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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_device.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #include <obd_class.h>
47 #include <lustre_mds.h>
48 #include <obd_support.h>
49 #include <lu_object.h>
50 #include <lustre_param.h>
51 #include <lustre_fid.h>
53 #include "mdd_internal.h"
55 const struct md_device_operations mdd_ops;
56 static struct lu_device_type mdd_device_type;
58 static const char mdd_root_dir_name[] = "ROOT";
59 static const char mdd_obf_dir_name[] = "fid";
61 /* Slab for MDD object allocation */
62 cfs_mem_cache_t *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)
123 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
124 mdd->mdd_md_dev.md_ops = &mdd_ops;
126 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
130 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
131 /* sync permission changes */
132 mdd->mdd_sync_permission = 1;
134 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
136 /* we are using service name but not mdd obd name
137 * for compatibility reasons.
138 * It is passed from MDT in lustre_cfg[2] buffer */
139 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
141 obd_disconnect(mdd->mdd_child_exp);
146 static struct lu_device *mdd_device_fini(const struct lu_env *env,
149 struct mdd_device *mdd = lu2mdd_dev(d);
153 lu_dev_del_linkage(d->ld_site, d);
155 rc = mdd_procfs_fini(mdd);
157 CERROR("proc fini error %d \n", rc);
163 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
164 struct llog_rec_hdr *hdr, void *data)
166 struct mdd_device *mdd = (struct mdd_device *)data;
167 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
169 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
170 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
173 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log"
174 DOSTID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
175 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
176 rec->cr.cr_name, POSTID(&llh->lgh_id.lgl_oi));
178 mdd->mdd_cl.mc_index = rec->cr.cr_index;
179 return LLOG_PROC_BREAK;
182 static int changelog_user_init_cb(const struct lu_env *env,
183 struct llog_handle *llh,
184 struct llog_rec_hdr *hdr, void *data)
186 struct mdd_device *mdd = (struct mdd_device *)data;
187 struct llog_changelog_user_rec *rec =
188 (struct llog_changelog_user_rec *)hdr;
190 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
191 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
193 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
194 " in log "DOSTID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
195 rec->cur_id, rec->cur_endrec, POSTID(&llh->lgh_id.lgl_oi));
197 spin_lock(&mdd->mdd_cl.mc_user_lock);
198 mdd->mdd_cl.mc_lastuser = rec->cur_id;
199 spin_unlock(&mdd->mdd_cl.mc_user_lock);
201 return LLOG_PROC_BREAK;
204 static int llog_changelog_cancel_cb(const struct lu_env *env,
205 struct llog_handle *llh,
206 struct llog_rec_hdr *hdr, void *data)
208 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
209 struct llog_cookie cookie;
210 long long endrec = *(long long *)data;
215 /* This is always a (sub)log, not the catalog */
216 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
218 if (rec->cr.cr_index > endrec)
219 /* records are in order, so we're done */
220 RETURN(LLOG_PROC_BREAK);
222 cookie.lgc_lgl = llh->lgh_id;
223 cookie.lgc_index = hdr->lrh_index;
225 /* cancel them one at a time. I suppose we could store up the cookies
226 * and cancel them all at once; probably more efficient, but this is
227 * done as a user call, so who cares... */
228 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
230 RETURN(rc < 0 ? rc : 0);
233 static int llog_changelog_cancel(const struct lu_env *env,
234 struct llog_ctxt *ctxt,
235 struct lov_stripe_md *lsm, int count,
236 struct llog_cookie *cookies, int flags)
238 struct llog_handle *cathandle = ctxt->loc_handle;
243 /* This should only be called with the catalog handle */
244 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
246 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
247 (void *)cookies, 0, 0);
249 /* 0 or 1 means we're done */
252 CERROR("%s: cancel idx %u of catalog "DOSTID" rc=%d\n",
253 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
254 POSTID(&cathandle->lgh_id.lgl_oi), rc);
259 static struct llog_operations changelog_orig_logops;
261 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
263 static int mdd_changelog_llog_init(const struct lu_env *env,
264 struct mdd_device *mdd)
266 struct obd_device *obd = mdd2obd_dev(mdd);
267 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
272 /* LU-2844 mdd setup failure should not cause umount oops */
273 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
276 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
277 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
278 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
279 obd, &changelog_orig_logops);
281 CERROR("%s: changelog llog setup failed: rc = %d\n",
286 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
289 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
292 GOTO(out_cleanup, rc);
294 rc = llog_cat_init_and_process(env, ctxt->loc_handle);
298 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
299 changelog_init_cb, mdd);
302 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
307 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
308 mdd->mdd_cl.mc_index);
310 /* setup user changelog */
311 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
312 obd, &changelog_orig_logops);
314 CERROR("%s: changelog users llog setup failed: rc = %d\n",
319 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
322 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
325 GOTO(out_ucleanup, rc);
327 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
328 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
330 rc = llog_cat_init_and_process(env, uctxt->loc_handle);
332 GOTO(out_uclose, rc);
334 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
335 changelog_user_init_cb, mdd);
337 CERROR("%s: changelog user init failed: rc = %d\n",
339 GOTO(out_uclose, rc);
342 /* If we have registered users, assume we want changelogs on */
343 if (mdd->mdd_cl.mc_lastuser > 0) {
344 rc = mdd_changelog_on(env, mdd, 1);
346 GOTO(out_uclose, rc);
349 llog_ctxt_put(uctxt);
352 llog_cat_close(env, uctxt->loc_handle);
354 llog_cleanup(env, uctxt);
356 llog_cat_close(env, ctxt->loc_handle);
358 llog_cleanup(env, ctxt);
362 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
364 struct obd_device *obd = mdd2obd_dev(mdd);
367 mdd->mdd_cl.mc_index = 0;
368 spin_lock_init(&mdd->mdd_cl.mc_lock);
369 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
370 mdd->mdd_cl.mc_flags = 0; /* off by default */
371 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
372 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
373 mdd->mdd_cl.mc_lastuser = 0;
375 rc = mdd_changelog_llog_init(env, mdd);
377 CERROR("%s: changelog setup during init failed: rc = %d\n",
379 mdd->mdd_cl.mc_flags |= CLM_ERR;
385 static void mdd_changelog_fini(const struct lu_env *env,
386 struct mdd_device *mdd)
388 struct obd_device *obd = mdd2obd_dev(mdd);
389 struct llog_ctxt *ctxt;
391 mdd->mdd_cl.mc_flags = 0;
393 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
395 llog_cat_close(env, ctxt->loc_handle);
396 llog_cleanup(env, ctxt);
398 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
400 llog_cat_close(env, ctxt->loc_handle);
401 llog_cleanup(env, ctxt);
405 int mdd_changelog_write_header(const struct lu_env *env,
406 struct mdd_device *mdd, int markerflags);
408 /* Start / stop recording */
409 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
413 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
414 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
415 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
416 CERROR("Changelogs cannot be enabled due to error "
417 "condition (see %s log).\n",
418 mdd2obd_dev(mdd)->obd_name);
421 spin_lock(&mdd->mdd_cl.mc_lock);
422 mdd->mdd_cl.mc_flags |= CLM_ON;
423 spin_unlock(&mdd->mdd_cl.mc_lock);
424 rc = mdd_changelog_write_header(env, mdd, CLM_START);
426 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
427 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
428 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
429 spin_lock(&mdd->mdd_cl.mc_lock);
430 mdd->mdd_cl.mc_flags &= ~CLM_ON;
431 spin_unlock(&mdd->mdd_cl.mc_lock);
436 /** Remove entries with indicies up to and including \a endrec from the
442 int mdd_changelog_llog_cancel(const struct lu_env *env,
443 struct mdd_device *mdd, long long endrec)
445 struct obd_device *obd = mdd2obd_dev(mdd);
446 struct llog_ctxt *ctxt;
447 long long unsigned cur;
450 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
454 spin_lock(&mdd->mdd_cl.mc_lock);
455 cur = (long long)mdd->mdd_cl.mc_index;
456 spin_unlock(&mdd->mdd_cl.mc_lock);
460 /* purge to "0" is shorthand for everything */
464 /* If purging all records, write a header entry so we don't have an
465 empty catalog and we're sure to have a valid starting index next
466 time. In case of crash, we just restart with old log so we're
469 /* XXX: transaction is started by llog itself */
470 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
475 /* Some records were purged, so reset repeat-access time (so we
476 record new mtime update records, so users can see a file has been
477 changed since the last purge) */
478 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
480 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
486 /** Add a CL_MARK record to the changelog
488 * \param markerflags - CLM_*
491 int mdd_changelog_write_header(const struct lu_env *env,
492 struct mdd_device *mdd, int markerflags)
494 struct obd_device *obd = mdd2obd_dev(mdd);
495 struct llog_changelog_rec *rec;
497 struct llog_ctxt *ctxt;
499 int len = strlen(obd->obd_name);
504 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
505 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
509 reclen = llog_data_len(sizeof(*rec) + len);
510 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
511 if (buf->lb_buf == NULL)
515 rec->cr.cr_flags = CLF_VERSION;
516 rec->cr.cr_type = CL_MARK;
517 rec->cr.cr_namelen = len;
518 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
519 /* Status and action flags */
520 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
521 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
522 rec->cr_hdr.lrh_type = CHANGELOG_REC;
523 rec->cr.cr_time = cl_time();
524 spin_lock(&mdd->mdd_cl.mc_lock);
525 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
526 spin_unlock(&mdd->mdd_cl.mc_lock);
528 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
531 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
536 /* assume on or off event; reset repeat-access time */
537 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
542 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
543 * are valid. We also check if object with passed fid exists or not.
545 static int obf_lookup(const struct lu_env *env, struct md_object *p,
546 const struct lu_name *lname, struct lu_fid *f,
547 struct md_op_spec *spec)
549 char *name = (char *)lname->ln_name;
550 struct mdd_device *mdd = mdo2mdd(p);
551 struct mdd_object *child;
557 sscanf(name, SFID, RFID(f));
558 if (!fid_is_sane(f)) {
559 CWARN("%s: bad FID format [%s], should be "DFID"\n",
560 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
561 (__u64)FID_SEQ_NORMAL, 1, 0);
562 GOTO(out, rc = -EINVAL);
565 if (!fid_is_norm(f)) {
566 CWARN("%s: "DFID" is invalid, sequence should be "
567 ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
568 (__u64)FID_SEQ_NORMAL);
569 GOTO(out, rc = -EINVAL);
572 /* Check if object with this fid exists */
573 child = mdd_object_find(env, mdd, f);
577 GOTO(out, rc = PTR_ERR(child));
579 if (mdd_object_exists(child) == 0)
582 mdd_object_put(env, child);
588 static int obf_create(const struct lu_env *env, struct md_object *pobj,
589 const struct lu_name *lname, struct md_object *child,
590 struct md_op_spec *spec, struct md_attr* ma)
595 static int obf_rename(const struct lu_env *env,
596 struct md_object *src_pobj, struct md_object *tgt_pobj,
597 const struct lu_fid *lf, const struct lu_name *lsname,
598 struct md_object *tobj, const struct lu_name *ltname,
604 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
605 struct md_object *src_obj, const struct lu_name *lname,
611 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
612 struct md_object *cobj, const struct lu_name *lname,
613 struct md_attr *ma, int no_name)
618 static struct md_dir_operations mdd_obf_dir_ops = {
619 .mdo_lookup = obf_lookup,
620 .mdo_create = obf_create,
621 .mdo_rename = obf_rename,
622 .mdo_link = obf_link,
623 .mdo_unlink = obf_unlink
626 static struct md_object *mdo_locate(const struct lu_env *env,
627 struct md_device *md,
628 const struct lu_fid *fid)
630 struct lu_object *obj;
631 struct md_object *mdo;
633 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
635 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
636 LASSERT(obj != NULL);
639 mdo = ERR_PTR(PTR_ERR(obj));
645 * Create special in-memory "fid" object for open-by-fid.
647 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
649 struct md_object *mdo;
650 struct mdd_object *mdd_obf;
651 struct lu_fid fid = LU_OBF_FID;
654 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
655 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
659 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
661 RETURN(PTR_ERR(mdo));
663 LASSERT(lu_object_exists(&mdo->mo_lu));
665 mdd_obf = md2mdd_obj(mdo);
666 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
667 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
672 /** Setup ".lustre" directory object */
673 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
675 struct md_object *mdo;
680 /* Create ".lustre" directory in ROOT. */
681 fid = LU_DOT_LUSTRE_FID;
682 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
684 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
688 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
690 RETURN(PTR_ERR(mdo));
691 LASSERT(lu_object_exists(&mdo->mo_lu));
693 m->mdd_dot_lustre = md2mdd_obj(mdo);
695 rc = mdd_obf_setup(env, m);
697 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
698 mdd2obd_dev(m)->obd_name, rc);
703 mdd_object_put(env, m->mdd_dot_lustre);
704 m->mdd_dot_lustre = NULL;
708 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
709 struct lustre_cfg *cfg)
711 mdd_lfsck_cleanup(env, m);
712 mdd_changelog_fini(env, m);
713 orph_index_fini(env, m);
714 if (m->mdd_dot_lustre_objs.mdd_obf)
715 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
716 if (m->mdd_dot_lustre)
717 mdd_object_put(env, m->mdd_dot_lustre);
718 if (m->mdd_los != NULL)
719 local_oid_storage_fini(env, m->mdd_los);
720 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
722 if (m->mdd_child_exp)
723 obd_disconnect(m->mdd_child_exp);
726 static int mdd_process_config(const struct lu_env *env,
727 struct lu_device *d, struct lustre_cfg *cfg)
729 struct mdd_device *m = lu2mdd_dev(d);
730 struct dt_device *dt = m->mdd_child;
731 struct lu_device *next = &dt->dd_lu_dev;
735 switch (cfg->lcfg_command) {
737 struct lprocfs_static_vars lvars;
739 lprocfs_mdd_init_vars(&lvars);
740 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
741 if (rc > 0 || rc == -ENOSYS)
742 /* we don't understand; pass it on */
743 rc = next->ld_ops->ldo_process_config(env, next, cfg);
747 rc = next->ld_ops->ldo_process_config(env, next, cfg);
750 dt_conf_get(env, dt, &m->mdd_dt_conf);
753 rc = next->ld_ops->ldo_process_config(env, next, cfg);
754 lu_dev_del_linkage(d->ld_site, d);
755 mdd_device_shutdown(env, m, cfg);
758 rc = next->ld_ops->ldo_process_config(env, next, cfg);
765 static int mdd_recovery_complete(const struct lu_env *env,
768 struct mdd_device *mdd = lu2mdd_dev(d);
769 struct lu_device *next;
773 LASSERT(mdd != NULL);
774 next = &mdd->mdd_child->dd_lu_dev;
776 /* XXX: orphans handling. */
777 __mdd_orphan_cleanup(env, mdd);
778 rc = next->ld_ops->ldo_recovery_complete(env, next);
783 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
784 const struct lu_fid *pfid, const char *name, __u32 mode,
787 struct dt_object *parent, *dto;
792 LASSERT(!fid_is_zero(pfid));
793 parent = dt_locate(env, mdd->mdd_bottom, pfid);
794 if (unlikely(IS_ERR(parent)))
795 RETURN(PTR_ERR(parent));
797 /* create local file/dir, if @fid is passed then try to use it */
798 if (fid_is_zero(fid))
799 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
802 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
806 GOTO(out_put, rc = PTR_ERR(dto));
807 *fid = *lu_object_fid(&dto->do_lu);
808 /* since stack is not fully set up the local_storage uses own stack
809 * and we should drop its object from cache */
810 lu_object_put_nocache(env, &dto->do_lu);
813 lu_object_put(env, &parent->do_lu);
817 static int mdd_prepare(const struct lu_env *env,
818 struct lu_device *pdev,
819 struct lu_device *cdev)
821 struct mdd_device *mdd = lu2mdd_dev(cdev);
822 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
828 rc = next->ld_ops->ldo_prepare(env, cdev, next);
832 /* Setup local dirs */
833 fid.f_seq = FID_SEQ_LOCAL_NAME;
836 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
841 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
845 if (mdd_seq_site(mdd)->ss_node_id == 0) {
847 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
848 mdd_root_dir_name, S_IFDIR |
849 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
851 CERROR("%s: create root fid failed: rc = %d\n",
852 mdd2obd_dev(mdd)->obd_name, rc);
855 mdd->mdd_root_fid = fid;
857 rc = mdd_dot_lustre_setup(env, mdd);
859 CERROR("%s: initializing .lustre failed: rc = %d\n",
860 mdd2obd_dev(mdd)->obd_name, rc);
864 rc = mdd_compat_fixes(env, mdd);
870 rc = orph_index_init(env, mdd);
874 rc = mdd_changelog_init(env, mdd);
876 CERROR("%s: failed to initialize changelog: rc = %d\n",
877 mdd2obd_dev(mdd)->obd_name, rc);
881 rc = mdd_lfsck_setup(env, mdd);
883 CERROR("%s: failed to initialize lfsck: rc = %d\n",
884 mdd2obd_dev(mdd)->obd_name, rc);
885 GOTO(out_changelog, rc);
889 mdd_changelog_fini(env, mdd);
891 orph_index_fini(env, mdd);
893 if (mdd_seq_site(mdd)->ss_node_id == 0) {
894 mdd_object_put(env, mdd->mdd_dot_lustre);
895 mdd->mdd_dot_lustre = NULL;
896 mdd_object_put(env, mdd->mdd_dot_lustre_objs.mdd_obf);
897 mdd->mdd_dot_lustre_objs.mdd_obf = NULL;
900 local_oid_storage_fini(env, mdd->mdd_los);
905 const struct lu_device_operations mdd_lu_ops = {
906 .ldo_object_alloc = mdd_object_alloc,
907 .ldo_process_config = mdd_process_config,
908 .ldo_recovery_complete = mdd_recovery_complete,
909 .ldo_prepare = mdd_prepare,
912 static int mdd_root_get(const struct lu_env *env,
913 struct md_device *m, struct lu_fid *f)
915 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
918 *f = mdd->mdd_root_fid;
923 * No permission check is needed.
925 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
926 struct obd_statfs *sfs)
928 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
933 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
939 * No permission check is needed.
941 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
942 int mode, unsigned long timeout, __u32 alg,
943 struct lustre_capa_key *keys)
945 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
949 /* need barrier for mds_capa_keys access. */
951 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
956 static int mdd_update_capa_key(const struct lu_env *env,
958 struct lustre_capa_key *key)
960 /* we do not support capabilities ... */
964 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
967 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
969 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
970 return (*h == NULL ? -ENOENT : 0);
973 static struct lu_device *mdd_device_free(const struct lu_env *env,
974 struct lu_device *lu)
976 struct mdd_device *m = lu2mdd_dev(lu);
979 LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
980 md_device_fini(&m->mdd_md_dev);
985 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
986 struct lu_device_type *t,
987 struct lustre_cfg *lcfg)
990 struct mdd_device *m;
994 l = ERR_PTR(-ENOMEM);
999 md_device_init(&m->mdd_md_dev, t);
1000 rc = mdd_init0(env, m, t, lcfg);
1002 mdd_device_free(env, l);
1011 * we use exports to track all mdd users
1013 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1014 struct obd_device *obd, struct obd_uuid *cluuid,
1015 struct obd_connect_data *data, void *localdata)
1017 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1018 struct lustre_handle conn;
1022 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1024 rc = class_connect(&conn, obd, cluuid);
1028 *exp = class_conn2export(&conn);
1030 /* Why should there ever be more than 1 connect? */
1031 LASSERT(mdd->mdd_connects == 0);
1032 mdd->mdd_connects++;
1038 * once last export (we don't count self-export) disappeared
1039 * mdd can be released
1041 static int mdd_obd_disconnect(struct obd_export *exp)
1043 struct obd_device *obd = exp->exp_obd;
1044 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1045 int rc, release = 0;
1048 mdd->mdd_connects--;
1049 if (mdd->mdd_connects == 0)
1052 rc = class_disconnect(exp);
1054 if (rc == 0 && release)
1055 class_manual_cleanup(obd);
1059 static int mdd_obd_health_check(const struct lu_env *env,
1060 struct obd_device *obd)
1062 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1067 rc = obd_health_check(env, mdd->mdd_child_exp->exp_obd);
1071 static struct obd_ops mdd_obd_device_ops = {
1072 .o_owner = THIS_MODULE,
1073 .o_connect = mdd_obd_connect,
1074 .o_disconnect = mdd_obd_disconnect,
1075 .o_health_check = mdd_obd_health_check
1079 * context key constructor/destructor:
1080 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1082 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1084 struct lu_context_key mdd_capainfo_key = {
1085 .lct_tags = LCT_SESSION,
1086 .lct_init = mdd_capainfo_key_init,
1087 .lct_fini = mdd_capainfo_key_fini
1090 struct md_capainfo *md_capainfo(const struct lu_env *env)
1092 /* NB, in mdt_init0 */
1093 if (env->le_ses == NULL)
1095 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1097 EXPORT_SYMBOL(md_capainfo);
1099 static int mdd_changelog_user_register(const struct lu_env *env,
1100 struct mdd_device *mdd, int *id)
1102 struct llog_ctxt *ctxt;
1103 struct llog_changelog_user_rec *rec;
1107 ctxt = llog_get_context(mdd2obd_dev(mdd),
1108 LLOG_CHANGELOG_USER_ORIG_CTXT);
1114 llog_ctxt_put(ctxt);
1118 /* Assume we want it on since somebody registered */
1119 rc = mdd_changelog_on(env, mdd, 1);
1123 rec->cur_hdr.lrh_len = sizeof(*rec);
1124 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1125 spin_lock(&mdd->mdd_cl.mc_user_lock);
1126 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1127 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1128 CERROR("Maximum number of changelog users exceeded!\n");
1129 GOTO(out, rc = -EOVERFLOW);
1131 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1132 rec->cur_endrec = mdd->mdd_cl.mc_index;
1133 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1135 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
1137 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1140 llog_ctxt_put(ctxt);
1144 struct mdd_changelog_user_data {
1145 __u64 mcud_endrec; /**< purge record for this user */
1146 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1148 __u32 mcud_minid; /**< user id with lowest rec reference */
1149 __u32 mcud_usercount;
1150 unsigned int mcud_found:1;
1152 #define MCUD_UNREGISTER -1LL
1155 * 1. Find the smallest record everyone is willing to purge
1156 * 2. Update the last purgeable record for this user
1158 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1159 struct llog_handle *llh,
1160 struct llog_rec_hdr *hdr, void *data)
1162 struct llog_changelog_user_rec *rec;
1163 struct mdd_changelog_user_data *mcud = data;
1168 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1170 rec = (struct llog_changelog_user_rec *)hdr;
1172 mcud->mcud_usercount++;
1174 /* If we have a new endrec for this id, use it for the following
1175 min check instead of its old value */
1176 if (rec->cur_id == mcud->mcud_id)
1177 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1179 /* Track the minimum referenced record */
1180 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1181 mcud->mcud_minid = rec->cur_id;
1182 mcud->mcud_minrec = rec->cur_endrec;
1185 if (rec->cur_id != mcud->mcud_id)
1188 /* Update this user's record */
1189 mcud->mcud_found = 1;
1191 /* Special case: unregister this user */
1192 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1193 struct llog_cookie cookie;
1195 cookie.lgc_lgl = llh->lgh_id;
1196 cookie.lgc_index = hdr->lrh_index;
1198 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1201 mcud->mcud_usercount--;
1206 /* Update the endrec */
1207 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1208 mcud->mcud_id, rec->cur_endrec);
1210 /* hdr+1 is loc of data */
1211 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1212 rc = llog_write(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1218 static int mdd_changelog_user_purge(const struct lu_env *env,
1219 struct mdd_device *mdd, int id,
1222 struct mdd_changelog_user_data data;
1223 struct llog_ctxt *ctxt;
1227 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1230 data.mcud_minid = 0;
1231 data.mcud_minrec = 0;
1232 data.mcud_usercount = 0;
1233 data.mcud_endrec = endrec;
1234 spin_lock(&mdd->mdd_cl.mc_lock);
1235 endrec = mdd->mdd_cl.mc_index;
1236 spin_unlock(&mdd->mdd_cl.mc_lock);
1237 if ((data.mcud_endrec == 0) ||
1238 ((data.mcud_endrec > endrec) &&
1239 (data.mcud_endrec != MCUD_UNREGISTER)))
1240 data.mcud_endrec = endrec;
1242 ctxt = llog_get_context(mdd2obd_dev(mdd),
1243 LLOG_CHANGELOG_USER_ORIG_CTXT);
1247 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1249 rc = llog_cat_process(env, ctxt->loc_handle,
1250 mdd_changelog_user_purge_cb, (void *)&data,
1252 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1253 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1254 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1255 data.mcud_minrec, data.mcud_minid);
1256 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1258 CWARN("Could not determine changelog records to purge; rc=%d\n",
1262 llog_ctxt_put(ctxt);
1264 if (!data.mcud_found) {
1265 CWARN("No entry for user %d. Last changelog reference is "
1266 LPD64" by changelog user %d\n", data.mcud_id,
1267 data.mcud_minrec, data.mcud_minid);
1271 if (!rc && data.mcud_usercount == 0)
1272 /* No more users; turn changelogs off */
1273 rc = mdd_changelog_on(env, mdd, 0);
1279 * May be called remotely from mdt_iocontrol_handle or locally from
1280 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1281 * an obd_ioctl_data format (but local ioctl handler does).
1283 * \param len - data len
1284 * \param karg - ioctl data, in kernel space
1286 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1287 unsigned int cmd, int len, void *karg)
1289 struct mdd_device *mdd;
1290 struct obd_ioctl_data *data = karg;
1294 mdd = lu2mdd_dev(&m->md_lu_dev);
1296 /* Doesn't use obd_ioctl_data */
1298 case OBD_IOC_CHANGELOG_CLEAR: {
1299 struct changelog_setinfo *cs = karg;
1300 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1304 case OBD_IOC_GET_MNTOPT: {
1305 mntopt_t *mntopts = (mntopt_t *)karg;
1306 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1309 case OBD_IOC_START_LFSCK: {
1310 rc = mdd_lfsck_start(env, &mdd->mdd_lfsck,
1311 (struct lfsck_start *)karg);
1314 case OBD_IOC_STOP_LFSCK: {
1315 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck, false);
1318 case OBD_IOC_PAUSE_LFSCK: {
1319 rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck, true);
1324 /* Below ioctls use obd_ioctl_data */
1325 if (len != sizeof(*data)) {
1326 CERROR("Bad ioctl size %d\n", len);
1329 if (data->ioc_version != OBD_IOCTL_VERSION) {
1330 CERROR("Bad magic %x != %x\n", data->ioc_version,
1336 case OBD_IOC_CHANGELOG_REG:
1337 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1339 case OBD_IOC_CHANGELOG_DEREG:
1340 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1350 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1351 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_capainfo_key);
1353 const struct md_device_operations mdd_ops = {
1354 .mdo_statfs = mdd_statfs,
1355 .mdo_root_get = mdd_root_get,
1356 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1357 .mdo_update_capa_key= mdd_update_capa_key,
1358 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1359 .mdo_iocontrol = mdd_iocontrol,
1362 static struct lu_device_type_operations mdd_device_type_ops = {
1363 .ldto_init = mdd_type_init,
1364 .ldto_fini = mdd_type_fini,
1366 .ldto_start = mdd_type_start,
1367 .ldto_stop = mdd_type_stop,
1369 .ldto_device_alloc = mdd_device_alloc,
1370 .ldto_device_free = mdd_device_free,
1372 .ldto_device_fini = mdd_device_fini
1375 static struct lu_device_type mdd_device_type = {
1376 .ldt_tags = LU_DEVICE_MD,
1377 .ldt_name = LUSTRE_MDD_NAME,
1378 .ldt_ops = &mdd_device_type_ops,
1379 .ldt_ctx_tags = LCT_MD_THREAD
1382 /* context key constructor: mdd_key_init */
1383 LU_KEY_INIT(mdd, struct mdd_thread_info);
1385 static void mdd_key_fini(const struct lu_context *ctx,
1386 struct lu_context_key *key, void *data)
1388 struct mdd_thread_info *info = data;
1390 lu_buf_free(&info->mti_big_buf);
1391 lu_buf_free(&info->mti_link_buf);
1396 /* context key: mdd_thread_key */
1397 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1399 static int __init mdd_mod_init(void)
1401 struct lprocfs_static_vars lvars;
1404 lprocfs_mdd_init_vars(&lvars);
1406 rc = lu_kmem_init(mdd_caches);
1410 changelog_orig_logops = llog_osd_ops;
1411 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
1412 changelog_orig_logops.lop_add = llog_cat_add_rec;
1413 changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1415 rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1416 LUSTRE_MDD_NAME, &mdd_device_type);
1418 lu_kmem_fini(mdd_caches);
1422 static void __exit mdd_mod_exit(void)
1424 class_unregister_type(LUSTRE_MDD_NAME);
1425 lu_kmem_fini(mdd_caches);
1428 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1429 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1430 MODULE_LICENSE("GPL");
1432 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);