4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_dir.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <obd_class.h>
46 #include <obd_support.h>
47 #include <lustre_mds.h>
48 #include <lustre_fid.h>
50 #include "mdd_internal.h"
52 static const char dot[] = ".";
53 static const char dotdot[] = "..";
55 static struct lu_name lname_dotdot = {
60 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
61 const struct lu_name *lname, struct lu_fid* fid,
63 static int mdd_declare_links_add(const struct lu_env *env,
64 struct mdd_object *mdd_obj,
65 struct thandle *handle);
66 static int mdd_links_add(const struct lu_env *env,
67 struct mdd_object *mdd_obj,
68 const struct lu_fid *pfid,
69 const struct lu_name *lname,
70 struct thandle *handle, int first);
71 static int mdd_links_rename(const struct lu_env *env,
72 struct mdd_object *mdd_obj,
73 const struct lu_fid *oldpfid,
74 const struct lu_name *oldlname,
75 const struct lu_fid *newpfid,
76 const struct lu_name *newlname,
77 struct thandle *handle);
80 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
81 const struct lu_name *lname, struct lu_fid* fid, int mask)
83 const char *name = lname->ln_name;
84 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
85 struct dynlock_handle *dlh;
88 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
89 if (unlikely(dlh == NULL))
91 rc = __mdd_lookup(env, pobj, lname, fid, mask);
92 mdd_pdo_read_unlock(env, mdd_obj, dlh);
97 int mdd_lookup(const struct lu_env *env,
98 struct md_object *pobj, const struct lu_name *lname,
99 struct lu_fid* fid, struct md_op_spec *spec)
103 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
107 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
110 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
114 * For root fid use special function, which does not compare version component
115 * of fid. Version component is different for root fids on all MDTs.
117 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
119 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
120 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
124 * return 1: if lf is the fid of the ancestor of p1;
127 * return -EREMOTE: if remote object is found, in this
128 * case fid of remote object is saved to @pf;
130 * otherwise: values < 0, errors.
132 static int mdd_is_parent(const struct lu_env *env,
133 struct mdd_device *mdd,
134 struct mdd_object *p1,
135 const struct lu_fid *lf,
138 struct mdd_object *parent = NULL;
143 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
144 pfid = &mdd_env_info(env)->mti_fid;
146 /* Check for root first. */
147 if (mdd_is_root(mdd, mdo2fid(p1)))
151 /* this is done recursively, bypass capa for each obj */
152 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
153 rc = mdd_parent_fid(env, p1, pfid);
156 if (mdd_is_root(mdd, pfid))
158 if (lu_fid_eq(pfid, lf))
161 mdd_object_put(env, parent);
162 parent = mdd_object_find(env, mdd, pfid);
164 /* cross-ref parent */
165 if (parent == NULL) {
168 GOTO(out, rc = -EREMOTE);
169 } else if (IS_ERR(parent))
170 GOTO(out, rc = PTR_ERR(parent));
175 if (parent && !IS_ERR(parent))
176 mdd_object_put(env, parent);
181 * No permission check is needed.
183 * returns 1: if fid is ancestor of @mo;
184 * returns 0: if fid is not a ancestor of @mo;
186 * returns EREMOTE if remote object is found, fid of remote object is saved to
189 * returns < 0: if error
191 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
192 const struct lu_fid *fid, struct lu_fid *sfid)
194 struct mdd_device *mdd = mdo2mdd(mo);
198 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
201 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
205 } else if (rc == 1) {
206 /* found @fid is parent */
214 * Check that @dir contains no entries except (possibly) dot and dotdot.
219 * -ENOTDIR not a directory object
220 * -ENOTEMPTY not empty
224 static int mdd_dir_is_empty(const struct lu_env *env,
225 struct mdd_object *dir)
228 struct dt_object *obj;
229 const struct dt_it_ops *iops;
233 obj = mdd_object_child(dir);
234 if (!dt_try_as_dir(env, obj))
237 iops = &obj->do_index_ops->dio_it;
238 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
240 result = iops->get(env, it, (const void *)"");
243 for (result = 0, i = 0; result == 0 && i < 3; ++i)
244 result = iops->next(env, it);
247 else if (result == +1)
249 } else if (result == 0)
251 * Huh? Index contains no zero key?
258 result = PTR_ERR(it);
262 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
264 struct mdd_device *m = mdd_obj2mdd_dev(obj);
265 struct lu_attr *la = &mdd_env_info(env)->mti_la;
269 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
274 * Subdir count limitation can be broken through.
276 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
277 !S_ISDIR(la->la_mode))
284 * Check whether it may create the cobj under the pobj.
287 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
288 struct mdd_object *cobj, int check_perm, int check_nlink)
293 if (cobj && mdd_object_exists(cobj))
296 if (mdd_is_dead_obj(pobj))
300 rc = mdd_permission_internal_locked(env, pobj, NULL,
301 MAY_WRITE | MAY_EXEC,
303 if (!rc && check_nlink)
304 rc = __mdd_may_link(env, pobj);
310 * Check whether can unlink from the pobj in the case of "cobj == NULL".
312 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
313 const struct lu_attr *attr)
318 if (mdd_is_dead_obj(pobj))
321 if ((attr->la_valid & LA_FLAGS) &&
322 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
325 rc = mdd_permission_internal_locked(env, pobj, NULL,
326 MAY_WRITE | MAY_EXEC,
331 if (mdd_is_append(pobj))
338 * pobj == NULL is remote ops case, under such case, pobj's
339 * VTX feature has been checked already, no need check again.
341 static inline int mdd_is_sticky(const struct lu_env *env,
342 struct mdd_object *pobj,
343 struct mdd_object *cobj)
345 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
346 struct md_ucred *uc = md_ucred(env);
350 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
354 if (!(tmp_la->la_mode & S_ISVTX) ||
355 (tmp_la->la_uid == uc->mu_fsuid))
359 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
363 if (tmp_la->la_uid == uc->mu_fsuid)
366 return !mdd_capable(uc, CFS_CAP_FOWNER);
370 * Check whether it may delete the cobj from the pobj.
373 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
374 struct mdd_object *cobj, struct lu_attr *cattr,
375 struct lu_attr *src_attr, int check_perm, int check_empty)
381 if (!mdd_object_exists(cobj))
384 if (mdd_is_dead_obj(cobj))
388 if (!mdd_object_exists(pobj))
391 if (mdd_is_dead_obj(pobj))
395 rc = mdd_permission_internal_locked(env, pobj, NULL,
396 MAY_WRITE | MAY_EXEC,
402 if (mdd_is_append(pobj))
406 if (mdd_is_sticky(env, pobj, cobj))
409 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
412 if ((cattr->la_valid & LA_FLAGS) &&
413 (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
416 /* additional check the rename case */
418 if (S_ISDIR(src_attr->la_mode)) {
419 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
421 if (!S_ISDIR(cattr->la_mode))
424 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
426 } else if (S_ISDIR(cattr->la_mode))
430 if (S_ISDIR(cattr->la_mode) && check_empty)
431 rc = mdd_dir_is_empty(env, cobj);
438 * has mdd_write_lock on src already, but not on tgt yet
440 int mdd_link_sanity_check(const struct lu_env *env,
441 struct mdd_object *tgt_obj,
442 const struct lu_name *lname,
443 struct mdd_object *src_obj)
445 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
449 if (!mdd_object_exists(src_obj))
452 if (mdd_is_dead_obj(src_obj))
455 /* Local ops, no lookup before link, check filename length here. */
456 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
457 RETURN(-ENAMETOOLONG);
459 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
462 if (S_ISDIR(mdd_object_type(src_obj)))
465 LASSERT(src_obj != tgt_obj);
467 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
472 rc = __mdd_may_link(env, src_obj);
477 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
478 const char *name, struct thandle *handle,
479 struct lustre_capa *capa)
481 struct dt_object *next = mdd_object_child(pobj);
485 if (dt_try_as_dir(env, next)) {
486 rc = next->do_index_ops->dio_delete(env, next,
487 (struct dt_key *)name,
495 static int __mdd_index_insert_only(const struct lu_env *env,
496 struct mdd_object *pobj,
497 const struct lu_fid *lf, const char *name,
498 struct thandle *handle,
499 struct lustre_capa *capa)
501 struct dt_object *next = mdd_object_child(pobj);
505 if (dt_try_as_dir(env, next)) {
506 struct md_ucred *uc = md_ucred(env);
508 rc = next->do_index_ops->dio_insert(env, next,
510 (const struct dt_key *)name,
511 handle, capa, uc->mu_cap &
512 CFS_CAP_SYS_RESOURCE_MASK);
519 /* insert named index, add reference if isdir */
520 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
521 const struct lu_fid *lf, const char *name, int is_dir,
522 struct thandle *handle, struct lustre_capa *capa)
527 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
528 if (rc == 0 && is_dir) {
529 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
530 mdo_ref_add(env, pobj, handle);
531 mdd_write_unlock(env, pobj);
536 /* delete named index, drop reference if isdir */
537 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
538 const char *name, int is_dir, struct thandle *handle,
539 struct lustre_capa *capa)
544 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
545 if (rc == 0 && is_dir) {
546 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
547 mdo_ref_del(env, pobj, handle);
548 mdd_write_unlock(env, pobj);
554 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
555 int reclen, struct thandle *handle)
559 /* XXX: this is a temporary solution to declare llog changes
560 * will be fixed in 2.3 with new llog implementation */
562 LASSERT(mdd->mdd_capa);
564 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
565 * have to set the parameter 'size' as INT_MAX or 0 to inform
566 * OSD that this record write is for a llog write or catalog
567 * header update, and osd declare function will reserve less
568 * credits for optimization purpose.
570 * Reserve 6 blocks for a llog write, since the llog file is
571 * usually small, reserve 2 blocks for catalog header update,
572 * because we know for sure that catalog header is already
575 * This hack should be removed in 2.3.
579 rc = dt_declare_record_write(env, mdd->mdd_capa,
580 DECLARE_LLOG_WRITE, 0, handle);
584 /* header will be updated as well */
585 rc = dt_declare_record_write(env, mdd->mdd_capa,
586 DECLARE_LLOG_WRITE, 0, handle);
590 /* also we should be able to create new plain log */
591 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
595 /* new record referencing new plain llog */
596 rc = dt_declare_record_write(env, mdd->mdd_capa,
597 DECLARE_LLOG_WRITE, 0, handle);
601 /* catalog's header will be updated as well */
602 rc = dt_declare_record_write(env, mdd->mdd_capa,
603 DECLARE_LLOG_REWRITE, 0, handle);
608 int mdd_declare_changelog_store(const struct lu_env *env,
609 struct mdd_device *mdd,
610 const struct lu_name *fname,
611 struct thandle *handle)
616 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
619 /* we'll be writing payload + llog header */
620 reclen = sizeof(struct llog_changelog_rec);
622 reclen += fname->ln_namelen;
623 reclen = llog_data_len(reclen);
625 return mdd_declare_llog_record(env, mdd, reclen, handle);
628 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
629 struct mdd_device *mdd,
630 const struct lu_name *tname,
631 const struct lu_name *sname,
632 struct thandle *handle)
637 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
640 /* we'll be writing payload + llog header */
641 reclen = sizeof(struct llog_changelog_ext_rec);
643 reclen += tname->ln_namelen;
645 reclen += 1 + sname->ln_namelen;
646 reclen = llog_data_len(reclen);
648 return mdd_declare_llog_record(env, mdd, reclen, handle);
651 /** Store a namespace change changelog record
652 * If this fails, we must fail the whole transaction; we don't
653 * want the change to commit without the log entry.
654 * \param target - mdd_object of change
655 * \param parent - parent dir/object
656 * \param tname - target name string
657 * \param handle - transacion handle
659 static int mdd_changelog_ns_store(const struct lu_env *env,
660 struct mdd_device *mdd,
661 enum changelog_rec_type type,
663 struct mdd_object *target,
664 struct mdd_object *parent,
665 const struct lu_name *tname,
666 struct thandle *handle)
668 struct llog_changelog_rec *rec;
675 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
677 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
680 LASSERT(target != NULL);
681 LASSERT(parent != NULL);
682 LASSERT(tname != NULL);
683 LASSERT(handle != NULL);
685 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
686 buf = mdd_buf_alloc(env, reclen);
687 if (buf->lb_buf == NULL)
689 rec = (struct llog_changelog_rec *)buf->lb_buf;
691 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
692 rec->cr.cr_type = (__u32)type;
693 rec->cr.cr_tfid = *mdo2fid(target);
694 rec->cr.cr_pfid = *mdo2fid(parent);
695 rec->cr.cr_namelen = tname->ln_namelen;
696 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
698 target->mod_cltime = cfs_time_current_64();
700 rc = mdd_changelog_llog_write(mdd, rec, handle);
702 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
703 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
704 PFID(&rec->cr.cr_pfid));
712 /** Store a namespace change changelog record
713 * If this fails, we must fail the whole transaction; we don't
714 * want the change to commit without the log entry.
715 * \param target - mdd_object of change
716 * \param tpfid - target parent dir/object fid
717 * \param sfid - source object fid
718 * \param spfid - source parent fid
719 * \param tname - target name string
720 * \param sname - source name string
721 * \param handle - transacion handle
723 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
724 struct mdd_device *mdd,
725 enum changelog_rec_type type,
727 struct mdd_object *target,
728 const struct lu_fid *tpfid,
729 const struct lu_fid *sfid,
730 const struct lu_fid *spfid,
731 const struct lu_name *tname,
732 const struct lu_name *sname,
733 struct thandle *handle)
735 struct llog_changelog_ext_rec *rec;
742 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
744 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
747 LASSERT(sfid != NULL);
748 LASSERT(tpfid != NULL);
749 LASSERT(tname != NULL);
750 LASSERT(handle != NULL);
752 reclen = sizeof(*rec) + tname->ln_namelen;
754 reclen += 1 + sname->ln_namelen;
755 reclen = llog_data_len(reclen);
756 buf = mdd_buf_alloc(env, reclen);
757 if (buf->lb_buf == NULL)
759 rec = (struct llog_changelog_ext_rec *)buf->lb_buf;
761 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
762 rec->cr.cr_type = (__u32)type;
763 rec->cr.cr_pfid = *tpfid;
764 rec->cr.cr_sfid = *sfid;
765 rec->cr.cr_spfid = *spfid;
766 rec->cr.cr_namelen = tname->ln_namelen;
767 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
769 LASSERT(sfid != NULL);
770 rec->cr.cr_name[tname->ln_namelen] = '\0';
771 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
773 rec->cr.cr_namelen += 1 + sname->ln_namelen;
776 if (likely(target != NULL)) {
777 rec->cr.cr_tfid = *mdo2fid(target);
778 target->mod_cltime = cfs_time_current_64();
780 fid_zero(&rec->cr.cr_tfid);
783 rc = mdd_changelog_ext_llog_write(mdd, rec, handle);
785 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
786 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
793 static int mdd_declare_link(const struct lu_env *env,
794 struct mdd_device *mdd,
795 struct mdd_object *p,
796 struct mdd_object *c,
797 const struct lu_name *name,
798 struct thandle *handle)
802 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
806 rc = mdo_declare_ref_add(env, c, handle);
810 rc = mdo_declare_attr_set(env, p, NULL, handle);
814 rc = mdo_declare_attr_set(env, c, NULL, handle);
818 rc = mdd_declare_links_add(env, c, handle);
822 rc = mdd_declare_changelog_store(env, mdd, name, handle);
827 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
828 struct md_object *src_obj, const struct lu_name *lname,
831 const char *name = lname->ln_name;
832 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
833 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
834 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
835 struct mdd_device *mdd = mdo2mdd(src_obj);
836 struct dynlock_handle *dlh;
837 struct thandle *handle;
838 #ifdef HAVE_QUOTA_SUPPORT
839 struct obd_device *obd = mdd->mdd_obd_dev;
840 struct obd_export *exp = md_quota(env)->mq_exp;
841 struct mds_obd *mds = &obd->u.mds;
842 unsigned int qids[MAXQUOTAS] = { 0, 0 };
843 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
848 #ifdef HAVE_QUOTA_SUPPORT
849 if (mds->mds_quota) {
850 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
852 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
855 mdd_data_get(env, mdd_tobj, &data);
856 quota_opc = FSFILT_OP_LINK;
857 mdd_quota_wrapper(la_tmp, qids);
858 /* get block quota for parent */
859 lquota_chkquota(mds_quota_interface_ref, obd, exp,
860 qids, rec_pending, 1, NULL,
861 LQUOTA_FLAGS_BLK, data, 1);
866 handle = mdd_trans_create(env, mdd);
868 GOTO(out_pending, rc = PTR_ERR(handle));
870 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
874 rc = mdd_trans_start(env, mdd, handle);
878 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
880 GOTO(out_trans, rc = -ENOMEM);
881 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
883 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
885 GOTO(out_unlock, rc);
887 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
889 mdd_object_capa(env, mdd_tobj));
891 GOTO(out_unlock, rc);
893 rc = mdo_ref_add(env, mdd_sobj, handle);
895 __mdd_index_delete_only(env, mdd_tobj, name, handle,
896 mdd_object_capa(env, mdd_tobj));
897 GOTO(out_unlock, rc);
900 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
901 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
903 la->la_valid = LA_CTIME | LA_MTIME;
904 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
906 GOTO(out_unlock, rc);
908 la->la_valid = LA_CTIME;
909 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
911 mdd_links_add(env, mdd_sobj,
912 mdo2fid(mdd_tobj), lname, handle, 0);
917 mdd_write_unlock(env, mdd_sobj);
918 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
921 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
922 mdd_tobj, lname, handle);
924 mdd_trans_stop(env, mdd, rc, handle);
926 #ifdef HAVE_QUOTA_SUPPORT
928 lquota_pending_commit(mds_quota_interface_ref, obd,
929 qids, rec_pending, 1);
930 /* Trigger dqacq for the parent owner. If failed,
931 * the next call for lquota_chkquota will process it. */
932 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
939 int mdd_declare_finish_unlink(const struct lu_env *env,
940 struct mdd_object *obj,
942 struct thandle *handle)
946 rc = orph_declare_index_insert(env, obj, handle);
950 return mdd_declare_object_kill(env, obj, ma, handle);
953 /* caller should take a lock before calling */
954 int mdd_finish_unlink(const struct lu_env *env,
955 struct mdd_object *obj, struct md_attr *ma,
960 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
963 LASSERT(mdd_write_locked(env, obj) != 0);
965 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
966 obj->mod_flags |= DEAD_OBJ;
967 /* add new orphan and the object
968 * will be deleted during mdd_close() */
969 if (obj->mod_count) {
970 rc = __mdd_orphan_add(env, obj, th);
972 CDEBUG(D_HA, "Object "DFID" is inserted into "
973 "orphan list, open count = %d\n",
974 PFID(mdd_object_fid(obj)),
977 CERROR("Object "DFID" fail to be an orphan, "
978 "open count = %d, maybe cause failed "
980 PFID(mdd_object_fid(obj)),
983 rc = mdd_object_kill(env, obj, ma, th);
989 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
996 * has mdd_write_lock on cobj already, but not on pobj yet
998 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
999 struct mdd_object *cobj, struct lu_attr *cattr)
1004 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1009 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1010 struct mdd_object *p, struct mdd_object *c,
1011 const struct lu_name *name, struct md_attr *ma,
1012 struct thandle *handle)
1016 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1020 rc = mdo_declare_ref_del(env, p, handle);
1024 rc = mdo_declare_ref_del(env, c, handle);
1028 rc = mdo_declare_ref_del(env, c, handle);
1032 rc = mdo_declare_attr_set(env, p, NULL, handle);
1036 rc = mdo_declare_attr_set(env, c, NULL, handle);
1040 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1044 rc = mdd_declare_links_add(env, c, handle);
1048 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1053 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1054 struct md_object *cobj, const struct lu_name *lname,
1057 const char *name = lname->ln_name;
1058 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1059 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1060 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1061 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
1062 struct mdd_device *mdd = mdo2mdd(pobj);
1063 struct dynlock_handle *dlh;
1064 struct thandle *handle;
1065 #ifdef HAVE_QUOTA_SUPPORT
1066 struct obd_device *obd = mdd->mdd_obd_dev;
1067 struct mds_obd *mds = &obd->u.mds;
1068 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1069 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1075 if (mdd_object_exists(mdd_cobj) <= 0)
1078 handle = mdd_trans_create(env, mdd);
1080 RETURN(PTR_ERR(handle));
1082 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1087 rc = mdd_trans_start(env, mdd, handle);
1091 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1093 GOTO(stop, rc = -ENOMEM);
1094 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1097 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1101 is_dir = S_ISDIR(cattr->la_mode);
1103 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1107 rc = mdo_ref_del(env, mdd_cobj, handle);
1109 __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj),
1111 mdd_object_capa(env, mdd_pobj));
1115 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1116 mdd_object_capa(env, mdd_pobj));
1122 mdo_ref_del(env, mdd_cobj, handle);
1124 /* fetch updated nlink */
1125 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1129 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1130 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1132 la->la_valid = LA_CTIME | LA_MTIME;
1133 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1137 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1138 /* update ctime of an unlinked file only if it is still
1139 * opened or a link still exists */
1140 la->la_valid = LA_CTIME;
1141 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1146 /* XXX: this transfer to ma will be removed with LOD/OSP */
1147 ma->ma_attr = *cattr;
1148 ma->ma_valid |= MA_INODE;
1149 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1151 /* fetch updated nlink */
1153 rc = mdd_la_get(env, mdd_cobj, cattr,
1154 mdd_object_capa(env, mdd_cobj));
1156 #ifdef HAVE_QUOTA_SUPPORT
1157 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1158 ma->ma_attr.la_nlink == 0) {
1159 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1161 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1163 mdd_quota_wrapper(la_tmp, qpids);
1164 if (mdd_cobj->mod_count == 0) {
1165 quota_opc = FSFILT_OP_UNLINK;
1166 mdd_quota_wrapper(&ma->ma_attr, qcids);
1168 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1174 /* old files may not have link ea; ignore errors */
1175 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
1176 lname, NULL, NULL, handle);
1178 /* if object is removed then we can't get its attrs, use last get */
1179 if (cattr->la_nlink == 0) {
1180 ma->ma_attr = *cattr;
1181 ma->ma_valid |= MA_INODE;
1185 mdd_write_unlock(env, mdd_cobj);
1186 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1190 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1191 if ((ma->ma_valid & MA_HSM) &&
1192 (ma->ma_hsm.mh_flags & HS_EXISTS))
1193 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1195 rc = mdd_changelog_ns_store(env, mdd,
1196 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1197 mdd_cobj, mdd_pobj, lname, handle);
1201 mdd_trans_stop(env, mdd, rc, handle);
1202 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 55, 0)
1203 if (rc == 0 && ma->ma_valid & MA_COOKIE && ma->ma_valid & MA_LOV &&
1204 ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_UNLINK_DESTROY)
1205 /* Since echo client is incapable of destorying ost object,
1206 * it will destory the object here. */
1207 rc = mdd_lovobj_unlink(env, mdd, mdd_cobj, la, ma, 1);
1209 #warning "please remove this after 2.4 (LOD/OSP)."
1212 #ifdef HAVE_QUOTA_SUPPORT
1214 /* Trigger dqrel on the owner of child and parent. If failed,
1215 * the next call for lquota_chkquota will process it. */
1216 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1223 * The permission has been checked when obj created, no need check again.
1225 static int mdd_cd_sanity_check(const struct lu_env *env,
1226 struct mdd_object *obj)
1231 if (!obj || mdd_is_dead_obj(obj))
1238 static int mdd_declare_create_data(const struct lu_env *env,
1239 struct mdd_device *mdd,
1240 struct mdd_object *obj,
1242 struct thandle *handle)
1244 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1248 buf->lb_len = lmm_size;
1249 rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
1254 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1259 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1260 struct md_object *cobj, const struct md_op_spec *spec,
1263 struct mdd_device *mdd = mdo2mdd(cobj);
1264 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1265 struct mdd_object *son = md2mdd_obj(cobj);
1266 struct lov_mds_md *lmm = NULL;
1268 struct thandle *handle;
1269 struct lu_attr *attr = &mdd_env_info(env)->mti_la_for_fix;
1273 rc = mdd_cd_sanity_check(env, son);
1277 if (!md_should_create(spec->sp_cr_flags))
1279 lmm_size = ma->ma_lmm_size;
1281 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, spec, ma);
1285 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1289 /* calling ->ah_make_hint() is used to transfer information from parent */
1290 mdd_object_make_hint(env, mdd_pobj, son, attr);
1292 handle = mdd_trans_create(env, mdd);
1294 GOTO(out_free, rc = PTR_ERR(handle));
1296 rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle);
1300 rc = mdd_trans_start(env, mdd, handle);
1305 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1306 * Should this be fixed?
1309 /* Replay creates has objects already */
1311 if (spec->no_create) {
1312 CDEBUG(D_INFO, "we already have lov ea\n");
1313 rc = mdd_lov_set_md(env, mdd_pobj, son,
1314 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1315 spec->u.sp_ea.eadatalen, handle, 0);
1318 /* No need mdd_lsm_sanity_check here */
1319 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1320 lmm_size, handle, 0);
1322 /* update lov_objid data, must be before transaction stop! */
1324 mdd_lov_objid_update(mdd, lmm);
1327 mdd_trans_stop(env, mdd, rc, handle);
1329 /* Finish mdd_lov_create() stuff. */
1330 /* if no_create == 0 (not replay), we free lmm allocated by
1331 * mdd_lov_create() */
1332 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1336 /* Get fid from name and parent */
1338 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1339 const struct lu_name *lname, struct lu_fid* fid, int mask)
1341 const char *name = lname->ln_name;
1342 const struct dt_key *key = (const struct dt_key *)name;
1343 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1344 struct mdd_device *m = mdo2mdd(pobj);
1345 struct dt_object *dir = mdd_object_child(mdd_obj);
1349 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1352 rc = mdd_object_exists(mdd_obj);
1353 if (unlikely(rc == 0))
1355 else if (unlikely(rc < 0)) {
1356 CERROR("Object "DFID" locates on remote server\n",
1357 PFID(mdo2fid(mdd_obj)));
1361 /* The common filename length check. */
1362 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1363 RETURN(-ENAMETOOLONG);
1365 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1370 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1371 dt_try_as_dir(env, dir))) {
1373 rc = dir->do_index_ops->dio_lookup(env, dir,
1374 (struct dt_rec *)fid, key,
1375 mdd_object_capa(env, mdd_obj));
1386 int mdd_declare_object_initialize(const struct lu_env *env,
1387 struct mdd_object *child,
1388 struct lu_attr *attr,
1389 struct thandle *handle)
1393 rc = mdo_declare_attr_set(env, child, attr, handle);
1394 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1395 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1398 rc = mdo_declare_ref_add(env, child, handle);
1402 mdd_declare_links_add(env, child, handle);
1407 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1408 const struct lu_name *lname, struct mdd_object *child,
1409 struct lu_attr *attr, struct thandle *handle,
1410 const struct md_op_spec *spec)
1416 * Update attributes for child.
1419 * (1) the valid bits should be converted between Lustre and Linux;
1420 * (2) maybe, the child attributes should be set in OSD when creation.
1423 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1427 if (S_ISDIR(attr->la_mode)) {
1428 /* Add "." and ".." for newly created dir */
1429 mdo_ref_add(env, child, handle);
1430 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1431 dot, handle, BYPASS_CAPA);
1433 rc = __mdd_index_insert_only(env, child, pfid,
1437 mdo_ref_del(env, child, handle);
1440 mdd_links_add(env, child, pfid, lname, handle, 1);
1445 /* has not lock on pobj yet */
1446 static int mdd_create_sanity_check(const struct lu_env *env,
1447 struct md_object *pobj,
1448 struct lu_attr *pattr,
1449 const struct lu_name *lname,
1450 struct lu_attr *cattr,
1451 struct md_op_spec *spec)
1453 struct mdd_thread_info *info = mdd_env_info(env);
1454 struct lu_fid *fid = &info->mti_fid;
1455 struct mdd_object *obj = md2mdd_obj(pobj);
1456 struct mdd_device *m = mdo2mdd(pobj);
1457 int lookup = spec->sp_cr_lookup;
1462 if (mdd_is_dead_obj(obj))
1466 * In some cases this lookup is not needed - we know before if name
1467 * exists or not because MDT performs lookup for it.
1468 * name length check is done in lookup.
1472 * Check if the name already exist, though it will be checked in
1473 * _index_insert also, for avoiding rolling back if exists
1476 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1477 MAY_WRITE | MAY_EXEC);
1479 RETURN(rc ? : -EEXIST);
1482 * Check WRITE permission for the parent.
1483 * EXEC permission have been checked
1484 * when lookup before create already.
1486 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1493 if (pattr->la_mode & S_ISGID) {
1494 cattr->la_gid = pattr->la_gid;
1495 if (S_ISDIR(cattr->la_mode)) {
1496 cattr->la_mode |= S_ISGID;
1497 cattr->la_valid |= LA_MODE;
1501 switch (cattr->la_mode & S_IFMT) {
1503 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1505 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1506 RETURN(-ENAMETOOLONG);
1525 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1526 struct mdd_object *p, struct mdd_object *c,
1527 const struct lu_name *name,
1528 struct lu_attr *attr, int lmm_size,
1530 struct thandle *handle,
1531 const struct md_op_spec *spec)
1533 struct mdd_thread_info *info = mdd_env_info(env);
1534 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1537 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1541 #ifdef CONFIG_FS_POSIX_ACL
1542 if (got_def_acl > 0) {
1543 struct lu_buf *acl_buf;
1545 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1546 /* if dir, then can inherit default ACl */
1547 if (S_ISDIR(attr->la_mode)) {
1548 rc = mdo_declare_xattr_set(env, c, acl_buf,
1549 XATTR_NAME_ACL_DEFAULT,
1555 rc = mdo_declare_attr_set(env, c, &info->mti_pattr, handle);
1559 rc = mdo_declare_xattr_set(env, c, acl_buf,
1560 XATTR_NAME_ACL_ACCESS, 0, handle);
1566 if (S_ISDIR(attr->la_mode)) {
1567 rc = mdo_declare_ref_add(env, p, handle);
1572 rc = mdd_declare_object_initialize(env, c, attr, handle);
1576 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1577 name->ln_name, handle);
1581 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1586 if (S_ISLNK(attr->la_mode)) {
1587 rc = dt_declare_record_write(env, mdd_object_child(c),
1588 strlen(spec->u.sp_symname), 0,
1594 rc = mdo_declare_attr_set(env, p, attr, handle);
1598 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1602 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1610 * Create object and insert it into namespace.
1612 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1613 const struct lu_name *lname, struct md_object *child,
1614 struct md_op_spec *spec, struct md_attr* ma)
1616 struct mdd_thread_info *info = mdd_env_info(env);
1617 struct lu_attr *la = &info->mti_la_for_fix;
1618 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1619 struct mdd_object *son = md2mdd_obj(child);
1620 struct mdd_device *mdd = mdo2mdd(pobj);
1621 struct lu_attr *attr = &ma->ma_attr;
1622 struct lov_mds_md *lmm = NULL;
1623 struct thandle *handle;
1624 struct lu_attr *pattr = &info->mti_pattr;
1625 struct dynlock_handle *dlh;
1626 const char *name = lname->ln_name;
1627 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1628 int got_def_acl = 0;
1629 #ifdef HAVE_QUOTA_SUPPORT
1630 struct obd_device *obd = mdd->mdd_obd_dev;
1631 struct obd_export *exp = md_quota(env)->mq_exp;
1632 struct mds_obd *mds = &obd->u.mds;
1633 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1634 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1635 int quota_opc = 0, block_count = 0;
1636 int inode_pending[MAXQUOTAS] = { 0, 0 };
1637 int block_pending[MAXQUOTAS] = { 0, 0 };
1638 int parent_pending[MAXQUOTAS] = { 0, 0 };
1643 * Two operations have to be performed:
1645 * - an allocation of a new object (->do_create()), and
1647 * - an insertion into a parent index (->dio_insert()).
1649 * Due to locking, operation order is not important, when both are
1650 * successful, *but* error handling cases are quite different:
1652 * - if insertion is done first, and following object creation fails,
1653 * insertion has to be rolled back, but this operation might fail
1654 * also leaving us with dangling index entry.
1656 * - if creation is done first, is has to be undone if insertion
1657 * fails, leaving us with leaked space, which is neither good, nor
1660 * It seems that creation-first is simplest solution, but it is
1661 * sub-optimal in the frequent
1666 * case, because second mkdir is bound to create object, only to
1667 * destroy it immediately.
1669 * To avoid this follow local file systems that do double lookup:
1671 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1673 * 1. create (mdd_object_create_internal())
1675 * 2. insert (__mdd_index_insert(), lookup again)
1678 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1682 /* Sanity checks before big job. */
1683 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
1687 #ifdef HAVE_QUOTA_SUPPORT
1688 if (mds->mds_quota) {
1691 quota_opc = FSFILT_OP_CREATE;
1692 mdd_quota_wrapper(attr, qcids);
1693 mdd_quota_wrapper(pattr, qpids);
1694 /* get file quota for child */
1695 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1696 qcids, inode_pending, 1, NULL, 0, NULL,
1698 switch (attr->la_mode & S_IFMT) {
1707 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1708 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1712 /* get block quota for child and parent */
1714 lquota_chkquota(mds_quota_interface_ref, obd,
1715 exp, qcids, block_pending,
1717 LQUOTA_FLAGS_BLK, NULL, 0);
1719 lquota_chkquota(mds_quota_interface_ref, obd,
1720 exp, qpids, parent_pending, 1,
1721 NULL, LQUOTA_FLAGS_BLK, NULL,
1726 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1727 GOTO(out_pending, rc = -EINPROGRESS);
1730 * No RPC inside the transaction, so OST objects should be created at
1733 if (S_ISREG(attr->la_mode)) {
1734 lmm_size = ma->ma_lmm_size;
1735 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1738 GOTO(out_pending, rc);
1741 if (!S_ISLNK(attr->la_mode)) {
1742 struct lu_buf *acl_buf;
1744 acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
1745 sizeof(info->mti_xattr_buf));
1746 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1747 rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
1748 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1749 mdd_read_unlock(env, mdd_pobj);
1752 else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
1756 mdd_object_make_hint(env, mdd_pobj, son, attr);
1758 handle = mdd_trans_create(env, mdd);
1760 GOTO(out_free, rc = PTR_ERR(handle));
1762 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
1763 got_def_acl, lmm_size, handle, spec);
1767 rc = mdd_trans_start(env, mdd, handle);
1771 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1773 GOTO(out_trans, rc = -ENOMEM);
1775 mdd_write_lock(env, son, MOR_TGT_CHILD);
1776 rc = mdd_object_create_internal(env, mdd_pobj, son, attr, handle, spec);
1778 mdd_write_unlock(env, son);
1784 #ifdef CONFIG_FS_POSIX_ACL
1786 struct lu_buf *acl_buf;
1788 acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
1789 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1791 mdd_write_unlock(env, son);
1797 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1798 son, attr, handle, spec);
1799 mdd_write_unlock(env, son);
1802 * Object has no links, so it will be destroyed when last
1803 * reference is released. (XXX not now.)
1809 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1810 name, S_ISDIR(attr->la_mode), handle,
1811 mdd_object_capa(env, mdd_pobj));
1817 /* No need mdd_lsm_sanity_check here */
1818 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1820 CERROR("error on stripe info copy %d \n", rc);
1824 if (S_ISLNK(attr->la_mode)) {
1825 struct md_ucred *uc = md_ucred(env);
1826 struct dt_object *dt = mdd_object_child(son);
1827 const char *target_name = spec->u.sp_symname;
1828 int sym_len = strlen(target_name);
1829 const struct lu_buf *buf;
1832 buf = mdd_buf_get_const(env, target_name, sym_len);
1833 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1834 mdd_object_capa(env, son),
1836 CFS_CAP_SYS_RESOURCE_MASK);
1841 GOTO(cleanup, rc = -EFAULT);
1845 la->la_valid = LA_CTIME | LA_MTIME;
1846 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1852 if (rc != 0 && created != 0) {
1855 if (inserted != 0) {
1856 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1857 S_ISDIR(attr->la_mode),
1858 handle, BYPASS_CAPA);
1863 mdd_write_lock(env, son, MOR_TGT_CHILD);
1864 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
1865 /* Drop the reference, no need to delete "."/"..",
1866 * because the object to be destroied directly. */
1867 rc2 = mdo_ref_del(env, son, handle);
1869 mdd_write_unlock(env, son);
1874 rc2 = mdo_ref_del(env, son, handle);
1876 mdd_write_unlock(env, son);
1880 mdo_destroy(env, son, handle);
1881 mdd_write_unlock(env, son);
1884 /* update lov_objid data, must be before transaction stop! */
1886 mdd_lov_objid_update(mdd, lmm);
1888 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1891 rc = mdd_changelog_ns_store(env, mdd,
1892 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1893 S_ISREG(attr->la_mode) ? CL_CREATE :
1894 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1895 0, son, mdd_pobj, lname, handle);
1897 mdd_trans_stop(env, mdd, rc, handle);
1899 /* finish lov_create stuff, free all temporary data */
1900 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1902 #ifdef HAVE_QUOTA_SUPPORT
1904 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1906 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1908 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1910 /* Trigger dqacq on the owner of child and parent. If failed,
1911 * the next call for lquota_chkquota will process it. */
1912 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1916 /* The child object shouldn't be cached anymore */
1918 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
1919 &child->mo_lu.lo_header->loh_flags);
1924 * Get locks on parents in proper order
1925 * RETURN: < 0 - error, rename_order if successful
1933 static int mdd_rename_order(const struct lu_env *env,
1934 struct mdd_device *mdd,
1935 struct mdd_object *src_pobj,
1936 struct mdd_object *tgt_pobj)
1938 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1942 if (src_pobj == tgt_pobj)
1943 RETURN(MDD_RN_SAME);
1945 /* compared the parent child relationship of src_p&tgt_p */
1946 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1948 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1951 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1964 /* has not mdd_write{read}_lock on any obj yet. */
1965 static int mdd_rename_sanity_check(const struct lu_env *env,
1966 struct mdd_object *src_pobj,
1967 struct mdd_object *tgt_pobj,
1968 struct mdd_object *sobj,
1969 struct mdd_object *tobj,
1970 struct lu_attr *so_attr,
1971 struct lu_attr *tg_attr)
1976 /* XXX: when get here, sobj must NOT be NULL,
1977 * the other case has been processed in cml_rename
1978 * before mdd_rename and enable MDS_PERM_BYPASS. */
1981 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
1985 /* XXX: when get here, "tobj == NULL" means tobj must
1986 * NOT exist (neither on remote MDS, such case has been
1987 * processed in cml_rename before mdd_rename and enable
1989 * So check may_create, but not check may_unlink. */
1991 rc = mdd_may_create(env, tgt_pobj, NULL,
1992 (src_pobj != tgt_pobj), 0);
1994 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
1995 (src_pobj != tgt_pobj), 1);
1997 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1998 S_ISDIR(so_attr->la_mode))
1999 rc = __mdd_may_link(env, tgt_pobj);
2004 static int mdd_declare_rename(const struct lu_env *env,
2005 struct mdd_device *mdd,
2006 struct mdd_object *mdd_spobj,
2007 struct mdd_object *mdd_tpobj,
2008 struct mdd_object *mdd_sobj,
2009 struct mdd_object *mdd_tobj,
2010 const struct lu_name *tname,
2011 const struct lu_name *sname,
2013 struct thandle *handle)
2021 /* name from source dir */
2022 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2026 /* .. from source child */
2027 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2028 /* source child can be directory,
2029 * counted by source dir's nlink */
2030 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2034 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2038 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2043 /* new target child can be directory,
2044 * counted by target dir's nlink */
2045 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2051 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2055 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2058 mdd_declare_links_add(env, mdd_sobj, handle);
2062 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2067 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2068 tname->ln_name, handle);
2072 /* name from target dir (old name), we declare it unconditionally
2073 * as mdd_rename() calls delete unconditionally as well. so just
2074 * to balance declarations vs calls to change ... */
2075 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2079 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2080 /* delete target child in target parent directory */
2081 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2085 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2086 /* target child can be directory,
2087 * delete "." reference in target child directory */
2088 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2092 /* delete ".." reference in target parent directory */
2093 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2098 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2102 mdd_declare_links_add(env, mdd_tobj, handle);
2106 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2111 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2118 /* src object can be remote that is why we use only fid and type of object */
2119 static int mdd_rename(const struct lu_env *env,
2120 struct md_object *src_pobj, struct md_object *tgt_pobj,
2121 const struct lu_fid *lf, const struct lu_name *lsname,
2122 struct md_object *tobj, const struct lu_name *ltname,
2125 const char *sname = lsname->ln_name;
2126 const char *tname = ltname->ln_name;
2127 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2128 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2129 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2130 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2131 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2132 struct mdd_device *mdd = mdo2mdd(src_pobj);
2133 struct mdd_object *mdd_sobj = NULL; /* source object */
2134 struct mdd_object *mdd_tobj = NULL;
2135 struct dynlock_handle *sdlh, *tdlh;
2136 struct thandle *handle;
2137 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2138 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2140 unsigned cl_flags = 0;
2143 #ifdef HAVE_QUOTA_SUPPORT
2144 struct obd_device *obd = mdd->mdd_obd_dev;
2145 struct obd_export *exp = md_quota(env)->mq_exp;
2146 struct mds_obd *mds = &obd->u.mds;
2147 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
2148 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
2149 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
2150 int quota_copc = 0, quota_popc = 0;
2151 int rec_pending[MAXQUOTAS] = { 0, 0 };
2156 mdd_tobj = md2mdd_obj(tobj);
2158 #ifdef HAVE_QUOTA_SUPPORT
2159 if (mds->mds_quota) {
2160 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2162 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2164 mdd_quota_wrapper(la_tmp, qspids);
2166 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2170 mdd_data_get(env, mdd_tpobj, &data);
2171 quota_popc = FSFILT_OP_LINK;
2172 mdd_quota_wrapper(la_tmp, qtpids);
2173 /* get block quota for target parent */
2174 lquota_chkquota(mds_quota_interface_ref,
2176 rec_pending, 1, NULL,
2184 mdd_sobj = mdd_object_find(env, mdd, lf);
2186 handle = mdd_trans_create(env, mdd);
2188 GOTO(out_pending, rc = PTR_ERR(handle));
2190 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2191 mdd_tobj, lsname, ltname, ma, handle);
2195 rc = mdd_trans_start(env, mdd, handle);
2199 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2200 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2202 GOTO(cleanup_unlocked, rc);
2204 /* Get locks in determined order */
2205 if (rc == MDD_RN_SAME) {
2206 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2207 sname, MOR_SRC_PARENT);
2208 /* check hashes to determine do we need one lock or two */
2209 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2210 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2214 } else if (rc == MDD_RN_SRCTGT) {
2215 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2216 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2218 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2219 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2221 if (sdlh == NULL || tdlh == NULL)
2222 GOTO(cleanup, rc = -ENOMEM);
2224 rc = mdd_la_get(env, mdd_sobj, so_attr,
2225 mdd_object_capa(env, mdd_sobj));
2230 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2231 mdd_object_capa(env, mdd_tobj));
2236 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2237 mdd_tobj, so_attr, tg_attr);
2241 is_dir = S_ISDIR(so_attr->la_mode);
2243 /* Remove source name from source directory */
2244 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2245 mdd_object_capa(env, mdd_spobj));
2249 /* "mv dir1 dir2" needs "dir1/.." link update */
2250 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2251 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2252 mdd_object_capa(env, mdd_sobj));
2254 GOTO(fixup_spobj2, rc);
2256 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2257 handle, mdd_object_capa(env, mdd_sobj));
2259 GOTO(fixup_spobj, rc);
2262 /* Remove target name from target directory
2263 * Here tobj can be remote one, so we do index_delete unconditionally
2264 * and -ENOENT is allowed.
2266 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2267 mdd_object_capa(env, mdd_tpobj));
2270 /* tname might been renamed to something else */
2271 GOTO(fixup_spobj, rc);
2274 GOTO(fixup_spobj, rc);
2277 /* Insert new fid with target name into target dir */
2278 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2279 mdd_object_capa(env, mdd_tpobj));
2281 GOTO(fixup_tpobj, rc);
2283 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2284 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2286 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2288 la->la_valid = LA_CTIME;
2289 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2291 GOTO(fixup_tpobj, rc);
2294 /* Remove old target object
2295 * For tobj is remote case cmm layer has processed
2296 * and set tobj to NULL then. So when tobj is NOT NULL,
2297 * it must be local one.
2299 if (tobj && mdd_object_exists(mdd_tobj)) {
2300 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2301 if (mdd_is_dead_obj(mdd_tobj)) {
2302 mdd_write_unlock(env, mdd_tobj);
2303 /* shld not be dead, something is wrong */
2304 CERROR("tobj is dead, something is wrong\n");
2308 mdo_ref_del(env, mdd_tobj, handle);
2310 /* Remove dot reference. */
2311 if (S_ISDIR(tg_attr->la_mode))
2312 mdo_ref_del(env, mdd_tobj, handle);
2314 /* fetch updated nlink */
2315 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2316 mdd_object_capa(env, mdd_tobj));
2318 GOTO(fixup_tpobj, rc);
2320 la->la_valid = LA_CTIME;
2321 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2323 GOTO(fixup_tpobj, rc);
2325 /* XXX: this transfer to ma will be removed with LOD/OSP */
2326 ma->ma_attr = *tg_attr;
2327 ma->ma_valid |= MA_INODE;
2328 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2329 mdd_write_unlock(env, mdd_tobj);
2331 GOTO(fixup_tpobj, rc);
2333 /* fetch updated nlink */
2334 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2335 mdd_object_capa(env, mdd_tobj));
2337 GOTO(fixup_tpobj, rc);
2338 /* XXX: this transfer to ma will be removed with LOD/OSP */
2339 ma->ma_attr = *tg_attr;
2340 ma->ma_valid |= MA_INODE;
2342 if (so_attr->la_nlink == 0) {
2343 cl_flags |= CLF_RENAME_LAST;
2344 #ifdef HAVE_QUOTA_SUPPORT
2345 if (mds->mds_quota && mdd_tobj->mod_count == 0) {
2346 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2347 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2353 la->la_valid = LA_CTIME | LA_MTIME;
2354 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2356 GOTO(fixup_tpobj, rc);
2358 if (mdd_spobj != mdd_tpobj) {
2359 la->la_valid = LA_CTIME | LA_MTIME;
2360 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2364 if (rc == 0 && mdd_sobj) {
2365 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2366 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2367 mdo2fid(mdd_tpobj), ltname, handle);
2369 /* Old files might not have EA entry */
2370 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2372 mdd_write_unlock(env, mdd_sobj);
2373 /* We don't fail the transaction if the link ea can't be
2374 updated -- fid2path will use alternate lookup method. */
2382 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2385 CWARN("tp obj fix error %d\n",rc2);
2387 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2388 !mdd_is_dead_obj(mdd_tobj)) {
2389 rc2 = __mdd_index_insert(env, mdd_tpobj,
2390 mdo2fid(mdd_tobj), tname,
2395 CWARN("tp obj fix error %d\n",rc2);
2400 if (rc && is_dir && mdd_sobj) {
2401 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2405 CWARN("sp obj dotdot delete error %d\n",rc2);
2408 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2409 dotdot, handle, BYPASS_CAPA);
2411 CWARN("sp obj dotdot insert error %d\n",rc2);
2416 rc2 = __mdd_index_insert(env, mdd_spobj,
2417 lf, sname, is_dir, handle, BYPASS_CAPA);
2419 CWARN("sp obj fix error %d\n",rc2);
2422 if (likely(tdlh) && sdlh != tdlh)
2423 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2425 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2428 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2429 mdd_tobj, tpobj_fid, lf,
2430 spobj_fid, ltname, lsname,
2434 mdd_trans_stop(env, mdd, rc, handle);
2436 mdd_object_put(env, mdd_sobj);
2438 #ifdef HAVE_QUOTA_SUPPORT
2439 if (mds->mds_quota) {
2441 lquota_pending_commit(mds_quota_interface_ref, obd,
2442 qtpids, rec_pending, 1);
2445 /* Trigger dqrel on the source owner of parent.
2446 * If failed, the next call for lquota_chkquota will
2448 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2449 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2451 /* Trigger dqrel on the target owner of child.
2452 * If failed, the next call for lquota_chkquota
2453 * will process it. */
2454 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2455 qtpids, rc, quota_copc);
2462 /** enable/disable storing of hardlink info */
2463 int mdd_linkea_enable = 1;
2464 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2465 "record hardlink info in EAs");
2467 /** Read the link EA into a temp buffer.
2468 * Uses the name_buf since it is generally large.
2469 * \retval IS_ERR err
2470 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2472 struct lu_buf *mdd_links_get(const struct lu_env *env,
2473 struct mdd_object *mdd_obj)
2476 struct lustre_capa *capa;
2477 struct link_ea_header *leh;
2480 /* First try a small buf */
2481 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2482 if (buf->lb_buf == NULL)
2483 return ERR_PTR(-ENOMEM);
2485 if (!mdd_object_exists(mdd_obj))
2486 return ERR_PTR(-ENODATA);
2488 capa = mdd_object_capa(env, mdd_obj);
2489 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2490 if (rc == -ERANGE) {
2491 /* Buf was too small, figure out what we need. */
2493 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2496 buf = mdd_buf_alloc(env, rc);
2497 if (buf->lb_buf == NULL)
2498 return ERR_PTR(-ENOMEM);
2499 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2505 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2506 leh->leh_magic = LINK_EA_MAGIC;
2507 leh->leh_reccount = __swab32(leh->leh_reccount);
2508 leh->leh_len = __swab64(leh->leh_len);
2509 /* entries are swabbed by mdd_lee_unpack */
2511 if (leh->leh_magic != LINK_EA_MAGIC)
2512 return ERR_PTR(-EINVAL);
2513 if (leh->leh_reccount == 0)
2514 return ERR_PTR(-ENODATA);
2519 /** Pack a link_ea_entry.
2520 * All elements are stored as chars to avoid alignment issues.
2521 * Numbers are always big-endian
2522 * \retval record length
2524 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2525 const struct lu_fid *pfid)
2527 struct lu_fid tmpfid;
2530 fid_cpu_to_be(&tmpfid, pfid);
2531 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2532 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2533 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2535 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2536 lee->lee_reclen[1] = reclen & 0xff;
2540 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2541 struct lu_name *lname, struct lu_fid *pfid)
2543 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2544 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2545 fid_be_to_cpu(pfid, pfid);
2546 lname->ln_name = lee->lee_name;
2547 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2550 /** Add a record to the end of link ea buf */
2551 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2552 const struct lu_fid *pfid,
2553 const struct lu_name *lname)
2555 struct link_ea_header *leh;
2556 struct link_ea_entry *lee;
2559 if (lname == NULL || pfid == NULL)
2562 /* Make sure our buf is big enough for the new one */
2564 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2565 if (leh->leh_len + reclen > buf->lb_len) {
2566 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2571 lee = buf->lb_buf + leh->leh_len;
2572 reclen = mdd_lee_pack(lee, lname, pfid);
2573 leh->leh_len += reclen;
2574 leh->leh_reccount++;
2578 static int mdd_declare_links_add(const struct lu_env *env,
2579 struct mdd_object *mdd_obj,
2580 struct thandle *handle)
2584 /* XXX: max size? */
2585 rc = mdo_declare_xattr_set(env, mdd_obj,
2586 mdd_buf_get_const(env, NULL, 4096),
2587 XATTR_NAME_LINK, 0, handle);
2592 /* For pathologic linkers, we don't want to spend lots of time scanning the
2593 * link ea. Limit ourseleves to something reasonable; links not in the EA
2594 * can be looked up via (slower) parent lookup.
2596 #define LINKEA_MAX_COUNT 128
2598 static int mdd_links_add(const struct lu_env *env,
2599 struct mdd_object *mdd_obj,
2600 const struct lu_fid *pfid,
2601 const struct lu_name *lname,
2602 struct thandle *handle, int first)
2605 struct link_ea_header *leh;
2609 if (!mdd_linkea_enable)
2612 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2615 if (rc != -ENODATA) {
2616 CERROR("link_ea read failed %d "DFID"\n", rc,
2617 PFID(mdd_object_fid(mdd_obj)));
2620 /* empty EA; start one */
2621 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2622 if (buf->lb_buf == NULL)
2625 leh->leh_magic = LINK_EA_MAGIC;
2626 leh->leh_len = sizeof(struct link_ea_header);
2627 leh->leh_reccount = 0;
2631 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2634 rc = __mdd_links_add(env, buf, pfid, lname);
2639 rc = mdo_xattr_set(env, mdd_obj,
2640 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2641 XATTR_NAME_LINK, 0, handle,
2642 mdd_object_capa(env, mdd_obj));
2645 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2646 PFID(mdd_object_fid(mdd_obj)));
2648 CERROR("link_ea add failed %d "DFID"\n", rc,
2649 PFID(mdd_object_fid(mdd_obj)));
2652 if (buf->lb_len > OBD_ALLOC_BIG)
2653 /* if we vmalloced a large buffer drop it */
2659 static int mdd_links_rename(const struct lu_env *env,
2660 struct mdd_object *mdd_obj,
2661 const struct lu_fid *oldpfid,
2662 const struct lu_name *oldlname,
2663 const struct lu_fid *newpfid,
2664 const struct lu_name *newlname,
2665 struct thandle *handle)
2668 struct link_ea_header *leh;
2669 struct link_ea_entry *lee;
2670 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2671 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2677 if (!mdd_linkea_enable)
2680 if (mdd_obj->mod_flags & DEAD_OBJ)
2681 /* No more links, don't bother */
2684 buf = mdd_links_get(env, mdd_obj);
2688 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2689 rc, PFID(mdd_object_fid(mdd_obj)));
2691 CERROR("link_ea read failed %d "DFID"\n",
2692 rc, PFID(mdd_object_fid(mdd_obj)));
2696 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2698 /* Find the old record */
2699 for(count = 0; count < leh->leh_reccount; count++) {
2700 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2701 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2702 lu_fid_eq(tmpfid, oldpfid) &&
2703 (strncmp(tmpname->ln_name, oldlname->ln_name,
2704 tmpname->ln_namelen) == 0))
2706 lee = (struct link_ea_entry *)((char *)lee + reclen);
2708 if ((count + 1) > leh->leh_reccount) {
2709 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2710 oldlname->ln_namelen, oldlname->ln_name);
2711 GOTO(out, rc = -ENOENT);
2714 /* Remove the old record */
2715 leh->leh_reccount--;
2716 leh->leh_len -= reclen;
2717 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2720 /* If renaming, add the new record */
2721 if (newpfid != NULL) {
2722 /* if the add fails, we still delete the out-of-date old link */
2723 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2727 rc = mdo_xattr_set(env, mdd_obj,
2728 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2729 XATTR_NAME_LINK, 0, handle,
2730 mdd_object_capa(env, mdd_obj));
2736 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2737 oldlname->ln_namelen, oldlname->ln_name, rc,
2738 PFID(mdd_object_fid(mdd_obj)));
2740 if (buf->lb_len > OBD_ALLOC_BIG)
2741 /* if we vmalloced a large buffer drop it */
2747 const struct md_dir_operations mdd_dir_ops = {
2748 .mdo_is_subdir = mdd_is_subdir,
2749 .mdo_lookup = mdd_lookup,
2750 .mdo_create = mdd_create,
2751 .mdo_rename = mdd_rename,
2752 .mdo_link = mdd_link,
2753 .mdo_unlink = mdd_unlink,
2754 .mdo_create_data = mdd_create_data,