4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_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 = {
60 static int __mdd_lookup(const struct lu_env *, struct md_object *,
61 const struct lu_name *, struct lu_fid*, int);
64 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
65 const struct lu_name *lname, struct lu_fid* fid, int mask)
67 const char *name = lname->ln_name;
68 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
69 struct dynlock_handle *dlh;
72 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
73 if (unlikely(dlh == NULL))
75 rc = __mdd_lookup(env, pobj, lname, fid, mask);
76 mdd_pdo_read_unlock(env, mdd_obj, dlh);
81 int mdd_lookup(const struct lu_env *env,
82 struct md_object *pobj, const struct lu_name *lname,
83 struct lu_fid* fid, struct md_op_spec *spec)
87 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
91 int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
94 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
98 * For root fid use special function, which does not compare version component
99 * of fid. Version component is different for root fids on all MDTs.
101 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
103 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
104 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
108 * return 1: if lf is the fid of the ancestor of p1;
111 * return -EREMOTE: if remote object is found, in this
112 * case fid of remote object is saved to @pf;
114 * otherwise: values < 0, errors.
116 static int mdd_is_parent(const struct lu_env *env,
117 struct mdd_device *mdd,
118 struct mdd_object *p1,
119 const struct lu_fid *lf,
122 struct mdd_object *parent = NULL;
127 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
128 pfid = &mdd_env_info(env)->mti_fid;
130 /* Check for root first. */
131 if (mdd_is_root(mdd, mdo2fid(p1)))
135 /* this is done recursively, bypass capa for each obj */
136 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
137 rc = mdd_parent_fid(env, p1, pfid);
140 if (mdd_is_root(mdd, pfid))
142 if (lu_fid_eq(pfid, lf))
145 mdd_object_put(env, parent);
147 parent = mdd_object_find(env, mdd, pfid);
148 if (IS_ERR(parent)) {
149 GOTO(out, rc = PTR_ERR(parent));
150 } else if (mdd_object_remote(parent)) {
151 /*FIXME: Because of the restriction of rename in Phase I.
152 * If the parent is remote, we just assumed lf is not the
153 * parent of P1 for now */
160 if (parent && !IS_ERR(parent))
161 mdd_object_put(env, parent);
166 * No permission check is needed.
168 * returns 1: if fid is ancestor of @mo;
169 * returns 0: if fid is not a ancestor of @mo;
171 * returns EREMOTE if remote object is found, fid of remote object is saved to
174 * returns < 0: if error
176 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
177 const struct lu_fid *fid, struct lu_fid *sfid)
179 struct mdd_device *mdd = mdo2mdd(mo);
183 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
186 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
190 } else if (rc == 1) {
191 /* found @fid is parent */
199 * Check that @dir contains no entries except (possibly) dot and dotdot.
204 * -ENOTDIR not a directory object
205 * -ENOTEMPTY not empty
209 static int mdd_dir_is_empty(const struct lu_env *env,
210 struct mdd_object *dir)
213 struct dt_object *obj;
214 const struct dt_it_ops *iops;
218 obj = mdd_object_child(dir);
219 if (!dt_try_as_dir(env, obj))
222 iops = &obj->do_index_ops->dio_it;
223 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
225 result = iops->get(env, it, (const void *)"");
228 for (result = 0, i = 0; result == 0 && i < 3; ++i)
229 result = iops->next(env, it);
232 else if (result == +1)
234 } else if (result == 0)
236 * Huh? Index contains no zero key?
243 result = PTR_ERR(it);
247 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
249 struct mdd_device *m = mdd_obj2mdd_dev(obj);
250 struct lu_attr *la = &mdd_env_info(env)->mti_la;
254 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
259 * Subdir count limitation can be broken through.
261 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
262 !S_ISDIR(la->la_mode))
269 * Check whether it may create the cobj under the pobj.
272 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
273 struct mdd_object *cobj, int check_perm, int check_nlink)
278 if (cobj && mdd_object_exists(cobj))
281 if (mdd_is_dead_obj(pobj))
285 rc = mdd_permission_internal_locked(env, pobj, NULL,
286 MAY_WRITE | MAY_EXEC,
288 if (!rc && check_nlink)
289 rc = __mdd_may_link(env, pobj);
295 * Check whether can unlink from the pobj in the case of "cobj == NULL".
297 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
298 const struct lu_attr *attr)
303 if (mdd_is_dead_obj(pobj))
306 if ((attr->la_valid & LA_FLAGS) &&
307 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
310 rc = mdd_permission_internal_locked(env, pobj, NULL,
311 MAY_WRITE | MAY_EXEC,
316 if (mdd_is_append(pobj))
323 * pobj == NULL is remote ops case, under such case, pobj's
324 * VTX feature has been checked already, no need check again.
326 static inline int mdd_is_sticky(const struct lu_env *env,
327 struct mdd_object *pobj,
328 struct mdd_object *cobj)
330 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
331 struct lu_ucred *uc = lu_ucred_assert(env);
335 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
339 if (!(tmp_la->la_mode & S_ISVTX) ||
340 (tmp_la->la_uid == uc->uc_fsuid))
344 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
348 if (tmp_la->la_uid == uc->uc_fsuid)
351 return !md_capable(uc, CFS_CAP_FOWNER);
354 static int mdd_may_delete_entry(const struct lu_env *env,
355 struct mdd_object *pobj, int check_perm)
359 LASSERT(pobj != NULL);
360 if (!mdd_object_exists(pobj))
363 if (mdd_is_dead_obj(pobj))
368 rc = mdd_permission_internal_locked(env, pobj, NULL,
369 MAY_WRITE | MAY_EXEC,
375 if (mdd_is_append(pobj))
382 * Check whether it may delete the cobj from the pobj.
385 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
386 struct mdd_object *cobj, struct lu_attr *cattr,
387 struct lu_attr *src_attr, int check_perm, int check_empty)
393 rc = mdd_may_delete_entry(env, pobj, check_perm);
401 if (!mdd_object_exists(cobj))
404 if (mdd_is_dead_obj(cobj))
408 if (mdd_is_sticky(env, pobj, cobj))
411 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
414 if ((cattr->la_valid & LA_FLAGS) &&
415 (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
418 /* additional check the rename case */
420 if (S_ISDIR(src_attr->la_mode)) {
421 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
423 if (!S_ISDIR(cattr->la_mode))
426 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
428 } else if (S_ISDIR(cattr->la_mode))
432 if (S_ISDIR(cattr->la_mode) && check_empty)
433 rc = mdd_dir_is_empty(env, cobj);
440 * has mdd_write_lock on src already, but not on tgt yet
442 int mdd_link_sanity_check(const struct lu_env *env,
443 struct mdd_object *tgt_obj,
444 const struct lu_name *lname,
445 struct mdd_object *src_obj)
447 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
451 if (!mdd_object_exists(src_obj))
454 if (mdd_is_dead_obj(src_obj))
457 /* Local ops, no lookup before link, check filename length here. */
458 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
459 RETURN(-ENAMETOOLONG);
461 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
464 if (S_ISDIR(mdd_object_type(src_obj)))
467 LASSERT(src_obj != tgt_obj);
469 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
474 rc = __mdd_may_link(env, src_obj);
479 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
480 const char *name, struct thandle *handle,
481 struct lustre_capa *capa)
483 struct dt_object *next = mdd_object_child(pobj);
487 if (dt_try_as_dir(env, next)) {
488 rc = next->do_index_ops->dio_delete(env, next,
489 (struct dt_key *)name,
497 static int __mdd_index_insert_only(const struct lu_env *env,
498 struct mdd_object *pobj,
499 const struct lu_fid *lf, const char *name,
500 struct thandle *handle,
501 struct lustre_capa *capa)
503 struct dt_object *next = mdd_object_child(pobj);
507 if (dt_try_as_dir(env, next)) {
508 struct lu_ucred *uc = lu_ucred_check(env);
511 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
512 rc = next->do_index_ops->dio_insert(env, next,
514 (const struct dt_key *)name,
515 handle, capa, ignore_quota);
522 /* insert named index, add reference if isdir */
523 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
524 const struct lu_fid *lf, const char *name, int is_dir,
525 struct thandle *handle, struct lustre_capa *capa)
530 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
531 if (rc == 0 && is_dir) {
532 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
533 mdo_ref_add(env, pobj, handle);
534 mdd_write_unlock(env, pobj);
539 /* delete named index, drop reference if isdir */
540 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
541 const char *name, int is_dir, struct thandle *handle,
542 struct lustre_capa *capa)
547 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
548 if (rc == 0 && is_dir) {
549 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
550 mdo_ref_del(env, pobj, handle);
551 mdd_write_unlock(env, pobj);
557 int mdd_declare_changelog_store(const struct lu_env *env,
558 struct mdd_device *mdd,
559 const struct lu_name *fname,
560 struct thandle *handle)
562 struct obd_device *obd = mdd2obd_dev(mdd);
563 struct llog_ctxt *ctxt;
564 struct llog_changelog_rec *rec;
570 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
573 reclen = llog_data_len(sizeof(*rec) +
574 (fname != NULL ? fname->ln_namelen : 0));
575 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
576 if (buf->lb_buf == NULL)
580 rec->cr_hdr.lrh_len = reclen;
581 rec->cr_hdr.lrh_type = CHANGELOG_REC;
583 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
587 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
593 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
594 struct mdd_device *mdd,
595 const struct lu_name *tname,
596 const struct lu_name *sname,
597 struct thandle *handle)
599 struct obd_device *obd = mdd2obd_dev(mdd);
600 struct llog_ctxt *ctxt;
601 struct llog_changelog_ext_rec *rec;
607 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
610 reclen = llog_data_len(sizeof(*rec) +
611 (tname != NULL ? tname->ln_namelen : 0) +
612 (sname != NULL ? 1 + sname->ln_namelen : 0));
613 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
614 if (buf->lb_buf == NULL)
618 rec->cr_hdr.lrh_len = reclen;
619 rec->cr_hdr.lrh_type = CHANGELOG_REC;
621 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
625 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
631 /** Add a changelog entry \a rec to the changelog llog
634 * \param handle - currently ignored since llogs start their own transaction;
635 * this will hopefully be fixed in llog rewrite
638 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
639 struct llog_changelog_rec *rec, struct thandle *th)
641 struct obd_device *obd = mdd2obd_dev(mdd);
642 struct llog_ctxt *ctxt;
645 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
646 rec->cr_hdr.lrh_type = CHANGELOG_REC;
647 rec->cr.cr_time = cl_time();
649 spin_lock(&mdd->mdd_cl.mc_lock);
650 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
651 * but as long as the MDD transactions are ordered correctly for e.g.
652 * rename conflicts, I don't think this should matter. */
653 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
654 spin_unlock(&mdd->mdd_cl.mc_lock);
656 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
660 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
667 /** Add a changelog_ext entry \a rec to the changelog llog
670 * \param handle - currently ignored since llogs start their own transaction;
671 * this will hopefully be fixed in llog rewrite
674 int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
675 struct llog_changelog_ext_rec *rec,
678 struct obd_device *obd = mdd2obd_dev(mdd);
679 struct llog_ctxt *ctxt;
682 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
683 /* llog_lvfs_write_rec sets the llog tail len */
684 rec->cr_hdr.lrh_type = CHANGELOG_REC;
685 rec->cr.cr_time = cl_time();
687 spin_lock(&mdd->mdd_cl.mc_lock);
688 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
689 * but as long as the MDD transactions are ordered correctly for e.g.
690 * rename conflicts, I don't think this should matter. */
691 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
692 spin_unlock(&mdd->mdd_cl.mc_lock);
694 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
698 /* nested journal transaction */
699 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
707 /** Store a namespace change changelog record
708 * If this fails, we must fail the whole transaction; we don't
709 * want the change to commit without the log entry.
710 * \param target - mdd_object of change
711 * \param parent - parent dir/object
712 * \param tname - target name string
713 * \param handle - transacion handle
715 int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd,
716 enum changelog_rec_type type, unsigned flags,
717 struct mdd_object *target, struct mdd_object *parent,
718 const struct lu_name *tname, struct thandle *handle)
720 struct llog_changelog_rec *rec;
727 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
729 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
732 LASSERT(target != NULL);
733 LASSERT(parent != NULL);
734 LASSERT(tname != NULL);
735 LASSERT(handle != NULL);
737 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
738 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
739 if (buf->lb_buf == NULL)
743 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
744 rec->cr.cr_type = (__u32)type;
745 rec->cr.cr_tfid = *mdo2fid(target);
746 rec->cr.cr_pfid = *mdo2fid(parent);
747 rec->cr.cr_namelen = tname->ln_namelen;
748 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
750 target->mod_cltime = cfs_time_current_64();
752 rc = mdd_changelog_store(env, mdd, rec, handle);
754 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
755 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
756 PFID(&rec->cr.cr_pfid));
764 /** Store a namespace change changelog record
765 * If this fails, we must fail the whole transaction; we don't
766 * want the change to commit without the log entry.
767 * \param target - mdd_object of change
768 * \param tpfid - target parent dir/object fid
769 * \param sfid - source object fid
770 * \param spfid - source parent fid
771 * \param tname - target name string
772 * \param sname - source name string
773 * \param handle - transacion handle
775 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
776 struct mdd_device *mdd,
777 enum changelog_rec_type type,
779 struct mdd_object *target,
780 const struct lu_fid *tpfid,
781 const struct lu_fid *sfid,
782 const struct lu_fid *spfid,
783 const struct lu_name *tname,
784 const struct lu_name *sname,
785 struct thandle *handle)
787 struct llog_changelog_ext_rec *rec;
794 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
796 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
799 LASSERT(sfid != NULL);
800 LASSERT(tpfid != NULL);
801 LASSERT(tname != NULL);
802 LASSERT(handle != NULL);
804 reclen = llog_data_len(sizeof(*rec) +
805 sname != NULL ? 1 + sname->ln_namelen : 0);
806 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
807 if (buf->lb_buf == NULL)
811 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
812 rec->cr.cr_type = (__u32)type;
813 rec->cr.cr_pfid = *tpfid;
814 rec->cr.cr_sfid = *sfid;
815 rec->cr.cr_spfid = *spfid;
816 rec->cr.cr_namelen = tname->ln_namelen;
817 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
819 rec->cr.cr_name[tname->ln_namelen] = '\0';
820 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
822 rec->cr.cr_namelen += 1 + sname->ln_namelen;
825 if (likely(target != NULL)) {
826 rec->cr.cr_tfid = *mdo2fid(target);
827 target->mod_cltime = cfs_time_current_64();
829 fid_zero(&rec->cr.cr_tfid);
832 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
834 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
835 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
842 static int __mdd_links_add(const struct lu_env *env,
843 struct mdd_object *mdd_obj,
844 struct linkea_data *ldata,
845 const struct lu_name *lname,
846 const struct lu_fid *pfid,
847 int first, int check)
851 if (ldata->ld_leh == NULL) {
852 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
856 rc = linkea_data_new(ldata,
857 &mdd_env_info(env)->mti_link_buf);
864 rc = linkea_links_find(ldata, lname, pfid);
865 if (rc && rc != -ENOENT)
871 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
872 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
876 linkea_add_buf(ldata, lname, tfid);
879 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
880 linkea_add_buf(ldata, lname, pfid);
882 return linkea_add_buf(ldata, lname, pfid);
885 static int __mdd_links_del(const struct lu_env *env,
886 struct mdd_object *mdd_obj,
887 struct linkea_data *ldata,
888 const struct lu_name *lname,
889 const struct lu_fid *pfid)
893 if (ldata->ld_leh == NULL) {
894 rc = mdd_links_read(env, mdd_obj, ldata);
899 rc = linkea_links_find(ldata, lname, pfid);
903 linkea_del_buf(ldata, lname);
907 static int mdd_linkea_prepare(const struct lu_env *env,
908 struct mdd_object *mdd_obj,
909 const struct lu_fid *oldpfid,
910 const struct lu_name *oldlname,
911 const struct lu_fid *newpfid,
912 const struct lu_name *newlname,
913 int first, int check,
914 struct linkea_data *ldata)
920 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
923 LASSERT(oldpfid != NULL || newpfid != NULL);
925 if (mdd_obj->mod_flags & DEAD_OBJ) {
926 /* Prevent linkea to be updated which is NOT necessary. */
927 ldata->ld_reclen = 0;
928 /* No more links, don't bother */
932 if (oldpfid != NULL) {
933 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
936 (rc != -ENODATA && rc != -ENOENT))
938 /* No changes done. */
943 /* If renaming, add the new record */
944 if (newpfid != NULL) {
945 /* even if the add fails, we still delete the out-of-date
947 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
953 rc = rc != 0 ? rc : rc2;
958 int mdd_links_rename(const struct lu_env *env,
959 struct mdd_object *mdd_obj,
960 const struct lu_fid *oldpfid,
961 const struct lu_name *oldlname,
962 const struct lu_fid *newpfid,
963 const struct lu_name *newlname,
964 struct thandle *handle,
965 struct linkea_data *ldata,
966 int first, int check)
973 ldata = &mdd_env_info(env)->mti_link_data;
974 memset(ldata, 0, sizeof(*ldata));
975 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
976 newpfid, newlname, first, check,
982 if (ldata->ld_reclen != 0)
983 rc = mdd_links_write(env, mdd_obj, ldata, handle);
990 if (rc == -EOVERFLOW || rc == -ENOENT || rc == -ENOSPC)
993 CDEBUG(error ? D_ERROR : D_OTHER,
994 "link_ea add '%.*s' failed %d "DFID"\n",
995 newlname->ln_namelen, newlname->ln_name,
996 rc, PFID(mdd_object_fid(mdd_obj)));
997 else if (newpfid == NULL)
998 CDEBUG(error ? D_ERROR : D_OTHER,
999 "link_ea del '%.*s' failed %d "DFID"\n",
1000 oldlname->ln_namelen, oldlname->ln_name,
1001 rc, PFID(mdd_object_fid(mdd_obj)));
1003 CDEBUG(error ? D_ERROR : D_OTHER,
1004 "link_ea rename '%.*s'->'%.*s' failed %d "
1006 oldlname->ln_namelen, oldlname->ln_name,
1007 newlname->ln_namelen, newlname->ln_name,
1008 rc, PFID(mdd_object_fid(mdd_obj)));
1011 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1012 /* if we vmalloced a large buffer drop it */
1013 lu_buf_free(ldata->ld_buf);
1018 static inline int mdd_links_add(const struct lu_env *env,
1019 struct mdd_object *mdd_obj,
1020 const struct lu_fid *pfid,
1021 const struct lu_name *lname,
1022 struct thandle *handle,
1023 struct linkea_data *data, int first)
1025 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1026 pfid, lname, handle, data, first, 0);
1029 static inline int mdd_links_del(const struct lu_env *env,
1030 struct mdd_object *mdd_obj,
1031 const struct lu_fid *pfid,
1032 const struct lu_name *lname,
1033 struct thandle *handle)
1035 return mdd_links_rename(env, mdd_obj, pfid, lname,
1036 NULL, NULL, handle, NULL, 0, 0);
1039 /** Read the link EA into a temp buffer.
1040 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1041 * A pointer to the buffer is stored in \a ldata::ld_buf.
1043 * \retval 0 or error
1045 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1046 struct linkea_data *ldata)
1050 /* First try a small buf */
1051 LASSERT(env != NULL);
1052 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1054 if (ldata->ld_buf->lb_buf == NULL)
1057 if (!mdd_object_exists(mdd_obj))
1060 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1062 if (rc == -ERANGE) {
1063 /* Buf was too small, figure out what we need. */
1064 lu_buf_free(ldata->ld_buf);
1065 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1066 XATTR_NAME_LINK, BYPASS_CAPA);
1069 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1070 if (ldata->ld_buf->lb_buf == NULL)
1072 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1073 XATTR_NAME_LINK, BYPASS_CAPA);
1082 /** Read the link EA into a temp buffer.
1083 * Uses the name_buf since it is generally large.
1084 * \retval IS_ERR err
1085 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1087 struct lu_buf *mdd_links_get(const struct lu_env *env,
1088 struct mdd_object *mdd_obj)
1090 struct linkea_data ldata = { 0 };
1093 rc = mdd_links_read(env, mdd_obj, &ldata);
1094 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1097 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1098 struct linkea_data *ldata, struct thandle *handle)
1100 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1101 ldata->ld_leh->leh_len);
1102 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1103 mdd_object_capa(env, mdd_obj));
1106 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1107 struct thandle *handle, struct linkea_data *ldata)
1113 if (ldata != NULL && ldata->ld_lee != NULL) {
1114 ea_len = ldata->ld_leh->leh_len;
1115 linkea = ldata->ld_buf->lb_buf;
1117 ea_len = DEFAULT_LINKEA_SIZE;
1121 /* XXX: max size? */
1122 rc = mdo_declare_xattr_set(env, mdd_obj,
1123 mdd_buf_get_const(env, linkea, ea_len),
1124 XATTR_NAME_LINK, 0, handle);
1128 static inline int mdd_declare_links_del(const struct lu_env *env,
1129 struct mdd_object *c,
1130 struct thandle *handle)
1134 /* For directory, the linkEA will be removed together
1135 * with the object. */
1136 if (!S_ISDIR(mdd_object_type(c)))
1137 rc = mdd_declare_links_add(env, c, handle, NULL);
1142 static int mdd_declare_link(const struct lu_env *env,
1143 struct mdd_device *mdd,
1144 struct mdd_object *p,
1145 struct mdd_object *c,
1146 const struct lu_name *name,
1147 struct thandle *handle,
1149 struct linkea_data *data)
1153 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
1157 rc = mdo_declare_ref_add(env, c, handle);
1161 la->la_valid = LA_CTIME | LA_MTIME;
1162 rc = mdo_declare_attr_set(env, p, la, handle);
1166 la->la_valid = LA_CTIME;
1167 rc = mdo_declare_attr_set(env, c, la, handle);
1171 rc = mdd_declare_links_add(env, c, handle, data);
1175 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1180 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1181 struct md_object *src_obj, const struct lu_name *lname,
1184 const char *name = lname->ln_name;
1185 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1186 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1187 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1188 struct mdd_device *mdd = mdo2mdd(src_obj);
1189 struct dynlock_handle *dlh;
1190 struct thandle *handle;
1191 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1195 handle = mdd_trans_create(env, mdd);
1197 GOTO(out_pending, rc = PTR_ERR(handle));
1199 memset(ldata, 0, sizeof(*ldata));
1201 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1202 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1204 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1209 rc = mdd_trans_start(env, mdd, handle);
1213 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
1215 GOTO(out_trans, rc = -ENOMEM);
1216 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1218 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
1220 GOTO(out_unlock, rc);
1222 rc = mdo_ref_add(env, mdd_sobj, handle);
1224 GOTO(out_unlock, rc);
1227 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1229 mdd_object_capa(env, mdd_tobj));
1231 mdo_ref_del(env, mdd_sobj, handle);
1232 GOTO(out_unlock, rc);
1235 la->la_valid = LA_CTIME | LA_MTIME;
1236 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1238 GOTO(out_unlock, rc);
1240 la->la_valid = LA_CTIME;
1241 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
1243 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1244 mdo2fid(mdd_tobj), lname, 0, 0,
1247 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1248 lname, handle, ldata, 0);
1249 /* The failure of links_add should not cause the link
1250 * failure, reset rc here */
1255 mdd_write_unlock(env, mdd_sobj);
1256 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
1259 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1260 mdd_tobj, lname, handle);
1262 mdd_trans_stop(env, mdd, rc, handle);
1264 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1265 /* if we vmalloced a large buffer drop it */
1266 lu_buf_free(ldata->ld_buf);
1271 int mdd_declare_finish_unlink(const struct lu_env *env,
1272 struct mdd_object *obj,
1274 struct thandle *handle)
1278 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1282 return mdo_declare_destroy(env, obj, handle);
1285 /* caller should take a lock before calling */
1286 int mdd_finish_unlink(const struct lu_env *env,
1287 struct mdd_object *obj, struct md_attr *ma,
1291 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1294 LASSERT(mdd_write_locked(env, obj) != 0);
1296 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1297 obj->mod_flags |= DEAD_OBJ;
1299 /* add new orphan and the object
1300 * will be deleted during mdd_close() */
1301 if (obj->mod_count) {
1302 rc = __mdd_orphan_add(env, obj, th);
1304 CDEBUG(D_HA, "Object "DFID" is inserted into "
1305 "orphan list, open count = %d\n",
1306 PFID(mdd_object_fid(obj)),
1309 CERROR("Object "DFID" fail to be an orphan, "
1310 "open count = %d, maybe cause failed "
1312 PFID(mdd_object_fid(obj)),
1315 rc = mdo_destroy(env, obj, th);
1324 * has mdd_write_lock on cobj already, but not on pobj yet
1326 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1327 struct mdd_object *cobj, struct lu_attr *cattr)
1332 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1337 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1338 struct mdd_object *p, struct mdd_object *c,
1339 const struct lu_name *name, struct md_attr *ma,
1340 struct thandle *handle, int no_name)
1342 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1345 if (likely(no_name == 0)) {
1346 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1351 rc = mdo_declare_ref_del(env, p, handle);
1355 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1356 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1357 la->la_valid = LA_CTIME | LA_MTIME;
1358 rc = mdo_declare_attr_set(env, p, la, handle);
1363 rc = mdo_declare_ref_del(env, c, handle);
1367 rc = mdo_declare_ref_del(env, c, handle);
1371 la->la_valid = LA_CTIME;
1372 rc = mdo_declare_attr_set(env, c, la, handle);
1376 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1380 rc = mdd_declare_links_del(env, c, handle);
1384 /* FIXME: need changelog for remove entry */
1385 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1392 * Delete name entry and the object.
1393 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1394 * does not exist for this object, and it could only happen during resending
1395 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1396 * is also needed in this case(needed by changelog), so we have to add another
1397 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1398 * the ENOENT failure should be able to be fixed by redo mechanism.
1400 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1401 struct md_object *cobj, const struct lu_name *lname,
1402 struct md_attr *ma, int no_name)
1404 const char *name = lname->ln_name;
1405 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1406 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1407 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1408 struct mdd_object *mdd_cobj = NULL;
1409 struct mdd_device *mdd = mdo2mdd(pobj);
1410 struct dynlock_handle *dlh;
1411 struct thandle *handle;
1415 /* cobj == NULL means only delete name entry */
1416 if (likely(cobj != NULL)) {
1417 mdd_cobj = md2mdd_obj(cobj);
1418 if (mdd_object_exists(mdd_cobj) == 0)
1420 /* currently it is assume, it could only delete
1421 * name entry of remote directory */
1425 handle = mdd_trans_create(env, mdd);
1427 RETURN(PTR_ERR(handle));
1429 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1430 lname, ma, handle, no_name);
1434 rc = mdd_trans_start(env, mdd, handle);
1438 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1440 GOTO(stop, rc = -ENOMEM);
1442 if (likely(mdd_cobj != NULL)) {
1443 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1446 rc = mdd_la_get(env, mdd_cobj, cattr,
1447 mdd_object_capa(env, mdd_cobj));
1451 is_dir = S_ISDIR(cattr->la_mode);
1455 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1459 if (likely(no_name == 0)) {
1460 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1461 mdd_object_capa(env, mdd_pobj));
1466 if (likely(mdd_cobj != NULL)) {
1467 rc = mdo_ref_del(env, mdd_cobj, handle);
1469 __mdd_index_insert_only(env, mdd_pobj,
1472 mdd_object_capa(env, mdd_pobj));
1478 mdo_ref_del(env, mdd_cobj, handle);
1480 /* fetch updated nlink */
1481 rc = mdd_la_get(env, mdd_cobj, cattr,
1482 mdd_object_capa(env, mdd_cobj));
1487 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1488 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1490 la->la_valid = LA_CTIME | LA_MTIME;
1491 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1495 /* Enough for only unlink the entry */
1496 if (unlikely(mdd_cobj == NULL)) {
1497 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1501 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1502 /* update ctime of an unlinked file only if it is still
1503 * opened or a link still exists */
1504 la->la_valid = LA_CTIME;
1505 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1510 /* XXX: this transfer to ma will be removed with LOD/OSP */
1511 ma->ma_attr = *cattr;
1512 ma->ma_valid |= MA_INODE;
1513 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1515 /* fetch updated nlink */
1517 rc = mdd_la_get(env, mdd_cobj, cattr,
1518 mdd_object_capa(env, mdd_cobj));
1521 /* old files may not have link ea; ignore errors */
1522 mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle);
1524 /* if object is removed then we can't get its attrs, use last get */
1525 if (cattr->la_nlink == 0) {
1526 ma->ma_attr = *cattr;
1527 ma->ma_valid |= MA_INODE;
1531 mdd_write_unlock(env, mdd_cobj);
1532 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1536 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1537 if ((ma->ma_valid & MA_HSM) &&
1538 (ma->ma_hsm.mh_flags & HS_EXISTS))
1539 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1541 rc = mdd_changelog_ns_store(env, mdd,
1542 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1543 mdd_cobj, mdd_pobj, lname, handle);
1547 mdd_trans_stop(env, mdd, rc, handle);
1553 * The permission has been checked when obj created, no need check again.
1555 static int mdd_cd_sanity_check(const struct lu_env *env,
1556 struct mdd_object *obj)
1561 if (!obj || mdd_is_dead_obj(obj))
1568 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1569 struct md_object *cobj, const struct md_op_spec *spec,
1572 struct mdd_device *mdd = mdo2mdd(cobj);
1573 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1574 struct mdd_object *son = md2mdd_obj(cobj);
1575 struct thandle *handle;
1576 const struct lu_buf *buf;
1577 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1581 rc = mdd_cd_sanity_check(env, son);
1585 if (!md_should_create(spec->sp_cr_flags))
1589 * there are following use cases for this function:
1590 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1591 * striping can be specified or not
1594 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1598 /* calling ->ah_make_hint() is used to transfer information from parent */
1599 mdd_object_make_hint(env, mdd_pobj, son, attr);
1601 handle = mdd_trans_create(env, mdd);
1603 GOTO(out_free, rc = PTR_ERR(handle));
1606 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1607 * Should this be fixed?
1609 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, no_create %u\n",
1610 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1611 spec->sp_cr_flags, spec->no_create);
1613 if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1614 /* replay case or lfs setstripe */
1615 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1616 spec->u.sp_ea.eadatalen);
1621 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1622 XATTR_NAME_LOV, 0, handle);
1626 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1630 rc = mdd_trans_start(env, mdd, handle);
1634 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1635 0, handle, mdd_object_capa(env, son));
1640 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1643 mdd_trans_stop(env, mdd, rc, handle);
1648 /* Get fid from name and parent */
1650 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1651 const struct lu_name *lname, struct lu_fid* fid, int mask)
1653 const char *name = lname->ln_name;
1654 const struct dt_key *key = (const struct dt_key *)name;
1655 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1656 struct mdd_device *m = mdo2mdd(pobj);
1657 struct dt_object *dir = mdd_object_child(mdd_obj);
1661 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1664 if (mdd_object_remote(mdd_obj)) {
1665 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
1666 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
1667 } else if (!mdd_object_exists(mdd_obj)) {
1671 /* The common filename length check. */
1672 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1673 RETURN(-ENAMETOOLONG);
1675 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1680 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1681 dt_try_as_dir(env, dir))) {
1683 rc = dir->do_index_ops->dio_lookup(env, dir,
1684 (struct dt_rec *)fid, key,
1685 mdd_object_capa(env, mdd_obj));
1696 static int mdd_declare_object_initialize(const struct lu_env *env,
1697 struct mdd_object *parent,
1698 struct mdd_object *child,
1699 struct lu_attr *attr,
1700 struct thandle *handle,
1701 struct linkea_data *ldata)
1707 * inode mode has been set in creation time, and it's based on umask,
1708 * la_mode and acl, don't set here again! (which will go wrong
1709 * because below function doesn't consider umask).
1710 * I'd suggest set all object attributes in creation time, see above.
1712 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1713 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1714 rc = mdo_declare_attr_set(env, child, attr, handle);
1715 attr->la_valid |= LA_MODE | LA_TYPE;
1716 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1717 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1720 rc = mdo_declare_ref_add(env, child, handle);
1722 rc = mdo_declare_index_insert(env, child, mdo2fid(parent),
1727 mdd_declare_links_add(env, child, handle, ldata);
1732 static int mdd_object_initialize(const struct lu_env *env,
1733 const struct lu_fid *pfid,
1734 const struct lu_name *lname,
1735 struct mdd_object *child,
1736 struct lu_attr *attr, struct thandle *handle,
1737 const struct md_op_spec *spec,
1738 struct linkea_data *ldata)
1744 * Update attributes for child.
1747 * (1) the valid bits should be converted between Lustre and Linux;
1748 * (2) maybe, the child attributes should be set in OSD when creation.
1751 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1752 /* arguments are supposed to stay the same */
1753 if (S_ISDIR(attr->la_mode)) {
1754 /* Add "." and ".." for newly created dir */
1755 mdo_ref_add(env, child, handle);
1756 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1757 dot, handle, BYPASS_CAPA);
1759 rc = __mdd_index_insert_only(env, child, pfid,
1763 mdo_ref_del(env, child, handle);
1767 mdd_links_add(env, child, pfid, lname, handle, ldata, 1);
1772 /* has not lock on pobj yet */
1773 static int mdd_create_sanity_check(const struct lu_env *env,
1774 struct md_object *pobj,
1775 struct lu_attr *pattr,
1776 const struct lu_name *lname,
1777 struct lu_attr *cattr,
1778 struct md_op_spec *spec)
1780 struct mdd_thread_info *info = mdd_env_info(env);
1781 struct lu_fid *fid = &info->mti_fid;
1782 struct mdd_object *obj = md2mdd_obj(pobj);
1783 struct mdd_device *m = mdo2mdd(pobj);
1788 if (mdd_is_dead_obj(obj))
1792 * In some cases this lookup is not needed - we know before if name
1793 * exists or not because MDT performs lookup for it.
1794 * name length check is done in lookup.
1796 if (spec->sp_cr_lookup) {
1798 * Check if the name already exist, though it will be checked in
1799 * _index_insert also, for avoiding rolling back if exists
1802 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1803 MAY_WRITE | MAY_EXEC);
1805 RETURN(rc ? : -EEXIST);
1808 * Check WRITE permission for the parent.
1809 * EXEC permission have been checked
1810 * when lookup before create already.
1812 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1819 if (pattr->la_mode & S_ISGID) {
1820 cattr->la_gid = pattr->la_gid;
1821 if (S_ISDIR(cattr->la_mode)) {
1822 cattr->la_mode |= S_ISGID;
1823 cattr->la_valid |= LA_MODE;
1827 switch (cattr->la_mode & S_IFMT) {
1829 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1831 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1832 RETURN(-ENAMETOOLONG);
1851 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1852 struct mdd_object *p, struct mdd_object *c,
1853 const struct lu_name *name,
1854 struct lu_attr *attr,
1856 struct thandle *handle,
1857 const struct md_op_spec *spec,
1858 struct linkea_data *ldata)
1862 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1866 #ifdef CONFIG_FS_POSIX_ACL
1867 if (got_def_acl > 0) {
1868 struct lu_buf *acl_buf;
1870 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1871 /* if dir, then can inherit default ACl */
1872 if (S_ISDIR(attr->la_mode)) {
1873 rc = mdo_declare_xattr_set(env, c, acl_buf,
1874 XATTR_NAME_ACL_DEFAULT,
1880 rc = mdo_declare_attr_set(env, c, attr, handle);
1884 rc = mdo_declare_xattr_set(env, c, acl_buf,
1885 XATTR_NAME_ACL_ACCESS, 0, handle);
1891 if (S_ISDIR(attr->la_mode)) {
1892 rc = mdo_declare_ref_add(env, p, handle);
1897 rc = mdd_declare_object_initialize(env, p, c, attr, handle, ldata);
1901 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1902 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
1904 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1905 name->ln_name, handle);
1909 /* replay case, create LOV EA from client data */
1910 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1911 const struct lu_buf *buf;
1913 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1914 spec->u.sp_ea.eadatalen);
1915 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1921 if (S_ISLNK(attr->la_mode)) {
1922 rc = dt_declare_record_write(env, mdd_object_child(c),
1923 strlen(spec->u.sp_symname), 0,
1929 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
1930 rc = mdo_declare_attr_set(env, p, attr, handle);
1935 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1943 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
1944 struct lu_attr *la, struct lu_buf *acl_buf,
1945 int *got_def_acl, int *reset_acl)
1950 if (S_ISLNK(la->la_mode))
1953 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
1954 rc = mdo_xattr_get(env, pobj, acl_buf,
1955 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1956 mdd_read_unlock(env, pobj);
1958 /* If there are default ACL, fix mode by default ACL */
1960 acl_buf->lb_len = rc;
1961 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
1965 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
1966 /* If there are no default ACL, fix mode by mask */
1967 struct lu_ucred *uc = lu_ucred(env);
1969 /* The create triggered by MDT internal events, such as
1970 * LFSCK reset, will not contain valid "uc". */
1971 if (unlikely(uc != NULL))
1972 la->la_mode &= ~uc->uc_umask;
1980 * Create object and insert it into namespace.
1982 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1983 const struct lu_name *lname, struct md_object *child,
1984 struct md_op_spec *spec, struct md_attr* ma)
1986 struct mdd_thread_info *info = mdd_env_info(env);
1987 struct lu_attr *la = &info->mti_la_for_fix;
1988 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1989 struct mdd_object *son = md2mdd_obj(child);
1990 struct mdd_device *mdd = mdo2mdd(pobj);
1991 struct lu_attr *attr = &ma->ma_attr;
1992 struct thandle *handle;
1993 struct lu_attr *pattr = &info->mti_pattr;
1994 struct lu_buf acl_buf;
1995 struct linkea_data *ldata = &info->mti_link_data;
1996 struct dynlock_handle *dlh;
1997 const char *name = lname->ln_name;
1998 int rc, created = 0, initialized = 0, inserted = 0;
1999 int got_def_acl = 0;
2004 * Two operations have to be performed:
2006 * - an allocation of a new object (->do_create()), and
2008 * - an insertion into a parent index (->dio_insert()).
2010 * Due to locking, operation order is not important, when both are
2011 * successful, *but* error handling cases are quite different:
2013 * - if insertion is done first, and following object creation fails,
2014 * insertion has to be rolled back, but this operation might fail
2015 * also leaving us with dangling index entry.
2017 * - if creation is done first, is has to be undone if insertion
2018 * fails, leaving us with leaked space, which is neither good, nor
2021 * It seems that creation-first is simplest solution, but it is
2022 * sub-optimal in the frequent
2027 * case, because second mkdir is bound to create object, only to
2028 * destroy it immediately.
2030 * To avoid this follow local file systems that do double lookup:
2032 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2034 * 1. create (mdd_object_create_internal())
2036 * 2. insert (__mdd_index_insert(), lookup again)
2039 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2043 /* Sanity checks before big job. */
2044 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2048 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2049 GOTO(out_free, rc = -EINPROGRESS);
2051 acl_buf.lb_buf = info->mti_xattr_buf;
2052 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2053 rc = mdd_acl_init(env, mdd_pobj, attr, &acl_buf, &got_def_acl,
2058 mdd_object_make_hint(env, mdd_pobj, son, attr);
2060 handle = mdd_trans_create(env, mdd);
2062 GOTO(out_free, rc = PTR_ERR(handle));
2064 memset(ldata, 0, sizeof(*ldata));
2065 mdd_linkea_prepare(env, son, NULL, NULL, mdd_object_fid(mdd_pobj),
2066 lname, 1, 0, ldata);
2067 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2068 got_def_acl, handle, spec, ldata);
2072 rc = mdd_trans_start(env, mdd, handle);
2076 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
2078 GOTO(out_trans, rc = -ENOMEM);
2080 mdd_write_lock(env, son, MOR_TGT_CHILD);
2081 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec);
2083 mdd_write_unlock(env, son);
2089 #ifdef CONFIG_FS_POSIX_ACL
2091 /* set default acl */
2092 if (S_ISDIR(attr->la_mode)) {
2093 LASSERTF(acl_buf.lb_len == got_def_acl,
2094 "invalid acl_buf: %p:%d got_def %d\n",
2095 acl_buf.lb_buf, (int)acl_buf.lb_len,
2097 rc = mdo_xattr_set(env, son, &acl_buf,
2098 XATTR_NAME_ACL_DEFAULT, 0,
2099 handle, BYPASS_CAPA);
2101 mdd_write_unlock(env, son);
2106 /* set its own acl */
2108 LASSERTF(acl_buf.lb_buf != NULL && acl_buf.lb_len != 0,
2109 "invalid acl_buf %p:%d\n", acl_buf.lb_buf,
2110 (int)acl_buf.lb_len);
2111 rc = mdo_xattr_set(env, son, &acl_buf,
2112 XATTR_NAME_ACL_ACCESS,
2113 0, handle, BYPASS_CAPA);
2115 mdd_write_unlock(env, son);
2122 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
2123 son, attr, handle, spec, ldata);
2126 * in case of replay we just set LOVEA provided by the client
2127 * XXX: I think it would be interesting to try "old" way where
2128 * MDT calls this xattr_set(LOV) in a different transaction.
2129 * probably this way we code can be made better.
2131 if (rc == 0 && (spec->no_create ||
2132 (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
2133 const struct lu_buf *buf;
2135 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2136 spec->u.sp_ea.eadatalen);
2137 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
2141 if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2142 rc = __mdd_orphan_add(env, son, handle);
2144 mdd_write_unlock(env, son);
2148 * Object has no links, so it will be destroyed when last
2149 * reference is released. (XXX not now.)
2155 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE))
2156 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2157 name, S_ISDIR(attr->la_mode), handle,
2158 mdd_object_capa(env, mdd_pobj));
2165 if (S_ISLNK(attr->la_mode)) {
2166 struct lu_ucred *uc = lu_ucred_assert(env);
2167 struct dt_object *dt = mdd_object_child(son);
2168 const char *target_name = spec->u.sp_symname;
2169 int sym_len = strlen(target_name);
2170 const struct lu_buf *buf;
2173 buf = mdd_buf_get_const(env, target_name, sym_len);
2174 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2175 mdd_object_capa(env, son),
2177 CFS_CAP_SYS_RESOURCE_MASK);
2182 GOTO(cleanup, rc = -EFAULT);
2185 /* volatile file creation does not update parent directory times */
2186 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2187 GOTO(cleanup, rc = 0);
2189 /* update parent directory mtime/ctime */
2191 la->la_valid = LA_CTIME | LA_MTIME;
2192 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
2198 if (rc != 0 && created != 0) {
2201 if (inserted != 0) {
2202 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2203 rc2 = __mdd_orphan_del(env, son, handle);
2205 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2206 S_ISDIR(attr->la_mode),
2207 handle, BYPASS_CAPA);
2212 mdd_write_lock(env, son, MOR_TGT_CHILD);
2213 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
2214 /* Drop the reference, no need to delete "."/"..",
2215 * because the object to be destroied directly. */
2216 rc2 = mdo_ref_del(env, son, handle);
2218 mdd_write_unlock(env, son);
2223 rc2 = mdo_ref_del(env, son, handle);
2225 mdd_write_unlock(env, son);
2229 mdo_destroy(env, son, handle);
2230 mdd_write_unlock(env, son);
2233 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
2235 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)))
2236 rc = mdd_changelog_ns_store(env, mdd,
2237 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2238 S_ISREG(attr->la_mode) ? CL_CREATE :
2239 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2240 0, son, mdd_pobj, lname, handle);
2242 mdd_trans_stop(env, mdd, rc, handle);
2244 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2245 /* if we vmalloced a large buffer drop it */
2246 lu_buf_free(ldata->ld_buf);
2248 /* The child object shouldn't be cached anymore */
2250 set_bit(LU_OBJECT_HEARD_BANSHEE,
2251 &child->mo_lu.lo_header->loh_flags);
2256 * Get locks on parents in proper order
2257 * RETURN: < 0 - error, rename_order if successful
2265 static int mdd_rename_order(const struct lu_env *env,
2266 struct mdd_device *mdd,
2267 struct mdd_object *src_pobj,
2268 struct mdd_object *tgt_pobj)
2270 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2274 if (src_pobj == tgt_pobj)
2275 RETURN(MDD_RN_SAME);
2277 /* compared the parent child relationship of src_p&tgt_p */
2278 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2280 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2283 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
2296 /* has not mdd_write{read}_lock on any obj yet. */
2297 static int mdd_rename_sanity_check(const struct lu_env *env,
2298 struct mdd_object *src_pobj,
2299 struct mdd_object *tgt_pobj,
2300 struct mdd_object *sobj,
2301 struct mdd_object *tobj,
2302 struct lu_attr *so_attr,
2303 struct lu_attr *tg_attr)
2308 /* XXX: when get here, sobj must NOT be NULL,
2309 * the other case has been processed in cld_rename
2310 * before mdd_rename and enable MDS_PERM_BYPASS. */
2313 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
2317 /* XXX: when get here, "tobj == NULL" means tobj must
2318 * NOT exist (neither on remote MDS, such case has been
2319 * processed in cld_rename before mdd_rename and enable
2321 * So check may_create, but not check may_unlink. */
2323 rc = mdd_may_create(env, tgt_pobj, NULL,
2324 (src_pobj != tgt_pobj), 0);
2326 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
2327 (src_pobj != tgt_pobj), 1);
2329 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2330 S_ISDIR(so_attr->la_mode))
2331 rc = __mdd_may_link(env, tgt_pobj);
2336 static int mdd_declare_rename(const struct lu_env *env,
2337 struct mdd_device *mdd,
2338 struct mdd_object *mdd_spobj,
2339 struct mdd_object *mdd_tpobj,
2340 struct mdd_object *mdd_sobj,
2341 struct mdd_object *mdd_tobj,
2342 const struct lu_name *tname,
2343 const struct lu_name *sname,
2345 struct linkea_data *ldata,
2346 struct thandle *handle)
2348 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2351 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2352 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2358 /* name from source dir */
2359 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2363 /* .. from source child */
2364 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2365 /* source child can be directory,
2366 * counted by source dir's nlink */
2367 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2370 if (mdd_spobj != mdd_tpobj) {
2371 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2376 rc = mdo_declare_index_insert(env, mdd_sobj,
2382 /* new target child can be directory,
2383 * counted by target dir's nlink */
2384 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2390 la->la_valid = LA_CTIME | LA_MTIME;
2391 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2395 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2399 la->la_valid = LA_CTIME;
2400 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2404 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2409 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2410 tname->ln_name, handle);
2414 /* name from target dir (old name), we declare it unconditionally
2415 * as mdd_rename() calls delete unconditionally as well. so just
2416 * to balance declarations vs calls to change ... */
2417 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2421 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2422 /* delete target child in target parent directory */
2423 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2427 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2428 /* target child can be directory,
2429 * delete "." reference in target child directory */
2430 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2434 /* delete ".." reference in target parent directory */
2435 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2440 la->la_valid = LA_CTIME;
2441 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2445 mdd_declare_links_del(env, mdd_tobj, handle);
2449 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2454 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2461 /* src object can be remote that is why we use only fid and type of object */
2462 static int mdd_rename(const struct lu_env *env,
2463 struct md_object *src_pobj, struct md_object *tgt_pobj,
2464 const struct lu_fid *lf, const struct lu_name *lsname,
2465 struct md_object *tobj, const struct lu_name *ltname,
2468 const char *sname = lsname->ln_name;
2469 const char *tname = ltname->ln_name;
2470 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2471 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2472 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2473 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2474 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2475 struct mdd_device *mdd = mdo2mdd(src_pobj);
2476 struct mdd_object *mdd_sobj = NULL; /* source object */
2477 struct mdd_object *mdd_tobj = NULL;
2478 struct dynlock_handle *sdlh = NULL, *tdlh = NULL;
2479 struct thandle *handle;
2480 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2481 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2482 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2485 bool tobj_locked = 0;
2486 unsigned cl_flags = 0;
2491 mdd_tobj = md2mdd_obj(tobj);
2493 mdd_sobj = mdd_object_find(env, mdd, lf);
2495 rc = mdd_la_get(env, mdd_sobj, so_attr,
2496 mdd_object_capa(env, mdd_sobj));
2498 GOTO(out_pending, rc);
2501 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2502 mdd_object_capa(env, mdd_tobj));
2504 GOTO(out_pending, rc);
2507 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2508 mdd_tobj, so_attr, tg_attr);
2510 GOTO(out_pending, rc);
2512 handle = mdd_trans_create(env, mdd);
2514 GOTO(out_pending, rc = PTR_ERR(handle));
2516 memset(ldata, 0, sizeof(*ldata));
2517 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2518 ltname, 1, 0, ldata);
2519 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2520 mdd_tobj, lsname, ltname, ma, ldata, handle);
2524 rc = mdd_trans_start(env, mdd, handle);
2528 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2529 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2531 GOTO(cleanup_unlocked, rc);
2533 /* Get locks in determined order */
2534 if (rc == MDD_RN_SAME) {
2535 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2536 sname, MOR_SRC_PARENT);
2537 /* check hashes to determine do we need one lock or two */
2538 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2539 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2543 } else if (rc == MDD_RN_SRCTGT) {
2544 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2545 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2547 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2548 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2550 if (sdlh == NULL || tdlh == NULL)
2551 GOTO(cleanup, rc = -ENOMEM);
2553 is_dir = S_ISDIR(so_attr->la_mode);
2555 /* Remove source name from source directory */
2556 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2557 mdd_object_capa(env, mdd_spobj));
2561 /* "mv dir1 dir2" needs "dir1/.." link update */
2562 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2563 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2564 mdd_object_capa(env, mdd_sobj));
2566 GOTO(fixup_spobj2, rc);
2568 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2569 handle, mdd_object_capa(env, mdd_sobj));
2571 GOTO(fixup_spobj, rc);
2574 /* Remove target name from target directory
2575 * Here tobj can be remote one, so we do index_delete unconditionally
2576 * and -ENOENT is allowed.
2578 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2579 mdd_object_capa(env, mdd_tpobj));
2582 /* tname might been renamed to something else */
2583 GOTO(fixup_spobj, rc);
2586 GOTO(fixup_spobj, rc);
2589 /* Insert new fid with target name into target dir */
2590 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2591 mdd_object_capa(env, mdd_tpobj));
2593 GOTO(fixup_tpobj, rc);
2595 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2596 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2598 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2600 la->la_valid = LA_CTIME;
2601 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2603 GOTO(fixup_tpobj, rc);
2606 /* Remove old target object
2607 * For tobj is remote case cmm layer has processed
2608 * and set tobj to NULL then. So when tobj is NOT NULL,
2609 * it must be local one.
2611 if (tobj && mdd_object_exists(mdd_tobj)) {
2612 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2614 if (mdd_is_dead_obj(mdd_tobj)) {
2615 /* shld not be dead, something is wrong */
2616 CERROR("tobj is dead, something is wrong\n");
2620 mdo_ref_del(env, mdd_tobj, handle);
2622 /* Remove dot reference. */
2623 if (S_ISDIR(tg_attr->la_mode))
2624 mdo_ref_del(env, mdd_tobj, handle);
2627 /* fetch updated nlink */
2628 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2629 mdd_object_capa(env, mdd_tobj));
2631 CERROR("%s: Failed to get nlink for tobj "
2633 mdd2obd_dev(mdd)->obd_name,
2634 PFID(tpobj_fid), rc);
2635 GOTO(fixup_tpobj, rc);
2638 la->la_valid = LA_CTIME;
2639 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2641 CERROR("%s: Failed to set ctime for tobj "
2643 mdd2obd_dev(mdd)->obd_name,
2644 PFID(tpobj_fid), rc);
2645 GOTO(fixup_tpobj, rc);
2648 /* XXX: this transfer to ma will be removed with LOD/OSP */
2649 ma->ma_attr = *tg_attr;
2650 ma->ma_valid |= MA_INODE;
2651 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2653 CERROR("%s: Failed to unlink tobj "
2655 mdd2obd_dev(mdd)->obd_name,
2656 PFID(tpobj_fid), rc);
2657 GOTO(fixup_tpobj, rc);
2660 /* fetch updated nlink */
2661 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2662 mdd_object_capa(env, mdd_tobj));
2664 CERROR("%s: Failed to get nlink for tobj "
2666 mdd2obd_dev(mdd)->obd_name,
2667 PFID(tpobj_fid), rc);
2668 GOTO(fixup_tpobj, rc);
2670 /* XXX: this transfer to ma will be removed with LOD/OSP */
2671 ma->ma_attr = *tg_attr;
2672 ma->ma_valid |= MA_INODE;
2674 if (so_attr->la_nlink == 0)
2675 cl_flags |= CLF_RENAME_LAST;
2678 la->la_valid = LA_CTIME | LA_MTIME;
2679 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2681 GOTO(fixup_tpobj, rc);
2683 if (mdd_spobj != mdd_tpobj) {
2684 la->la_valid = LA_CTIME | LA_MTIME;
2685 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2689 if (rc == 0 && mdd_sobj) {
2690 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2691 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2692 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2695 /* Old files might not have EA entry */
2696 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2697 lsname, handle, NULL, 0);
2698 mdd_write_unlock(env, mdd_sobj);
2699 /* We don't fail the transaction if the link ea can't be
2700 updated -- fid2path will use alternate lookup method. */
2708 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2711 CWARN("tp obj fix error %d\n",rc2);
2713 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2714 !mdd_is_dead_obj(mdd_tobj)) {
2716 mdo_ref_add(env, mdd_tobj, handle);
2718 mdo_ref_add(env, mdd_tobj, handle);
2721 rc2 = __mdd_index_insert(env, mdd_tpobj,
2722 mdo2fid(mdd_tobj), tname,
2727 CWARN("tp obj fix error %d\n",rc2);
2732 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2733 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2737 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2738 mdd2obd_dev(mdd)->obd_name, rc2);
2741 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2742 dotdot, handle, BYPASS_CAPA);
2744 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2745 mdd2obd_dev(mdd)->obd_name, rc2);
2750 rc2 = __mdd_index_insert(env, mdd_spobj,
2751 lf, sname, is_dir, handle, BYPASS_CAPA);
2753 CWARN("sp obj fix error %d\n",rc2);
2757 mdd_write_unlock(env, mdd_tobj);
2758 if (likely(tdlh) && sdlh != tdlh)
2759 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2761 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2764 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2765 mdd_tobj, tpobj_fid, lf,
2766 spobj_fid, ltname, lsname,
2770 mdd_trans_stop(env, mdd, rc, handle);
2772 mdd_object_put(env, mdd_sobj);
2776 const struct md_dir_operations mdd_dir_ops = {
2777 .mdo_is_subdir = mdd_is_subdir,
2778 .mdo_lookup = mdd_lookup,
2779 .mdo_create = mdd_create,
2780 .mdo_rename = mdd_rename,
2781 .mdo_link = mdd_link,
2782 .mdo_unlink = mdd_unlink,
2783 .mdo_create_data = mdd_create_data,