1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #include <linux/jbd.h>
51 #include <obd_class.h>
52 #include <lustre_ver.h>
53 #include <obd_support.h>
54 #include <lprocfs_status.h>
56 #include <lustre_disk.h>
57 #include <lustre_fid.h>
58 #include <linux/ldiskfs_fs.h>
59 #include <lustre_mds.h>
60 #include <lustre/lustre_idl.h>
61 #include <lustre_disk.h> /* for changelogs */
62 #include <lustre_param.h>
63 #include <lustre_fid.h>
65 #include "mdd_internal.h"
67 const struct md_device_operations mdd_ops;
68 static struct lu_device_type mdd_device_type;
70 static const char mdd_root_dir_name[] = "ROOT";
71 static const char mdd_obf_dir_name[] = "fid";
73 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
74 const char *name, struct lu_device *next)
76 struct mdd_device *mdd = lu2mdd_dev(d);
80 mdd->mdd_child = lu2dt_dev(next);
82 /* Prepare transactions callbacks. */
83 mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
84 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
85 mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb;
86 mdd->mdd_txn_cb.dtc_cookie = mdd;
87 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
88 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
90 rc = mdd_procfs_init(mdd, name);
94 static struct lu_device *mdd_device_fini(const struct lu_env *env,
97 struct mdd_device *mdd = lu2mdd_dev(d);
98 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
101 rc = mdd_procfs_fini(mdd);
103 CERROR("proc fini error %d \n", rc);
109 static void mdd_changelog_fini(const struct lu_env *env,
110 struct mdd_device *mdd);
112 static void mdd_device_shutdown(const struct lu_env *env,
113 struct mdd_device *m, struct lustre_cfg *cfg)
116 mdd_changelog_fini(env, m);
117 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
118 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
119 mdd_object_put(env, m->mdd_dot_lustre);
121 mdd_fini_obd(env, m, cfg);
122 orph_index_fini(env, m);
123 /* remove upcall device*/
124 md_upcall_fini(&m->mdd_md_dev);
128 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
131 struct mdd_device *mdd = (struct mdd_device *)data;
132 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
135 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
136 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
139 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
140 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
141 rec->cr_type, rec->cr_namelen, rec->cr_name,
142 llh->lgh_id.lgl_oid);
144 mdd->mdd_cl.mc_index = rec->cr_index;
145 RETURN(LLOG_PROC_BREAK);
148 static int changelog_user_init_cb(struct llog_handle *llh,
149 struct llog_rec_hdr *hdr, void *data)
151 struct mdd_device *mdd = (struct mdd_device *)data;
152 struct llog_changelog_user_rec *rec =
153 (struct llog_changelog_user_rec *)hdr;
156 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
157 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
159 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
160 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
161 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
163 spin_lock(&mdd->mdd_cl.mc_user_lock);
164 mdd->mdd_cl.mc_lastuser = rec->cur_id;
165 spin_unlock(&mdd->mdd_cl.mc_user_lock);
167 RETURN(LLOG_PROC_BREAK);
171 static int mdd_changelog_llog_init(struct mdd_device *mdd)
173 struct obd_device *obd = mdd2obd_dev(mdd);
174 struct llog_ctxt *ctxt;
177 /* Find last changelog entry number */
178 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
180 CERROR("no changelog context\n");
183 if (!ctxt->loc_handle) {
188 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
192 CERROR("changelog init failed: %d\n", rc);
195 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
196 mdd->mdd_cl.mc_index);
198 /* Find last changelog user id */
199 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
201 CERROR("no changelog user context\n");
204 if (!ctxt->loc_handle) {
209 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
214 CERROR("changelog user init failed: %d\n", rc);
218 /* If we have registered users, assume we want changelogs on */
219 if (mdd->mdd_cl.mc_lastuser > 0)
220 rc = mdd_changelog_on(mdd, 1);
225 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
229 mdd->mdd_cl.mc_index = 0;
230 spin_lock_init(&mdd->mdd_cl.mc_lock);
231 cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
232 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
233 mdd->mdd_cl.mc_flags = 0; /* off by default */
234 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
235 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
236 mdd->mdd_cl.mc_lastuser = 0;
238 rc = mdd_changelog_llog_init(mdd);
240 CERROR("Changelog setup during init failed %d\n", rc);
241 mdd->mdd_cl.mc_flags |= CLM_ERR;
247 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
249 mdd->mdd_cl.mc_flags = 0;
252 /* Start / stop recording */
253 int mdd_changelog_on(struct mdd_device *mdd, int on)
257 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
258 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
259 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
260 CERROR("Changelogs cannot be enabled due to error "
261 "condition (see %s log).\n",
262 mdd2obd_dev(mdd)->obd_name);
265 spin_lock(&mdd->mdd_cl.mc_lock);
266 mdd->mdd_cl.mc_flags |= CLM_ON;
267 spin_unlock(&mdd->mdd_cl.mc_lock);
268 rc = mdd_changelog_write_header(mdd, CLM_START);
270 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
271 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
272 rc = mdd_changelog_write_header(mdd, CLM_FINI);
273 spin_lock(&mdd->mdd_cl.mc_lock);
274 mdd->mdd_cl.mc_flags &= ~CLM_ON;
275 spin_unlock(&mdd->mdd_cl.mc_lock);
280 /** Add a changelog entry \a rec to the changelog llog
283 * \param handle - currently ignored since llogs start their own transaction;
284 * this will hopefully be fixed in llog rewrite
287 int mdd_changelog_llog_write(struct mdd_device *mdd,
288 struct llog_changelog_rec *rec,
289 struct thandle *handle)
291 struct obd_device *obd = mdd2obd_dev(mdd);
292 struct llog_ctxt *ctxt;
295 if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
298 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
299 /* llog_lvfs_write_rec sets the llog tail len */
300 rec->cr_hdr.lrh_type = CHANGELOG_REC;
301 rec->cr_time = cfs_time_current_64();
302 spin_lock(&mdd->mdd_cl.mc_lock);
303 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
304 but as long as the MDD transactions are ordered correctly for e.g.
305 rename conflicts, I don't think this should matter. */
306 rec->cr_index = ++mdd->mdd_cl.mc_index;
307 spin_unlock(&mdd->mdd_cl.mc_lock);
308 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
312 /* nested journal transaction */
313 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
316 cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
321 /** Remove entries with indicies up to and including \a endrec from the
327 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
329 struct obd_device *obd = mdd2obd_dev(mdd);
330 struct llog_ctxt *ctxt;
331 long long unsigned cur;
334 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
338 spin_lock(&mdd->mdd_cl.mc_lock);
339 cur = (long long)mdd->mdd_cl.mc_index;
340 spin_unlock(&mdd->mdd_cl.mc_lock);
344 /* purge to "0" is shorthand for everything */
348 /* If purging all records, write a header entry so we don't have an
349 empty catalog and we're sure to have a valid starting index next
350 time. In case of crash, we just restart with old log so we're
353 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
358 /* Some records were purged, so reset repeat-access time (so we
359 record new mtime update records, so users can see a file has been
360 changed since the last purge) */
361 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
363 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
369 /** Add a CL_MARK record to the changelog
371 * \param markerflags - CLM_*
374 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
376 struct obd_device *obd = mdd2obd_dev(mdd);
377 struct llog_changelog_rec *rec;
379 int len = strlen(obd->obd_name);
383 reclen = llog_data_len(sizeof(*rec) + len);
384 OBD_ALLOC(rec, reclen);
388 rec->cr_flags = CLF_VERSION;
389 rec->cr_type = CL_MARK;
390 rec->cr_namelen = len;
391 memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
392 /* Status and action flags */
393 rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
395 rc = mdd_changelog_llog_write(mdd, rec, NULL);
397 /* assume on or off event; reset repeat-access time */
398 mdd->mdd_cl.mc_starttime = rec->cr_time;
400 OBD_FREE(rec, reclen);
405 * Create ".lustre" directory.
407 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
409 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
410 struct md_object *mdo;
413 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
414 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
415 mdd_root_dir_name, mdd_dot_lustre_name,
416 fid, &dt_directory_features);
417 /* .lustre dir may be already present */
418 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
420 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
421 mdd_dot_lustre_name, PFID(fid), rc);
426 lu_object_put(env, &mdo->mo_lu);
431 static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
434 struct mdd_object *mdd_obj = md2mdd_obj(obj);
436 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
439 static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
440 const struct md_attr *ma)
445 static int dot_lustre_xattr_get(const struct lu_env *env,
446 struct md_object *obj, struct lu_buf *buf,
452 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
455 struct mdd_object *mdd_obj = md2mdd_obj(obj);
457 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
458 mdd_obj->mod_count++;
459 mdd_write_unlock(env, mdd_obj);
464 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
465 char *path, int pathlen, __u64 *recno, int *linkno)
470 static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
473 struct mdd_object *mdd_obj = md2mdd_obj(obj);
475 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
476 mdd_obj->mod_count--;
477 mdd_write_unlock(env, mdd_obj);
482 static struct md_object_operations mdd_dot_lustre_obj_ops = {
483 .moo_attr_get = dot_lustre_attr_get,
484 .moo_attr_set = dot_lustre_attr_set,
485 .moo_xattr_get = dot_lustre_xattr_get,
486 .moo_open = dot_lustre_mdd_open,
487 .moo_close = dot_lustre_close,
488 .moo_readpage = mdd_readpage,
489 .moo_path = dot_lustre_path
492 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
493 const struct lu_name *lname, struct lu_fid *f,
494 struct md_op_spec *spec)
496 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
504 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
505 const struct lu_name *lname,
506 struct md_object *child, struct md_op_spec *spec,
512 static int dot_lustre_rename(const struct lu_env *env,
513 struct md_object *src_pobj,
514 struct md_object *tgt_pobj,
515 const struct lu_fid *lf,
516 const struct lu_name *lsname,
517 struct md_object *tobj,
518 const struct lu_name *ltname, struct md_attr *ma)
523 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
524 struct md_object *src_obj,
525 const struct lu_name *lname, struct md_attr *ma)
530 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
531 struct md_object *cobj, const struct lu_name *lname,
537 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
538 .mdo_lookup = dot_lustre_lookup,
539 .mdo_create = dot_lustre_create,
540 .mdo_rename = dot_lustre_rename,
541 .mdo_link = dot_lustre_link,
542 .mdo_unlink = dot_lustre_unlink,
545 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
550 if (ma->ma_need & MA_INODE) {
551 struct mdd_device *mdd = mdo2mdd(obj);
553 /* "fid" is a virtual object and hence does not have any "real"
554 * attributes. So we reuse attributes of .lustre for "fid" dir */
555 ma->ma_need |= MA_INODE;
556 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
559 ma->ma_valid |= MA_INODE;
562 /* "fid" directory does not have any striping information. */
563 if (ma->ma_need & MA_LOV) {
564 struct mdd_object *mdd_obj = md2mdd_obj(obj);
566 if (ma->ma_valid & MA_LOV)
569 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
570 S_ISDIR(mdd_object_type(mdd_obj))))
573 if (ma->ma_need & MA_LOV_DEF) {
574 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
577 ma->ma_valid |= MA_LOV;
586 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
587 const struct md_attr *ma)
592 static int obf_xattr_get(const struct lu_env *env,
593 struct md_object *obj, struct lu_buf *buf,
599 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
602 struct mdd_object *mdd_obj = md2mdd_obj(obj);
604 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
605 mdd_obj->mod_count++;
606 mdd_write_unlock(env, mdd_obj);
611 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
614 struct mdd_object *mdd_obj = md2mdd_obj(obj);
616 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
617 mdd_obj->mod_count--;
618 mdd_write_unlock(env, mdd_obj);
623 /** Nothing to list in "fid" directory */
624 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
625 const struct lu_rdpg *rdpg)
630 static int obf_path(const struct lu_env *env, struct md_object *obj,
631 char *path, int pathlen, __u64 *recno, int *linkno)
636 static struct md_object_operations mdd_obf_obj_ops = {
637 .moo_attr_get = obf_attr_get,
638 .moo_attr_set = obf_attr_set,
639 .moo_xattr_get = obf_xattr_get,
640 .moo_open = obf_mdd_open,
641 .moo_close = obf_mdd_close,
642 .moo_readpage = obf_mdd_readpage,
647 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
648 * are valid. We also check if object with passed fid exists or not.
650 static int obf_lookup(const struct lu_env *env, struct md_object *p,
651 const struct lu_name *lname, struct lu_fid *f,
652 struct md_op_spec *spec)
654 char *name = (char *)lname->ln_name;
655 struct mdd_device *mdd = mdo2mdd(p);
656 struct mdd_object *child;
662 sscanf(name, SFID, &(f->f_seq), &(f->f_oid),
664 if (!fid_is_sane(f)) {
665 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
667 GOTO(out, rc = -EINVAL);
670 /* Check if object with this fid exists */
671 child = mdd_object_find(env, mdd, f);
675 GOTO(out, rc = PTR_ERR(child));
677 if (mdd_object_exists(child) == 0)
680 mdd_object_put(env, child);
686 static int obf_create(const struct lu_env *env, struct md_object *pobj,
687 const struct lu_name *lname, struct md_object *child,
688 struct md_op_spec *spec, struct md_attr* ma)
693 static int obf_rename(const struct lu_env *env,
694 struct md_object *src_pobj, struct md_object *tgt_pobj,
695 const struct lu_fid *lf, const struct lu_name *lsname,
696 struct md_object *tobj, const struct lu_name *ltname,
702 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
703 struct md_object *src_obj, const struct lu_name *lname,
709 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
710 struct md_object *cobj, const struct lu_name *lname,
716 static struct md_dir_operations mdd_obf_dir_ops = {
717 .mdo_lookup = obf_lookup,
718 .mdo_create = obf_create,
719 .mdo_rename = obf_rename,
720 .mdo_link = obf_link,
721 .mdo_unlink = obf_unlink
725 * Create special in-memory "fid" object for open-by-fid.
727 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
729 struct mdd_object *mdd_obf;
730 struct lu_object *obf_lu_obj;
733 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
735 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
736 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
737 GOTO(out, rc = -ENOENT);
739 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
740 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
741 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
742 /* Don't allow objects to be created in "fid" dir */
743 mdd_obf->mod_flags |= IMMUTE_OBJ;
745 obf_lu_obj = mdd2lu_obj(mdd_obf);
746 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
752 /** Setup ".lustre" directory object */
753 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
755 struct dt_object *dt_dot_lustre;
756 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
759 rc = create_dot_lustre_dir(env, m);
763 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
764 mdd_dot_lustre_name, fid);
765 if (IS_ERR(dt_dot_lustre)) {
766 rc = PTR_ERR(dt_dot_lustre);
770 /* references are released in mdd_device_shutdown() */
771 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
774 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
775 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
777 rc = mdd_obf_setup(env, m);
779 CERROR("Error initializing \"fid\" object - %d.\n", rc);
785 static int mdd_process_config(const struct lu_env *env,
786 struct lu_device *d, struct lustre_cfg *cfg)
788 struct mdd_device *m = lu2mdd_dev(d);
789 struct dt_device *dt = m->mdd_child;
790 struct lu_device *next = &dt->dd_lu_dev;
794 switch (cfg->lcfg_command) {
796 struct lprocfs_static_vars lvars;
798 lprocfs_mdd_init_vars(&lvars);
799 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
800 if (rc > 0 || rc == -ENOSYS)
801 /* we don't understand; pass it on */
802 rc = next->ld_ops->ldo_process_config(env, next, cfg);
806 rc = next->ld_ops->ldo_process_config(env, next, cfg);
809 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
811 rc = mdd_init_obd(env, m, cfg);
813 CERROR("lov init error %d \n", rc);
816 rc = mdd_txn_init_credits(env, m);
820 mdd_changelog_init(env, m);
823 mdd_device_shutdown(env, m, cfg);
825 rc = next->ld_ops->ldo_process_config(env, next, cfg);
833 static int mdd_lov_set_nextid(const struct lu_env *env,
834 struct mdd_device *mdd)
836 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
840 LASSERT(mds->mds_lov_objids != NULL);
841 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
842 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
843 mds->mds_lov_objids, NULL);
848 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
849 struct mdd_device *mdd)
851 /* XXX: to be implemented! */
856 static int mdd_recovery_complete(const struct lu_env *env,
859 struct mdd_device *mdd = lu2mdd_dev(d);
860 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
861 struct obd_device *obd = mdd2obd_dev(mdd);
865 LASSERT(mdd != NULL);
866 LASSERT(obd != NULL);
868 /* XXX: Do we need this in new stack? */
869 rc = mdd_lov_set_nextid(env, mdd);
871 CERROR("mdd_lov_set_nextid() failed %d\n",
876 /* XXX: cleanup unlink. */
877 rc = mdd_cleanup_unlink_llog(env, mdd);
879 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
884 /* Call that with obd_recovering = 1 just to update objids */
885 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
886 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
888 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
889 obd->obd_recovering = 0;
890 obd->obd_type->typ_dt_ops->o_postrecov(obd);
892 /* XXX: orphans handling. */
893 __mdd_orphan_cleanup(env, mdd);
894 rc = next->ld_ops->ldo_recovery_complete(env, next);
899 static int mdd_prepare(const struct lu_env *env,
900 struct lu_device *pdev,
901 struct lu_device *cdev)
903 struct mdd_device *mdd = lu2mdd_dev(cdev);
904 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
905 struct dt_object *root;
909 rc = next->ld_ops->ldo_prepare(env, cdev, next);
913 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
914 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
917 LASSERT(root != NULL);
918 lu_object_put(env, &root->do_lu);
919 rc = orph_index_init(env, mdd);
926 rc = mdd_dot_lustre_setup(env, mdd);
928 CERROR("Error(%d) initializing .lustre objects\n", rc);
936 const struct lu_device_operations mdd_lu_ops = {
937 .ldo_object_alloc = mdd_object_alloc,
938 .ldo_process_config = mdd_process_config,
939 .ldo_recovery_complete = mdd_recovery_complete,
940 .ldo_prepare = mdd_prepare,
944 * No permission check is needed.
946 static int mdd_root_get(const struct lu_env *env,
947 struct md_device *m, struct lu_fid *f)
949 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
952 *f = mdd->mdd_root_fid;
957 * No permission check is needed.
959 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
962 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
967 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
973 * No permission check is needed.
975 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
976 int *md_size, int *cookie_size)
978 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
981 *md_size = mdd_lov_mdsize(env, mdd);
982 *cookie_size = mdd_lov_cookiesize(env, mdd);
987 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
988 int mode, unsigned long timeout, __u32 alg,
989 struct lustre_capa_key *keys)
991 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
992 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
996 mds->mds_capa_keys = keys;
997 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1002 static int mdd_update_capa_key(const struct lu_env *env,
1003 struct md_device *m,
1004 struct lustre_capa_key *key)
1006 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1007 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
1011 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1012 sizeof(*key), key, NULL);
1016 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1019 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1021 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1022 return (*h == NULL ? -ENOENT : 0);
1025 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1026 struct lu_device_type *t,
1027 struct lustre_cfg *lcfg)
1029 struct lu_device *l;
1030 struct mdd_device *m;
1034 l = ERR_PTR(-ENOMEM);
1036 md_device_init(&m->mdd_md_dev, t);
1038 l->ld_ops = &mdd_lu_ops;
1039 m->mdd_md_dev.md_ops = &mdd_ops;
1040 md_upcall_init(&m->mdd_md_dev, NULL);
1046 static struct lu_device *mdd_device_free(const struct lu_env *env,
1047 struct lu_device *lu)
1049 struct mdd_device *m = lu2mdd_dev(lu);
1050 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1053 LASSERT(atomic_read(&lu->ld_ref) == 0);
1054 md_device_fini(&m->mdd_md_dev);
1059 static struct obd_ops mdd_obd_device_ops = {
1060 .o_owner = THIS_MODULE
1063 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1064 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1066 static struct lu_context_key mdd_ucred_key = {
1067 .lct_tags = LCT_SESSION,
1068 .lct_init = mdd_ucred_key_init,
1069 .lct_fini = mdd_ucred_key_fini
1072 struct md_ucred *md_ucred(const struct lu_env *env)
1074 LASSERT(env->le_ses != NULL);
1075 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1077 EXPORT_SYMBOL(md_ucred);
1080 * context key constructor/destructor:
1081 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1083 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1085 struct lu_context_key mdd_capainfo_key = {
1086 .lct_tags = LCT_SESSION,
1087 .lct_init = mdd_capainfo_key_init,
1088 .lct_fini = mdd_capainfo_key_fini
1091 struct md_capainfo *md_capainfo(const struct lu_env *env)
1093 /* NB, in mdt_init0 */
1094 if (env->le_ses == NULL)
1096 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1098 EXPORT_SYMBOL(md_capainfo);
1100 static int mdd_changelog_user_register(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),LLOG_CHANGELOG_USER_ORIG_CTXT);
1113 llog_ctxt_put(ctxt);
1117 /* Assume we want it on since somebody registered */
1118 rc = mdd_changelog_on(mdd, 1);
1122 rec->cur_hdr.lrh_len = sizeof(*rec);
1123 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1124 spin_lock(&mdd->mdd_cl.mc_user_lock);
1125 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1126 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1127 CERROR("Maximum number of changelog users exceeded!\n");
1128 GOTO(out, rc = -EOVERFLOW);
1130 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1131 rec->cur_endrec = mdd->mdd_cl.mc_index;
1132 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1134 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1136 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1139 llog_ctxt_put(ctxt);
1143 struct mdd_changelog_user_data {
1144 __u64 mcud_endrec; /**< purge record for this user */
1145 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1147 __u32 mcud_minid; /**< user id with lowest rec reference */
1148 __u32 mcud_usercount;
1151 #define MCUD_UNREGISTER -1LL
1154 * 1. Find the smallest record everyone is willing to purge
1155 * 2. Update the last purgeable record for this user
1157 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1158 struct llog_rec_hdr *hdr, void *data)
1160 struct llog_changelog_user_rec *rec;
1161 struct mdd_changelog_user_data *mcud =
1162 (struct mdd_changelog_user_data *)data;
1166 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1168 rec = (struct llog_changelog_user_rec *)hdr;
1170 mcud->mcud_usercount++;
1172 /* If we have a new endrec for this id, use it for the following
1173 min check instead of its old value */
1174 if (rec->cur_id == mcud->mcud_id)
1175 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1177 /* Track the minimum referenced record */
1178 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1179 mcud->mcud_minid = rec->cur_id;
1180 mcud->mcud_minrec = rec->cur_endrec;
1183 if (rec->cur_id != mcud->mcud_id)
1186 /* Update this user's record */
1187 mcud->mcud_found = 1;
1189 /* Special case: unregister this user */
1190 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1191 struct llog_cookie cookie;
1192 cookie.lgc_lgl = llh->lgh_id;
1193 cookie.lgc_index = hdr->lrh_index;
1194 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1197 mcud->mcud_usercount--;
1201 /* Update the endrec */
1202 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1203 mcud->mcud_id, rec->cur_endrec);
1205 /* hdr+1 is loc of data */
1206 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1207 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1213 static int mdd_changelog_user_purge(struct mdd_device *mdd, int id,
1216 struct mdd_changelog_user_data data;
1217 struct llog_ctxt *ctxt;
1221 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec);
1224 data.mcud_minid = 0;
1225 data.mcud_minrec = 0;
1226 data.mcud_usercount = 0;
1227 data.mcud_endrec = endrec;
1228 spin_lock(&mdd->mdd_cl.mc_lock);
1229 endrec = mdd->mdd_cl.mc_index;
1230 spin_unlock(&mdd->mdd_cl.mc_lock);
1231 if ((data.mcud_endrec == 0) ||
1232 ((data.mcud_endrec > endrec) &&
1233 (data.mcud_endrec != MCUD_UNREGISTER)))
1234 data.mcud_endrec = endrec;
1236 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1239 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1241 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1242 (void *)&data, 0, 0);
1243 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1244 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1245 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1246 data.mcud_minrec, data.mcud_minid);
1247 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1249 CWARN("Could not determine changelog records to purge; rc=%d\n",
1253 llog_ctxt_put(ctxt);
1255 if (!data.mcud_found) {
1256 CWARN("No entry for user %d. Last changelog reference is "
1257 LPD64" by changelog user %d\n", data.mcud_id,
1258 data.mcud_minrec, data.mcud_minid);
1262 if (!rc && data.mcud_usercount == 0)
1263 /* No more users; turn changelogs off */
1264 rc = mdd_changelog_on(mdd, 0);
1270 * May be called remotely from mdt_iocontrol_handle or locally from
1271 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce or
1272 * swab an obd_ioctl_data format (but local ioctl handler does).
1274 * \param len - data len
1275 * \param karg - ioctl data, in kernel space
1277 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1278 unsigned int cmd, int len, void *karg)
1280 struct mdd_device *mdd;
1281 struct obd_ioctl_data *data = karg;
1285 mdd = lu2mdd_dev(&m->md_lu_dev);
1287 /* Doesn't use obd_ioctl_data */
1288 if (cmd == OBD_IOC_CHANGELOG_CLEAR) {
1289 struct changelog_setinfo *cs = karg;
1290 if (len != sizeof(*cs)) {
1291 CERROR("Bad changelog_clear ioctl size %d\n", len);
1294 rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno);
1298 /* Below ioctls use obd_ioctl_data */
1299 if (len != sizeof(*data)) {
1300 CERROR("Bad ioctl size %d\n", len);
1303 if (data->ioc_version != OBD_IOCTL_VERSION) {
1304 CERROR("Bad magic %x != %x\n", data->ioc_version,
1310 case OBD_IOC_CHANGELOG_REG:
1311 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1313 case OBD_IOC_CHANGELOG_DEREG:
1314 rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1,
1324 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1325 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
1327 const struct md_device_operations mdd_ops = {
1328 .mdo_statfs = mdd_statfs,
1329 .mdo_root_get = mdd_root_get,
1330 .mdo_maxsize_get = mdd_maxsize_get,
1331 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1332 .mdo_update_capa_key= mdd_update_capa_key,
1333 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1334 .mdo_iocontrol = mdd_iocontrol,
1335 #ifdef HAVE_QUOTA_SUPPORT
1337 .mqo_notify = mdd_quota_notify,
1338 .mqo_setup = mdd_quota_setup,
1339 .mqo_cleanup = mdd_quota_cleanup,
1340 .mqo_recovery = mdd_quota_recovery,
1341 .mqo_check = mdd_quota_check,
1342 .mqo_on = mdd_quota_on,
1343 .mqo_off = mdd_quota_off,
1344 .mqo_setinfo = mdd_quota_setinfo,
1345 .mqo_getinfo = mdd_quota_getinfo,
1346 .mqo_setquota = mdd_quota_setquota,
1347 .mqo_getquota = mdd_quota_getquota,
1348 .mqo_getoinfo = mdd_quota_getoinfo,
1349 .mqo_getoquota = mdd_quota_getoquota,
1350 .mqo_invalidate = mdd_quota_invalidate,
1351 .mqo_finvalidate = mdd_quota_finvalidate
1356 static struct lu_device_type_operations mdd_device_type_ops = {
1357 .ldto_init = mdd_type_init,
1358 .ldto_fini = mdd_type_fini,
1360 .ldto_start = mdd_type_start,
1361 .ldto_stop = mdd_type_stop,
1363 .ldto_device_alloc = mdd_device_alloc,
1364 .ldto_device_free = mdd_device_free,
1366 .ldto_device_init = mdd_device_init,
1367 .ldto_device_fini = mdd_device_fini
1370 static struct lu_device_type mdd_device_type = {
1371 .ldt_tags = LU_DEVICE_MD,
1372 .ldt_name = LUSTRE_MDD_NAME,
1373 .ldt_ops = &mdd_device_type_ops,
1374 .ldt_ctx_tags = LCT_MD_THREAD
1377 /* context key constructor: mdd_key_init */
1378 LU_KEY_INIT(mdd, struct mdd_thread_info);
1380 static void mdd_key_fini(const struct lu_context *ctx,
1381 struct lu_context_key *key, void *data)
1383 struct mdd_thread_info *info = data;
1384 if (info->mti_max_lmm != NULL)
1385 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1386 if (info->mti_max_cookie != NULL)
1387 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1388 mdd_buf_put(&info->mti_big_buf);
1393 /* context key: mdd_thread_key */
1394 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1396 static struct lu_local_obj_desc llod_capa_key = {
1397 .llod_name = CAPA_KEYS,
1398 .llod_oid = MDD_CAPA_KEYS_OID,
1402 static struct lu_local_obj_desc llod_mdd_orphan = {
1403 .llod_name = orph_index_name,
1404 .llod_oid = MDD_ORPHAN_OID,
1406 .llod_feat = &dt_directory_features,
1409 static struct lu_local_obj_desc llod_mdd_root = {
1410 .llod_name = mdd_root_dir_name,
1411 .llod_oid = MDD_ROOT_INDEX_OID,
1413 .llod_feat = &dt_directory_features,
1416 static int __init mdd_mod_init(void)
1418 struct lprocfs_static_vars lvars;
1419 lprocfs_mdd_init_vars(&lvars);
1421 llo_local_obj_register(&llod_capa_key);
1422 llo_local_obj_register(&llod_mdd_orphan);
1423 llo_local_obj_register(&llod_mdd_root);
1425 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1426 LUSTRE_MDD_NAME, &mdd_device_type);
1429 static void __exit mdd_mod_exit(void)
1431 llo_local_obj_unregister(&llod_capa_key);
1432 llo_local_obj_unregister(&llod_mdd_orphan);
1433 llo_local_obj_unregister(&llod_mdd_root);
1435 class_unregister_type(LUSTRE_MDD_NAME);
1438 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1439 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1440 MODULE_LICENSE("GPL");
1442 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);