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
487 changelog_init_cb, mdd);
490 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
495 CDEBUG(D_IOCTL, "changelog starting index=%llu\n",
496 mdd->mdd_cl.mc_index);
498 /* setup user changelog */
499 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
500 obd, &changelog_orig_logops);
502 CERROR("%s: changelog users llog setup failed: rc = %d\n",
507 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
510 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
513 GOTO(out_ucleanup, rc);
515 rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL);
517 GOTO(out_uclose, rc);
519 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
520 changelog_user_init_cb, mdd);
522 CERROR("%s: changelog user init failed: rc = %d\n",
524 GOTO(out_uclose, rc);
527 /* If we have registered users, assume we want changelogs on */
528 if (mdd->mdd_cl.mc_lastuser > 0) {
529 rc = mdd_changelog_on(env, mdd);
531 GOTO(out_uclose, rc);
534 /* find and clear any orphan changelog records (1st record index <
535 * smallest of all users current index), likely to come from an
536 * interrupted manual or GC-thread purge, as its user record had
538 * XXX we may wait for a still registered user clear operation to
539 * do the job, but it may then take a long time to reach the user's
540 * real targetted records if a huge purge backlog is still to be
541 * processed as a long time idle user record could have been deleted
542 * XXX we may need to run end of purge as a separate thread
544 rc = llog_cat_process(env, ctxt->loc_handle, changelog_detect_orphan_cb,
545 &changelog_orphan, 0, 0);
547 CERROR("%s: changelog detect orphan failed: rc = %d\n",
549 GOTO(out_uclose, rc);
551 rc = llog_cat_process(env, uctxt->loc_handle,
552 changelog_user_detect_orphan_cb,
555 CERROR("%s: changelog user detect orphan failed: rc = %d\n",
557 GOTO(out_uclose, rc);
559 if (unlikely(changelog_orphan.index < user_orphan.index)) {
560 struct changelog_cancel_cookie cl_cookie = {
561 .endrec = user_orphan.index,
565 CWARN("%s : orphan changelog records found, starting from "
566 "index %llu to index %llu, being cleared now\n",
567 obd->obd_name, changelog_orphan.index, user_orphan.index);
569 /* XXX we may need to run end of purge as a separate thread */
570 rc = llog_changelog_cancel(env, ctxt, &cl_cookie);
572 CERROR("%s: purge of changelog orphan records failed: "
573 "rc = %d\n", obd->obd_name, rc);
574 GOTO(out_uclose, rc);
579 llog_ctxt_put(uctxt);
582 llog_cat_close(env, uctxt->loc_handle);
584 llog_cleanup(env, uctxt);
586 llog_cat_close(env, ctxt->loc_handle);
588 llog_cleanup(env, ctxt);
592 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
594 struct obd_device *obd = mdd2obd_dev(mdd);
597 mdd->mdd_cl.mc_index = 0;
598 spin_lock_init(&mdd->mdd_cl.mc_lock);
599 mdd->mdd_cl.mc_starttime = ktime_get();
600 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
601 mdd->mdd_cl.mc_lastuser = 0;
603 /* ensure a GC check will, and a thread run may, occur upon start */
604 mdd->mdd_cl.mc_gc_time = 0;
605 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NONE;
607 rc = mdd_changelog_llog_init(env, mdd);
609 CERROR("%s: changelog setup during init failed: rc = %d\n",
611 mdd->mdd_cl.mc_flags |= CLM_ERR;
617 static void mdd_changelog_fini(const struct lu_env *env,
618 struct mdd_device *mdd)
620 struct obd_device *obd = mdd2obd_dev(mdd);
621 struct llog_ctxt *ctxt;
623 if (mdd->mdd_cl.mc_flags & CLM_CLEANUP_DONE)
625 mdd->mdd_cl.mc_flags = CLM_CLEANUP_DONE;
628 /* stop GC-thread if running */
629 spin_lock(&mdd->mdd_cl.mc_lock);
630 if (likely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE)) {
631 /* avoid any attempt to run a GC-thread */
632 mdd->mdd_cl.mc_gc_task = current;
633 spin_unlock(&mdd->mdd_cl.mc_lock);
635 struct task_struct *gc_task;
637 if (unlikely(mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NEED ||
638 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_START)) {
639 /* need to wait for birthing GC-thread to be started
640 * and to have set mc_gc_task to itself
642 spin_unlock(&mdd->mdd_cl.mc_lock);
643 /* Add a tiny sleep */
644 schedule_timeout_uninterruptible(1);
645 /* go back to fully check if GC-thread has started or
646 * even already exited or if a new one is starting...
650 /* take a reference on task_struct to avoid it to be freed
653 gc_task = mdd->mdd_cl.mc_gc_task;
654 get_task_struct(gc_task);
655 spin_unlock(&mdd->mdd_cl.mc_lock);
656 kthread_stop(gc_task);
657 put_task_struct(gc_task);
660 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
662 llog_cat_close(env, ctxt->loc_handle);
663 llog_cleanup(env, ctxt);
665 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
667 llog_cat_close(env, ctxt->loc_handle);
668 llog_cleanup(env, ctxt);
672 /** Remove entries with indicies up to and including \a endrec from the
679 mdd_changelog_llog_cancel(const struct lu_env *env, struct mdd_device *mdd,
682 struct obd_device *obd = mdd2obd_dev(mdd);
683 struct llog_ctxt *ctxt;
684 long long unsigned cur;
685 struct changelog_cancel_cookie cookie;
688 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
692 spin_lock(&mdd->mdd_cl.mc_lock);
693 cur = (long long)mdd->mdd_cl.mc_index;
694 spin_unlock(&mdd->mdd_cl.mc_lock);
698 /* purge to "0" is shorthand for everything */
702 /* If purging all records, write a header entry so we don't have an
703 empty catalog and we're sure to have a valid starting index next
704 time. In case of crash, we just restart with old log so we're
707 /* XXX: transaction is started by llog itself */
708 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
713 /* Some records were purged, so reset repeat-access time (so we
714 record new mtime update records, so users can see a file has been
715 changed since the last purge) */
716 mdd->mdd_cl.mc_starttime = ktime_get();
718 cookie.endrec = endrec;
720 rc = llog_changelog_cancel(env, ctxt, &cookie);
726 /** Add a CL_MARK record to the changelog
728 * \param markerflags - CLM_*
731 int mdd_changelog_write_header(const struct lu_env *env,
732 struct mdd_device *mdd, int markerflags)
734 struct obd_device *obd = mdd2obd_dev(mdd);
735 struct llog_changelog_rec *rec;
737 struct llog_ctxt *ctxt;
739 int len = strlen(obd->obd_name);
744 if (mdd->mdd_cl.mc_mask & BIT(CL_MARK)) {
745 mdd->mdd_cl.mc_starttime = ktime_get();
749 reclen = llog_data_len(sizeof(*rec) + len);
750 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_chlg_buf, reclen);
751 if (buf->lb_buf == NULL)
755 rec->cr.cr_flags = CLF_VERSION;
756 rec->cr.cr_type = CL_MARK;
757 rec->cr.cr_namelen = len;
758 memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen);
759 /* Status and action flags */
760 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
761 rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) +
763 rec->cr_hdr.lrh_type = CHANGELOG_REC;
764 rec->cr.cr_time = cl_time();
766 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
769 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL);
774 /* assume on or off event; reset repeat-access time */
775 mdd->mdd_cl.mc_starttime = ktime_get();
780 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
781 * are valid. We also check if object with passed fid exists or not.
783 static int obf_lookup(const struct lu_env *env, struct md_object *p,
784 const struct lu_name *lname, struct lu_fid *f,
785 struct md_op_spec *spec)
787 char *name = (char *)lname->ln_name;
788 struct mdd_device *mdd = mdo2mdd(p);
789 struct mdd_object *child;
795 sscanf(name, SFID, RFID(f));
797 GOTO(out, rc = -ENOENT);
799 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
800 !fid_seq_is_dot(f->f_seq))
801 GOTO(out, rc = -ENOENT);
803 /* Check if object with this fid exists */
804 child = mdd_object_find(env, mdd, f);
806 GOTO(out, rc = PTR_ERR(child));
808 if (mdd_object_exists(child) == 0)
811 mdd_object_put(env, child);
817 static int mdd_dummy_create(const struct lu_env *env,
818 struct md_object *pobj,
819 const struct lu_name *lname,
820 struct md_object *child,
821 struct md_op_spec *spec,
827 static int mdd_dummy_rename(const struct lu_env *env,
828 struct md_object *src_pobj,
829 struct md_object *tgt_pobj,
830 const struct lu_fid *lf,
831 const struct lu_name *lsname,
832 struct md_object *tobj,
833 const struct lu_name *ltname,
839 static int mdd_dummy_link(const struct lu_env *env,
840 struct md_object *tgt_obj,
841 struct md_object *src_obj,
842 const struct lu_name *lname,
848 static int mdd_dummy_unlink(const struct lu_env *env,
849 struct md_object *pobj,
850 struct md_object *cobj,
851 const struct lu_name *lname,
858 int mdd_create(const struct lu_env *env, struct md_object *pobj,
859 const struct lu_name *lname, struct md_object *child,
860 struct md_op_spec *spec, struct md_attr *ma);
861 static int mdd_obf_create(const struct lu_env *env, struct md_object *pobj,
862 const struct lu_name *lname, struct md_object *child,
863 struct md_op_spec *spec, struct md_attr *ma)
865 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
866 return mdd_create(env, pobj, lname, child, spec, ma);
870 static const struct md_dir_operations mdd_obf_dir_ops = {
871 .mdo_lookup = obf_lookup,
872 .mdo_create = mdd_obf_create,
873 .mdo_rename = mdd_dummy_rename,
874 .mdo_link = mdd_dummy_link,
875 .mdo_unlink = mdd_dummy_unlink
878 static const struct md_dir_operations mdd_lpf_dir_ops = {
879 .mdo_lookup = mdd_lookup,
880 .mdo_create = mdd_dummy_create,
881 .mdo_rename = mdd_dummy_rename,
882 .mdo_link = mdd_dummy_link,
883 .mdo_unlink = mdd_dummy_unlink
886 static struct md_object *mdo_locate(const struct lu_env *env,
887 struct md_device *md,
888 const struct lu_fid *fid)
890 struct lu_object *obj;
891 struct md_object *mdo;
893 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
895 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
896 LASSERT(obj != NULL);
904 static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m)
906 struct md_object *mdo;
907 struct mdd_object *mdd_lpf;
908 struct lu_fid fid = LU_LPF_FID;
912 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
913 mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR,
918 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
920 RETURN(PTR_ERR(mdo));
922 LASSERT(lu_object_exists(&mdo->mo_lu));
924 mdd_lpf = md2mdd_obj(mdo);
925 mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops;
926 m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf;
932 * Create special in-memory "fid" object for open-by-fid.
934 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
936 struct md_object *mdo;
937 struct mdd_object *mdd_obf;
938 struct lu_fid fid = LU_OBF_FID;
941 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
942 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
946 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
948 RETURN(PTR_ERR(mdo));
950 LASSERT(lu_object_exists(&mdo->mo_lu));
952 mdd_obf = md2mdd_obj(mdo);
953 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
954 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
959 static void mdd_dot_lustre_cleanup(const struct lu_env *env,
960 struct mdd_device *m)
962 if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) {
963 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf);
964 m->mdd_dot_lustre_objs.mdd_lpf = NULL;
966 if (m->mdd_dot_lustre_objs.mdd_obf != NULL) {
967 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
968 m->mdd_dot_lustre_objs.mdd_obf = NULL;
970 if (m->mdd_dot_lustre != NULL) {
971 mdd_object_put(env, m->mdd_dot_lustre);
972 m->mdd_dot_lustre = NULL;
976 /** Setup ".lustre" directory object */
977 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
979 struct md_object *mdo;
984 /* Create ".lustre" directory in ROOT. */
985 fid = LU_DOT_LUSTRE_FID;
986 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
988 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
992 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
994 RETURN(PTR_ERR(mdo));
995 LASSERT(lu_object_exists(&mdo->mo_lu));
997 m->mdd_dot_lustre = md2mdd_obj(mdo);
999 rc = mdd_obf_setup(env, m);
1001 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
1002 mdd2obd_dev(m)->obd_name, rc);
1006 rc = mdd_lpf_setup(env, m);
1008 CERROR("%s: error initializing \"lost+found\": rc = %d.\n",
1009 mdd2obd_dev(m)->obd_name, rc);
1016 mdd_dot_lustre_cleanup(env, m);
1022 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
1023 * object in obd_device
1025 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
1026 struct mdd_device *m)
1028 struct obd_device *obd = mdd2obd_dev(m);
1029 struct llog_ctxt *ctxt = NULL;
1033 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1034 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
1036 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
1037 obd, &llog_common_cat_ops);
1039 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
1044 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1047 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
1050 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
1052 GOTO(out_cleanup, rc);
1055 rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL);
1057 GOTO(out_close, rc);
1059 llog_ctxt_put(ctxt);
1063 llog_cat_close(env, ctxt->loc_handle);
1064 ctxt->loc_handle = NULL;
1066 llog_cleanup(env, ctxt);
1072 * cleanup the context created by llog_setup_named()
1074 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
1075 struct mdd_device *m)
1077 struct obd_device *obd = mdd2obd_dev(m);
1078 struct llog_ctxt *lctxt;
1081 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
1083 llog_cat_close(env, lctxt->loc_handle);
1084 lctxt->loc_handle = NULL;
1085 llog_cleanup(env, lctxt);
1091 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
1092 struct lustre_cfg *cfg)
1094 barrier_deregister(m->mdd_bottom);
1095 lfsck_degister(env, m->mdd_bottom);
1096 mdd_hsm_actions_llog_fini(env, m);
1097 mdd_changelog_fini(env, m);
1098 mdd_orphan_index_fini(env, m);
1099 mdd_dot_lustre_cleanup(env, m);
1100 if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) {
1101 nm_config_file_deregister_tgt(env,
1102 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file);
1103 mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL;
1105 if (m->mdd_los != NULL) {
1106 local_oid_storage_fini(env, m->mdd_los);
1109 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
1111 if (m->mdd_child_exp)
1112 obd_disconnect(m->mdd_child_exp);
1115 static int mdd_process_config(const struct lu_env *env,
1116 struct lu_device *d, struct lustre_cfg *cfg)
1118 struct mdd_device *m = lu2mdd_dev(d);
1119 struct dt_device *dt = m->mdd_child;
1120 struct lu_device *next = &dt->dd_lu_dev;
1124 switch (cfg->lcfg_command) {
1128 count = class_modify_config(cfg, PARAM_MDD, &m->mdd_kobj);
1129 rc = count > 0 ? 0 : count;
1131 /* we don't understand; pass it on */
1132 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1136 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1139 dt_conf_get(env, dt, &m->mdd_dt_conf);
1141 case LCFG_PRE_CLEANUP:
1142 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1143 mdd_generic_thread_stop(&m->mdd_orphan_cleanup_thread);
1146 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1147 lu_dev_del_linkage(d->ld_site, d);
1148 mdd_device_shutdown(env, m, cfg);
1151 rc = next->ld_ops->ldo_process_config(env, next, cfg);
1158 static int mdd_recovery_complete(const struct lu_env *env,
1159 struct lu_device *d)
1161 struct mdd_device *mdd = lu2mdd_dev(d);
1162 struct lu_device *next;
1166 LASSERT(mdd != NULL);
1167 next = &mdd->mdd_child->dd_lu_dev;
1169 if (!mdd->mdd_bottom->dd_rdonly)
1170 mdd_orphan_cleanup(env, mdd);
1171 rc = next->ld_ops->ldo_recovery_complete(env, next);
1176 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
1177 const struct lu_fid *pfid, const char *name,
1178 __u32 mode, struct lu_fid *fid)
1180 struct dt_object *parent, *dto;
1185 LASSERT(!fid_is_zero(pfid));
1186 parent = dt_locate(env, mdd->mdd_bottom, pfid);
1187 if (unlikely(IS_ERR(parent)))
1188 RETURN(PTR_ERR(parent));
1190 /* create local file/dir, if @fid is passed then try to use it */
1191 if (fid_is_zero(fid))
1192 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
1195 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
1199 GOTO(out_put, rc = PTR_ERR(dto));
1200 *fid = *lu_object_fid(&dto->do_lu);
1201 /* since stack is not fully set up the local_storage uses own stack
1202 * and we should drop its object from cache */
1203 dt_object_put_nocache(env, dto);
1206 dt_object_put(env, parent);
1210 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
1211 enum lfsck_events event)
1216 static int mdd_prepare(const struct lu_env *env,
1217 struct lu_device *pdev,
1218 struct lu_device *cdev)
1220 struct mdd_device *mdd = lu2mdd_dev(cdev);
1221 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
1222 struct nm_config_file *nodemap_config;
1223 struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt;
1229 rc = next->ld_ops->ldo_prepare(env, cdev, next);
1233 /* Setup local dirs */
1234 fid.f_seq = FID_SEQ_LOCAL_NAME;
1237 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
1242 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
1247 if (mdd_seq_site(mdd)->ss_node_id == 0) {
1248 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
1249 mdd_root_dir_name, S_IFDIR |
1250 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
1252 CERROR("%s: create root fid failed: rc = %d\n",
1253 mdd2obd_dev(mdd)->obd_name, rc);
1257 mdd->mdd_root_fid = fid;
1258 rc = mdd_dot_lustre_setup(env, mdd);
1260 CERROR("%s: initializing .lustre failed: rc = %d\n",
1261 mdd2obd_dev(mdd)->obd_name, rc);
1265 /* Normal client usually send root access to MDT0 directly,
1266 * the root FID on non-MDT0 will only be used by echo client. */
1267 mdd->mdd_root_fid = fid;
1270 rc = mdd_orphan_index_init(env, mdd);
1274 rc = mdd_changelog_init(env, mdd);
1276 CERROR("%s: failed to initialize changelog: rc = %d\n",
1277 mdd2obd_dev(mdd)->obd_name, rc);
1281 rc = mdd_hsm_actions_llog_init(env, mdd);
1283 GOTO(out_changelog, rc);
1285 nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom,
1287 if (IS_ERR(nodemap_config)) {
1288 rc = PTR_ERR(nodemap_config);
1292 obt->obt_nodemap_config_file = nodemap_config;
1295 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
1296 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
1299 CERROR("%s: failed to initialize lfsck: rc = %d\n",
1300 mdd2obd_dev(mdd)->obd_name, rc);
1301 GOTO(out_nodemap, rc);
1304 rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child);
1306 CERROR("%s: failed to register to barrier: rc = %d\n",
1307 mdd2obd_dev(mdd)->obd_name, rc);
1308 GOTO(out_lfsck, rc);
1314 lfsck_degister(env, mdd->mdd_bottom);
1316 nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file);
1317 obt->obt_nodemap_config_file = NULL;
1319 mdd_hsm_actions_llog_fini(env, mdd);
1321 mdd_changelog_fini(env, mdd);
1323 mdd_orphan_index_fini(env, mdd);
1325 if (mdd_seq_site(mdd)->ss_node_id == 0)
1326 mdd_dot_lustre_cleanup(env, mdd);
1328 local_oid_storage_fini(env, mdd->mdd_los);
1329 mdd->mdd_los = NULL;
1335 * Implementation of lu_device_operations::ldo_fid_alloc() for MDD.
1337 * Find corresponding device by passed parent and name, and allocate FID from
1340 * see include/lu_object.h for the details.
1342 static int mdd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1343 struct lu_fid *fid, struct lu_object *parent,
1344 const struct lu_name *name)
1346 struct mdd_device *mdd = lu2mdd_dev(d);
1347 struct lu_object *o = lu_object_next(parent);
1349 return dt_fid_alloc(env, mdd->mdd_child, fid, o, name);
1352 const struct lu_device_operations mdd_lu_ops = {
1353 .ldo_object_alloc = mdd_object_alloc,
1354 .ldo_process_config = mdd_process_config,
1355 .ldo_recovery_complete = mdd_recovery_complete,
1356 .ldo_prepare = mdd_prepare,
1357 .ldo_fid_alloc = mdd_fid_alloc,
1360 static int mdd_root_get(const struct lu_env *env,
1361 struct md_device *m, struct lu_fid *f)
1363 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1366 *f = mdd->mdd_root_fid;
1371 * No permission check is needed.
1373 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1374 struct obd_statfs *sfs)
1376 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1381 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs, NULL);
1383 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1388 static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env,
1389 struct md_device *m)
1391 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1393 return &mdd->mdd_dt_conf;
1396 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1399 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1401 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1402 return (*h == NULL ? -ENOENT : 0);
1405 static struct lu_device *mdd_device_free(const struct lu_env *env,
1406 struct lu_device *lu)
1408 struct mdd_device *m = lu2mdd_dev(lu);
1411 LASSERT(atomic_read(&lu->ld_ref) == 0);
1412 md_device_fini(&m->mdd_md_dev);
1417 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1418 struct lu_device_type *t,
1419 struct lustre_cfg *lcfg)
1421 struct lu_device *l;
1422 struct mdd_device *m;
1426 l = ERR_PTR(-ENOMEM);
1431 md_device_init(&m->mdd_md_dev, t);
1432 rc = mdd_init0(env, m, t, lcfg);
1434 mdd_device_free(env, l);
1443 * we use exports to track all mdd users
1445 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1446 struct obd_device *obd, struct obd_uuid *cluuid,
1447 struct obd_connect_data *data, void *localdata)
1449 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1450 struct lustre_handle conn;
1454 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1456 rc = class_connect(&conn, obd, cluuid);
1460 *exp = class_conn2export(&conn);
1462 /* Why should there ever be more than 1 connect? */
1463 LASSERT(mdd->mdd_connects == 0);
1464 mdd->mdd_connects++;
1470 * once last export (we don't count self-export) disappeared
1471 * mdd can be released
1473 static int mdd_obd_disconnect(struct obd_export *exp)
1475 struct obd_device *obd = exp->exp_obd;
1476 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1477 int rc, release = 0;
1480 mdd->mdd_connects--;
1481 if (mdd->mdd_connects == 0)
1484 rc = class_disconnect(exp);
1486 if (rc == 0 && release)
1487 class_manual_cleanup(obd);
1491 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1492 __u32 keylen, void *key, __u32 *vallen, void *val)
1496 if (KEY_IS(KEY_OSP_CONNECTED)) {
1497 struct obd_device *obd = exp->exp_obd;
1498 struct mdd_device *mdd;
1500 if (!obd->obd_set_up || obd->obd_stopping)
1503 mdd = lu2mdd_dev(obd->obd_lu_dev);
1505 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1513 static int mdd_obd_set_info_async(const struct lu_env *env,
1514 struct obd_export *exp,
1515 __u32 keylen, void *key,
1516 __u32 vallen, void *val,
1517 struct ptlrpc_request_set *set)
1519 struct obd_device *obd = exp->exp_obd;
1520 struct mdd_device *mdd;
1523 if (!obd->obd_set_up || obd->obd_stopping)
1526 mdd = lu2mdd_dev(obd->obd_lu_dev);
1528 rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key,
1533 static const struct obd_ops mdd_obd_device_ops = {
1534 .o_owner = THIS_MODULE,
1535 .o_connect = mdd_obd_connect,
1536 .o_disconnect = mdd_obd_disconnect,
1537 .o_get_info = mdd_obd_get_info,
1538 .o_set_info_async = mdd_obd_set_info_async,
1541 static int mdd_changelog_user_register(const struct lu_env *env,
1542 struct mdd_device *mdd, int *id)
1544 struct llog_ctxt *ctxt;
1545 struct llog_changelog_user_rec *rec;
1549 ctxt = llog_get_context(mdd2obd_dev(mdd),
1550 LLOG_CHANGELOG_USER_ORIG_CTXT);
1556 llog_ctxt_put(ctxt);
1560 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1562 rec->cur_hdr.lrh_len = sizeof(*rec);
1563 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1564 spin_lock(&mdd->mdd_cl.mc_user_lock);
1565 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1566 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1567 CERROR("Maximum number of changelog users exceeded!\n");
1568 GOTO(out, rc = -EOVERFLOW);
1570 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1571 mdd->mdd_cl.mc_users++;
1572 rec->cur_endrec = mdd->mdd_cl.mc_index;
1574 rec->cur_time = (__u32)ktime_get_real_seconds();
1575 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1578 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1580 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL);
1582 CWARN("%s: Failed to register changelog user %d: rc=%d\n",
1583 mdd2obd_dev(mdd)->obd_name, *id, rc);
1584 spin_lock(&mdd->mdd_cl.mc_user_lock);
1585 mdd->mdd_cl.mc_users--;
1586 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1590 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1592 /* Assume we want it on since somebody registered */
1593 rc = mdd_changelog_on(env, mdd);
1599 llog_ctxt_put(ctxt);
1603 struct mdd_changelog_user_purge {
1604 struct mdd_device *mcup_mdd;
1606 __u32 mcup_usercount;
1612 * changelog_user_purge callback
1614 * Is called once per user.
1616 * Check to see the user requested is available from rec.
1617 * Truncate the changelog.
1618 * Keep track of the total number of users (calls).
1620 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1621 struct llog_handle *llh,
1622 struct llog_rec_hdr *hdr, void *data)
1624 struct llog_changelog_user_rec *rec;
1625 struct mdd_changelog_user_purge *mcup = data;
1626 struct llog_cookie cookie;
1631 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1634 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1636 mcup->mcup_usercount++;
1638 if (rec->cur_id != mcup->mcup_id) {
1639 /* truncate to the lowest endrec that is not this user */
1640 mcup->mcup_minrec = min(mcup->mcup_minrec,
1645 /* Unregister this user */
1646 cookie.lgc_lgl = llh->lgh_id;
1647 cookie.lgc_index = hdr->lrh_index;
1649 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1652 mcup->mcup_found = true;
1653 mcup->mcup_usercount--;
1654 spin_lock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1655 mcup->mcup_mdd->mdd_cl.mc_users--;
1656 spin_unlock(&mcup->mcup_mdd->mdd_cl.mc_user_lock);
1662 int mdd_changelog_user_purge(const struct lu_env *env,
1663 struct mdd_device *mdd, __u32 id)
1665 struct mdd_changelog_user_purge mcup = {
1668 .mcup_found = false,
1669 .mcup_usercount = 0,
1670 .mcup_minrec = ULLONG_MAX,
1672 struct llog_ctxt *ctxt;
1677 CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n",
1678 mdd2obd_dev(mdd)->obd_name, id);
1680 ctxt = llog_get_context(mdd2obd_dev(mdd),
1681 LLOG_CHANGELOG_USER_ORIG_CTXT);
1683 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1684 GOTO(out, rc = -ENXIO);
1686 rc = llog_cat_process(env, ctxt->loc_handle,
1687 mdd_changelog_user_purge_cb, &mcup,
1690 OBD_FAIL_TIMEOUT(OBD_FAIL_LLOG_PURGE_DELAY, cfs_fail_val);
1692 if ((rc == 0) && (mcup.mcup_usercount == 0)) {
1693 spin_lock(&mdd->mdd_cl.mc_user_lock);
1694 if (mdd->mdd_cl.mc_users == 0) {
1695 /* No more users; turn changelogs off */
1696 CDEBUG(D_IOCTL, "turning off changelogs\n");
1697 rc = mdd_changelog_off(env, mdd);
1699 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1702 if ((rc == 0) && mcup.mcup_found) {
1703 CDEBUG(D_IOCTL, "%s: Purging changelog entries for user %d "
1705 mdd2obd_dev(mdd)->obd_name, id, mcup.mcup_minrec);
1706 /* Cancelling record 0 destroys the entire changelog, make sure
1707 we don't do that unless we mean it. */
1708 if (mcup.mcup_minrec != 0 || mcup.mcup_usercount == 0) {
1709 rc = mdd_changelog_llog_cancel(env, mdd,
1713 CWARN("%s: No changelog for user %u; rc=%d\n",
1714 mdd2obd_dev(mdd)->obd_name, id, rc);
1715 GOTO(out, rc = -ENOENT);
1718 CFS_RACE(CFS_FAIL_CHLOG_USER_REG_UNREG_RACE);
1723 llog_ctxt_put(ctxt);
1728 struct mdd_changelog_user_clear {
1736 * changelog_clear callback
1738 * Is called once per user.
1740 * Check to see the user requested is available from rec.
1741 * Check the oldest (smallest) record for boundary conditions.
1742 * Truncate the changelog.
1744 static int mdd_changelog_clear_cb(const struct lu_env *env,
1745 struct llog_handle *llh,
1746 struct llog_rec_hdr *hdr,
1749 struct llog_changelog_user_rec *rec;
1750 struct mdd_changelog_user_clear *mcuc = data;
1755 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
1758 rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr);
1760 /* Does the changelog id match the requested id? */
1761 if (rec->cur_id != mcuc->mcuc_id) {
1762 mcuc->mcuc_minrec = min(mcuc->mcuc_minrec,
1767 /* cur_endrec is the oldest purgeable record, make sure we're newer */
1768 if (rec->cur_endrec > mcuc->mcuc_endrec) {
1769 CDEBUG(D_IOCTL, "Request %llu out of range: %llu\n",
1770 mcuc->mcuc_endrec, rec->cur_endrec);
1774 /* Flag that we've met all the range and user checks.
1775 * We now know the record to flush.
1777 rec->cur_endrec = mcuc->mcuc_endrec;
1779 rec->cur_time = (__u32)ktime_get_real_seconds();
1780 if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER))
1783 mcuc->mcuc_flush = true;
1785 CDEBUG(D_IOCTL, "Rewriting changelog user %u endrec to %llu\n",
1786 mcuc->mcuc_id, rec->cur_endrec);
1788 /* Update the endrec */
1789 rc = llog_write(env, llh, hdr, hdr->lrh_index);
1795 * Clear a changelog up to entry specified by endrec for user id.
1797 static int mdd_changelog_clear(const struct lu_env *env,
1798 struct mdd_device *mdd, __u32 id,
1801 struct mdd_changelog_user_clear mcuc = {
1803 .mcuc_minrec = endrec,
1804 .mcuc_flush = false,
1806 struct llog_ctxt *ctxt;
1812 CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n",
1813 mdd2obd_dev(mdd)->obd_name, id, endrec);
1814 /* start_rec is the newest (largest value) entry in the changelogs*/
1815 spin_lock(&mdd->mdd_cl.mc_lock);
1816 start_rec = mdd->mdd_cl.mc_index;
1817 spin_unlock(&mdd->mdd_cl.mc_lock);
1819 if (start_rec < endrec) {
1820 CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\
1821 "address out of range\n", mdd2obd_dev(mdd)->obd_name);
1826 mcuc.mcuc_endrec = start_rec;
1827 mcuc.mcuc_minrec = ULLONG_MAX;
1829 mcuc.mcuc_endrec = endrec;
1832 ctxt = llog_get_context(mdd2obd_dev(mdd),
1833 LLOG_CHANGELOG_USER_ORIG_CTXT);
1835 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
1836 GOTO(out, rc = -ENXIO);
1838 rc = llog_cat_process(env, ctxt->loc_handle,
1839 mdd_changelog_clear_cb, (void *)&mcuc,
1842 if (rc == -EINVAL) {
1843 CDEBUG(D_IOCTL, "%s: No changelog recnum <= %llu to clear\n",
1844 mdd2obd_dev(mdd)->obd_name, (unsigned long long) endrec);
1846 } else if (rc < 0) {
1847 CWARN("%s: Failure to clear the changelog for user %d: %d\n",
1848 mdd2obd_dev(mdd)->obd_name, id, rc);
1849 } else if (mcuc.mcuc_flush) {
1850 /* Cancelling record 0 destroys the entire changelog, make sure
1851 we don't do that unless we mean it. */
1852 if (mcuc.mcuc_minrec != 0) {
1853 CDEBUG(D_IOCTL, "%s: Purging changelog entries up "\
1854 "to %llu\n", mdd2obd_dev(mdd)->obd_name,
1857 rc = mdd_changelog_llog_cancel(env, mdd,
1861 CDEBUG(D_IOCTL, "%s: No entry for user %d\n",
1862 mdd2obd_dev(mdd)->obd_name, id);
1869 llog_ctxt_put(ctxt);
1875 * May be called remotely from mdt_iocontrol_handle or locally from
1876 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1877 * an obd_ioctl_data format (but local ioctl handler does).
1879 * \param len - data len
1880 * \param karg - ioctl data, in kernel space
1882 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1883 unsigned int cmd, int len, void *karg)
1885 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1886 struct obd_ioctl_data *data = karg;
1890 /* Doesn't use obd_ioctl_data */
1892 case OBD_IOC_CHANGELOG_CLEAR: {
1893 struct changelog_setinfo *cs = karg;
1895 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1896 RETURN(-EINPROGRESS);
1898 rc = mdd_changelog_clear(env, mdd, cs->cs_id,
1900 barrier_exit(mdd->mdd_bottom);
1903 case OBD_IOC_START_LFSCK: {
1904 rc = lfsck_start(env, mdd->mdd_bottom,
1905 (struct lfsck_start_param *)karg);
1908 case OBD_IOC_STOP_LFSCK: {
1909 rc = lfsck_stop(env, mdd->mdd_bottom,
1910 (struct lfsck_stop *)karg);
1913 case OBD_IOC_QUERY_LFSCK: {
1914 rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL,
1915 (struct lfsck_query *)karg);
1920 /* Below ioctls use obd_ioctl_data */
1921 if (len != sizeof(*data)) {
1922 CERROR("Bad ioctl size %d\n", len);
1925 if (data->ioc_version != OBD_IOCTL_VERSION) {
1926 CERROR("Bad magic %x != %x\n", data->ioc_version,
1932 case OBD_IOC_CHANGELOG_REG:
1933 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1934 RETURN(-EINPROGRESS);
1936 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1937 barrier_exit(mdd->mdd_bottom);
1939 case OBD_IOC_CHANGELOG_DEREG:
1940 if (unlikely(!barrier_entry(mdd->mdd_bottom)))
1941 RETURN(-EINPROGRESS);
1943 /* explicitly clear changelog first, to protect from crash in
1944 * the middle of purge that would lead to unregistered consumer
1945 * but pending changelog entries
1947 rc = mdd_changelog_clear(env, mdd, data->ioc_u32_1, 0);
1949 rc = mdd_changelog_user_purge(env,
1950 mdd, data->ioc_u32_1);
1952 barrier_exit(mdd->mdd_bottom);
1961 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1962 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
1964 static const struct md_device_operations mdd_ops = {
1965 .mdo_statfs = mdd_statfs,
1966 .mdo_root_get = mdd_root_get,
1967 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1968 .mdo_iocontrol = mdd_iocontrol,
1969 .mdo_dtconf_get = mdd_dtconf_get,
1972 static const struct lu_device_type_operations mdd_device_type_ops = {
1973 .ldto_init = mdd_type_init,
1974 .ldto_fini = mdd_type_fini,
1976 .ldto_start = mdd_type_start,
1977 .ldto_stop = mdd_type_stop,
1979 .ldto_device_alloc = mdd_device_alloc,
1980 .ldto_device_free = mdd_device_free,
1982 .ldto_device_fini = mdd_device_fini
1985 static struct lu_device_type mdd_device_type = {
1986 .ldt_tags = LU_DEVICE_MD,
1987 .ldt_name = LUSTRE_MDD_NAME,
1988 .ldt_ops = &mdd_device_type_ops,
1989 .ldt_ctx_tags = LCT_MD_THREAD
1992 /* context key constructor: mdd_key_init */
1993 LU_KEY_INIT(mdd, struct mdd_thread_info);
1995 static void mdd_key_fini(const struct lu_context *ctx,
1996 struct lu_context_key *key, void *data)
1998 struct mdd_thread_info *info = data;
2000 lu_buf_free(&info->mti_big_buf);
2001 lu_buf_free(&info->mti_link_buf);
2002 lu_buf_free(&info->mti_xattr_buf);
2003 lu_buf_free(&info->mti_chlg_buf);
2008 /* context key: mdd_thread_key */
2009 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
2011 int mdd_generic_thread_start(struct mdd_generic_thread *thread,
2012 int (*func)(void *), void *data, char *name)
2014 struct task_struct *task;
2016 LASSERT(thread->mgt_init == false);
2017 init_completion(&thread->mgt_started);
2018 init_completion(&thread->mgt_finished);
2019 thread->mgt_data = data;
2020 thread->mgt_abort = false;
2021 thread->mgt_init = true;
2023 task = kthread_run(func, thread, name);
2025 complete(&thread->mgt_finished);
2026 return PTR_ERR(task);
2028 wait_for_completion(&thread->mgt_started);
2032 void mdd_generic_thread_stop(struct mdd_generic_thread *thread)
2034 if (thread->mgt_init == true) {
2035 thread->mgt_abort = true;
2036 wait_for_completion(&thread->mgt_finished);
2040 static int __init mdd_init(void)
2044 rc = lu_kmem_init(mdd_caches);
2048 changelog_orig_logops = llog_common_cat_ops;
2049 changelog_orig_logops.lop_write_rec = mdd_changelog_write_rec;
2051 rc = class_register_type(&mdd_obd_device_ops, NULL, false,
2052 LUSTRE_MDD_NAME, &mdd_device_type);
2054 lu_kmem_fini(mdd_caches);
2058 static void __exit mdd_exit(void)
2060 class_unregister_type(LUSTRE_MDD_NAME);
2061 lu_kmem_fini(mdd_caches);
2064 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2065 MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")");
2066 MODULE_VERSION(LUSTRE_VERSION_STRING);
2067 MODULE_LICENSE("GPL");
2069 module_init(mdd_init);
2070 module_exit(mdd_exit);