1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_device.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #include <linux/jbd.h>
51 #include <obd_class.h>
52 #include <lustre_ver.h>
53 #include <obd_support.h>
54 #include <lprocfs_status.h>
56 #include <lustre_disk.h>
57 #include <lustre_fid.h>
58 #include <linux/ldiskfs_fs.h>
59 #include <lustre_mds.h>
60 #include <lustre/lustre_idl.h>
61 #include <lustre_disk.h> /* for changelogs */
62 #include <lustre_param.h>
63 #include <lustre_fid.h>
65 #include "mdd_internal.h"
67 const struct md_device_operations mdd_ops;
68 static struct lu_device_type mdd_device_type;
70 static const char mdd_root_dir_name[] = "ROOT";
71 static const char mdd_obf_dir_name[] = "fid";
73 static int mdd_device_init(const struct lu_env *env, struct lu_device *d,
74 const char *name, struct lu_device *next)
76 struct mdd_device *mdd = lu2mdd_dev(d);
80 mdd->mdd_child = lu2dt_dev(next);
82 /* Prepare transactions callbacks. */
83 mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb;
84 mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb;
85 mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb;
86 mdd->mdd_txn_cb.dtc_cookie = mdd;
87 CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage);
88 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
90 rc = mdd_procfs_init(mdd, name);
94 static struct lu_device *mdd_device_fini(const struct lu_env *env,
97 struct mdd_device *mdd = lu2mdd_dev(d);
98 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
101 rc = mdd_procfs_fini(mdd);
103 CERROR("proc fini error %d \n", rc);
109 static void mdd_changelog_fini(const struct lu_env *env,
110 struct mdd_device *mdd);
112 static void mdd_device_shutdown(const struct lu_env *env,
113 struct mdd_device *m, struct lustre_cfg *cfg)
116 mdd_changelog_fini(env, m);
117 dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb);
118 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
119 mdd_object_put(env, m->mdd_dot_lustre);
121 mdd_fini_obd(env, m, cfg);
122 orph_index_fini(env, m);
123 /* remove upcall device*/
124 md_upcall_fini(&m->mdd_md_dev);
128 static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
131 struct mdd_device *mdd = (struct mdd_device *)data;
132 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
135 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
136 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
139 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
140 hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index,
141 rec->cr_type, rec->cr_namelen, rec->cr_name,
142 llh->lgh_id.lgl_oid);
144 mdd->mdd_cl.mc_index = rec->cr_index;
145 RETURN(LLOG_PROC_BREAK);
148 static int changelog_user_init_cb(struct llog_handle *llh,
149 struct llog_rec_hdr *hdr, void *data)
151 struct mdd_device *mdd = (struct mdd_device *)data;
152 struct llog_changelog_user_rec *rec =
153 (struct llog_changelog_user_rec *)hdr;
156 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
157 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
159 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
160 " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
161 rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
163 spin_lock(&mdd->mdd_cl.mc_user_lock);
164 mdd->mdd_cl.mc_lastuser = rec->cur_id;
165 spin_unlock(&mdd->mdd_cl.mc_user_lock);
167 RETURN(LLOG_PROC_BREAK);
171 static int mdd_changelog_llog_init(struct mdd_device *mdd)
173 struct obd_device *obd = mdd2obd_dev(mdd);
174 struct llog_ctxt *ctxt;
177 /* Find last changelog entry number */
178 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
180 CERROR("no changelog context\n");
183 if (!ctxt->loc_handle) {
188 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd);
192 CERROR("changelog init failed: %d\n", rc);
195 CDEBUG(D_INODE, "changelog starting index="LPU64"\n",
196 mdd->mdd_cl.mc_index);
198 /* Find last changelog user id */
199 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
201 CERROR("no changelog user context\n");
204 if (!ctxt->loc_handle) {
209 rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb,
214 CERROR("changelog user init failed: %d\n", rc);
220 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
224 mdd->mdd_cl.mc_index = 0;
225 spin_lock_init(&mdd->mdd_cl.mc_lock);
226 cfs_waitq_init(&mdd->mdd_cl.mc_waitq);
227 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
228 mdd->mdd_cl.mc_flags = 0; /* off by default */
229 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
230 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
231 mdd->mdd_cl.mc_lastuser = 0;
233 rc = mdd_changelog_llog_init(mdd);
235 CERROR("Changelog setup during init failed %d\n", rc);
236 mdd->mdd_cl.mc_flags |= CLM_ERR;
242 static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd)
244 mdd->mdd_cl.mc_flags = 0;
247 /** Add a changelog entry \a rec to the changelog llog
250 * \param handle - currently ignored since llogs start their own transaction;
251 * this will hopefully be fixed in llog rewrite
254 int mdd_changelog_llog_write(struct mdd_device *mdd,
255 struct llog_changelog_rec *rec,
256 struct thandle *handle)
258 struct obd_device *obd = mdd2obd_dev(mdd);
259 struct llog_ctxt *ctxt;
262 if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0)
265 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen);
266 /* llog_lvfs_write_rec sets the llog tail len */
267 rec->cr_hdr.lrh_type = CHANGELOG_REC;
268 rec->cr_time = cfs_time_current_64();
269 spin_lock(&mdd->mdd_cl.mc_lock);
270 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
271 but as long as the MDD transactions are ordered correctly for e.g.
272 rename conflicts, I don't think this should matter. */
273 rec->cr_index = ++mdd->mdd_cl.mc_index;
274 spin_unlock(&mdd->mdd_cl.mc_lock);
275 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
279 /* nested journal transaction */
280 rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0);
283 cfs_waitq_signal(&mdd->mdd_cl.mc_waitq);
288 /** Remove entries with indicies up to and including \a endrec from the
294 int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec)
296 struct obd_device *obd = mdd2obd_dev(mdd);
297 struct llog_ctxt *ctxt;
298 long long unsigned cur;
301 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
305 spin_lock(&mdd->mdd_cl.mc_lock);
306 cur = (long long)mdd->mdd_cl.mc_index;
307 spin_unlock(&mdd->mdd_cl.mc_lock);
311 /* purge to "0" is shorthand for everything */
315 /* If purging all records, write a header entry so we don't have an
316 empty catalog and we're sure to have a valid starting index next
317 time. In case of crash, we just restart with old log so we're
320 rc = mdd_changelog_write_header(mdd, CLM_PURGE);
325 /* Some records were purged, so reset repeat-access time (so we
326 record new mtime update records, so users can see a file has been
327 changed since the last purge) */
328 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
330 rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
336 /** Add a CL_MARK record to the changelog
338 * \param markerflags - CLM_*
341 int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags)
343 struct obd_device *obd = mdd2obd_dev(mdd);
344 struct llog_changelog_rec *rec;
346 int len = strlen(obd->obd_name);
350 reclen = llog_data_len(sizeof(*rec) + len);
351 OBD_ALLOC(rec, reclen);
355 rec->cr_flags = CLF_VERSION;
356 rec->cr_type = CL_MARK;
357 rec->cr_namelen = len;
358 memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen);
359 /* Status and action flags */
360 rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
362 rc = mdd_changelog_llog_write(mdd, rec, NULL);
364 /* assume on or off event; reset repeat-access time */
365 mdd->mdd_cl.mc_starttime = rec->cr_time;
367 OBD_FREE(rec, reclen);
372 * Create ".lustre" directory.
374 static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
376 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
377 struct md_object *mdo;
380 memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
381 mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
382 mdd_root_dir_name, mdd_dot_lustre_name,
383 fid, &dt_directory_features);
384 /* .lustre dir may be already present */
385 if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
387 CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
388 mdd_dot_lustre_name, PFID(fid), rc);
393 lu_object_put(env, &mdo->mo_lu);
398 static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj,
401 struct mdd_object *mdd_obj = md2mdd_obj(obj);
403 return mdd_attr_get_internal_locked(env, mdd_obj, ma);
406 static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj,
407 const struct md_attr *ma)
412 static int dot_lustre_xattr_get(const struct lu_env *env,
413 struct md_object *obj, struct lu_buf *buf,
419 static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
422 struct mdd_object *mdd_obj = md2mdd_obj(obj);
424 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
425 mdd_obj->mod_count++;
426 mdd_write_unlock(env, mdd_obj);
431 static int dot_lustre_path(const struct lu_env *env, struct md_object *obj,
432 char *path, int pathlen, __u64 *recno, int *linkno)
437 static int dot_lustre_close(const struct lu_env *env, struct md_object *obj,
440 struct mdd_object *mdd_obj = md2mdd_obj(obj);
442 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
443 mdd_obj->mod_count--;
444 mdd_write_unlock(env, mdd_obj);
449 static struct md_object_operations mdd_dot_lustre_obj_ops = {
450 .moo_attr_get = dot_lustre_attr_get,
451 .moo_attr_set = dot_lustre_attr_set,
452 .moo_xattr_get = dot_lustre_xattr_get,
453 .moo_open = dot_lustre_mdd_open,
454 .moo_close = dot_lustre_close,
455 .moo_readpage = mdd_readpage,
456 .moo_path = dot_lustre_path
459 static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p,
460 const struct lu_name *lname, struct lu_fid *f,
461 struct md_op_spec *spec)
463 if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
471 static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj,
472 const struct lu_name *lname,
473 struct md_object *child, struct md_op_spec *spec,
479 static int dot_lustre_rename(const struct lu_env *env,
480 struct md_object *src_pobj,
481 struct md_object *tgt_pobj,
482 const struct lu_fid *lf,
483 const struct lu_name *lsname,
484 struct md_object *tobj,
485 const struct lu_name *ltname, struct md_attr *ma)
490 static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj,
491 struct md_object *src_obj,
492 const struct lu_name *lname, struct md_attr *ma)
497 static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj,
498 struct md_object *cobj, const struct lu_name *lname,
504 static struct md_dir_operations mdd_dot_lustre_dir_ops = {
505 .mdo_lookup = dot_lustre_lookup,
506 .mdo_create = dot_lustre_create,
507 .mdo_rename = dot_lustre_rename,
508 .mdo_link = dot_lustre_link,
509 .mdo_unlink = dot_lustre_unlink,
512 static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
517 if (ma->ma_need & MA_INODE) {
518 struct mdd_device *mdd = mdo2mdd(obj);
520 /* "fid" is a virtual object and hence does not have any "real"
521 * attributes. So we reuse attributes of .lustre for "fid" dir */
522 ma->ma_need |= MA_INODE;
523 rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
526 ma->ma_valid |= MA_INODE;
529 /* "fid" directory does not have any striping information. */
530 if (ma->ma_need & MA_LOV) {
531 struct mdd_object *mdd_obj = md2mdd_obj(obj);
533 if (ma->ma_valid & MA_LOV)
536 if (!(S_ISREG(mdd_object_type(mdd_obj)) ||
537 S_ISDIR(mdd_object_type(mdd_obj))))
540 if (ma->ma_need & MA_LOV_DEF) {
541 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
544 ma->ma_valid |= MA_LOV;
553 static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
554 const struct md_attr *ma)
559 static int obf_xattr_get(const struct lu_env *env,
560 struct md_object *obj, struct lu_buf *buf,
566 static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
569 struct mdd_object *mdd_obj = md2mdd_obj(obj);
571 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
572 mdd_obj->mod_count++;
573 mdd_write_unlock(env, mdd_obj);
578 static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
581 struct mdd_object *mdd_obj = md2mdd_obj(obj);
583 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
584 mdd_obj->mod_count--;
585 mdd_write_unlock(env, mdd_obj);
590 /** Nothing to list in "fid" directory */
591 static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
592 const struct lu_rdpg *rdpg)
597 static int obf_path(const struct lu_env *env, struct md_object *obj,
598 char *path, int pathlen, __u64 *recno, int *linkno)
603 static struct md_object_operations mdd_obf_obj_ops = {
604 .moo_attr_get = obf_attr_get,
605 .moo_attr_set = obf_attr_set,
606 .moo_xattr_get = obf_xattr_get,
607 .moo_open = obf_mdd_open,
608 .moo_close = obf_mdd_close,
609 .moo_readpage = obf_mdd_readpage,
614 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
615 * are valid. We also check if object with passed fid exists or not.
617 static int obf_lookup(const struct lu_env *env, struct md_object *p,
618 const struct lu_name *lname, struct lu_fid *f,
619 struct md_op_spec *spec)
621 char *name = (char *)lname->ln_name;
622 struct mdd_device *mdd = mdo2mdd(p);
623 struct mdd_object *child;
629 sscanf(name, SFID, &(f->f_seq), &(f->f_oid),
631 if (!fid_is_sane(f)) {
632 CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name,
634 GOTO(out, rc = -EINVAL);
637 /* Check if object with this fid exists */
638 child = mdd_object_find(env, mdd, f);
642 GOTO(out, rc = PTR_ERR(child));
644 if (mdd_object_exists(child) == 0)
647 mdd_object_put(env, child);
653 static int obf_create(const struct lu_env *env, struct md_object *pobj,
654 const struct lu_name *lname, struct md_object *child,
655 struct md_op_spec *spec, struct md_attr* ma)
660 static int obf_rename(const struct lu_env *env,
661 struct md_object *src_pobj, struct md_object *tgt_pobj,
662 const struct lu_fid *lf, const struct lu_name *lsname,
663 struct md_object *tobj, const struct lu_name *ltname,
669 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
670 struct md_object *src_obj, const struct lu_name *lname,
676 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
677 struct md_object *cobj, const struct lu_name *lname,
683 static struct md_dir_operations mdd_obf_dir_ops = {
684 .mdo_lookup = obf_lookup,
685 .mdo_create = obf_create,
686 .mdo_rename = obf_rename,
687 .mdo_link = obf_link,
688 .mdo_unlink = obf_unlink
692 * Create special in-memory "fid" object for open-by-fid.
694 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
696 struct mdd_object *mdd_obf;
697 struct lu_object *obf_lu_obj;
700 m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
702 if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
703 IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
704 GOTO(out, rc = -ENOENT);
706 mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
707 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
708 mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
709 /* Don't allow objects to be created in "fid" dir */
710 mdd_obf->mod_flags |= IMMUTE_OBJ;
712 obf_lu_obj = mdd2lu_obj(mdd_obf);
713 obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
719 /** Setup ".lustre" directory object */
720 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
722 struct dt_object *dt_dot_lustre;
723 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
726 rc = create_dot_lustre_dir(env, m);
730 dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
731 mdd_dot_lustre_name, fid);
732 if (IS_ERR(dt_dot_lustre)) {
733 rc = PTR_ERR(dt_dot_lustre);
737 /* references are released in mdd_device_shutdown() */
738 m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
741 m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
742 m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
744 rc = mdd_obf_setup(env, m);
746 CERROR("Error initializing \"fid\" object - %d.\n", rc);
752 static int mdd_process_config(const struct lu_env *env,
753 struct lu_device *d, struct lustre_cfg *cfg)
755 struct mdd_device *m = lu2mdd_dev(d);
756 struct dt_device *dt = m->mdd_child;
757 struct lu_device *next = &dt->dd_lu_dev;
761 switch (cfg->lcfg_command) {
763 struct lprocfs_static_vars lvars;
765 lprocfs_mdd_init_vars(&lvars);
766 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
767 if (rc > 0 || rc == -ENOSYS)
768 /* we don't understand; pass it on */
769 rc = next->ld_ops->ldo_process_config(env, next, cfg);
773 rc = next->ld_ops->ldo_process_config(env, next, cfg);
776 dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf);
778 rc = mdd_init_obd(env, m, cfg);
780 CERROR("lov init error %d \n", rc);
783 rc = mdd_txn_init_credits(env, m);
787 mdd_changelog_init(env, m);
790 mdd_device_shutdown(env, m, cfg);
792 rc = next->ld_ops->ldo_process_config(env, next, cfg);
800 static int mdd_lov_set_nextid(const struct lu_env *env,
801 struct mdd_device *mdd)
803 struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds;
807 LASSERT(mds->mds_lov_objids != NULL);
808 rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID),
809 KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count,
810 mds->mds_lov_objids, NULL);
815 static int mdd_cleanup_unlink_llog(const struct lu_env *env,
816 struct mdd_device *mdd)
818 /* XXX: to be implemented! */
823 static int mdd_recovery_complete(const struct lu_env *env,
826 struct mdd_device *mdd = lu2mdd_dev(d);
827 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
828 struct obd_device *obd = mdd2obd_dev(mdd);
832 LASSERT(mdd != NULL);
833 LASSERT(obd != NULL);
835 /* XXX: Do we need this in new stack? */
836 rc = mdd_lov_set_nextid(env, mdd);
838 CERROR("mdd_lov_set_nextid() failed %d\n",
843 /* XXX: cleanup unlink. */
844 rc = mdd_cleanup_unlink_llog(env, mdd);
846 CERROR("mdd_cleanup_unlink_llog() failed %d\n",
851 /* Call that with obd_recovering = 1 just to update objids */
852 obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ?
853 OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL);
855 /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */
856 obd->obd_recovering = 0;
857 obd->obd_type->typ_dt_ops->o_postrecov(obd);
859 /* XXX: orphans handling. */
860 __mdd_orphan_cleanup(env, mdd);
861 rc = next->ld_ops->ldo_recovery_complete(env, next);
866 static int mdd_prepare(const struct lu_env *env,
867 struct lu_device *pdev,
868 struct lu_device *cdev)
870 struct mdd_device *mdd = lu2mdd_dev(cdev);
871 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
872 struct dt_object *root;
876 rc = next->ld_ops->ldo_prepare(env, cdev, next);
880 dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb);
881 root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name,
884 LASSERT(root != NULL);
885 lu_object_put(env, &root->do_lu);
886 rc = orph_index_init(env, mdd);
893 rc = mdd_dot_lustre_setup(env, mdd);
895 CERROR("Error(%d) initializing .lustre objects\n", rc);
903 const struct lu_device_operations mdd_lu_ops = {
904 .ldo_object_alloc = mdd_object_alloc,
905 .ldo_process_config = mdd_process_config,
906 .ldo_recovery_complete = mdd_recovery_complete,
907 .ldo_prepare = mdd_prepare,
911 * No permission check is needed.
913 static int mdd_root_get(const struct lu_env *env,
914 struct md_device *m, struct lu_fid *f)
916 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
919 *f = mdd->mdd_root_fid;
924 * No permission check is needed.
926 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
929 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
934 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
940 * No permission check is needed.
942 static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m,
943 int *md_size, int *cookie_size)
945 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
948 *md_size = mdd_lov_mdsize(env, mdd);
949 *cookie_size = mdd_lov_cookiesize(env, mdd);
954 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
955 int mode, unsigned long timeout, __u32 alg,
956 struct lustre_capa_key *keys)
958 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
959 struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds;
963 mds->mds_capa_keys = keys;
964 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
969 static int mdd_update_capa_key(const struct lu_env *env,
971 struct lustre_capa_key *key)
973 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
974 struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp;
978 rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
979 sizeof(*key), key, NULL);
983 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
986 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
988 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
989 return (*h == NULL ? -ENOENT : 0);
992 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
993 struct lu_device_type *t,
994 struct lustre_cfg *lcfg)
997 struct mdd_device *m;
1001 l = ERR_PTR(-ENOMEM);
1003 md_device_init(&m->mdd_md_dev, t);
1005 l->ld_ops = &mdd_lu_ops;
1006 m->mdd_md_dev.md_ops = &mdd_ops;
1007 md_upcall_init(&m->mdd_md_dev, NULL);
1013 static struct lu_device *mdd_device_free(const struct lu_env *env,
1014 struct lu_device *lu)
1016 struct mdd_device *m = lu2mdd_dev(lu);
1017 struct lu_device *next = &m->mdd_child->dd_lu_dev;
1020 LASSERT(atomic_read(&lu->ld_ref) == 0);
1021 md_device_fini(&m->mdd_md_dev);
1026 static struct obd_ops mdd_obd_device_ops = {
1027 .o_owner = THIS_MODULE
1030 /* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */
1031 LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred);
1033 static struct lu_context_key mdd_ucred_key = {
1034 .lct_tags = LCT_SESSION,
1035 .lct_init = mdd_ucred_key_init,
1036 .lct_fini = mdd_ucred_key_fini
1039 struct md_ucred *md_ucred(const struct lu_env *env)
1041 LASSERT(env->le_ses != NULL);
1042 return lu_context_key_get(env->le_ses, &mdd_ucred_key);
1044 EXPORT_SYMBOL(md_ucred);
1047 * context key constructor/destructor:
1048 * mdd_capainfo_key_init, mdd_capainfo_key_fini
1050 LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo);
1052 struct lu_context_key mdd_capainfo_key = {
1053 .lct_tags = LCT_SESSION,
1054 .lct_init = mdd_capainfo_key_init,
1055 .lct_fini = mdd_capainfo_key_fini
1058 struct md_capainfo *md_capainfo(const struct lu_env *env)
1060 /* NB, in mdt_init0 */
1061 if (env->le_ses == NULL)
1063 return lu_context_key_get(env->le_ses, &mdd_capainfo_key);
1065 EXPORT_SYMBOL(md_capainfo);
1067 static int mdd_changelog_user_register(struct mdd_device *mdd, int *id)
1069 struct llog_ctxt *ctxt;
1070 struct llog_changelog_user_rec *rec;
1074 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1080 llog_ctxt_put(ctxt);
1084 rec->cur_hdr.lrh_len = sizeof(*rec);
1085 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1086 rec->cur_endrec = 0ULL;
1087 spin_lock(&mdd->mdd_cl.mc_user_lock);
1088 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1089 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1090 CERROR("Maximum number of changelog users exceeded!\n");
1091 GOTO(out, rc = -EOVERFLOW);
1093 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1094 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1095 rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0);
1097 CDEBUG(D_INODE, "Registered changelog user %d\n", *id);
1100 llog_ctxt_put(ctxt);
1104 struct mdd_changelog_user_data {
1105 __u64 mcud_endrec; /**< purge record for this user */
1106 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1108 __u32 mcud_minid; /**< user id with lowest rec reference */
1113 * 1. Find the smallest record everyone is willing to purge
1114 * 2. Update the last purgeable record for this user
1116 static int mdd_changelog_user_purge_cb(struct llog_handle *llh,
1117 struct llog_rec_hdr *hdr, void *data)
1119 struct llog_changelog_user_rec *rec;
1120 struct mdd_changelog_user_data *mcud =
1121 (struct mdd_changelog_user_data *)data;
1125 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1127 rec = (struct llog_changelog_user_rec *)hdr;
1129 /* If we have a new endrec for this id, use it for the min check */
1130 if (rec->cur_id == mcud->mcud_id)
1131 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1133 /* Track the minimum referenced record */
1134 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1135 mcud->mcud_minid = rec->cur_id;
1136 mcud->mcud_minrec = rec->cur_endrec;
1139 if (rec->cur_id != mcud->mcud_id)
1142 /* Update this user's record */
1143 mcud->mcud_found = 1;
1145 /* Special case: unregister this user if endrec == -1 */
1146 if (mcud->mcud_endrec == -1) {
1147 struct llog_cookie cookie;
1148 cookie.lgc_lgl = llh->lgh_id;
1149 cookie.lgc_index = hdr->lrh_index;
1150 rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle,
1155 /* Update the endrec */
1156 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1157 mcud->mcud_id, rec->cur_endrec);
1159 /* hdr+1 is loc of data */
1160 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1161 rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1),
1167 static int mdd_changelog_user_purge(struct mdd_device *mdd, int id,
1170 struct mdd_changelog_user_data data;
1171 struct llog_ctxt *ctxt;
1175 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec);
1177 ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT);
1180 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1183 data.mcud_endrec = endrec;
1184 data.mcud_minid = 0;
1185 data.mcud_minrec = 0;
1186 rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb,
1187 (void *)&data, 0, 0);
1188 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1189 CDEBUG(D_INODE, "Purging CL entries up to "LPD64
1190 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1191 data.mcud_minrec, data.mcud_minid);
1192 rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec);
1194 CWARN("Could not determine changelog records to purge; rc=%d\n",
1198 if (!data.mcud_found) {
1199 CWARN("No entry for user %d. Last changelog reference is "
1200 LPD64" by changelog user %d\n", data.mcud_id,
1201 data.mcud_minrec, data.mcud_minid);
1205 llog_ctxt_put(ctxt);
1210 * May be called remotely from mdt_iocontrol_handle or locally from
1211 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce or
1212 * swab an obd_ioctl_data format (but local ioctl handler does).
1214 * \param len - data len
1215 * \param karg - ioctl data, in kernel space
1217 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1218 unsigned int cmd, int len, void *karg)
1220 struct mdd_device *mdd;
1221 struct obd_ioctl_data *data = karg;
1225 mdd = lu2mdd_dev(&m->md_lu_dev);
1227 /* Doesn't use obd_ioctl_data */
1228 if (cmd == OBD_IOC_CHANGELOG_CLEAR) {
1229 struct changelog_setinfo *cs = karg;
1230 if (len != sizeof(*cs)) {
1231 CERROR("Bad changelog_clear ioctl size %d\n", len);
1234 rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno);
1238 /* Below ioctls use obd_ioctl_data */
1239 if (len != sizeof(*data)) {
1240 CERROR("Bad ioctl size %d\n", len);
1243 if (data->ioc_version != OBD_IOCTL_VERSION) {
1244 CERROR("Bad magic %x != %x\n", data->ioc_version,
1250 case OBD_IOC_CHANGELOG_REG:
1251 rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1);
1253 case OBD_IOC_CHANGELOG_DEREG:
1254 rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1, -1);
1263 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1264 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key);
1266 const struct md_device_operations mdd_ops = {
1267 .mdo_statfs = mdd_statfs,
1268 .mdo_root_get = mdd_root_get,
1269 .mdo_maxsize_get = mdd_maxsize_get,
1270 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1271 .mdo_update_capa_key= mdd_update_capa_key,
1272 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1273 .mdo_iocontrol = mdd_iocontrol,
1274 #ifdef HAVE_QUOTA_SUPPORT
1276 .mqo_notify = mdd_quota_notify,
1277 .mqo_setup = mdd_quota_setup,
1278 .mqo_cleanup = mdd_quota_cleanup,
1279 .mqo_recovery = mdd_quota_recovery,
1280 .mqo_check = mdd_quota_check,
1281 .mqo_on = mdd_quota_on,
1282 .mqo_off = mdd_quota_off,
1283 .mqo_setinfo = mdd_quota_setinfo,
1284 .mqo_getinfo = mdd_quota_getinfo,
1285 .mqo_setquota = mdd_quota_setquota,
1286 .mqo_getquota = mdd_quota_getquota,
1287 .mqo_getoinfo = mdd_quota_getoinfo,
1288 .mqo_getoquota = mdd_quota_getoquota,
1289 .mqo_invalidate = mdd_quota_invalidate,
1290 .mqo_finvalidate = mdd_quota_finvalidate
1295 static struct lu_device_type_operations mdd_device_type_ops = {
1296 .ldto_init = mdd_type_init,
1297 .ldto_fini = mdd_type_fini,
1299 .ldto_start = mdd_type_start,
1300 .ldto_stop = mdd_type_stop,
1302 .ldto_device_alloc = mdd_device_alloc,
1303 .ldto_device_free = mdd_device_free,
1305 .ldto_device_init = mdd_device_init,
1306 .ldto_device_fini = mdd_device_fini
1309 static struct lu_device_type mdd_device_type = {
1310 .ldt_tags = LU_DEVICE_MD,
1311 .ldt_name = LUSTRE_MDD_NAME,
1312 .ldt_ops = &mdd_device_type_ops,
1313 .ldt_ctx_tags = LCT_MD_THREAD
1316 /* context key constructor: mdd_key_init */
1317 LU_KEY_INIT(mdd, struct mdd_thread_info);
1319 static void mdd_key_fini(const struct lu_context *ctx,
1320 struct lu_context_key *key, void *data)
1322 struct mdd_thread_info *info = data;
1323 if (info->mti_max_lmm != NULL)
1324 OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size);
1325 if (info->mti_max_cookie != NULL)
1326 OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size);
1327 mdd_buf_put(&info->mti_big_buf);
1332 /* context key: mdd_thread_key */
1333 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1335 static struct lu_local_obj_desc llod_capa_key = {
1336 .llod_name = CAPA_KEYS,
1337 .llod_oid = MDD_CAPA_KEYS_OID,
1341 static struct lu_local_obj_desc llod_mdd_orphan = {
1342 .llod_name = orph_index_name,
1343 .llod_oid = MDD_ORPHAN_OID,
1345 .llod_feat = &dt_directory_features,
1348 static struct lu_local_obj_desc llod_mdd_root = {
1349 .llod_name = mdd_root_dir_name,
1350 .llod_oid = MDD_ROOT_INDEX_OID,
1352 .llod_feat = &dt_directory_features,
1355 static int __init mdd_mod_init(void)
1357 struct lprocfs_static_vars lvars;
1358 lprocfs_mdd_init_vars(&lvars);
1360 llo_local_obj_register(&llod_capa_key);
1361 llo_local_obj_register(&llod_mdd_orphan);
1362 llo_local_obj_register(&llod_mdd_root);
1364 return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars,
1365 LUSTRE_MDD_NAME, &mdd_device_type);
1368 static void __exit mdd_mod_exit(void)
1370 llo_local_obj_unregister(&llod_capa_key);
1371 llo_local_obj_unregister(&llod_mdd_orphan);
1372 llo_local_obj_unregister(&llod_mdd_root);
1374 class_unregister_type(LUSTRE_MDD_NAME);
1377 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1378 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1379 MODULE_LICENSE("GPL");
1381 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);