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, 2012, Whamcloud, Inc.
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@clusterfs.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 = {
60 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
61 const struct lu_name *lname, struct lu_fid* fid,
63 static int mdd_declare_links_add(const struct lu_env *env,
64 struct mdd_object *mdd_obj,
65 struct thandle *handle);
66 static int mdd_links_add(const struct lu_env *env,
67 struct mdd_object *mdd_obj,
68 const struct lu_fid *pfid,
69 const struct lu_name *lname,
70 struct thandle *handle, int first);
71 static int mdd_links_rename(const struct lu_env *env,
72 struct mdd_object *mdd_obj,
73 const struct lu_fid *oldpfid,
74 const struct lu_name *oldlname,
75 const struct lu_fid *newpfid,
76 const struct lu_name *newlname,
77 struct thandle *handle);
80 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
81 const struct lu_name *lname, struct lu_fid* fid, int mask)
83 const char *name = lname->ln_name;
84 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
85 struct dynlock_handle *dlh;
88 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
89 if (unlikely(dlh == NULL))
91 rc = __mdd_lookup(env, pobj, lname, fid, mask);
92 mdd_pdo_read_unlock(env, mdd_obj, dlh);
97 int mdd_lookup(const struct lu_env *env,
98 struct md_object *pobj, const struct lu_name *lname,
99 struct lu_fid* fid, struct md_op_spec *spec)
103 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
107 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
110 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
114 * For root fid use special function, which does not compare version component
115 * of fid. Version component is different for root fids on all MDTs.
117 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
119 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
120 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
124 * return 1: if lf is the fid of the ancestor of p1;
127 * return -EREMOTE: if remote object is found, in this
128 * case fid of remote object is saved to @pf;
130 * otherwise: values < 0, errors.
132 static int mdd_is_parent(const struct lu_env *env,
133 struct mdd_device *mdd,
134 struct mdd_object *p1,
135 const struct lu_fid *lf,
138 struct mdd_object *parent = NULL;
143 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
144 pfid = &mdd_env_info(env)->mti_fid;
146 /* Check for root first. */
147 if (mdd_is_root(mdd, mdo2fid(p1)))
151 /* this is done recursively, bypass capa for each obj */
152 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
153 rc = mdd_parent_fid(env, p1, pfid);
156 if (mdd_is_root(mdd, pfid))
158 if (lu_fid_eq(pfid, lf))
161 mdd_object_put(env, parent);
162 parent = mdd_object_find(env, mdd, pfid);
164 /* cross-ref parent */
165 if (parent == NULL) {
168 GOTO(out, rc = -EREMOTE);
169 } else if (IS_ERR(parent))
170 GOTO(out, rc = PTR_ERR(parent));
175 if (parent && !IS_ERR(parent))
176 mdd_object_put(env, parent);
181 * No permission check is needed.
183 * returns 1: if fid is ancestor of @mo;
184 * returns 0: if fid is not a ancestor of @mo;
186 * returns EREMOTE if remote object is found, fid of remote object is saved to
189 * returns < 0: if error
191 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
192 const struct lu_fid *fid, struct lu_fid *sfid)
194 struct mdd_device *mdd = mdo2mdd(mo);
198 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
201 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
205 } else if (rc == 1) {
206 /* found @fid is parent */
214 * Check that @dir contains no entries except (possibly) dot and dotdot.
219 * -ENOTDIR not a directory object
220 * -ENOTEMPTY not empty
224 static int mdd_dir_is_empty(const struct lu_env *env,
225 struct mdd_object *dir)
228 struct dt_object *obj;
229 const struct dt_it_ops *iops;
233 obj = mdd_object_child(dir);
234 if (!dt_try_as_dir(env, obj))
237 iops = &obj->do_index_ops->dio_it;
238 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
240 result = iops->get(env, it, (const void *)"");
243 for (result = 0, i = 0; result == 0 && i < 3; ++i)
244 result = iops->next(env, it);
247 else if (result == +1)
249 } else if (result == 0)
251 * Huh? Index contains no zero key?
258 result = PTR_ERR(it);
262 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
264 struct mdd_device *m = mdd_obj2mdd_dev(obj);
265 struct lu_attr *la = &mdd_env_info(env)->mti_la;
269 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
274 * Subdir count limitation can be broken through.
276 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
277 !S_ISDIR(la->la_mode))
284 * Check whether it may create the cobj under the pobj.
287 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
288 struct mdd_object *cobj, int check_perm, int check_nlink)
293 if (cobj && mdd_object_exists(cobj))
296 if (mdd_is_dead_obj(pobj))
300 rc = mdd_permission_internal_locked(env, pobj, NULL,
301 MAY_WRITE | MAY_EXEC,
303 if (!rc && check_nlink)
304 rc = __mdd_may_link(env, pobj);
310 * Check whether can unlink from the pobj in the case of "cobj == NULL".
312 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
313 const struct lu_attr *attr)
318 if (mdd_is_dead_obj(pobj))
321 if ((attr->la_valid & LA_FLAGS) &&
322 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
325 rc = mdd_permission_internal_locked(env, pobj, NULL,
326 MAY_WRITE | MAY_EXEC,
331 if (mdd_is_append(pobj))
338 * pobj == NULL is remote ops case, under such case, pobj's
339 * VTX feature has been checked already, no need check again.
341 static inline int mdd_is_sticky(const struct lu_env *env,
342 struct mdd_object *pobj,
343 struct mdd_object *cobj)
345 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
346 struct md_ucred *uc = md_ucred(env);
350 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
354 if (!(tmp_la->la_mode & S_ISVTX) ||
355 (tmp_la->la_uid == uc->mu_fsuid))
359 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
363 if (tmp_la->la_uid == uc->mu_fsuid)
366 return !mdd_capable(uc, CFS_CAP_FOWNER);
370 * Check whether it may delete the cobj from the pobj.
373 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
374 struct mdd_object *cobj, struct lu_attr *cattr,
375 struct lu_attr *src_attr, int check_perm, int check_empty)
381 if (!mdd_object_exists(cobj))
384 if (mdd_is_dead_obj(cobj))
388 if (!mdd_object_exists(pobj))
391 if (mdd_is_dead_obj(pobj))
395 rc = mdd_permission_internal_locked(env, pobj, NULL,
396 MAY_WRITE | MAY_EXEC,
402 if (mdd_is_append(pobj))
406 if (mdd_is_sticky(env, pobj, cobj))
409 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
412 if ((cattr->la_valid & LA_FLAGS) &&
413 (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
416 /* additional check the rename case */
418 if (S_ISDIR(src_attr->la_mode)) {
419 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
421 if (!S_ISDIR(cattr->la_mode))
424 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
426 } else if (S_ISDIR(cattr->la_mode))
430 if (S_ISDIR(cattr->la_mode) && check_empty)
431 rc = mdd_dir_is_empty(env, cobj);
438 * has mdd_write_lock on src already, but not on tgt yet
440 int mdd_link_sanity_check(const struct lu_env *env,
441 struct mdd_object *tgt_obj,
442 const struct lu_name *lname,
443 struct mdd_object *src_obj)
445 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
449 if (!mdd_object_exists(src_obj))
452 if (mdd_is_dead_obj(src_obj))
455 /* Local ops, no lookup before link, check filename length here. */
456 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
457 RETURN(-ENAMETOOLONG);
459 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
462 if (S_ISDIR(mdd_object_type(src_obj)))
465 LASSERT(src_obj != tgt_obj);
467 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
472 rc = __mdd_may_link(env, src_obj);
477 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
478 const char *name, struct thandle *handle,
479 struct lustre_capa *capa)
481 struct dt_object *next = mdd_object_child(pobj);
485 if (dt_try_as_dir(env, next)) {
486 rc = next->do_index_ops->dio_delete(env, next,
487 (struct dt_key *)name,
495 static int __mdd_index_insert_only(const struct lu_env *env,
496 struct mdd_object *pobj,
497 const struct lu_fid *lf, const char *name,
498 struct thandle *handle,
499 struct lustre_capa *capa)
501 struct dt_object *next = mdd_object_child(pobj);
505 if (dt_try_as_dir(env, next)) {
506 struct md_ucred *uc = md_ucred(env);
508 rc = next->do_index_ops->dio_insert(env, next,
510 (const struct dt_key *)name,
511 handle, capa, uc->mu_cap &
512 CFS_CAP_SYS_RESOURCE_MASK);
519 /* insert named index, add reference if isdir */
520 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
521 const struct lu_fid *lf, const char *name, int is_dir,
522 struct thandle *handle, struct lustre_capa *capa)
527 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
528 if (rc == 0 && is_dir) {
529 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
530 mdo_ref_add(env, pobj, handle);
531 mdd_write_unlock(env, pobj);
536 /* delete named index, drop reference if isdir */
537 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
538 const char *name, int is_dir, struct thandle *handle,
539 struct lustre_capa *capa)
544 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
545 if (rc == 0 && is_dir) {
546 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
547 mdo_ref_del(env, pobj, handle);
548 mdd_write_unlock(env, pobj);
554 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
555 int reclen, struct thandle *handle)
559 /* XXX: this is a temporary solution to declare llog changes
560 * will be fixed in 2.3 with new llog implementation */
562 LASSERT(mdd->mdd_capa);
564 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
565 * have to set the parameter 'size' as INT_MAX or 0 to inform
566 * OSD that this record write is for a llog write or catalog
567 * header update, and osd declare function will reserve less
568 * credits for optimization purpose.
570 * Reserve 6 blocks for a llog write, since the llog file is
571 * usually small, reserve 2 blocks for catalog header update,
572 * because we know for sure that catalog header is already
575 * This hack should be removed in 2.3.
579 rc = dt_declare_record_write(env, mdd->mdd_capa,
580 DECLARE_LLOG_WRITE, 0, handle);
584 /* header will be updated as well */
585 rc = dt_declare_record_write(env, mdd->mdd_capa,
586 DECLARE_LLOG_WRITE, 0, handle);
590 /* also we should be able to create new plain log */
591 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
595 /* new record referencing new plain llog */
596 rc = dt_declare_record_write(env, mdd->mdd_capa,
597 DECLARE_LLOG_WRITE, 0, handle);
601 /* catalog's header will be updated as well */
602 rc = dt_declare_record_write(env, mdd->mdd_capa,
603 DECLARE_LLOG_REWRITE, 0, handle);
608 int mdd_declare_changelog_store(const struct lu_env *env,
609 struct mdd_device *mdd,
610 const struct lu_name *fname,
611 struct thandle *handle)
613 struct obd_device *obd = mdd2obd_dev(mdd);
614 struct llog_ctxt *ctxt;
615 struct llog_changelog_rec *rec;
621 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
624 reclen = llog_data_len(sizeof(*rec) +
625 (fname != NULL ? fname->ln_namelen : 0));
626 buf = mdd_buf_alloc(env, reclen);
627 if (buf->lb_buf == NULL)
631 rec->cr_hdr.lrh_len = reclen;
632 rec->cr_hdr.lrh_type = CHANGELOG_REC;
634 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
638 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
644 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
645 struct mdd_device *mdd,
646 const struct lu_name *tname,
647 const struct lu_name *sname,
648 struct thandle *handle)
650 struct obd_device *obd = mdd2obd_dev(mdd);
651 struct llog_ctxt *ctxt;
652 struct llog_changelog_ext_rec *rec;
658 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
661 reclen = llog_data_len(sizeof(*rec) +
662 (tname != NULL ? tname->ln_namelen : 0) +
663 (sname != NULL ? 1 + sname->ln_namelen : 0));
664 buf = mdd_buf_alloc(env, reclen);
665 if (buf->lb_buf == NULL)
669 rec->cr_hdr.lrh_len = reclen;
670 rec->cr_hdr.lrh_type = CHANGELOG_REC;
672 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
676 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
682 /** Add a changelog entry \a rec to the changelog llog
685 * \param handle - currently ignored since llogs start their own transaction;
686 * this will hopefully be fixed in llog rewrite
689 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
690 struct llog_changelog_rec *rec, struct thandle *th)
692 struct obd_device *obd = mdd2obd_dev(mdd);
693 struct llog_ctxt *ctxt;
696 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
697 rec->cr_hdr.lrh_type = CHANGELOG_REC;
698 rec->cr.cr_time = cl_time();
700 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
701 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
702 * but as long as the MDD transactions are ordered correctly for e.g.
703 * rename conflicts, I don't think this should matter. */
704 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
705 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
707 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
711 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
718 /** Add a changelog_ext entry \a rec to the changelog llog
721 * \param handle - currently ignored since llogs start their own transaction;
722 * this will hopefully be fixed in llog rewrite
725 int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
726 struct llog_changelog_ext_rec *rec,
729 struct obd_device *obd = mdd2obd_dev(mdd);
730 struct llog_ctxt *ctxt;
733 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
734 /* llog_lvfs_write_rec sets the llog tail len */
735 rec->cr_hdr.lrh_type = CHANGELOG_REC;
736 rec->cr.cr_time = cl_time();
738 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
739 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
740 * but as long as the MDD transactions are ordered correctly for e.g.
741 * rename conflicts, I don't think this should matter. */
742 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
743 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
745 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
749 /* nested journal transaction */
750 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
758 /** Store a namespace change changelog record
759 * If this fails, we must fail the whole transaction; we don't
760 * want the change to commit without the log entry.
761 * \param target - mdd_object of change
762 * \param parent - parent dir/object
763 * \param tname - target name string
764 * \param handle - transacion handle
766 static int mdd_changelog_ns_store(const struct lu_env *env,
767 struct mdd_device *mdd,
768 enum changelog_rec_type type,
770 struct mdd_object *target,
771 struct mdd_object *parent,
772 const struct lu_name *tname,
773 struct thandle *handle)
775 struct llog_changelog_rec *rec;
782 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
784 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
787 LASSERT(target != NULL);
788 LASSERT(parent != NULL);
789 LASSERT(tname != NULL);
790 LASSERT(handle != NULL);
792 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
793 buf = mdd_buf_alloc(env, reclen);
794 if (buf->lb_buf == NULL)
798 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
799 rec->cr.cr_type = (__u32)type;
800 rec->cr.cr_tfid = *mdo2fid(target);
801 rec->cr.cr_pfid = *mdo2fid(parent);
802 rec->cr.cr_namelen = tname->ln_namelen;
803 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
805 target->mod_cltime = cfs_time_current_64();
807 rc = mdd_changelog_store(env, mdd, rec, handle);
809 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
810 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
811 PFID(&rec->cr.cr_pfid));
819 /** Store a namespace change changelog record
820 * If this fails, we must fail the whole transaction; we don't
821 * want the change to commit without the log entry.
822 * \param target - mdd_object of change
823 * \param tpfid - target parent dir/object fid
824 * \param sfid - source object fid
825 * \param spfid - source parent fid
826 * \param tname - target name string
827 * \param sname - source name string
828 * \param handle - transacion handle
830 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
831 struct mdd_device *mdd,
832 enum changelog_rec_type type,
834 struct mdd_object *target,
835 const struct lu_fid *tpfid,
836 const struct lu_fid *sfid,
837 const struct lu_fid *spfid,
838 const struct lu_name *tname,
839 const struct lu_name *sname,
840 struct thandle *handle)
842 struct llog_changelog_ext_rec *rec;
849 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
851 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
854 LASSERT(sfid != NULL);
855 LASSERT(tpfid != NULL);
856 LASSERT(tname != NULL);
857 LASSERT(handle != NULL);
859 reclen = llog_data_len(sizeof(*rec) +
860 sname != NULL ? 1 + sname->ln_namelen : 0);
861 buf = mdd_buf_alloc(env, reclen);
862 if (buf->lb_buf == NULL)
866 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
867 rec->cr.cr_type = (__u32)type;
868 rec->cr.cr_pfid = *tpfid;
869 rec->cr.cr_sfid = *sfid;
870 rec->cr.cr_spfid = *spfid;
871 rec->cr.cr_namelen = tname->ln_namelen;
872 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
874 LASSERT(sfid != NULL);
875 rec->cr.cr_name[tname->ln_namelen] = '\0';
876 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
878 rec->cr.cr_namelen += 1 + sname->ln_namelen;
881 if (likely(target != NULL)) {
882 rec->cr.cr_tfid = *mdo2fid(target);
883 target->mod_cltime = cfs_time_current_64();
885 fid_zero(&rec->cr.cr_tfid);
888 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
890 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
891 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
898 static int mdd_declare_link(const struct lu_env *env,
899 struct mdd_device *mdd,
900 struct mdd_object *p,
901 struct mdd_object *c,
902 const struct lu_name *name,
903 struct thandle *handle)
907 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
911 rc = mdo_declare_ref_add(env, c, handle);
915 rc = mdo_declare_attr_set(env, p, NULL, handle);
919 rc = mdo_declare_attr_set(env, c, NULL, handle);
923 rc = mdd_declare_links_add(env, c, handle);
927 rc = mdd_declare_changelog_store(env, mdd, name, handle);
932 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
933 struct md_object *src_obj, const struct lu_name *lname,
936 const char *name = lname->ln_name;
937 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
938 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
939 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
940 struct mdd_device *mdd = mdo2mdd(src_obj);
941 struct dynlock_handle *dlh;
942 struct thandle *handle;
946 handle = mdd_trans_create(env, mdd);
948 GOTO(out_pending, rc = PTR_ERR(handle));
950 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
954 rc = mdd_trans_start(env, mdd, handle);
958 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
960 GOTO(out_trans, rc = -ENOMEM);
961 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
963 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
965 GOTO(out_unlock, rc);
967 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
969 mdd_object_capa(env, mdd_tobj));
971 GOTO(out_unlock, rc);
973 rc = mdo_ref_add(env, mdd_sobj, handle);
975 __mdd_index_delete_only(env, mdd_tobj, name, handle,
976 mdd_object_capa(env, mdd_tobj));
977 GOTO(out_unlock, rc);
980 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
981 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
983 la->la_valid = LA_CTIME | LA_MTIME;
984 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
986 GOTO(out_unlock, rc);
988 la->la_valid = LA_CTIME;
989 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
991 mdd_links_add(env, mdd_sobj,
992 mdo2fid(mdd_tobj), lname, handle, 0);
997 mdd_write_unlock(env, mdd_sobj);
998 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
1001 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1002 mdd_tobj, lname, handle);
1004 mdd_trans_stop(env, mdd, rc, handle);
1009 int mdd_declare_finish_unlink(const struct lu_env *env,
1010 struct mdd_object *obj,
1012 struct thandle *handle)
1016 rc = orph_declare_index_insert(env, obj, handle);
1020 return mdo_declare_destroy(env, obj, handle);
1023 /* caller should take a lock before calling */
1024 int mdd_finish_unlink(const struct lu_env *env,
1025 struct mdd_object *obj, struct md_attr *ma,
1029 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1032 LASSERT(mdd_write_locked(env, obj) != 0);
1034 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
1035 obj->mod_flags |= DEAD_OBJ;
1036 /* add new orphan and the object
1037 * will be deleted during mdd_close() */
1038 if (obj->mod_count) {
1039 rc = __mdd_orphan_add(env, obj, th);
1041 CDEBUG(D_HA, "Object "DFID" is inserted into "
1042 "orphan list, open count = %d\n",
1043 PFID(mdd_object_fid(obj)),
1046 CERROR("Object "DFID" fail to be an orphan, "
1047 "open count = %d, maybe cause failed "
1049 PFID(mdd_object_fid(obj)),
1052 rc = mdo_destroy(env, obj, th);
1061 * has mdd_write_lock on cobj already, but not on pobj yet
1063 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1064 struct mdd_object *cobj, struct lu_attr *cattr)
1069 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1074 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1075 struct mdd_object *p, struct mdd_object *c,
1076 const struct lu_name *name, struct md_attr *ma,
1077 struct thandle *handle)
1081 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1085 rc = mdo_declare_ref_del(env, p, handle);
1089 rc = mdo_declare_ref_del(env, c, handle);
1093 rc = mdo_declare_ref_del(env, c, handle);
1097 rc = mdo_declare_attr_set(env, p, NULL, handle);
1101 rc = mdo_declare_attr_set(env, c, NULL, handle);
1105 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1109 rc = mdd_declare_links_add(env, c, handle);
1113 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1118 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1119 struct md_object *cobj, const struct lu_name *lname,
1122 const char *name = lname->ln_name;
1123 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1124 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1125 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1126 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
1127 struct mdd_device *mdd = mdo2mdd(pobj);
1128 struct dynlock_handle *dlh;
1129 struct thandle *handle;
1133 if (mdd_object_exists(mdd_cobj) <= 0)
1136 handle = mdd_trans_create(env, mdd);
1138 RETURN(PTR_ERR(handle));
1140 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1145 rc = mdd_trans_start(env, mdd, handle);
1149 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1151 GOTO(stop, rc = -ENOMEM);
1152 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1155 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1159 is_dir = S_ISDIR(cattr->la_mode);
1161 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1165 rc = mdo_ref_del(env, mdd_cobj, handle);
1167 __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj),
1169 mdd_object_capa(env, mdd_pobj));
1173 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1174 mdd_object_capa(env, mdd_pobj));
1180 mdo_ref_del(env, mdd_cobj, handle);
1182 /* fetch updated nlink */
1183 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1187 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1188 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1190 la->la_valid = LA_CTIME | LA_MTIME;
1191 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1195 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1196 /* update ctime of an unlinked file only if it is still
1197 * opened or a link still exists */
1198 la->la_valid = LA_CTIME;
1199 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1204 /* XXX: this transfer to ma will be removed with LOD/OSP */
1205 ma->ma_attr = *cattr;
1206 ma->ma_valid |= MA_INODE;
1207 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1209 /* fetch updated nlink */
1211 rc = mdd_la_get(env, mdd_cobj, cattr,
1212 mdd_object_capa(env, mdd_cobj));
1215 /* old files may not have link ea; ignore errors */
1216 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
1217 lname, NULL, NULL, handle);
1219 /* if object is removed then we can't get its attrs, use last get */
1220 if (cattr->la_nlink == 0) {
1221 ma->ma_attr = *cattr;
1222 ma->ma_valid |= MA_INODE;
1226 mdd_write_unlock(env, mdd_cobj);
1227 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1231 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1232 if ((ma->ma_valid & MA_HSM) &&
1233 (ma->ma_hsm.mh_flags & HS_EXISTS))
1234 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1236 rc = mdd_changelog_ns_store(env, mdd,
1237 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1238 mdd_cobj, mdd_pobj, lname, handle);
1242 mdd_trans_stop(env, mdd, rc, handle);
1248 * The permission has been checked when obj created, no need check again.
1250 static int mdd_cd_sanity_check(const struct lu_env *env,
1251 struct mdd_object *obj)
1256 if (!obj || mdd_is_dead_obj(obj))
1263 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1264 struct md_object *cobj, const struct md_op_spec *spec,
1267 struct mdd_device *mdd = mdo2mdd(cobj);
1268 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1269 struct mdd_object *son = md2mdd_obj(cobj);
1270 struct thandle *handle;
1271 const struct lu_buf *buf;
1272 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1276 /* do not let users to create stripes via .lustre/
1277 * mdd_obf_setup() sets IMMUTE_OBJ on this directory */
1278 if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ)
1281 rc = mdd_cd_sanity_check(env, son);
1285 if (!md_should_create(spec->sp_cr_flags))
1289 * there are following use cases for this function:
1290 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1291 * striping can be specified or not
1294 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1298 /* calling ->ah_make_hint() is used to transfer information from parent */
1299 mdd_object_make_hint(env, mdd_pobj, son, attr);
1301 handle = mdd_trans_create(env, mdd);
1303 GOTO(out_free, rc = PTR_ERR(handle));
1306 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1307 * Should this be fixed?
1309 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, no_create %u\n",
1310 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1311 spec->sp_cr_flags, spec->no_create);
1313 if (spec->no_create) {
1315 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1316 spec->u.sp_ea.eadatalen);
1317 } else if (!(spec->sp_cr_flags & MDS_OPEN_HAS_OBJS)) {
1318 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1320 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1321 spec->u.sp_ea.eadatalen);
1326 /* MDS_OPEN_HAS_OBJS is not used anymore ? */
1330 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1331 XATTR_NAME_LOV, 0, handle);
1335 rc = mdd_trans_start(env, mdd, handle);
1339 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1340 0, handle, mdd_object_capa(env, son));
1342 mdd_trans_stop(env, mdd, rc, handle);
1347 /* Get fid from name and parent */
1349 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1350 const struct lu_name *lname, struct lu_fid* fid, int mask)
1352 const char *name = lname->ln_name;
1353 const struct dt_key *key = (const struct dt_key *)name;
1354 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1355 struct mdd_device *m = mdo2mdd(pobj);
1356 struct dt_object *dir = mdd_object_child(mdd_obj);
1360 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1363 rc = mdd_object_exists(mdd_obj);
1364 if (unlikely(rc == 0))
1366 else if (unlikely(rc < 0)) {
1367 CERROR("Object "DFID" locates on remote server\n",
1368 PFID(mdo2fid(mdd_obj)));
1372 /* The common filename length check. */
1373 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1374 RETURN(-ENAMETOOLONG);
1376 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1381 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1382 dt_try_as_dir(env, dir))) {
1384 rc = dir->do_index_ops->dio_lookup(env, dir,
1385 (struct dt_rec *)fid, key,
1386 mdd_object_capa(env, mdd_obj));
1397 int mdd_declare_object_initialize(const struct lu_env *env,
1398 struct mdd_object *child,
1399 struct lu_attr *attr,
1400 struct thandle *handle)
1404 rc = mdo_declare_attr_set(env, child, attr, handle);
1405 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1406 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1409 rc = mdo_declare_ref_add(env, child, handle);
1413 mdd_declare_links_add(env, child, handle);
1418 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1419 const struct lu_name *lname, struct mdd_object *child,
1420 struct lu_attr *attr, struct thandle *handle,
1421 const struct md_op_spec *spec)
1427 * Update attributes for child.
1430 * (1) the valid bits should be converted between Lustre and Linux;
1431 * (2) maybe, the child attributes should be set in OSD when creation.
1434 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1438 if (S_ISDIR(attr->la_mode)) {
1439 /* Add "." and ".." for newly created dir */
1440 mdo_ref_add(env, child, handle);
1441 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1442 dot, handle, BYPASS_CAPA);
1444 rc = __mdd_index_insert_only(env, child, pfid,
1448 mdo_ref_del(env, child, handle);
1451 mdd_links_add(env, child, pfid, lname, handle, 1);
1456 /* has not lock on pobj yet */
1457 static int mdd_create_sanity_check(const struct lu_env *env,
1458 struct md_object *pobj,
1459 struct lu_attr *pattr,
1460 const struct lu_name *lname,
1461 struct lu_attr *cattr,
1462 struct md_op_spec *spec)
1464 struct mdd_thread_info *info = mdd_env_info(env);
1465 struct lu_fid *fid = &info->mti_fid;
1466 struct mdd_object *obj = md2mdd_obj(pobj);
1467 struct mdd_device *m = mdo2mdd(pobj);
1468 int lookup = spec->sp_cr_lookup;
1473 if (mdd_is_dead_obj(obj))
1477 * In some cases this lookup is not needed - we know before if name
1478 * exists or not because MDT performs lookup for it.
1479 * name length check is done in lookup.
1483 * Check if the name already exist, though it will be checked in
1484 * _index_insert also, for avoiding rolling back if exists
1487 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1488 MAY_WRITE | MAY_EXEC);
1490 RETURN(rc ? : -EEXIST);
1493 * Check WRITE permission for the parent.
1494 * EXEC permission have been checked
1495 * when lookup before create already.
1497 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1504 if (pattr->la_mode & S_ISGID) {
1505 cattr->la_gid = pattr->la_gid;
1506 if (S_ISDIR(cattr->la_mode)) {
1507 cattr->la_mode |= S_ISGID;
1508 cattr->la_valid |= LA_MODE;
1512 switch (cattr->la_mode & S_IFMT) {
1514 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1516 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1517 RETURN(-ENAMETOOLONG);
1536 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1537 struct mdd_object *p, struct mdd_object *c,
1538 const struct lu_name *name,
1539 struct lu_attr *attr,
1541 struct thandle *handle,
1542 const struct md_op_spec *spec)
1544 struct mdd_thread_info *info = mdd_env_info(env);
1547 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1551 #ifdef CONFIG_FS_POSIX_ACL
1552 if (got_def_acl > 0) {
1553 struct lu_buf *acl_buf;
1555 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1556 /* if dir, then can inherit default ACl */
1557 if (S_ISDIR(attr->la_mode)) {
1558 rc = mdo_declare_xattr_set(env, c, acl_buf,
1559 XATTR_NAME_ACL_DEFAULT,
1565 rc = mdo_declare_attr_set(env, c, &info->mti_pattr, handle);
1569 rc = mdo_declare_xattr_set(env, c, acl_buf,
1570 XATTR_NAME_ACL_ACCESS, 0, handle);
1576 if (S_ISDIR(attr->la_mode)) {
1577 rc = mdo_declare_ref_add(env, p, handle);
1582 rc = mdd_declare_object_initialize(env, c, attr, handle);
1586 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1587 name->ln_name, handle);
1591 /* replay case, create LOV EA from client data */
1592 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1593 const struct lu_buf *buf;
1595 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1596 spec->u.sp_ea.eadatalen);
1597 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1603 if (S_ISLNK(attr->la_mode)) {
1604 rc = dt_declare_record_write(env, mdd_object_child(c),
1605 strlen(spec->u.sp_symname), 0,
1611 rc = mdo_declare_attr_set(env, p, attr, handle);
1615 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1625 * Create object and insert it into namespace.
1627 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1628 const struct lu_name *lname, struct md_object *child,
1629 struct md_op_spec *spec, struct md_attr* ma)
1631 struct mdd_thread_info *info = mdd_env_info(env);
1632 struct lu_attr *la = &info->mti_la_for_fix;
1633 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1634 struct mdd_object *son = md2mdd_obj(child);
1635 struct mdd_device *mdd = mdo2mdd(pobj);
1636 struct lu_attr *attr = &ma->ma_attr;
1637 struct thandle *handle;
1638 struct lu_attr *pattr = &info->mti_pattr;
1639 struct dynlock_handle *dlh;
1640 const char *name = lname->ln_name;
1641 int rc, created = 0, initialized = 0, inserted = 0;
1642 int got_def_acl = 0;
1646 * Two operations have to be performed:
1648 * - an allocation of a new object (->do_create()), and
1650 * - an insertion into a parent index (->dio_insert()).
1652 * Due to locking, operation order is not important, when both are
1653 * successful, *but* error handling cases are quite different:
1655 * - if insertion is done first, and following object creation fails,
1656 * insertion has to be rolled back, but this operation might fail
1657 * also leaving us with dangling index entry.
1659 * - if creation is done first, is has to be undone if insertion
1660 * fails, leaving us with leaked space, which is neither good, nor
1663 * It seems that creation-first is simplest solution, but it is
1664 * sub-optimal in the frequent
1669 * case, because second mkdir is bound to create object, only to
1670 * destroy it immediately.
1672 * To avoid this follow local file systems that do double lookup:
1674 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1676 * 1. create (mdd_object_create_internal())
1678 * 2. insert (__mdd_index_insert(), lookup again)
1681 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1685 /* Sanity checks before big job. */
1686 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
1690 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1691 GOTO(out_free, rc = -EINPROGRESS);
1693 if (!S_ISLNK(attr->la_mode)) {
1694 struct lu_buf *acl_buf;
1696 acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
1697 sizeof(info->mti_xattr_buf));
1698 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1699 rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
1700 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1701 mdd_read_unlock(env, mdd_pobj);
1704 else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
1708 mdd_object_make_hint(env, mdd_pobj, son, attr);
1710 handle = mdd_trans_create(env, mdd);
1712 GOTO(out_free, rc = PTR_ERR(handle));
1714 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
1715 got_def_acl, handle, spec);
1719 rc = mdd_trans_start(env, mdd, handle);
1723 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1725 GOTO(out_trans, rc = -ENOMEM);
1727 mdd_write_lock(env, son, MOR_TGT_CHILD);
1728 rc = mdd_object_create_internal(env, mdd_pobj, son, attr, handle, spec);
1730 mdd_write_unlock(env, son);
1736 #ifdef CONFIG_FS_POSIX_ACL
1738 struct lu_buf *acl_buf;
1740 acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
1741 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1743 mdd_write_unlock(env, son);
1749 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1750 son, attr, handle, spec);
1753 * in case of replay we just set LOVEA provided by the client
1754 * XXX: I think it would be interesting to try "old" way where
1755 * MDT calls this xattr_set(LOV) in a different transaction.
1756 * probably this way we code can be made better.
1759 (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
1760 const struct lu_buf *buf;
1762 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1763 spec->u.sp_ea.eadatalen);
1764 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
1767 mdd_write_unlock(env, son);
1770 * Object has no links, so it will be destroyed when last
1771 * reference is released. (XXX not now.)
1777 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1778 name, S_ISDIR(attr->la_mode), handle,
1779 mdd_object_capa(env, mdd_pobj));
1785 if (S_ISLNK(attr->la_mode)) {
1786 struct md_ucred *uc = md_ucred(env);
1787 struct dt_object *dt = mdd_object_child(son);
1788 const char *target_name = spec->u.sp_symname;
1789 int sym_len = strlen(target_name);
1790 const struct lu_buf *buf;
1793 buf = mdd_buf_get_const(env, target_name, sym_len);
1794 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1795 mdd_object_capa(env, son),
1797 CFS_CAP_SYS_RESOURCE_MASK);
1802 GOTO(cleanup, rc = -EFAULT);
1806 la->la_valid = LA_CTIME | LA_MTIME;
1807 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1813 if (rc != 0 && created != 0) {
1816 if (inserted != 0) {
1817 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1818 S_ISDIR(attr->la_mode),
1819 handle, BYPASS_CAPA);
1824 mdd_write_lock(env, son, MOR_TGT_CHILD);
1825 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
1826 /* Drop the reference, no need to delete "."/"..",
1827 * because the object to be destroied directly. */
1828 rc2 = mdo_ref_del(env, son, handle);
1830 mdd_write_unlock(env, son);
1835 rc2 = mdo_ref_del(env, son, handle);
1837 mdd_write_unlock(env, son);
1841 mdo_destroy(env, son, handle);
1842 mdd_write_unlock(env, son);
1845 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1848 rc = mdd_changelog_ns_store(env, mdd,
1849 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1850 S_ISREG(attr->la_mode) ? CL_CREATE :
1851 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1852 0, son, mdd_pobj, lname, handle);
1854 mdd_trans_stop(env, mdd, rc, handle);
1856 /* The child object shouldn't be cached anymore */
1858 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
1859 &child->mo_lu.lo_header->loh_flags);
1864 * Get locks on parents in proper order
1865 * RETURN: < 0 - error, rename_order if successful
1873 static int mdd_rename_order(const struct lu_env *env,
1874 struct mdd_device *mdd,
1875 struct mdd_object *src_pobj,
1876 struct mdd_object *tgt_pobj)
1878 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1882 if (src_pobj == tgt_pobj)
1883 RETURN(MDD_RN_SAME);
1885 /* compared the parent child relationship of src_p&tgt_p */
1886 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1888 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1891 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1904 /* has not mdd_write{read}_lock on any obj yet. */
1905 static int mdd_rename_sanity_check(const struct lu_env *env,
1906 struct mdd_object *src_pobj,
1907 struct mdd_object *tgt_pobj,
1908 struct mdd_object *sobj,
1909 struct mdd_object *tobj,
1910 struct lu_attr *so_attr,
1911 struct lu_attr *tg_attr)
1916 /* XXX: when get here, sobj must NOT be NULL,
1917 * the other case has been processed in cml_rename
1918 * before mdd_rename and enable MDS_PERM_BYPASS. */
1921 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
1925 /* XXX: when get here, "tobj == NULL" means tobj must
1926 * NOT exist (neither on remote MDS, such case has been
1927 * processed in cml_rename before mdd_rename and enable
1929 * So check may_create, but not check may_unlink. */
1931 rc = mdd_may_create(env, tgt_pobj, NULL,
1932 (src_pobj != tgt_pobj), 0);
1934 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
1935 (src_pobj != tgt_pobj), 1);
1937 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1938 S_ISDIR(so_attr->la_mode))
1939 rc = __mdd_may_link(env, tgt_pobj);
1944 static int mdd_declare_rename(const struct lu_env *env,
1945 struct mdd_device *mdd,
1946 struct mdd_object *mdd_spobj,
1947 struct mdd_object *mdd_tpobj,
1948 struct mdd_object *mdd_sobj,
1949 struct mdd_object *mdd_tobj,
1950 const struct lu_name *tname,
1951 const struct lu_name *sname,
1953 struct thandle *handle)
1961 /* name from source dir */
1962 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
1966 /* .. from source child */
1967 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
1968 /* source child can be directory,
1969 * counted by source dir's nlink */
1970 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
1974 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
1978 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
1983 /* new target child can be directory,
1984 * counted by target dir's nlink */
1985 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
1991 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
1995 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
1998 mdd_declare_links_add(env, mdd_sobj, handle);
2002 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2007 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2008 tname->ln_name, handle);
2012 /* name from target dir (old name), we declare it unconditionally
2013 * as mdd_rename() calls delete unconditionally as well. so just
2014 * to balance declarations vs calls to change ... */
2015 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2019 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2020 /* delete target child in target parent directory */
2021 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2025 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2026 /* target child can be directory,
2027 * delete "." reference in target child directory */
2028 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2032 /* delete ".." reference in target parent directory */
2033 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2038 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2042 mdd_declare_links_add(env, mdd_tobj, handle);
2046 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2051 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2058 /* src object can be remote that is why we use only fid and type of object */
2059 static int mdd_rename(const struct lu_env *env,
2060 struct md_object *src_pobj, struct md_object *tgt_pobj,
2061 const struct lu_fid *lf, const struct lu_name *lsname,
2062 struct md_object *tobj, const struct lu_name *ltname,
2065 const char *sname = lsname->ln_name;
2066 const char *tname = ltname->ln_name;
2067 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2068 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2069 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2070 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2071 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2072 struct mdd_device *mdd = mdo2mdd(src_pobj);
2073 struct mdd_object *mdd_sobj = NULL; /* source object */
2074 struct mdd_object *mdd_tobj = NULL;
2075 struct dynlock_handle *sdlh, *tdlh;
2076 struct thandle *handle;
2077 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2078 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2081 bool tobj_locked = 0;
2082 unsigned cl_flags = 0;
2087 mdd_tobj = md2mdd_obj(tobj);
2089 mdd_sobj = mdd_object_find(env, mdd, lf);
2091 handle = mdd_trans_create(env, mdd);
2093 GOTO(out_pending, rc = PTR_ERR(handle));
2095 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2096 mdd_tobj, lsname, ltname, ma, handle);
2100 rc = mdd_trans_start(env, mdd, handle);
2104 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2105 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2107 GOTO(cleanup_unlocked, rc);
2109 /* Get locks in determined order */
2110 if (rc == MDD_RN_SAME) {
2111 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2112 sname, MOR_SRC_PARENT);
2113 /* check hashes to determine do we need one lock or two */
2114 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2115 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2119 } else if (rc == MDD_RN_SRCTGT) {
2120 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2121 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2123 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2124 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2126 if (sdlh == NULL || tdlh == NULL)
2127 GOTO(cleanup, rc = -ENOMEM);
2129 rc = mdd_la_get(env, mdd_sobj, so_attr,
2130 mdd_object_capa(env, mdd_sobj));
2135 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2136 mdd_object_capa(env, mdd_tobj));
2141 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2142 mdd_tobj, so_attr, tg_attr);
2146 is_dir = S_ISDIR(so_attr->la_mode);
2148 /* Remove source name from source directory */
2149 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2150 mdd_object_capa(env, mdd_spobj));
2154 /* "mv dir1 dir2" needs "dir1/.." link update */
2155 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2156 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2157 mdd_object_capa(env, mdd_sobj));
2159 GOTO(fixup_spobj2, rc);
2161 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2162 handle, mdd_object_capa(env, mdd_sobj));
2164 GOTO(fixup_spobj, rc);
2167 /* Remove target name from target directory
2168 * Here tobj can be remote one, so we do index_delete unconditionally
2169 * and -ENOENT is allowed.
2171 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2172 mdd_object_capa(env, mdd_tpobj));
2175 /* tname might been renamed to something else */
2176 GOTO(fixup_spobj, rc);
2179 GOTO(fixup_spobj, rc);
2182 /* Insert new fid with target name into target dir */
2183 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2184 mdd_object_capa(env, mdd_tpobj));
2186 GOTO(fixup_tpobj, rc);
2188 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2189 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2191 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2193 la->la_valid = LA_CTIME;
2194 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2196 GOTO(fixup_tpobj, rc);
2199 /* Remove old target object
2200 * For tobj is remote case cmm layer has processed
2201 * and set tobj to NULL then. So when tobj is NOT NULL,
2202 * it must be local one.
2204 if (tobj && mdd_object_exists(mdd_tobj)) {
2205 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2207 if (mdd_is_dead_obj(mdd_tobj)) {
2208 /* shld not be dead, something is wrong */
2209 CERROR("tobj is dead, something is wrong\n");
2213 mdo_ref_del(env, mdd_tobj, handle);
2215 /* Remove dot reference. */
2216 if (S_ISDIR(tg_attr->la_mode))
2217 mdo_ref_del(env, mdd_tobj, handle);
2220 /* fetch updated nlink */
2221 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2222 mdd_object_capa(env, mdd_tobj));
2224 CERROR("%s: Failed to get nlink for tobj "
2226 mdd2obd_dev(mdd)->obd_name,
2227 PFID(tpobj_fid), rc);
2228 GOTO(fixup_tpobj, rc);
2231 la->la_valid = LA_CTIME;
2232 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2234 CERROR("%s: Failed to set ctime for tobj "
2236 mdd2obd_dev(mdd)->obd_name,
2237 PFID(tpobj_fid), rc);
2238 GOTO(fixup_tpobj, rc);
2241 /* XXX: this transfer to ma will be removed with LOD/OSP */
2242 ma->ma_attr = *tg_attr;
2243 ma->ma_valid |= MA_INODE;
2244 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2246 CERROR("%s: Failed to unlink tobj "
2248 mdd2obd_dev(mdd)->obd_name,
2249 PFID(tpobj_fid), rc);
2250 GOTO(fixup_tpobj, rc);
2253 /* fetch updated nlink */
2254 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2255 mdd_object_capa(env, mdd_tobj));
2257 CERROR("%s: Failed to get nlink for tobj "
2259 mdd2obd_dev(mdd)->obd_name,
2260 PFID(tpobj_fid), rc);
2261 GOTO(fixup_tpobj, rc);
2263 /* XXX: this transfer to ma will be removed with LOD/OSP */
2264 ma->ma_attr = *tg_attr;
2265 ma->ma_valid |= MA_INODE;
2267 if (so_attr->la_nlink == 0)
2268 cl_flags |= CLF_RENAME_LAST;
2271 la->la_valid = LA_CTIME | LA_MTIME;
2272 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2274 GOTO(fixup_tpobj, rc);
2276 if (mdd_spobj != mdd_tpobj) {
2277 la->la_valid = LA_CTIME | LA_MTIME;
2278 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2282 if (rc == 0 && mdd_sobj) {
2283 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2284 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2285 mdo2fid(mdd_tpobj), ltname, handle);
2287 /* Old files might not have EA entry */
2288 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2290 mdd_write_unlock(env, mdd_sobj);
2291 /* We don't fail the transaction if the link ea can't be
2292 updated -- fid2path will use alternate lookup method. */
2300 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2303 CWARN("tp obj fix error %d\n",rc2);
2305 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2306 !mdd_is_dead_obj(mdd_tobj)) {
2308 mdo_ref_add(env, mdd_tobj, handle);
2310 mdo_ref_add(env, mdd_tobj, handle);
2313 rc2 = __mdd_index_insert(env, mdd_tpobj,
2314 mdo2fid(mdd_tobj), tname,
2319 CWARN("tp obj fix error %d\n",rc2);
2324 if (rc && is_dir && mdd_sobj) {
2325 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2329 CWARN("sp obj dotdot delete error %d\n",rc2);
2332 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2333 dotdot, handle, BYPASS_CAPA);
2335 CWARN("sp obj dotdot insert error %d\n",rc2);
2340 rc2 = __mdd_index_insert(env, mdd_spobj,
2341 lf, sname, is_dir, handle, BYPASS_CAPA);
2343 CWARN("sp obj fix error %d\n",rc2);
2347 mdd_write_unlock(env, mdd_tobj);
2348 if (likely(tdlh) && sdlh != tdlh)
2349 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2351 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2354 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2355 mdd_tobj, tpobj_fid, lf,
2356 spobj_fid, ltname, lsname,
2360 mdd_trans_stop(env, mdd, rc, handle);
2362 mdd_object_put(env, mdd_sobj);
2367 /** enable/disable storing of hardlink info */
2368 int mdd_linkea_enable = 1;
2369 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2370 "record hardlink info in EAs");
2372 /** Read the link EA into a temp buffer.
2373 * Uses the name_buf since it is generally large.
2374 * \retval IS_ERR err
2375 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2377 struct lu_buf *mdd_links_get(const struct lu_env *env,
2378 struct mdd_object *mdd_obj)
2381 struct lustre_capa *capa;
2382 struct link_ea_header *leh;
2385 /* First try a small buf */
2386 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2387 if (buf->lb_buf == NULL)
2388 return ERR_PTR(-ENOMEM);
2390 if (!mdd_object_exists(mdd_obj))
2391 return ERR_PTR(-ENODATA);
2393 capa = mdd_object_capa(env, mdd_obj);
2394 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2395 if (rc == -ERANGE) {
2396 /* Buf was too small, figure out what we need. */
2398 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2401 buf = mdd_buf_alloc(env, rc);
2402 if (buf->lb_buf == NULL)
2403 return ERR_PTR(-ENOMEM);
2404 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2410 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2411 leh->leh_magic = LINK_EA_MAGIC;
2412 leh->leh_reccount = __swab32(leh->leh_reccount);
2413 leh->leh_len = __swab64(leh->leh_len);
2414 /* entries are swabbed by mdd_lee_unpack */
2416 if (leh->leh_magic != LINK_EA_MAGIC)
2417 return ERR_PTR(-EINVAL);
2418 if (leh->leh_reccount == 0)
2419 return ERR_PTR(-ENODATA);
2424 /** Pack a link_ea_entry.
2425 * All elements are stored as chars to avoid alignment issues.
2426 * Numbers are always big-endian
2427 * \retval record length
2429 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2430 const struct lu_fid *pfid)
2432 struct lu_fid tmpfid;
2435 fid_cpu_to_be(&tmpfid, pfid);
2436 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2437 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2438 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2440 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2441 lee->lee_reclen[1] = reclen & 0xff;
2445 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2446 struct lu_name *lname, struct lu_fid *pfid)
2448 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2449 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2450 fid_be_to_cpu(pfid, pfid);
2451 lname->ln_name = lee->lee_name;
2452 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2455 /** Add a record to the end of link ea buf */
2456 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2457 const struct lu_fid *pfid,
2458 const struct lu_name *lname)
2460 struct link_ea_header *leh;
2461 struct link_ea_entry *lee;
2464 if (lname == NULL || pfid == NULL)
2467 /* Make sure our buf is big enough for the new one */
2469 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2470 if (leh->leh_len + reclen > buf->lb_len) {
2471 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2476 lee = buf->lb_buf + leh->leh_len;
2477 reclen = mdd_lee_pack(lee, lname, pfid);
2478 leh->leh_len += reclen;
2479 leh->leh_reccount++;
2483 static int mdd_declare_links_add(const struct lu_env *env,
2484 struct mdd_object *mdd_obj,
2485 struct thandle *handle)
2489 /* XXX: max size? */
2490 rc = mdo_declare_xattr_set(env, mdd_obj,
2491 mdd_buf_get_const(env, NULL, 4096),
2492 XATTR_NAME_LINK, 0, handle);
2497 /* For pathologic linkers, we don't want to spend lots of time scanning the
2498 * link ea. Limit ourseleves to something reasonable; links not in the EA
2499 * can be looked up via (slower) parent lookup.
2501 #define LINKEA_MAX_COUNT 128
2503 static int mdd_links_add(const struct lu_env *env,
2504 struct mdd_object *mdd_obj,
2505 const struct lu_fid *pfid,
2506 const struct lu_name *lname,
2507 struct thandle *handle, int first)
2510 struct link_ea_header *leh;
2514 if (!mdd_linkea_enable)
2517 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2520 if (rc != -ENODATA) {
2521 CERROR("link_ea read failed %d "DFID"\n", rc,
2522 PFID(mdd_object_fid(mdd_obj)));
2525 /* empty EA; start one */
2526 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2527 if (buf->lb_buf == NULL)
2530 leh->leh_magic = LINK_EA_MAGIC;
2531 leh->leh_len = sizeof(struct link_ea_header);
2532 leh->leh_reccount = 0;
2536 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2539 rc = __mdd_links_add(env, buf, pfid, lname);
2544 rc = mdo_xattr_set(env, mdd_obj,
2545 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2546 XATTR_NAME_LINK, 0, handle,
2547 mdd_object_capa(env, mdd_obj));
2550 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2551 PFID(mdd_object_fid(mdd_obj)));
2553 CERROR("link_ea add failed %d "DFID"\n", rc,
2554 PFID(mdd_object_fid(mdd_obj)));
2557 if (buf->lb_len > OBD_ALLOC_BIG)
2558 /* if we vmalloced a large buffer drop it */
2564 static int mdd_links_rename(const struct lu_env *env,
2565 struct mdd_object *mdd_obj,
2566 const struct lu_fid *oldpfid,
2567 const struct lu_name *oldlname,
2568 const struct lu_fid *newpfid,
2569 const struct lu_name *newlname,
2570 struct thandle *handle)
2573 struct link_ea_header *leh;
2574 struct link_ea_entry *lee;
2575 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2576 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2582 if (!mdd_linkea_enable)
2585 if (mdd_obj->mod_flags & DEAD_OBJ)
2586 /* No more links, don't bother */
2589 buf = mdd_links_get(env, mdd_obj);
2593 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2594 rc, PFID(mdd_object_fid(mdd_obj)));
2596 CERROR("link_ea read failed %d "DFID"\n",
2597 rc, PFID(mdd_object_fid(mdd_obj)));
2601 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2603 /* Find the old record */
2604 for(count = 0; count < leh->leh_reccount; count++) {
2605 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2606 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2607 lu_fid_eq(tmpfid, oldpfid) &&
2608 (strncmp(tmpname->ln_name, oldlname->ln_name,
2609 tmpname->ln_namelen) == 0))
2611 lee = (struct link_ea_entry *)((char *)lee + reclen);
2613 if ((count + 1) > leh->leh_reccount) {
2614 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2615 oldlname->ln_namelen, oldlname->ln_name);
2616 GOTO(out, rc = -ENOENT);
2619 /* Remove the old record */
2620 leh->leh_reccount--;
2621 leh->leh_len -= reclen;
2622 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2625 /* If renaming, add the new record */
2626 if (newpfid != NULL) {
2627 /* if the add fails, we still delete the out-of-date old link */
2628 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2632 rc = mdo_xattr_set(env, mdd_obj,
2633 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2634 XATTR_NAME_LINK, 0, handle,
2635 mdd_object_capa(env, mdd_obj));
2641 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2642 oldlname->ln_namelen, oldlname->ln_name, rc,
2643 PFID(mdd_object_fid(mdd_obj)));
2645 if (buf->lb_len > OBD_ALLOC_BIG)
2646 /* if we vmalloced a large buffer drop it */
2652 const struct md_dir_operations mdd_dir_ops = {
2653 .mdo_is_subdir = mdd_is_subdir,
2654 .mdo_lookup = mdd_lookup,
2655 .mdo_create = mdd_create,
2656 .mdo_rename = mdd_rename,
2657 .mdo_link = mdd_link,
2658 .mdo_unlink = mdd_unlink,
2659 .mdo_create_data = mdd_create_data,