4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, 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_llog_record(const struct lu_env *env, struct mdd_device *mdd,
558 int reclen, struct thandle *handle)
562 /* XXX: this is a temporary solution to declare llog changes
563 * will be fixed in 2.3 with new llog implementation */
565 LASSERT(mdd->mdd_capa);
567 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
568 * have to set the parameter 'size' as INT_MAX or 0 to inform
569 * OSD that this record write is for a llog write or catalog
570 * header update, and osd declare function will reserve less
571 * credits for optimization purpose.
573 * Reserve 6 blocks for a llog write, since the llog file is
574 * usually small, reserve 2 blocks for catalog header update,
575 * because we know for sure that catalog header is already
578 * This hack should be removed in 2.3.
582 rc = dt_declare_record_write(env, mdd->mdd_capa,
583 DECLARE_LLOG_WRITE, 0, handle);
587 /* header will be updated as well */
588 rc = dt_declare_record_write(env, mdd->mdd_capa,
589 DECLARE_LLOG_WRITE, 0, handle);
593 /* also we should be able to create new plain log */
594 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
598 /* new record referencing new plain llog */
599 rc = dt_declare_record_write(env, mdd->mdd_capa,
600 DECLARE_LLOG_WRITE, 0, handle);
604 /* catalog's header will be updated as well */
605 rc = dt_declare_record_write(env, mdd->mdd_capa,
606 DECLARE_LLOG_REWRITE, 0, handle);
611 int mdd_declare_changelog_store(const struct lu_env *env,
612 struct mdd_device *mdd,
613 const struct lu_name *fname,
614 struct thandle *handle)
616 struct obd_device *obd = mdd2obd_dev(mdd);
617 struct llog_ctxt *ctxt;
618 struct llog_changelog_rec *rec;
624 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
627 reclen = llog_data_len(sizeof(*rec) +
628 (fname != NULL ? fname->ln_namelen : 0));
629 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
630 if (buf->lb_buf == NULL)
634 rec->cr_hdr.lrh_len = reclen;
635 rec->cr_hdr.lrh_type = CHANGELOG_REC;
637 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
641 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
647 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
648 struct mdd_device *mdd,
649 const struct lu_name *tname,
650 const struct lu_name *sname,
651 struct thandle *handle)
653 struct obd_device *obd = mdd2obd_dev(mdd);
654 struct llog_ctxt *ctxt;
655 struct llog_changelog_ext_rec *rec;
661 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
664 reclen = llog_data_len(sizeof(*rec) +
665 (tname != NULL ? tname->ln_namelen : 0) +
666 (sname != NULL ? 1 + sname->ln_namelen : 0));
667 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
668 if (buf->lb_buf == NULL)
672 rec->cr_hdr.lrh_len = reclen;
673 rec->cr_hdr.lrh_type = CHANGELOG_REC;
675 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
679 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
685 /** Add a changelog entry \a rec to the changelog llog
688 * \param handle - currently ignored since llogs start their own transaction;
689 * this will hopefully be fixed in llog rewrite
692 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
693 struct llog_changelog_rec *rec, struct thandle *th)
695 struct obd_device *obd = mdd2obd_dev(mdd);
696 struct llog_ctxt *ctxt;
699 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
700 rec->cr_hdr.lrh_type = CHANGELOG_REC;
701 rec->cr.cr_time = cl_time();
703 spin_lock(&mdd->mdd_cl.mc_lock);
704 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
705 * but as long as the MDD transactions are ordered correctly for e.g.
706 * rename conflicts, I don't think this should matter. */
707 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
708 spin_unlock(&mdd->mdd_cl.mc_lock);
710 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
714 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
721 /** Add a changelog_ext entry \a rec to the changelog llog
724 * \param handle - currently ignored since llogs start their own transaction;
725 * this will hopefully be fixed in llog rewrite
728 int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
729 struct llog_changelog_ext_rec *rec,
732 struct obd_device *obd = mdd2obd_dev(mdd);
733 struct llog_ctxt *ctxt;
736 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
737 /* llog_lvfs_write_rec sets the llog tail len */
738 rec->cr_hdr.lrh_type = CHANGELOG_REC;
739 rec->cr.cr_time = cl_time();
741 spin_lock(&mdd->mdd_cl.mc_lock);
742 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
743 * but as long as the MDD transactions are ordered correctly for e.g.
744 * rename conflicts, I don't think this should matter. */
745 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
746 spin_unlock(&mdd->mdd_cl.mc_lock);
748 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
752 /* nested journal transaction */
753 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
761 /** Store a namespace change changelog record
762 * If this fails, we must fail the whole transaction; we don't
763 * want the change to commit without the log entry.
764 * \param target - mdd_object of change
765 * \param parent - parent dir/object
766 * \param tname - target name string
767 * \param handle - transacion handle
769 int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd,
770 enum changelog_rec_type type, unsigned flags,
771 struct mdd_object *target, struct mdd_object *parent,
772 const struct lu_name *tname, struct thandle *handle)
774 struct llog_changelog_rec *rec;
781 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
783 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
786 LASSERT(target != NULL);
787 LASSERT(parent != NULL);
788 LASSERT(tname != NULL);
789 LASSERT(handle != NULL);
791 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
792 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
793 if (buf->lb_buf == NULL)
797 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
798 rec->cr.cr_type = (__u32)type;
799 rec->cr.cr_tfid = *mdo2fid(target);
800 rec->cr.cr_pfid = *mdo2fid(parent);
801 rec->cr.cr_namelen = tname->ln_namelen;
802 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
804 target->mod_cltime = cfs_time_current_64();
806 rc = mdd_changelog_store(env, mdd, rec, handle);
808 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
809 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
810 PFID(&rec->cr.cr_pfid));
818 /** Store a namespace change changelog record
819 * If this fails, we must fail the whole transaction; we don't
820 * want the change to commit without the log entry.
821 * \param target - mdd_object of change
822 * \param tpfid - target parent dir/object fid
823 * \param sfid - source object fid
824 * \param spfid - source parent fid
825 * \param tname - target name string
826 * \param sname - source name string
827 * \param handle - transacion handle
829 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
830 struct mdd_device *mdd,
831 enum changelog_rec_type type,
833 struct mdd_object *target,
834 const struct lu_fid *tpfid,
835 const struct lu_fid *sfid,
836 const struct lu_fid *spfid,
837 const struct lu_name *tname,
838 const struct lu_name *sname,
839 struct thandle *handle)
841 struct llog_changelog_ext_rec *rec;
848 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
850 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
853 LASSERT(sfid != NULL);
854 LASSERT(tpfid != NULL);
855 LASSERT(tname != NULL);
856 LASSERT(handle != NULL);
858 reclen = llog_data_len(sizeof(*rec) +
859 sname != NULL ? 1 + sname->ln_namelen : 0);
860 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
861 if (buf->lb_buf == NULL)
865 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
866 rec->cr.cr_type = (__u32)type;
867 rec->cr.cr_pfid = *tpfid;
868 rec->cr.cr_sfid = *sfid;
869 rec->cr.cr_spfid = *spfid;
870 rec->cr.cr_namelen = tname->ln_namelen;
871 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
873 rec->cr.cr_name[tname->ln_namelen] = '\0';
874 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
876 rec->cr.cr_namelen += 1 + sname->ln_namelen;
879 if (likely(target != NULL)) {
880 rec->cr.cr_tfid = *mdo2fid(target);
881 target->mod_cltime = cfs_time_current_64();
883 fid_zero(&rec->cr.cr_tfid);
886 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
888 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
889 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
896 static int __mdd_links_add(const struct lu_env *env,
897 struct mdd_object *mdd_obj,
898 struct linkea_data *ldata,
899 const struct lu_name *lname,
900 const struct lu_fid *pfid,
901 int first, int check)
905 if (ldata->ld_leh == NULL) {
906 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
910 rc = linkea_data_new(ldata,
911 &mdd_env_info(env)->mti_link_buf);
918 rc = linkea_links_find(ldata, lname, pfid);
919 if (rc && rc != -ENOENT)
925 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
926 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
930 linkea_add_buf(ldata, lname, tfid);
933 return linkea_add_buf(ldata, lname, pfid);
936 static int __mdd_links_del(const struct lu_env *env,
937 struct mdd_object *mdd_obj,
938 struct linkea_data *ldata,
939 const struct lu_name *lname,
940 const struct lu_fid *pfid)
944 if (ldata->ld_leh == NULL) {
945 rc = mdd_links_read(env, mdd_obj, ldata);
950 rc = linkea_links_find(ldata, lname, pfid);
954 linkea_del_buf(ldata, lname);
958 static int mdd_linkea_prepare(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 int first, int check,
965 struct linkea_data *ldata)
971 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
974 LASSERT(oldpfid != NULL || newpfid != NULL);
976 if (mdd_obj->mod_flags & DEAD_OBJ)
977 /* No more links, don't bother */
980 if (oldpfid != NULL) {
981 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
984 (rc != -ENODATA && rc != -ENOENT))
986 /* No changes done. */
991 /* If renaming, add the new record */
992 if (newpfid != NULL) {
993 /* even if the add fails, we still delete the out-of-date
995 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1001 rc = rc != 0 ? rc : rc2;
1006 int mdd_links_rename(const struct lu_env *env,
1007 struct mdd_object *mdd_obj,
1008 const struct lu_fid *oldpfid,
1009 const struct lu_name *oldlname,
1010 const struct lu_fid *newpfid,
1011 const struct lu_name *newlname,
1012 struct thandle *handle,
1013 struct linkea_data *ldata,
1014 int first, int check)
1020 if (ldata == NULL) {
1021 ldata = &mdd_env_info(env)->mti_link_data;
1022 memset(ldata, 0, sizeof(*ldata));
1023 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1024 newpfid, newlname, first, check,
1030 if (ldata->ld_lee != NULL)
1031 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1038 if (rc == -EOVERFLOW || rc == -ENOENT)
1040 if (oldpfid == NULL)
1041 CDEBUG(error ? D_ERROR : D_OTHER,
1042 "link_ea add '%.*s' failed %d "DFID"\n",
1043 newlname->ln_namelen, newlname->ln_name,
1044 rc, PFID(mdd_object_fid(mdd_obj)));
1045 else if (newpfid == NULL)
1046 CDEBUG(error ? D_ERROR : D_OTHER,
1047 "link_ea del '%.*s' failed %d "DFID"\n",
1048 oldlname->ln_namelen, oldlname->ln_name,
1049 rc, PFID(mdd_object_fid(mdd_obj)));
1051 CDEBUG(error ? D_ERROR : D_OTHER,
1052 "link_ea rename '%.*s'->'%.*s' failed %d "
1054 oldlname->ln_namelen, oldlname->ln_name,
1055 newlname->ln_namelen, newlname->ln_name,
1056 rc, PFID(mdd_object_fid(mdd_obj)));
1059 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1060 /* if we vmalloced a large buffer drop it */
1061 lu_buf_free(ldata->ld_buf);
1066 static inline int mdd_links_add(const struct lu_env *env,
1067 struct mdd_object *mdd_obj,
1068 const struct lu_fid *pfid,
1069 const struct lu_name *lname,
1070 struct thandle *handle,
1071 struct linkea_data *data, int first)
1073 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1074 pfid, lname, handle, data, first, 0);
1077 static inline int mdd_links_del(const struct lu_env *env,
1078 struct mdd_object *mdd_obj,
1079 const struct lu_fid *pfid,
1080 const struct lu_name *lname,
1081 struct thandle *handle)
1083 return mdd_links_rename(env, mdd_obj, pfid, lname,
1084 NULL, NULL, handle, NULL, 0, 0);
1087 /** Read the link EA into a temp buffer.
1088 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1089 * A pointer to the buffer is stored in \a ldata::ld_buf.
1091 * \retval 0 or error
1093 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1094 struct linkea_data *ldata)
1098 /* First try a small buf */
1099 LASSERT(env != NULL);
1100 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1102 if (ldata->ld_buf->lb_buf == NULL)
1105 if (!mdd_object_exists(mdd_obj))
1108 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1110 if (rc == -ERANGE) {
1111 /* Buf was too small, figure out what we need. */
1112 lu_buf_free(ldata->ld_buf);
1113 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1114 XATTR_NAME_LINK, BYPASS_CAPA);
1117 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1118 if (ldata->ld_buf->lb_buf == NULL)
1120 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1121 XATTR_NAME_LINK, BYPASS_CAPA);
1130 /** Read the link EA into a temp buffer.
1131 * Uses the name_buf since it is generally large.
1132 * \retval IS_ERR err
1133 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1135 struct lu_buf *mdd_links_get(const struct lu_env *env,
1136 struct mdd_object *mdd_obj)
1138 struct linkea_data ldata = { 0 };
1141 rc = mdd_links_read(env, mdd_obj, &ldata);
1142 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1145 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1146 struct linkea_data *ldata, struct thandle *handle)
1148 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1149 ldata->ld_leh->leh_len);
1150 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1151 mdd_object_capa(env, mdd_obj));
1154 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1155 struct thandle *handle, struct linkea_data *ldata)
1161 if (ldata != NULL && ldata->ld_lee != NULL) {
1162 ea_len = ldata->ld_leh->leh_len;
1163 linkea = ldata->ld_buf->lb_buf;
1169 /* XXX: max size? */
1170 rc = mdo_declare_xattr_set(env, mdd_obj,
1171 mdd_buf_get_const(env, linkea, ea_len),
1172 XATTR_NAME_LINK, 0, handle);
1176 static inline int mdd_declare_links_del(const struct lu_env *env,
1177 struct mdd_object *c,
1178 struct thandle *handle)
1182 /* For directory, the linkEA will be removed together
1183 * with the object. */
1184 if (!S_ISDIR(mdd_object_type(c)))
1185 rc = mdd_declare_links_add(env, c, handle, NULL);
1190 static int mdd_declare_link(const struct lu_env *env,
1191 struct mdd_device *mdd,
1192 struct mdd_object *p,
1193 struct mdd_object *c,
1194 const struct lu_name *name,
1195 struct thandle *handle,
1197 struct linkea_data *data)
1201 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
1205 rc = mdo_declare_ref_add(env, c, handle);
1209 la->la_valid = LA_CTIME | LA_MTIME;
1210 rc = mdo_declare_attr_set(env, p, la, handle);
1214 la->la_valid = LA_CTIME;
1215 rc = mdo_declare_attr_set(env, c, la, handle);
1219 rc = mdd_declare_links_add(env, c, handle, data);
1223 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1228 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1229 struct md_object *src_obj, const struct lu_name *lname,
1232 const char *name = lname->ln_name;
1233 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1234 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1235 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1236 struct mdd_device *mdd = mdo2mdd(src_obj);
1237 struct dynlock_handle *dlh;
1238 struct thandle *handle;
1239 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1243 handle = mdd_trans_create(env, mdd);
1245 GOTO(out_pending, rc = PTR_ERR(handle));
1247 memset(ldata, 0, sizeof(*ldata));
1249 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1250 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1252 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1257 rc = mdd_trans_start(env, mdd, handle);
1261 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
1263 GOTO(out_trans, rc = -ENOMEM);
1264 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1266 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
1268 GOTO(out_unlock, rc);
1270 rc = mdo_ref_add(env, mdd_sobj, handle);
1272 GOTO(out_unlock, rc);
1275 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1277 mdd_object_capa(env, mdd_tobj));
1279 mdo_ref_del(env, mdd_sobj, handle);
1280 GOTO(out_unlock, rc);
1283 la->la_valid = LA_CTIME | LA_MTIME;
1284 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1286 GOTO(out_unlock, rc);
1288 la->la_valid = LA_CTIME;
1289 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
1291 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1292 mdo2fid(mdd_tobj), lname, 0, 0,
1295 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1296 lname, handle, ldata, 0);
1297 /* The failure of links_add should not cause the link
1298 * failure, reset rc here */
1303 mdd_write_unlock(env, mdd_sobj);
1304 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
1307 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1308 mdd_tobj, lname, handle);
1310 mdd_trans_stop(env, mdd, rc, handle);
1312 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1313 /* if we vmalloced a large buffer drop it */
1314 lu_buf_free(ldata->ld_buf);
1319 int mdd_declare_finish_unlink(const struct lu_env *env,
1320 struct mdd_object *obj,
1322 struct thandle *handle)
1326 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1330 return mdo_declare_destroy(env, obj, handle);
1333 /* caller should take a lock before calling */
1334 int mdd_finish_unlink(const struct lu_env *env,
1335 struct mdd_object *obj, struct md_attr *ma,
1339 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1342 LASSERT(mdd_write_locked(env, obj) != 0);
1344 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
1345 obj->mod_flags |= DEAD_OBJ;
1346 /* add new orphan and the object
1347 * will be deleted during mdd_close() */
1348 if (obj->mod_count) {
1349 rc = __mdd_orphan_add(env, obj, th);
1351 CDEBUG(D_HA, "Object "DFID" is inserted into "
1352 "orphan list, open count = %d\n",
1353 PFID(mdd_object_fid(obj)),
1356 CERROR("Object "DFID" fail to be an orphan, "
1357 "open count = %d, maybe cause failed "
1359 PFID(mdd_object_fid(obj)),
1362 rc = mdo_destroy(env, obj, th);
1371 * has mdd_write_lock on cobj already, but not on pobj yet
1373 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1374 struct mdd_object *cobj, struct lu_attr *cattr)
1379 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1384 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1385 struct mdd_object *p, struct mdd_object *c,
1386 const struct lu_name *name, struct md_attr *ma,
1387 struct thandle *handle, int no_name)
1389 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1392 if (likely(no_name == 0)) {
1393 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1398 rc = mdo_declare_ref_del(env, p, handle);
1402 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1403 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1404 la->la_valid = LA_CTIME | LA_MTIME;
1405 rc = mdo_declare_attr_set(env, p, la, handle);
1410 rc = mdo_declare_ref_del(env, c, handle);
1414 rc = mdo_declare_ref_del(env, c, handle);
1418 la->la_valid = LA_CTIME;
1419 rc = mdo_declare_attr_set(env, c, la, handle);
1423 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1427 rc = mdd_declare_links_del(env, c, handle);
1431 /* FIXME: need changelog for remove entry */
1432 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1439 * Delete name entry and the object.
1440 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1441 * does not exist for this object, and it could only happen during resending
1442 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1443 * is also needed in this case(needed by changelog), so we have to add another
1444 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1445 * the ENOENT failure should be able to be fixed by redo mechanism.
1447 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1448 struct md_object *cobj, const struct lu_name *lname,
1449 struct md_attr *ma, int no_name)
1451 const char *name = lname->ln_name;
1452 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1453 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1454 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1455 struct mdd_object *mdd_cobj = NULL;
1456 struct mdd_device *mdd = mdo2mdd(pobj);
1457 struct dynlock_handle *dlh;
1458 struct thandle *handle;
1462 /* cobj == NULL means only delete name entry */
1463 if (likely(cobj != NULL)) {
1464 mdd_cobj = md2mdd_obj(cobj);
1465 if (mdd_object_exists(mdd_cobj) == 0)
1467 /* currently it is assume, it could only delete
1468 * name entry of remote directory */
1472 handle = mdd_trans_create(env, mdd);
1474 RETURN(PTR_ERR(handle));
1476 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1477 lname, ma, handle, no_name);
1481 rc = mdd_trans_start(env, mdd, handle);
1485 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1487 GOTO(stop, rc = -ENOMEM);
1489 if (likely(mdd_cobj != NULL)) {
1490 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1493 rc = mdd_la_get(env, mdd_cobj, cattr,
1494 mdd_object_capa(env, mdd_cobj));
1498 is_dir = S_ISDIR(cattr->la_mode);
1502 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1506 if (likely(no_name == 0)) {
1507 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1508 mdd_object_capa(env, mdd_pobj));
1513 if (likely(mdd_cobj != NULL)) {
1514 rc = mdo_ref_del(env, mdd_cobj, handle);
1516 __mdd_index_insert_only(env, mdd_pobj,
1519 mdd_object_capa(env, mdd_pobj));
1525 mdo_ref_del(env, mdd_cobj, handle);
1527 /* fetch updated nlink */
1528 rc = mdd_la_get(env, mdd_cobj, cattr,
1529 mdd_object_capa(env, mdd_cobj));
1534 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1535 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1537 la->la_valid = LA_CTIME | LA_MTIME;
1538 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1542 /* Enough for only unlink the entry */
1543 if (unlikely(mdd_cobj == NULL)) {
1544 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1548 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1549 /* update ctime of an unlinked file only if it is still
1550 * opened or a link still exists */
1551 la->la_valid = LA_CTIME;
1552 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1557 /* XXX: this transfer to ma will be removed with LOD/OSP */
1558 ma->ma_attr = *cattr;
1559 ma->ma_valid |= MA_INODE;
1560 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1562 /* fetch updated nlink */
1564 rc = mdd_la_get(env, mdd_cobj, cattr,
1565 mdd_object_capa(env, mdd_cobj));
1568 /* old files may not have link ea; ignore errors */
1569 mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle);
1571 /* if object is removed then we can't get its attrs, use last get */
1572 if (cattr->la_nlink == 0) {
1573 ma->ma_attr = *cattr;
1574 ma->ma_valid |= MA_INODE;
1578 mdd_write_unlock(env, mdd_cobj);
1579 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1583 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1584 if ((ma->ma_valid & MA_HSM) &&
1585 (ma->ma_hsm.mh_flags & HS_EXISTS))
1586 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1588 rc = mdd_changelog_ns_store(env, mdd,
1589 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1590 mdd_cobj, mdd_pobj, lname, handle);
1594 mdd_trans_stop(env, mdd, rc, handle);
1600 * The permission has been checked when obj created, no need check again.
1602 static int mdd_cd_sanity_check(const struct lu_env *env,
1603 struct mdd_object *obj)
1608 if (!obj || mdd_is_dead_obj(obj))
1615 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1616 struct md_object *cobj, const struct md_op_spec *spec,
1619 struct mdd_device *mdd = mdo2mdd(cobj);
1620 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1621 struct mdd_object *son = md2mdd_obj(cobj);
1622 struct thandle *handle;
1623 const struct lu_buf *buf;
1624 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1628 /* do not let users to create stripes via .lustre/
1629 * mdd_obf_setup() sets IMMUTE_OBJ on this directory */
1630 if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ)
1633 rc = mdd_cd_sanity_check(env, son);
1637 if (!md_should_create(spec->sp_cr_flags))
1641 * there are following use cases for this function:
1642 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1643 * striping can be specified or not
1646 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1650 /* calling ->ah_make_hint() is used to transfer information from parent */
1651 mdd_object_make_hint(env, mdd_pobj, son, attr);
1653 handle = mdd_trans_create(env, mdd);
1655 GOTO(out_free, rc = PTR_ERR(handle));
1658 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1659 * Should this be fixed?
1661 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, no_create %u\n",
1662 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1663 spec->sp_cr_flags, spec->no_create);
1665 if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1666 /* replay case or lfs setstripe */
1667 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1668 spec->u.sp_ea.eadatalen);
1673 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1674 XATTR_NAME_LOV, 0, handle);
1678 rc = mdd_trans_start(env, mdd, handle);
1682 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1683 0, handle, mdd_object_capa(env, son));
1685 mdd_trans_stop(env, mdd, rc, handle);
1690 /* Get fid from name and parent */
1692 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1693 const struct lu_name *lname, struct lu_fid* fid, int mask)
1695 const char *name = lname->ln_name;
1696 const struct dt_key *key = (const struct dt_key *)name;
1697 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1698 struct mdd_device *m = mdo2mdd(pobj);
1699 struct dt_object *dir = mdd_object_child(mdd_obj);
1703 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1706 if (mdd_object_remote(mdd_obj)) {
1707 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
1708 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
1709 } else if (!mdd_object_exists(mdd_obj)) {
1713 /* The common filename length check. */
1714 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1715 RETURN(-ENAMETOOLONG);
1717 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1722 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1723 dt_try_as_dir(env, dir))) {
1725 rc = dir->do_index_ops->dio_lookup(env, dir,
1726 (struct dt_rec *)fid, key,
1727 mdd_object_capa(env, mdd_obj));
1738 static int mdd_declare_object_initialize(const struct lu_env *env,
1739 struct mdd_object *parent,
1740 struct mdd_object *child,
1741 struct lu_attr *attr,
1742 struct thandle *handle,
1743 struct linkea_data *ldata)
1749 * inode mode has been set in creation time, and it's based on umask,
1750 * la_mode and acl, don't set here again! (which will go wrong
1751 * because below function doesn't consider umask).
1752 * I'd suggest set all object attributes in creation time, see above.
1754 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1755 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1756 rc = mdo_declare_attr_set(env, child, attr, handle);
1757 attr->la_valid |= LA_MODE | LA_TYPE;
1758 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1759 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1762 rc = mdo_declare_ref_add(env, child, handle);
1764 rc = mdo_declare_index_insert(env, child, mdo2fid(parent),
1768 if (rc == 0 && (fid_is_norm(mdo2fid(child)) ||
1769 fid_is_dot_lustre(mdo2fid(child)) ||
1770 fid_is_root(mdo2fid(child))))
1771 mdd_declare_links_add(env, child, handle, ldata);
1776 static int mdd_object_initialize(const struct lu_env *env,
1777 const struct lu_fid *pfid,
1778 const struct lu_name *lname,
1779 struct mdd_object *child,
1780 struct lu_attr *attr, struct thandle *handle,
1781 const struct md_op_spec *spec,
1782 struct linkea_data *ldata)
1788 * Update attributes for child.
1791 * (1) the valid bits should be converted between Lustre and Linux;
1792 * (2) maybe, the child attributes should be set in OSD when creation.
1795 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1796 /* arguments are supposed to stay the same */
1797 if (S_ISDIR(attr->la_mode)) {
1798 /* Add "." and ".." for newly created dir */
1799 mdo_ref_add(env, child, handle);
1800 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1801 dot, handle, BYPASS_CAPA);
1803 rc = __mdd_index_insert_only(env, child, pfid,
1807 mdo_ref_del(env, child, handle);
1810 if (rc == 0 && (fid_is_norm(mdo2fid(child)) ||
1811 fid_is_dot_lustre(mdo2fid(child)) ||
1812 fid_is_root(mdo2fid(child))))
1813 mdd_links_add(env, child, pfid, lname, handle, ldata, 1);
1818 /* has not lock on pobj yet */
1819 static int mdd_create_sanity_check(const struct lu_env *env,
1820 struct md_object *pobj,
1821 struct lu_attr *pattr,
1822 const struct lu_name *lname,
1823 struct lu_attr *cattr,
1824 struct md_op_spec *spec)
1826 struct mdd_thread_info *info = mdd_env_info(env);
1827 struct lu_fid *fid = &info->mti_fid;
1828 struct mdd_object *obj = md2mdd_obj(pobj);
1829 struct mdd_device *m = mdo2mdd(pobj);
1834 if (mdd_is_dead_obj(obj))
1838 * In some cases this lookup is not needed - we know before if name
1839 * exists or not because MDT performs lookup for it.
1840 * name length check is done in lookup.
1842 if (spec->sp_cr_lookup) {
1844 * Check if the name already exist, though it will be checked in
1845 * _index_insert also, for avoiding rolling back if exists
1848 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1849 MAY_WRITE | MAY_EXEC);
1851 RETURN(rc ? : -EEXIST);
1854 * Check WRITE permission for the parent.
1855 * EXEC permission have been checked
1856 * when lookup before create already.
1858 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1865 if (pattr->la_mode & S_ISGID) {
1866 cattr->la_gid = pattr->la_gid;
1867 if (S_ISDIR(cattr->la_mode)) {
1868 cattr->la_mode |= S_ISGID;
1869 cattr->la_valid |= LA_MODE;
1873 switch (cattr->la_mode & S_IFMT) {
1875 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1877 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1878 RETURN(-ENAMETOOLONG);
1897 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1898 struct mdd_object *p, struct mdd_object *c,
1899 const struct lu_name *name,
1900 struct lu_attr *attr,
1902 struct thandle *handle,
1903 const struct md_op_spec *spec,
1904 struct linkea_data *ldata)
1908 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1912 #ifdef CONFIG_FS_POSIX_ACL
1913 if (got_def_acl > 0) {
1914 struct lu_buf *acl_buf;
1916 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1917 /* if dir, then can inherit default ACl */
1918 if (S_ISDIR(attr->la_mode)) {
1919 rc = mdo_declare_xattr_set(env, c, acl_buf,
1920 XATTR_NAME_ACL_DEFAULT,
1926 rc = mdo_declare_attr_set(env, c, attr, handle);
1930 rc = mdo_declare_xattr_set(env, c, acl_buf,
1931 XATTR_NAME_ACL_ACCESS, 0, handle);
1937 if (S_ISDIR(attr->la_mode)) {
1938 rc = mdo_declare_ref_add(env, p, handle);
1943 rc = mdd_declare_object_initialize(env, p, c, attr, handle, ldata);
1947 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1948 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
1950 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1951 name->ln_name, handle);
1955 /* replay case, create LOV EA from client data */
1956 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1957 const struct lu_buf *buf;
1959 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1960 spec->u.sp_ea.eadatalen);
1961 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1967 if (S_ISLNK(attr->la_mode)) {
1968 rc = dt_declare_record_write(env, mdd_object_child(c),
1969 strlen(spec->u.sp_symname), 0,
1975 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
1976 rc = mdo_declare_attr_set(env, p, attr, handle);
1981 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1989 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
1990 struct lu_attr *la, struct lu_buf *acl_buf,
1991 int *got_def_acl, int *reset_acl)
1996 if (S_ISLNK(la->la_mode))
1999 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2000 rc = mdo_xattr_get(env, pobj, acl_buf,
2001 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
2002 mdd_read_unlock(env, pobj);
2004 /* If there are default ACL, fix mode by default ACL */
2006 acl_buf->lb_len = rc;
2007 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2011 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2012 /* If there are no default ACL, fix mode by mask */
2013 struct lu_ucred *uc = lu_ucred(env);
2015 /* The create triggered by MDT internal events, such as
2016 * LFSCK reset, will not contain valid "uc". */
2017 if (unlikely(uc != NULL))
2018 la->la_mode &= ~uc->uc_umask;
2026 * Create object and insert it into namespace.
2028 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2029 const struct lu_name *lname, struct md_object *child,
2030 struct md_op_spec *spec, struct md_attr* ma)
2032 struct mdd_thread_info *info = mdd_env_info(env);
2033 struct lu_attr *la = &info->mti_la_for_fix;
2034 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2035 struct mdd_object *son = md2mdd_obj(child);
2036 struct mdd_device *mdd = mdo2mdd(pobj);
2037 struct lu_attr *attr = &ma->ma_attr;
2038 struct thandle *handle;
2039 struct lu_attr *pattr = &info->mti_pattr;
2040 struct lu_buf acl_buf;
2041 struct linkea_data *ldata = &info->mti_link_data;
2042 struct dynlock_handle *dlh;
2043 const char *name = lname->ln_name;
2044 int rc, created = 0, initialized = 0, inserted = 0;
2045 int got_def_acl = 0;
2050 * Two operations have to be performed:
2052 * - an allocation of a new object (->do_create()), and
2054 * - an insertion into a parent index (->dio_insert()).
2056 * Due to locking, operation order is not important, when both are
2057 * successful, *but* error handling cases are quite different:
2059 * - if insertion is done first, and following object creation fails,
2060 * insertion has to be rolled back, but this operation might fail
2061 * also leaving us with dangling index entry.
2063 * - if creation is done first, is has to be undone if insertion
2064 * fails, leaving us with leaked space, which is neither good, nor
2067 * It seems that creation-first is simplest solution, but it is
2068 * sub-optimal in the frequent
2073 * case, because second mkdir is bound to create object, only to
2074 * destroy it immediately.
2076 * To avoid this follow local file systems that do double lookup:
2078 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2080 * 1. create (mdd_object_create_internal())
2082 * 2. insert (__mdd_index_insert(), lookup again)
2085 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2089 /* Sanity checks before big job. */
2090 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2094 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2095 GOTO(out_free, rc = -EINPROGRESS);
2097 acl_buf.lb_buf = info->mti_xattr_buf;
2098 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2099 rc = mdd_acl_init(env, mdd_pobj, attr, &acl_buf, &got_def_acl,
2104 mdd_object_make_hint(env, mdd_pobj, son, attr);
2106 handle = mdd_trans_create(env, mdd);
2108 GOTO(out_free, rc = PTR_ERR(handle));
2110 memset(ldata, 0, sizeof(*ldata));
2111 mdd_linkea_prepare(env, son, NULL, NULL, mdd_object_fid(mdd_pobj),
2112 lname, 1, 0, ldata);
2113 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2114 got_def_acl, handle, spec, ldata);
2118 rc = mdd_trans_start(env, mdd, handle);
2122 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
2124 GOTO(out_trans, rc = -ENOMEM);
2126 mdd_write_lock(env, son, MOR_TGT_CHILD);
2127 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec);
2129 mdd_write_unlock(env, son);
2135 #ifdef CONFIG_FS_POSIX_ACL
2137 /* set default acl */
2138 if (S_ISDIR(attr->la_mode)) {
2139 LASSERTF(acl_buf.lb_len == got_def_acl,
2140 "invalid acl_buf: %p:%d got_def %d\n",
2141 acl_buf.lb_buf, (int)acl_buf.lb_len,
2143 rc = mdo_xattr_set(env, son, &acl_buf,
2144 XATTR_NAME_ACL_DEFAULT, 0,
2145 handle, BYPASS_CAPA);
2147 mdd_write_unlock(env, son);
2152 /* set its own acl */
2154 LASSERTF(acl_buf.lb_buf != NULL && acl_buf.lb_len != 0,
2155 "invalid acl_buf %p:%d\n", acl_buf.lb_buf,
2156 (int)acl_buf.lb_len);
2157 rc = mdo_xattr_set(env, son, &acl_buf,
2158 XATTR_NAME_ACL_ACCESS,
2159 0, handle, BYPASS_CAPA);
2161 mdd_write_unlock(env, son);
2168 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
2169 son, attr, handle, spec, ldata);
2172 * in case of replay we just set LOVEA provided by the client
2173 * XXX: I think it would be interesting to try "old" way where
2174 * MDT calls this xattr_set(LOV) in a different transaction.
2175 * probably this way we code can be made better.
2177 if (rc == 0 && (spec->no_create ||
2178 (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
2179 const struct lu_buf *buf;
2181 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2182 spec->u.sp_ea.eadatalen);
2183 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
2187 if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2188 rc = __mdd_orphan_add(env, son, handle);
2190 mdd_write_unlock(env, son);
2194 * Object has no links, so it will be destroyed when last
2195 * reference is released. (XXX not now.)
2201 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE))
2202 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2203 name, S_ISDIR(attr->la_mode), handle,
2204 mdd_object_capa(env, mdd_pobj));
2211 if (S_ISLNK(attr->la_mode)) {
2212 struct lu_ucred *uc = lu_ucred_assert(env);
2213 struct dt_object *dt = mdd_object_child(son);
2214 const char *target_name = spec->u.sp_symname;
2215 int sym_len = strlen(target_name);
2216 const struct lu_buf *buf;
2219 buf = mdd_buf_get_const(env, target_name, sym_len);
2220 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2221 mdd_object_capa(env, son),
2223 CFS_CAP_SYS_RESOURCE_MASK);
2228 GOTO(cleanup, rc = -EFAULT);
2231 /* volatile file creation does not update parent directory times */
2232 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2233 GOTO(cleanup, rc = 0);
2235 /* update parent directory mtime/ctime */
2237 la->la_valid = LA_CTIME | LA_MTIME;
2238 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
2244 if (rc != 0 && created != 0) {
2247 if (inserted != 0) {
2248 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2249 rc2 = __mdd_orphan_del(env, son, handle);
2251 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2252 S_ISDIR(attr->la_mode),
2253 handle, BYPASS_CAPA);
2258 mdd_write_lock(env, son, MOR_TGT_CHILD);
2259 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
2260 /* Drop the reference, no need to delete "."/"..",
2261 * because the object to be destroied directly. */
2262 rc2 = mdo_ref_del(env, son, handle);
2264 mdd_write_unlock(env, son);
2269 rc2 = mdo_ref_del(env, son, handle);
2271 mdd_write_unlock(env, son);
2275 mdo_destroy(env, son, handle);
2276 mdd_write_unlock(env, son);
2279 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
2281 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)))
2282 rc = mdd_changelog_ns_store(env, mdd,
2283 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2284 S_ISREG(attr->la_mode) ? CL_CREATE :
2285 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2286 0, son, mdd_pobj, lname, handle);
2288 mdd_trans_stop(env, mdd, rc, handle);
2290 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2291 /* if we vmalloced a large buffer drop it */
2292 lu_buf_free(ldata->ld_buf);
2294 /* The child object shouldn't be cached anymore */
2296 set_bit(LU_OBJECT_HEARD_BANSHEE,
2297 &child->mo_lu.lo_header->loh_flags);
2302 * Get locks on parents in proper order
2303 * RETURN: < 0 - error, rename_order if successful
2311 static int mdd_rename_order(const struct lu_env *env,
2312 struct mdd_device *mdd,
2313 struct mdd_object *src_pobj,
2314 struct mdd_object *tgt_pobj)
2316 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2320 if (src_pobj == tgt_pobj)
2321 RETURN(MDD_RN_SAME);
2323 /* compared the parent child relationship of src_p&tgt_p */
2324 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2326 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2329 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
2342 /* has not mdd_write{read}_lock on any obj yet. */
2343 static int mdd_rename_sanity_check(const struct lu_env *env,
2344 struct mdd_object *src_pobj,
2345 struct mdd_object *tgt_pobj,
2346 struct mdd_object *sobj,
2347 struct mdd_object *tobj,
2348 struct lu_attr *so_attr,
2349 struct lu_attr *tg_attr)
2354 /* XXX: when get here, sobj must NOT be NULL,
2355 * the other case has been processed in cld_rename
2356 * before mdd_rename and enable MDS_PERM_BYPASS. */
2359 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
2363 /* XXX: when get here, "tobj == NULL" means tobj must
2364 * NOT exist (neither on remote MDS, such case has been
2365 * processed in cld_rename before mdd_rename and enable
2367 * So check may_create, but not check may_unlink. */
2369 rc = mdd_may_create(env, tgt_pobj, NULL,
2370 (src_pobj != tgt_pobj), 0);
2372 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
2373 (src_pobj != tgt_pobj), 1);
2375 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2376 S_ISDIR(so_attr->la_mode))
2377 rc = __mdd_may_link(env, tgt_pobj);
2382 static int mdd_declare_rename(const struct lu_env *env,
2383 struct mdd_device *mdd,
2384 struct mdd_object *mdd_spobj,
2385 struct mdd_object *mdd_tpobj,
2386 struct mdd_object *mdd_sobj,
2387 struct mdd_object *mdd_tobj,
2388 const struct lu_name *tname,
2389 const struct lu_name *sname,
2391 struct linkea_data *ldata,
2392 struct thandle *handle)
2394 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2397 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2398 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2404 /* name from source dir */
2405 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2409 /* .. from source child */
2410 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2411 /* source child can be directory,
2412 * counted by source dir's nlink */
2413 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2416 if (mdd_spobj != mdd_tpobj) {
2417 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2422 rc = mdo_declare_index_insert(env, mdd_sobj,
2428 /* new target child can be directory,
2429 * counted by target dir's nlink */
2430 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2436 la->la_valid = LA_CTIME | LA_MTIME;
2437 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2441 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2445 la->la_valid = LA_CTIME;
2446 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2450 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2455 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2456 tname->ln_name, handle);
2460 /* name from target dir (old name), we declare it unconditionally
2461 * as mdd_rename() calls delete unconditionally as well. so just
2462 * to balance declarations vs calls to change ... */
2463 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2467 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2468 /* delete target child in target parent directory */
2469 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2473 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2474 /* target child can be directory,
2475 * delete "." reference in target child directory */
2476 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2480 /* delete ".." reference in target parent directory */
2481 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2486 la->la_valid = LA_CTIME;
2487 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2491 mdd_declare_links_del(env, mdd_tobj, handle);
2495 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2500 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2507 /* src object can be remote that is why we use only fid and type of object */
2508 static int mdd_rename(const struct lu_env *env,
2509 struct md_object *src_pobj, struct md_object *tgt_pobj,
2510 const struct lu_fid *lf, const struct lu_name *lsname,
2511 struct md_object *tobj, const struct lu_name *ltname,
2514 const char *sname = lsname->ln_name;
2515 const char *tname = ltname->ln_name;
2516 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2517 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2518 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2519 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2520 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2521 struct mdd_device *mdd = mdo2mdd(src_pobj);
2522 struct mdd_object *mdd_sobj = NULL; /* source object */
2523 struct mdd_object *mdd_tobj = NULL;
2524 struct dynlock_handle *sdlh = NULL, *tdlh = NULL;
2525 struct thandle *handle;
2526 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2527 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2528 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2531 bool tobj_locked = 0;
2532 unsigned cl_flags = 0;
2537 mdd_tobj = md2mdd_obj(tobj);
2539 mdd_sobj = mdd_object_find(env, mdd, lf);
2541 rc = mdd_la_get(env, mdd_sobj, so_attr,
2542 mdd_object_capa(env, mdd_sobj));
2544 GOTO(out_pending, rc);
2547 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2548 mdd_object_capa(env, mdd_tobj));
2550 GOTO(out_pending, rc);
2553 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2554 mdd_tobj, so_attr, tg_attr);
2556 GOTO(out_pending, rc);
2558 handle = mdd_trans_create(env, mdd);
2560 GOTO(out_pending, rc = PTR_ERR(handle));
2562 memset(ldata, 0, sizeof(*ldata));
2563 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2564 ltname, 1, 0, ldata);
2565 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2566 mdd_tobj, lsname, ltname, ma, ldata, handle);
2570 rc = mdd_trans_start(env, mdd, handle);
2574 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2575 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2577 GOTO(cleanup_unlocked, rc);
2579 /* Get locks in determined order */
2580 if (rc == MDD_RN_SAME) {
2581 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2582 sname, MOR_SRC_PARENT);
2583 /* check hashes to determine do we need one lock or two */
2584 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2585 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2589 } else if (rc == MDD_RN_SRCTGT) {
2590 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2591 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2593 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2594 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2596 if (sdlh == NULL || tdlh == NULL)
2597 GOTO(cleanup, rc = -ENOMEM);
2599 is_dir = S_ISDIR(so_attr->la_mode);
2601 /* Remove source name from source directory */
2602 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2603 mdd_object_capa(env, mdd_spobj));
2607 /* "mv dir1 dir2" needs "dir1/.." link update */
2608 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2609 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2610 mdd_object_capa(env, mdd_sobj));
2612 GOTO(fixup_spobj2, rc);
2614 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2615 handle, mdd_object_capa(env, mdd_sobj));
2617 GOTO(fixup_spobj, rc);
2620 /* Remove target name from target directory
2621 * Here tobj can be remote one, so we do index_delete unconditionally
2622 * and -ENOENT is allowed.
2624 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2625 mdd_object_capa(env, mdd_tpobj));
2628 /* tname might been renamed to something else */
2629 GOTO(fixup_spobj, rc);
2632 GOTO(fixup_spobj, rc);
2635 /* Insert new fid with target name into target dir */
2636 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2637 mdd_object_capa(env, mdd_tpobj));
2639 GOTO(fixup_tpobj, rc);
2641 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2642 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2644 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2646 la->la_valid = LA_CTIME;
2647 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2649 GOTO(fixup_tpobj, rc);
2652 /* Remove old target object
2653 * For tobj is remote case cmm layer has processed
2654 * and set tobj to NULL then. So when tobj is NOT NULL,
2655 * it must be local one.
2657 if (tobj && mdd_object_exists(mdd_tobj)) {
2658 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2660 if (mdd_is_dead_obj(mdd_tobj)) {
2661 /* shld not be dead, something is wrong */
2662 CERROR("tobj is dead, something is wrong\n");
2666 mdo_ref_del(env, mdd_tobj, handle);
2668 /* Remove dot reference. */
2669 if (S_ISDIR(tg_attr->la_mode))
2670 mdo_ref_del(env, mdd_tobj, handle);
2673 /* fetch updated nlink */
2674 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2675 mdd_object_capa(env, mdd_tobj));
2677 CERROR("%s: Failed to get nlink for tobj "
2679 mdd2obd_dev(mdd)->obd_name,
2680 PFID(tpobj_fid), rc);
2681 GOTO(fixup_tpobj, rc);
2684 la->la_valid = LA_CTIME;
2685 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2687 CERROR("%s: Failed to set ctime for tobj "
2689 mdd2obd_dev(mdd)->obd_name,
2690 PFID(tpobj_fid), rc);
2691 GOTO(fixup_tpobj, rc);
2694 /* XXX: this transfer to ma will be removed with LOD/OSP */
2695 ma->ma_attr = *tg_attr;
2696 ma->ma_valid |= MA_INODE;
2697 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2699 CERROR("%s: Failed to unlink tobj "
2701 mdd2obd_dev(mdd)->obd_name,
2702 PFID(tpobj_fid), rc);
2703 GOTO(fixup_tpobj, rc);
2706 /* fetch updated nlink */
2707 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2708 mdd_object_capa(env, mdd_tobj));
2710 CERROR("%s: Failed to get nlink for tobj "
2712 mdd2obd_dev(mdd)->obd_name,
2713 PFID(tpobj_fid), rc);
2714 GOTO(fixup_tpobj, rc);
2716 /* XXX: this transfer to ma will be removed with LOD/OSP */
2717 ma->ma_attr = *tg_attr;
2718 ma->ma_valid |= MA_INODE;
2720 if (so_attr->la_nlink == 0)
2721 cl_flags |= CLF_RENAME_LAST;
2724 la->la_valid = LA_CTIME | LA_MTIME;
2725 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2727 GOTO(fixup_tpobj, rc);
2729 if (mdd_spobj != mdd_tpobj) {
2730 la->la_valid = LA_CTIME | LA_MTIME;
2731 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2735 if (rc == 0 && mdd_sobj) {
2736 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2737 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2738 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2741 /* Old files might not have EA entry */
2742 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2743 lsname, handle, NULL, 0);
2744 mdd_write_unlock(env, mdd_sobj);
2745 /* We don't fail the transaction if the link ea can't be
2746 updated -- fid2path will use alternate lookup method. */
2754 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2757 CWARN("tp obj fix error %d\n",rc2);
2759 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2760 !mdd_is_dead_obj(mdd_tobj)) {
2762 mdo_ref_add(env, mdd_tobj, handle);
2764 mdo_ref_add(env, mdd_tobj, handle);
2767 rc2 = __mdd_index_insert(env, mdd_tpobj,
2768 mdo2fid(mdd_tobj), tname,
2773 CWARN("tp obj fix error %d\n",rc2);
2778 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2779 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2783 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2784 mdd2obd_dev(mdd)->obd_name, rc2);
2787 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2788 dotdot, handle, BYPASS_CAPA);
2790 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2791 mdd2obd_dev(mdd)->obd_name, rc2);
2796 rc2 = __mdd_index_insert(env, mdd_spobj,
2797 lf, sname, is_dir, handle, BYPASS_CAPA);
2799 CWARN("sp obj fix error %d\n",rc2);
2803 mdd_write_unlock(env, mdd_tobj);
2804 if (likely(tdlh) && sdlh != tdlh)
2805 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2807 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2810 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2811 mdd_tobj, tpobj_fid, lf,
2812 spobj_fid, ltname, lsname,
2816 mdd_trans_stop(env, mdd, rc, handle);
2818 mdd_object_put(env, mdd_sobj);
2822 const struct md_dir_operations mdd_dir_ops = {
2823 .mdo_is_subdir = mdd_is_subdir,
2824 .mdo_lookup = mdd_lookup,
2825 .mdo_create = mdd_create,
2826 .mdo_rename = mdd_rename,
2827 .mdo_link = mdd_link,
2828 .mdo_unlink = mdd_unlink,
2829 .mdo_create_data = mdd_create_data,