4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
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>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #include <obd_class.h>
47 #include <lustre_mds.h>
48 #include <obd_support.h>
49 #include <lu_object.h>
50 #include <lustre_param.h>
51 #include <lustre_fid.h>
53 #include "mdd_internal.h"
55 const struct md_device_operations mdd_ops;
56 static struct lu_device_type mdd_device_type;
58 static const char mdd_root_dir_name[] = "ROOT";
59 static const char mdd_obf_dir_name[] = "fid";
61 /* Slab for MDD object allocation */
62 struct kmem_cache *mdd_object_kmem;
64 static struct lu_kmem_descr mdd_caches[] = {
66 .ckd_cache = &mdd_object_kmem,
67 .ckd_name = "mdd_obj",
68 .ckd_size = sizeof(struct mdd_object)
75 static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m,
78 struct obd_connect_data *data = NULL;
79 struct lu_device *lud = mdd2lu_dev(m);
80 struct obd_device *obd;
84 LASSERT(m->mdd_child_exp == NULL);
86 OBD_ALLOC(data, sizeof(*data));
88 GOTO(out, rc = -ENOMEM);
90 obd = class_name2obd(nextdev);
92 CERROR("can't locate next device: %s\n", nextdev);
93 GOTO(out, rc = -ENOTCONN);
96 data->ocd_connect_flags = OBD_CONNECT_VERSION;
97 data->ocd_version = LUSTRE_VERSION_CODE;
99 rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL);
101 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
105 lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site;
106 LASSERT(lud->ld_site);
107 m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev);
108 m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev);
109 lu_dev_add_linkage(lud->ld_site, lud);
113 OBD_FREE(data, sizeof(*data));
117 static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd,
118 struct lu_device_type *t, struct lustre_cfg *lcfg)
123 mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops;
124 mdd->mdd_md_dev.md_ops = &mdd_ops;
126 rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3));
130 mdd->mdd_atime_diff = MAX_ATIME_DIFF;
131 /* sync permission changes */
132 mdd->mdd_sync_permission = 1;
134 dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf);
136 /* we are using service name but not mdd obd name
137 * for compatibility reasons.
138 * It is passed from MDT in lustre_cfg[2] buffer */
139 rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2));
141 obd_disconnect(mdd->mdd_child_exp);
146 static struct lu_device *mdd_device_fini(const struct lu_env *env,
149 struct mdd_device *mdd = lu2mdd_dev(d);
153 lu_dev_del_linkage(d->ld_site, d);
155 rc = mdd_procfs_fini(mdd);
157 CERROR("proc fini error %d \n", rc);
163 static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
164 struct llog_rec_hdr *hdr, void *data)
166 struct mdd_device *mdd = (struct mdd_device *)data;
167 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
169 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
170 LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
173 "seeing record at index %d/%d/"LPU64" t=%x %.*s in log"
174 DOSTID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index,
175 rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen,
176 rec->cr.cr_name, POSTID(&llh->lgh_id.lgl_oi));
178 mdd->mdd_cl.mc_index = rec->cr.cr_index;
179 return LLOG_PROC_BREAK;
182 static int changelog_user_init_cb(const struct lu_env *env,
183 struct llog_handle *llh,
184 struct llog_rec_hdr *hdr, void *data)
186 struct mdd_device *mdd = (struct mdd_device *)data;
187 struct llog_changelog_user_rec *rec =
188 (struct llog_changelog_user_rec *)hdr;
190 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
191 LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
193 CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
194 " in log "DOSTID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
195 rec->cur_id, rec->cur_endrec, POSTID(&llh->lgh_id.lgl_oi));
197 spin_lock(&mdd->mdd_cl.mc_user_lock);
198 mdd->mdd_cl.mc_lastuser = rec->cur_id;
199 if (rec->cur_endrec > mdd->mdd_cl.mc_index)
200 mdd->mdd_cl.mc_index = rec->cur_endrec;
201 spin_unlock(&mdd->mdd_cl.mc_user_lock);
203 return LLOG_PROC_BREAK;
206 static int llog_changelog_cancel_cb(const struct lu_env *env,
207 struct llog_handle *llh,
208 struct llog_rec_hdr *hdr, void *data)
210 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
211 struct llog_cookie cookie;
212 long long endrec = *(long long *)data;
217 /* This is always a (sub)log, not the catalog */
218 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
220 if (rec->cr.cr_index > endrec)
221 /* records are in order, so we're done */
222 RETURN(LLOG_PROC_BREAK);
224 cookie.lgc_lgl = llh->lgh_id;
225 cookie.lgc_index = hdr->lrh_index;
227 /* cancel them one at a time. I suppose we could store up the cookies
228 * and cancel them all at once; probably more efficient, but this is
229 * done as a user call, so who cares... */
230 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
232 RETURN(rc < 0 ? rc : 0);
235 static int llog_changelog_cancel(const struct lu_env *env,
236 struct llog_ctxt *ctxt,
237 struct lov_stripe_md *lsm, int count,
238 struct llog_cookie *cookies, int flags)
240 struct llog_handle *cathandle = ctxt->loc_handle;
245 /* This should only be called with the catalog handle */
246 LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
248 rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
249 (void *)cookies, 0, 0);
251 /* 0 or 1 means we're done */
254 CERROR("%s: cancel idx %u of catalog "DOSTID" rc=%d\n",
255 ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
256 POSTID(&cathandle->lgh_id.lgl_oi), rc);
261 static struct llog_operations changelog_orig_logops;
263 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
265 static int mdd_changelog_llog_init(const struct lu_env *env,
266 struct mdd_device *mdd)
268 struct obd_device *obd = mdd2obd_dev(mdd);
269 struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
274 /* LU-2844 mdd setup failure should not cause umount oops */
275 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT))
278 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
279 obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
280 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
281 obd, &changelog_orig_logops);
283 CERROR("%s: changelog llog setup failed: rc = %d\n",
288 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
291 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
294 GOTO(out_cleanup, rc);
296 rc = llog_cat_init_and_process(env, ctxt->loc_handle);
300 rc = llog_cat_reverse_process(env, ctxt->loc_handle,
301 changelog_init_cb, mdd);
304 CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
309 CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
310 mdd->mdd_cl.mc_index);
312 /* setup user changelog */
313 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
314 obd, &changelog_orig_logops);
316 CERROR("%s: changelog users llog setup failed: rc = %d\n",
321 uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
324 rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
327 GOTO(out_ucleanup, rc);
329 uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
330 uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
332 rc = llog_cat_init_and_process(env, uctxt->loc_handle);
334 GOTO(out_uclose, rc);
336 rc = llog_cat_reverse_process(env, uctxt->loc_handle,
337 changelog_user_init_cb, mdd);
339 CERROR("%s: changelog user init failed: rc = %d\n",
341 GOTO(out_uclose, rc);
344 /* If we have registered users, assume we want changelogs on */
345 if (mdd->mdd_cl.mc_lastuser > 0) {
346 rc = mdd_changelog_on(env, mdd, 1);
348 GOTO(out_uclose, rc);
351 llog_ctxt_put(uctxt);
354 llog_cat_close(env, uctxt->loc_handle);
356 llog_cleanup(env, uctxt);
358 llog_cat_close(env, ctxt->loc_handle);
360 llog_cleanup(env, ctxt);
364 static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
366 struct obd_device *obd = mdd2obd_dev(mdd);
369 mdd->mdd_cl.mc_index = 0;
370 spin_lock_init(&mdd->mdd_cl.mc_lock);
371 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
372 mdd->mdd_cl.mc_flags = 0; /* off by default */
373 mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
374 spin_lock_init(&mdd->mdd_cl.mc_user_lock);
375 mdd->mdd_cl.mc_lastuser = 0;
377 rc = mdd_changelog_llog_init(env, mdd);
379 CERROR("%s: changelog setup during init failed: rc = %d\n",
381 mdd->mdd_cl.mc_flags |= CLM_ERR;
387 static void mdd_changelog_fini(const struct lu_env *env,
388 struct mdd_device *mdd)
390 struct obd_device *obd = mdd2obd_dev(mdd);
391 struct llog_ctxt *ctxt;
393 mdd->mdd_cl.mc_flags = 0;
395 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
397 llog_cat_close(env, ctxt->loc_handle);
398 llog_cleanup(env, ctxt);
400 ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
402 llog_cat_close(env, ctxt->loc_handle);
403 llog_cleanup(env, ctxt);
407 int mdd_changelog_write_header(const struct lu_env *env,
408 struct mdd_device *mdd, int markerflags);
410 /* Start / stop recording */
411 int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
415 if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
416 LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
417 if (mdd->mdd_cl.mc_flags & CLM_ERR) {
418 CERROR("Changelogs cannot be enabled due to error "
419 "condition (see %s log).\n",
420 mdd2obd_dev(mdd)->obd_name);
423 spin_lock(&mdd->mdd_cl.mc_lock);
424 mdd->mdd_cl.mc_flags |= CLM_ON;
425 spin_unlock(&mdd->mdd_cl.mc_lock);
426 rc = mdd_changelog_write_header(env, mdd, CLM_START);
428 } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
429 LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
430 rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
431 spin_lock(&mdd->mdd_cl.mc_lock);
432 mdd->mdd_cl.mc_flags &= ~CLM_ON;
433 spin_unlock(&mdd->mdd_cl.mc_lock);
438 /** Remove entries with indicies up to and including \a endrec from the
444 int mdd_changelog_llog_cancel(const struct lu_env *env,
445 struct mdd_device *mdd, long long endrec)
447 struct obd_device *obd = mdd2obd_dev(mdd);
448 struct llog_ctxt *ctxt;
449 long long unsigned cur;
452 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
456 spin_lock(&mdd->mdd_cl.mc_lock);
457 cur = (long long)mdd->mdd_cl.mc_index;
458 spin_unlock(&mdd->mdd_cl.mc_lock);
462 /* purge to "0" is shorthand for everything */
466 /* If purging all records, write a header entry so we don't have an
467 empty catalog and we're sure to have a valid starting index next
468 time. In case of crash, we just restart with old log so we're
471 /* XXX: transaction is started by llog itself */
472 rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
477 /* Some records were purged, so reset repeat-access time (so we
478 record new mtime update records, so users can see a file has been
479 changed since the last purge) */
480 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
482 rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
488 /** Add a CL_MARK record to the changelog
490 * \param markerflags - CLM_*
493 int mdd_changelog_write_header(const struct lu_env *env,
494 struct mdd_device *mdd, int markerflags)
496 struct obd_device *obd = mdd2obd_dev(mdd);
497 struct llog_changelog_rec *rec;
499 struct llog_ctxt *ctxt;
501 int len = strlen(obd->obd_name);
506 if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
507 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
511 reclen = llog_data_len(sizeof(*rec) + len);
512 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
513 if (buf->lb_buf == NULL)
517 rec->cr.cr_flags = CLF_VERSION;
518 rec->cr.cr_type = CL_MARK;
519 rec->cr.cr_namelen = len;
520 memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
521 /* Status and action flags */
522 rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
523 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
524 rec->cr_hdr.lrh_type = CHANGELOG_REC;
525 rec->cr.cr_time = cl_time();
526 spin_lock(&mdd->mdd_cl.mc_lock);
527 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
528 spin_unlock(&mdd->mdd_cl.mc_lock);
530 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
533 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
538 /* assume on or off event; reset repeat-access time */
539 mdd->mdd_cl.mc_starttime = cfs_time_current_64();
544 * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
545 * are valid. We also check if object with passed fid exists or not.
547 static int obf_lookup(const struct lu_env *env, struct md_object *p,
548 const struct lu_name *lname, struct lu_fid *f,
549 struct md_op_spec *spec)
551 char *name = (char *)lname->ln_name;
552 struct mdd_device *mdd = mdo2mdd(p);
553 struct mdd_object *child;
559 sscanf(name, SFID, RFID(f));
560 if (!fid_is_sane(f)) {
561 CWARN("%s: bad FID format [%s], should be "DFID"\n",
562 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
563 (__u64)FID_SEQ_NORMAL, 1, 0);
564 GOTO(out, rc = -EINVAL);
567 if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) &&
568 !fid_seq_is_dot(f->f_seq)) {
569 CWARN("%s: "DFID" is invalid, sequence should be "
570 ">= "LPX64" or within ["LPX64","LPX64"].\n",
571 mdd2obd_dev(mdd)->obd_name, PFID(f),
572 (__u64)FID_SEQ_NORMAL, (__u64)FID_SEQ_IGIF,
573 (__u64)FID_SEQ_IGIF_MAX);
574 GOTO(out, rc = -EINVAL);
577 /* Check if object with this fid exists */
578 child = mdd_object_find(env, mdd, f);
582 GOTO(out, rc = PTR_ERR(child));
584 if (mdd_object_exists(child) == 0)
587 mdd_object_put(env, child);
593 static int obf_create(const struct lu_env *env, struct md_object *pobj,
594 const struct lu_name *lname, struct md_object *child,
595 struct md_op_spec *spec, struct md_attr* ma)
600 static int obf_rename(const struct lu_env *env,
601 struct md_object *src_pobj, struct md_object *tgt_pobj,
602 const struct lu_fid *lf, const struct lu_name *lsname,
603 struct md_object *tobj, const struct lu_name *ltname,
609 static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
610 struct md_object *src_obj, const struct lu_name *lname,
616 static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
617 struct md_object *cobj, const struct lu_name *lname,
618 struct md_attr *ma, int no_name)
623 static struct md_dir_operations mdd_obf_dir_ops = {
624 .mdo_lookup = obf_lookup,
625 .mdo_create = obf_create,
626 .mdo_rename = obf_rename,
627 .mdo_link = obf_link,
628 .mdo_unlink = obf_unlink
631 static struct md_object *mdo_locate(const struct lu_env *env,
632 struct md_device *md,
633 const struct lu_fid *fid)
635 struct lu_object *obj;
636 struct md_object *mdo;
638 obj = lu_object_find(env, &md->md_lu_dev, fid, NULL);
640 obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type);
641 LASSERT(obj != NULL);
644 mdo = ERR_PTR(PTR_ERR(obj));
650 * Create special in-memory "fid" object for open-by-fid.
652 static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
654 struct md_object *mdo;
655 struct mdd_object *mdd_obf;
656 struct lu_fid fid = LU_OBF_FID;
659 rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre),
660 mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid);
664 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
666 RETURN(PTR_ERR(mdo));
668 LASSERT(lu_object_exists(&mdo->mo_lu));
670 mdd_obf = md2mdd_obj(mdo);
671 mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
672 m->mdd_dot_lustre_objs.mdd_obf = mdd_obf;
677 /** Setup ".lustre" directory object */
678 static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
680 struct md_object *mdo;
685 /* Create ".lustre" directory in ROOT. */
686 fid = LU_DOT_LUSTRE_FID;
687 rc = mdd_local_file_create(env, m, &m->mdd_root_fid,
689 S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO,
693 mdo = mdo_locate(env, &m->mdd_md_dev, &fid);
695 RETURN(PTR_ERR(mdo));
696 LASSERT(lu_object_exists(&mdo->mo_lu));
698 m->mdd_dot_lustre = md2mdd_obj(mdo);
700 rc = mdd_obf_setup(env, m);
702 CERROR("%s: error initializing \"fid\" object: rc = %d.\n",
703 mdd2obd_dev(m)->obd_name, rc);
708 mdd_object_put(env, m->mdd_dot_lustre);
709 m->mdd_dot_lustre = NULL;
714 static struct llog_operations hsm_actions_logops;
717 * set llog methods and create LLOG_AGENT_ORIG_CTXT llog
718 * object in obd_device
720 static int mdd_hsm_actions_llog_init(const struct lu_env *env,
721 struct mdd_device *m)
723 struct obd_device *obd = mdd2obd_dev(m);
724 struct llog_ctxt *ctxt = NULL;
728 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
729 obd->obd_lvfs_ctxt.dt = m->mdd_bottom;
731 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT,
732 obd, &hsm_actions_logops);
734 CERROR("%s: hsm actions llog setup failed: rc = %d\n",
739 ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
742 rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
745 CERROR("%s: hsm actions llog open_create failed: rc = %d\n",
747 GOTO(out_cleanup, rc);
750 rc = llog_cat_init_and_process(env, ctxt->loc_handle);
758 llog_cat_close(env, ctxt->loc_handle);
759 ctxt->loc_handle = NULL;
761 llog_cleanup(env, ctxt);
767 * cleanup the context created by llog_setup_named()
769 static int mdd_hsm_actions_llog_fini(const struct lu_env *env,
770 struct mdd_device *m)
772 struct obd_device *obd = mdd2obd_dev(m);
773 struct llog_ctxt *lctxt;
776 lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT);
778 llog_cat_close(env, lctxt->loc_handle);
779 lctxt->loc_handle = NULL;
780 llog_cleanup(env, lctxt);
786 static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m,
787 struct lustre_cfg *cfg)
789 lfsck_degister(env, m->mdd_bottom);
790 mdd_hsm_actions_llog_fini(env, m);
791 mdd_changelog_fini(env, m);
792 orph_index_fini(env, m);
793 if (m->mdd_dot_lustre_objs.mdd_obf)
794 mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf);
795 if (m->mdd_dot_lustre)
796 mdd_object_put(env, m->mdd_dot_lustre);
797 if (m->mdd_los != NULL)
798 local_oid_storage_fini(env, m->mdd_los);
799 lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0);
801 if (m->mdd_child_exp)
802 obd_disconnect(m->mdd_child_exp);
805 static int mdd_process_config(const struct lu_env *env,
806 struct lu_device *d, struct lustre_cfg *cfg)
808 struct mdd_device *m = lu2mdd_dev(d);
809 struct dt_device *dt = m->mdd_child;
810 struct lu_device *next = &dt->dd_lu_dev;
814 switch (cfg->lcfg_command) {
816 struct lprocfs_static_vars lvars;
818 lprocfs_mdd_init_vars(&lvars);
819 rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m);
820 if (rc > 0 || rc == -ENOSYS)
821 /* we don't understand; pass it on */
822 rc = next->ld_ops->ldo_process_config(env, next, cfg);
826 rc = next->ld_ops->ldo_process_config(env, next, cfg);
829 dt_conf_get(env, dt, &m->mdd_dt_conf);
832 rc = next->ld_ops->ldo_process_config(env, next, cfg);
833 lu_dev_del_linkage(d->ld_site, d);
834 mdd_device_shutdown(env, m, cfg);
837 rc = next->ld_ops->ldo_process_config(env, next, cfg);
844 static int mdd_recovery_complete(const struct lu_env *env,
847 struct mdd_device *mdd = lu2mdd_dev(d);
848 struct lu_device *next;
852 LASSERT(mdd != NULL);
853 next = &mdd->mdd_child->dd_lu_dev;
855 /* XXX: orphans handling. */
856 __mdd_orphan_cleanup(env, mdd);
857 rc = next->ld_ops->ldo_recovery_complete(env, next);
862 int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd,
863 const struct lu_fid *pfid, const char *name, __u32 mode,
866 struct dt_object *parent, *dto;
871 LASSERT(!fid_is_zero(pfid));
872 parent = dt_locate(env, mdd->mdd_bottom, pfid);
873 if (unlikely(IS_ERR(parent)))
874 RETURN(PTR_ERR(parent));
876 /* create local file/dir, if @fid is passed then try to use it */
877 if (fid_is_zero(fid))
878 dto = local_file_find_or_create(env, mdd->mdd_los, parent,
881 dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom,
885 GOTO(out_put, rc = PTR_ERR(dto));
886 *fid = *lu_object_fid(&dto->do_lu);
887 /* since stack is not fully set up the local_storage uses own stack
888 * and we should drop its object from cache */
889 lu_object_put_nocache(env, &dto->do_lu);
892 lu_object_put(env, &parent->do_lu);
896 static int mdd_lfsck_out_notify(const struct lu_env *env, void *data,
897 enum lfsck_events event)
902 static int mdd_prepare(const struct lu_env *env,
903 struct lu_device *pdev,
904 struct lu_device *cdev)
906 struct mdd_device *mdd = lu2mdd_dev(cdev);
907 struct lu_device *next = &mdd->mdd_child->dd_lu_dev;
913 rc = next->ld_ops->ldo_prepare(env, cdev, next);
917 /* Setup local dirs */
918 fid.f_seq = FID_SEQ_LOCAL_NAME;
921 rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid,
926 rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid);
931 if (mdd_seq_site(mdd)->ss_node_id == 0) {
932 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
933 mdd_root_dir_name, S_IFDIR |
934 S_IRUGO | S_IWUSR | S_IXUGO, &fid);
936 CERROR("%s: create root fid failed: rc = %d\n",
937 mdd2obd_dev(mdd)->obd_name, rc);
941 mdd->mdd_root_fid = fid;
942 rc = mdd_dot_lustre_setup(env, mdd);
944 CERROR("%s: initializing .lustre failed: rc = %d\n",
945 mdd2obd_dev(mdd)->obd_name, rc);
949 rc = mdd_compat_fixes(env, mdd);
954 /* Normal client usually send root access to MDT0 directly,
955 * the root FID on non-MDT0 will only be used by echo client. */
956 mdd->mdd_root_fid = fid;
959 rc = orph_index_init(env, mdd);
963 rc = mdd_changelog_init(env, mdd);
965 CERROR("%s: failed to initialize changelog: rc = %d\n",
966 mdd2obd_dev(mdd)->obd_name, rc);
970 rc = mdd_hsm_actions_llog_init(env, mdd);
972 GOTO(out_changelog, rc);
974 rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child,
975 mdd2obd_dev(mdd), mdd_lfsck_out_notify,
978 CERROR("%s: failed to initialize lfsck: rc = %d\n",
979 mdd2obd_dev(mdd)->obd_name, rc);
984 mdd_hsm_actions_llog_fini(env, mdd);
986 mdd_changelog_fini(env, mdd);
988 orph_index_fini(env, mdd);
990 if (mdd_seq_site(mdd)->ss_node_id == 0) {
991 mdd_object_put(env, mdd->mdd_dot_lustre);
992 mdd->mdd_dot_lustre = NULL;
993 mdd_object_put(env, mdd->mdd_dot_lustre_objs.mdd_obf);
994 mdd->mdd_dot_lustre_objs.mdd_obf = NULL;
997 local_oid_storage_fini(env, mdd->mdd_los);
1003 const struct lu_device_operations mdd_lu_ops = {
1004 .ldo_object_alloc = mdd_object_alloc,
1005 .ldo_process_config = mdd_process_config,
1006 .ldo_recovery_complete = mdd_recovery_complete,
1007 .ldo_prepare = mdd_prepare,
1010 static int mdd_root_get(const struct lu_env *env,
1011 struct md_device *m, struct lu_fid *f)
1013 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1016 *f = mdd->mdd_root_fid;
1021 * No permission check is needed.
1023 static int mdd_statfs(const struct lu_env *env, struct md_device *m,
1024 struct obd_statfs *sfs)
1026 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1031 rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs);
1033 sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX);
1039 * No permission check is needed.
1041 static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m,
1042 int mode, unsigned long timeout, __u32 alg,
1043 struct lustre_capa_key *keys)
1045 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1049 /* need barrier for mds_capa_keys access. */
1051 rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode,
1052 timeout, alg, keys);
1056 static int mdd_maxeasize_get(const struct lu_env *env, struct md_device *m,
1059 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1062 *easize = mdd->mdd_dt_conf.ddp_max_ea_size;
1067 static int mdd_update_capa_key(const struct lu_env *env,
1068 struct md_device *m,
1069 struct lustre_capa_key *key)
1071 /* we do not support capabilities ... */
1075 static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
1078 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
1080 *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx);
1081 return (*h == NULL ? -ENOENT : 0);
1084 static struct lu_device *mdd_device_free(const struct lu_env *env,
1085 struct lu_device *lu)
1087 struct mdd_device *m = lu2mdd_dev(lu);
1090 LASSERT(atomic_read(&lu->ld_ref) == 0);
1091 md_device_fini(&m->mdd_md_dev);
1096 static struct lu_device *mdd_device_alloc(const struct lu_env *env,
1097 struct lu_device_type *t,
1098 struct lustre_cfg *lcfg)
1100 struct lu_device *l;
1101 struct mdd_device *m;
1105 l = ERR_PTR(-ENOMEM);
1110 md_device_init(&m->mdd_md_dev, t);
1111 rc = mdd_init0(env, m, t, lcfg);
1113 mdd_device_free(env, l);
1122 * we use exports to track all mdd users
1124 static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1125 struct obd_device *obd, struct obd_uuid *cluuid,
1126 struct obd_connect_data *data, void *localdata)
1128 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1129 struct lustre_handle conn;
1133 CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects);
1135 rc = class_connect(&conn, obd, cluuid);
1139 *exp = class_conn2export(&conn);
1141 /* Why should there ever be more than 1 connect? */
1142 LASSERT(mdd->mdd_connects == 0);
1143 mdd->mdd_connects++;
1149 * once last export (we don't count self-export) disappeared
1150 * mdd can be released
1152 static int mdd_obd_disconnect(struct obd_export *exp)
1154 struct obd_device *obd = exp->exp_obd;
1155 struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev);
1156 int rc, release = 0;
1159 mdd->mdd_connects--;
1160 if (mdd->mdd_connects == 0)
1163 rc = class_disconnect(exp);
1165 if (rc == 0 && release)
1166 class_manual_cleanup(obd);
1170 static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1171 __u32 keylen, void *key, __u32 *vallen, void *val,
1172 struct lov_stripe_md *lsm)
1176 if (KEY_IS(KEY_OSP_CONNECTED)) {
1177 struct obd_device *obd = exp->exp_obd;
1178 struct mdd_device *mdd;
1180 if (!obd->obd_set_up || obd->obd_stopping)
1183 mdd = lu2mdd_dev(obd->obd_lu_dev);
1185 rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen,
1193 static struct obd_ops mdd_obd_device_ops = {
1194 .o_owner = THIS_MODULE,
1195 .o_connect = mdd_obd_connect,
1196 .o_disconnect = mdd_obd_disconnect,
1197 .o_get_info = mdd_obd_get_info,
1200 static int mdd_changelog_user_register(const struct lu_env *env,
1201 struct mdd_device *mdd, int *id)
1203 struct llog_ctxt *ctxt;
1204 struct llog_changelog_user_rec *rec;
1208 ctxt = llog_get_context(mdd2obd_dev(mdd),
1209 LLOG_CHANGELOG_USER_ORIG_CTXT);
1215 llog_ctxt_put(ctxt);
1219 /* Assume we want it on since somebody registered */
1220 rc = mdd_changelog_on(env, mdd, 1);
1224 rec->cur_hdr.lrh_len = sizeof(*rec);
1225 rec->cur_hdr.lrh_type = CHANGELOG_USER_REC;
1226 spin_lock(&mdd->mdd_cl.mc_user_lock);
1227 if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) {
1228 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1229 CERROR("Maximum number of changelog users exceeded!\n");
1230 GOTO(out, rc = -EOVERFLOW);
1232 *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser;
1233 rec->cur_endrec = mdd->mdd_cl.mc_index;
1234 spin_unlock(&mdd->mdd_cl.mc_user_lock);
1236 rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL);
1238 CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id);
1241 llog_ctxt_put(ctxt);
1245 struct mdd_changelog_user_data {
1246 __u64 mcud_endrec; /**< purge record for this user */
1247 __u64 mcud_minrec; /**< lowest changelog recno still referenced */
1249 __u32 mcud_minid; /**< user id with lowest rec reference */
1250 __u32 mcud_usercount;
1251 unsigned int mcud_found:1;
1253 #define MCUD_UNREGISTER -1LL
1256 * 1. Find the smallest record everyone is willing to purge
1257 * 2. Update the last purgeable record for this user
1259 static int mdd_changelog_user_purge_cb(const struct lu_env *env,
1260 struct llog_handle *llh,
1261 struct llog_rec_hdr *hdr, void *data)
1263 struct llog_changelog_user_rec *rec;
1264 struct mdd_changelog_user_data *mcud = data;
1269 LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
1271 rec = (struct llog_changelog_user_rec *)hdr;
1273 mcud->mcud_usercount++;
1275 /* If we have a new endrec for this id, use it for the following
1276 min check instead of its old value */
1277 if (rec->cur_id == mcud->mcud_id)
1278 rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec);
1280 /* Track the minimum referenced record */
1281 if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) {
1282 mcud->mcud_minid = rec->cur_id;
1283 mcud->mcud_minrec = rec->cur_endrec;
1286 if (rec->cur_id != mcud->mcud_id)
1289 /* Update this user's record */
1290 mcud->mcud_found = 1;
1292 /* Special case: unregister this user */
1293 if (mcud->mcud_endrec == MCUD_UNREGISTER) {
1294 struct llog_cookie cookie;
1296 cookie.lgc_lgl = llh->lgh_id;
1297 cookie.lgc_index = hdr->lrh_index;
1299 rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle,
1302 mcud->mcud_usercount--;
1307 /* Update the endrec */
1308 CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n",
1309 mcud->mcud_id, rec->cur_endrec);
1311 /* hdr+1 is loc of data */
1312 hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail);
1313 rc = llog_write(env, llh, hdr, NULL, 0, (void *)(hdr + 1),
1319 static int mdd_changelog_user_purge(const struct lu_env *env,
1320 struct mdd_device *mdd, int id,
1323 struct mdd_changelog_user_data data;
1324 struct llog_ctxt *ctxt;
1328 CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec);
1331 data.mcud_minid = 0;
1332 data.mcud_minrec = 0;
1333 data.mcud_usercount = 0;
1334 data.mcud_endrec = endrec;
1335 spin_lock(&mdd->mdd_cl.mc_lock);
1336 endrec = mdd->mdd_cl.mc_index;
1337 spin_unlock(&mdd->mdd_cl.mc_lock);
1338 if ((data.mcud_endrec == 0) ||
1339 ((data.mcud_endrec > endrec) &&
1340 (data.mcud_endrec != MCUD_UNREGISTER)))
1341 data.mcud_endrec = endrec;
1343 ctxt = llog_get_context(mdd2obd_dev(mdd),
1344 LLOG_CHANGELOG_USER_ORIG_CTXT);
1348 LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
1350 rc = llog_cat_process(env, ctxt->loc_handle,
1351 mdd_changelog_user_purge_cb, (void *)&data,
1353 if ((rc >= 0) && (data.mcud_minrec > 0)) {
1354 CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64
1355 ", referenced by "CHANGELOG_USER_PREFIX"%d\n",
1356 data.mcud_minrec, data.mcud_minid);
1357 rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec);
1359 CWARN("Could not determine changelog records to purge; rc=%d\n",
1363 llog_ctxt_put(ctxt);
1365 if (!data.mcud_found) {
1366 CWARN("No entry for user %d. Last changelog reference is "
1367 LPD64" by changelog user %d\n", data.mcud_id,
1368 data.mcud_minrec, data.mcud_minid);
1372 if (!rc && data.mcud_usercount == 0)
1373 /* No more users; turn changelogs off */
1374 rc = mdd_changelog_on(env, mdd, 0);
1380 * May be called remotely from mdt_iocontrol_handle or locally from
1381 * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce
1382 * an obd_ioctl_data format (but local ioctl handler does).
1384 * \param len - data len
1385 * \param karg - ioctl data, in kernel space
1387 static int mdd_iocontrol(const struct lu_env *env, struct md_device *m,
1388 unsigned int cmd, int len, void *karg)
1390 struct mdd_device *mdd;
1391 struct obd_ioctl_data *data = karg;
1395 mdd = lu2mdd_dev(&m->md_lu_dev);
1397 /* Doesn't use obd_ioctl_data */
1399 case OBD_IOC_CHANGELOG_CLEAR: {
1400 struct changelog_setinfo *cs = karg;
1401 rc = mdd_changelog_user_purge(env, mdd, cs->cs_id,
1405 case OBD_IOC_GET_MNTOPT: {
1406 mntopt_t *mntopts = (mntopt_t *)karg;
1407 *mntopts = mdd->mdd_dt_conf.ddp_mntopts;
1410 case OBD_IOC_START_LFSCK: {
1411 rc = lfsck_start(env, mdd->mdd_bottom,
1412 (struct lfsck_start_param *)karg);
1415 case OBD_IOC_STOP_LFSCK: {
1416 rc = lfsck_stop(env, mdd->mdd_bottom,
1417 (struct lfsck_stop *)karg);
1422 /* Below ioctls use obd_ioctl_data */
1423 if (len != sizeof(*data)) {
1424 CERROR("Bad ioctl size %d\n", len);
1427 if (data->ioc_version != OBD_IOCTL_VERSION) {
1428 CERROR("Bad magic %x != %x\n", data->ioc_version,
1434 case OBD_IOC_CHANGELOG_REG:
1435 rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1);
1437 case OBD_IOC_CHANGELOG_DEREG:
1438 rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1,
1448 /* type constructor/destructor: mdd_type_init, mdd_type_fini */
1449 LU_TYPE_INIT_FINI(mdd, &mdd_thread_key);
1451 const struct md_device_operations mdd_ops = {
1452 .mdo_statfs = mdd_statfs,
1453 .mdo_root_get = mdd_root_get,
1454 .mdo_init_capa_ctxt = mdd_init_capa_ctxt,
1455 .mdo_update_capa_key= mdd_update_capa_key,
1456 .mdo_llog_ctxt_get = mdd_llog_ctxt_get,
1457 .mdo_iocontrol = mdd_iocontrol,
1458 .mdo_maxeasize_get = mdd_maxeasize_get,
1461 static struct lu_device_type_operations mdd_device_type_ops = {
1462 .ldto_init = mdd_type_init,
1463 .ldto_fini = mdd_type_fini,
1465 .ldto_start = mdd_type_start,
1466 .ldto_stop = mdd_type_stop,
1468 .ldto_device_alloc = mdd_device_alloc,
1469 .ldto_device_free = mdd_device_free,
1471 .ldto_device_fini = mdd_device_fini
1474 static struct lu_device_type mdd_device_type = {
1475 .ldt_tags = LU_DEVICE_MD,
1476 .ldt_name = LUSTRE_MDD_NAME,
1477 .ldt_ops = &mdd_device_type_ops,
1478 .ldt_ctx_tags = LCT_MD_THREAD
1481 /* context key constructor: mdd_key_init */
1482 LU_KEY_INIT(mdd, struct mdd_thread_info);
1484 static void mdd_key_fini(const struct lu_context *ctx,
1485 struct lu_context_key *key, void *data)
1487 struct mdd_thread_info *info = data;
1489 lu_buf_free(&info->mti_big_buf);
1490 lu_buf_free(&info->mti_link_buf);
1495 /* context key: mdd_thread_key */
1496 LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD);
1498 static int __init mdd_mod_init(void)
1500 struct lprocfs_static_vars lvars;
1503 lprocfs_mdd_init_vars(&lvars);
1505 rc = lu_kmem_init(mdd_caches);
1509 changelog_orig_logops = llog_osd_ops;
1510 changelog_orig_logops.lop_cancel = llog_changelog_cancel;
1511 changelog_orig_logops.lop_add = llog_cat_add_rec;
1512 changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
1514 hsm_actions_logops = llog_osd_ops;
1515 hsm_actions_logops.lop_add = llog_cat_add_rec;
1516 hsm_actions_logops.lop_declare_add = llog_cat_declare_add_rec;
1518 rc = class_register_type(&mdd_obd_device_ops, NULL, NULL,
1519 #ifndef HAVE_ONLY_PROCFS_SEQ
1522 LUSTRE_MDD_NAME, &mdd_device_type);
1524 lu_kmem_fini(mdd_caches);
1528 static void __exit mdd_mod_exit(void)
1530 class_unregister_type(LUSTRE_MDD_NAME);
1531 lu_kmem_fini(mdd_caches);
1534 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1535 MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")");
1536 MODULE_LICENSE("GPL");
1538 cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit);