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 if (m->mdd_dot_lustre_objs.mdd_obf)
120 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
121 if (m->mdd_dot_lustre)
122 mdd_object_put(env, m->mdd_dot_lustre);
124 mdd_fini_obd(env, m, cfg);
125 orph_index_fini(env, m);
126 /* remove upcall device*/
127 md_upcall_fini(&m->mdd_md_dev);
131 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
134 struct mdd_device *mdd = (struct mdd_device *)data;
135 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
138 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
139 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
142 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
143 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
144 rec->cr_type, rec->cr_namelen, rec->cr_name,
145 llh->lgh_id.lgl_oid);
147 mdd->mdd_cl.mc_index = rec->cr_index;
148 RETURN(LLOG_PROC_BREAK);
151 static int changelog_user_init_cb(struct llog_handle *llh,
152 struct llog_rec_hdr *hdr, void *data)
154 struct mdd_device *mdd = (struct mdd_device *)data;
155 struct llog_changelog_user_rec *rec =
156 (struct llog_changelog_user_rec *)hdr;
159 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
160 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
162 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
163 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
164 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
166 spin_lock(&mdd->mdd_cl.mc_user_lock);
167 mdd->mdd_cl.mc_lastuser = rec->cur_id;
168 spin_unlock(&mdd->mdd_cl.mc_user_lock);
170 RETURN(LLOG_PROC_BREAK);
174 static int mdd_changelog_llog_init(struct mdd_device *mdd)
176 struct obd_device *obd = mdd2obd_dev(mdd);
177 struct llog_ctxt *ctxt;
180 /* Find last changelog entry number */
181 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
183 CERROR("no changelog context\n");
186 if (!ctxt->loc_handle) {
191 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
195 CERROR("changelog init failed: %d\n", rc);
198 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
199 mdd->mdd_cl.mc_index);
201 /* Find last changelog user id */
202 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
204 CERROR("no changelog user context\n");
207 if (!ctxt->loc_handle) {
212 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
217 CERROR("changelog user init failed: %d\n", rc);
221 /* If we have registered users, assume we want changelogs on */
222 if (mdd->mdd_cl.mc_lastuser > 0)
223 rc = mdd_changelog_on(mdd, 1);
228 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
232 mdd->mdd_cl.mc_index = 0;
233 spin_lock_init(&mdd->mdd_cl.mc_lock);
234 cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
235 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
236 mdd->mdd_cl.mc_flags = 0; /* off by default */
237 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
238 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
239 mdd->mdd_cl.mc_lastuser = 0;
241 rc = mdd_changelog_llog_init(mdd);
243 CERROR("Changelog setup during init failed %d\n", rc);
244 mdd->mdd_cl.mc_flags |= CLM_ERR;
250 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
252 mdd->mdd_cl.mc_flags = 0;
255 /* Start / stop recording */
256 int mdd_changelog_on(struct mdd_device *mdd, int on)
260 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
261 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
262 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
263 CERROR("Changelogs cannot be enabled due to error "
264 "condition (see %s log).\n",
265 mdd2obd_dev(mdd)->obd_name);
268 spin_lock(&mdd->mdd_cl.mc_lock);
269 mdd->mdd_cl.mc_flags |= CLM_ON;
270 spin_unlock(&mdd->mdd_cl.mc_lock);
271 rc = mdd_changelog_write_header(mdd, CLM_START);
273 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
274 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
275 rc = mdd_changelog_write_header(mdd, CLM_FINI);
276 spin_lock(&mdd->mdd_cl.mc_lock);
277 mdd->mdd_cl.mc_flags &= ~CLM_ON;
278 spin_unlock(&mdd->mdd_cl.mc_lock);
283 /** Add a changelog entry \a rec to the changelog llog
286 * \param handle - currently ignored since llogs start their own transaction;
287 * this will hopefully be fixed in llog rewrite
290 int mdd_changelog_llog_write(struct mdd_device *mdd,
291 struct llog_changelog_rec *rec,
292 struct thandle *handle)
294 struct obd_device *obd = mdd2obd_dev(mdd);
295 struct llog_ctxt *ctxt;
298 if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
301 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
302 /* llog_lvfs_write_rec sets the llog tail len */
303 rec->cr_hdr.lrh_type = CHANGELOG_REC;
304 rec->cr_time = cfs_time_current_64();
305 spin_lock(&mdd->mdd_cl.mc_lock);
306 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
307 but as long as the MDD transactions are ordered correctly for e.g.
308 rename conflicts, I don't think this should matter. */
309 rec->cr_index = ++mdd->mdd_cl.mc_index;
310 spin_unlock(&mdd->mdd_cl.mc_lock);
311 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
315 /* nested journal transaction */
316 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
319 cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
324 /** Remove entries with indicies up to and including \a endrec from the
330 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
332 struct obd_device *obd = mdd2obd_dev(mdd);
333 struct llog_ctxt *ctxt;
334 long long unsigned cur;
337 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
341 spin_lock(&mdd->mdd_cl.mc_lock);
342 cur = (long long)mdd->mdd_cl.mc_index;
343 spin_unlock(&mdd->mdd_cl.mc_lock);
347 /* purge to "0" is shorthand for everything */
351 /* If purging all records, write a header entry so we don't have an
352 empty catalog and we're sure to have a valid starting index next
353 time. In case of crash, we just restart with old log so we're
356 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
361 /* Some records were purged, so reset repeat-access time (so we
362 record new mtime update records, so users can see a file has been
363 changed since the last purge) */
364 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
366 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
372 /** Add a CL_MARK record to the changelog
374 * \param markerflags - CLM_*
377 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
379 struct obd_device *obd = mdd2obd_dev(mdd);
380 struct llog_changelog_rec *rec;
382 int len = strlen(obd->obd_name);
386 reclen = llog_data_len(sizeof(*rec) + len);
387 OBD_ALLOC(rec, reclen);
391 rec->cr_flags = CLF_VERSION;
392 rec->cr_type = CL_MARK;
393 rec->cr_namelen = len;
394 memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
395 /* Status and action flags */
396 rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
398 rc = mdd_changelog_llog_write(mdd, rec, NULL);
400 /* assume on or off event; reset repeat-access time */
401 mdd->mdd_cl.mc_starttime = rec->cr_time;
403 OBD_FREE(rec, reclen);
408 * Create ".lustre" directory.
410 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
412 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
413 struct md_object *mdo;
416 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
417 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
418 mdd_root_dir_name, dot_lustre_name,
419 fid, &dt_directory_features);
420 /* .lustre dir may be already present */
421 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
423 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
424 dot_lustre_name, PFID(fid), rc);
429 lu_object_put(env, &mdo->mo_lu);
434 static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
437 struct mdd_object *mdd_obj = md2mdd_obj(obj);
439 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
442 static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
443 const struct md_attr *ma)
448 static int dot_lustre_xattr_get(const struct lu_env *env,
449 struct md_object *obj, struct lu_buf *buf,
455 static int dot_lustre_xattr_list(const struct lu_env *env,
456 struct md_object *obj, struct lu_buf *buf)
461 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
464 struct mdd_object *mdd_obj = md2mdd_obj(obj);
466 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
467 mdd_obj->mod_count++;
468 mdd_write_unlock(env, mdd_obj);
473 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
474 char *path, int pathlen, __u64 *recno, int *linkno)
479 static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
482 struct mdd_object *mdd_obj = md2mdd_obj(obj);
484 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
485 mdd_obj->mod_count--;
486 mdd_write_unlock(env, mdd_obj);
491 static dt_obj_version_t dot_lustre_version_get(const struct lu_env *env,
492 struct md_object *obj)
497 static void dot_lustre_version_set(const struct lu_env *env,
498 struct md_object *obj,
499 dt_obj_version_t version)
505 static struct md_object_operations mdd_dot_lustre_obj_ops = {
506 .moo_attr_get = dot_lustre_attr_get,
507 .moo_attr_set = dot_lustre_attr_set,
508 .moo_xattr_get = dot_lustre_xattr_get,
509 .moo_xattr_list = dot_lustre_xattr_list,
510 .moo_open = dot_lustre_mdd_open,
511 .moo_close = dot_lustre_close,
512 .moo_readpage = mdd_readpage,
513 .moo_version_get = dot_lustre_version_get,
514 .moo_version_set = dot_lustre_version_set,
515 .moo_path = dot_lustre_path
518 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
519 const struct lu_name *lname, struct lu_fid *f,
520 struct md_op_spec *spec)
522 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
530 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
531 const struct lu_name *lname,
532 struct md_object *child, struct md_op_spec *spec,
538 static int dot_lustre_rename(const struct lu_env *env,
539 struct md_object *src_pobj,
540 struct md_object *tgt_pobj,
541 const struct lu_fid *lf,
542 const struct lu_name *lsname,
543 struct md_object *tobj,
544 const struct lu_name *ltname, struct md_attr *ma)
549 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
550 struct md_object *src_obj,
551 const struct lu_name *lname, struct md_attr *ma)
556 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
557 struct md_object *cobj, const struct lu_name *lname,
563 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
564 .mdo_lookup = dot_lustre_lookup,
565 .mdo_create = dot_lustre_create,
566 .mdo_rename = dot_lustre_rename,
567 .mdo_link = dot_lustre_link,
568 .mdo_unlink = dot_lustre_unlink,
571 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
576 if (ma->ma_need & MA_INODE) {
577 struct mdd_device *mdd = mdo2mdd(obj);
579 /* "fid" is a virtual object and hence does not have any "real"
580 * attributes. So we reuse attributes of .lustre for "fid" dir */
581 ma->ma_need |= MA_INODE;
582 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
585 ma->ma_valid |= MA_INODE;
588 /* "fid" directory does not have any striping information. */
589 if (ma->ma_need & MA_LOV) {
590 struct mdd_object *mdd_obj = md2mdd_obj(obj);
592 if (ma->ma_valid & MA_LOV)
595 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
596 S_ISDIR(mdd_object_type(mdd_obj))))
599 if (ma->ma_need & MA_LOV_DEF) {
600 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
603 ma->ma_valid |= MA_LOV;
612 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
613 const struct md_attr *ma)
618 static int obf_xattr_get(const struct lu_env *env,
619 struct md_object *obj, struct lu_buf *buf,
625 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
628 struct mdd_object *mdd_obj = md2mdd_obj(obj);
630 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
631 mdd_obj->mod_count++;
632 mdd_write_unlock(env, mdd_obj);
637 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
640 struct mdd_object *mdd_obj = md2mdd_obj(obj);
642 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
643 mdd_obj->mod_count--;
644 mdd_write_unlock(env, mdd_obj);
649 /** Nothing to list in "fid" directory */
650 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
651 const struct lu_rdpg *rdpg)
656 static int obf_path(const struct lu_env *env, struct md_object *obj,
657 char *path, int pathlen, __u64 *recno, int *linkno)
662 static struct md_object_operations mdd_obf_obj_ops = {
663 .moo_attr_get = obf_attr_get,
664 .moo_attr_set = obf_attr_set,
665 .moo_xattr_get = obf_xattr_get,
666 .moo_open = obf_mdd_open,
667 .moo_close = obf_mdd_close,
668 .moo_readpage = obf_mdd_readpage,
673 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
674 * are valid. We also check if object with passed fid exists or not.
676 static int obf_lookup(const struct lu_env *env, struct md_object *p,
677 const struct lu_name *lname, struct lu_fid *f,
678 struct md_op_spec *spec)
680 char *name = (char *)lname->ln_name;
681 struct mdd_device *mdd = mdo2mdd(p);
682 struct mdd_object *child;
688 sscanf(name, SFID, RFID(f));
689 if (!fid_is_sane(f)) {
690 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
692 GOTO(out, rc = -EINVAL);
695 /* Check if object with this fid exists */
696 child = mdd_object_find(env, mdd, f);
700 GOTO(out, rc = PTR_ERR(child));
702 if (mdd_object_exists(child) == 0)
705 mdd_object_put(env, child);
711 static int obf_create(const struct lu_env *env, struct md_object *pobj,
712 const struct lu_name *lname, struct md_object *child,
713 struct md_op_spec *spec, struct md_attr* ma)
718 static int obf_rename(const struct lu_env *env,
719 struct md_object *src_pobj, struct md_object *tgt_pobj,
720 const struct lu_fid *lf, const struct lu_name *lsname,
721 struct md_object *tobj, const struct lu_name *ltname,
727 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
728 struct md_object *src_obj, const struct lu_name *lname,
734 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
735 struct md_object *cobj, const struct lu_name *lname,
741 static struct md_dir_operations mdd_obf_dir_ops = {
742 .mdo_lookup = obf_lookup,
743 .mdo_create = obf_create,
744 .mdo_rename = obf_rename,
745 .mdo_link = obf_link,
746 .mdo_unlink = obf_unlink
750 * Create special in-memory "fid" object for open-by-fid.
752 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
754 struct mdd_object *mdd_obf;
755 struct lu_object *obf_lu_obj;
758 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
760 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
761 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
762 GOTO(out, rc = -ENOENT);
764 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
765 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
766 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
767 /* Don't allow objects to be created in "fid" dir */
768 mdd_obf->mod_flags |= IMMUTE_OBJ;
770 obf_lu_obj = mdd2lu_obj(mdd_obf);
771 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
777 /** Setup ".lustre" directory object */
778 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
780 struct dt_object *dt_dot_lustre;
781 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
784 rc = create_dot_lustre_dir(env, m);
788 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
789 dot_lustre_name, fid);
790 if (IS_ERR(dt_dot_lustre)) {
791 rc = PTR_ERR(dt_dot_lustre);
795 /* references are released in mdd_device_shutdown() */
796 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
799 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
800 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
802 rc = mdd_obf_setup(env, m);
804 CERROR("Error initializing \"fid\" object - %d.\n", rc);
810 static int mdd_process_config(const struct lu_env *env,
811 struct lu_device *d, struct lustre_cfg *cfg)
813 struct mdd_device *m = lu2mdd_dev(d);
814 struct dt_device *dt = m->mdd_child;
815 struct lu_device *next = &dt->dd_lu_dev;
819 switch (cfg->lcfg_command) {
821 struct lprocfs_static_vars lvars;
823 lprocfs_mdd_init_vars(&lvars);
824 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
825 if (rc > 0 || rc == -ENOSYS)
826 /* we don't understand; pass it on */
827 rc = next->ld_ops->ldo_process_config(env, next, cfg);
831 rc = next->ld_ops->ldo_process_config(env, next, cfg);
834 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
836 rc = mdd_init_obd(env, m, cfg);
838 CERROR("lov init error %d \n", rc);
841 rc = mdd_txn_init_credits(env, m);
845 mdd_changelog_init(env, m);
848 mdd_device_shutdown(env, m, cfg);
850 rc = next->ld_ops->ldo_process_config(env, next, cfg);
858 static int mdd_lov_set_nextid(const struct lu_env *env,
859 struct mdd_device *mdd)
861 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
865 LASSERT(mds->mds_lov_objids != NULL);
866 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
867 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
868 mds->mds_lov_objids, NULL);
873 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
874 struct mdd_device *mdd)
876 /* XXX: to be implemented! */
881 static int mdd_recovery_complete(const struct lu_env *env,
884 struct mdd_device *mdd = lu2mdd_dev(d);
885 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
886 struct obd_device *obd = mdd2obd_dev(mdd);
890 LASSERT(mdd != NULL);
891 LASSERT(obd != NULL);
893 /* XXX: Do we need this in new stack? */
894 rc = mdd_lov_set_nextid(env, mdd);
896 CERROR("mdd_lov_set_nextid() failed %d\n",
901 /* XXX: cleanup unlink. */
902 rc = mdd_cleanup_unlink_llog(env, mdd);
904 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
909 /* Call that with obd_recovering = 1 just to update objids */
910 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
911 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
913 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
914 obd->obd_recovering = 0;
915 obd->obd_type->typ_dt_ops->o_postrecov(obd);
917 /* XXX: orphans handling. */
918 __mdd_orphan_cleanup(env, mdd);
919 rc = next->ld_ops->ldo_recovery_complete(env, next);
924 static int mdd_prepare(const struct lu_env *env,
925 struct lu_device *pdev,
926 struct lu_device *cdev)
928 struct mdd_device *mdd = lu2mdd_dev(cdev);
929 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
930 struct dt_object *root;
934 rc = next->ld_ops->ldo_prepare(env, cdev, next);
938 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
939 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
942 LASSERT(root != NULL);
943 lu_object_put(env, &root->do_lu);
944 rc = orph_index_init(env, mdd);
951 rc = mdd_dot_lustre_setup(env, mdd);
953 CERROR("Error(%d) initializing .lustre objects\n", rc);
961 const struct lu_device_operations mdd_lu_ops = {
962 .ldo_object_alloc = mdd_object_alloc,
963 .ldo_process_config = mdd_process_config,
964 .ldo_recovery_complete = mdd_recovery_complete,
965 .ldo_prepare = mdd_prepare,
969 * No permission check is needed.
971 static int mdd_root_get(const struct lu_env *env,
972 struct md_device *m, struct lu_fid *f)
974 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
977 *f = mdd->mdd_root_fid;
982 * No permission check is needed.
984 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
987 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
992 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
998 * No permission check is needed.
1000 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
1001 int *md_size, int *cookie_size)
1003 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1006 *md_size = mdd_lov_mdsize(env, mdd);
1007 *cookie_size = mdd_lov_cookiesize(env, mdd);
1012 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1013 int mode, unsigned long timeout, __u32 alg,
1014 struct lustre_capa_key *keys)
1016 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1017 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1021 mds->mds_capa_keys = keys;
1022 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1023 timeout, alg, keys);
1027 static int mdd_update_capa_key(const struct lu_env *env,
1028 struct md_device *m,
1029 struct lustre_capa_key *key)
1031 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1032 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1033 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
1037 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1038 sizeof(info), &info, NULL);
1042 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1045 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1047 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1048 return (*h == NULL ? -ENOENT : 0);
1051 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1052 struct lu_device_type *t,
1053 struct lustre_cfg *lcfg)
1055 struct lu_device *l;
1056 struct mdd_device *m;
1060 l = ERR_PTR(-ENOMEM);
1062 md_device_init(&m->mdd_md_dev, t);
1064 l->ld_ops = &mdd_lu_ops;
1065 m->mdd_md_dev.md_ops = &mdd_ops;
1066 md_upcall_init(&m->mdd_md_dev, NULL);
1072 static struct lu_device *mdd_device_free(const struct lu_env *env,
1073 struct lu_device *lu)
1075 struct mdd_device *m = lu2mdd_dev(lu);
1076 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1079 LASSERT(atomic_read(&lu->ld_ref) == 0);
1080 md_device_fini(&m->mdd_md_dev);
1085 static struct obd_ops mdd_obd_device_ops = {
1086 .o_owner = THIS_MODULE
1089 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1090 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1092 static struct lu_context_key mdd_ucred_key = {
1093 .lct_tags = LCT_SESSION,
1094 .lct_init = mdd_ucred_key_init,
1095 .lct_fini = mdd_ucred_key_fini
1098 struct md_ucred *md_ucred(const struct lu_env *env)
1100 LASSERT(env->le_ses != NULL);
1101 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1103 EXPORT_SYMBOL(md_ucred);
1106 * context key constructor/destructor:
1107 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1109 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1111 struct lu_context_key mdd_capainfo_key = {
1112 .lct_tags = LCT_SESSION,
1113 .lct_init = mdd_capainfo_key_init,
1114 .lct_fini = mdd_capainfo_key_fini
1117 struct md_capainfo *md_capainfo(const struct lu_env *env)
1119 /* NB, in mdt_init0 */
1120 if (env->le_ses == NULL)
1122 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1124 EXPORT_SYMBOL(md_capainfo);
1126 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1128 struct llog_ctxt *ctxt;
1129 struct llog_changelog_user_rec *rec;
1133 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1139 llog_ctxt_put(ctxt);
1143 /* Assume we want it on since somebody registered */
1144 rc = mdd_changelog_on(mdd, 1);
1148 rec->cur_hdr.lrh_len = sizeof(*rec);
1149 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1150 spin_lock(&mdd->mdd_cl.mc_user_lock);
1151 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1152 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1153 CERROR("Maximum number of changelog users exceeded!\n");
1154 GOTO(out, rc = -EOVERFLOW);
1156 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1157 rec->cur_endrec = mdd->mdd_cl.mc_index;
1158 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1160 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1162 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1165 llog_ctxt_put(ctxt);
1169 struct mdd_changelog_user_data {
1170 __u64 mcud_endrec; /**< purge record for this user */
1171 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1173 __u32 mcud_minid; /**< user id with lowest rec reference */
1174 __u32 mcud_usercount;
1177 #define MCUD_UNREGISTER -1LL
1180 * 1. Find the smallest record everyone is willing to purge
1181 * 2. Update the last purgeable record for this user
1183 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1184 struct llog_rec_hdr *hdr, void *data)
1186 struct llog_changelog_user_rec *rec;
1187 struct mdd_changelog_user_data *mcud =
1188 (struct mdd_changelog_user_data *)data;
1192 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1194 rec = (struct llog_changelog_user_rec *)hdr;
1196 mcud->mcud_usercount++;
1198 /* If we have a new endrec for this id, use it for the following
1199 min check instead of its old value */
1200 if (rec->cur_id == mcud->mcud_id)
1201 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1203 /* Track the minimum referenced record */
1204 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1205 mcud->mcud_minid = rec->cur_id;
1206 mcud->mcud_minrec = rec->cur_endrec;
1209 if (rec->cur_id != mcud->mcud_id)
1212 /* Update this user's record */
1213 mcud->mcud_found = 1;
1215 /* Special case: unregister this user */
1216 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1217 struct llog_cookie cookie;
1218 cookie.lgc_lgl = llh->lgh_id;
1219 cookie.lgc_index = hdr->lrh_index;
1220 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1223 mcud->mcud_usercount--;
1227 /* Update the endrec */
1228 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1229 mcud->mcud_id, rec->cur_endrec);
1231 /* hdr+1 is loc of data */
1232 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1233 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1239 static int mdd_changelog_user_purge(struct mdd_device *mdd, int id,
1242 struct mdd_changelog_user_data data;
1243 struct llog_ctxt *ctxt;
1247 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec);
1250 data.mcud_minid = 0;
1251 data.mcud_minrec = 0;
1252 data.mcud_usercount = 0;
1253 data.mcud_endrec = endrec;
1254 spin_lock(&mdd->mdd_cl.mc_lock);
1255 endrec = mdd->mdd_cl.mc_index;
1256 spin_unlock(&mdd->mdd_cl.mc_lock);
1257 if ((data.mcud_endrec == 0) ||
1258 ((data.mcud_endrec > endrec) &&
1259 (data.mcud_endrec != MCUD_UNREGISTER)))
1260 data.mcud_endrec = endrec;
1262 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1265 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1267 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1268 (void *)&data, 0, 0);
1269 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1270 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1271 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1272 data.mcud_minrec, data.mcud_minid);
1273 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1275 CWARN("Could not determine changelog records to purge; rc=%d\n",
1279 llog_ctxt_put(ctxt);
1281 if (!data.mcud_found) {
1282 CWARN("No entry for user %d. Last changelog reference is "
1283 LPD64" by changelog user %d\n", data.mcud_id,
1284 data.mcud_minrec, data.mcud_minid);
1288 if (!rc && data.mcud_usercount == 0)
1289 /* No more users; turn changelogs off */
1290 rc = mdd_changelog_on(mdd, 0);
1296 * May be called remotely from mdt_iocontrol_handle or locally from
1297 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1298 * an obd_ioctl_data format (but local ioctl handler does).
1300 * \param len - data len
1301 * \param karg - ioctl data, in kernel space
1303 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1304 unsigned int cmd, int len, void *karg)
1306 struct mdd_device *mdd;
1307 struct obd_ioctl_data *data = karg;
1311 mdd = lu2mdd_dev(&m->md_lu_dev);
1313 /* Doesn't use obd_ioctl_data */
1314 if (cmd == OBD_IOC_CHANGELOG_CLEAR) {
1315 struct changelog_setinfo *cs = karg;
1316 rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno);
1320 /* Below ioctls use obd_ioctl_data */
1321 if (len != sizeof(*data)) {
1322 CERROR("Bad ioctl size %d\n", len);
1325 if (data->ioc_version != OBD_IOCTL_VERSION) {
1326 CERROR("Bad magic %x != %x\n", data->ioc_version,
1332 case OBD_IOC_CHANGELOG_REG:
1333 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1335 case OBD_IOC_CHANGELOG_DEREG:
1336 rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1,
1346 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1347 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
1349 const struct md_device_operations mdd_ops = {
1350 .mdo_statfs = mdd_statfs,
1351 .mdo_root_get = mdd_root_get,
1352 .mdo_maxsize_get = mdd_maxsize_get,
1353 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1354 .mdo_update_capa_key= mdd_update_capa_key,
1355 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1356 .mdo_iocontrol = mdd_iocontrol,
1357 #ifdef HAVE_QUOTA_SUPPORT
1359 .mqo_notify = mdd_quota_notify,
1360 .mqo_setup = mdd_quota_setup,
1361 .mqo_cleanup = mdd_quota_cleanup,
1362 .mqo_recovery = mdd_quota_recovery,
1363 .mqo_check = mdd_quota_check,
1364 .mqo_on = mdd_quota_on,
1365 .mqo_off = mdd_quota_off,
1366 .mqo_setinfo = mdd_quota_setinfo,
1367 .mqo_getinfo = mdd_quota_getinfo,
1368 .mqo_setquota = mdd_quota_setquota,
1369 .mqo_getquota = mdd_quota_getquota,
1370 .mqo_getoinfo = mdd_quota_getoinfo,
1371 .mqo_getoquota = mdd_quota_getoquota,
1372 .mqo_invalidate = mdd_quota_invalidate,
1373 .mqo_finvalidate = mdd_quota_finvalidate
1378 static struct lu_device_type_operations mdd_device_type_ops = {
1379 .ldto_init = mdd_type_init,
1380 .ldto_fini = mdd_type_fini,
1382 .ldto_start = mdd_type_start,
1383 .ldto_stop = mdd_type_stop,
1385 .ldto_device_alloc = mdd_device_alloc,
1386 .ldto_device_free = mdd_device_free,
1388 .ldto_device_init = mdd_device_init,
1389 .ldto_device_fini = mdd_device_fini
1392 static struct lu_device_type mdd_device_type = {
1393 .ldt_tags = LU_DEVICE_MD,
1394 .ldt_name = LUSTRE_MDD_NAME,
1395 .ldt_ops = &mdd_device_type_ops,
1396 .ldt_ctx_tags = LCT_MD_THREAD
1399 /* context key constructor: mdd_key_init */
1400 LU_KEY_INIT(mdd, struct mdd_thread_info);
1402 static void mdd_key_fini(const struct lu_context *ctx,
1403 struct lu_context_key *key, void *data)
1405 struct mdd_thread_info *info = data;
1406 if (info->mti_max_lmm != NULL)
1407 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1408 if (info->mti_max_cookie != NULL)
1409 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1410 mdd_buf_put(&info->mti_big_buf);
1415 /* context key: mdd_thread_key */
1416 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1418 static struct lu_local_obj_desc llod_capa_key = {
1419 .llod_name = CAPA_KEYS,
1420 .llod_oid = MDD_CAPA_KEYS_OID,
1424 static struct lu_local_obj_desc llod_mdd_orphan = {
1425 .llod_name = orph_index_name,
1426 .llod_oid = MDD_ORPHAN_OID,
1428 .llod_feat = &dt_directory_features,
1431 static struct lu_local_obj_desc llod_mdd_root = {
1432 .llod_name = mdd_root_dir_name,
1433 .llod_oid = MDD_ROOT_INDEX_OID,
1435 .llod_feat = &dt_directory_features,
1438 static int __init mdd_mod_init(void)
1440 struct lprocfs_static_vars lvars;
1441 lprocfs_mdd_init_vars(&lvars);
1443 llo_local_obj_register(&llod_capa_key);
1444 llo_local_obj_register(&llod_mdd_orphan);
1445 llo_local_obj_register(&llod_mdd_root);
1447 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1448 LUSTRE_MDD_NAME, &mdd_device_type);
1451 static void __exit mdd_mod_exit(void)
1453 llo_local_obj_unregister(&llod_capa_key);
1454 llo_local_obj_unregister(&llod_mdd_orphan);
1455 llo_local_obj_unregister(&llod_mdd_root);
1457 class_unregister_type(LUSTRE_MDD_NAME);
1460 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1461 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1462 MODULE_LICENSE("GPL");
1464 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);