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;
841 handle = mdd_trans_create(env, mdd);
843 GOTO(out_pending, rc = PTR_ERR(handle));
845 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
849 rc = mdd_trans_start(env, mdd, handle);
853 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
855 GOTO(out_trans, rc = -ENOMEM);
856 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
858 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
860 GOTO(out_unlock, rc);
862 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
864 mdd_object_capa(env, mdd_tobj));
866 GOTO(out_unlock, rc);
868 rc = mdo_ref_add(env, mdd_sobj, handle);
870 __mdd_index_delete_only(env, mdd_tobj, name, handle,
871 mdd_object_capa(env, mdd_tobj));
872 GOTO(out_unlock, rc);
875 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
876 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
878 la->la_valid = LA_CTIME | LA_MTIME;
879 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
881 GOTO(out_unlock, rc);
883 la->la_valid = LA_CTIME;
884 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
886 mdd_links_add(env, mdd_sobj,
887 mdo2fid(mdd_tobj), lname, handle, 0);
892 mdd_write_unlock(env, mdd_sobj);
893 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
896 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
897 mdd_tobj, lname, handle);
899 mdd_trans_stop(env, mdd, rc, handle);
904 int mdd_declare_finish_unlink(const struct lu_env *env,
905 struct mdd_object *obj,
907 struct thandle *handle)
911 rc = orph_declare_index_insert(env, obj, handle);
915 return mdo_declare_destroy(env, obj, handle);
918 /* caller should take a lock before calling */
919 int mdd_finish_unlink(const struct lu_env *env,
920 struct mdd_object *obj, struct md_attr *ma,
924 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
927 LASSERT(mdd_write_locked(env, obj) != 0);
929 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
930 obj->mod_flags |= DEAD_OBJ;
931 /* add new orphan and the object
932 * will be deleted during mdd_close() */
933 if (obj->mod_count) {
934 rc = __mdd_orphan_add(env, obj, th);
936 CDEBUG(D_HA, "Object "DFID" is inserted into "
937 "orphan list, open count = %d\n",
938 PFID(mdd_object_fid(obj)),
941 CERROR("Object "DFID" fail to be an orphan, "
942 "open count = %d, maybe cause failed "
944 PFID(mdd_object_fid(obj)),
947 rc = mdo_destroy(env, obj, th);
956 * has mdd_write_lock on cobj already, but not on pobj yet
958 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
959 struct mdd_object *cobj, struct lu_attr *cattr)
964 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
969 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
970 struct mdd_object *p, struct mdd_object *c,
971 const struct lu_name *name, struct md_attr *ma,
972 struct thandle *handle)
976 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
980 rc = mdo_declare_ref_del(env, p, handle);
984 rc = mdo_declare_ref_del(env, c, handle);
988 rc = mdo_declare_ref_del(env, c, handle);
992 rc = mdo_declare_attr_set(env, p, NULL, handle);
996 rc = mdo_declare_attr_set(env, c, NULL, handle);
1000 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1004 rc = mdd_declare_links_add(env, c, handle);
1008 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1013 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1014 struct md_object *cobj, const struct lu_name *lname,
1017 const char *name = lname->ln_name;
1018 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1019 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1020 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1021 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
1022 struct mdd_device *mdd = mdo2mdd(pobj);
1023 struct dynlock_handle *dlh;
1024 struct thandle *handle;
1028 if (mdd_object_exists(mdd_cobj) <= 0)
1031 handle = mdd_trans_create(env, mdd);
1033 RETURN(PTR_ERR(handle));
1035 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1040 rc = mdd_trans_start(env, mdd, handle);
1044 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1046 GOTO(stop, rc = -ENOMEM);
1047 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1050 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1054 is_dir = S_ISDIR(cattr->la_mode);
1056 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1060 rc = mdo_ref_del(env, mdd_cobj, handle);
1062 __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj),
1064 mdd_object_capa(env, mdd_pobj));
1068 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1069 mdd_object_capa(env, mdd_pobj));
1075 mdo_ref_del(env, mdd_cobj, handle);
1077 /* fetch updated nlink */
1078 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1082 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1083 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1085 la->la_valid = LA_CTIME | LA_MTIME;
1086 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1090 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1091 /* update ctime of an unlinked file only if it is still
1092 * opened or a link still exists */
1093 la->la_valid = LA_CTIME;
1094 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1099 /* XXX: this transfer to ma will be removed with LOD/OSP */
1100 ma->ma_attr = *cattr;
1101 ma->ma_valid |= MA_INODE;
1102 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1104 /* fetch updated nlink */
1106 rc = mdd_la_get(env, mdd_cobj, cattr,
1107 mdd_object_capa(env, mdd_cobj));
1110 /* old files may not have link ea; ignore errors */
1111 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
1112 lname, NULL, NULL, handle);
1114 /* if object is removed then we can't get its attrs, use last get */
1115 if (cattr->la_nlink == 0) {
1116 ma->ma_attr = *cattr;
1117 ma->ma_valid |= MA_INODE;
1121 mdd_write_unlock(env, mdd_cobj);
1122 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1126 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1127 if ((ma->ma_valid & MA_HSM) &&
1128 (ma->ma_hsm.mh_flags & HS_EXISTS))
1129 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1131 rc = mdd_changelog_ns_store(env, mdd,
1132 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1133 mdd_cobj, mdd_pobj, lname, handle);
1137 mdd_trans_stop(env, mdd, rc, handle);
1143 * The permission has been checked when obj created, no need check again.
1145 static int mdd_cd_sanity_check(const struct lu_env *env,
1146 struct mdd_object *obj)
1151 if (!obj || mdd_is_dead_obj(obj))
1158 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1159 struct md_object *cobj, const struct md_op_spec *spec,
1162 struct mdd_device *mdd = mdo2mdd(cobj);
1163 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1164 struct mdd_object *son = md2mdd_obj(cobj);
1165 struct thandle *handle;
1166 const struct lu_buf *buf;
1167 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1171 /* do not let users to create stripes via .lustre/
1172 * mdd_obf_setup() sets IMMUTE_OBJ on this directory */
1173 if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ)
1176 rc = mdd_cd_sanity_check(env, son);
1180 if (!md_should_create(spec->sp_cr_flags))
1184 * there are following use cases for this function:
1185 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1186 * striping can be specified or not
1189 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1193 /* calling ->ah_make_hint() is used to transfer information from parent */
1194 mdd_object_make_hint(env, mdd_pobj, son, attr);
1196 handle = mdd_trans_create(env, mdd);
1198 GOTO(out_free, rc = PTR_ERR(handle));
1201 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1202 * Should this be fixed?
1204 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, no_create %u\n",
1205 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1206 spec->sp_cr_flags, spec->no_create);
1208 if (spec->no_create) {
1210 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1211 spec->u.sp_ea.eadatalen);
1212 } else if (!(spec->sp_cr_flags & MDS_OPEN_HAS_OBJS)) {
1213 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1215 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1216 spec->u.sp_ea.eadatalen);
1221 /* MDS_OPEN_HAS_OBJS is not used anymore ? */
1225 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1226 XATTR_NAME_LOV, 0, handle);
1230 rc = mdd_trans_start(env, mdd, handle);
1234 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1235 0, handle, mdd_object_capa(env, son));
1237 mdd_trans_stop(env, mdd, rc, handle);
1242 /* Get fid from name and parent */
1244 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1245 const struct lu_name *lname, struct lu_fid* fid, int mask)
1247 const char *name = lname->ln_name;
1248 const struct dt_key *key = (const struct dt_key *)name;
1249 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1250 struct mdd_device *m = mdo2mdd(pobj);
1251 struct dt_object *dir = mdd_object_child(mdd_obj);
1255 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1258 rc = mdd_object_exists(mdd_obj);
1259 if (unlikely(rc == 0))
1261 else if (unlikely(rc < 0)) {
1262 CERROR("Object "DFID" locates on remote server\n",
1263 PFID(mdo2fid(mdd_obj)));
1267 /* The common filename length check. */
1268 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1269 RETURN(-ENAMETOOLONG);
1271 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1276 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1277 dt_try_as_dir(env, dir))) {
1279 rc = dir->do_index_ops->dio_lookup(env, dir,
1280 (struct dt_rec *)fid, key,
1281 mdd_object_capa(env, mdd_obj));
1292 int mdd_declare_object_initialize(const struct lu_env *env,
1293 struct mdd_object *child,
1294 struct lu_attr *attr,
1295 struct thandle *handle)
1299 rc = mdo_declare_attr_set(env, child, attr, handle);
1300 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1301 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1304 rc = mdo_declare_ref_add(env, child, handle);
1308 mdd_declare_links_add(env, child, handle);
1313 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1314 const struct lu_name *lname, struct mdd_object *child,
1315 struct lu_attr *attr, struct thandle *handle,
1316 const struct md_op_spec *spec)
1322 * Update attributes for child.
1325 * (1) the valid bits should be converted between Lustre and Linux;
1326 * (2) maybe, the child attributes should be set in OSD when creation.
1329 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1333 if (S_ISDIR(attr->la_mode)) {
1334 /* Add "." and ".." for newly created dir */
1335 mdo_ref_add(env, child, handle);
1336 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1337 dot, handle, BYPASS_CAPA);
1339 rc = __mdd_index_insert_only(env, child, pfid,
1343 mdo_ref_del(env, child, handle);
1346 mdd_links_add(env, child, pfid, lname, handle, 1);
1351 /* has not lock on pobj yet */
1352 static int mdd_create_sanity_check(const struct lu_env *env,
1353 struct md_object *pobj,
1354 struct lu_attr *pattr,
1355 const struct lu_name *lname,
1356 struct lu_attr *cattr,
1357 struct md_op_spec *spec)
1359 struct mdd_thread_info *info = mdd_env_info(env);
1360 struct lu_fid *fid = &info->mti_fid;
1361 struct mdd_object *obj = md2mdd_obj(pobj);
1362 struct mdd_device *m = mdo2mdd(pobj);
1363 int lookup = spec->sp_cr_lookup;
1368 if (mdd_is_dead_obj(obj))
1372 * In some cases this lookup is not needed - we know before if name
1373 * exists or not because MDT performs lookup for it.
1374 * name length check is done in lookup.
1378 * Check if the name already exist, though it will be checked in
1379 * _index_insert also, for avoiding rolling back if exists
1382 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1383 MAY_WRITE | MAY_EXEC);
1385 RETURN(rc ? : -EEXIST);
1388 * Check WRITE permission for the parent.
1389 * EXEC permission have been checked
1390 * when lookup before create already.
1392 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1399 if (pattr->la_mode & S_ISGID) {
1400 cattr->la_gid = pattr->la_gid;
1401 if (S_ISDIR(cattr->la_mode)) {
1402 cattr->la_mode |= S_ISGID;
1403 cattr->la_valid |= LA_MODE;
1407 switch (cattr->la_mode & S_IFMT) {
1409 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1411 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1412 RETURN(-ENAMETOOLONG);
1431 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1432 struct mdd_object *p, struct mdd_object *c,
1433 const struct lu_name *name,
1434 struct lu_attr *attr,
1436 struct thandle *handle,
1437 const struct md_op_spec *spec)
1439 struct mdd_thread_info *info = mdd_env_info(env);
1442 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1446 #ifdef CONFIG_FS_POSIX_ACL
1447 if (got_def_acl > 0) {
1448 struct lu_buf *acl_buf;
1450 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1451 /* if dir, then can inherit default ACl */
1452 if (S_ISDIR(attr->la_mode)) {
1453 rc = mdo_declare_xattr_set(env, c, acl_buf,
1454 XATTR_NAME_ACL_DEFAULT,
1460 rc = mdo_declare_attr_set(env, c, &info->mti_pattr, handle);
1464 rc = mdo_declare_xattr_set(env, c, acl_buf,
1465 XATTR_NAME_ACL_ACCESS, 0, handle);
1471 if (S_ISDIR(attr->la_mode)) {
1472 rc = mdo_declare_ref_add(env, p, handle);
1477 rc = mdd_declare_object_initialize(env, c, attr, handle);
1481 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1482 name->ln_name, handle);
1486 /* replay case, create LOV EA from client data */
1487 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1488 const struct lu_buf *buf;
1490 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1491 spec->u.sp_ea.eadatalen);
1492 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1498 if (S_ISLNK(attr->la_mode)) {
1499 rc = dt_declare_record_write(env, mdd_object_child(c),
1500 strlen(spec->u.sp_symname), 0,
1506 rc = mdo_declare_attr_set(env, p, attr, handle);
1510 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1520 * Create object and insert it into namespace.
1522 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1523 const struct lu_name *lname, struct md_object *child,
1524 struct md_op_spec *spec, struct md_attr* ma)
1526 struct mdd_thread_info *info = mdd_env_info(env);
1527 struct lu_attr *la = &info->mti_la_for_fix;
1528 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1529 struct mdd_object *son = md2mdd_obj(child);
1530 struct mdd_device *mdd = mdo2mdd(pobj);
1531 struct lu_attr *attr = &ma->ma_attr;
1532 struct thandle *handle;
1533 struct lu_attr *pattr = &info->mti_pattr;
1534 struct dynlock_handle *dlh;
1535 const char *name = lname->ln_name;
1536 int rc, created = 0, initialized = 0, inserted = 0;
1537 int got_def_acl = 0;
1541 * Two operations have to be performed:
1543 * - an allocation of a new object (->do_create()), and
1545 * - an insertion into a parent index (->dio_insert()).
1547 * Due to locking, operation order is not important, when both are
1548 * successful, *but* error handling cases are quite different:
1550 * - if insertion is done first, and following object creation fails,
1551 * insertion has to be rolled back, but this operation might fail
1552 * also leaving us with dangling index entry.
1554 * - if creation is done first, is has to be undone if insertion
1555 * fails, leaving us with leaked space, which is neither good, nor
1558 * It seems that creation-first is simplest solution, but it is
1559 * sub-optimal in the frequent
1564 * case, because second mkdir is bound to create object, only to
1565 * destroy it immediately.
1567 * To avoid this follow local file systems that do double lookup:
1569 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1571 * 1. create (mdd_object_create_internal())
1573 * 2. insert (__mdd_index_insert(), lookup again)
1576 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1580 /* Sanity checks before big job. */
1581 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
1585 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1586 GOTO(out_free, rc = -EINPROGRESS);
1588 if (!S_ISLNK(attr->la_mode)) {
1589 struct lu_buf *acl_buf;
1591 acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
1592 sizeof(info->mti_xattr_buf));
1593 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1594 rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
1595 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1596 mdd_read_unlock(env, mdd_pobj);
1599 else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
1603 mdd_object_make_hint(env, mdd_pobj, son, attr);
1605 handle = mdd_trans_create(env, mdd);
1607 GOTO(out_free, rc = PTR_ERR(handle));
1609 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
1610 got_def_acl, handle, spec);
1614 rc = mdd_trans_start(env, mdd, handle);
1618 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1620 GOTO(out_trans, rc = -ENOMEM);
1622 mdd_write_lock(env, son, MOR_TGT_CHILD);
1623 rc = mdd_object_create_internal(env, mdd_pobj, son, attr, handle, spec);
1625 mdd_write_unlock(env, son);
1631 #ifdef CONFIG_FS_POSIX_ACL
1633 struct lu_buf *acl_buf;
1635 acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
1636 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1638 mdd_write_unlock(env, son);
1644 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1645 son, attr, handle, spec);
1648 * in case of replay we just set LOVEA provided by the client
1649 * XXX: I think it would be interesting to try "old" way where
1650 * MDT calls this xattr_set(LOV) in a different transaction.
1651 * probably this way we code can be made better.
1654 (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
1655 const struct lu_buf *buf;
1657 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1658 spec->u.sp_ea.eadatalen);
1659 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
1662 mdd_write_unlock(env, son);
1665 * Object has no links, so it will be destroyed when last
1666 * reference is released. (XXX not now.)
1672 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1673 name, S_ISDIR(attr->la_mode), handle,
1674 mdd_object_capa(env, mdd_pobj));
1680 if (S_ISLNK(attr->la_mode)) {
1681 struct md_ucred *uc = md_ucred(env);
1682 struct dt_object *dt = mdd_object_child(son);
1683 const char *target_name = spec->u.sp_symname;
1684 int sym_len = strlen(target_name);
1685 const struct lu_buf *buf;
1688 buf = mdd_buf_get_const(env, target_name, sym_len);
1689 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1690 mdd_object_capa(env, son),
1692 CFS_CAP_SYS_RESOURCE_MASK);
1697 GOTO(cleanup, rc = -EFAULT);
1701 la->la_valid = LA_CTIME | LA_MTIME;
1702 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1708 if (rc != 0 && created != 0) {
1711 if (inserted != 0) {
1712 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1713 S_ISDIR(attr->la_mode),
1714 handle, BYPASS_CAPA);
1719 mdd_write_lock(env, son, MOR_TGT_CHILD);
1720 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
1721 /* Drop the reference, no need to delete "."/"..",
1722 * because the object to be destroied directly. */
1723 rc2 = mdo_ref_del(env, son, handle);
1725 mdd_write_unlock(env, son);
1730 rc2 = mdo_ref_del(env, son, handle);
1732 mdd_write_unlock(env, son);
1736 mdo_destroy(env, son, handle);
1737 mdd_write_unlock(env, son);
1740 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1743 rc = mdd_changelog_ns_store(env, mdd,
1744 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1745 S_ISREG(attr->la_mode) ? CL_CREATE :
1746 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1747 0, son, mdd_pobj, lname, handle);
1749 mdd_trans_stop(env, mdd, rc, handle);
1751 /* The child object shouldn't be cached anymore */
1753 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
1754 &child->mo_lu.lo_header->loh_flags);
1759 * Get locks on parents in proper order
1760 * RETURN: < 0 - error, rename_order if successful
1768 static int mdd_rename_order(const struct lu_env *env,
1769 struct mdd_device *mdd,
1770 struct mdd_object *src_pobj,
1771 struct mdd_object *tgt_pobj)
1773 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1777 if (src_pobj == tgt_pobj)
1778 RETURN(MDD_RN_SAME);
1780 /* compared the parent child relationship of src_p&tgt_p */
1781 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1783 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1786 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1799 /* has not mdd_write{read}_lock on any obj yet. */
1800 static int mdd_rename_sanity_check(const struct lu_env *env,
1801 struct mdd_object *src_pobj,
1802 struct mdd_object *tgt_pobj,
1803 struct mdd_object *sobj,
1804 struct mdd_object *tobj,
1805 struct lu_attr *so_attr,
1806 struct lu_attr *tg_attr)
1811 /* XXX: when get here, sobj must NOT be NULL,
1812 * the other case has been processed in cml_rename
1813 * before mdd_rename and enable MDS_PERM_BYPASS. */
1816 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
1820 /* XXX: when get here, "tobj == NULL" means tobj must
1821 * NOT exist (neither on remote MDS, such case has been
1822 * processed in cml_rename before mdd_rename and enable
1824 * So check may_create, but not check may_unlink. */
1826 rc = mdd_may_create(env, tgt_pobj, NULL,
1827 (src_pobj != tgt_pobj), 0);
1829 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
1830 (src_pobj != tgt_pobj), 1);
1832 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1833 S_ISDIR(so_attr->la_mode))
1834 rc = __mdd_may_link(env, tgt_pobj);
1839 static int mdd_declare_rename(const struct lu_env *env,
1840 struct mdd_device *mdd,
1841 struct mdd_object *mdd_spobj,
1842 struct mdd_object *mdd_tpobj,
1843 struct mdd_object *mdd_sobj,
1844 struct mdd_object *mdd_tobj,
1845 const struct lu_name *tname,
1846 const struct lu_name *sname,
1848 struct thandle *handle)
1856 /* name from source dir */
1857 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
1861 /* .. from source child */
1862 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
1863 /* source child can be directory,
1864 * counted by source dir's nlink */
1865 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
1869 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
1873 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
1878 /* new target child can be directory,
1879 * counted by target dir's nlink */
1880 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
1886 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
1890 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
1893 mdd_declare_links_add(env, mdd_sobj, handle);
1897 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
1902 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
1903 tname->ln_name, handle);
1907 /* name from target dir (old name), we declare it unconditionally
1908 * as mdd_rename() calls delete unconditionally as well. so just
1909 * to balance declarations vs calls to change ... */
1910 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
1914 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
1915 /* delete target child in target parent directory */
1916 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
1920 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
1921 /* target child can be directory,
1922 * delete "." reference in target child directory */
1923 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
1927 /* delete ".." reference in target parent directory */
1928 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
1933 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
1937 mdd_declare_links_add(env, mdd_tobj, handle);
1941 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
1946 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
1953 /* src object can be remote that is why we use only fid and type of object */
1954 static int mdd_rename(const struct lu_env *env,
1955 struct md_object *src_pobj, struct md_object *tgt_pobj,
1956 const struct lu_fid *lf, const struct lu_name *lsname,
1957 struct md_object *tobj, const struct lu_name *ltname,
1960 const char *sname = lsname->ln_name;
1961 const char *tname = ltname->ln_name;
1962 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1963 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
1964 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
1965 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1966 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1967 struct mdd_device *mdd = mdo2mdd(src_pobj);
1968 struct mdd_object *mdd_sobj = NULL; /* source object */
1969 struct mdd_object *mdd_tobj = NULL;
1970 struct dynlock_handle *sdlh, *tdlh;
1971 struct thandle *handle;
1972 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1973 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1975 unsigned cl_flags = 0;
1980 mdd_tobj = md2mdd_obj(tobj);
1982 mdd_sobj = mdd_object_find(env, mdd, lf);
1984 handle = mdd_trans_create(env, mdd);
1986 GOTO(out_pending, rc = PTR_ERR(handle));
1988 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
1989 mdd_tobj, lsname, ltname, ma, handle);
1993 rc = mdd_trans_start(env, mdd, handle);
1997 /* FIXME: Should consider tobj and sobj too in rename_lock. */
1998 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2000 GOTO(cleanup_unlocked, rc);
2002 /* Get locks in determined order */
2003 if (rc == MDD_RN_SAME) {
2004 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2005 sname, MOR_SRC_PARENT);
2006 /* check hashes to determine do we need one lock or two */
2007 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2008 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2012 } else if (rc == MDD_RN_SRCTGT) {
2013 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2014 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2016 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2017 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2019 if (sdlh == NULL || tdlh == NULL)
2020 GOTO(cleanup, rc = -ENOMEM);
2022 rc = mdd_la_get(env, mdd_sobj, so_attr,
2023 mdd_object_capa(env, mdd_sobj));
2028 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2029 mdd_object_capa(env, mdd_tobj));
2034 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2035 mdd_tobj, so_attr, tg_attr);
2039 is_dir = S_ISDIR(so_attr->la_mode);
2041 /* Remove source name from source directory */
2042 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2043 mdd_object_capa(env, mdd_spobj));
2047 /* "mv dir1 dir2" needs "dir1/.." link update */
2048 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2049 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2050 mdd_object_capa(env, mdd_sobj));
2052 GOTO(fixup_spobj2, rc);
2054 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2055 handle, mdd_object_capa(env, mdd_sobj));
2057 GOTO(fixup_spobj, rc);
2060 /* Remove target name from target directory
2061 * Here tobj can be remote one, so we do index_delete unconditionally
2062 * and -ENOENT is allowed.
2064 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2065 mdd_object_capa(env, mdd_tpobj));
2068 /* tname might been renamed to something else */
2069 GOTO(fixup_spobj, rc);
2072 GOTO(fixup_spobj, rc);
2075 /* Insert new fid with target name into target dir */
2076 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2077 mdd_object_capa(env, mdd_tpobj));
2079 GOTO(fixup_tpobj, rc);
2081 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2082 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2084 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2086 la->la_valid = LA_CTIME;
2087 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2089 GOTO(fixup_tpobj, rc);
2092 /* Remove old target object
2093 * For tobj is remote case cmm layer has processed
2094 * and set tobj to NULL then. So when tobj is NOT NULL,
2095 * it must be local one.
2097 if (tobj && mdd_object_exists(mdd_tobj)) {
2098 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2099 if (mdd_is_dead_obj(mdd_tobj)) {
2100 mdd_write_unlock(env, mdd_tobj);
2101 /* shld not be dead, something is wrong */
2102 CERROR("tobj is dead, something is wrong\n");
2106 mdo_ref_del(env, mdd_tobj, handle);
2108 /* Remove dot reference. */
2109 if (S_ISDIR(tg_attr->la_mode))
2110 mdo_ref_del(env, mdd_tobj, handle);
2112 /* fetch updated nlink */
2113 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2114 mdd_object_capa(env, mdd_tobj));
2116 GOTO(fixup_tpobj, rc);
2118 la->la_valid = LA_CTIME;
2119 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2121 GOTO(fixup_tpobj, rc);
2123 /* XXX: this transfer to ma will be removed with LOD/OSP */
2124 ma->ma_attr = *tg_attr;
2125 ma->ma_valid |= MA_INODE;
2126 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2127 mdd_write_unlock(env, mdd_tobj);
2129 GOTO(fixup_tpobj, rc);
2131 /* fetch updated nlink */
2132 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2133 mdd_object_capa(env, mdd_tobj));
2135 GOTO(fixup_tpobj, rc);
2136 /* XXX: this transfer to ma will be removed with LOD/OSP */
2137 ma->ma_attr = *tg_attr;
2138 ma->ma_valid |= MA_INODE;
2140 if (so_attr->la_nlink == 0)
2141 cl_flags |= CLF_RENAME_LAST;
2144 la->la_valid = LA_CTIME | LA_MTIME;
2145 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2147 GOTO(fixup_tpobj, rc);
2149 if (mdd_spobj != mdd_tpobj) {
2150 la->la_valid = LA_CTIME | LA_MTIME;
2151 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2155 if (rc == 0 && mdd_sobj) {
2156 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2157 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2158 mdo2fid(mdd_tpobj), ltname, handle);
2160 /* Old files might not have EA entry */
2161 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2163 mdd_write_unlock(env, mdd_sobj);
2164 /* We don't fail the transaction if the link ea can't be
2165 updated -- fid2path will use alternate lookup method. */
2173 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2176 CWARN("tp obj fix error %d\n",rc2);
2178 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2179 !mdd_is_dead_obj(mdd_tobj)) {
2180 rc2 = __mdd_index_insert(env, mdd_tpobj,
2181 mdo2fid(mdd_tobj), tname,
2186 CWARN("tp obj fix error %d\n",rc2);
2191 if (rc && is_dir && mdd_sobj) {
2192 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2196 CWARN("sp obj dotdot delete error %d\n",rc2);
2199 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2200 dotdot, handle, BYPASS_CAPA);
2202 CWARN("sp obj dotdot insert error %d\n",rc2);
2207 rc2 = __mdd_index_insert(env, mdd_spobj,
2208 lf, sname, is_dir, handle, BYPASS_CAPA);
2210 CWARN("sp obj fix error %d\n",rc2);
2213 if (likely(tdlh) && sdlh != tdlh)
2214 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2216 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2219 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2220 mdd_tobj, tpobj_fid, lf,
2221 spobj_fid, ltname, lsname,
2225 mdd_trans_stop(env, mdd, rc, handle);
2227 mdd_object_put(env, mdd_sobj);
2232 /** enable/disable storing of hardlink info */
2233 int mdd_linkea_enable = 1;
2234 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2235 "record hardlink info in EAs");
2237 /** Read the link EA into a temp buffer.
2238 * Uses the name_buf since it is generally large.
2239 * \retval IS_ERR err
2240 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2242 struct lu_buf *mdd_links_get(const struct lu_env *env,
2243 struct mdd_object *mdd_obj)
2246 struct lustre_capa *capa;
2247 struct link_ea_header *leh;
2250 /* First try a small buf */
2251 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2252 if (buf->lb_buf == NULL)
2253 return ERR_PTR(-ENOMEM);
2255 if (!mdd_object_exists(mdd_obj))
2256 return ERR_PTR(-ENODATA);
2258 capa = mdd_object_capa(env, mdd_obj);
2259 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2260 if (rc == -ERANGE) {
2261 /* Buf was too small, figure out what we need. */
2263 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2266 buf = mdd_buf_alloc(env, rc);
2267 if (buf->lb_buf == NULL)
2268 return ERR_PTR(-ENOMEM);
2269 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2275 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2276 leh->leh_magic = LINK_EA_MAGIC;
2277 leh->leh_reccount = __swab32(leh->leh_reccount);
2278 leh->leh_len = __swab64(leh->leh_len);
2279 /* entries are swabbed by mdd_lee_unpack */
2281 if (leh->leh_magic != LINK_EA_MAGIC)
2282 return ERR_PTR(-EINVAL);
2283 if (leh->leh_reccount == 0)
2284 return ERR_PTR(-ENODATA);
2289 /** Pack a link_ea_entry.
2290 * All elements are stored as chars to avoid alignment issues.
2291 * Numbers are always big-endian
2292 * \retval record length
2294 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2295 const struct lu_fid *pfid)
2297 struct lu_fid tmpfid;
2300 fid_cpu_to_be(&tmpfid, pfid);
2301 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2302 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2303 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2305 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2306 lee->lee_reclen[1] = reclen & 0xff;
2310 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2311 struct lu_name *lname, struct lu_fid *pfid)
2313 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2314 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2315 fid_be_to_cpu(pfid, pfid);
2316 lname->ln_name = lee->lee_name;
2317 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2320 /** Add a record to the end of link ea buf */
2321 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2322 const struct lu_fid *pfid,
2323 const struct lu_name *lname)
2325 struct link_ea_header *leh;
2326 struct link_ea_entry *lee;
2329 if (lname == NULL || pfid == NULL)
2332 /* Make sure our buf is big enough for the new one */
2334 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2335 if (leh->leh_len + reclen > buf->lb_len) {
2336 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2341 lee = buf->lb_buf + leh->leh_len;
2342 reclen = mdd_lee_pack(lee, lname, pfid);
2343 leh->leh_len += reclen;
2344 leh->leh_reccount++;
2348 static int mdd_declare_links_add(const struct lu_env *env,
2349 struct mdd_object *mdd_obj,
2350 struct thandle *handle)
2354 /* XXX: max size? */
2355 rc = mdo_declare_xattr_set(env, mdd_obj,
2356 mdd_buf_get_const(env, NULL, 4096),
2357 XATTR_NAME_LINK, 0, handle);
2362 /* For pathologic linkers, we don't want to spend lots of time scanning the
2363 * link ea. Limit ourseleves to something reasonable; links not in the EA
2364 * can be looked up via (slower) parent lookup.
2366 #define LINKEA_MAX_COUNT 128
2368 static int mdd_links_add(const struct lu_env *env,
2369 struct mdd_object *mdd_obj,
2370 const struct lu_fid *pfid,
2371 const struct lu_name *lname,
2372 struct thandle *handle, int first)
2375 struct link_ea_header *leh;
2379 if (!mdd_linkea_enable)
2382 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2385 if (rc != -ENODATA) {
2386 CERROR("link_ea read failed %d "DFID"\n", rc,
2387 PFID(mdd_object_fid(mdd_obj)));
2390 /* empty EA; start one */
2391 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2392 if (buf->lb_buf == NULL)
2395 leh->leh_magic = LINK_EA_MAGIC;
2396 leh->leh_len = sizeof(struct link_ea_header);
2397 leh->leh_reccount = 0;
2401 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2404 rc = __mdd_links_add(env, buf, pfid, lname);
2409 rc = mdo_xattr_set(env, mdd_obj,
2410 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2411 XATTR_NAME_LINK, 0, handle,
2412 mdd_object_capa(env, mdd_obj));
2415 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2416 PFID(mdd_object_fid(mdd_obj)));
2418 CERROR("link_ea add failed %d "DFID"\n", rc,
2419 PFID(mdd_object_fid(mdd_obj)));
2422 if (buf->lb_len > OBD_ALLOC_BIG)
2423 /* if we vmalloced a large buffer drop it */
2429 static int mdd_links_rename(const struct lu_env *env,
2430 struct mdd_object *mdd_obj,
2431 const struct lu_fid *oldpfid,
2432 const struct lu_name *oldlname,
2433 const struct lu_fid *newpfid,
2434 const struct lu_name *newlname,
2435 struct thandle *handle)
2438 struct link_ea_header *leh;
2439 struct link_ea_entry *lee;
2440 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2441 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2447 if (!mdd_linkea_enable)
2450 if (mdd_obj->mod_flags & DEAD_OBJ)
2451 /* No more links, don't bother */
2454 buf = mdd_links_get(env, mdd_obj);
2458 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2459 rc, PFID(mdd_object_fid(mdd_obj)));
2461 CERROR("link_ea read failed %d "DFID"\n",
2462 rc, PFID(mdd_object_fid(mdd_obj)));
2466 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2468 /* Find the old record */
2469 for(count = 0; count < leh->leh_reccount; count++) {
2470 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2471 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2472 lu_fid_eq(tmpfid, oldpfid) &&
2473 (strncmp(tmpname->ln_name, oldlname->ln_name,
2474 tmpname->ln_namelen) == 0))
2476 lee = (struct link_ea_entry *)((char *)lee + reclen);
2478 if ((count + 1) > leh->leh_reccount) {
2479 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2480 oldlname->ln_namelen, oldlname->ln_name);
2481 GOTO(out, rc = -ENOENT);
2484 /* Remove the old record */
2485 leh->leh_reccount--;
2486 leh->leh_len -= reclen;
2487 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2490 /* If renaming, add the new record */
2491 if (newpfid != NULL) {
2492 /* if the add fails, we still delete the out-of-date old link */
2493 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2497 rc = mdo_xattr_set(env, mdd_obj,
2498 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2499 XATTR_NAME_LINK, 0, handle,
2500 mdd_object_capa(env, mdd_obj));
2506 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2507 oldlname->ln_namelen, oldlname->ln_name, rc,
2508 PFID(mdd_object_fid(mdd_obj)));
2510 if (buf->lb_len > OBD_ALLOC_BIG)
2511 /* if we vmalloced a large buffer drop it */
2517 const struct md_dir_operations mdd_dir_ops = {
2518 .mdo_is_subdir = mdd_is_subdir,
2519 .mdo_lookup = mdd_lookup,
2520 .mdo_create = mdd_create,
2521 .mdo_rename = mdd_rename,
2522 .mdo_link = mdd_link,
2523 .mdo_unlink = mdd_unlink,
2524 .mdo_create_data = mdd_create_data,