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,
951 rc = rc != 0 ? rc : rc2;
956 int mdd_links_rename(const struct lu_env *env,
957 struct mdd_object *mdd_obj,
958 const struct lu_fid *oldpfid,
959 const struct lu_name *oldlname,
960 const struct lu_fid *newpfid,
961 const struct lu_name *newlname,
962 struct thandle *handle,
963 struct linkea_data *ldata,
964 int first, int check)
971 ldata = &mdd_env_info(env)->mti_link_data;
972 memset(ldata, 0, sizeof(*ldata));
973 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
974 newpfid, newlname, first, check,
980 if (ldata->ld_reclen != 0)
981 rc = mdd_links_write(env, mdd_obj, ldata, handle);
988 if (rc == -EOVERFLOW || rc == -ENOSPC)
991 CDEBUG(error ? D_ERROR : D_OTHER,
992 "link_ea add '%.*s' failed %d "DFID"\n",
993 newlname->ln_namelen, newlname->ln_name,
994 rc, PFID(mdd_object_fid(mdd_obj)));
995 else if (newpfid == NULL)
996 CDEBUG(error ? D_ERROR : D_OTHER,
997 "link_ea del '%.*s' failed %d "DFID"\n",
998 oldlname->ln_namelen, oldlname->ln_name,
999 rc, PFID(mdd_object_fid(mdd_obj)));
1001 CDEBUG(error ? D_ERROR : D_OTHER,
1002 "link_ea rename '%.*s'->'%.*s' failed %d "
1004 oldlname->ln_namelen, oldlname->ln_name,
1005 newlname->ln_namelen, newlname->ln_name,
1006 rc, PFID(mdd_object_fid(mdd_obj)));
1009 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1010 /* if we vmalloced a large buffer drop it */
1011 lu_buf_free(ldata->ld_buf);
1016 static inline int mdd_links_add(const struct lu_env *env,
1017 struct mdd_object *mdd_obj,
1018 const struct lu_fid *pfid,
1019 const struct lu_name *lname,
1020 struct thandle *handle,
1021 struct linkea_data *data, int first)
1023 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1024 pfid, lname, handle, data, first, 0);
1027 static inline int mdd_links_del(const struct lu_env *env,
1028 struct mdd_object *mdd_obj,
1029 const struct lu_fid *pfid,
1030 const struct lu_name *lname,
1031 struct thandle *handle)
1033 return mdd_links_rename(env, mdd_obj, pfid, lname,
1034 NULL, NULL, handle, NULL, 0, 0);
1037 /** Read the link EA into a temp buffer.
1038 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1039 * A pointer to the buffer is stored in \a ldata::ld_buf.
1041 * \retval 0 or error
1043 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1044 struct linkea_data *ldata)
1048 /* First try a small buf */
1049 LASSERT(env != NULL);
1050 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1052 if (ldata->ld_buf->lb_buf == NULL)
1055 if (!mdd_object_exists(mdd_obj))
1058 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1060 if (rc == -ERANGE) {
1061 /* Buf was too small, figure out what we need. */
1062 lu_buf_free(ldata->ld_buf);
1063 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1064 XATTR_NAME_LINK, BYPASS_CAPA);
1067 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1068 if (ldata->ld_buf->lb_buf == NULL)
1070 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1071 XATTR_NAME_LINK, BYPASS_CAPA);
1076 return linkea_init(ldata);
1079 /** Read the link EA into a temp buffer.
1080 * Uses the name_buf since it is generally large.
1081 * \retval IS_ERR err
1082 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1084 struct lu_buf *mdd_links_get(const struct lu_env *env,
1085 struct mdd_object *mdd_obj)
1087 struct linkea_data ldata = { 0 };
1090 rc = mdd_links_read(env, mdd_obj, &ldata);
1091 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1094 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1095 struct linkea_data *ldata, struct thandle *handle)
1097 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1098 ldata->ld_leh->leh_len);
1099 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1100 mdd_object_capa(env, mdd_obj));
1103 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1104 struct thandle *handle, struct linkea_data *ldata)
1110 if (ldata != NULL && ldata->ld_lee != NULL) {
1111 ea_len = ldata->ld_leh->leh_len;
1112 linkea = ldata->ld_buf->lb_buf;
1114 ea_len = DEFAULT_LINKEA_SIZE;
1118 /* XXX: max size? */
1119 rc = mdo_declare_xattr_set(env, mdd_obj,
1120 mdd_buf_get_const(env, linkea, ea_len),
1121 XATTR_NAME_LINK, 0, handle);
1125 static inline int mdd_declare_links_del(const struct lu_env *env,
1126 struct mdd_object *c,
1127 struct thandle *handle)
1131 /* For directory, the linkEA will be removed together
1132 * with the object. */
1133 if (!S_ISDIR(mdd_object_type(c)))
1134 rc = mdd_declare_links_add(env, c, handle, NULL);
1139 static int mdd_declare_link(const struct lu_env *env,
1140 struct mdd_device *mdd,
1141 struct mdd_object *p,
1142 struct mdd_object *c,
1143 const struct lu_name *name,
1144 struct thandle *handle,
1146 struct linkea_data *data)
1150 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
1154 rc = mdo_declare_ref_add(env, c, handle);
1158 la->la_valid = LA_CTIME | LA_MTIME;
1159 rc = mdo_declare_attr_set(env, p, la, handle);
1163 la->la_valid = LA_CTIME;
1164 rc = mdo_declare_attr_set(env, c, la, handle);
1168 rc = mdd_declare_links_add(env, c, handle, data);
1172 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1177 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1178 struct md_object *src_obj, const struct lu_name *lname,
1181 const char *name = lname->ln_name;
1182 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1183 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1184 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1185 struct mdd_device *mdd = mdo2mdd(src_obj);
1186 struct dynlock_handle *dlh;
1187 struct thandle *handle;
1188 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1192 handle = mdd_trans_create(env, mdd);
1194 GOTO(out_pending, rc = PTR_ERR(handle));
1196 memset(ldata, 0, sizeof(*ldata));
1198 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1199 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1201 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1206 rc = mdd_trans_start(env, mdd, handle);
1210 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
1212 GOTO(out_trans, rc = -ENOMEM);
1213 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1215 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
1217 GOTO(out_unlock, rc);
1219 rc = mdo_ref_add(env, mdd_sobj, handle);
1221 GOTO(out_unlock, rc);
1224 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1226 mdd_object_capa(env, mdd_tobj));
1228 mdo_ref_del(env, mdd_sobj, handle);
1229 GOTO(out_unlock, rc);
1232 la->la_valid = LA_CTIME | LA_MTIME;
1233 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1235 GOTO(out_unlock, rc);
1237 la->la_valid = LA_CTIME;
1238 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
1240 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1241 mdo2fid(mdd_tobj), lname, 0, 0,
1244 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1245 lname, handle, ldata, 0);
1246 /* The failure of links_add should not cause the link
1247 * failure, reset rc here */
1252 mdd_write_unlock(env, mdd_sobj);
1253 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
1256 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1257 mdd_tobj, lname, handle);
1259 mdd_trans_stop(env, mdd, rc, handle);
1261 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1262 /* if we vmalloced a large buffer drop it */
1263 lu_buf_free(ldata->ld_buf);
1268 int mdd_declare_finish_unlink(const struct lu_env *env,
1269 struct mdd_object *obj,
1271 struct thandle *handle)
1275 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1279 return mdo_declare_destroy(env, obj, handle);
1282 /* caller should take a lock before calling */
1283 int mdd_finish_unlink(const struct lu_env *env,
1284 struct mdd_object *obj, struct md_attr *ma,
1288 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1291 LASSERT(mdd_write_locked(env, obj) != 0);
1293 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1294 obj->mod_flags |= DEAD_OBJ;
1296 /* add new orphan and the object
1297 * will be deleted during mdd_close() */
1298 if (obj->mod_count) {
1299 rc = __mdd_orphan_add(env, obj, th);
1301 CDEBUG(D_HA, "Object "DFID" is inserted into "
1302 "orphan list, open count = %d\n",
1303 PFID(mdd_object_fid(obj)),
1306 CERROR("Object "DFID" fail to be an orphan, "
1307 "open count = %d, maybe cause failed "
1309 PFID(mdd_object_fid(obj)),
1312 rc = mdo_destroy(env, obj, th);
1321 * has mdd_write_lock on cobj already, but not on pobj yet
1323 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1324 struct mdd_object *cobj, struct lu_attr *cattr)
1329 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1334 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1335 struct mdd_object *p, struct mdd_object *c,
1336 const struct lu_name *name, struct md_attr *ma,
1337 struct thandle *handle, int no_name)
1339 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1342 if (likely(no_name == 0)) {
1343 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1348 rc = mdo_declare_ref_del(env, p, handle);
1352 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1353 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1354 la->la_valid = LA_CTIME | LA_MTIME;
1355 rc = mdo_declare_attr_set(env, p, la, handle);
1360 rc = mdo_declare_ref_del(env, c, handle);
1364 rc = mdo_declare_ref_del(env, c, handle);
1368 la->la_valid = LA_CTIME;
1369 rc = mdo_declare_attr_set(env, c, la, handle);
1373 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1377 rc = mdd_declare_links_del(env, c, handle);
1381 /* FIXME: need changelog for remove entry */
1382 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1389 * test if a file has an HSM archive
1390 * if HSM attributes are not found in ma update them from
1393 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1394 struct mdd_object *obj,
1399 if (!(ma->ma_valid & MA_HSM)) {
1400 /* no HSM MD provided, read xattr */
1401 struct lu_buf *hsm_buf;
1402 const size_t buflen = sizeof(struct hsm_attrs);
1405 hsm_buf = mdd_buf_get(env, NULL, 0);
1406 lu_buf_alloc(hsm_buf, buflen);
1407 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM,
1408 mdd_object_capa(env, obj));
1409 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1410 lu_buf_free(hsm_buf);
1414 ma->ma_valid = MA_HSM;
1416 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1422 * Delete name entry and the object.
1423 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1424 * does not exist for this object, and it could only happen during resending
1425 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1426 * is also needed in this case(needed by changelog), so we have to add another
1427 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1428 * the ENOENT failure should be able to be fixed by redo mechanism.
1430 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1431 struct md_object *cobj, const struct lu_name *lname,
1432 struct md_attr *ma, int no_name)
1434 const char *name = lname->ln_name;
1435 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1436 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1437 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1438 struct mdd_object *mdd_cobj = NULL;
1439 struct mdd_device *mdd = mdo2mdd(pobj);
1440 struct dynlock_handle *dlh;
1441 struct thandle *handle;
1445 /* cobj == NULL means only delete name entry */
1446 if (likely(cobj != NULL)) {
1447 mdd_cobj = md2mdd_obj(cobj);
1448 if (mdd_object_exists(mdd_cobj) == 0)
1450 /* currently it is assume, it could only delete
1451 * name entry of remote directory */
1455 handle = mdd_trans_create(env, mdd);
1457 RETURN(PTR_ERR(handle));
1459 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1460 lname, ma, handle, no_name);
1464 rc = mdd_trans_start(env, mdd, handle);
1468 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1470 GOTO(stop, rc = -ENOMEM);
1472 if (likely(mdd_cobj != NULL)) {
1473 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1476 rc = mdd_la_get(env, mdd_cobj, cattr,
1477 mdd_object_capa(env, mdd_cobj));
1481 is_dir = S_ISDIR(cattr->la_mode);
1485 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1489 if (likely(no_name == 0)) {
1490 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1491 mdd_object_capa(env, mdd_pobj));
1496 if (likely(mdd_cobj != NULL)) {
1497 rc = mdo_ref_del(env, mdd_cobj, handle);
1499 __mdd_index_insert_only(env, mdd_pobj,
1502 mdd_object_capa(env, mdd_pobj));
1508 mdo_ref_del(env, mdd_cobj, handle);
1510 /* fetch updated nlink */
1511 rc = mdd_la_get(env, mdd_cobj, cattr,
1512 mdd_object_capa(env, mdd_cobj));
1517 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1518 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1520 la->la_valid = LA_CTIME | LA_MTIME;
1521 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1525 /* Enough for only unlink the entry */
1526 if (unlikely(mdd_cobj == NULL)) {
1527 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1531 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1532 /* update ctime of an unlinked file only if it is still
1533 * opened or a link still exists */
1534 la->la_valid = LA_CTIME;
1535 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1540 /* XXX: this transfer to ma will be removed with LOD/OSP */
1541 ma->ma_attr = *cattr;
1542 ma->ma_valid |= MA_INODE;
1543 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1545 /* fetch updated nlink */
1547 rc = mdd_la_get(env, mdd_cobj, cattr,
1548 mdd_object_capa(env, mdd_cobj));
1551 /* old files may not have link ea; ignore errors */
1552 mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle);
1554 /* if object is removed then we can't get its attrs, use last get */
1555 if (cattr->la_nlink == 0) {
1556 ma->ma_attr = *cattr;
1557 ma->ma_valid |= MA_INODE;
1561 mdd_write_unlock(env, mdd_cobj);
1562 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1566 if (cattr->la_nlink == 0) {
1567 cl_flags |= CLF_UNLINK_LAST;
1568 /* search for an existing archive */
1569 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1570 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1573 rc = mdd_changelog_ns_store(env, mdd,
1574 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1575 mdd_cobj, mdd_pobj, lname, handle);
1579 mdd_trans_stop(env, mdd, rc, handle);
1585 * The permission has been checked when obj created, no need check again.
1587 static int mdd_cd_sanity_check(const struct lu_env *env,
1588 struct mdd_object *obj)
1593 if (!obj || mdd_is_dead_obj(obj))
1600 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1601 struct md_object *cobj, const struct md_op_spec *spec,
1604 struct mdd_device *mdd = mdo2mdd(cobj);
1605 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1606 struct mdd_object *son = md2mdd_obj(cobj);
1607 struct thandle *handle;
1608 const struct lu_buf *buf;
1609 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1613 rc = mdd_cd_sanity_check(env, son);
1617 if (!md_should_create(spec->sp_cr_flags))
1621 * there are following use cases for this function:
1622 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1623 * striping can be specified or not
1626 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1630 /* calling ->ah_make_hint() is used to transfer information from parent */
1631 mdd_object_make_hint(env, mdd_pobj, son, attr);
1633 handle = mdd_trans_create(env, mdd);
1635 GOTO(out_free, rc = PTR_ERR(handle));
1638 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1639 * Should this be fixed?
1641 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1642 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1643 spec->sp_cr_flags, spec->no_create);
1645 if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1646 /* replay case or lfs setstripe */
1647 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1648 spec->u.sp_ea.eadatalen);
1653 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1654 XATTR_NAME_LOV, 0, handle);
1658 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1662 rc = mdd_trans_start(env, mdd, handle);
1666 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1667 0, handle, mdd_object_capa(env, son));
1672 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1675 mdd_trans_stop(env, mdd, rc, handle);
1680 /* Get fid from name and parent */
1682 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1683 const struct lu_name *lname, struct lu_fid* fid, int mask)
1685 const char *name = lname->ln_name;
1686 const struct dt_key *key = (const struct dt_key *)name;
1687 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1688 struct mdd_device *m = mdo2mdd(pobj);
1689 struct dt_object *dir = mdd_object_child(mdd_obj);
1693 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1696 if (mdd_object_remote(mdd_obj)) {
1697 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
1698 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
1699 } else if (!mdd_object_exists(mdd_obj)) {
1703 /* The common filename length check. */
1704 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1705 RETURN(-ENAMETOOLONG);
1707 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1712 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1713 dt_try_as_dir(env, dir))) {
1715 rc = dir->do_index_ops->dio_lookup(env, dir,
1716 (struct dt_rec *)fid, key,
1717 mdd_object_capa(env, mdd_obj));
1728 static int mdd_declare_object_initialize(const struct lu_env *env,
1729 struct mdd_object *parent,
1730 struct mdd_object *child,
1731 struct lu_attr *attr,
1732 struct thandle *handle,
1733 struct linkea_data *ldata)
1739 * inode mode has been set in creation time, and it's based on umask,
1740 * la_mode and acl, don't set here again! (which will go wrong
1741 * because below function doesn't consider umask).
1742 * I'd suggest set all object attributes in creation time, see above.
1744 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1745 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1746 rc = mdo_declare_attr_set(env, child, attr, handle);
1747 attr->la_valid |= LA_MODE | LA_TYPE;
1748 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1749 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1752 rc = mdo_declare_ref_add(env, child, handle);
1754 rc = mdo_declare_index_insert(env, child, mdo2fid(parent),
1759 mdd_declare_links_add(env, child, handle, ldata);
1764 static int mdd_object_initialize(const struct lu_env *env,
1765 const struct lu_fid *pfid,
1766 const struct lu_name *lname,
1767 struct mdd_object *child,
1768 struct lu_attr *attr, struct thandle *handle,
1769 const struct md_op_spec *spec,
1770 struct linkea_data *ldata)
1776 * Update attributes for child.
1779 * (1) the valid bits should be converted between Lustre and Linux;
1780 * (2) maybe, the child attributes should be set in OSD when creation.
1783 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1784 /* arguments are supposed to stay the same */
1785 if (S_ISDIR(attr->la_mode)) {
1786 /* Add "." and ".." for newly created dir */
1787 mdo_ref_add(env, child, handle);
1788 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1789 dot, handle, BYPASS_CAPA);
1791 rc = __mdd_index_insert_only(env, child, pfid,
1795 mdo_ref_del(env, child, handle);
1799 mdd_links_add(env, child, pfid, lname, handle, ldata, 1);
1804 /* has not lock on pobj yet */
1805 static int mdd_create_sanity_check(const struct lu_env *env,
1806 struct md_object *pobj,
1807 struct lu_attr *pattr,
1808 const struct lu_name *lname,
1809 struct lu_attr *cattr,
1810 struct md_op_spec *spec)
1812 struct mdd_thread_info *info = mdd_env_info(env);
1813 struct lu_fid *fid = &info->mti_fid;
1814 struct mdd_object *obj = md2mdd_obj(pobj);
1815 struct mdd_device *m = mdo2mdd(pobj);
1820 if (mdd_is_dead_obj(obj))
1824 * In some cases this lookup is not needed - we know before if name
1825 * exists or not because MDT performs lookup for it.
1826 * name length check is done in lookup.
1828 if (spec->sp_cr_lookup) {
1830 * Check if the name already exist, though it will be checked in
1831 * _index_insert also, for avoiding rolling back if exists
1834 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1835 MAY_WRITE | MAY_EXEC);
1837 RETURN(rc ? : -EEXIST);
1840 * Check WRITE permission for the parent.
1841 * EXEC permission have been checked
1842 * when lookup before create already.
1844 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1851 if (pattr->la_mode & S_ISGID) {
1852 cattr->la_gid = pattr->la_gid;
1853 if (S_ISDIR(cattr->la_mode)) {
1854 cattr->la_mode |= S_ISGID;
1855 cattr->la_valid |= LA_MODE;
1859 switch (cattr->la_mode & S_IFMT) {
1861 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1863 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1864 RETURN(-ENAMETOOLONG);
1883 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1884 struct mdd_object *p, struct mdd_object *c,
1885 const struct lu_name *name,
1886 struct lu_attr *attr,
1887 struct thandle *handle,
1888 const struct md_op_spec *spec,
1889 struct linkea_data *ldata,
1890 struct lu_buf *def_acl_buf,
1891 struct lu_buf *acl_buf)
1895 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1899 #ifdef CONFIG_FS_POSIX_ACL
1900 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1901 /* if dir, then can inherit default ACl */
1902 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1903 XATTR_NAME_ACL_DEFAULT,
1909 if (acl_buf->lb_len > 0) {
1910 rc = mdo_declare_attr_set(env, c, attr, handle);
1914 rc = mdo_declare_xattr_set(env, c, acl_buf,
1915 XATTR_NAME_ACL_ACCESS, 0, handle);
1921 if (S_ISDIR(attr->la_mode)) {
1922 rc = mdo_declare_ref_add(env, p, handle);
1927 rc = mdd_declare_object_initialize(env, p, c, attr, handle, ldata);
1931 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1932 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
1934 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1935 name->ln_name, handle);
1939 /* replay case, create LOV EA from client data */
1940 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1941 const struct lu_buf *buf;
1943 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1944 spec->u.sp_ea.eadatalen);
1945 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1951 if (S_ISLNK(attr->la_mode)) {
1952 rc = dt_declare_record_write(env, mdd_object_child(c),
1953 strlen(spec->u.sp_symname), 0,
1959 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
1960 rc = mdo_declare_attr_set(env, p, attr, handle);
1965 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1973 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
1974 struct lu_attr *la, struct lu_buf *def_acl_buf,
1975 struct lu_buf *acl_buf)
1980 if (S_ISLNK(la->la_mode)) {
1981 acl_buf->lb_len = 0;
1982 def_acl_buf->lb_len = 0;
1986 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
1987 rc = mdo_xattr_get(env, pobj, def_acl_buf,
1988 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1989 mdd_read_unlock(env, pobj);
1991 /* If there are default ACL, fix mode/ACL by default ACL */
1992 def_acl_buf->lb_len = rc;
1993 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
1994 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
1995 acl_buf->lb_len = rc;
1996 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
1999 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2000 /* If there are no default ACL, fix mode by mask */
2001 struct lu_ucred *uc = lu_ucred(env);
2003 /* The create triggered by MDT internal events, such as
2004 * LFSCK reset, will not contain valid "uc". */
2005 if (unlikely(uc != NULL))
2006 la->la_mode &= ~uc->uc_umask;
2008 acl_buf->lb_len = 0;
2009 def_acl_buf->lb_len = 0;
2016 * Create object and insert it into namespace.
2018 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2019 const struct lu_name *lname, struct md_object *child,
2020 struct md_op_spec *spec, struct md_attr* ma)
2022 struct mdd_thread_info *info = mdd_env_info(env);
2023 struct lu_attr *la = &info->mti_la_for_fix;
2024 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2025 struct mdd_object *son = md2mdd_obj(child);
2026 struct mdd_device *mdd = mdo2mdd(pobj);
2027 struct lu_attr *attr = &ma->ma_attr;
2028 struct thandle *handle;
2029 struct lu_attr *pattr = &info->mti_pattr;
2030 struct lu_buf acl_buf;
2031 struct lu_buf def_acl_buf;
2032 struct linkea_data *ldata = &info->mti_link_data;
2033 struct dynlock_handle *dlh;
2034 const char *name = lname->ln_name;
2035 int rc, created = 0, initialized = 0, inserted = 0;
2039 * Two operations have to be performed:
2041 * - an allocation of a new object (->do_create()), and
2043 * - an insertion into a parent index (->dio_insert()).
2045 * Due to locking, operation order is not important, when both are
2046 * successful, *but* error handling cases are quite different:
2048 * - if insertion is done first, and following object creation fails,
2049 * insertion has to be rolled back, but this operation might fail
2050 * also leaving us with dangling index entry.
2052 * - if creation is done first, is has to be undone if insertion
2053 * fails, leaving us with leaked space, which is neither good, nor
2056 * It seems that creation-first is simplest solution, but it is
2057 * sub-optimal in the frequent
2062 * case, because second mkdir is bound to create object, only to
2063 * destroy it immediately.
2065 * To avoid this follow local file systems that do double lookup:
2067 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2069 * 1. create (mdd_object_create_internal())
2071 * 2. insert (__mdd_index_insert(), lookup again)
2074 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2078 /* Sanity checks before big job. */
2079 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2083 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2084 GOTO(out_free, rc = -EINPROGRESS);
2086 acl_buf.lb_buf = info->mti_xattr_buf;
2087 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2088 def_acl_buf.lb_buf = info->mti_key;
2089 def_acl_buf.lb_len = sizeof(info->mti_key);
2090 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2094 mdd_object_make_hint(env, mdd_pobj, son, attr);
2096 handle = mdd_trans_create(env, mdd);
2098 GOTO(out_free, rc = PTR_ERR(handle));
2100 memset(ldata, 0, sizeof(*ldata));
2101 mdd_linkea_prepare(env, son, NULL, NULL, mdd_object_fid(mdd_pobj),
2102 lname, 1, 0, ldata);
2104 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2105 handle, spec, ldata, &def_acl_buf, &acl_buf);
2109 rc = mdd_trans_start(env, mdd, handle);
2113 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
2115 GOTO(out_trans, rc = -ENOMEM);
2117 mdd_write_lock(env, son, MOR_TGT_CHILD);
2118 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec);
2120 mdd_write_unlock(env, son);
2126 #ifdef CONFIG_FS_POSIX_ACL
2127 if (def_acl_buf.lb_len > 0 && S_ISDIR(attr->la_mode)) {
2128 /* set default acl */
2129 rc = mdo_xattr_set(env, son, &def_acl_buf,
2130 XATTR_NAME_ACL_DEFAULT, 0,
2131 handle, BYPASS_CAPA);
2133 mdd_write_unlock(env, son);
2137 /* set its own acl */
2138 if (acl_buf.lb_len > 0) {
2139 rc = mdo_xattr_set(env, son, &acl_buf,
2140 XATTR_NAME_ACL_ACCESS,
2141 0, handle, BYPASS_CAPA);
2143 mdd_write_unlock(env, son);
2149 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
2150 son, attr, handle, spec, ldata);
2153 * in case of replay we just set LOVEA provided by the client
2154 * XXX: I think it would be interesting to try "old" way where
2155 * MDT calls this xattr_set(LOV) in a different transaction.
2156 * probably this way we code can be made better.
2158 if (rc == 0 && (spec->no_create ||
2159 (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
2160 const struct lu_buf *buf;
2162 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2163 spec->u.sp_ea.eadatalen);
2164 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
2168 if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2169 rc = __mdd_orphan_add(env, son, handle);
2171 mdd_write_unlock(env, son);
2175 * Object has no links, so it will be destroyed when last
2176 * reference is released. (XXX not now.)
2182 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE))
2183 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2184 name, S_ISDIR(attr->la_mode), handle,
2185 mdd_object_capa(env, mdd_pobj));
2192 if (S_ISLNK(attr->la_mode)) {
2193 struct lu_ucred *uc = lu_ucred_assert(env);
2194 struct dt_object *dt = mdd_object_child(son);
2195 const char *target_name = spec->u.sp_symname;
2196 int sym_len = strlen(target_name);
2197 const struct lu_buf *buf;
2200 buf = mdd_buf_get_const(env, target_name, sym_len);
2201 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2202 mdd_object_capa(env, son),
2204 CFS_CAP_SYS_RESOURCE_MASK);
2209 GOTO(cleanup, rc = -EFAULT);
2212 /* volatile file creation does not update parent directory times */
2213 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2214 GOTO(cleanup, rc = 0);
2216 /* update parent directory mtime/ctime */
2218 la->la_valid = LA_CTIME | LA_MTIME;
2219 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
2225 if (rc != 0 && created != 0) {
2228 if (inserted != 0) {
2229 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2230 rc2 = __mdd_orphan_del(env, son, handle);
2232 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2233 S_ISDIR(attr->la_mode),
2234 handle, BYPASS_CAPA);
2239 mdd_write_lock(env, son, MOR_TGT_CHILD);
2240 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
2241 /* Drop the reference, no need to delete "."/"..",
2242 * because the object to be destroied directly. */
2243 rc2 = mdo_ref_del(env, son, handle);
2245 mdd_write_unlock(env, son);
2250 rc2 = mdo_ref_del(env, son, handle);
2252 mdd_write_unlock(env, son);
2256 mdo_destroy(env, son, handle);
2257 mdd_write_unlock(env, son);
2260 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
2262 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)))
2263 rc = mdd_changelog_ns_store(env, mdd,
2264 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2265 S_ISREG(attr->la_mode) ? CL_CREATE :
2266 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2267 0, son, mdd_pobj, lname, handle);
2269 mdd_trans_stop(env, mdd, rc, handle);
2271 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2272 /* if we vmalloced a large buffer drop it */
2273 lu_buf_free(ldata->ld_buf);
2275 /* The child object shouldn't be cached anymore */
2277 set_bit(LU_OBJECT_HEARD_BANSHEE,
2278 &child->mo_lu.lo_header->loh_flags);
2283 * Get locks on parents in proper order
2284 * RETURN: < 0 - error, rename_order if successful
2292 static int mdd_rename_order(const struct lu_env *env,
2293 struct mdd_device *mdd,
2294 struct mdd_object *src_pobj,
2295 struct mdd_object *tgt_pobj)
2297 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2301 if (src_pobj == tgt_pobj)
2302 RETURN(MDD_RN_SAME);
2304 /* compared the parent child relationship of src_p&tgt_p */
2305 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2307 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2310 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
2323 /* has not mdd_write{read}_lock on any obj yet. */
2324 static int mdd_rename_sanity_check(const struct lu_env *env,
2325 struct mdd_object *src_pobj,
2326 struct mdd_object *tgt_pobj,
2327 struct mdd_object *sobj,
2328 struct mdd_object *tobj,
2329 struct lu_attr *so_attr,
2330 struct lu_attr *tg_attr)
2335 /* XXX: when get here, sobj must NOT be NULL,
2336 * the other case has been processed in cld_rename
2337 * before mdd_rename and enable MDS_PERM_BYPASS. */
2340 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
2344 /* XXX: when get here, "tobj == NULL" means tobj must
2345 * NOT exist (neither on remote MDS, such case has been
2346 * processed in cld_rename before mdd_rename and enable
2348 * So check may_create, but not check may_unlink. */
2350 rc = mdd_may_create(env, tgt_pobj, NULL,
2351 (src_pobj != tgt_pobj), 0);
2353 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
2354 (src_pobj != tgt_pobj), 1);
2356 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2357 S_ISDIR(so_attr->la_mode))
2358 rc = __mdd_may_link(env, tgt_pobj);
2363 static int mdd_declare_rename(const struct lu_env *env,
2364 struct mdd_device *mdd,
2365 struct mdd_object *mdd_spobj,
2366 struct mdd_object *mdd_tpobj,
2367 struct mdd_object *mdd_sobj,
2368 struct mdd_object *mdd_tobj,
2369 const struct lu_name *tname,
2370 const struct lu_name *sname,
2372 struct linkea_data *ldata,
2373 struct thandle *handle)
2375 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2378 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2379 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2385 /* name from source dir */
2386 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2390 /* .. from source child */
2391 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2392 /* source child can be directory,
2393 * counted by source dir's nlink */
2394 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2397 if (mdd_spobj != mdd_tpobj) {
2398 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2403 rc = mdo_declare_index_insert(env, mdd_sobj,
2409 /* new target child can be directory,
2410 * counted by target dir's nlink */
2411 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2417 la->la_valid = LA_CTIME | LA_MTIME;
2418 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2422 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2426 la->la_valid = LA_CTIME;
2427 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2431 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2436 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2437 tname->ln_name, handle);
2441 /* name from target dir (old name), we declare it unconditionally
2442 * as mdd_rename() calls delete unconditionally as well. so just
2443 * to balance declarations vs calls to change ... */
2444 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2448 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2449 /* delete target child in target parent directory */
2450 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2454 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2455 /* target child can be directory,
2456 * delete "." reference in target child directory */
2457 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2461 /* delete ".." reference in target parent directory */
2462 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2467 la->la_valid = LA_CTIME;
2468 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2472 rc = mdd_declare_links_del(env, mdd_tobj, handle);
2476 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2481 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2488 /* src object can be remote that is why we use only fid and type of object */
2489 static int mdd_rename(const struct lu_env *env,
2490 struct md_object *src_pobj, struct md_object *tgt_pobj,
2491 const struct lu_fid *lf, const struct lu_name *lsname,
2492 struct md_object *tobj, const struct lu_name *ltname,
2495 const char *sname = lsname->ln_name;
2496 const char *tname = ltname->ln_name;
2497 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2498 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2499 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2500 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2501 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2502 struct mdd_device *mdd = mdo2mdd(src_pobj);
2503 struct mdd_object *mdd_sobj = NULL; /* source object */
2504 struct mdd_object *mdd_tobj = NULL;
2505 struct dynlock_handle *sdlh = NULL, *tdlh = NULL;
2506 struct thandle *handle;
2507 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2508 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2509 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2512 bool tobj_locked = 0;
2513 unsigned cl_flags = 0;
2518 mdd_tobj = md2mdd_obj(tobj);
2520 mdd_sobj = mdd_object_find(env, mdd, lf);
2522 rc = mdd_la_get(env, mdd_sobj, so_attr,
2523 mdd_object_capa(env, mdd_sobj));
2525 GOTO(out_pending, rc);
2528 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2529 mdd_object_capa(env, mdd_tobj));
2531 GOTO(out_pending, rc);
2534 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2535 mdd_tobj, so_attr, tg_attr);
2537 GOTO(out_pending, rc);
2539 handle = mdd_trans_create(env, mdd);
2541 GOTO(out_pending, rc = PTR_ERR(handle));
2543 memset(ldata, 0, sizeof(*ldata));
2544 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2545 ltname, 1, 0, ldata);
2546 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2547 mdd_tobj, lsname, ltname, ma, ldata, handle);
2551 rc = mdd_trans_start(env, mdd, handle);
2555 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2556 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2558 GOTO(cleanup_unlocked, rc);
2560 /* Get locks in determined order */
2561 if (rc == MDD_RN_SAME) {
2562 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2563 sname, MOR_SRC_PARENT);
2564 /* check hashes to determine do we need one lock or two */
2565 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2566 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2570 } else if (rc == MDD_RN_SRCTGT) {
2571 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2572 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2574 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2575 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2577 if (sdlh == NULL || tdlh == NULL)
2578 GOTO(cleanup, rc = -ENOMEM);
2580 is_dir = S_ISDIR(so_attr->la_mode);
2582 /* Remove source name from source directory */
2583 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2584 mdd_object_capa(env, mdd_spobj));
2588 /* "mv dir1 dir2" needs "dir1/.." link update */
2589 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2590 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2591 mdd_object_capa(env, mdd_sobj));
2593 GOTO(fixup_spobj2, rc);
2595 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2596 handle, mdd_object_capa(env, mdd_sobj));
2598 GOTO(fixup_spobj, rc);
2601 /* Remove target name from target directory
2602 * Here tobj can be remote one, so we do index_delete unconditionally
2603 * and -ENOENT is allowed.
2605 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2606 mdd_object_capa(env, mdd_tpobj));
2609 /* tname might been renamed to something else */
2610 GOTO(fixup_spobj, rc);
2613 GOTO(fixup_spobj, rc);
2616 /* Insert new fid with target name into target dir */
2617 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2618 mdd_object_capa(env, mdd_tpobj));
2620 GOTO(fixup_tpobj, rc);
2622 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2623 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2625 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2627 la->la_valid = LA_CTIME;
2628 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2630 GOTO(fixup_tpobj, rc);
2633 /* Remove old target object
2634 * For tobj is remote case cmm layer has processed
2635 * and set tobj to NULL then. So when tobj is NOT NULL,
2636 * it must be local one.
2638 if (tobj && mdd_object_exists(mdd_tobj)) {
2639 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2641 if (mdd_is_dead_obj(mdd_tobj)) {
2642 /* shld not be dead, something is wrong */
2643 CERROR("tobj is dead, something is wrong\n");
2647 mdo_ref_del(env, mdd_tobj, handle);
2649 /* Remove dot reference. */
2650 if (S_ISDIR(tg_attr->la_mode))
2651 mdo_ref_del(env, mdd_tobj, handle);
2654 /* fetch updated nlink */
2655 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2656 mdd_object_capa(env, mdd_tobj));
2658 CERROR("%s: Failed to get nlink for tobj "
2660 mdd2obd_dev(mdd)->obd_name,
2661 PFID(tpobj_fid), rc);
2662 GOTO(fixup_tpobj, rc);
2665 la->la_valid = LA_CTIME;
2666 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2668 CERROR("%s: Failed to set ctime for tobj "
2670 mdd2obd_dev(mdd)->obd_name,
2671 PFID(tpobj_fid), rc);
2672 GOTO(fixup_tpobj, rc);
2675 /* XXX: this transfer to ma will be removed with LOD/OSP */
2676 ma->ma_attr = *tg_attr;
2677 ma->ma_valid |= MA_INODE;
2678 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2680 CERROR("%s: Failed to unlink tobj "
2682 mdd2obd_dev(mdd)->obd_name,
2683 PFID(tpobj_fid), rc);
2684 GOTO(fixup_tpobj, rc);
2687 /* fetch updated nlink */
2688 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2689 mdd_object_capa(env, mdd_tobj));
2691 CERROR("%s: Failed to get nlink for tobj "
2693 mdd2obd_dev(mdd)->obd_name,
2694 PFID(tpobj_fid), rc);
2695 GOTO(fixup_tpobj, rc);
2697 /* XXX: this transfer to ma will be removed with LOD/OSP */
2698 ma->ma_attr = *tg_attr;
2699 ma->ma_valid |= MA_INODE;
2701 if (tg_attr->la_nlink == 0) {
2702 cl_flags |= CLF_RENAME_LAST;
2703 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2704 cl_flags |= CLF_RENAME_LAST_EXISTS;
2708 la->la_valid = LA_CTIME | LA_MTIME;
2709 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2711 GOTO(fixup_tpobj, rc);
2713 if (mdd_spobj != mdd_tpobj) {
2714 la->la_valid = LA_CTIME | LA_MTIME;
2715 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2719 if (rc == 0 && mdd_sobj) {
2720 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2721 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2722 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2725 /* Old files might not have EA entry */
2726 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2727 lsname, handle, NULL, 0);
2728 mdd_write_unlock(env, mdd_sobj);
2729 /* We don't fail the transaction if the link ea can't be
2730 updated -- fid2path will use alternate lookup method. */
2738 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2741 CWARN("tp obj fix error %d\n",rc2);
2743 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2744 !mdd_is_dead_obj(mdd_tobj)) {
2746 mdo_ref_add(env, mdd_tobj, handle);
2748 mdo_ref_add(env, mdd_tobj, handle);
2751 rc2 = __mdd_index_insert(env, mdd_tpobj,
2752 mdo2fid(mdd_tobj), tname,
2757 CWARN("tp obj fix error %d\n",rc2);
2762 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2763 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2767 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2768 mdd2obd_dev(mdd)->obd_name, rc2);
2771 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2772 dotdot, handle, BYPASS_CAPA);
2774 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2775 mdd2obd_dev(mdd)->obd_name, rc2);
2780 rc2 = __mdd_index_insert(env, mdd_spobj,
2781 lf, sname, is_dir, handle, BYPASS_CAPA);
2783 CWARN("sp obj fix error %d\n",rc2);
2787 mdd_write_unlock(env, mdd_tobj);
2788 if (likely(tdlh) && sdlh != tdlh)
2789 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2791 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2794 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2795 mdd_tobj, tpobj_fid, lf,
2796 spobj_fid, ltname, lsname,
2800 mdd_trans_stop(env, mdd, rc, handle);
2802 mdd_object_put(env, mdd_sobj);
2806 const struct md_dir_operations mdd_dir_ops = {
2807 .mdo_is_subdir = mdd_is_subdir,
2808 .mdo_lookup = mdd_lookup,
2809 .mdo_create = mdd_create,
2810 .mdo_rename = mdd_rename,
2811 .mdo_link = mdd_link,
2812 .mdo_unlink = mdd_unlink,
2813 .mdo_create_data = mdd_create_data,