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);
1323 rc = mdd_attr_get_internal_locked(env, son, ma);
1325 /* update lov_objid data, must be before transaction stop! */
1327 mdd_lov_objid_update(mdd, lmm);
1330 mdd_trans_stop(env, mdd, rc, handle);
1332 /* Finish mdd_lov_create() stuff. */
1333 /* if no_create == 0 (not replay), we free lmm allocated by
1334 * mdd_lov_create() */
1335 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1339 /* Get fid from name and parent */
1341 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1342 const struct lu_name *lname, struct lu_fid* fid, int mask)
1344 const char *name = lname->ln_name;
1345 const struct dt_key *key = (const struct dt_key *)name;
1346 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1347 struct mdd_device *m = mdo2mdd(pobj);
1348 struct dt_object *dir = mdd_object_child(mdd_obj);
1352 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1355 rc = mdd_object_exists(mdd_obj);
1356 if (unlikely(rc == 0))
1358 else if (unlikely(rc < 0)) {
1359 CERROR("Object "DFID" locates on remote server\n",
1360 PFID(mdo2fid(mdd_obj)));
1364 /* The common filename length check. */
1365 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1366 RETURN(-ENAMETOOLONG);
1368 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1373 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1374 dt_try_as_dir(env, dir))) {
1376 rc = dir->do_index_ops->dio_lookup(env, dir,
1377 (struct dt_rec *)fid, key,
1378 mdd_object_capa(env, mdd_obj));
1389 int mdd_declare_object_initialize(const struct lu_env *env,
1390 struct mdd_object *child,
1391 struct lu_attr *attr,
1392 struct thandle *handle)
1396 rc = mdo_declare_attr_set(env, child, attr, handle);
1397 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1398 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1401 rc = mdo_declare_ref_add(env, child, handle);
1405 mdd_declare_links_add(env, child, handle);
1410 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1411 const struct lu_name *lname, struct mdd_object *child,
1412 struct lu_attr *attr, struct thandle *handle,
1413 const struct md_op_spec *spec)
1419 * Update attributes for child.
1422 * (1) the valid bits should be converted between Lustre and Linux;
1423 * (2) maybe, the child attributes should be set in OSD when creation.
1426 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1430 if (S_ISDIR(attr->la_mode)) {
1431 /* Add "." and ".." for newly created dir */
1432 mdo_ref_add(env, child, handle);
1433 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1434 dot, handle, BYPASS_CAPA);
1436 rc = __mdd_index_insert_only(env, child, pfid,
1440 mdo_ref_del(env, child, handle);
1443 mdd_links_add(env, child, pfid, lname, handle, 1);
1448 /* has not lock on pobj yet */
1449 static int mdd_create_sanity_check(const struct lu_env *env,
1450 struct md_object *pobj,
1451 struct lu_attr *pattr,
1452 const struct lu_name *lname,
1453 struct lu_attr *cattr,
1454 struct md_op_spec *spec)
1456 struct mdd_thread_info *info = mdd_env_info(env);
1457 struct lu_fid *fid = &info->mti_fid;
1458 struct mdd_object *obj = md2mdd_obj(pobj);
1459 struct mdd_device *m = mdo2mdd(pobj);
1460 int lookup = spec->sp_cr_lookup;
1465 if (mdd_is_dead_obj(obj))
1469 * In some cases this lookup is not needed - we know before if name
1470 * exists or not because MDT performs lookup for it.
1471 * name length check is done in lookup.
1475 * Check if the name already exist, though it will be checked in
1476 * _index_insert also, for avoiding rolling back if exists
1479 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1480 MAY_WRITE | MAY_EXEC);
1482 RETURN(rc ? : -EEXIST);
1485 * Check WRITE permission for the parent.
1486 * EXEC permission have been checked
1487 * when lookup before create already.
1489 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1496 if (pattr->la_mode & S_ISGID) {
1497 cattr->la_gid = pattr->la_gid;
1498 if (S_ISDIR(cattr->la_mode)) {
1499 cattr->la_mode |= S_ISGID;
1500 cattr->la_valid |= LA_MODE;
1504 switch (cattr->la_mode & S_IFMT) {
1506 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1508 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1509 RETURN(-ENAMETOOLONG);
1528 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1529 struct mdd_object *p, struct mdd_object *c,
1530 const struct lu_name *name,
1531 struct lu_attr *attr, int lmm_size,
1533 struct thandle *handle,
1534 const struct md_op_spec *spec)
1536 struct mdd_thread_info *info = mdd_env_info(env);
1537 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1540 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1544 #ifdef CONFIG_FS_POSIX_ACL
1545 if (got_def_acl > 0) {
1546 struct lu_buf *acl_buf;
1548 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1549 /* if dir, then can inherit default ACl */
1550 if (S_ISDIR(attr->la_mode)) {
1551 rc = mdo_declare_xattr_set(env, c, acl_buf,
1552 XATTR_NAME_ACL_DEFAULT,
1558 rc = mdo_declare_attr_set(env, c, &info->mti_pattr, handle);
1562 rc = mdo_declare_xattr_set(env, c, acl_buf,
1563 XATTR_NAME_ACL_ACCESS, 0, handle);
1569 if (S_ISDIR(attr->la_mode)) {
1570 rc = mdo_declare_ref_add(env, p, handle);
1575 rc = mdd_declare_object_initialize(env, c, attr, handle);
1579 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1580 name->ln_name, handle);
1584 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1589 if (S_ISLNK(attr->la_mode)) {
1590 rc = dt_declare_record_write(env, mdd_object_child(c),
1591 strlen(spec->u.sp_symname), 0,
1597 rc = mdo_declare_attr_set(env, p, attr, handle);
1601 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1605 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1613 * Create object and insert it into namespace.
1615 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1616 const struct lu_name *lname, struct md_object *child,
1617 struct md_op_spec *spec, struct md_attr* ma)
1619 struct mdd_thread_info *info = mdd_env_info(env);
1620 struct lu_attr *la = &info->mti_la_for_fix;
1621 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1622 struct mdd_object *son = md2mdd_obj(child);
1623 struct mdd_device *mdd = mdo2mdd(pobj);
1624 struct lu_attr *attr = &ma->ma_attr;
1625 struct lov_mds_md *lmm = NULL;
1626 struct thandle *handle;
1627 struct lu_attr *pattr = &info->mti_pattr;
1628 struct dynlock_handle *dlh;
1629 const char *name = lname->ln_name;
1630 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1631 int got_def_acl = 0;
1632 #ifdef HAVE_QUOTA_SUPPORT
1633 struct obd_device *obd = mdd->mdd_obd_dev;
1634 struct obd_export *exp = md_quota(env)->mq_exp;
1635 struct mds_obd *mds = &obd->u.mds;
1636 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1637 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1638 int quota_opc = 0, block_count = 0;
1639 int inode_pending[MAXQUOTAS] = { 0, 0 };
1640 int block_pending[MAXQUOTAS] = { 0, 0 };
1641 int parent_pending[MAXQUOTAS] = { 0, 0 };
1646 * Two operations have to be performed:
1648 * - an allocation of a new object (->do_create()), and
1650 * - an insertion into a parent index (->dio_insert()).
1652 * Due to locking, operation order is not important, when both are
1653 * successful, *but* error handling cases are quite different:
1655 * - if insertion is done first, and following object creation fails,
1656 * insertion has to be rolled back, but this operation might fail
1657 * also leaving us with dangling index entry.
1659 * - if creation is done first, is has to be undone if insertion
1660 * fails, leaving us with leaked space, which is neither good, nor
1663 * It seems that creation-first is simplest solution, but it is
1664 * sub-optimal in the frequent
1669 * case, because second mkdir is bound to create object, only to
1670 * destroy it immediately.
1672 * To avoid this follow local file systems that do double lookup:
1674 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1676 * 1. create (mdd_object_create_internal())
1678 * 2. insert (__mdd_index_insert(), lookup again)
1681 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1685 /* Sanity checks before big job. */
1686 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
1690 #ifdef HAVE_QUOTA_SUPPORT
1691 if (mds->mds_quota) {
1694 quota_opc = FSFILT_OP_CREATE;
1695 mdd_quota_wrapper(attr, qcids);
1696 mdd_quota_wrapper(pattr, qpids);
1697 /* get file quota for child */
1698 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1699 qcids, inode_pending, 1, NULL, 0, NULL,
1701 switch (attr->la_mode & S_IFMT) {
1710 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1711 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1715 /* get block quota for child and parent */
1717 lquota_chkquota(mds_quota_interface_ref, obd,
1718 exp, qcids, block_pending,
1720 LQUOTA_FLAGS_BLK, NULL, 0);
1722 lquota_chkquota(mds_quota_interface_ref, obd,
1723 exp, qpids, parent_pending, 1,
1724 NULL, LQUOTA_FLAGS_BLK, NULL,
1729 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1730 GOTO(out_pending, rc = -EINPROGRESS);
1733 * No RPC inside the transaction, so OST objects should be created at
1736 if (S_ISREG(attr->la_mode)) {
1737 lmm_size = ma->ma_lmm_size;
1738 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1741 GOTO(out_pending, rc);
1744 if (!S_ISLNK(attr->la_mode)) {
1745 struct lu_buf *acl_buf;
1747 acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
1748 sizeof(info->mti_xattr_buf));
1749 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1750 rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
1751 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1752 mdd_read_unlock(env, mdd_pobj);
1755 else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
1759 mdd_object_make_hint(env, mdd_pobj, son, attr);
1761 handle = mdd_trans_create(env, mdd);
1763 GOTO(out_free, rc = PTR_ERR(handle));
1765 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
1766 got_def_acl, lmm_size, handle, spec);
1770 rc = mdd_trans_start(env, mdd, handle);
1774 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1776 GOTO(out_trans, rc = -ENOMEM);
1778 mdd_write_lock(env, son, MOR_TGT_CHILD);
1779 rc = mdd_object_create_internal(env, mdd_pobj, son, attr, handle, spec);
1781 mdd_write_unlock(env, son);
1787 #ifdef CONFIG_FS_POSIX_ACL
1789 struct lu_buf *acl_buf;
1791 acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
1792 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1794 mdd_write_unlock(env, son);
1800 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1801 son, attr, handle, spec);
1802 mdd_write_unlock(env, son);
1805 * Object has no links, so it will be destroyed when last
1806 * reference is released. (XXX not now.)
1812 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1813 name, S_ISDIR(attr->la_mode), handle,
1814 mdd_object_capa(env, mdd_pobj));
1820 /* No need mdd_lsm_sanity_check here */
1821 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1823 CERROR("error on stripe info copy %d \n", rc);
1826 if (lmm && lmm_size > 0) {
1827 /* Set Lov here, do not get lmm again later */
1828 if (lmm_size > ma->ma_lmm_size) {
1829 /* Reply buffer is smaller, need bigger one */
1830 mdd_max_lmm_buffer(env, lmm_size);
1831 if (unlikely(info->mti_max_lmm == NULL))
1832 GOTO(cleanup, rc = -ENOMEM);
1833 ma->ma_lmm = info->mti_max_lmm;
1834 ma->ma_big_lmm_used = 1;
1836 memcpy(ma->ma_lmm, lmm, lmm_size);
1837 ma->ma_lmm_size = lmm_size;
1838 ma->ma_valid |= MA_LOV;
1841 if (S_ISLNK(attr->la_mode)) {
1842 struct md_ucred *uc = md_ucred(env);
1843 struct dt_object *dt = mdd_object_child(son);
1844 const char *target_name = spec->u.sp_symname;
1845 int sym_len = strlen(target_name);
1846 const struct lu_buf *buf;
1849 buf = mdd_buf_get_const(env, target_name, sym_len);
1850 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1851 mdd_object_capa(env, son),
1853 CFS_CAP_SYS_RESOURCE_MASK);
1858 GOTO(cleanup, rc = -EFAULT);
1862 la->la_valid = LA_CTIME | LA_MTIME;
1863 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1867 /* Return attr back. */
1868 rc = mdd_attr_get_internal_locked(env, son, ma);
1871 if (rc != 0 && created != 0) {
1874 if (inserted != 0) {
1875 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1876 S_ISDIR(attr->la_mode),
1877 handle, BYPASS_CAPA);
1882 mdd_write_lock(env, son, MOR_TGT_CHILD);
1883 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
1884 /* Drop the reference, no need to delete "."/"..",
1885 * because the object to be destroied directly. */
1886 rc2 = mdo_ref_del(env, son, handle);
1888 mdd_write_unlock(env, son);
1893 rc2 = mdo_ref_del(env, son, handle);
1895 mdd_write_unlock(env, son);
1899 mdo_destroy(env, son, handle);
1900 mdd_write_unlock(env, son);
1903 /* update lov_objid data, must be before transaction stop! */
1905 mdd_lov_objid_update(mdd, lmm);
1907 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1910 rc = mdd_changelog_ns_store(env, mdd,
1911 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1912 S_ISREG(attr->la_mode) ? CL_CREATE :
1913 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1914 0, son, mdd_pobj, lname, handle);
1916 mdd_trans_stop(env, mdd, rc, handle);
1918 /* finish lov_create stuff, free all temporary data */
1919 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1921 #ifdef HAVE_QUOTA_SUPPORT
1923 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1925 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1927 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1929 /* Trigger dqacq on the owner of child and parent. If failed,
1930 * the next call for lquota_chkquota will process it. */
1931 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1935 /* The child object shouldn't be cached anymore */
1937 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
1938 &child->mo_lu.lo_header->loh_flags);
1943 * Get locks on parents in proper order
1944 * RETURN: < 0 - error, rename_order if successful
1952 static int mdd_rename_order(const struct lu_env *env,
1953 struct mdd_device *mdd,
1954 struct mdd_object *src_pobj,
1955 struct mdd_object *tgt_pobj)
1957 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1961 if (src_pobj == tgt_pobj)
1962 RETURN(MDD_RN_SAME);
1964 /* compared the parent child relationship of src_p&tgt_p */
1965 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1967 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1970 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1983 /* has not mdd_write{read}_lock on any obj yet. */
1984 static int mdd_rename_sanity_check(const struct lu_env *env,
1985 struct mdd_object *src_pobj,
1986 struct mdd_object *tgt_pobj,
1987 struct mdd_object *sobj,
1988 struct mdd_object *tobj,
1989 struct lu_attr *so_attr,
1990 struct lu_attr *tg_attr)
1995 /* XXX: when get here, sobj must NOT be NULL,
1996 * the other case has been processed in cml_rename
1997 * before mdd_rename and enable MDS_PERM_BYPASS. */
2000 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
2004 /* XXX: when get here, "tobj == NULL" means tobj must
2005 * NOT exist (neither on remote MDS, such case has been
2006 * processed in cml_rename before mdd_rename and enable
2008 * So check may_create, but not check may_unlink. */
2010 rc = mdd_may_create(env, tgt_pobj, NULL,
2011 (src_pobj != tgt_pobj), 0);
2013 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
2014 (src_pobj != tgt_pobj), 1);
2016 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2017 S_ISDIR(so_attr->la_mode))
2018 rc = __mdd_may_link(env, tgt_pobj);
2023 static int mdd_declare_rename(const struct lu_env *env,
2024 struct mdd_device *mdd,
2025 struct mdd_object *mdd_spobj,
2026 struct mdd_object *mdd_tpobj,
2027 struct mdd_object *mdd_sobj,
2028 struct mdd_object *mdd_tobj,
2029 const struct lu_name *tname,
2030 const struct lu_name *sname,
2032 struct thandle *handle)
2040 /* name from source dir */
2041 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2045 /* .. from source child */
2046 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2047 /* source child can be directory,
2048 * counted by source dir's nlink */
2049 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2053 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2057 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2062 /* new target child can be directory,
2063 * counted by target dir's nlink */
2064 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2070 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2074 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2077 mdd_declare_links_add(env, mdd_sobj, handle);
2081 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2086 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2087 tname->ln_name, handle);
2091 /* name from target dir (old name), we declare it unconditionally
2092 * as mdd_rename() calls delete unconditionally as well. so just
2093 * to balance declarations vs calls to change ... */
2094 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2098 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2099 /* delete target child in target parent directory */
2100 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2104 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2105 /* target child can be directory,
2106 * delete "." reference in target child directory */
2107 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2111 /* delete ".." reference in target parent directory */
2112 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2117 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2121 mdd_declare_links_add(env, mdd_tobj, handle);
2125 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2130 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2137 /* src object can be remote that is why we use only fid and type of object */
2138 static int mdd_rename(const struct lu_env *env,
2139 struct md_object *src_pobj, struct md_object *tgt_pobj,
2140 const struct lu_fid *lf, const struct lu_name *lsname,
2141 struct md_object *tobj, const struct lu_name *ltname,
2144 const char *sname = lsname->ln_name;
2145 const char *tname = ltname->ln_name;
2146 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2147 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2148 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2149 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2150 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2151 struct mdd_device *mdd = mdo2mdd(src_pobj);
2152 struct mdd_object *mdd_sobj = NULL; /* source object */
2153 struct mdd_object *mdd_tobj = NULL;
2154 struct dynlock_handle *sdlh, *tdlh;
2155 struct thandle *handle;
2156 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2157 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2159 unsigned cl_flags = 0;
2162 #ifdef HAVE_QUOTA_SUPPORT
2163 struct obd_device *obd = mdd->mdd_obd_dev;
2164 struct obd_export *exp = md_quota(env)->mq_exp;
2165 struct mds_obd *mds = &obd->u.mds;
2166 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
2167 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
2168 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
2169 int quota_copc = 0, quota_popc = 0;
2170 int rec_pending[MAXQUOTAS] = { 0, 0 };
2175 mdd_tobj = md2mdd_obj(tobj);
2177 #ifdef HAVE_QUOTA_SUPPORT
2178 if (mds->mds_quota) {
2179 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2181 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2183 mdd_quota_wrapper(la_tmp, qspids);
2185 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2189 mdd_data_get(env, mdd_tpobj, &data);
2190 quota_popc = FSFILT_OP_LINK;
2191 mdd_quota_wrapper(la_tmp, qtpids);
2192 /* get block quota for target parent */
2193 lquota_chkquota(mds_quota_interface_ref,
2195 rec_pending, 1, NULL,
2203 mdd_sobj = mdd_object_find(env, mdd, lf);
2205 handle = mdd_trans_create(env, mdd);
2207 GOTO(out_pending, rc = PTR_ERR(handle));
2209 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2210 mdd_tobj, lsname, ltname, ma, handle);
2214 rc = mdd_trans_start(env, mdd, handle);
2218 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2219 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2221 GOTO(cleanup_unlocked, rc);
2223 /* Get locks in determined order */
2224 if (rc == MDD_RN_SAME) {
2225 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2226 sname, MOR_SRC_PARENT);
2227 /* check hashes to determine do we need one lock or two */
2228 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2229 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2233 } else if (rc == MDD_RN_SRCTGT) {
2234 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2235 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2237 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2238 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2240 if (sdlh == NULL || tdlh == NULL)
2241 GOTO(cleanup, rc = -ENOMEM);
2243 rc = mdd_la_get(env, mdd_sobj, so_attr,
2244 mdd_object_capa(env, mdd_sobj));
2249 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2250 mdd_object_capa(env, mdd_tobj));
2255 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2256 mdd_tobj, so_attr, tg_attr);
2260 is_dir = S_ISDIR(so_attr->la_mode);
2262 /* Remove source name from source directory */
2263 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2264 mdd_object_capa(env, mdd_spobj));
2268 /* "mv dir1 dir2" needs "dir1/.." link update */
2269 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2270 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2271 mdd_object_capa(env, mdd_sobj));
2273 GOTO(fixup_spobj2, rc);
2275 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2276 handle, mdd_object_capa(env, mdd_sobj));
2278 GOTO(fixup_spobj, rc);
2281 /* Remove target name from target directory
2282 * Here tobj can be remote one, so we do index_delete unconditionally
2283 * and -ENOENT is allowed.
2285 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2286 mdd_object_capa(env, mdd_tpobj));
2289 /* tname might been renamed to something else */
2290 GOTO(fixup_spobj, rc);
2293 GOTO(fixup_spobj, rc);
2296 /* Insert new fid with target name into target dir */
2297 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2298 mdd_object_capa(env, mdd_tpobj));
2300 GOTO(fixup_tpobj, rc);
2302 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2303 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2305 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2307 la->la_valid = LA_CTIME;
2308 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2310 GOTO(fixup_tpobj, rc);
2313 /* Remove old target object
2314 * For tobj is remote case cmm layer has processed
2315 * and set tobj to NULL then. So when tobj is NOT NULL,
2316 * it must be local one.
2318 if (tobj && mdd_object_exists(mdd_tobj)) {
2319 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2320 if (mdd_is_dead_obj(mdd_tobj)) {
2321 mdd_write_unlock(env, mdd_tobj);
2322 /* shld not be dead, something is wrong */
2323 CERROR("tobj is dead, something is wrong\n");
2327 mdo_ref_del(env, mdd_tobj, handle);
2329 /* Remove dot reference. */
2330 if (S_ISDIR(tg_attr->la_mode))
2331 mdo_ref_del(env, mdd_tobj, handle);
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);
2339 la->la_valid = LA_CTIME;
2340 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2342 GOTO(fixup_tpobj, rc);
2344 /* XXX: this transfer to ma will be removed with LOD/OSP */
2345 ma->ma_attr = *tg_attr;
2346 ma->ma_valid |= MA_INODE;
2347 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2348 mdd_write_unlock(env, mdd_tobj);
2350 GOTO(fixup_tpobj, rc);
2352 /* fetch updated nlink */
2353 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2354 mdd_object_capa(env, mdd_tobj));
2356 GOTO(fixup_tpobj, rc);
2357 /* XXX: this transfer to ma will be removed with LOD/OSP */
2358 ma->ma_attr = *tg_attr;
2359 ma->ma_valid |= MA_INODE;
2361 if (so_attr->la_nlink == 0) {
2362 cl_flags |= CLF_RENAME_LAST;
2363 #ifdef HAVE_QUOTA_SUPPORT
2364 if (mds->mds_quota && mdd_tobj->mod_count == 0) {
2365 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2366 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2372 la->la_valid = LA_CTIME | LA_MTIME;
2373 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2375 GOTO(fixup_tpobj, rc);
2377 if (mdd_spobj != mdd_tpobj) {
2378 la->la_valid = LA_CTIME | LA_MTIME;
2379 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2383 if (rc == 0 && mdd_sobj) {
2384 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2385 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2386 mdo2fid(mdd_tpobj), ltname, handle);
2388 /* Old files might not have EA entry */
2389 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2391 mdd_write_unlock(env, mdd_sobj);
2392 /* We don't fail the transaction if the link ea can't be
2393 updated -- fid2path will use alternate lookup method. */
2401 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2404 CWARN("tp obj fix error %d\n",rc2);
2406 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2407 !mdd_is_dead_obj(mdd_tobj)) {
2408 rc2 = __mdd_index_insert(env, mdd_tpobj,
2409 mdo2fid(mdd_tobj), tname,
2414 CWARN("tp obj fix error %d\n",rc2);
2419 if (rc && is_dir && mdd_sobj) {
2420 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2424 CWARN("sp obj dotdot delete error %d\n",rc2);
2427 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2428 dotdot, handle, BYPASS_CAPA);
2430 CWARN("sp obj dotdot insert error %d\n",rc2);
2435 rc2 = __mdd_index_insert(env, mdd_spobj,
2436 lf, sname, is_dir, handle, BYPASS_CAPA);
2438 CWARN("sp obj fix error %d\n",rc2);
2441 if (likely(tdlh) && sdlh != tdlh)
2442 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2444 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2447 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2448 mdd_tobj, tpobj_fid, lf,
2449 spobj_fid, ltname, lsname,
2453 mdd_trans_stop(env, mdd, rc, handle);
2455 mdd_object_put(env, mdd_sobj);
2457 #ifdef HAVE_QUOTA_SUPPORT
2458 if (mds->mds_quota) {
2460 lquota_pending_commit(mds_quota_interface_ref, obd,
2461 qtpids, rec_pending, 1);
2464 /* Trigger dqrel on the source owner of parent.
2465 * If failed, the next call for lquota_chkquota will
2467 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2468 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2470 /* Trigger dqrel on the target owner of child.
2471 * If failed, the next call for lquota_chkquota
2472 * will process it. */
2473 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2474 qtpids, rc, quota_copc);
2481 /** enable/disable storing of hardlink info */
2482 int mdd_linkea_enable = 1;
2483 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2484 "record hardlink info in EAs");
2486 /** Read the link EA into a temp buffer.
2487 * Uses the name_buf since it is generally large.
2488 * \retval IS_ERR err
2489 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2491 struct lu_buf *mdd_links_get(const struct lu_env *env,
2492 struct mdd_object *mdd_obj)
2495 struct lustre_capa *capa;
2496 struct link_ea_header *leh;
2499 /* First try a small buf */
2500 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2501 if (buf->lb_buf == NULL)
2502 return ERR_PTR(-ENOMEM);
2504 if (!mdd_object_exists(mdd_obj))
2505 return ERR_PTR(-ENODATA);
2507 capa = mdd_object_capa(env, mdd_obj);
2508 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2509 if (rc == -ERANGE) {
2510 /* Buf was too small, figure out what we need. */
2512 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2515 buf = mdd_buf_alloc(env, rc);
2516 if (buf->lb_buf == NULL)
2517 return ERR_PTR(-ENOMEM);
2518 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2524 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2525 leh->leh_magic = LINK_EA_MAGIC;
2526 leh->leh_reccount = __swab32(leh->leh_reccount);
2527 leh->leh_len = __swab64(leh->leh_len);
2528 /* entries are swabbed by mdd_lee_unpack */
2530 if (leh->leh_magic != LINK_EA_MAGIC)
2531 return ERR_PTR(-EINVAL);
2532 if (leh->leh_reccount == 0)
2533 return ERR_PTR(-ENODATA);
2538 /** Pack a link_ea_entry.
2539 * All elements are stored as chars to avoid alignment issues.
2540 * Numbers are always big-endian
2541 * \retval record length
2543 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2544 const struct lu_fid *pfid)
2546 struct lu_fid tmpfid;
2549 fid_cpu_to_be(&tmpfid, pfid);
2550 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2551 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2552 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2554 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2555 lee->lee_reclen[1] = reclen & 0xff;
2559 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2560 struct lu_name *lname, struct lu_fid *pfid)
2562 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2563 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2564 fid_be_to_cpu(pfid, pfid);
2565 lname->ln_name = lee->lee_name;
2566 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2569 /** Add a record to the end of link ea buf */
2570 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2571 const struct lu_fid *pfid,
2572 const struct lu_name *lname)
2574 struct link_ea_header *leh;
2575 struct link_ea_entry *lee;
2578 if (lname == NULL || pfid == NULL)
2581 /* Make sure our buf is big enough for the new one */
2583 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2584 if (leh->leh_len + reclen > buf->lb_len) {
2585 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2590 lee = buf->lb_buf + leh->leh_len;
2591 reclen = mdd_lee_pack(lee, lname, pfid);
2592 leh->leh_len += reclen;
2593 leh->leh_reccount++;
2597 static int mdd_declare_links_add(const struct lu_env *env,
2598 struct mdd_object *mdd_obj,
2599 struct thandle *handle)
2603 /* XXX: max size? */
2604 rc = mdo_declare_xattr_set(env, mdd_obj,
2605 mdd_buf_get_const(env, NULL, 4096),
2606 XATTR_NAME_LINK, 0, handle);
2611 /* For pathologic linkers, we don't want to spend lots of time scanning the
2612 * link ea. Limit ourseleves to something reasonable; links not in the EA
2613 * can be looked up via (slower) parent lookup.
2615 #define LINKEA_MAX_COUNT 128
2617 static int mdd_links_add(const struct lu_env *env,
2618 struct mdd_object *mdd_obj,
2619 const struct lu_fid *pfid,
2620 const struct lu_name *lname,
2621 struct thandle *handle, int first)
2624 struct link_ea_header *leh;
2628 if (!mdd_linkea_enable)
2631 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2634 if (rc != -ENODATA) {
2635 CERROR("link_ea read failed %d "DFID"\n", rc,
2636 PFID(mdd_object_fid(mdd_obj)));
2639 /* empty EA; start one */
2640 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2641 if (buf->lb_buf == NULL)
2644 leh->leh_magic = LINK_EA_MAGIC;
2645 leh->leh_len = sizeof(struct link_ea_header);
2646 leh->leh_reccount = 0;
2650 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2653 rc = __mdd_links_add(env, buf, pfid, lname);
2658 rc = mdo_xattr_set(env, mdd_obj,
2659 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2660 XATTR_NAME_LINK, 0, handle,
2661 mdd_object_capa(env, mdd_obj));
2664 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2665 PFID(mdd_object_fid(mdd_obj)));
2667 CERROR("link_ea add failed %d "DFID"\n", rc,
2668 PFID(mdd_object_fid(mdd_obj)));
2671 if (buf->lb_len > OBD_ALLOC_BIG)
2672 /* if we vmalloced a large buffer drop it */
2678 static int mdd_links_rename(const struct lu_env *env,
2679 struct mdd_object *mdd_obj,
2680 const struct lu_fid *oldpfid,
2681 const struct lu_name *oldlname,
2682 const struct lu_fid *newpfid,
2683 const struct lu_name *newlname,
2684 struct thandle *handle)
2687 struct link_ea_header *leh;
2688 struct link_ea_entry *lee;
2689 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2690 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2696 if (!mdd_linkea_enable)
2699 if (mdd_obj->mod_flags & DEAD_OBJ)
2700 /* No more links, don't bother */
2703 buf = mdd_links_get(env, mdd_obj);
2707 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2708 rc, PFID(mdd_object_fid(mdd_obj)));
2710 CERROR("link_ea read failed %d "DFID"\n",
2711 rc, PFID(mdd_object_fid(mdd_obj)));
2715 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2717 /* Find the old record */
2718 for(count = 0; count < leh->leh_reccount; count++) {
2719 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2720 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2721 lu_fid_eq(tmpfid, oldpfid) &&
2722 (strncmp(tmpname->ln_name, oldlname->ln_name,
2723 tmpname->ln_namelen) == 0))
2725 lee = (struct link_ea_entry *)((char *)lee + reclen);
2727 if ((count + 1) > leh->leh_reccount) {
2728 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2729 oldlname->ln_namelen, oldlname->ln_name);
2730 GOTO(out, rc = -ENOENT);
2733 /* Remove the old record */
2734 leh->leh_reccount--;
2735 leh->leh_len -= reclen;
2736 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2739 /* If renaming, add the new record */
2740 if (newpfid != NULL) {
2741 /* if the add fails, we still delete the out-of-date old link */
2742 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2746 rc = mdo_xattr_set(env, mdd_obj,
2747 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2748 XATTR_NAME_LINK, 0, handle,
2749 mdd_object_capa(env, mdd_obj));
2755 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2756 oldlname->ln_namelen, oldlname->ln_name, rc,
2757 PFID(mdd_object_fid(mdd_obj)));
2759 if (buf->lb_len > OBD_ALLOC_BIG)
2760 /* if we vmalloced a large buffer drop it */
2766 const struct md_dir_operations mdd_dir_ops = {
2767 .mdo_is_subdir = mdd_is_subdir,
2768 .mdo_lookup = mdd_lookup,
2769 .mdo_create = mdd_create,
2770 .mdo_rename = mdd_rename,
2771 .mdo_link = mdd_link,
2772 .mdo_unlink = mdd_unlink,
2773 .mdo_create_data = mdd_create_data,