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 mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD;
88 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
89 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
91 rc = mdd_procfs_init(mdd, name);
95 static struct lu_device *mdd_device_fini(const struct lu_env *env,
98 struct mdd_device *mdd = lu2mdd_dev(d);
99 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
102 rc = mdd_procfs_fini(mdd);
104 CERROR("proc fini error %d \n", rc);
110 static void mdd_changelog_fini(const struct lu_env *env,
111 struct mdd_device *mdd);
113 static void mdd_device_shutdown(const struct lu_env *env,
114 struct mdd_device *m, struct lustre_cfg *cfg)
117 mdd_changelog_fini(env, m);
118 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
119 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
120 mdd_object_put(env, m->mdd_dot_lustre);
122 mdd_fini_obd(env, m, cfg);
123 orph_index_fini(env, m);
124 /* remove upcall device*/
125 md_upcall_fini(&m->mdd_md_dev);
129 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
132 struct mdd_device *mdd = (struct mdd_device *)data;
133 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
136 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
137 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
140 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
141 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
142 rec->cr_type, rec->cr_namelen, rec->cr_name,
143 llh->lgh_id.lgl_oid);
145 mdd->mdd_cl.mc_index = rec->cr_index;
146 RETURN(LLOG_PROC_BREAK);
149 static int changelog_user_init_cb(struct llog_handle *llh,
150 struct llog_rec_hdr *hdr, void *data)
152 struct mdd_device *mdd = (struct mdd_device *)data;
153 struct llog_changelog_user_rec *rec =
154 (struct llog_changelog_user_rec *)hdr;
157 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
158 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
160 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
161 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
162 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
164 spin_lock(&mdd->mdd_cl.mc_user_lock);
165 mdd->mdd_cl.mc_lastuser = rec->cur_id;
166 spin_unlock(&mdd->mdd_cl.mc_user_lock);
168 RETURN(LLOG_PROC_BREAK);
172 static int mdd_changelog_llog_init(struct mdd_device *mdd)
174 struct obd_device *obd = mdd2obd_dev(mdd);
175 struct llog_ctxt *ctxt;
178 /* Find last changelog entry number */
179 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
181 CERROR("no changelog context\n");
184 if (!ctxt->loc_handle) {
189 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
193 CERROR("changelog init failed: %d\n", rc);
196 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
197 mdd->mdd_cl.mc_index);
199 /* Find last changelog user id */
200 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
202 CERROR("no changelog user context\n");
205 if (!ctxt->loc_handle) {
210 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
215 CERROR("changelog user init failed: %d\n", rc);
219 /* If we have registered users, assume we want changelogs on */
220 if (mdd->mdd_cl.mc_lastuser > 0)
221 rc = mdd_changelog_on(mdd, 1);
226 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
230 mdd->mdd_cl.mc_index = 0;
231 spin_lock_init(&mdd->mdd_cl.mc_lock);
232 cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
233 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
234 mdd->mdd_cl.mc_flags = 0; /* off by default */
235 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
236 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
237 mdd->mdd_cl.mc_lastuser = 0;
239 rc = mdd_changelog_llog_init(mdd);
241 CERROR("Changelog setup during init failed %d\n", rc);
242 mdd->mdd_cl.mc_flags |= CLM_ERR;
248 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
250 mdd->mdd_cl.mc_flags = 0;
253 /* Start / stop recording */
254 int mdd_changelog_on(struct mdd_device *mdd, int on)
258 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
259 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
260 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
261 CERROR("Changelogs cannot be enabled due to error "
262 "condition (see %s log).\n",
263 mdd2obd_dev(mdd)->obd_name);
266 spin_lock(&mdd->mdd_cl.mc_lock);
267 mdd->mdd_cl.mc_flags |= CLM_ON;
268 spin_unlock(&mdd->mdd_cl.mc_lock);
269 rc = mdd_changelog_write_header(mdd, CLM_START);
271 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
272 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
273 rc = mdd_changelog_write_header(mdd, CLM_FINI);
274 spin_lock(&mdd->mdd_cl.mc_lock);
275 mdd->mdd_cl.mc_flags &= ~CLM_ON;
276 spin_unlock(&mdd->mdd_cl.mc_lock);
281 /** Add a changelog entry \a rec to the changelog llog
284 * \param handle - currently ignored since llogs start their own transaction;
285 * this will hopefully be fixed in llog rewrite
288 int mdd_changelog_llog_write(struct mdd_device *mdd,
289 struct llog_changelog_rec *rec,
290 struct thandle *handle)
292 struct obd_device *obd = mdd2obd_dev(mdd);
293 struct llog_ctxt *ctxt;
296 if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
299 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
300 /* llog_lvfs_write_rec sets the llog tail len */
301 rec->cr_hdr.lrh_type = CHANGELOG_REC;
302 rec->cr_time = cfs_time_current_64();
303 spin_lock(&mdd->mdd_cl.mc_lock);
304 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
305 but as long as the MDD transactions are ordered correctly for e.g.
306 rename conflicts, I don't think this should matter. */
307 rec->cr_index = ++mdd->mdd_cl.mc_index;
308 spin_unlock(&mdd->mdd_cl.mc_lock);
309 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
313 /* nested journal transaction */
314 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
317 cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
322 /** Remove entries with indicies up to and including \a endrec from the
328 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
330 struct obd_device *obd = mdd2obd_dev(mdd);
331 struct llog_ctxt *ctxt;
332 long long unsigned cur;
335 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
339 spin_lock(&mdd->mdd_cl.mc_lock);
340 cur = (long long)mdd->mdd_cl.mc_index;
341 spin_unlock(&mdd->mdd_cl.mc_lock);
345 /* purge to "0" is shorthand for everything */
349 /* If purging all records, write a header entry so we don't have an
350 empty catalog and we're sure to have a valid starting index next
351 time. In case of crash, we just restart with old log so we're
354 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
359 /* Some records were purged, so reset repeat-access time (so we
360 record new mtime update records, so users can see a file has been
361 changed since the last purge) */
362 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
364 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
370 /** Add a CL_MARK record to the changelog
372 * \param markerflags - CLM_*
375 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
377 struct obd_device *obd = mdd2obd_dev(mdd);
378 struct llog_changelog_rec *rec;
380 int len = strlen(obd->obd_name);
384 reclen = llog_data_len(sizeof(*rec) + len);
385 OBD_ALLOC(rec, reclen);
389 rec->cr_flags = CLF_VERSION;
390 rec->cr_type = CL_MARK;
391 rec->cr_namelen = len;
392 memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
393 /* Status and action flags */
394 rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
396 rc = mdd_changelog_llog_write(mdd, rec, NULL);
398 /* assume on or off event; reset repeat-access time */
399 mdd->mdd_cl.mc_starttime = rec->cr_time;
401 OBD_FREE(rec, reclen);
406 * Create ".lustre" directory.
408 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
410 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
411 struct md_object *mdo;
414 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
415 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
416 mdd_root_dir_name, mdd_dot_lustre_name,
417 fid, &dt_directory_features);
418 /* .lustre dir may be already present */
419 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
421 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
422 mdd_dot_lustre_name, PFID(fid), rc);
427 lu_object_put(env, &mdo->mo_lu);
432 static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
435 struct mdd_object *mdd_obj = md2mdd_obj(obj);
437 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
440 static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
441 const struct md_attr *ma)
446 static int dot_lustre_xattr_get(const struct lu_env *env,
447 struct md_object *obj, struct lu_buf *buf,
453 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
456 struct mdd_object *mdd_obj = md2mdd_obj(obj);
458 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
459 mdd_obj->mod_count++;
460 mdd_write_unlock(env, mdd_obj);
465 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
466 char *path, int pathlen, __u64 *recno, int *linkno)
471 static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
474 struct mdd_object *mdd_obj = md2mdd_obj(obj);
476 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
477 mdd_obj->mod_count--;
478 mdd_write_unlock(env, mdd_obj);
483 static dt_obj_version_t dot_lustre_version_get(const struct lu_env *env,
484 struct md_object *obj)
489 static void dot_lustre_version_set(const struct lu_env *env,
490 struct md_object *obj,
491 dt_obj_version_t version)
497 static struct md_object_operations mdd_dot_lustre_obj_ops = {
498 .moo_attr_get = dot_lustre_attr_get,
499 .moo_attr_set = dot_lustre_attr_set,
500 .moo_xattr_get = dot_lustre_xattr_get,
501 .moo_open = dot_lustre_mdd_open,
502 .moo_close = dot_lustre_close,
503 .moo_readpage = mdd_readpage,
504 .moo_version_get = dot_lustre_version_get,
505 .moo_version_set = dot_lustre_version_set,
506 .moo_path = dot_lustre_path
509 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
510 const struct lu_name *lname, struct lu_fid *f,
511 struct md_op_spec *spec)
513 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
521 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
522 const struct lu_name *lname,
523 struct md_object *child, struct md_op_spec *spec,
529 static int dot_lustre_rename(const struct lu_env *env,
530 struct md_object *src_pobj,
531 struct md_object *tgt_pobj,
532 const struct lu_fid *lf,
533 const struct lu_name *lsname,
534 struct md_object *tobj,
535 const struct lu_name *ltname, struct md_attr *ma)
540 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
541 struct md_object *src_obj,
542 const struct lu_name *lname, struct md_attr *ma)
547 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
548 struct md_object *cobj, const struct lu_name *lname,
554 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
555 .mdo_lookup = dot_lustre_lookup,
556 .mdo_create = dot_lustre_create,
557 .mdo_rename = dot_lustre_rename,
558 .mdo_link = dot_lustre_link,
559 .mdo_unlink = dot_lustre_unlink,
562 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
567 if (ma->ma_need & MA_INODE) {
568 struct mdd_device *mdd = mdo2mdd(obj);
570 /* "fid" is a virtual object and hence does not have any "real"
571 * attributes. So we reuse attributes of .lustre for "fid" dir */
572 ma->ma_need |= MA_INODE;
573 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
576 ma->ma_valid |= MA_INODE;
579 /* "fid" directory does not have any striping information. */
580 if (ma->ma_need & MA_LOV) {
581 struct mdd_object *mdd_obj = md2mdd_obj(obj);
583 if (ma->ma_valid & MA_LOV)
586 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
587 S_ISDIR(mdd_object_type(mdd_obj))))
590 if (ma->ma_need & MA_LOV_DEF) {
591 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
594 ma->ma_valid |= MA_LOV;
603 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
604 const struct md_attr *ma)
609 static int obf_xattr_get(const struct lu_env *env,
610 struct md_object *obj, struct lu_buf *buf,
616 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
619 struct mdd_object *mdd_obj = md2mdd_obj(obj);
621 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
622 mdd_obj->mod_count++;
623 mdd_write_unlock(env, mdd_obj);
628 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
631 struct mdd_object *mdd_obj = md2mdd_obj(obj);
633 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
634 mdd_obj->mod_count--;
635 mdd_write_unlock(env, mdd_obj);
640 /** Nothing to list in "fid" directory */
641 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
642 const struct lu_rdpg *rdpg)
647 static int obf_path(const struct lu_env *env, struct md_object *obj,
648 char *path, int pathlen, __u64 *recno, int *linkno)
653 static struct md_object_operations mdd_obf_obj_ops = {
654 .moo_attr_get = obf_attr_get,
655 .moo_attr_set = obf_attr_set,
656 .moo_xattr_get = obf_xattr_get,
657 .moo_open = obf_mdd_open,
658 .moo_close = obf_mdd_close,
659 .moo_readpage = obf_mdd_readpage,
664 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
665 * are valid. We also check if object with passed fid exists or not.
667 static int obf_lookup(const struct lu_env *env, struct md_object *p,
668 const struct lu_name *lname, struct lu_fid *f,
669 struct md_op_spec *spec)
671 char *name = (char *)lname->ln_name;
672 struct mdd_device *mdd = mdo2mdd(p);
673 struct mdd_object *child;
679 sscanf(name, SFID, &(f->f_seq), &(f->f_oid),
681 if (!fid_is_sane(f)) {
682 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
684 GOTO(out, rc = -EINVAL);
687 /* Check if object with this fid exists */
688 child = mdd_object_find(env, mdd, f);
692 GOTO(out, rc = PTR_ERR(child));
694 if (mdd_object_exists(child) == 0)
697 mdd_object_put(env, child);
703 static int obf_create(const struct lu_env *env, struct md_object *pobj,
704 const struct lu_name *lname, struct md_object *child,
705 struct md_op_spec *spec, struct md_attr* ma)
710 static int obf_rename(const struct lu_env *env,
711 struct md_object *src_pobj, struct md_object *tgt_pobj,
712 const struct lu_fid *lf, const struct lu_name *lsname,
713 struct md_object *tobj, const struct lu_name *ltname,
719 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
720 struct md_object *src_obj, const struct lu_name *lname,
726 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
727 struct md_object *cobj, const struct lu_name *lname,
733 static struct md_dir_operations mdd_obf_dir_ops = {
734 .mdo_lookup = obf_lookup,
735 .mdo_create = obf_create,
736 .mdo_rename = obf_rename,
737 .mdo_link = obf_link,
738 .mdo_unlink = obf_unlink
742 * Create special in-memory "fid" object for open-by-fid.
744 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
746 struct mdd_object *mdd_obf;
747 struct lu_object *obf_lu_obj;
750 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
752 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
753 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
754 GOTO(out, rc = -ENOENT);
756 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
757 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
758 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
759 /* Don't allow objects to be created in "fid" dir */
760 mdd_obf->mod_flags |= IMMUTE_OBJ;
762 obf_lu_obj = mdd2lu_obj(mdd_obf);
763 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
769 /** Setup ".lustre" directory object */
770 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
772 struct dt_object *dt_dot_lustre;
773 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
776 rc = create_dot_lustre_dir(env, m);
780 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
781 mdd_dot_lustre_name, fid);
782 if (IS_ERR(dt_dot_lustre)) {
783 rc = PTR_ERR(dt_dot_lustre);
787 /* references are released in mdd_device_shutdown() */
788 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
791 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
792 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
794 rc = mdd_obf_setup(env, m);
796 CERROR("Error initializing \"fid\" object - %d.\n", rc);
802 static int mdd_process_config(const struct lu_env *env,
803 struct lu_device *d, struct lustre_cfg *cfg)
805 struct mdd_device *m = lu2mdd_dev(d);
806 struct dt_device *dt = m->mdd_child;
807 struct lu_device *next = &dt->dd_lu_dev;
811 switch (cfg->lcfg_command) {
813 struct lprocfs_static_vars lvars;
815 lprocfs_mdd_init_vars(&lvars);
816 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
817 if (rc > 0 || rc == -ENOSYS)
818 /* we don't understand; pass it on */
819 rc = next->ld_ops->ldo_process_config(env, next, cfg);
823 rc = next->ld_ops->ldo_process_config(env, next, cfg);
826 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
828 rc = mdd_init_obd(env, m, cfg);
830 CERROR("lov init error %d \n", rc);
833 rc = mdd_txn_init_credits(env, m);
837 mdd_changelog_init(env, m);
840 mdd_device_shutdown(env, m, cfg);
842 rc = next->ld_ops->ldo_process_config(env, next, cfg);
850 static int mdd_lov_set_nextid(const struct lu_env *env,
851 struct mdd_device *mdd)
853 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
857 LASSERT(mds->mds_lov_objids != NULL);
858 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
859 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
860 mds->mds_lov_objids, NULL);
865 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
866 struct mdd_device *mdd)
868 /* XXX: to be implemented! */
873 static int mdd_recovery_complete(const struct lu_env *env,
876 struct mdd_device *mdd = lu2mdd_dev(d);
877 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
878 struct obd_device *obd = mdd2obd_dev(mdd);
882 LASSERT(mdd != NULL);
883 LASSERT(obd != NULL);
885 /* XXX: Do we need this in new stack? */
886 rc = mdd_lov_set_nextid(env, mdd);
888 CERROR("mdd_lov_set_nextid() failed %d\n",
893 /* XXX: cleanup unlink. */
894 rc = mdd_cleanup_unlink_llog(env, mdd);
896 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
901 /* Call that with obd_recovering = 1 just to update objids */
902 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
903 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
905 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
906 obd->obd_recovering = 0;
907 obd->obd_type->typ_dt_ops->o_postrecov(obd);
909 /* XXX: orphans handling. */
910 __mdd_orphan_cleanup(env, mdd);
911 rc = next->ld_ops->ldo_recovery_complete(env, next);
916 static int mdd_prepare(const struct lu_env *env,
917 struct lu_device *pdev,
918 struct lu_device *cdev)
920 struct mdd_device *mdd = lu2mdd_dev(cdev);
921 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
922 struct dt_object *root;
926 rc = next->ld_ops->ldo_prepare(env, cdev, next);
930 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
931 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
934 LASSERT(root != NULL);
935 lu_object_put(env, &root->do_lu);
936 rc = orph_index_init(env, mdd);
943 rc = mdd_dot_lustre_setup(env, mdd);
945 CERROR("Error(%d) initializing .lustre objects\n", rc);
953 const struct lu_device_operations mdd_lu_ops = {
954 .ldo_object_alloc = mdd_object_alloc,
955 .ldo_process_config = mdd_process_config,
956 .ldo_recovery_complete = mdd_recovery_complete,
957 .ldo_prepare = mdd_prepare,
961 * No permission check is needed.
963 static int mdd_root_get(const struct lu_env *env,
964 struct md_device *m, struct lu_fid *f)
966 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
969 *f = mdd->mdd_root_fid;
974 * No permission check is needed.
976 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
979 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
984 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
990 * No permission check is needed.
992 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
993 int *md_size, int *cookie_size)
995 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
998 *md_size = mdd_lov_mdsize(env, mdd);
999 *cookie_size = mdd_lov_cookiesize(env, mdd);
1004 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1005 int mode, unsigned long timeout, __u32 alg,
1006 struct lustre_capa_key *keys)
1008 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1009 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1013 mds->mds_capa_keys = keys;
1014 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1015 timeout, alg, keys);
1019 static int mdd_update_capa_key(const struct lu_env *env,
1020 struct md_device *m,
1021 struct lustre_capa_key *key)
1023 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1024 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1025 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
1029 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1030 sizeof(info), &info, NULL);
1034 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1037 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1039 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1040 return (*h == NULL ? -ENOENT : 0);
1043 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1044 struct lu_device_type *t,
1045 struct lustre_cfg *lcfg)
1047 struct lu_device *l;
1048 struct mdd_device *m;
1052 l = ERR_PTR(-ENOMEM);
1054 md_device_init(&m->mdd_md_dev, t);
1056 l->ld_ops = &mdd_lu_ops;
1057 m->mdd_md_dev.md_ops = &mdd_ops;
1058 md_upcall_init(&m->mdd_md_dev, NULL);
1064 static struct lu_device *mdd_device_free(const struct lu_env *env,
1065 struct lu_device *lu)
1067 struct mdd_device *m = lu2mdd_dev(lu);
1068 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1071 LASSERT(atomic_read(&lu->ld_ref) == 0);
1072 md_device_fini(&m->mdd_md_dev);
1077 static struct obd_ops mdd_obd_device_ops = {
1078 .o_owner = THIS_MODULE
1081 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1082 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1084 static struct lu_context_key mdd_ucred_key = {
1085 .lct_tags = LCT_SESSION,
1086 .lct_init = mdd_ucred_key_init,
1087 .lct_fini = mdd_ucred_key_fini
1090 struct md_ucred *md_ucred(const struct lu_env *env)
1092 LASSERT(env->le_ses != NULL);
1093 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1095 EXPORT_SYMBOL(md_ucred);
1098 * context key constructor/destructor:
1099 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1101 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1103 struct lu_context_key mdd_capainfo_key = {
1104 .lct_tags = LCT_SESSION,
1105 .lct_init = mdd_capainfo_key_init,
1106 .lct_fini = mdd_capainfo_key_fini
1109 struct md_capainfo *md_capainfo(const struct lu_env *env)
1111 /* NB, in mdt_init0 */
1112 if (env->le_ses == NULL)
1114 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1116 EXPORT_SYMBOL(md_capainfo);
1118 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1120 struct llog_ctxt *ctxt;
1121 struct llog_changelog_user_rec *rec;
1125 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1131 llog_ctxt_put(ctxt);
1135 /* Assume we want it on since somebody registered */
1136 rc = mdd_changelog_on(mdd, 1);
1140 rec->cur_hdr.lrh_len = sizeof(*rec);
1141 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1142 spin_lock(&mdd->mdd_cl.mc_user_lock);
1143 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1144 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1145 CERROR("Maximum number of changelog users exceeded!\n");
1146 GOTO(out, rc = -EOVERFLOW);
1148 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1149 rec->cur_endrec = mdd->mdd_cl.mc_index;
1150 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1152 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1154 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1157 llog_ctxt_put(ctxt);
1161 struct mdd_changelog_user_data {
1162 __u64 mcud_endrec; /**< purge record for this user */
1163 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1165 __u32 mcud_minid; /**< user id with lowest rec reference */
1166 __u32 mcud_usercount;
1169 #define MCUD_UNREGISTER -1LL
1172 * 1. Find the smallest record everyone is willing to purge
1173 * 2. Update the last purgeable record for this user
1175 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1176 struct llog_rec_hdr *hdr, void *data)
1178 struct llog_changelog_user_rec *rec;
1179 struct mdd_changelog_user_data *mcud =
1180 (struct mdd_changelog_user_data *)data;
1184 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1186 rec = (struct llog_changelog_user_rec *)hdr;
1188 mcud->mcud_usercount++;
1190 /* If we have a new endrec for this id, use it for the following
1191 min check instead of its old value */
1192 if (rec->cur_id == mcud->mcud_id)
1193 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1195 /* Track the minimum referenced record */
1196 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1197 mcud->mcud_minid = rec->cur_id;
1198 mcud->mcud_minrec = rec->cur_endrec;
1201 if (rec->cur_id != mcud->mcud_id)
1204 /* Update this user's record */
1205 mcud->mcud_found = 1;
1207 /* Special case: unregister this user */
1208 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1209 struct llog_cookie cookie;
1210 cookie.lgc_lgl = llh->lgh_id;
1211 cookie.lgc_index = hdr->lrh_index;
1212 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1215 mcud->mcud_usercount--;
1219 /* Update the endrec */
1220 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1221 mcud->mcud_id, rec->cur_endrec);
1223 /* hdr+1 is loc of data */
1224 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1225 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1231 static int mdd_changelog_user_purge(struct mdd_device *mdd, int id,
1234 struct mdd_changelog_user_data data;
1235 struct llog_ctxt *ctxt;
1239 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec);
1242 data.mcud_minid = 0;
1243 data.mcud_minrec = 0;
1244 data.mcud_usercount = 0;
1245 data.mcud_endrec = endrec;
1246 spin_lock(&mdd->mdd_cl.mc_lock);
1247 endrec = mdd->mdd_cl.mc_index;
1248 spin_unlock(&mdd->mdd_cl.mc_lock);
1249 if ((data.mcud_endrec == 0) ||
1250 ((data.mcud_endrec > endrec) &&
1251 (data.mcud_endrec != MCUD_UNREGISTER)))
1252 data.mcud_endrec = endrec;
1254 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1257 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1259 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1260 (void *)&data, 0, 0);
1261 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1262 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1263 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1264 data.mcud_minrec, data.mcud_minid);
1265 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1267 CWARN("Could not determine changelog records to purge; rc=%d\n",
1271 llog_ctxt_put(ctxt);
1273 if (!data.mcud_found) {
1274 CWARN("No entry for user %d. Last changelog reference is "
1275 LPD64" by changelog user %d\n", data.mcud_id,
1276 data.mcud_minrec, data.mcud_minid);
1280 if (!rc && data.mcud_usercount == 0)
1281 /* No more users; turn changelogs off */
1282 rc = mdd_changelog_on(mdd, 0);
1288 * May be called remotely from mdt_iocontrol_handle or locally from
1289 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1290 * an obd_ioctl_data format (but local ioctl handler does).
1292 * \param len - data len
1293 * \param karg - ioctl data, in kernel space
1295 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1296 unsigned int cmd, int len, void *karg)
1298 struct mdd_device *mdd;
1299 struct obd_ioctl_data *data = karg;
1303 mdd = lu2mdd_dev(&m->md_lu_dev);
1305 /* Doesn't use obd_ioctl_data */
1306 if (cmd == OBD_IOC_CHANGELOG_CLEAR) {
1307 struct changelog_setinfo *cs = karg;
1308 rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno);
1312 /* Below ioctls use obd_ioctl_data */
1313 if (len != sizeof(*data)) {
1314 CERROR("Bad ioctl size %d\n", len);
1317 if (data->ioc_version != OBD_IOCTL_VERSION) {
1318 CERROR("Bad magic %x != %x\n", data->ioc_version,
1324 case OBD_IOC_CHANGELOG_REG:
1325 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1327 case OBD_IOC_CHANGELOG_DEREG:
1328 rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1,
1338 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1339 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
1341 const struct md_device_operations mdd_ops = {
1342 .mdo_statfs = mdd_statfs,
1343 .mdo_root_get = mdd_root_get,
1344 .mdo_maxsize_get = mdd_maxsize_get,
1345 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1346 .mdo_update_capa_key= mdd_update_capa_key,
1347 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1348 .mdo_iocontrol = mdd_iocontrol,
1349 #ifdef HAVE_QUOTA_SUPPORT
1351 .mqo_notify = mdd_quota_notify,
1352 .mqo_setup = mdd_quota_setup,
1353 .mqo_cleanup = mdd_quota_cleanup,
1354 .mqo_recovery = mdd_quota_recovery,
1355 .mqo_check = mdd_quota_check,
1356 .mqo_on = mdd_quota_on,
1357 .mqo_off = mdd_quota_off,
1358 .mqo_setinfo = mdd_quota_setinfo,
1359 .mqo_getinfo = mdd_quota_getinfo,
1360 .mqo_setquota = mdd_quota_setquota,
1361 .mqo_getquota = mdd_quota_getquota,
1362 .mqo_getoinfo = mdd_quota_getoinfo,
1363 .mqo_getoquota = mdd_quota_getoquota,
1364 .mqo_invalidate = mdd_quota_invalidate,
1365 .mqo_finvalidate = mdd_quota_finvalidate
1370 static struct lu_device_type_operations mdd_device_type_ops = {
1371 .ldto_init = mdd_type_init,
1372 .ldto_fini = mdd_type_fini,
1374 .ldto_start = mdd_type_start,
1375 .ldto_stop = mdd_type_stop,
1377 .ldto_device_alloc = mdd_device_alloc,
1378 .ldto_device_free = mdd_device_free,
1380 .ldto_device_init = mdd_device_init,
1381 .ldto_device_fini = mdd_device_fini
1384 static struct lu_device_type mdd_device_type = {
1385 .ldt_tags = LU_DEVICE_MD,
1386 .ldt_name = LUSTRE_MDD_NAME,
1387 .ldt_ops = &mdd_device_type_ops,
1388 .ldt_ctx_tags = LCT_MD_THREAD
1391 /* context key constructor: mdd_key_init */
1392 LU_KEY_INIT(mdd, struct mdd_thread_info);
1394 static void mdd_key_fini(const struct lu_context *ctx,
1395 struct lu_context_key *key, void *data)
1397 struct mdd_thread_info *info = data;
1398 if (info->mti_max_lmm != NULL)
1399 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1400 if (info->mti_max_cookie != NULL)
1401 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1402 mdd_buf_put(&info->mti_big_buf);
1407 /* context key: mdd_thread_key */
1408 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1410 static struct lu_local_obj_desc llod_capa_key = {
1411 .llod_name = CAPA_KEYS,
1412 .llod_oid = MDD_CAPA_KEYS_OID,
1416 static struct lu_local_obj_desc llod_mdd_orphan = {
1417 .llod_name = orph_index_name,
1418 .llod_oid = MDD_ORPHAN_OID,
1420 .llod_feat = &dt_directory_features,
1423 static struct lu_local_obj_desc llod_mdd_root = {
1424 .llod_name = mdd_root_dir_name,
1425 .llod_oid = MDD_ROOT_INDEX_OID,
1427 .llod_feat = &dt_directory_features,
1430 static int __init mdd_mod_init(void)
1432 struct lprocfs_static_vars lvars;
1433 lprocfs_mdd_init_vars(&lvars);
1435 llo_local_obj_register(&llod_capa_key);
1436 llo_local_obj_register(&llod_mdd_orphan);
1437 llo_local_obj_register(&llod_mdd_root);
1439 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1440 LUSTRE_MDD_NAME, &mdd_device_type);
1443 static void __exit mdd_mod_exit(void)
1445 llo_local_obj_unregister(&llod_capa_key);
1446 llo_local_obj_unregister(&llod_mdd_orphan);
1447 llo_local_obj_unregister(&llod_mdd_root);
1449 class_unregister_type(LUSTRE_MDD_NAME);
1452 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1453 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1454 MODULE_LICENSE("GPL");
1456 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);