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, 2014, 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_dir.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@intel.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <obd_class.h>
46 #include <obd_support.h>
47 #include <lustre_mds.h>
48 #include <lustre_fid.h>
50 #include "mdd_internal.h"
52 static const char dot[] = ".";
53 static const char dotdot[] = "..";
55 static struct lu_name lname_dotdot = {
61 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
63 if (!lu_name_is_valid(ln))
65 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
71 /* Get FID from name and parent */
73 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
74 const struct lu_attr *pattr, const struct lu_name *lname,
75 struct lu_fid* fid, int mask)
77 const char *name = lname->ln_name;
78 const struct dt_key *key = (const struct dt_key *)name;
79 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
80 struct mdd_device *m = mdo2mdd(pobj);
81 struct dt_object *dir = mdd_object_child(mdd_obj);
85 if (unlikely(mdd_is_dead_obj(mdd_obj)))
88 if (mdd_object_remote(mdd_obj)) {
89 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
90 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
91 } else if (!mdd_object_exists(mdd_obj)) {
95 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
100 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
101 dt_try_as_dir(env, dir)))
102 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
109 int mdd_lookup(const struct lu_env *env,
110 struct md_object *pobj, const struct lu_name *lname,
111 struct lu_fid *fid, struct md_op_spec *spec)
113 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
117 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
121 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
122 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
126 static inline int mdd_parent_fid(const struct lu_env *env,
127 struct mdd_object *obj,
128 const struct lu_attr *attr,
131 return __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
135 * For root fid use special function, which does not compare version component
136 * of fid. Version component is different for root fids on all MDTs.
138 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
140 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
141 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
145 * return 1: if lf is the fid of the ancestor of p1;
148 * return -EREMOTE: if remote object is found, in this
149 * case fid of remote object is saved to @pf;
151 * otherwise: values < 0, errors.
153 static int mdd_is_parent(const struct lu_env *env,
154 struct mdd_device *mdd,
155 struct mdd_object *p1,
156 const struct lu_attr *attr,
157 const struct lu_fid *lf,
160 struct mdd_object *parent = NULL;
165 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
166 pfid = &mdd_env_info(env)->mti_fid;
168 /* Check for root first. */
169 if (mdd_is_root(mdd, mdo2fid(p1)))
173 /* this is done recursively */
174 rc = mdd_parent_fid(env, p1, attr, pfid);
177 if (mdd_is_root(mdd, pfid))
179 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
181 if (lu_fid_eq(pfid, lf))
184 mdd_object_put(env, parent);
186 parent = mdd_object_find(env, mdd, pfid);
187 if (IS_ERR(parent)) {
188 GOTO(out, rc = PTR_ERR(parent));
189 } else if (mdd_object_remote(parent)) {
190 /*FIXME: Because of the restriction of rename in Phase I.
191 * If the parent is remote, we just assumed lf is not the
192 * parent of P1 for now */
199 if (parent && !IS_ERR(parent))
200 mdd_object_put(env, parent);
205 * No permission check is needed.
207 * returns 1: if fid is ancestor of @mo;
208 * returns 0: if fid is not a ancestor of @mo;
210 * returns EREMOTE if remote object is found, fid of remote object is saved to
213 * returns < 0: if error
215 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
216 const struct lu_fid *fid, struct lu_fid *sfid)
218 struct mdd_device *mdd = mdo2mdd(mo);
219 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
223 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
226 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
230 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
234 } else if (rc == 1) {
235 /* found @fid is parent */
243 * Check that @dir contains no entries except (possibly) dot and dotdot.
248 * -ENOTDIR not a directory object
249 * -ENOTEMPTY not empty
253 static int mdd_dir_is_empty(const struct lu_env *env,
254 struct mdd_object *dir)
257 struct dt_object *obj;
258 const struct dt_it_ops *iops;
262 obj = mdd_object_child(dir);
263 if (!dt_try_as_dir(env, obj))
266 iops = &obj->do_index_ops->dio_it;
267 it = iops->init(env, obj, LUDA_64BITHASH);
269 result = iops->get(env, it, (const struct dt_key *)"");
272 for (result = 0, i = 0; result == 0 && i < 3; ++i)
273 result = iops->next(env, it);
276 else if (result == 1)
278 } else if (result == 0)
280 * Huh? Index contains no zero key?
287 result = PTR_ERR(it);
292 * Determine if the target object can be hard linked, and right now it only
293 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
294 * directory nlink count might exceed the maximum link count(see
295 * osd_object_ref_add), so it only check nlink for non-directories.
297 * \param[in] env thread environment
298 * \param[in] obj object being linked to
299 * \param[in] la attributes of \a obj
301 * \retval 0 if \a obj can be hard linked
302 * \retval negative error if \a obj is a directory or has too
305 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
306 const struct lu_attr *la)
308 struct mdd_device *m = mdd_obj2mdd_dev(obj);
313 /* Subdir count limitation can be broken through
314 * (see osd_object_ref_add), so only check non-directory here. */
315 if (!S_ISDIR(la->la_mode) &&
316 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
323 * Check whether it may create the cobj under the pobj.
325 * \param[in] env execution environment
326 * \param[in] pobj the parent directory
327 * \param[in] pattr the attribute of the parent directory
328 * \param[in] cobj the child to be created
329 * \param[in] check_perm if check WRITE|EXEC permission for parent
331 * \retval = 0 create the child under this dir is allowed
332 * \retval negative errno create the child under this dir is
335 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
336 const struct lu_attr *pattr, struct mdd_object *cobj,
339 struct mdd_thread_info *info = mdd_env_info(env);
344 if (cobj && mdd_object_exists(cobj))
347 if (mdd_is_dead_obj(pobj))
350 /* If the parent is a sub-stripe, check whether it is dead */
351 xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key));
352 rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV);
353 if (unlikely(rc > 0)) {
354 struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf;
356 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE &&
357 le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD)
363 rc = mdd_permission_internal_locked(env, pobj, pattr,
364 MAY_WRITE | MAY_EXEC,
370 * Check whether can unlink from the pobj in the case of "cobj == NULL".
372 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
373 const struct lu_attr *pattr, const struct lu_attr *attr)
378 if (mdd_is_dead_obj(pobj))
381 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
384 rc = mdd_permission_internal_locked(env, pobj, pattr,
385 MAY_WRITE | MAY_EXEC,
390 if (pattr->la_flags & LUSTRE_APPEND_FL)
397 * pobj == NULL is remote ops case, under such case, pobj's
398 * VTX feature has been checked already, no need check again.
400 static inline int mdd_is_sticky(const struct lu_env *env,
401 struct mdd_object *pobj,
402 const struct lu_attr *pattr,
403 struct mdd_object *cobj,
404 const struct lu_attr *cattr)
406 struct lu_ucred *uc = lu_ucred_assert(env);
409 LASSERT(pattr != NULL);
410 if (!(pattr->la_mode & S_ISVTX) ||
411 (pattr->la_uid == uc->uc_fsuid))
415 LASSERT(cattr != NULL);
416 if (cattr->la_uid == uc->uc_fsuid)
419 return !md_capable(uc, CFS_CAP_FOWNER);
422 static int mdd_may_delete_entry(const struct lu_env *env,
423 struct mdd_object *pobj,
424 const struct lu_attr *pattr,
429 LASSERT(pobj != NULL);
430 if (!mdd_object_exists(pobj))
433 if (mdd_is_dead_obj(pobj))
438 rc = mdd_permission_internal_locked(env, pobj, pattr,
439 MAY_WRITE | MAY_EXEC,
445 if (pattr->la_flags & LUSTRE_APPEND_FL)
452 * Check whether it may delete the cobj from the pobj.
455 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
456 const struct lu_attr *tpattr, struct mdd_object *tobj,
457 const struct lu_attr *tattr, const struct lu_attr *cattr,
458 int check_perm, int check_empty)
464 LASSERT(tpattr != NULL);
465 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
473 if (!mdd_object_exists(tobj))
476 if (mdd_is_dead_obj(tobj))
479 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
482 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
485 /* additional check the rename case */
487 if (S_ISDIR(cattr->la_mode)) {
488 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
490 if (!S_ISDIR(tattr->la_mode))
493 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
495 } else if (S_ISDIR(tattr->la_mode))
499 if (S_ISDIR(tattr->la_mode) && check_empty)
500 rc = mdd_dir_is_empty(env, tobj);
506 * Check whether it can create the link file(linked to @src_obj) under
507 * the target directory(@tgt_obj), and src_obj has been locked by
510 * \param[in] env execution environment
511 * \param[in] tgt_obj the target directory
512 * \param[in] tattr attributes of target directory
513 * \param[in] lname the link name
514 * \param[in] src_obj source object for link
515 * \param[in] cattr attributes for source object
517 * \retval = 0 it is allowed to create the link file under tgt_obj
518 * \retval negative error not allowed to create the link file
520 static int mdd_link_sanity_check(const struct lu_env *env,
521 struct mdd_object *tgt_obj,
522 const struct lu_attr *tattr,
523 const struct lu_name *lname,
524 struct mdd_object *src_obj,
525 const struct lu_attr *cattr)
527 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
531 if (!mdd_object_exists(src_obj))
534 if (mdd_is_dead_obj(src_obj))
537 /* Local ops, no lookup before link, check filename length here. */
538 rc = mdd_name_check(m, lname);
542 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
545 if (S_ISDIR(mdd_object_type(src_obj)))
548 LASSERT(src_obj != tgt_obj);
549 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
553 rc = __mdd_may_link(env, src_obj, cattr);
558 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
559 const char *name, struct thandle *handle)
561 struct dt_object *next = mdd_object_child(pobj);
565 if (dt_try_as_dir(env, next))
566 rc = dt_delete(env, next, (struct dt_key *)name, handle);
573 static int __mdd_index_insert_only(const struct lu_env *env,
574 struct mdd_object *pobj,
575 const struct lu_fid *lf, __u32 type,
577 struct thandle *handle)
579 struct dt_object *next = mdd_object_child(pobj);
583 if (dt_try_as_dir(env, next)) {
584 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
585 struct lu_ucred *uc = lu_ucred_check(env);
589 rec->rec_type = type;
590 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
591 rc = dt_insert(env, next, (const struct dt_rec *)rec,
592 (const struct dt_key *)name, handle,
600 /* insert named index, add reference if isdir */
601 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
602 const struct lu_fid *lf, __u32 type,
603 const char *name, struct thandle *handle)
608 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
609 if (rc == 0 && S_ISDIR(type)) {
610 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
611 mdo_ref_add(env, pobj, handle);
612 mdd_write_unlock(env, pobj);
618 /* delete named index, drop reference if isdir */
619 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
620 const char *name, int is_dir,
621 struct thandle *handle)
626 rc = __mdd_index_delete_only(env, pobj, name, handle);
627 if (rc == 0 && is_dir) {
628 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
629 mdo_ref_del(env, pobj, handle);
630 mdd_write_unlock(env, pobj);
636 static int mdd_llog_record_calc_size(const struct lu_env *env,
637 const struct lu_name *tname,
638 const struct lu_name *sname)
640 const struct lu_ucred *uc = lu_ucred(env);
641 enum changelog_rec_flags crf = 0;
642 size_t hdr_size = sizeof(struct llog_changelog_rec) -
643 sizeof(struct changelog_rec);
648 if (uc != NULL && uc->uc_jobid[0] != '\0')
651 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
652 (tname != NULL ? tname->ln_namelen : 0) +
653 (sname != NULL ? 1 + sname->ln_namelen : 0));
656 int mdd_declare_changelog_store(const struct lu_env *env,
657 struct mdd_device *mdd,
658 const struct lu_name *tname,
659 const struct lu_name *sname,
660 struct thandle *handle)
662 struct obd_device *obd = mdd2obd_dev(mdd);
663 struct llog_ctxt *ctxt;
664 struct llog_changelog_rec *rec;
670 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
673 reclen = mdd_llog_record_calc_size(env, tname, sname);
674 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
675 if (buf->lb_buf == NULL)
679 rec->cr_hdr.lrh_len = reclen;
680 rec->cr_hdr.lrh_type = CHANGELOG_REC;
682 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
686 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
692 /** Add a changelog entry \a rec to the changelog llog
695 * \param handle - currently ignored since llogs start their own transaction;
696 * this will hopefully be fixed in llog rewrite
699 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
700 struct llog_changelog_rec *rec, struct thandle *th)
702 struct obd_device *obd = mdd2obd_dev(mdd);
703 struct llog_ctxt *ctxt;
706 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
707 changelog_rec_varsize(&rec->cr));
709 /* llog_lvfs_write_rec sets the llog tail len */
710 rec->cr_hdr.lrh_type = CHANGELOG_REC;
711 rec->cr.cr_time = cl_time();
713 spin_lock(&mdd->mdd_cl.mc_lock);
714 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
715 * but as long as the MDD transactions are ordered correctly for e.g.
716 * rename conflicts, I don't think this should matter. */
717 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
718 spin_unlock(&mdd->mdd_cl.mc_lock);
720 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
724 /* nested journal transaction */
725 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
733 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
734 const struct lu_fid *sfid,
735 const struct lu_fid *spfid,
736 const struct lu_name *sname)
738 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
739 size_t extsize = sname->ln_namelen + 1;
741 LASSERT(sfid != NULL);
742 LASSERT(spfid != NULL);
743 LASSERT(sname != NULL);
745 rnm->cr_sfid = *sfid;
746 rnm->cr_spfid = *spfid;
748 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
749 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
750 rec->cr_namelen += extsize;
753 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
755 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
757 if (jobid == NULL || jobid[0] == '\0')
760 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
763 /** Store a namespace change changelog record
764 * If this fails, we must fail the whole transaction; we don't
765 * want the change to commit without the log entry.
766 * \param target - mdd_object of change
767 * \param tpfid - target parent dir/object fid
768 * \param sfid - source object fid
769 * \param spfid - source parent fid
770 * \param tname - target name string
771 * \param sname - source name string
772 * \param handle - transacion handle
774 int mdd_changelog_ns_store(const struct lu_env *env,
775 struct mdd_device *mdd,
776 enum changelog_rec_type type,
777 enum changelog_rec_flags crf,
778 struct mdd_object *target,
779 const struct lu_fid *tpfid,
780 const struct lu_fid *sfid,
781 const struct lu_fid *spfid,
782 const struct lu_name *tname,
783 const struct lu_name *sname,
784 struct thandle *handle)
786 const struct lu_ucred *uc = lu_ucred(env);
787 struct llog_changelog_rec *rec;
794 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
797 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
800 LASSERT(tpfid != NULL);
801 LASSERT(tname != NULL);
802 LASSERT(handle != NULL);
804 reclen = mdd_llog_record_calc_size(env, tname, sname);
805 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
806 if (buf->lb_buf == NULL)
812 if (uc != NULL && uc->uc_jobid[0] != '\0')
820 rec->cr.cr_flags = crf;
821 rec->cr.cr_type = (__u32)type;
822 rec->cr.cr_pfid = *tpfid;
823 rec->cr.cr_namelen = tname->ln_namelen;
824 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
826 if (crf & CLF_RENAME)
827 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
830 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
832 if (likely(target != NULL)) {
833 rec->cr.cr_tfid = *mdo2fid(target);
834 target->mod_cltime = cfs_time_current_64();
836 fid_zero(&rec->cr.cr_tfid);
839 rc = mdd_changelog_store(env, mdd, rec, handle);
841 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
842 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
849 static int __mdd_links_add(const struct lu_env *env,
850 struct mdd_object *mdd_obj,
851 struct linkea_data *ldata,
852 const struct lu_name *lname,
853 const struct lu_fid *pfid,
854 int first, int check)
858 if (ldata->ld_leh == NULL) {
859 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
863 rc = linkea_data_new(ldata,
864 &mdd_env_info(env)->mti_link_buf);
871 rc = linkea_links_find(ldata, lname, pfid);
872 if (rc && rc != -ENOENT)
878 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
879 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
883 linkea_add_buf(ldata, lname, tfid);
886 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
887 linkea_add_buf(ldata, lname, pfid);
889 return linkea_add_buf(ldata, lname, pfid);
892 static int __mdd_links_del(const struct lu_env *env,
893 struct mdd_object *mdd_obj,
894 struct linkea_data *ldata,
895 const struct lu_name *lname,
896 const struct lu_fid *pfid)
900 if (ldata->ld_leh == NULL) {
901 rc = mdd_links_read(env, mdd_obj, ldata);
906 rc = linkea_links_find(ldata, lname, pfid);
910 linkea_del_buf(ldata, lname);
914 static int mdd_linkea_prepare(const struct lu_env *env,
915 struct mdd_object *mdd_obj,
916 const struct lu_fid *oldpfid,
917 const struct lu_name *oldlname,
918 const struct lu_fid *newpfid,
919 const struct lu_name *newlname,
920 int first, int check,
921 struct linkea_data *ldata)
927 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
930 LASSERT(oldpfid != NULL || newpfid != NULL);
932 if (mdd_obj->mod_flags & DEAD_OBJ) {
933 /* Prevent linkea to be updated which is NOT necessary. */
934 ldata->ld_reclen = 0;
935 /* No more links, don't bother */
939 if (oldpfid != NULL) {
940 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
943 (rc != -ENODATA && rc != -ENOENT))
945 /* No changes done. */
950 /* If renaming, add the new record */
951 if (newpfid != NULL) {
952 /* even if the add fails, we still delete the out-of-date
954 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
958 rc = rc != 0 ? rc : rc2;
963 int mdd_links_rename(const struct lu_env *env,
964 struct mdd_object *mdd_obj,
965 const struct lu_fid *oldpfid,
966 const struct lu_name *oldlname,
967 const struct lu_fid *newpfid,
968 const struct lu_name *newlname,
969 struct thandle *handle,
970 struct linkea_data *ldata,
971 int first, int check)
977 ldata = &mdd_env_info(env)->mti_link_data;
978 memset(ldata, 0, sizeof(*ldata));
979 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
980 newpfid, newlname, first, check,
986 if (ldata->ld_reclen != 0)
987 rc = mdd_links_write(env, mdd_obj, ldata, handle);
992 if (rc == -EOVERFLOW || rc == -ENOSPC)
995 CDEBUG(error ? D_ERROR : D_OTHER,
996 "link_ea add '%.*s' failed %d "DFID"\n",
997 newlname->ln_namelen, newlname->ln_name,
998 rc, PFID(mdd_object_fid(mdd_obj)));
999 else if (newpfid == NULL)
1000 CDEBUG(error ? D_ERROR : D_OTHER,
1001 "link_ea del '%.*s' failed %d "DFID"\n",
1002 oldlname->ln_namelen, oldlname->ln_name,
1003 rc, PFID(mdd_object_fid(mdd_obj)));
1005 CDEBUG(error ? D_ERROR : D_OTHER,
1006 "link_ea rename '%.*s'->'%.*s' failed %d "
1008 oldlname->ln_namelen, oldlname->ln_name,
1009 newlname->ln_namelen, newlname->ln_name,
1010 rc, PFID(mdd_object_fid(mdd_obj)));
1013 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1014 /* if we vmalloced a large buffer drop it */
1015 lu_buf_free(ldata->ld_buf);
1020 static inline int mdd_links_add(const struct lu_env *env,
1021 struct mdd_object *mdd_obj,
1022 const struct lu_fid *pfid,
1023 const struct lu_name *lname,
1024 struct thandle *handle,
1025 struct linkea_data *ldata, int first)
1027 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1028 pfid, lname, handle, ldata, first, 0);
1031 static inline int mdd_links_del(const struct lu_env *env,
1032 struct mdd_object *mdd_obj,
1033 const struct lu_fid *pfid,
1034 const struct lu_name *lname,
1035 struct thandle *handle)
1037 return mdd_links_rename(env, mdd_obj, pfid, lname,
1038 NULL, NULL, handle, NULL, 0, 0);
1041 /** Read the link EA into a temp buffer.
1042 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1043 * A pointer to the buffer is stored in \a ldata::ld_buf.
1045 * \retval 0 or error
1047 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1048 struct linkea_data *ldata)
1052 if (!mdd_object_exists(mdd_obj))
1055 /* First try a small buf */
1056 LASSERT(env != NULL);
1057 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1059 if (ldata->ld_buf->lb_buf == NULL)
1062 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
1063 if (rc == -ERANGE) {
1064 /* Buf was too small, figure out what we need. */
1065 lu_buf_free(ldata->ld_buf);
1066 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1070 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1071 if (ldata->ld_buf->lb_buf == NULL)
1073 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1077 lu_buf_free(ldata->ld_buf);
1078 ldata->ld_buf = NULL;
1082 return linkea_init(ldata);
1085 /** Read the link EA into a temp buffer.
1086 * Uses the name_buf since it is generally large.
1087 * \retval IS_ERR err
1088 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1090 struct lu_buf *mdd_links_get(const struct lu_env *env,
1091 struct mdd_object *mdd_obj)
1093 struct linkea_data ldata = { NULL };
1096 rc = mdd_links_read(env, mdd_obj, &ldata);
1097 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1100 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1101 struct linkea_data *ldata, struct thandle *handle)
1103 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1104 ldata->ld_leh->leh_len);
1107 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1110 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1111 if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) &&
1112 mdd_object_remote(mdd_obj) == 0) {
1113 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1115 /* XXX: If the linkEA is overflow, then we need to notify the
1116 * namespace LFSCK to skip "nlink" attribute verification
1117 * on this object to avoid the "nlink" to be shrinked by
1118 * wrong. It may be not good an interaction with LFSCK
1119 * like this. We will consider to replace it with other
1120 * mechanism in future. LU-5802. */
1121 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK,
1122 LFSCK_TYPE_NAMESPACE);
1123 lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1130 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1131 struct thandle *handle, struct linkea_data *ldata,
1132 enum mdd_links_add_overflow overflow)
1138 if (ldata != NULL && ldata->ld_leh != NULL) {
1139 ea_len = ldata->ld_leh->leh_len;
1140 linkea = ldata->ld_buf->lb_buf;
1142 ea_len = DEFAULT_LINKEA_SIZE;
1146 /* XXX: max size? */
1147 rc = mdo_declare_xattr_set(env, mdd_obj,
1148 mdd_buf_get_const(env, linkea, ea_len),
1149 XATTR_NAME_LINK, 0, handle);
1153 if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) {
1154 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1156 /* XXX: If the linkEA is overflow, then we need to notify the
1157 * namespace LFSCK to skip "nlink" attribute verification
1158 * on this object to avoid the "nlink" to be shrinked by
1159 * wrong. It may be not good an interaction with LFSCK
1160 * like this. We will consider to replace it with other
1161 * mechanism in future. LU-5802. */
1162 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE,
1163 LFSCK_TYPE_NAMESPACE);
1164 rc = lfsck_in_notify(env,
1165 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1172 static inline int mdd_declare_links_del(const struct lu_env *env,
1173 struct mdd_object *c,
1174 struct thandle *handle)
1178 /* For directory, the linkEA will be removed together
1179 * with the object. */
1180 if (!S_ISDIR(mdd_object_type(c)))
1181 rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE);
1186 static int mdd_declare_link(const struct lu_env *env,
1187 struct mdd_device *mdd,
1188 struct mdd_object *p,
1189 struct mdd_object *c,
1190 const struct lu_name *name,
1191 struct thandle *handle,
1193 struct linkea_data *data)
1197 rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
1198 name->ln_name, handle);
1202 rc = mdo_declare_ref_add(env, c, handle);
1206 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1207 rc = mdo_declare_ref_add(env, c, handle);
1212 la->la_valid = LA_CTIME | LA_MTIME;
1213 rc = mdo_declare_attr_set(env, p, la, handle);
1217 la->la_valid = LA_CTIME;
1218 rc = mdo_declare_attr_set(env, c, la, handle);
1222 rc = mdd_declare_links_add(env, c, handle, data,
1223 S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE);
1227 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1232 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1233 struct md_object *src_obj, const struct lu_name *lname,
1236 const char *name = lname->ln_name;
1237 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1238 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1239 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1240 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1241 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1242 struct mdd_device *mdd = mdo2mdd(src_obj);
1243 struct thandle *handle;
1244 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1248 rc = mdd_la_get(env, mdd_sobj, cattr);
1252 rc = mdd_la_get(env, mdd_tobj, tattr);
1256 handle = mdd_trans_create(env, mdd);
1258 GOTO(out_pending, rc = PTR_ERR(handle));
1260 memset(ldata, 0, sizeof(*ldata));
1262 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1263 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1265 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1270 rc = mdd_trans_start(env, mdd, handle);
1274 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1275 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1278 GOTO(out_unlock, rc);
1280 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1281 rc = mdo_ref_add(env, mdd_sobj, handle);
1283 GOTO(out_unlock, rc);
1286 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1287 rc = mdo_ref_add(env, mdd_sobj, handle);
1289 GOTO(out_unlock, rc);
1292 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) {
1293 struct lu_fid tfid = *mdo2fid(mdd_sobj);
1296 rc = __mdd_index_insert_only(env, mdd_tobj, &tfid,
1297 mdd_object_type(mdd_sobj),
1300 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1301 mdd_object_type(mdd_sobj),
1306 mdo_ref_del(env, mdd_sobj, handle);
1307 GOTO(out_unlock, rc);
1310 la->la_valid = LA_CTIME | LA_MTIME;
1311 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1313 GOTO(out_unlock, rc);
1315 la->la_valid = LA_CTIME;
1316 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1318 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1319 mdo2fid(mdd_tobj), lname, 0, 0,
1322 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1323 lname, handle, ldata, 0);
1324 /* The failure of links_add should not cause the link
1325 * failure, reset rc here */
1330 mdd_write_unlock(env, mdd_sobj);
1332 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1333 mdo2fid(mdd_tobj), NULL, NULL,
1334 lname, NULL, handle);
1336 mdd_trans_stop(env, mdd, rc, handle);
1338 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1339 /* if we vmalloced a large buffer drop it */
1340 lu_buf_free(ldata->ld_buf);
1345 static int mdd_mark_dead_object(const struct lu_env *env,
1346 struct mdd_object *obj, struct thandle *handle,
1349 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1353 obj->mod_flags |= DEAD_OBJ;
1355 if (!S_ISDIR(mdd_object_type(obj)))
1358 attr->la_valid = LA_FLAGS;
1359 attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
1362 rc = mdo_declare_attr_set(env, obj, attr, handle);
1364 rc = mdo_attr_set(env, obj, attr, handle);
1369 static int mdd_declare_finish_unlink(const struct lu_env *env,
1370 struct mdd_object *obj,
1371 struct thandle *handle)
1375 rc = mdd_mark_dead_object(env, obj, handle, true);
1379 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1383 rc = mdo_declare_destroy(env, obj, handle);
1387 return mdd_declare_links_del(env, obj, handle);
1390 /* caller should take a lock before calling */
1391 int mdd_finish_unlink(const struct lu_env *env,
1392 struct mdd_object *obj, struct md_attr *ma,
1393 const struct mdd_object *pobj,
1394 const struct lu_name *lname,
1398 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1401 LASSERT(mdd_write_locked(env, obj) != 0);
1403 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1404 rc = mdd_mark_dead_object(env, obj, th, false);
1408 /* add new orphan and the object
1409 * will be deleted during mdd_close() */
1410 if (obj->mod_count) {
1411 rc = __mdd_orphan_add(env, obj, th);
1413 CDEBUG(D_HA, "Object "DFID" is inserted into "
1414 "orphan list, open count = %d\n",
1415 PFID(mdd_object_fid(obj)),
1418 CERROR("Object "DFID" fail to be an orphan, "
1419 "open count = %d, maybe cause failed "
1421 PFID(mdd_object_fid(obj)),
1424 rc = mdo_destroy(env, obj, th);
1426 } else if (!is_dir) {
1427 /* old files may not have link ea; ignore errors */
1428 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1436 * has mdd_write_lock on cobj already, but not on pobj yet
1438 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1439 const struct lu_attr *pattr,
1440 struct mdd_object *cobj,
1441 const struct lu_attr *cattr)
1446 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1451 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1452 struct mdd_object *p, struct mdd_object *c,
1453 const struct lu_name *name, struct md_attr *ma,
1454 struct thandle *handle, int no_name, int is_dir)
1456 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1459 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1460 if (likely(no_name == 0)) {
1461 rc = mdo_declare_index_delete(env, p, name->ln_name,
1468 rc = mdo_declare_ref_del(env, p, handle);
1474 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1475 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1476 la->la_valid = LA_CTIME | LA_MTIME;
1477 rc = mdo_declare_attr_set(env, p, la, handle);
1482 rc = mdo_declare_ref_del(env, c, handle);
1486 rc = mdo_declare_ref_del(env, c, handle);
1490 la->la_valid = LA_CTIME;
1491 rc = mdo_declare_attr_set(env, c, la, handle);
1495 rc = mdd_declare_finish_unlink(env, c, handle);
1499 /* FIXME: need changelog for remove entry */
1500 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1507 * test if a file has an HSM archive
1508 * if HSM attributes are not found in ma update them from
1511 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1512 struct mdd_object *obj,
1517 if (!(ma->ma_valid & MA_HSM)) {
1518 /* no HSM MD provided, read xattr */
1519 struct lu_buf *hsm_buf;
1520 const size_t buflen = sizeof(struct hsm_attrs);
1523 hsm_buf = mdd_buf_get(env, NULL, 0);
1524 lu_buf_alloc(hsm_buf, buflen);
1525 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1526 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1527 lu_buf_free(hsm_buf);
1531 ma->ma_valid = MA_HSM;
1533 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1539 * Delete name entry and the object.
1540 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1541 * does not exist for this object, and it could only happen during resending
1542 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1543 * is also needed in this case(needed by changelog), so we have to add another
1544 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1545 * the ENOENT failure should be able to be fixed by redo mechanism.
1547 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1548 struct md_object *cobj, const struct lu_name *lname,
1549 struct md_attr *ma, int no_name)
1551 const char *name = lname->ln_name;
1552 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1553 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1554 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1555 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1556 struct mdd_object *mdd_cobj = NULL;
1557 struct mdd_device *mdd = mdo2mdd(pobj);
1558 struct thandle *handle;
1562 /* cobj == NULL means only delete name entry */
1563 if (likely(cobj != NULL)) {
1564 mdd_cobj = md2mdd_obj(cobj);
1565 if (mdd_object_exists(mdd_cobj) == 0)
1569 rc = mdd_la_get(env, mdd_pobj, pattr);
1573 if (likely(mdd_cobj != NULL)) {
1575 rc = mdd_la_get(env, mdd_cobj, cattr);
1579 is_dir = S_ISDIR(cattr->la_mode);
1582 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1586 handle = mdd_trans_create(env, mdd);
1588 RETURN(PTR_ERR(handle));
1590 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1591 lname, ma, handle, no_name, is_dir);
1595 rc = mdd_trans_start(env, mdd, handle);
1599 if (likely(mdd_cobj != NULL))
1600 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1602 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1603 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1608 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1609 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1610 GOTO(cleanup, rc = 0);
1612 if (likely(mdd_cobj != NULL)) {
1613 rc = mdo_ref_del(env, mdd_cobj, handle);
1615 __mdd_index_insert_only(env, mdd_pobj,
1617 mdd_object_type(mdd_cobj),
1624 mdo_ref_del(env, mdd_cobj, handle);
1626 /* fetch updated nlink */
1627 rc = mdd_la_get(env, mdd_cobj, cattr);
1632 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1633 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1635 la->la_valid = LA_CTIME | LA_MTIME;
1636 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1640 /* Enough for only unlink the entry */
1641 if (unlikely(mdd_cobj == NULL))
1644 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1645 /* update ctime of an unlinked file only if it is still
1646 * opened or a link still exists */
1647 la->la_valid = LA_CTIME;
1648 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1653 /* XXX: this transfer to ma will be removed with LOD/OSP */
1654 ma->ma_attr = *cattr;
1655 ma->ma_valid |= MA_INODE;
1656 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1658 /* fetch updated nlink */
1660 rc = mdd_la_get(env, mdd_cobj, cattr);
1662 /* if object is removed then we can't get its attrs, use last get */
1663 if (cattr->la_nlink == 0) {
1664 ma->ma_attr = *cattr;
1665 ma->ma_valid |= MA_INODE;
1669 if (likely(mdd_cobj != NULL))
1670 mdd_write_unlock(env, mdd_cobj);
1675 if (cattr->la_nlink == 0) {
1676 cl_flags |= CLF_UNLINK_LAST;
1677 /* search for an existing archive */
1678 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1679 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1682 rc = mdd_changelog_ns_store(env, mdd,
1683 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1684 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1689 mdd_trans_stop(env, mdd, rc, handle);
1695 * The permission has been checked when obj created, no need check again.
1697 static int mdd_cd_sanity_check(const struct lu_env *env,
1698 struct mdd_object *obj)
1703 if (!obj || mdd_is_dead_obj(obj))
1709 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1710 struct md_object *cobj, const struct md_op_spec *spec,
1713 struct mdd_device *mdd = mdo2mdd(cobj);
1714 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1715 struct mdd_object *son = md2mdd_obj(cobj);
1716 struct thandle *handle;
1717 const struct lu_buf *buf;
1718 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1719 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1723 rc = mdd_cd_sanity_check(env, son);
1727 if (!md_should_create(spec->sp_cr_flags))
1731 * there are following use cases for this function:
1732 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1733 * striping can be specified or not
1736 rc = mdd_la_get(env, son, attr);
1740 /* calling ->ah_make_hint() is used to transfer information from parent */
1741 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1743 handle = mdd_trans_create(env, mdd);
1745 GOTO(out_free, rc = PTR_ERR(handle));
1748 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1749 * Should this be fixed?
1751 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1752 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1753 spec->sp_cr_flags, spec->no_create);
1755 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1756 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1757 spec->u.sp_ea.eadatalen);
1762 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1763 XATTR_NAME_LOV, 0, handle);
1767 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1771 rc = mdd_trans_start(env, mdd, handle);
1775 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1781 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1784 mdd_trans_stop(env, mdd, rc, handle);
1789 static int mdd_declare_object_initialize(const struct lu_env *env,
1790 struct mdd_object *parent,
1791 struct mdd_object *child,
1792 struct lu_attr *attr,
1793 struct thandle *handle)
1799 * inode mode has been set in creation time, and it's based on umask,
1800 * la_mode and acl, don't set here again! (which will go wrong
1801 * because below function doesn't consider umask).
1802 * I'd suggest set all object attributes in creation time, see above.
1804 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1805 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1806 rc = mdo_declare_attr_set(env, child, attr, handle);
1807 attr->la_valid |= LA_MODE | LA_TYPE;
1808 if (rc != 0 || !S_ISDIR(attr->la_mode))
1811 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1816 rc = mdo_declare_ref_add(env, child, handle);
1820 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1826 static int mdd_object_initialize(const struct lu_env *env,
1827 const struct lu_fid *pfid,
1828 struct mdd_object *child,
1829 struct lu_attr *attr, struct thandle *handle,
1830 const struct md_op_spec *spec)
1835 if (S_ISDIR(attr->la_mode)) {
1836 /* Add "." and ".." for newly created dir */
1837 mdo_ref_add(env, child, handle);
1838 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1839 S_IFDIR, dot, handle);
1841 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1844 mdo_ref_del(env, child, handle);
1851 * This function checks whether it can create a file/dir under the
1852 * directory(@pobj). The directory(@pobj) is not being locked by
1855 * \param[in] env execution environment
1856 * \param[in] pobj the directory to create files
1857 * \param[in] pattr the attributes of the directory
1858 * \param[in] lname the name of the created file/dir
1859 * \param[in] cattr the attributes of the file/dir
1860 * \param[in] spec create specification
1862 * \retval = 0 it is allowed to create file/dir under
1864 * \retval negative error not allowed to create file/dir
1865 * under the directory
1867 static int mdd_create_sanity_check(const struct lu_env *env,
1868 struct md_object *pobj,
1869 const struct lu_attr *pattr,
1870 const struct lu_name *lname,
1871 struct lu_attr *cattr,
1872 struct md_op_spec *spec)
1874 struct mdd_thread_info *info = mdd_env_info(env);
1875 struct lu_fid *fid = &info->mti_fid;
1876 struct mdd_object *obj = md2mdd_obj(pobj);
1877 struct mdd_device *m = mdo2mdd(pobj);
1878 bool check_perm = true;
1883 if (mdd_is_dead_obj(obj))
1887 * In some cases this lookup is not needed - we know before if name
1888 * exists or not because MDT performs lookup for it.
1889 * name length check is done in lookup.
1891 if (spec->sp_cr_lookup) {
1893 * Check if the name already exist, though it will be checked in
1894 * _index_insert also, for avoiding rolling back if exists
1897 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1898 MAY_WRITE | MAY_EXEC);
1900 RETURN(rc ? : -EEXIST);
1902 /* Permission is already being checked in mdd_lookup */
1906 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1911 if (pattr->la_mode & S_ISGID) {
1912 cattr->la_gid = pattr->la_gid;
1913 if (S_ISDIR(cattr->la_mode)) {
1914 cattr->la_mode |= S_ISGID;
1915 cattr->la_valid |= LA_MODE;
1919 rc = mdd_name_check(m, lname);
1923 switch (cattr->la_mode & S_IFMT) {
1925 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1927 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1928 RETURN(-ENAMETOOLONG);
1947 static int mdd_declare_object_create(const struct lu_env *env,
1948 struct mdd_device *mdd,
1949 struct mdd_object *p, struct mdd_object *c,
1950 struct lu_attr *attr,
1951 struct thandle *handle,
1952 const struct md_op_spec *spec,
1953 struct lu_buf *def_acl_buf,
1954 struct lu_buf *acl_buf,
1955 struct dt_allocation_hint *hint)
1959 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
1964 #ifdef CONFIG_FS_POSIX_ACL
1965 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1966 /* if dir, then can inherit default ACl */
1967 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1968 XATTR_NAME_ACL_DEFAULT,
1974 if (acl_buf->lb_len > 0) {
1975 rc = mdo_declare_attr_set(env, c, attr, handle);
1979 rc = mdo_declare_xattr_set(env, c, acl_buf,
1980 XATTR_NAME_ACL_ACCESS, 0, handle);
1985 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
1989 /* replay case, create LOV EA from client data */
1990 if (spec->no_create ||
1991 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
1992 const struct lu_buf *buf;
1994 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1995 spec->u.sp_ea.eadatalen);
1996 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2002 if (S_ISLNK(attr->la_mode)) {
2003 const char *target_name = spec->u.sp_symname;
2004 int sym_len = strlen(target_name);
2005 const struct lu_buf *buf;
2007 buf = mdd_buf_get_const(env, target_name, sym_len);
2008 rc = dt_declare_record_write(env, mdd_object_child(c),
2017 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2018 struct mdd_object *p, struct mdd_object *c,
2019 const struct lu_name *name,
2020 struct lu_attr *attr,
2021 struct thandle *handle,
2022 const struct md_op_spec *spec,
2023 struct linkea_data *ldata,
2024 struct lu_buf *def_acl_buf,
2025 struct lu_buf *acl_buf,
2026 struct dt_allocation_hint *hint)
2030 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2031 def_acl_buf, acl_buf, hint);
2035 if (S_ISDIR(attr->la_mode)) {
2036 rc = mdo_declare_ref_add(env, p, handle);
2041 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2042 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2046 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2048 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2049 name->ln_name, handle);
2053 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2058 la->la_valid = LA_CTIME | LA_MTIME;
2059 rc = mdo_declare_attr_set(env, p, la, handle);
2063 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2068 /* XXX: For remote create, it should indicate the remote RPC
2069 * will be sent after local transaction is finished, which
2070 * is not very nice, but it will be removed once we fully support
2072 if (mdd_object_remote(p) && handle->th_update != NULL)
2073 handle->th_update->tu_sent_after_local_trans = 1;
2078 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2079 struct lu_attr *la, struct lu_buf *def_acl_buf,
2080 struct lu_buf *acl_buf)
2085 if (S_ISLNK(la->la_mode)) {
2086 acl_buf->lb_len = 0;
2087 def_acl_buf->lb_len = 0;
2091 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2092 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2093 XATTR_NAME_ACL_DEFAULT);
2094 mdd_read_unlock(env, pobj);
2096 /* If there are default ACL, fix mode/ACL by default ACL */
2097 def_acl_buf->lb_len = rc;
2098 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2099 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2100 acl_buf->lb_len = rc;
2101 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2104 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2105 /* If there are no default ACL, fix mode by mask */
2106 struct lu_ucred *uc = lu_ucred(env);
2108 /* The create triggered by MDT internal events, such as
2109 * LFSCK reset, will not contain valid "uc". */
2110 if (unlikely(uc != NULL))
2111 la->la_mode &= ~uc->uc_umask;
2113 acl_buf->lb_len = 0;
2114 def_acl_buf->lb_len = 0;
2121 * Create a metadata object and initialize it, set acl, xattr.
2123 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2124 struct mdd_object *son, struct lu_attr *attr,
2125 struct md_op_spec *spec, struct lu_buf *acl_buf,
2126 struct lu_buf *def_acl_buf,
2127 struct dt_allocation_hint *hint,
2128 struct thandle *handle)
2132 mdd_write_lock(env, son, MOR_TGT_CHILD);
2133 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2138 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2139 * created in declare phase, they also needs to be added to master
2140 * object as sub-directory entry. So it has to initialize the master
2141 * object, then set dir striped EA.(in mdo_xattr_set) */
2142 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2145 GOTO(err_destroy, rc);
2148 * in case of replay we just set LOVEA provided by the client
2149 * XXX: I think it would be interesting to try "old" way where
2150 * MDT calls this xattr_set(LOV) in a different transaction.
2151 * probably this way we code can be made better.
2154 /* During creation, there are only a few cases we need do xattr_set to
2156 * 1. regular file: see comments above.
2157 * 2. dir: inherit default striping or pool settings from parent.
2158 * 3. create striped directory with provided stripeEA.
2159 * 4. create striped directory because inherit default layout from the
2162 if (spec->no_create ||
2163 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2164 S_ISDIR(attr->la_mode)) {
2165 const struct lu_buf *buf;
2167 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2168 spec->u.sp_ea.eadatalen);
2169 rc = mdo_xattr_set(env, son, buf,
2170 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2174 GOTO(err_destroy, rc);
2177 #ifdef CONFIG_FS_POSIX_ACL
2178 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2179 S_ISDIR(attr->la_mode)) {
2180 /* set default acl */
2181 rc = mdo_xattr_set(env, son, def_acl_buf,
2182 XATTR_NAME_ACL_DEFAULT, 0,
2185 GOTO(err_destroy, rc);
2187 /* set its own acl */
2188 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2189 rc = mdo_xattr_set(env, son, acl_buf,
2190 XATTR_NAME_ACL_ACCESS,
2193 GOTO(err_destroy, rc);
2197 if (S_ISLNK(attr->la_mode)) {
2198 struct lu_ucred *uc = lu_ucred_assert(env);
2199 struct dt_object *dt = mdd_object_child(son);
2200 const char *target_name = spec->u.sp_symname;
2201 int sym_len = strlen(target_name);
2202 const struct lu_buf *buf;
2205 buf = mdd_buf_get_const(env, target_name, sym_len);
2206 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2208 CFS_CAP_SYS_RESOURCE_MASK);
2213 GOTO(err_initlized, rc = -EFAULT);
2217 if (unlikely(rc != 0)) {
2219 if (S_ISDIR(attr->la_mode)) {
2220 /* Drop the reference, no need to delete "."/"..",
2221 * because the object to be destroied directly. */
2222 rc2 = mdo_ref_del(env, son, handle);
2226 rc2 = mdo_ref_del(env, son, handle);
2230 mdo_destroy(env, son, handle);
2233 mdd_write_unlock(env, son);
2238 * Create object and insert it into namespace.
2240 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2241 const struct lu_name *lname, struct md_object *child,
2242 struct md_op_spec *spec, struct md_attr* ma)
2244 struct mdd_thread_info *info = mdd_env_info(env);
2245 struct lu_attr *la = &info->mti_la_for_fix;
2246 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2247 struct mdd_object *son = md2mdd_obj(child);
2248 struct mdd_device *mdd = mdo2mdd(pobj);
2249 struct lu_attr *attr = &ma->ma_attr;
2250 struct thandle *handle;
2251 struct lu_attr *pattr = &info->mti_pattr;
2252 struct lu_buf acl_buf;
2253 struct lu_buf def_acl_buf;
2254 struct linkea_data *ldata = &info->mti_link_data;
2255 const char *name = lname->ln_name;
2256 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2262 * Two operations have to be performed:
2264 * - an allocation of a new object (->do_create()), and
2266 * - an insertion into a parent index (->dio_insert()).
2268 * Due to locking, operation order is not important, when both are
2269 * successful, *but* error handling cases are quite different:
2271 * - if insertion is done first, and following object creation fails,
2272 * insertion has to be rolled back, but this operation might fail
2273 * also leaving us with dangling index entry.
2275 * - if creation is done first, is has to be undone if insertion
2276 * fails, leaving us with leaked space, which is neither good, nor
2279 * It seems that creation-first is simplest solution, but it is
2280 * sub-optimal in the frequent
2285 * case, because second mkdir is bound to create object, only to
2286 * destroy it immediately.
2288 * To avoid this follow local file systems that do double lookup:
2290 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2292 * 1. create (mdd_object_create_internal())
2294 * 2. insert (__mdd_index_insert(), lookup again)
2297 rc = mdd_la_get(env, mdd_pobj, pattr);
2301 /* Sanity checks before big job. */
2302 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2306 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2307 GOTO(out_free, rc = -EINPROGRESS);
2309 handle = mdd_trans_create(env, mdd);
2311 GOTO(out_free, rc = PTR_ERR(handle));
2313 acl_buf.lb_buf = info->mti_xattr_buf;
2314 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2315 def_acl_buf.lb_buf = info->mti_key;
2316 def_acl_buf.lb_len = sizeof(info->mti_key);
2317 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2321 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2323 memset(ldata, 0, sizeof(*ldata));
2324 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2325 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2328 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2329 &tfid, lname, 1, 0, ldata);
2331 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2332 mdd_object_fid(mdd_pobj),
2333 lname, 1, 0, ldata);
2336 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2337 handle, spec, ldata, &def_acl_buf, &acl_buf,
2342 rc = mdd_trans_start(env, mdd, handle);
2346 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2347 &def_acl_buf, hint, handle);
2351 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2352 mdd_write_lock(env, son, MOR_TGT_CHILD);
2353 rc = __mdd_orphan_add(env, son, handle);
2354 GOTO(out_volatile, rc);
2356 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2357 attr->la_mode, name, handle);
2359 GOTO(err_created, rc);
2361 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2364 /* update parent directory mtime/ctime */
2366 la->la_valid = LA_CTIME | LA_MTIME;
2367 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2369 GOTO(err_insert, rc);
2377 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2378 rc2 = __mdd_orphan_del(env, son, handle);
2380 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2381 S_ISDIR(attr->la_mode),
2387 mdd_write_lock(env, son, MOR_TGT_CHILD);
2388 if (S_ISDIR(attr->la_mode)) {
2389 /* Drop the reference, no need to delete "."/"..",
2390 * because the object is to be destroyed directly. */
2391 rc2 = mdo_ref_del(env, son, handle);
2393 mdd_write_unlock(env, son);
2398 /* For volatile files drop one link immediately, since there is
2399 * no filename in the namespace, and save any error returned. */
2400 rc2 = mdo_ref_del(env, son, handle);
2402 mdd_write_unlock(env, son);
2403 if (unlikely(rc == 0))
2408 /* Don't destroy the volatile object on success */
2409 if (likely(rc != 0))
2410 mdo_destroy(env, son, handle);
2411 mdd_write_unlock(env, son);
2414 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2415 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2416 rc = mdd_changelog_ns_store(env, mdd,
2417 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2418 S_ISREG(attr->la_mode) ? CL_CREATE :
2419 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2420 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2423 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2427 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2428 /* if we vmalloced a large buffer drop it */
2429 lu_buf_free(ldata->ld_buf);
2431 /* The child object shouldn't be cached anymore */
2433 set_bit(LU_OBJECT_HEARD_BANSHEE,
2434 &child->mo_lu.lo_header->loh_flags);
2439 * Get locks on parents in proper order
2440 * RETURN: < 0 - error, rename_order if successful
2448 static int mdd_rename_order(const struct lu_env *env,
2449 struct mdd_device *mdd,
2450 struct mdd_object *src_pobj,
2451 const struct lu_attr *pattr,
2452 struct mdd_object *tgt_pobj)
2454 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2458 if (src_pobj == tgt_pobj)
2459 RETURN(MDD_RN_SAME);
2461 /* compared the parent child relationship of src_p&tgt_p */
2462 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2464 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2467 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2481 /* has not mdd_write{read}_lock on any obj yet. */
2482 static int mdd_rename_sanity_check(const struct lu_env *env,
2483 struct mdd_object *src_pobj,
2484 const struct lu_attr *pattr,
2485 struct mdd_object *tgt_pobj,
2486 const struct lu_attr *tpattr,
2487 struct mdd_object *sobj,
2488 const struct lu_attr *cattr,
2489 struct mdd_object *tobj,
2490 const struct lu_attr *tattr)
2495 /* XXX: when get here, sobj must NOT be NULL,
2496 * the other case has been processed in cld_rename
2497 * before mdd_rename and enable MDS_PERM_BYPASS. */
2500 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2504 /* XXX: when get here, "tobj == NULL" means tobj must
2505 * NOT exist (neither on remote MDS, such case has been
2506 * processed in cld_rename before mdd_rename and enable
2508 * So check may_create, but not check may_unlink. */
2510 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2511 (src_pobj != tgt_pobj));
2513 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2514 (src_pobj != tgt_pobj), 1);
2516 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2517 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2522 static int mdd_declare_rename(const struct lu_env *env,
2523 struct mdd_device *mdd,
2524 struct mdd_object *mdd_spobj,
2525 struct mdd_object *mdd_tpobj,
2526 struct mdd_object *mdd_sobj,
2527 struct mdd_object *mdd_tobj,
2528 const struct lu_name *tname,
2529 const struct lu_name *sname,
2531 struct linkea_data *ldata,
2532 struct thandle *handle)
2534 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2537 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2538 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2544 /* name from source dir */
2545 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2549 /* .. from source child */
2550 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2551 /* source child can be directory,
2552 * counted by source dir's nlink */
2553 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2556 if (mdd_spobj != mdd_tpobj) {
2557 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2562 rc = mdo_declare_index_insert(env, mdd_sobj,
2564 S_IFDIR, dotdot, handle);
2569 /* new target child can be directory,
2570 * counted by target dir's nlink */
2571 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2576 la->la_valid = LA_CTIME | LA_MTIME;
2577 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2581 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2585 la->la_valid = LA_CTIME;
2586 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2590 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
2591 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2596 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2597 mdd_object_type(mdd_sobj),
2598 tname->ln_name, handle);
2602 /* name from target dir (old name), we declare it unconditionally
2603 * as mdd_rename() calls delete unconditionally as well. so just
2604 * to balance declarations vs calls to change ... */
2605 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2609 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2610 /* delete target child in target parent directory */
2611 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2615 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2616 /* target child can be directory,
2617 * delete "." reference in target child directory */
2618 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2622 /* delete ".." reference in target parent directory */
2623 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2628 la->la_valid = LA_CTIME;
2629 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2633 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2638 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2645 /* src object can be remote that is why we use only fid and type of object */
2646 static int mdd_rename(const struct lu_env *env,
2647 struct md_object *src_pobj, struct md_object *tgt_pobj,
2648 const struct lu_fid *lf, const struct lu_name *lsname,
2649 struct md_object *tobj, const struct lu_name *ltname,
2652 const char *sname = lsname->ln_name;
2653 const char *tname = ltname->ln_name;
2654 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2655 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2656 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2657 struct mdd_device *mdd = mdo2mdd(src_pobj);
2658 struct mdd_object *mdd_sobj = NULL; /* source object */
2659 struct mdd_object *mdd_tobj = NULL;
2660 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2661 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2662 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2663 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2664 struct thandle *handle;
2665 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2666 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2667 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2670 bool tobj_locked = 0;
2671 unsigned cl_flags = 0;
2676 mdd_tobj = md2mdd_obj(tobj);
2678 mdd_sobj = mdd_object_find(env, mdd, lf);
2680 rc = mdd_la_get(env, mdd_sobj, cattr);
2682 GOTO(out_pending, rc);
2684 rc = mdd_la_get(env, mdd_spobj, pattr);
2686 GOTO(out_pending, rc);
2689 rc = mdd_la_get(env, mdd_tobj, tattr);
2691 GOTO(out_pending, rc);
2694 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2696 GOTO(out_pending, rc);
2698 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2699 mdd_sobj, cattr, mdd_tobj, tattr);
2701 GOTO(out_pending, rc);
2703 rc = mdd_name_check(mdd, ltname);
2705 GOTO(out_pending, rc);
2707 handle = mdd_trans_create(env, mdd);
2709 GOTO(out_pending, rc = PTR_ERR(handle));
2711 memset(ldata, 0, sizeof(*ldata));
2712 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2713 ltname, 1, 0, ldata);
2714 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2715 mdd_tobj, lsname, ltname, ma, ldata, handle);
2719 rc = mdd_trans_start(env, mdd, handle);
2723 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2724 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2726 GOTO(cleanup_unlocked, rc);
2728 is_dir = S_ISDIR(cattr->la_mode);
2730 /* Remove source name from source directory */
2731 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2735 /* "mv dir1 dir2" needs "dir1/.." link update */
2736 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2737 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2739 GOTO(fixup_spobj2, rc);
2741 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2744 GOTO(fixup_spobj, rc);
2747 /* Remove target name from target directory
2748 * Here tobj can be remote one, so we do index_delete unconditionally
2749 * and -ENOENT is allowed.
2751 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2754 /* tname might been renamed to something else */
2755 GOTO(fixup_spobj, rc);
2758 GOTO(fixup_spobj, rc);
2761 /* Insert new fid with target name into target dir */
2762 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2765 GOTO(fixup_tpobj, rc);
2767 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2768 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2770 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2772 la->la_valid = LA_CTIME;
2773 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2775 GOTO(fixup_tpobj, rc);
2778 /* Remove old target object
2779 * For tobj is remote case cmm layer has processed
2780 * and set tobj to NULL then. So when tobj is NOT NULL,
2781 * it must be local one.
2783 if (tobj && mdd_object_exists(mdd_tobj)) {
2784 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2786 if (mdd_is_dead_obj(mdd_tobj)) {
2787 /* shld not be dead, something is wrong */
2788 CERROR("tobj is dead, something is wrong\n");
2792 mdo_ref_del(env, mdd_tobj, handle);
2794 /* Remove dot reference. */
2795 if (S_ISDIR(tattr->la_mode))
2796 mdo_ref_del(env, mdd_tobj, handle);
2799 /* fetch updated nlink */
2800 rc = mdd_la_get(env, mdd_tobj, tattr);
2802 CERROR("%s: Failed to get nlink for tobj "
2804 mdd2obd_dev(mdd)->obd_name,
2805 PFID(tpobj_fid), rc);
2806 GOTO(fixup_tpobj, rc);
2809 la->la_valid = LA_CTIME;
2810 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2812 CERROR("%s: Failed to set ctime for tobj "
2814 mdd2obd_dev(mdd)->obd_name,
2815 PFID(tpobj_fid), rc);
2816 GOTO(fixup_tpobj, rc);
2819 /* XXX: this transfer to ma will be removed with LOD/OSP */
2820 ma->ma_attr = *tattr;
2821 ma->ma_valid |= MA_INODE;
2822 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2825 CERROR("%s: Failed to unlink tobj "
2827 mdd2obd_dev(mdd)->obd_name,
2828 PFID(tpobj_fid), rc);
2829 GOTO(fixup_tpobj, rc);
2832 /* fetch updated nlink */
2833 rc = mdd_la_get(env, mdd_tobj, tattr);
2835 CERROR("%s: Failed to get nlink for tobj "
2837 mdd2obd_dev(mdd)->obd_name,
2838 PFID(tpobj_fid), rc);
2839 GOTO(fixup_tpobj, rc);
2841 /* XXX: this transfer to ma will be removed with LOD/OSP */
2842 ma->ma_attr = *tattr;
2843 ma->ma_valid |= MA_INODE;
2845 if (tattr->la_nlink == 0) {
2846 cl_flags |= CLF_RENAME_LAST;
2847 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2848 cl_flags |= CLF_RENAME_LAST_EXISTS;
2852 la->la_valid = LA_CTIME | LA_MTIME;
2853 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2855 GOTO(fixup_tpobj, rc);
2857 if (mdd_spobj != mdd_tpobj) {
2858 la->la_valid = LA_CTIME | LA_MTIME;
2859 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2862 if (rc == 0 && mdd_sobj) {
2863 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2864 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2865 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2868 /* Old files might not have EA entry */
2869 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2870 lsname, handle, NULL, 0);
2871 mdd_write_unlock(env, mdd_sobj);
2872 /* We don't fail the transaction if the link ea can't be
2873 updated -- fid2path will use alternate lookup method. */
2881 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2883 CWARN("tp obj fix error %d\n",rc2);
2885 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2886 !mdd_is_dead_obj(mdd_tobj)) {
2888 mdo_ref_add(env, mdd_tobj, handle);
2890 mdo_ref_add(env, mdd_tobj, handle);
2893 rc2 = __mdd_index_insert(env, mdd_tpobj,
2895 mdd_object_type(mdd_tobj),
2898 CWARN("tp obj fix error: rc = %d\n", rc2);
2903 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2904 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2906 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2907 mdd2obd_dev(mdd)->obd_name, rc2);
2910 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
2913 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2914 mdd2obd_dev(mdd)->obd_name, rc2);
2919 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
2920 mdd_object_type(mdd_sobj), sname,
2923 CWARN("sp obj fix error: rc = %d\n", rc2);
2928 mdd_write_unlock(env, mdd_tobj);
2932 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
2933 mdd_tobj, tpobj_fid, lf, spobj_fid,
2934 ltname, lsname, handle);
2937 mdd_trans_stop(env, mdd, rc, handle);
2940 mdd_object_put(env, mdd_sobj);
2945 * During migration once the parent FID has been changed,
2946 * we need update the parent FID in linkea.
2948 static int mdd_linkea_update_child_internal(const struct lu_env *env,
2949 struct mdd_object *parent,
2950 struct mdd_object *child,
2951 const char *name, int namelen,
2952 struct thandle *handle,
2955 struct mdd_thread_info *info = mdd_env_info(env);
2956 struct linkea_data ldata = { NULL };
2957 struct lu_buf *buf = &info->mti_link_buf;
2963 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2964 if (buf->lb_buf == NULL)
2968 rc = mdd_links_read(env, child, &ldata);
2970 if (rc == -ENOENT || rc == -ENODATA)
2975 LASSERT(ldata.ld_leh != NULL);
2976 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
2977 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
2978 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
2979 struct lu_name lname;
2982 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
2985 if (strncmp(lname.ln_name, name, namelen) != 0 ||
2986 lu_fid_eq(&fid, mdd_object_fid(parent))) {
2987 ldata.ld_lee = (struct link_ea_entry *)
2988 ((char *)ldata.ld_lee +
2993 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
2994 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
2995 lname.ln_namelen, lname.ln_name,
2996 PFID(mdd_object_fid(parent)));
2997 /* update to the new parent fid */
2998 linkea_entry_pack(ldata.ld_lee, &lname,
2999 mdd_object_fid(parent));
3001 rc = mdd_declare_links_add(env, child, handle, &ldata,
3004 rc = mdd_links_write(env, child, &ldata, handle);
3010 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3011 struct mdd_object *parent,
3012 struct mdd_object *child,
3013 const char *name, int namelen,
3014 struct thandle *handle)
3016 return mdd_linkea_update_child_internal(env, parent, child, name,
3017 namelen, handle, true);
3020 static int mdd_linkea_update_child(const struct lu_env *env,
3021 struct mdd_object *parent,
3022 struct mdd_object *child,
3023 const char *name, int namelen,
3024 struct thandle *handle)
3026 return mdd_linkea_update_child_internal(env, parent, child, name,
3027 namelen, handle, false);
3030 static int mdd_update_linkea_internal(const struct lu_env *env,
3031 struct mdd_object *mdd_pobj,
3032 struct mdd_object *mdd_sobj,
3033 struct mdd_object *mdd_tobj,
3034 const struct lu_name *child_name,
3035 struct thandle *handle,
3038 struct mdd_thread_info *info = mdd_env_info(env);
3039 struct linkea_data *ldata = &info->mti_link_data;
3044 rc = mdd_links_read(env, mdd_sobj, ldata);
3046 if (rc == -ENOENT || rc == -ENODATA)
3051 /* If it is mulitple links file, we need update the name entry for
3053 LASSERT(ldata->ld_leh != NULL);
3054 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3055 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3056 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3057 struct mdd_object *pobj;
3058 struct lu_name lname;
3061 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3063 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3065 pobj = mdd_object_find(env, mdd, &fid);
3067 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3068 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3073 if (!mdd_object_exists(pobj)) {
3074 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3075 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3079 if (pobj == mdd_pobj &&
3080 lname.ln_namelen == child_name->ln_namelen &&
3081 strncmp(lname.ln_name, child_name->ln_name,
3082 lname.ln_namelen) == 0) {
3083 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3084 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3089 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3090 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3091 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3094 /* Remove source name from source directory */
3095 /* Insert new fid with target name into target dir */
3096 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3101 rc = mdo_declare_index_insert(env, pobj,
3102 mdd_object_fid(mdd_tobj),
3103 mdd_object_type(mdd_tobj),
3104 lname.ln_name, handle);
3108 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3112 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3116 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3121 rc = __mdd_index_insert(env, pobj,
3122 mdd_object_fid(mdd_tobj),
3123 mdd_object_type(mdd_tobj),
3124 lname.ln_name, handle);
3128 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3129 rc = mdo_ref_add(env, mdd_tobj, handle);
3130 mdd_write_unlock(env, mdd_tobj);
3134 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3135 mdo_ref_del(env, mdd_sobj, handle);
3136 mdd_write_unlock(env, mdd_sobj);
3139 mdd_object_put(env, pobj);
3147 static int mdd_migrate_xattrs(const struct lu_env *env,
3148 struct mdd_object *mdd_sobj,
3149 struct mdd_object *mdd_tobj)
3151 struct mdd_thread_info *info = mdd_env_info(env);
3152 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3154 struct thandle *handle;
3160 struct lu_buf list_xbuf;
3163 /* retrieve xattr list from the old object */
3164 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3165 if (list_xsize == -ENODATA)
3171 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3172 if (info->mti_big_buf.lb_buf == NULL)
3175 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3176 list_xbuf.lb_len = list_xsize;
3177 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3182 xname = list_xbuf.lb_buf;
3184 xlen = strnlen(xname, rem - 1) + 1;
3185 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3186 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3187 strcmp(XATTR_NAME_LMV, xname) == 0)
3190 /* For directory, if there are default layout, migrate here */
3191 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3192 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3195 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3196 if (xsize == -ENODATA)
3201 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3202 if (info->mti_link_buf.lb_buf == NULL)
3203 GOTO(out, rc = -ENOMEM);
3205 xbuf.lb_len = xsize;
3206 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3207 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3213 handle = mdd_trans_create(env, mdd);
3215 GOTO(out, rc = PTR_ERR(handle));
3217 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3220 GOTO(stop_trans, rc);
3221 /* Note: this transaction is part of migration, and it is not
3222 * the last step of migration, so we set th_local = 1 to avoid
3223 * update last rcvd for this transaction */
3224 handle->th_local = 1;
3225 rc = mdd_trans_start(env, mdd, handle);
3227 GOTO(stop_trans, rc);
3229 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3231 GOTO(stop_trans, rc = 0);
3234 GOTO(stop_trans, rc);
3236 mdd_trans_stop(env, mdd, rc, handle);
3241 memmove(xname, xname + xlen, rem);
3247 static int mdd_declare_migrate_create(const struct lu_env *env,
3248 struct mdd_object *mdd_pobj,
3249 struct mdd_object *mdd_sobj,
3250 struct mdd_object *mdd_tobj,
3251 struct md_op_spec *spec,
3253 union lmv_mds_md *mgr_ea,
3254 struct linkea_data *ldata,
3255 struct thandle *handle)
3257 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3258 const struct lu_buf *buf;
3262 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3263 handle, spec, NULL);
3267 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3272 if (S_ISLNK(la->la_mode)) {
3273 const char *target_name = spec->u.sp_symname;
3274 int sym_len = strlen(target_name);
3275 const struct lu_buf *buf;
3277 buf = mdd_buf_get_const(env, target_name, sym_len);
3278 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3282 } else if (S_ISDIR(la->la_mode)) {
3283 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3289 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3290 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3291 spec->u.sp_ea.eadatalen);
3292 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3298 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3299 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3300 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3305 la_flag->la_valid = LA_FLAGS;
3306 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3307 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3312 static int mdd_migrate_create(const struct lu_env *env,
3313 struct mdd_object *mdd_pobj,
3314 struct mdd_object *mdd_sobj,
3315 struct mdd_object *mdd_tobj,
3316 const struct lu_name *lname,
3319 struct mdd_thread_info *info = mdd_env_info(env);
3320 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3321 struct md_op_spec *spec = &info->mti_spec;
3322 struct lu_buf lmm_buf = { NULL };
3323 struct lu_buf link_buf = { NULL };
3324 const struct lu_buf *buf;
3325 struct thandle *handle;
3326 struct lmv_mds_md_v1 *mgr_ea;
3327 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3328 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3330 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3334 /* prepare spec for create */
3335 memset(spec, 0, sizeof(*spec));
3336 spec->sp_cr_lookup = 0;
3337 spec->sp_feat = &dt_directory_features;
3338 if (S_ISLNK(la->la_mode)) {
3339 buf = lu_buf_check_and_alloc(
3340 &mdd_env_info(env)->mti_big_buf,
3343 link_buf.lb_len = la->la_size + 1;
3344 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3345 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3347 rc = rc != 0 ? rc : -EFAULT;
3348 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3349 mdd2obd_dev(mdd)->obd_name,
3350 PFID(mdd_object_fid(mdd_sobj)), rc);
3353 spec->u.sp_symname = link_buf.lb_buf;
3354 } else if S_ISREG(la->la_mode) {
3355 /* retrieve lov of the old object */
3356 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3357 if (rc != 0 && rc != -ENODATA)
3359 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3360 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3361 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3362 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3364 } else if (S_ISDIR(la->la_mode)) {
3365 rc = mdd_links_read(env, mdd_sobj, ldata);
3366 if (rc < 0 && rc != -ENODATA)
3370 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3371 memset(mgr_ea, 0, sizeof(*mgr_ea));
3372 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3373 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3374 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3375 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3376 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3377 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3379 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3381 handle = mdd_trans_create(env, mdd);
3383 GOTO(out_free, rc = PTR_ERR(handle));
3385 /* Note: this transaction is part of migration, and it is not
3386 * the last step of migration, so we set th_local = 1 to avoid
3387 * update last rcvd for this transaction */
3388 handle->th_local = 1;
3389 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3391 (union lmv_mds_md *)info->mti_xattr_buf,
3394 GOTO(stop_trans, rc);
3396 rc = mdd_trans_start(env, mdd, handle);
3398 GOTO(stop_trans, rc);
3400 /* create the target object */
3401 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3404 GOTO(stop_trans, rc);
3406 if (S_ISDIR(la->la_mode)) {
3407 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3409 GOTO(stop_trans, rc);
3412 /* Set MIGRATE EA on the source inode, so once the migration needs
3413 * to be re-done during failover, the re-do process can locate the
3414 * target object which is already being created. */
3415 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3416 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3417 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
3419 GOTO(stop_trans, rc);
3421 /* Set immutable flag, so any modification is disabled until
3422 * the migration is done. Once the migration is interrupted,
3423 * if the resume process find the migrating object has both
3424 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3425 * flag and approve the migration */
3426 la_flag->la_valid = LA_FLAGS;
3427 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3428 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3431 mdd_trans_stop(env, mdd, rc, handle);
3433 if (lmm_buf.lb_buf != NULL)
3434 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3438 static int mdd_migrate_entries(const struct lu_env *env,
3439 struct mdd_object *mdd_sobj,
3440 struct mdd_object *mdd_tobj)
3442 struct dt_object *next = mdd_object_child(mdd_sobj);
3443 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3444 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3445 struct thandle *handle;
3447 const struct dt_it_ops *iops;
3450 struct lu_dirent *ent;
3453 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3457 if (!dt_try_as_dir(env, next))
3458 GOTO(out_ent, rc = -ENOTDIR);
3460 * iterate directories
3462 iops = &next->do_index_ops->dio_it;
3463 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3465 GOTO(out_ent, rc = PTR_ERR(it));
3467 rc = iops->load(env, it, 0);
3469 rc = iops->next(env, it);
3473 * At this point and across for-loop:
3475 * rc == 0 -> ok, proceed.
3476 * rc > 0 -> end of directory.
3480 struct mdd_object *child;
3481 char *name = mdd_env_info(env)->mti_key;
3485 bool target_exist = false;
3487 len = iops->key_size(env, it);
3491 result = iops->rec(env, it, (struct dt_rec *)ent,
3492 LUDA_FID | LUDA_TYPE);
3493 if (result == -ESTALE)
3500 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3501 recsize = le16_to_cpu(ent->lde_reclen);
3503 /* Insert new fid with target name into target dir */
3504 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3505 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3506 ent->lde_name[1] == '.'))
3509 child = mdd_object_find(env, mdd, &ent->lde_fid);
3511 GOTO(out, rc = PTR_ERR(child));
3513 is_dir = S_ISDIR(mdd_object_type(child));
3515 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3517 /* Check whether the name has been inserted to the target */
3518 if (dt_try_as_dir(env, dt_tobj)) {
3519 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3521 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3522 (struct dt_key *)name);
3523 if (unlikely(rc == 0))
3524 target_exist = true;
3527 handle = mdd_trans_create(env, mdd);
3529 GOTO(out, rc = PTR_ERR(handle));
3531 /* Note: this transaction is part of migration, and it is not
3532 * the last step of migration, so we set th_local = 1 to avoid
3533 * updating last rcvd for this transaction */
3534 handle->th_local = 1;
3535 if (likely(!target_exist)) {
3536 rc = mdo_declare_index_insert(env, mdd_tobj,
3538 mdd_object_type(child),
3544 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3550 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3555 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3559 /* Update .. for child */
3560 rc = mdo_declare_index_delete(env, child, dotdot,
3565 rc = mdo_declare_index_insert(env, child,
3566 mdd_object_fid(mdd_tobj),
3567 S_IFDIR, dotdot, handle);
3572 rc = mdd_linkea_declare_update_child(env, mdd_tobj,
3579 rc = mdd_trans_start(env, mdd, handle);
3581 CERROR("%s: transaction start failed: rc = %d\n",
3582 mdd2obd_dev(mdd)->obd_name, rc);
3586 if (likely(!target_exist)) {
3587 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3588 mdd_object_type(child),
3594 rc = mdo_ref_add(env, mdd_tobj, handle);
3601 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3606 rc = __mdd_index_delete_only(env, child, dotdot,
3611 rc = __mdd_index_insert_only(env, child,
3612 mdd_object_fid(mdd_tobj), S_IFDIR,
3618 rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
3619 strlen(name), handle);
3622 mdd_object_put(env, child);
3623 mdd_trans_stop(env, mdd, rc, handle);
3627 result = iops->next(env, it);
3628 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3629 GOTO(out, rc = -EINTR);
3631 if (result == -ESTALE)
3633 } while (result == 0);
3636 iops->fini(env, it);
3638 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3642 static int mdd_declare_update_linkea(const struct lu_env *env,
3643 struct mdd_object *mdd_pobj,
3644 struct mdd_object *mdd_sobj,
3645 struct mdd_object *mdd_tobj,
3646 struct thandle *handle,
3647 const struct lu_name *child_name)
3649 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3650 child_name, handle, 1);
3653 static int mdd_update_linkea(const struct lu_env *env,
3654 struct mdd_object *mdd_pobj,
3655 struct mdd_object *mdd_sobj,
3656 struct mdd_object *mdd_tobj,
3657 struct thandle *handle,
3658 const struct lu_name *child_name)
3660 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3661 child_name, handle, 0);
3664 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3665 struct mdd_object *mdd_pobj,
3666 struct mdd_object *mdd_sobj,
3667 struct mdd_object *mdd_tobj,
3668 const struct lu_name *lname,
3670 struct lu_attr *parent_la,
3671 struct thandle *handle)
3673 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3676 /* Revert IMMUTABLE flag */
3677 la_flag->la_valid = LA_FLAGS;
3678 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3679 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3683 /* delete entry from source dir */
3684 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3688 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3689 mdd_tobj, handle, lname);
3693 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3694 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3700 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3701 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3707 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3708 mdd_object_type(mdd_tobj),
3709 lname->ln_name, handle);
3713 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3714 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3719 /* delete old object */
3720 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3724 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3725 /* delete old object */
3726 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3729 /* set nlink to 0 */
3730 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3735 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3739 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3744 static int mdd_migrate_update_name(const struct lu_env *env,
3745 struct mdd_object *mdd_pobj,
3746 struct mdd_object *mdd_sobj,
3747 struct mdd_object *mdd_tobj,
3748 const struct lu_name *lname,
3751 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3752 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3753 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3754 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3755 struct thandle *handle;
3756 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3757 const char *name = lname->ln_name;
3761 /* update time for parent */
3762 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3763 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3764 p_la->la_valid = LA_CTIME;
3766 rc = mdd_la_get(env, mdd_sobj, so_attr);
3770 handle = mdd_trans_create(env, mdd);
3772 RETURN(PTR_ERR(handle));
3774 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3775 lname, so_attr, p_la, handle);
3777 /* If the migration can not be fit in one transaction, just
3778 * leave it in the original MDT */
3780 GOTO(stop_trans, rc = 0);
3782 GOTO(stop_trans, rc);
3785 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3786 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3787 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3788 PFID(mdd_object_fid(mdd_tobj)));
3790 rc = mdd_trans_start(env, mdd, handle);
3792 GOTO(stop_trans, rc);
3794 /* Revert IMMUTABLE flag */
3795 la_flag->la_valid = LA_FLAGS;
3796 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3797 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3799 GOTO(stop_trans, rc);
3801 /* Remove source name from source directory */
3802 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
3804 GOTO(stop_trans, rc);
3806 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3809 GOTO(stop_trans, rc);
3811 if (S_ISREG(so_attr->la_mode)) {
3812 if (so_attr->la_nlink == 1) {
3813 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3815 if (rc != 0 && rc != -ENODATA)
3816 GOTO(stop_trans, rc);
3820 /* Insert new fid with target name into target dir */
3821 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
3822 mdd_object_type(mdd_tobj), name, handle);
3824 GOTO(stop_trans, rc);
3826 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3829 GOTO(stop_trans, rc);
3831 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3832 mdo_ref_del(env, mdd_sobj, handle);
3834 mdo_ref_del(env, mdd_sobj, handle);
3836 /* Get the attr again after ref_del */
3837 rc = mdd_la_get(env, mdd_sobj, so_attr);
3839 GOTO(out_unlock, rc);
3841 ma->ma_attr = *so_attr;
3842 ma->ma_valid |= MA_INODE;
3843 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3845 GOTO(out_unlock, rc);
3847 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3849 GOTO(out_unlock, rc);
3852 mdd_write_unlock(env, mdd_sobj);
3855 mdd_trans_stop(env, mdd, rc, handle);
3861 * Check whether we should migrate the file/dir
3863 * < 0 permission check failed or other error.
3864 * = 0 the file can be migrated.
3865 * > 0 the file does not need to be migrated, mostly
3866 * for multiple link file
3868 static int mdd_migrate_sanity_check(const struct lu_env *env,
3869 struct mdd_object *pobj,
3870 const struct lu_attr *pattr,
3871 struct mdd_object *sobj,
3872 struct lu_attr *sattr)
3874 struct mdd_thread_info *info = mdd_env_info(env);
3875 struct linkea_data *ldata = &info->mti_link_data;
3877 struct lu_buf *mgr_buf;
3883 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3884 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3885 if (mgr_buf->lb_buf == NULL)
3888 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
3890 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3892 /* If the object has migrateEA, it means IMMUTE flag
3893 * is being set by previous migration process, so it
3894 * needs to override the IMMUTE flag, otherwise the
3895 * following sanity check will fail */
3896 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
3897 LMV_HASH_FLAG_MIGRATION) {
3898 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3900 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3901 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3902 mdd2obd_dev(mdd)->obd_name,
3903 PFID(mdd_object_fid(sobj)));
3907 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3908 sobj, sattr, NULL, NULL);
3912 /* Then it will check if the file should be migrated. If the file
3913 * has mulitple links, we only need migrate the file if all of its
3914 * entries has been migrated to the remote MDT */
3915 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3918 rc = mdd_links_read(env, sobj, ldata);
3920 /* For multiple links files, if there are no linkEA data at all,
3921 * means the file might be created before linkEA is enabled, and
3922 * all all of its links should not be migrated yet, otherwise
3923 * it should have some linkEA there */
3924 if (rc == -ENOENT || rc == -ENODATA)
3929 /* If it is mulitple links file, we need update the name entry for
3931 LASSERT(ldata->ld_leh != NULL);
3932 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3933 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3934 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3935 struct mdd_object *lpobj;
3936 struct lu_name lname;
3939 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3941 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3943 lpobj = mdd_object_find(env, mdd, &fid);
3944 if (IS_ERR(lpobj)) {
3945 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3946 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3951 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
3952 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
3953 PFID(&fid), lname.ln_namelen, lname.ln_name);
3954 mdd_object_put(env, lpobj);
3958 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
3959 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
3960 lname.ln_name, PFID(&fid));
3961 mdd_object_put(env, lpobj);
3969 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
3970 struct md_object *sobj, const struct lu_name *lname,
3971 struct md_object *tobj, struct md_attr *ma)
3973 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
3974 struct mdd_device *mdd = mdo2mdd(pobj);
3975 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
3976 struct mdd_object *mdd_tobj = NULL;
3977 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3978 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3982 /* If the file will being migrated, it will check whether
3983 * the file is being opened by someone else right now */
3984 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
3985 if (mdd_sobj->mod_count >= 1) {
3986 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
3987 mdd2obd_dev(mdd)->obd_name,
3988 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3989 mdd_sobj->mod_count, -EBUSY);
3990 mdd_read_unlock(env, mdd_sobj);
3991 GOTO(put, rc = -EBUSY);
3993 mdd_read_unlock(env, mdd_sobj);
3995 rc = mdd_la_get(env, mdd_sobj, so_attr);
3999 rc = mdd_la_get(env, mdd_pobj, pattr);
4003 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4010 /* Sigh, it is impossible to finish all of migration in a single
4011 * transaction, for example migrating big directory entries to the
4012 * new MDT, it needs insert all of name entries of children in the
4015 * So migration will be done in multiple steps and transactions.
4017 * 1. create an orphan object on the remote MDT in one transaction.
4018 * 2. migrate extend attributes to the new target file/directory.
4019 * 3. For directory, migrate the entries to the new MDT and update
4020 * linkEA of each children. Because we can not migrate all entries
4021 * in a single transaction, so the migrating directory will become
4022 * a striped directory during migration, so once the process is
4023 * interrupted, the directory is still accessible. (During lookup,
4024 * client will locate the name by searching both original and target
4026 * 4. Finally, update the name/FID to point to the new file/directory
4027 * in a separate transaction.
4030 /* step 1: Check whether the orphan object has been created, and create
4031 * orphan object on the remote MDT if needed */
4032 mdd_tobj = md2mdd_obj(tobj);
4033 if (!mdd_object_exists(mdd_tobj)) {
4034 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4040 /* step 2: migrate xattr */
4041 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4045 /* step 3: migrate name entries to the orphan object */
4046 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4047 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4050 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4051 OBD_FAIL_MDS_REINT_NET_REP)))
4054 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4057 /* step 4: update name entry to the new object */
4058 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4066 const struct md_dir_operations mdd_dir_ops = {
4067 .mdo_is_subdir = mdd_is_subdir,
4068 .mdo_lookup = mdd_lookup,
4069 .mdo_create = mdd_create,
4070 .mdo_rename = mdd_rename,
4071 .mdo_link = mdd_link,
4072 .mdo_unlink = mdd_unlink,
4073 .mdo_create_data = mdd_create_data,
4074 .mdo_migrate = mdd_migrate,