4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_dir.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@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 inline 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 inline int mdd_links_del(const struct lu_env *env,
72 struct mdd_object *mdd_obj,
73 const struct lu_fid *pfid,
74 const struct lu_name *lname,
75 struct thandle *handle);
76 static int mdd_links_rename(const struct lu_env *env,
77 struct mdd_object *mdd_obj,
78 const struct lu_fid *oldpfid,
79 const struct lu_name *oldlname,
80 const struct lu_fid *newpfid,
81 const struct lu_name *newlname,
82 struct thandle *handle,
83 int first, int check);
86 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
87 const struct lu_name *lname, struct lu_fid* fid, int mask)
89 const char *name = lname->ln_name;
90 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
91 struct dynlock_handle *dlh;
94 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
95 if (unlikely(dlh == NULL))
97 rc = __mdd_lookup(env, pobj, lname, fid, mask);
98 mdd_pdo_read_unlock(env, mdd_obj, dlh);
103 int mdd_lookup(const struct lu_env *env,
104 struct md_object *pobj, const struct lu_name *lname,
105 struct lu_fid* fid, struct md_op_spec *spec)
109 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
113 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
116 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
120 * For root fid use special function, which does not compare version component
121 * of fid. Version component is different for root fids on all MDTs.
123 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
125 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
126 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
130 * return 1: if lf is the fid of the ancestor of p1;
133 * return -EREMOTE: if remote object is found, in this
134 * case fid of remote object is saved to @pf;
136 * otherwise: values < 0, errors.
138 static int mdd_is_parent(const struct lu_env *env,
139 struct mdd_device *mdd,
140 struct mdd_object *p1,
141 const struct lu_fid *lf,
144 struct mdd_object *parent = NULL;
149 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
150 pfid = &mdd_env_info(env)->mti_fid;
152 /* Check for root first. */
153 if (mdd_is_root(mdd, mdo2fid(p1)))
157 /* this is done recursively, bypass capa for each obj */
158 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
159 rc = mdd_parent_fid(env, p1, pfid);
162 if (mdd_is_root(mdd, pfid))
164 if (lu_fid_eq(pfid, lf))
167 mdd_object_put(env, parent);
168 parent = mdd_object_find(env, mdd, pfid);
170 /* cross-ref parent */
171 if (parent == NULL) {
174 GOTO(out, rc = -EREMOTE);
175 } else if (IS_ERR(parent))
176 GOTO(out, rc = PTR_ERR(parent));
181 if (parent && !IS_ERR(parent))
182 mdd_object_put(env, parent);
187 * No permission check is needed.
189 * returns 1: if fid is ancestor of @mo;
190 * returns 0: if fid is not a ancestor of @mo;
192 * returns EREMOTE if remote object is found, fid of remote object is saved to
195 * returns < 0: if error
197 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
198 const struct lu_fid *fid, struct lu_fid *sfid)
200 struct mdd_device *mdd = mdo2mdd(mo);
204 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
207 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
211 } else if (rc == 1) {
212 /* found @fid is parent */
220 * Check that @dir contains no entries except (possibly) dot and dotdot.
225 * -ENOTDIR not a directory object
226 * -ENOTEMPTY not empty
230 static int mdd_dir_is_empty(const struct lu_env *env,
231 struct mdd_object *dir)
234 struct dt_object *obj;
235 const struct dt_it_ops *iops;
239 obj = mdd_object_child(dir);
240 if (!dt_try_as_dir(env, obj))
243 iops = &obj->do_index_ops->dio_it;
244 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
246 result = iops->get(env, it, (const void *)"");
249 for (result = 0, i = 0; result == 0 && i < 3; ++i)
250 result = iops->next(env, it);
253 else if (result == +1)
255 } else if (result == 0)
257 * Huh? Index contains no zero key?
264 result = PTR_ERR(it);
268 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
270 struct mdd_device *m = mdd_obj2mdd_dev(obj);
271 struct lu_attr *la = &mdd_env_info(env)->mti_la;
275 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
280 * Subdir count limitation can be broken through.
282 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
283 !S_ISDIR(la->la_mode))
290 * Check whether it may create the cobj under the pobj.
293 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
294 struct mdd_object *cobj, int check_perm, int check_nlink)
299 if (cobj && mdd_object_exists(cobj))
302 if (mdd_is_dead_obj(pobj))
306 rc = mdd_permission_internal_locked(env, pobj, NULL,
307 MAY_WRITE | MAY_EXEC,
309 if (!rc && check_nlink)
310 rc = __mdd_may_link(env, pobj);
316 * Check whether can unlink from the pobj in the case of "cobj == NULL".
318 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
319 const struct lu_attr *attr)
324 if (mdd_is_dead_obj(pobj))
327 if ((attr->la_valid & LA_FLAGS) &&
328 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
331 rc = mdd_permission_internal_locked(env, pobj, NULL,
332 MAY_WRITE | MAY_EXEC,
337 if (mdd_is_append(pobj))
344 * pobj == NULL is remote ops case, under such case, pobj's
345 * VTX feature has been checked already, no need check again.
347 static inline int mdd_is_sticky(const struct lu_env *env,
348 struct mdd_object *pobj,
349 struct mdd_object *cobj)
351 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
352 struct lu_ucred *uc = lu_ucred_assert(env);
356 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
360 if (!(tmp_la->la_mode & S_ISVTX) ||
361 (tmp_la->la_uid == uc->uc_fsuid))
365 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
369 if (tmp_la->la_uid == uc->uc_fsuid)
372 return !md_capable(uc, CFS_CAP_FOWNER);
375 static int mdd_may_delete_entry(const struct lu_env *env,
376 struct mdd_object *pobj, int check_perm)
380 LASSERT(pobj != NULL);
381 if (!mdd_object_exists(pobj))
384 if (mdd_is_dead_obj(pobj))
389 rc = mdd_permission_internal_locked(env, pobj, NULL,
390 MAY_WRITE | MAY_EXEC,
396 if (mdd_is_append(pobj))
403 * Check whether it may delete the cobj from the pobj.
406 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
407 struct mdd_object *cobj, struct lu_attr *cattr,
408 struct lu_attr *src_attr, int check_perm, int check_empty)
414 rc = mdd_may_delete_entry(env, pobj, check_perm);
422 if (!mdd_object_exists(cobj))
425 if (mdd_is_dead_obj(cobj))
429 if (mdd_is_sticky(env, pobj, cobj))
432 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
435 if ((cattr->la_valid & LA_FLAGS) &&
436 (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
439 /* additional check the rename case */
441 if (S_ISDIR(src_attr->la_mode)) {
442 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
444 if (!S_ISDIR(cattr->la_mode))
447 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
449 } else if (S_ISDIR(cattr->la_mode))
453 if (S_ISDIR(cattr->la_mode) && check_empty)
454 rc = mdd_dir_is_empty(env, cobj);
461 * has mdd_write_lock on src already, but not on tgt yet
463 int mdd_link_sanity_check(const struct lu_env *env,
464 struct mdd_object *tgt_obj,
465 const struct lu_name *lname,
466 struct mdd_object *src_obj)
468 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
472 if (!mdd_object_exists(src_obj))
475 if (mdd_is_dead_obj(src_obj))
478 /* Local ops, no lookup before link, check filename length here. */
479 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
480 RETURN(-ENAMETOOLONG);
482 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
485 if (S_ISDIR(mdd_object_type(src_obj)))
488 LASSERT(src_obj != tgt_obj);
490 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
495 rc = __mdd_may_link(env, src_obj);
500 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
501 const char *name, struct thandle *handle,
502 struct lustre_capa *capa)
504 struct dt_object *next = mdd_object_child(pobj);
508 if (dt_try_as_dir(env, next)) {
509 rc = next->do_index_ops->dio_delete(env, next,
510 (struct dt_key *)name,
518 static int __mdd_index_insert_only(const struct lu_env *env,
519 struct mdd_object *pobj,
520 const struct lu_fid *lf, const char *name,
521 struct thandle *handle,
522 struct lustre_capa *capa)
524 struct dt_object *next = mdd_object_child(pobj);
528 if (dt_try_as_dir(env, next)) {
529 struct lu_ucred *uc = lu_ucred_check(env);
532 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
533 rc = next->do_index_ops->dio_insert(env, next,
535 (const struct dt_key *)name,
536 handle, capa, ignore_quota);
543 /* insert named index, add reference if isdir */
544 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
545 const struct lu_fid *lf, const char *name, int is_dir,
546 struct thandle *handle, struct lustre_capa *capa)
551 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
552 if (rc == 0 && is_dir) {
553 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
554 mdo_ref_add(env, pobj, handle);
555 mdd_write_unlock(env, pobj);
560 /* delete named index, drop reference if isdir */
561 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
562 const char *name, int is_dir, struct thandle *handle,
563 struct lustre_capa *capa)
568 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
569 if (rc == 0 && is_dir) {
570 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
571 mdo_ref_del(env, pobj, handle);
572 mdd_write_unlock(env, pobj);
578 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
579 int reclen, struct thandle *handle)
583 /* XXX: this is a temporary solution to declare llog changes
584 * will be fixed in 2.3 with new llog implementation */
586 LASSERT(mdd->mdd_capa);
588 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
589 * have to set the parameter 'size' as INT_MAX or 0 to inform
590 * OSD that this record write is for a llog write or catalog
591 * header update, and osd declare function will reserve less
592 * credits for optimization purpose.
594 * Reserve 6 blocks for a llog write, since the llog file is
595 * usually small, reserve 2 blocks for catalog header update,
596 * because we know for sure that catalog header is already
599 * This hack should be removed in 2.3.
603 rc = dt_declare_record_write(env, mdd->mdd_capa,
604 DECLARE_LLOG_WRITE, 0, handle);
608 /* header will be updated as well */
609 rc = dt_declare_record_write(env, mdd->mdd_capa,
610 DECLARE_LLOG_WRITE, 0, handle);
614 /* also we should be able to create new plain log */
615 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
619 /* new record referencing new plain llog */
620 rc = dt_declare_record_write(env, mdd->mdd_capa,
621 DECLARE_LLOG_WRITE, 0, handle);
625 /* catalog's header will be updated as well */
626 rc = dt_declare_record_write(env, mdd->mdd_capa,
627 DECLARE_LLOG_REWRITE, 0, handle);
632 int mdd_declare_changelog_store(const struct lu_env *env,
633 struct mdd_device *mdd,
634 const struct lu_name *fname,
635 struct thandle *handle)
637 struct obd_device *obd = mdd2obd_dev(mdd);
638 struct llog_ctxt *ctxt;
639 struct llog_changelog_rec *rec;
645 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
648 reclen = llog_data_len(sizeof(*rec) +
649 (fname != NULL ? fname->ln_namelen : 0));
650 buf = mdd_buf_alloc(env, reclen);
651 if (buf->lb_buf == NULL)
655 rec->cr_hdr.lrh_len = reclen;
656 rec->cr_hdr.lrh_type = CHANGELOG_REC;
658 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
662 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
668 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
669 struct mdd_device *mdd,
670 const struct lu_name *tname,
671 const struct lu_name *sname,
672 struct thandle *handle)
674 struct obd_device *obd = mdd2obd_dev(mdd);
675 struct llog_ctxt *ctxt;
676 struct llog_changelog_ext_rec *rec;
682 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
685 reclen = llog_data_len(sizeof(*rec) +
686 (tname != NULL ? tname->ln_namelen : 0) +
687 (sname != NULL ? 1 + sname->ln_namelen : 0));
688 buf = mdd_buf_alloc(env, reclen);
689 if (buf->lb_buf == NULL)
693 rec->cr_hdr.lrh_len = reclen;
694 rec->cr_hdr.lrh_type = CHANGELOG_REC;
696 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
700 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
706 /** Add a changelog entry \a rec to the changelog llog
709 * \param handle - currently ignored since llogs start their own transaction;
710 * this will hopefully be fixed in llog rewrite
713 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
714 struct llog_changelog_rec *rec, struct thandle *th)
716 struct obd_device *obd = mdd2obd_dev(mdd);
717 struct llog_ctxt *ctxt;
720 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
721 rec->cr_hdr.lrh_type = CHANGELOG_REC;
722 rec->cr.cr_time = cl_time();
724 spin_lock(&mdd->mdd_cl.mc_lock);
725 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
726 * but as long as the MDD transactions are ordered correctly for e.g.
727 * rename conflicts, I don't think this should matter. */
728 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
729 spin_unlock(&mdd->mdd_cl.mc_lock);
731 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
735 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
742 /** Add a changelog_ext entry \a rec to the changelog llog
745 * \param handle - currently ignored since llogs start their own transaction;
746 * this will hopefully be fixed in llog rewrite
749 int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
750 struct llog_changelog_ext_rec *rec,
753 struct obd_device *obd = mdd2obd_dev(mdd);
754 struct llog_ctxt *ctxt;
757 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
758 /* llog_lvfs_write_rec sets the llog tail len */
759 rec->cr_hdr.lrh_type = CHANGELOG_REC;
760 rec->cr.cr_time = cl_time();
762 spin_lock(&mdd->mdd_cl.mc_lock);
763 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
764 * but as long as the MDD transactions are ordered correctly for e.g.
765 * rename conflicts, I don't think this should matter. */
766 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
767 spin_unlock(&mdd->mdd_cl.mc_lock);
769 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
773 /* nested journal transaction */
774 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
782 /** Store a namespace change changelog record
783 * If this fails, we must fail the whole transaction; we don't
784 * want the change to commit without the log entry.
785 * \param target - mdd_object of change
786 * \param parent - parent dir/object
787 * \param tname - target name string
788 * \param handle - transacion handle
790 static int mdd_changelog_ns_store(const struct lu_env *env,
791 struct mdd_device *mdd,
792 enum changelog_rec_type type,
794 struct mdd_object *target,
795 struct mdd_object *parent,
796 const struct lu_name *tname,
797 struct thandle *handle)
799 struct llog_changelog_rec *rec;
806 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
808 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
811 LASSERT(target != NULL);
812 LASSERT(parent != NULL);
813 LASSERT(tname != NULL);
814 LASSERT(handle != NULL);
816 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
817 buf = mdd_buf_alloc(env, reclen);
818 if (buf->lb_buf == NULL)
822 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
823 rec->cr.cr_type = (__u32)type;
824 rec->cr.cr_tfid = *mdo2fid(target);
825 rec->cr.cr_pfid = *mdo2fid(parent);
826 rec->cr.cr_namelen = tname->ln_namelen;
827 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
829 target->mod_cltime = cfs_time_current_64();
831 rc = mdd_changelog_store(env, mdd, rec, handle);
833 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
834 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
835 PFID(&rec->cr.cr_pfid));
843 /** Store a namespace change changelog record
844 * If this fails, we must fail the whole transaction; we don't
845 * want the change to commit without the log entry.
846 * \param target - mdd_object of change
847 * \param tpfid - target parent dir/object fid
848 * \param sfid - source object fid
849 * \param spfid - source parent fid
850 * \param tname - target name string
851 * \param sname - source name string
852 * \param handle - transacion handle
854 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
855 struct mdd_device *mdd,
856 enum changelog_rec_type type,
858 struct mdd_object *target,
859 const struct lu_fid *tpfid,
860 const struct lu_fid *sfid,
861 const struct lu_fid *spfid,
862 const struct lu_name *tname,
863 const struct lu_name *sname,
864 struct thandle *handle)
866 struct llog_changelog_ext_rec *rec;
873 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
875 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
878 LASSERT(sfid != NULL);
879 LASSERT(tpfid != NULL);
880 LASSERT(tname != NULL);
881 LASSERT(handle != NULL);
883 reclen = llog_data_len(sizeof(*rec) +
884 sname != NULL ? 1 + sname->ln_namelen : 0);
885 buf = mdd_buf_alloc(env, reclen);
886 if (buf->lb_buf == NULL)
890 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
891 rec->cr.cr_type = (__u32)type;
892 rec->cr.cr_pfid = *tpfid;
893 rec->cr.cr_sfid = *sfid;
894 rec->cr.cr_spfid = *spfid;
895 rec->cr.cr_namelen = tname->ln_namelen;
896 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
898 rec->cr.cr_name[tname->ln_namelen] = '\0';
899 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
901 rec->cr.cr_namelen += 1 + sname->ln_namelen;
904 if (likely(target != NULL)) {
905 rec->cr.cr_tfid = *mdo2fid(target);
906 target->mod_cltime = cfs_time_current_64();
908 fid_zero(&rec->cr.cr_tfid);
911 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
913 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
914 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
921 static int mdd_declare_link(const struct lu_env *env,
922 struct mdd_device *mdd,
923 struct mdd_object *p,
924 struct mdd_object *c,
925 const struct lu_name *name,
926 struct thandle *handle)
930 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
934 rc = mdo_declare_ref_add(env, c, handle);
938 rc = mdo_declare_attr_set(env, p, NULL, handle);
942 rc = mdo_declare_attr_set(env, c, NULL, handle);
946 rc = mdd_declare_links_add(env, c, handle);
950 rc = mdd_declare_changelog_store(env, mdd, name, handle);
955 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
956 struct md_object *src_obj, const struct lu_name *lname,
959 const char *name = lname->ln_name;
960 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
961 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
962 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
963 struct mdd_device *mdd = mdo2mdd(src_obj);
964 struct dynlock_handle *dlh;
965 struct thandle *handle;
969 handle = mdd_trans_create(env, mdd);
971 GOTO(out_pending, rc = PTR_ERR(handle));
973 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
977 rc = mdd_trans_start(env, mdd, handle);
981 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
983 GOTO(out_trans, rc = -ENOMEM);
984 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
986 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
988 GOTO(out_unlock, rc);
990 rc = mdo_ref_add(env, mdd_sobj, handle);
992 GOTO(out_unlock, rc);
995 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
997 mdd_object_capa(env, mdd_tobj));
999 mdo_ref_del(env, mdd_sobj, handle);
1000 GOTO(out_unlock, rc);
1003 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1004 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1006 la->la_valid = LA_CTIME | LA_MTIME;
1007 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1009 GOTO(out_unlock, rc);
1011 la->la_valid = LA_CTIME;
1012 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
1014 mdd_links_add(env, mdd_sobj,
1015 mdo2fid(mdd_tobj), lname, handle, 0);
1020 mdd_write_unlock(env, mdd_sobj);
1021 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
1024 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1025 mdd_tobj, lname, handle);
1027 mdd_trans_stop(env, mdd, rc, handle);
1032 int mdd_declare_finish_unlink(const struct lu_env *env,
1033 struct mdd_object *obj,
1035 struct thandle *handle)
1039 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1043 return mdo_declare_destroy(env, obj, handle);
1046 /* caller should take a lock before calling */
1047 int mdd_finish_unlink(const struct lu_env *env,
1048 struct mdd_object *obj, struct md_attr *ma,
1052 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1055 LASSERT(mdd_write_locked(env, obj) != 0);
1057 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
1058 obj->mod_flags |= DEAD_OBJ;
1059 /* add new orphan and the object
1060 * will be deleted during mdd_close() */
1061 if (obj->mod_count) {
1062 rc = __mdd_orphan_add(env, obj, th);
1064 CDEBUG(D_HA, "Object "DFID" is inserted into "
1065 "orphan list, open count = %d\n",
1066 PFID(mdd_object_fid(obj)),
1069 CERROR("Object "DFID" fail to be an orphan, "
1070 "open count = %d, maybe cause failed "
1072 PFID(mdd_object_fid(obj)),
1075 rc = mdo_destroy(env, obj, th);
1084 * has mdd_write_lock on cobj already, but not on pobj yet
1086 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1087 struct mdd_object *cobj, struct lu_attr *cattr)
1092 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1097 static inline int mdd_declare_links_del(const struct lu_env *env,
1098 struct mdd_object *c,
1099 struct thandle *handle)
1103 /* For directory, the linkEA will be removed together with the object. */
1104 if (!S_ISDIR(mdd_object_type(c)))
1105 rc = mdd_declare_links_add(env, c, handle);
1110 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1111 struct mdd_object *p, struct mdd_object *c,
1112 const struct lu_name *name, struct md_attr *ma,
1113 struct thandle *handle)
1115 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1118 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1122 rc = mdo_declare_ref_del(env, p, handle);
1126 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1127 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1128 la->la_valid = LA_CTIME | LA_MTIME;
1129 rc = mdo_declare_attr_set(env, p, la, handle);
1134 rc = mdo_declare_ref_del(env, c, handle);
1138 rc = mdo_declare_ref_del(env, c, handle);
1142 rc = mdo_declare_attr_set(env, c, NULL, handle);
1146 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1150 rc = mdd_declare_links_del(env, c, handle);
1154 /* FIXME: need changelog for remove entry */
1155 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1161 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1162 struct md_object *cobj, const struct lu_name *lname,
1165 const char *name = lname->ln_name;
1166 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1167 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1168 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1169 struct mdd_object *mdd_cobj = NULL;
1170 struct mdd_device *mdd = mdo2mdd(pobj);
1171 struct dynlock_handle *dlh;
1172 struct thandle *handle;
1176 /* cobj == NULL means only delete name entry */
1177 if (likely(cobj != NULL)) {
1178 mdd_cobj = md2mdd_obj(cobj);
1179 if (mdd_object_exists(mdd_cobj) == 0)
1181 /* currently it is assume, it could only delete
1182 * name entry of remote directory */
1186 handle = mdd_trans_create(env, mdd);
1188 RETURN(PTR_ERR(handle));
1190 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1195 rc = mdd_trans_start(env, mdd, handle);
1199 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1201 GOTO(stop, rc = -ENOMEM);
1203 if (likely(mdd_cobj != NULL)) {
1204 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1207 rc = mdd_la_get(env, mdd_cobj, cattr,
1208 mdd_object_capa(env, mdd_cobj));
1212 is_dir = S_ISDIR(cattr->la_mode);
1216 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1220 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1221 mdd_object_capa(env, mdd_pobj));
1225 if (likely(mdd_cobj != NULL)) {
1226 rc = mdo_ref_del(env, mdd_cobj, handle);
1228 __mdd_index_insert_only(env, mdd_pobj,
1231 mdd_object_capa(env, mdd_pobj));
1237 mdo_ref_del(env, mdd_cobj, handle);
1239 /* fetch updated nlink */
1240 rc = mdd_la_get(env, mdd_cobj, cattr,
1241 mdd_object_capa(env, mdd_cobj));
1246 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1247 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1249 la->la_valid = LA_CTIME | LA_MTIME;
1250 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1254 /* Enough for only unlink the entry */
1255 if (unlikely(mdd_cobj == NULL)) {
1256 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1260 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1261 /* update ctime of an unlinked file only if it is still
1262 * opened or a link still exists */
1263 la->la_valid = LA_CTIME;
1264 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1269 /* XXX: this transfer to ma will be removed with LOD/OSP */
1270 ma->ma_attr = *cattr;
1271 ma->ma_valid |= MA_INODE;
1272 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1274 /* fetch updated nlink */
1276 rc = mdd_la_get(env, mdd_cobj, cattr,
1277 mdd_object_capa(env, mdd_cobj));
1280 /* old files may not have link ea; ignore errors */
1281 mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle);
1283 /* if object is removed then we can't get its attrs, use last get */
1284 if (cattr->la_nlink == 0) {
1285 ma->ma_attr = *cattr;
1286 ma->ma_valid |= MA_INODE;
1290 mdd_write_unlock(env, mdd_cobj);
1291 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1295 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1296 if ((ma->ma_valid & MA_HSM) &&
1297 (ma->ma_hsm.mh_flags & HS_EXISTS))
1298 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1300 rc = mdd_changelog_ns_store(env, mdd,
1301 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1302 mdd_cobj, mdd_pobj, lname, handle);
1306 mdd_trans_stop(env, mdd, rc, handle);
1312 * The permission has been checked when obj created, no need check again.
1314 static int mdd_cd_sanity_check(const struct lu_env *env,
1315 struct mdd_object *obj)
1320 if (!obj || mdd_is_dead_obj(obj))
1327 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1328 struct md_object *cobj, const struct md_op_spec *spec,
1331 struct mdd_device *mdd = mdo2mdd(cobj);
1332 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1333 struct mdd_object *son = md2mdd_obj(cobj);
1334 struct thandle *handle;
1335 const struct lu_buf *buf;
1336 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1340 /* do not let users to create stripes via .lustre/
1341 * mdd_obf_setup() sets IMMUTE_OBJ on this directory */
1342 if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ)
1345 rc = mdd_cd_sanity_check(env, son);
1349 if (!md_should_create(spec->sp_cr_flags))
1353 * there are following use cases for this function:
1354 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1355 * striping can be specified or not
1358 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1362 /* calling ->ah_make_hint() is used to transfer information from parent */
1363 mdd_object_make_hint(env, mdd_pobj, son, attr);
1365 handle = mdd_trans_create(env, mdd);
1367 GOTO(out_free, rc = PTR_ERR(handle));
1370 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1371 * Should this be fixed?
1373 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, no_create %u\n",
1374 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1375 spec->sp_cr_flags, spec->no_create);
1377 if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1378 /* replay case or lfs setstripe */
1379 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1380 spec->u.sp_ea.eadatalen);
1385 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1386 XATTR_NAME_LOV, 0, handle);
1390 rc = mdd_trans_start(env, mdd, handle);
1394 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1395 0, handle, mdd_object_capa(env, son));
1397 mdd_trans_stop(env, mdd, rc, handle);
1402 /* Get fid from name and parent */
1404 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1405 const struct lu_name *lname, struct lu_fid* fid, int mask)
1407 const char *name = lname->ln_name;
1408 const struct dt_key *key = (const struct dt_key *)name;
1409 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1410 struct mdd_device *m = mdo2mdd(pobj);
1411 struct dt_object *dir = mdd_object_child(mdd_obj);
1415 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1418 rc = mdd_object_exists(mdd_obj);
1419 if (unlikely(rc == 0))
1421 else if (unlikely(rc < 0)) {
1422 CERROR("Object "DFID" locates on remote server\n",
1423 PFID(mdo2fid(mdd_obj)));
1427 /* The common filename length check. */
1428 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1429 RETURN(-ENAMETOOLONG);
1431 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1436 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1437 dt_try_as_dir(env, dir))) {
1439 rc = dir->do_index_ops->dio_lookup(env, dir,
1440 (struct dt_rec *)fid, key,
1441 mdd_object_capa(env, mdd_obj));
1452 int mdd_declare_object_initialize(const struct lu_env *env,
1453 struct mdd_object *child,
1454 struct lu_attr *attr,
1455 struct thandle *handle)
1460 * inode mode has been set in creation time, and it's based on umask,
1461 * la_mode and acl, don't set here again! (which will go wrong
1462 * because below function doesn't consider umask).
1463 * I'd suggest set all object attributes in creation time, see above.
1465 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1466 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1467 rc = mdo_declare_attr_set(env, child, attr, handle);
1468 attr->la_valid |= LA_MODE | LA_TYPE;
1469 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1470 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1473 rc = mdo_declare_ref_add(env, child, handle);
1477 mdd_declare_links_add(env, child, handle);
1482 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1483 const struct lu_name *lname, struct mdd_object *child,
1484 struct lu_attr *attr, struct thandle *handle,
1485 const struct md_op_spec *spec)
1491 * Update attributes for child.
1494 * (1) the valid bits should be converted between Lustre and Linux;
1495 * (2) maybe, the child attributes should be set in OSD when creation.
1499 * inode mode has been set in creation time, and it's based on umask,
1500 * la_mode and acl, don't set here again! (which will go wrong
1501 * because below function doesn't consider umask).
1502 * I'd suggest set all object attributes in creation time, see above.
1504 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1505 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1506 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1507 /* arguments are supposed to stay the same */
1508 attr->la_valid |= LA_MODE | LA_TYPE;
1512 if (S_ISDIR(attr->la_mode)) {
1513 /* Add "." and ".." for newly created dir */
1514 mdo_ref_add(env, child, handle);
1515 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1516 dot, handle, BYPASS_CAPA);
1518 rc = __mdd_index_insert_only(env, child, pfid,
1522 mdo_ref_del(env, child, handle);
1525 mdd_links_add(env, child, pfid, lname, handle, 1);
1530 /* has not lock on pobj yet */
1531 static int mdd_create_sanity_check(const struct lu_env *env,
1532 struct md_object *pobj,
1533 struct lu_attr *pattr,
1534 const struct lu_name *lname,
1535 struct lu_attr *cattr,
1536 struct md_op_spec *spec)
1538 struct mdd_thread_info *info = mdd_env_info(env);
1539 struct lu_fid *fid = &info->mti_fid;
1540 struct mdd_object *obj = md2mdd_obj(pobj);
1541 struct mdd_device *m = mdo2mdd(pobj);
1546 if (mdd_is_dead_obj(obj))
1550 * In some cases this lookup is not needed - we know before if name
1551 * exists or not because MDT performs lookup for it.
1552 * name length check is done in lookup.
1554 if (spec->sp_cr_lookup) {
1556 * Check if the name already exist, though it will be checked in
1557 * _index_insert also, for avoiding rolling back if exists
1560 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1561 MAY_WRITE | MAY_EXEC);
1563 RETURN(rc ? : -EEXIST);
1566 * Check WRITE permission for the parent.
1567 * EXEC permission have been checked
1568 * when lookup before create already.
1570 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1577 if (pattr->la_mode & S_ISGID) {
1578 cattr->la_gid = pattr->la_gid;
1579 if (S_ISDIR(cattr->la_mode)) {
1580 cattr->la_mode |= S_ISGID;
1581 cattr->la_valid |= LA_MODE;
1585 switch (cattr->la_mode & S_IFMT) {
1587 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1589 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1590 RETURN(-ENAMETOOLONG);
1609 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1610 struct mdd_object *p, struct mdd_object *c,
1611 const struct lu_name *name,
1612 struct lu_attr *attr,
1614 struct thandle *handle,
1615 const struct md_op_spec *spec)
1619 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1623 #ifdef CONFIG_FS_POSIX_ACL
1624 if (got_def_acl > 0) {
1625 struct lu_buf *acl_buf;
1627 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1628 /* if dir, then can inherit default ACl */
1629 if (S_ISDIR(attr->la_mode)) {
1630 rc = mdo_declare_xattr_set(env, c, acl_buf,
1631 XATTR_NAME_ACL_DEFAULT,
1637 rc = mdo_declare_attr_set(env, c, attr, handle);
1641 rc = mdo_declare_xattr_set(env, c, acl_buf,
1642 XATTR_NAME_ACL_ACCESS, 0, handle);
1648 if (S_ISDIR(attr->la_mode)) {
1649 rc = mdo_declare_ref_add(env, p, handle);
1654 rc = mdd_declare_object_initialize(env, c, attr, handle);
1658 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1659 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
1661 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1662 name->ln_name, handle);
1666 /* replay case, create LOV EA from client data */
1667 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1668 const struct lu_buf *buf;
1670 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1671 spec->u.sp_ea.eadatalen);
1672 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1678 if (S_ISLNK(attr->la_mode)) {
1679 rc = dt_declare_record_write(env, mdd_object_child(c),
1680 strlen(spec->u.sp_symname), 0,
1686 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
1687 rc = mdo_declare_attr_set(env, p, attr, handle);
1692 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1701 * Create object and insert it into namespace.
1703 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1704 const struct lu_name *lname, struct md_object *child,
1705 struct md_op_spec *spec, struct md_attr* ma)
1707 struct mdd_thread_info *info = mdd_env_info(env);
1708 struct lu_attr *la = &info->mti_la_for_fix;
1709 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1710 struct mdd_object *son = md2mdd_obj(child);
1711 struct mdd_device *mdd = mdo2mdd(pobj);
1712 struct lu_attr *attr = &ma->ma_attr;
1713 struct thandle *handle;
1714 struct lu_attr *pattr = &info->mti_pattr;
1715 struct dynlock_handle *dlh;
1716 const char *name = lname->ln_name;
1717 int rc, created = 0, initialized = 0, inserted = 0;
1718 int got_def_acl = 0;
1722 * Two operations have to be performed:
1724 * - an allocation of a new object (->do_create()), and
1726 * - an insertion into a parent index (->dio_insert()).
1728 * Due to locking, operation order is not important, when both are
1729 * successful, *but* error handling cases are quite different:
1731 * - if insertion is done first, and following object creation fails,
1732 * insertion has to be rolled back, but this operation might fail
1733 * also leaving us with dangling index entry.
1735 * - if creation is done first, is has to be undone if insertion
1736 * fails, leaving us with leaked space, which is neither good, nor
1739 * It seems that creation-first is simplest solution, but it is
1740 * sub-optimal in the frequent
1745 * case, because second mkdir is bound to create object, only to
1746 * destroy it immediately.
1748 * To avoid this follow local file systems that do double lookup:
1750 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1752 * 1. create (mdd_object_create_internal())
1754 * 2. insert (__mdd_index_insert(), lookup again)
1757 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1761 /* Sanity checks before big job. */
1762 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
1766 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1767 GOTO(out_free, rc = -EINPROGRESS);
1769 if (!S_ISLNK(attr->la_mode)) {
1770 struct lu_buf *acl_buf;
1772 acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
1773 sizeof(info->mti_xattr_buf));
1774 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1775 rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
1776 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1777 mdd_read_unlock(env, mdd_pobj);
1780 else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
1784 mdd_object_make_hint(env, mdd_pobj, son, attr);
1786 handle = mdd_trans_create(env, mdd);
1788 GOTO(out_free, rc = PTR_ERR(handle));
1790 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
1791 got_def_acl, handle, spec);
1795 rc = mdd_trans_start(env, mdd, handle);
1799 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1801 GOTO(out_trans, rc = -ENOMEM);
1803 mdd_write_lock(env, son, MOR_TGT_CHILD);
1804 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec);
1806 mdd_write_unlock(env, son);
1812 #ifdef CONFIG_FS_POSIX_ACL
1814 struct lu_buf *acl_buf;
1816 acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
1817 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1819 mdd_write_unlock(env, son);
1825 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1826 son, attr, handle, spec);
1829 * in case of replay we just set LOVEA provided by the client
1830 * XXX: I think it would be interesting to try "old" way where
1831 * MDT calls this xattr_set(LOV) in a different transaction.
1832 * probably this way we code can be made better.
1834 if (rc == 0 && (spec->no_create ||
1835 (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
1836 const struct lu_buf *buf;
1838 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1839 spec->u.sp_ea.eadatalen);
1840 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
1844 if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1845 rc = __mdd_orphan_add(env, son, handle);
1847 mdd_write_unlock(env, son);
1851 * Object has no links, so it will be destroyed when last
1852 * reference is released. (XXX not now.)
1858 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE))
1859 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1860 name, S_ISDIR(attr->la_mode), handle,
1861 mdd_object_capa(env, mdd_pobj));
1868 if (S_ISLNK(attr->la_mode)) {
1869 struct lu_ucred *uc = lu_ucred_assert(env);
1870 struct dt_object *dt = mdd_object_child(son);
1871 const char *target_name = spec->u.sp_symname;
1872 int sym_len = strlen(target_name);
1873 const struct lu_buf *buf;
1876 buf = mdd_buf_get_const(env, target_name, sym_len);
1877 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1878 mdd_object_capa(env, son),
1880 CFS_CAP_SYS_RESOURCE_MASK);
1885 GOTO(cleanup, rc = -EFAULT);
1888 /* volatile file creation does not update parent directory times */
1889 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1890 GOTO(cleanup, rc = 0);
1892 /* update parent directory mtime/ctime */
1894 la->la_valid = LA_CTIME | LA_MTIME;
1895 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1901 if (rc != 0 && created != 0) {
1904 if (inserted != 0) {
1905 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1906 rc2 = __mdd_orphan_del(env, son, handle);
1908 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1909 S_ISDIR(attr->la_mode),
1910 handle, BYPASS_CAPA);
1915 mdd_write_lock(env, son, MOR_TGT_CHILD);
1916 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
1917 /* Drop the reference, no need to delete "."/"..",
1918 * because the object to be destroied directly. */
1919 rc2 = mdo_ref_del(env, son, handle);
1921 mdd_write_unlock(env, son);
1926 rc2 = mdo_ref_del(env, son, handle);
1928 mdd_write_unlock(env, son);
1932 mdo_destroy(env, son, handle);
1933 mdd_write_unlock(env, son);
1936 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1938 if (rc == 0 && fid_is_client_mdt_visible(mdo2fid(son)))
1939 rc = mdd_changelog_ns_store(env, mdd,
1940 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1941 S_ISREG(attr->la_mode) ? CL_CREATE :
1942 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1943 0, son, mdd_pobj, lname, handle);
1945 mdd_trans_stop(env, mdd, rc, handle);
1947 /* The child object shouldn't be cached anymore */
1949 set_bit(LU_OBJECT_HEARD_BANSHEE,
1950 &child->mo_lu.lo_header->loh_flags);
1955 * Get locks on parents in proper order
1956 * RETURN: < 0 - error, rename_order if successful
1964 static int mdd_rename_order(const struct lu_env *env,
1965 struct mdd_device *mdd,
1966 struct mdd_object *src_pobj,
1967 struct mdd_object *tgt_pobj)
1969 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1973 if (src_pobj == tgt_pobj)
1974 RETURN(MDD_RN_SAME);
1976 /* compared the parent child relationship of src_p&tgt_p */
1977 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1979 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1982 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1995 /* has not mdd_write{read}_lock on any obj yet. */
1996 static int mdd_rename_sanity_check(const struct lu_env *env,
1997 struct mdd_object *src_pobj,
1998 struct mdd_object *tgt_pobj,
1999 struct mdd_object *sobj,
2000 struct mdd_object *tobj,
2001 struct lu_attr *so_attr,
2002 struct lu_attr *tg_attr)
2007 /* XXX: when get here, sobj must NOT be NULL,
2008 * the other case has been processed in cml_rename
2009 * before mdd_rename and enable MDS_PERM_BYPASS. */
2012 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
2016 /* XXX: when get here, "tobj == NULL" means tobj must
2017 * NOT exist (neither on remote MDS, such case has been
2018 * processed in cml_rename before mdd_rename and enable
2020 * So check may_create, but not check may_unlink. */
2022 rc = mdd_may_create(env, tgt_pobj, NULL,
2023 (src_pobj != tgt_pobj), 0);
2025 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
2026 (src_pobj != tgt_pobj), 1);
2028 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2029 S_ISDIR(so_attr->la_mode))
2030 rc = __mdd_may_link(env, tgt_pobj);
2035 static int mdd_declare_rename(const struct lu_env *env,
2036 struct mdd_device *mdd,
2037 struct mdd_object *mdd_spobj,
2038 struct mdd_object *mdd_tpobj,
2039 struct mdd_object *mdd_sobj,
2040 struct mdd_object *mdd_tobj,
2041 const struct lu_name *tname,
2042 const struct lu_name *sname,
2044 struct thandle *handle)
2052 /* name from source dir */
2053 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2057 /* .. from source child */
2058 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2059 /* source child can be directory,
2060 * counted by source dir's nlink */
2061 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2065 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2069 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2074 /* new target child can be directory,
2075 * counted by target dir's nlink */
2076 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2082 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2086 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2089 mdd_declare_links_add(env, mdd_sobj, handle);
2093 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2098 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2099 tname->ln_name, handle);
2103 /* name from target dir (old name), we declare it unconditionally
2104 * as mdd_rename() calls delete unconditionally as well. so just
2105 * to balance declarations vs calls to change ... */
2106 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2110 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2111 /* delete target child in target parent directory */
2112 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2116 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2117 /* target child can be directory,
2118 * delete "." reference in target child directory */
2119 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2123 /* delete ".." reference in target parent directory */
2124 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2129 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2133 mdd_declare_links_add(env, mdd_tobj, handle);
2137 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2142 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2149 /* src object can be remote that is why we use only fid and type of object */
2150 static int mdd_rename(const struct lu_env *env,
2151 struct md_object *src_pobj, struct md_object *tgt_pobj,
2152 const struct lu_fid *lf, const struct lu_name *lsname,
2153 struct md_object *tobj, const struct lu_name *ltname,
2156 const char *sname = lsname->ln_name;
2157 const char *tname = ltname->ln_name;
2158 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2159 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2160 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2161 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2162 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2163 struct mdd_device *mdd = mdo2mdd(src_pobj);
2164 struct mdd_object *mdd_sobj = NULL; /* source object */
2165 struct mdd_object *mdd_tobj = NULL;
2166 struct dynlock_handle *sdlh, *tdlh;
2167 struct thandle *handle;
2168 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2169 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2172 bool tobj_locked = 0;
2173 unsigned cl_flags = 0;
2178 mdd_tobj = md2mdd_obj(tobj);
2180 mdd_sobj = mdd_object_find(env, mdd, lf);
2182 handle = mdd_trans_create(env, mdd);
2184 GOTO(out_pending, rc = PTR_ERR(handle));
2186 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2187 mdd_tobj, lsname, ltname, ma, handle);
2191 rc = mdd_trans_start(env, mdd, handle);
2195 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2196 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2198 GOTO(cleanup_unlocked, rc);
2200 /* Get locks in determined order */
2201 if (rc == MDD_RN_SAME) {
2202 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2203 sname, MOR_SRC_PARENT);
2204 /* check hashes to determine do we need one lock or two */
2205 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2206 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2210 } else if (rc == MDD_RN_SRCTGT) {
2211 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2212 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2214 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2215 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2217 if (sdlh == NULL || tdlh == NULL)
2218 GOTO(cleanup, rc = -ENOMEM);
2220 rc = mdd_la_get(env, mdd_sobj, so_attr,
2221 mdd_object_capa(env, mdd_sobj));
2226 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2227 mdd_object_capa(env, mdd_tobj));
2232 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2233 mdd_tobj, so_attr, tg_attr);
2237 is_dir = S_ISDIR(so_attr->la_mode);
2239 /* Remove source name from source directory */
2240 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2241 mdd_object_capa(env, mdd_spobj));
2245 /* "mv dir1 dir2" needs "dir1/.." link update */
2246 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2247 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2248 mdd_object_capa(env, mdd_sobj));
2250 GOTO(fixup_spobj2, rc);
2252 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2253 handle, mdd_object_capa(env, mdd_sobj));
2255 GOTO(fixup_spobj, rc);
2258 /* Remove target name from target directory
2259 * Here tobj can be remote one, so we do index_delete unconditionally
2260 * and -ENOENT is allowed.
2262 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2263 mdd_object_capa(env, mdd_tpobj));
2266 /* tname might been renamed to something else */
2267 GOTO(fixup_spobj, rc);
2270 GOTO(fixup_spobj, rc);
2273 /* Insert new fid with target name into target dir */
2274 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2275 mdd_object_capa(env, mdd_tpobj));
2277 GOTO(fixup_tpobj, rc);
2279 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2280 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2282 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2284 la->la_valid = LA_CTIME;
2285 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2287 GOTO(fixup_tpobj, rc);
2290 /* Remove old target object
2291 * For tobj is remote case cmm layer has processed
2292 * and set tobj to NULL then. So when tobj is NOT NULL,
2293 * it must be local one.
2295 if (tobj && mdd_object_exists(mdd_tobj)) {
2296 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2298 if (mdd_is_dead_obj(mdd_tobj)) {
2299 /* shld not be dead, something is wrong */
2300 CERROR("tobj is dead, something is wrong\n");
2304 mdo_ref_del(env, mdd_tobj, handle);
2306 /* Remove dot reference. */
2307 if (S_ISDIR(tg_attr->la_mode))
2308 mdo_ref_del(env, mdd_tobj, handle);
2311 /* fetch updated nlink */
2312 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2313 mdd_object_capa(env, mdd_tobj));
2315 CERROR("%s: Failed to get nlink for tobj "
2317 mdd2obd_dev(mdd)->obd_name,
2318 PFID(tpobj_fid), rc);
2319 GOTO(fixup_tpobj, rc);
2322 la->la_valid = LA_CTIME;
2323 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2325 CERROR("%s: Failed to set ctime for tobj "
2327 mdd2obd_dev(mdd)->obd_name,
2328 PFID(tpobj_fid), rc);
2329 GOTO(fixup_tpobj, rc);
2332 /* XXX: this transfer to ma will be removed with LOD/OSP */
2333 ma->ma_attr = *tg_attr;
2334 ma->ma_valid |= MA_INODE;
2335 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2337 CERROR("%s: Failed to unlink tobj "
2339 mdd2obd_dev(mdd)->obd_name,
2340 PFID(tpobj_fid), rc);
2341 GOTO(fixup_tpobj, rc);
2344 /* fetch updated nlink */
2345 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2346 mdd_object_capa(env, mdd_tobj));
2348 CERROR("%s: Failed to get nlink for tobj "
2350 mdd2obd_dev(mdd)->obd_name,
2351 PFID(tpobj_fid), rc);
2352 GOTO(fixup_tpobj, rc);
2354 /* XXX: this transfer to ma will be removed with LOD/OSP */
2355 ma->ma_attr = *tg_attr;
2356 ma->ma_valid |= MA_INODE;
2358 if (so_attr->la_nlink == 0)
2359 cl_flags |= CLF_RENAME_LAST;
2362 la->la_valid = LA_CTIME | LA_MTIME;
2363 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2365 GOTO(fixup_tpobj, rc);
2367 if (mdd_spobj != mdd_tpobj) {
2368 la->la_valid = LA_CTIME | LA_MTIME;
2369 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2373 if (rc == 0 && mdd_sobj) {
2374 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2375 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2376 mdo2fid(mdd_tpobj), ltname, handle, 0, 0);
2378 /* Old files might not have EA entry */
2379 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2381 mdd_write_unlock(env, mdd_sobj);
2382 /* We don't fail the transaction if the link ea can't be
2383 updated -- fid2path will use alternate lookup method. */
2391 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2394 CWARN("tp obj fix error %d\n",rc2);
2396 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2397 !mdd_is_dead_obj(mdd_tobj)) {
2399 mdo_ref_add(env, mdd_tobj, handle);
2401 mdo_ref_add(env, mdd_tobj, handle);
2404 rc2 = __mdd_index_insert(env, mdd_tpobj,
2405 mdo2fid(mdd_tobj), tname,
2410 CWARN("tp obj fix error %d\n",rc2);
2415 if (rc && is_dir && mdd_sobj) {
2416 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2420 CWARN("sp obj dotdot delete error %d\n",rc2);
2423 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2424 dotdot, handle, BYPASS_CAPA);
2426 CWARN("sp obj dotdot insert error %d\n",rc2);
2431 rc2 = __mdd_index_insert(env, mdd_spobj,
2432 lf, sname, is_dir, handle, BYPASS_CAPA);
2434 CWARN("sp obj fix error %d\n",rc2);
2438 mdd_write_unlock(env, mdd_tobj);
2439 if (likely(tdlh) && sdlh != tdlh)
2440 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2442 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2445 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2446 mdd_tobj, tpobj_fid, lf,
2447 spobj_fid, ltname, lsname,
2451 mdd_trans_stop(env, mdd, rc, handle);
2453 mdd_object_put(env, mdd_sobj);
2459 * The data that link search is done on.
2461 struct mdd_link_data {
2463 * Buffer to keep link EA body.
2465 struct lu_buf *ml_buf;
2467 * The matched header, entry and its lenght in the EA
2469 struct link_ea_header *ml_leh;
2470 struct link_ea_entry *ml_lee;
2474 static int mdd_links_new(const struct lu_env *env,
2475 struct mdd_link_data *ldata)
2477 ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2478 if (ldata->ml_buf->lb_buf == NULL)
2480 ldata->ml_leh = ldata->ml_buf->lb_buf;
2481 ldata->ml_leh->leh_magic = LINK_EA_MAGIC;
2482 ldata->ml_leh->leh_len = sizeof(struct link_ea_header);
2483 ldata->ml_leh->leh_reccount = 0;
2487 /** Read the link EA into a temp buffer.
2488 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
2489 * A pointer to the buffer is stored in \a ldata::ml_buf.
2491 * \retval 0 or error
2493 int mdd_links_read(const struct lu_env *env,
2494 struct mdd_object *mdd_obj,
2495 struct mdd_link_data *ldata)
2497 struct lustre_capa *capa;
2498 struct link_ea_header *leh;
2501 /* First try a small buf */
2502 LASSERT(env != NULL);
2503 ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2504 if (ldata->ml_buf->lb_buf == NULL)
2507 if (!mdd_object_exists(mdd_obj))
2510 capa = mdd_object_capa(env, mdd_obj);
2511 rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
2512 XATTR_NAME_LINK, capa);
2513 if (rc == -ERANGE) {
2514 /* Buf was too small, figure out what we need. */
2515 mdd_buf_put(ldata->ml_buf);
2516 rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
2517 XATTR_NAME_LINK, capa);
2520 ldata->ml_buf = mdd_buf_alloc(env, rc);
2521 if (ldata->ml_buf->lb_buf == NULL)
2523 rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
2524 XATTR_NAME_LINK, capa);
2529 leh = ldata->ml_buf->lb_buf;
2530 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2531 leh->leh_magic = LINK_EA_MAGIC;
2532 leh->leh_reccount = __swab32(leh->leh_reccount);
2533 leh->leh_len = __swab64(leh->leh_len);
2534 /* entries are swabbed by mdd_lee_unpack */
2536 if (leh->leh_magic != LINK_EA_MAGIC)
2538 if (leh->leh_reccount == 0)
2541 ldata->ml_leh = leh;
2545 /** Read the link EA into a temp buffer.
2546 * Uses the name_buf since it is generally large.
2547 * \retval IS_ERR err
2548 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2550 struct lu_buf *mdd_links_get(const struct lu_env *env,
2551 struct mdd_object *mdd_obj)
2553 struct mdd_link_data ldata = { 0 };
2556 rc = mdd_links_read(env, mdd_obj, &ldata);
2557 return rc ? ERR_PTR(rc) : ldata.ml_buf;
2560 static int mdd_links_write(const struct lu_env *env,
2561 struct mdd_object *mdd_obj,
2562 struct mdd_link_data *ldata,
2563 struct thandle *handle)
2565 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ml_buf->lb_buf,
2566 ldata->ml_leh->leh_len);
2567 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
2568 mdd_object_capa(env, mdd_obj));
2571 /** Pack a link_ea_entry.
2572 * All elements are stored as chars to avoid alignment issues.
2573 * Numbers are always big-endian
2574 * \retval record length
2576 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2577 const struct lu_fid *pfid)
2579 struct lu_fid tmpfid;
2582 fid_cpu_to_be(&tmpfid, pfid);
2583 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2584 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2585 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2587 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2588 lee->lee_reclen[1] = reclen & 0xff;
2592 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2593 struct lu_name *lname, struct lu_fid *pfid)
2595 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2596 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2597 fid_be_to_cpu(pfid, pfid);
2598 lname->ln_name = lee->lee_name;
2599 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2602 static int mdd_declare_links_add(const struct lu_env *env,
2603 struct mdd_object *mdd_obj,
2604 struct thandle *handle)
2608 /* XXX: max size? */
2609 rc = mdo_declare_xattr_set(env, mdd_obj,
2610 mdd_buf_get_const(env, NULL, 4096),
2611 XATTR_NAME_LINK, 0, handle);
2616 /* For pathologic linkers, we don't want to spend lots of time scanning the
2617 * link ea. Limit ourseleves to something reasonable; links not in the EA
2618 * can be looked up via (slower) parent lookup.
2620 #define LINKEA_MAX_COUNT 128
2622 /** Add a record to the end of link ea buf */
2623 static int mdd_links_add_buf(const struct lu_env *env,
2624 struct mdd_link_data *ldata,
2625 const struct lu_name *lname,
2626 const struct lu_fid *pfid)
2628 LASSERT(ldata->ml_leh != NULL);
2630 if (lname == NULL || pfid == NULL)
2633 /* Make sure our buf is big enough for the new one */
2634 if (ldata->ml_leh->leh_reccount > LINKEA_MAX_COUNT)
2637 ldata->ml_reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2638 if (ldata->ml_leh->leh_len + ldata->ml_reclen >
2639 ldata->ml_buf->lb_len) {
2640 if (mdd_buf_grow(env, ldata->ml_leh->leh_len +
2641 ldata->ml_reclen) < 0)
2645 ldata->ml_leh = ldata->ml_buf->lb_buf;
2646 ldata->ml_lee = ldata->ml_buf->lb_buf + ldata->ml_leh->leh_len;
2647 ldata->ml_reclen = mdd_lee_pack(ldata->ml_lee, lname, pfid);
2648 ldata->ml_leh->leh_len += ldata->ml_reclen;
2649 ldata->ml_leh->leh_reccount++;
2650 CDEBUG(D_INODE, "New link_ea name '%.*s' is added\n",
2651 lname->ln_namelen, lname->ln_name);
2655 /** Del the current record from the link ea buf */
2656 static void mdd_links_del_buf(const struct lu_env *env,
2657 struct mdd_link_data *ldata,
2658 const struct lu_name *lname)
2660 LASSERT(ldata->ml_leh != NULL);
2662 ldata->ml_leh->leh_reccount--;
2663 ldata->ml_leh->leh_len -= ldata->ml_reclen;
2664 memmove(ldata->ml_lee, (char *)ldata->ml_lee + ldata->ml_reclen,
2665 (char *)ldata->ml_leh + ldata->ml_leh->leh_len -
2666 (char *)ldata->ml_lee);
2667 CDEBUG(D_INODE, "Old link_ea name '%.*s' is removed\n",
2668 lname->ln_namelen, lname->ln_name);
2673 * Check if such a link exists in linkEA.
2675 * \param mdd_obj object being handled
2676 * \param pfid parent fid the link to be found for
2677 * \param lname name in the parent's directory entry pointing to this object
2678 * \param ldata link data the search to be done on
2681 * \retval -ENOENT link does not exist
2682 * \retval -ve on error
2684 static int mdd_links_find(const struct lu_env *env,
2685 struct mdd_object *mdd_obj,
2686 struct mdd_link_data *ldata,
2687 const struct lu_name *lname,
2688 const struct lu_fid *pfid)
2690 struct lu_name *tmpname = &mdd_env_info(env)->mti_name2;
2691 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2694 LASSERT(ldata->ml_leh != NULL);
2697 ldata->ml_lee = (struct link_ea_entry *)(ldata->ml_leh + 1);
2699 for (count = 0; count < ldata->ml_leh->leh_reccount; count++) {
2700 mdd_lee_unpack(ldata->ml_lee, &ldata->ml_reclen,
2702 if (tmpname->ln_namelen == lname->ln_namelen &&
2703 lu_fid_eq(tmpfid, pfid) &&
2704 (strncmp(tmpname->ln_name, lname->ln_name,
2705 tmpname->ln_namelen) == 0))
2707 ldata->ml_lee = (struct link_ea_entry *)((char *)ldata->ml_lee +
2711 if (count == ldata->ml_leh->leh_reccount) {
2712 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2713 lname->ln_namelen, lname->ln_name);
2719 static int __mdd_links_add(const struct lu_env *env,
2720 struct mdd_object *mdd_obj,
2721 struct mdd_link_data *ldata,
2722 const struct lu_name *lname,
2723 const struct lu_fid *pfid,
2724 int first, int check)
2728 if (ldata->ml_leh == NULL) {
2729 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
2733 rc = mdd_links_new(env, ldata);
2740 rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid);
2741 if (rc && rc != -ENOENT)
2747 return mdd_links_add_buf(env, ldata, lname, pfid);
2750 static int __mdd_links_del(const struct lu_env *env,
2751 struct mdd_object *mdd_obj,
2752 struct mdd_link_data *ldata,
2753 const struct lu_name *lname,
2754 const struct lu_fid *pfid)
2758 if (ldata->ml_leh == NULL) {
2759 rc = mdd_links_read(env, mdd_obj, ldata);
2764 rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid);
2768 mdd_links_del_buf(env, ldata, lname);
2772 static int mdd_links_rename(const struct lu_env *env,
2773 struct mdd_object *mdd_obj,
2774 const struct lu_fid *oldpfid,
2775 const struct lu_name *oldlname,
2776 const struct lu_fid *newpfid,
2777 const struct lu_name *newlname,
2778 struct thandle *handle,
2779 int first, int check)
2781 struct mdd_link_data ldata = { 0 };
2787 LASSERT(oldpfid != NULL || newpfid != NULL);
2789 if (mdd_obj->mod_flags & DEAD_OBJ)
2790 /* No more links, don't bother */
2793 if (oldpfid != NULL) {
2794 rc = __mdd_links_del(env, mdd_obj, &ldata,
2798 (rc != -ENODATA && rc != -ENOENT))
2800 /* No changes done. */
2807 /* If renaming, add the new record */
2808 if (newpfid != NULL) {
2809 /* even if the add fails, we still delete the out-of-date
2811 rc2 = __mdd_links_add(env, mdd_obj, &ldata,
2812 newlname, newpfid, first, check);
2820 rc = mdd_links_write(env, mdd_obj, &ldata, handle);
2827 if (rc == -EOVERFLOW || rc == - ENOENT)
2829 if (oldpfid == NULL)
2830 CDEBUG(error ? D_ERROR : D_OTHER,
2831 "link_ea add '%.*s' failed %d "DFID"\n",
2832 newlname->ln_namelen, newlname->ln_name,
2833 rc, PFID(mdd_object_fid(mdd_obj)));
2834 else if (newpfid == NULL)
2835 CDEBUG(error ? D_ERROR : D_OTHER,
2836 "link_ea del '%.*s' failed %d "DFID"\n",
2837 oldlname->ln_namelen, oldlname->ln_name,
2838 rc, PFID(mdd_object_fid(mdd_obj)));
2840 CDEBUG(error ? D_ERROR : D_OTHER,
2841 "link_ea rename '%.*s'->'%.*s' failed %d "
2843 oldlname->ln_namelen, oldlname->ln_name,
2844 newlname->ln_namelen, newlname->ln_name,
2845 rc, PFID(mdd_object_fid(mdd_obj)));
2848 if (ldata.ml_buf && ldata.ml_buf->lb_len > OBD_ALLOC_BIG)
2849 /* if we vmalloced a large buffer drop it */
2850 mdd_buf_put(ldata.ml_buf);
2855 static inline int mdd_links_add(const struct lu_env *env,
2856 struct mdd_object *mdd_obj,
2857 const struct lu_fid *pfid,
2858 const struct lu_name *lname,
2859 struct thandle *handle, int first)
2861 return mdd_links_rename(env, mdd_obj, NULL, NULL,
2862 pfid, lname, handle, first, 0);
2865 static inline int mdd_links_del(const struct lu_env *env,
2866 struct mdd_object *mdd_obj,
2867 const struct lu_fid *pfid,
2868 const struct lu_name *lname,
2869 struct thandle *handle)
2871 return mdd_links_rename(env, mdd_obj, pfid, lname,
2872 NULL, NULL, handle, 0, 0);
2875 const struct md_dir_operations mdd_dir_ops = {
2876 .mdo_is_subdir = mdd_is_subdir,
2877 .mdo_lookup = mdd_lookup,
2878 .mdo_create = mdd_create,
2879 .mdo_rename = mdd_rename,
2880 .mdo_link = mdd_link,
2881 .mdo_unlink = mdd_unlink,
2882 .mdo_create_data = mdd_create_data,