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, 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 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_xattr_list(const struct lu_env *env,
454 struct md_object *obj, struct lu_buf *buf)
459 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
462 struct mdd_object *mdd_obj = md2mdd_obj(obj);
464 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
465 mdd_obj->mod_count++;
466 mdd_write_unlock(env, mdd_obj);
471 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
472 char *path, int pathlen, __u64 *recno, int *linkno)
477 static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
480 struct mdd_object *mdd_obj = md2mdd_obj(obj);
482 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
483 mdd_obj->mod_count--;
484 mdd_write_unlock(env, mdd_obj);
489 static dt_obj_version_t dot_lustre_version_get(const struct lu_env *env,
490 struct md_object *obj)
495 static void dot_lustre_version_set(const struct lu_env *env,
496 struct md_object *obj,
497 dt_obj_version_t version)
503 static struct md_object_operations mdd_dot_lustre_obj_ops = {
504 .moo_attr_get = dot_lustre_attr_get,
505 .moo_attr_set = dot_lustre_attr_set,
506 .moo_xattr_get = dot_lustre_xattr_get,
507 .moo_xattr_list = dot_lustre_xattr_list,
508 .moo_open = dot_lustre_mdd_open,
509 .moo_close = dot_lustre_close,
510 .moo_readpage = mdd_readpage,
511 .moo_version_get = dot_lustre_version_get,
512 .moo_version_set = dot_lustre_version_set,
513 .moo_path = dot_lustre_path
516 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
517 const struct lu_name *lname, struct lu_fid *f,
518 struct md_op_spec *spec)
520 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
528 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
529 const struct lu_name *lname,
530 struct md_object *child, struct md_op_spec *spec,
536 static int dot_lustre_rename(const struct lu_env *env,
537 struct md_object *src_pobj,
538 struct md_object *tgt_pobj,
539 const struct lu_fid *lf,
540 const struct lu_name *lsname,
541 struct md_object *tobj,
542 const struct lu_name *ltname, struct md_attr *ma)
547 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
548 struct md_object *src_obj,
549 const struct lu_name *lname, struct md_attr *ma)
554 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
555 struct md_object *cobj, const struct lu_name *lname,
561 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
562 .mdo_lookup = dot_lustre_lookup,
563 .mdo_create = dot_lustre_create,
564 .mdo_rename = dot_lustre_rename,
565 .mdo_link = dot_lustre_link,
566 .mdo_unlink = dot_lustre_unlink,
569 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
574 if (ma->ma_need & MA_INODE) {
575 struct mdd_device *mdd = mdo2mdd(obj);
577 /* "fid" is a virtual object and hence does not have any "real"
578 * attributes. So we reuse attributes of .lustre for "fid" dir */
579 ma->ma_need |= MA_INODE;
580 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
583 ma->ma_valid |= MA_INODE;
586 /* "fid" directory does not have any striping information. */
587 if (ma->ma_need & MA_LOV) {
588 struct mdd_object *mdd_obj = md2mdd_obj(obj);
590 if (ma->ma_valid & MA_LOV)
593 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
594 S_ISDIR(mdd_object_type(mdd_obj))))
597 if (ma->ma_need & MA_LOV_DEF) {
598 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
601 ma->ma_valid |= MA_LOV;
610 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
611 const struct md_attr *ma)
616 static int obf_xattr_get(const struct lu_env *env,
617 struct md_object *obj, struct lu_buf *buf,
623 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
626 struct mdd_object *mdd_obj = md2mdd_obj(obj);
628 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
629 mdd_obj->mod_count++;
630 mdd_write_unlock(env, mdd_obj);
635 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
638 struct mdd_object *mdd_obj = md2mdd_obj(obj);
640 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
641 mdd_obj->mod_count--;
642 mdd_write_unlock(env, mdd_obj);
647 /** Nothing to list in "fid" directory */
648 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
649 const struct lu_rdpg *rdpg)
654 static int obf_path(const struct lu_env *env, struct md_object *obj,
655 char *path, int pathlen, __u64 *recno, int *linkno)
660 static struct md_object_operations mdd_obf_obj_ops = {
661 .moo_attr_get = obf_attr_get,
662 .moo_attr_set = obf_attr_set,
663 .moo_xattr_get = obf_xattr_get,
664 .moo_open = obf_mdd_open,
665 .moo_close = obf_mdd_close,
666 .moo_readpage = obf_mdd_readpage,
671 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
672 * are valid. We also check if object with passed fid exists or not.
674 static int obf_lookup(const struct lu_env *env, struct md_object *p,
675 const struct lu_name *lname, struct lu_fid *f,
676 struct md_op_spec *spec)
678 char *name = (char *)lname->ln_name;
679 struct mdd_device *mdd = mdo2mdd(p);
680 struct mdd_object *child;
686 sscanf(name, SFID, RFID(f));
687 if (!fid_is_sane(f)) {
688 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
690 GOTO(out, rc = -EINVAL);
693 /* Check if object with this fid exists */
694 child = mdd_object_find(env, mdd, f);
698 GOTO(out, rc = PTR_ERR(child));
700 if (mdd_object_exists(child) == 0)
703 mdd_object_put(env, child);
709 static int obf_create(const struct lu_env *env, struct md_object *pobj,
710 const struct lu_name *lname, struct md_object *child,
711 struct md_op_spec *spec, struct md_attr* ma)
716 static int obf_rename(const struct lu_env *env,
717 struct md_object *src_pobj, struct md_object *tgt_pobj,
718 const struct lu_fid *lf, const struct lu_name *lsname,
719 struct md_object *tobj, const struct lu_name *ltname,
725 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
726 struct md_object *src_obj, const struct lu_name *lname,
732 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
733 struct md_object *cobj, const struct lu_name *lname,
739 static struct md_dir_operations mdd_obf_dir_ops = {
740 .mdo_lookup = obf_lookup,
741 .mdo_create = obf_create,
742 .mdo_rename = obf_rename,
743 .mdo_link = obf_link,
744 .mdo_unlink = obf_unlink
748 * Create special in-memory "fid" object for open-by-fid.
750 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
752 struct mdd_object *mdd_obf;
753 struct lu_object *obf_lu_obj;
756 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
758 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
759 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
760 GOTO(out, rc = -ENOENT);
762 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
763 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
764 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
765 /* Don't allow objects to be created in "fid" dir */
766 mdd_obf->mod_flags |= IMMUTE_OBJ;
768 obf_lu_obj = mdd2lu_obj(mdd_obf);
769 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
775 /** Setup ".lustre" directory object */
776 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
778 struct dt_object *dt_dot_lustre;
779 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
782 rc = create_dot_lustre_dir(env, m);
786 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
787 dot_lustre_name, fid);
788 if (IS_ERR(dt_dot_lustre)) {
789 rc = PTR_ERR(dt_dot_lustre);
793 /* references are released in mdd_device_shutdown() */
794 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
797 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
798 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
800 rc = mdd_obf_setup(env, m);
802 CERROR("Error initializing \"fid\" object - %d.\n", rc);
808 static int mdd_process_config(const struct lu_env *env,
809 struct lu_device *d, struct lustre_cfg *cfg)
811 struct mdd_device *m = lu2mdd_dev(d);
812 struct dt_device *dt = m->mdd_child;
813 struct lu_device *next = &dt->dd_lu_dev;
817 switch (cfg->lcfg_command) {
819 struct lprocfs_static_vars lvars;
821 lprocfs_mdd_init_vars(&lvars);
822 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
823 if (rc > 0 || rc == -ENOSYS)
824 /* we don't understand; pass it on */
825 rc = next->ld_ops->ldo_process_config(env, next, cfg);
829 rc = next->ld_ops->ldo_process_config(env, next, cfg);
832 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
834 rc = mdd_init_obd(env, m, cfg);
836 CERROR("lov init error %d \n", rc);
839 rc = mdd_txn_init_credits(env, m);
843 mdd_changelog_init(env, m);
846 mdd_device_shutdown(env, m, cfg);
848 rc = next->ld_ops->ldo_process_config(env, next, cfg);
856 static int mdd_lov_set_nextid(const struct lu_env *env,
857 struct mdd_device *mdd)
859 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
863 LASSERT(mds->mds_lov_objids != NULL);
864 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
865 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
866 mds->mds_lov_objids, NULL);
871 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
872 struct mdd_device *mdd)
874 /* XXX: to be implemented! */
879 static int mdd_recovery_complete(const struct lu_env *env,
882 struct mdd_device *mdd = lu2mdd_dev(d);
883 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
884 struct obd_device *obd = mdd2obd_dev(mdd);
888 LASSERT(mdd != NULL);
889 LASSERT(obd != NULL);
891 /* XXX: Do we need this in new stack? */
892 rc = mdd_lov_set_nextid(env, mdd);
894 CERROR("mdd_lov_set_nextid() failed %d\n",
899 /* XXX: cleanup unlink. */
900 rc = mdd_cleanup_unlink_llog(env, mdd);
902 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
907 /* Call that with obd_recovering = 1 just to update objids */
908 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
909 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
911 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
912 obd->obd_recovering = 0;
913 obd->obd_type->typ_dt_ops->o_postrecov(obd);
915 /* XXX: orphans handling. */
916 __mdd_orphan_cleanup(env, mdd);
917 rc = next->ld_ops->ldo_recovery_complete(env, next);
922 static int mdd_prepare(const struct lu_env *env,
923 struct lu_device *pdev,
924 struct lu_device *cdev)
926 struct mdd_device *mdd = lu2mdd_dev(cdev);
927 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
928 struct dt_object *root;
932 rc = next->ld_ops->ldo_prepare(env, cdev, next);
936 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
937 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
940 LASSERT(root != NULL);
941 lu_object_put(env, &root->do_lu);
942 rc = orph_index_init(env, mdd);
949 rc = mdd_dot_lustre_setup(env, mdd);
951 CERROR("Error(%d) initializing .lustre objects\n", rc);
959 const struct lu_device_operations mdd_lu_ops = {
960 .ldo_object_alloc = mdd_object_alloc,
961 .ldo_process_config = mdd_process_config,
962 .ldo_recovery_complete = mdd_recovery_complete,
963 .ldo_prepare = mdd_prepare,
967 * No permission check is needed.
969 static int mdd_root_get(const struct lu_env *env,
970 struct md_device *m, struct lu_fid *f)
972 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
975 *f = mdd->mdd_root_fid;
980 * No permission check is needed.
982 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
985 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
990 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
996 * No permission check is needed.
998 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
999 int *md_size, int *cookie_size)
1001 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1004 *md_size = mdd_lov_mdsize(env, mdd);
1005 *cookie_size = mdd_lov_cookiesize(env, mdd);
1010 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1011 int mode, unsigned long timeout, __u32 alg,
1012 struct lustre_capa_key *keys)
1014 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1015 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
1019 mds->mds_capa_keys = keys;
1020 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1021 timeout, alg, keys);
1025 static int mdd_update_capa_key(const struct lu_env *env,
1026 struct md_device *m,
1027 struct lustre_capa_key *key)
1029 struct mds_capa_info info = { .uuid = NULL, .capa = key };
1030 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1031 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
1035 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
1036 sizeof(info), &info, NULL);
1040 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1043 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1045 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1046 return (*h == NULL ? -ENOENT : 0);
1049 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1050 struct lu_device_type *t,
1051 struct lustre_cfg *lcfg)
1053 struct lu_device *l;
1054 struct mdd_device *m;
1058 l = ERR_PTR(-ENOMEM);
1060 md_device_init(&m->mdd_md_dev, t);
1062 l->ld_ops = &mdd_lu_ops;
1063 m->mdd_md_dev.md_ops = &mdd_ops;
1064 md_upcall_init(&m->mdd_md_dev, NULL);
1070 static struct lu_device *mdd_device_free(const struct lu_env *env,
1071 struct lu_device *lu)
1073 struct mdd_device *m = lu2mdd_dev(lu);
1074 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1077 LASSERT(atomic_read(&lu->ld_ref) == 0);
1078 md_device_fini(&m->mdd_md_dev);
1083 static struct obd_ops mdd_obd_device_ops = {
1084 .o_owner = THIS_MODULE
1087 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1088 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1090 static struct lu_context_key mdd_ucred_key = {
1091 .lct_tags = LCT_SESSION,
1092 .lct_init = mdd_ucred_key_init,
1093 .lct_fini = mdd_ucred_key_fini
1096 struct md_ucred *md_ucred(const struct lu_env *env)
1098 LASSERT(env->le_ses != NULL);
1099 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1101 EXPORT_SYMBOL(md_ucred);
1104 * context key constructor/destructor:
1105 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1107 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1109 struct lu_context_key mdd_capainfo_key = {
1110 .lct_tags = LCT_SESSION,
1111 .lct_init = mdd_capainfo_key_init,
1112 .lct_fini = mdd_capainfo_key_fini
1115 struct md_capainfo *md_capainfo(const struct lu_env *env)
1117 /* NB, in mdt_init0 */
1118 if (env->le_ses == NULL)
1120 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1122 EXPORT_SYMBOL(md_capainfo);
1124 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1126 struct llog_ctxt *ctxt;
1127 struct llog_changelog_user_rec *rec;
1131 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1137 llog_ctxt_put(ctxt);
1141 /* Assume we want it on since somebody registered */
1142 rc = mdd_changelog_on(mdd, 1);
1146 rec->cur_hdr.lrh_len = sizeof(*rec);
1147 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1148 spin_lock(&mdd->mdd_cl.mc_user_lock);
1149 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1150 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1151 CERROR("Maximum number of changelog users exceeded!\n");
1152 GOTO(out, rc = -EOVERFLOW);
1154 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1155 rec->cur_endrec = mdd->mdd_cl.mc_index;
1156 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1158 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1160 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1163 llog_ctxt_put(ctxt);
1167 struct mdd_changelog_user_data {
1168 __u64 mcud_endrec; /**< purge record for this user */
1169 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1171 __u32 mcud_minid; /**< user id with lowest rec reference */
1172 __u32 mcud_usercount;
1175 #define MCUD_UNREGISTER -1LL
1178 * 1. Find the smallest record everyone is willing to purge
1179 * 2. Update the last purgeable record for this user
1181 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1182 struct llog_rec_hdr *hdr, void *data)
1184 struct llog_changelog_user_rec *rec;
1185 struct mdd_changelog_user_data *mcud =
1186 (struct mdd_changelog_user_data *)data;
1190 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1192 rec = (struct llog_changelog_user_rec *)hdr;
1194 mcud->mcud_usercount++;
1196 /* If we have a new endrec for this id, use it for the following
1197 min check instead of its old value */
1198 if (rec->cur_id == mcud->mcud_id)
1199 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1201 /* Track the minimum referenced record */
1202 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1203 mcud->mcud_minid = rec->cur_id;
1204 mcud->mcud_minrec = rec->cur_endrec;
1207 if (rec->cur_id != mcud->mcud_id)
1210 /* Update this user's record */
1211 mcud->mcud_found = 1;
1213 /* Special case: unregister this user */
1214 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1215 struct llog_cookie cookie;
1216 cookie.lgc_lgl = llh->lgh_id;
1217 cookie.lgc_index = hdr->lrh_index;
1218 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1221 mcud->mcud_usercount--;
1225 /* Update the endrec */
1226 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1227 mcud->mcud_id, rec->cur_endrec);
1229 /* hdr+1 is loc of data */
1230 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1231 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1237 static int mdd_changelog_user_purge(struct mdd_device *mdd, int id,
1240 struct mdd_changelog_user_data data;
1241 struct llog_ctxt *ctxt;
1245 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec);
1248 data.mcud_minid = 0;
1249 data.mcud_minrec = 0;
1250 data.mcud_usercount = 0;
1251 data.mcud_endrec = endrec;
1252 spin_lock(&mdd->mdd_cl.mc_lock);
1253 endrec = mdd->mdd_cl.mc_index;
1254 spin_unlock(&mdd->mdd_cl.mc_lock);
1255 if ((data.mcud_endrec == 0) ||
1256 ((data.mcud_endrec > endrec) &&
1257 (data.mcud_endrec != MCUD_UNREGISTER)))
1258 data.mcud_endrec = endrec;
1260 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1263 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1265 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1266 (void *)&data, 0, 0);
1267 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1268 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1269 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1270 data.mcud_minrec, data.mcud_minid);
1271 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1273 CWARN("Could not determine changelog records to purge; rc=%d\n",
1277 llog_ctxt_put(ctxt);
1279 if (!data.mcud_found) {
1280 CWARN("No entry for user %d. Last changelog reference is "
1281 LPD64" by changelog user %d\n", data.mcud_id,
1282 data.mcud_minrec, data.mcud_minid);
1286 if (!rc && data.mcud_usercount == 0)
1287 /* No more users; turn changelogs off */
1288 rc = mdd_changelog_on(mdd, 0);
1294 * May be called remotely from mdt_iocontrol_handle or locally from
1295 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1296 * an obd_ioctl_data format (but local ioctl handler does).
1298 * \param len - data len
1299 * \param karg - ioctl data, in kernel space
1301 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1302 unsigned int cmd, int len, void *karg)
1304 struct mdd_device *mdd;
1305 struct obd_ioctl_data *data = karg;
1309 mdd = lu2mdd_dev(&m->md_lu_dev);
1311 /* Doesn't use obd_ioctl_data */
1312 if (cmd == OBD_IOC_CHANGELOG_CLEAR) {
1313 struct changelog_setinfo *cs = karg;
1314 rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno);
1318 /* Below ioctls use obd_ioctl_data */
1319 if (len != sizeof(*data)) {
1320 CERROR("Bad ioctl size %d\n", len);
1323 if (data->ioc_version != OBD_IOCTL_VERSION) {
1324 CERROR("Bad magic %x != %x\n", data->ioc_version,
1330 case OBD_IOC_CHANGELOG_REG:
1331 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1333 case OBD_IOC_CHANGELOG_DEREG:
1334 rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1,
1344 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1345 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
1347 const struct md_device_operations mdd_ops = {
1348 .mdo_statfs = mdd_statfs,
1349 .mdo_root_get = mdd_root_get,
1350 .mdo_maxsize_get = mdd_maxsize_get,
1351 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1352 .mdo_update_capa_key= mdd_update_capa_key,
1353 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1354 .mdo_iocontrol = mdd_iocontrol,
1355 #ifdef HAVE_QUOTA_SUPPORT
1357 .mqo_notify = mdd_quota_notify,
1358 .mqo_setup = mdd_quota_setup,
1359 .mqo_cleanup = mdd_quota_cleanup,
1360 .mqo_recovery = mdd_quota_recovery,
1361 .mqo_check = mdd_quota_check,
1362 .mqo_on = mdd_quota_on,
1363 .mqo_off = mdd_quota_off,
1364 .mqo_setinfo = mdd_quota_setinfo,
1365 .mqo_getinfo = mdd_quota_getinfo,
1366 .mqo_setquota = mdd_quota_setquota,
1367 .mqo_getquota = mdd_quota_getquota,
1368 .mqo_getoinfo = mdd_quota_getoinfo,
1369 .mqo_getoquota = mdd_quota_getoquota,
1370 .mqo_invalidate = mdd_quota_invalidate,
1371 .mqo_finvalidate = mdd_quota_finvalidate
1376 static struct lu_device_type_operations mdd_device_type_ops = {
1377 .ldto_init = mdd_type_init,
1378 .ldto_fini = mdd_type_fini,
1380 .ldto_start = mdd_type_start,
1381 .ldto_stop = mdd_type_stop,
1383 .ldto_device_alloc = mdd_device_alloc,
1384 .ldto_device_free = mdd_device_free,
1386 .ldto_device_init = mdd_device_init,
1387 .ldto_device_fini = mdd_device_fini
1390 static struct lu_device_type mdd_device_type = {
1391 .ldt_tags = LU_DEVICE_MD,
1392 .ldt_name = LUSTRE_MDD_NAME,
1393 .ldt_ops = &mdd_device_type_ops,
1394 .ldt_ctx_tags = LCT_MD_THREAD
1397 /* context key constructor: mdd_key_init */
1398 LU_KEY_INIT(mdd, struct mdd_thread_info);
1400 static void mdd_key_fini(const struct lu_context *ctx,
1401 struct lu_context_key *key, void *data)
1403 struct mdd_thread_info *info = data;
1404 if (info->mti_max_lmm != NULL)
1405 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1406 if (info->mti_max_cookie != NULL)
1407 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1408 mdd_buf_put(&info->mti_big_buf);
1413 /* context key: mdd_thread_key */
1414 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1416 static struct lu_local_obj_desc llod_capa_key = {
1417 .llod_name = CAPA_KEYS,
1418 .llod_oid = MDD_CAPA_KEYS_OID,
1422 static struct lu_local_obj_desc llod_mdd_orphan = {
1423 .llod_name = orph_index_name,
1424 .llod_oid = MDD_ORPHAN_OID,
1426 .llod_feat = &dt_directory_features,
1429 static struct lu_local_obj_desc llod_mdd_root = {
1430 .llod_name = mdd_root_dir_name,
1431 .llod_oid = MDD_ROOT_INDEX_OID,
1433 .llod_feat = &dt_directory_features,
1436 static int __init mdd_mod_init(void)
1438 struct lprocfs_static_vars lvars;
1439 lprocfs_mdd_init_vars(&lvars);
1441 llo_local_obj_register(&llod_capa_key);
1442 llo_local_obj_register(&llod_mdd_orphan);
1443 llo_local_obj_register(&llod_mdd_root);
1445 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1446 LUSTRE_MDD_NAME, &mdd_device_type);
1449 static void __exit mdd_mod_exit(void)
1451 llo_local_obj_unregister(&llod_capa_key);
1452 llo_local_obj_unregister(&llod_mdd_orphan);
1453 llo_local_obj_unregister(&llod_mdd_root);
1455 class_unregister_type(LUSTRE_MDD_NAME);
1458 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1459 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1460 MODULE_LICENSE("GPL");
1462 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);