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 !mdd_capable(uc, CFS_CAP_FOWNER);
376 * Check whether it may delete the cobj from the pobj.
379 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
380 struct mdd_object *cobj, struct lu_attr *cattr,
381 struct lu_attr *src_attr, int check_perm, int check_empty)
387 if (!mdd_object_exists(cobj))
390 if (mdd_is_dead_obj(cobj))
394 if (!mdd_object_exists(pobj))
397 if (mdd_is_dead_obj(pobj))
401 rc = mdd_permission_internal_locked(env, pobj, NULL,
402 MAY_WRITE | MAY_EXEC,
408 if (mdd_is_append(pobj))
412 if (mdd_is_sticky(env, pobj, cobj))
415 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
418 if ((cattr->la_valid & LA_FLAGS) &&
419 (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
422 /* additional check the rename case */
424 if (S_ISDIR(src_attr->la_mode)) {
425 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
427 if (!S_ISDIR(cattr->la_mode))
430 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
432 } else if (S_ISDIR(cattr->la_mode))
436 if (S_ISDIR(cattr->la_mode) && check_empty)
437 rc = mdd_dir_is_empty(env, cobj);
444 * has mdd_write_lock on src already, but not on tgt yet
446 int mdd_link_sanity_check(const struct lu_env *env,
447 struct mdd_object *tgt_obj,
448 const struct lu_name *lname,
449 struct mdd_object *src_obj)
451 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
455 if (!mdd_object_exists(src_obj))
458 if (mdd_is_dead_obj(src_obj))
461 /* Local ops, no lookup before link, check filename length here. */
462 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
463 RETURN(-ENAMETOOLONG);
465 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
468 if (S_ISDIR(mdd_object_type(src_obj)))
471 LASSERT(src_obj != tgt_obj);
473 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
478 rc = __mdd_may_link(env, src_obj);
483 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
484 const char *name, struct thandle *handle,
485 struct lustre_capa *capa)
487 struct dt_object *next = mdd_object_child(pobj);
491 if (dt_try_as_dir(env, next)) {
492 rc = next->do_index_ops->dio_delete(env, next,
493 (struct dt_key *)name,
501 static int __mdd_index_insert_only(const struct lu_env *env,
502 struct mdd_object *pobj,
503 const struct lu_fid *lf, const char *name,
504 struct thandle *handle,
505 struct lustre_capa *capa)
507 struct dt_object *next = mdd_object_child(pobj);
511 if (dt_try_as_dir(env, next)) {
512 struct lu_ucred *uc = lu_ucred_check(env);
515 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
516 rc = next->do_index_ops->dio_insert(env, next,
518 (const struct dt_key *)name,
519 handle, capa, ignore_quota);
526 /* insert named index, add reference if isdir */
527 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
528 const struct lu_fid *lf, const char *name, int is_dir,
529 struct thandle *handle, struct lustre_capa *capa)
534 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
535 if (rc == 0 && is_dir) {
536 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
537 mdo_ref_add(env, pobj, handle);
538 mdd_write_unlock(env, pobj);
543 /* delete named index, drop reference if isdir */
544 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
545 const char *name, int is_dir, struct thandle *handle,
546 struct lustre_capa *capa)
551 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
552 if (rc == 0 && is_dir) {
553 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
554 mdo_ref_del(env, pobj, handle);
555 mdd_write_unlock(env, pobj);
561 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
562 int reclen, struct thandle *handle)
566 /* XXX: this is a temporary solution to declare llog changes
567 * will be fixed in 2.3 with new llog implementation */
569 LASSERT(mdd->mdd_capa);
571 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
572 * have to set the parameter 'size' as INT_MAX or 0 to inform
573 * OSD that this record write is for a llog write or catalog
574 * header update, and osd declare function will reserve less
575 * credits for optimization purpose.
577 * Reserve 6 blocks for a llog write, since the llog file is
578 * usually small, reserve 2 blocks for catalog header update,
579 * because we know for sure that catalog header is already
582 * This hack should be removed in 2.3.
586 rc = dt_declare_record_write(env, mdd->mdd_capa,
587 DECLARE_LLOG_WRITE, 0, handle);
591 /* header will be updated as well */
592 rc = dt_declare_record_write(env, mdd->mdd_capa,
593 DECLARE_LLOG_WRITE, 0, handle);
597 /* also we should be able to create new plain log */
598 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
602 /* new record referencing new plain llog */
603 rc = dt_declare_record_write(env, mdd->mdd_capa,
604 DECLARE_LLOG_WRITE, 0, handle);
608 /* catalog's header will be updated as well */
609 rc = dt_declare_record_write(env, mdd->mdd_capa,
610 DECLARE_LLOG_REWRITE, 0, handle);
615 int mdd_declare_changelog_store(const struct lu_env *env,
616 struct mdd_device *mdd,
617 const struct lu_name *fname,
618 struct thandle *handle)
620 struct obd_device *obd = mdd2obd_dev(mdd);
621 struct llog_ctxt *ctxt;
622 struct llog_changelog_rec *rec;
628 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
631 reclen = llog_data_len(sizeof(*rec) +
632 (fname != NULL ? fname->ln_namelen : 0));
633 buf = mdd_buf_alloc(env, reclen);
634 if (buf->lb_buf == NULL)
638 rec->cr_hdr.lrh_len = reclen;
639 rec->cr_hdr.lrh_type = CHANGELOG_REC;
641 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
645 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
651 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
652 struct mdd_device *mdd,
653 const struct lu_name *tname,
654 const struct lu_name *sname,
655 struct thandle *handle)
657 struct obd_device *obd = mdd2obd_dev(mdd);
658 struct llog_ctxt *ctxt;
659 struct llog_changelog_ext_rec *rec;
665 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
668 reclen = llog_data_len(sizeof(*rec) +
669 (tname != NULL ? tname->ln_namelen : 0) +
670 (sname != NULL ? 1 + sname->ln_namelen : 0));
671 buf = mdd_buf_alloc(env, reclen);
672 if (buf->lb_buf == NULL)
676 rec->cr_hdr.lrh_len = reclen;
677 rec->cr_hdr.lrh_type = CHANGELOG_REC;
679 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
683 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
689 /** Add a changelog entry \a rec to the changelog llog
692 * \param handle - currently ignored since llogs start their own transaction;
693 * this will hopefully be fixed in llog rewrite
696 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
697 struct llog_changelog_rec *rec, struct thandle *th)
699 struct obd_device *obd = mdd2obd_dev(mdd);
700 struct llog_ctxt *ctxt;
703 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
704 rec->cr_hdr.lrh_type = CHANGELOG_REC;
705 rec->cr.cr_time = cl_time();
707 spin_lock(&mdd->mdd_cl.mc_lock);
708 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
709 * but as long as the MDD transactions are ordered correctly for e.g.
710 * rename conflicts, I don't think this should matter. */
711 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
712 spin_unlock(&mdd->mdd_cl.mc_lock);
714 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
718 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
725 /** Add a changelog_ext entry \a rec to the changelog llog
728 * \param handle - currently ignored since llogs start their own transaction;
729 * this will hopefully be fixed in llog rewrite
732 int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
733 struct llog_changelog_ext_rec *rec,
736 struct obd_device *obd = mdd2obd_dev(mdd);
737 struct llog_ctxt *ctxt;
740 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
741 /* llog_lvfs_write_rec sets the llog tail len */
742 rec->cr_hdr.lrh_type = CHANGELOG_REC;
743 rec->cr.cr_time = cl_time();
745 spin_lock(&mdd->mdd_cl.mc_lock);
746 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
747 * but as long as the MDD transactions are ordered correctly for e.g.
748 * rename conflicts, I don't think this should matter. */
749 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
750 spin_unlock(&mdd->mdd_cl.mc_lock);
752 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
756 /* nested journal transaction */
757 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
765 /** Store a namespace change changelog record
766 * If this fails, we must fail the whole transaction; we don't
767 * want the change to commit without the log entry.
768 * \param target - mdd_object of change
769 * \param parent - parent dir/object
770 * \param tname - target name string
771 * \param handle - transacion handle
773 static int mdd_changelog_ns_store(const struct lu_env *env,
774 struct mdd_device *mdd,
775 enum changelog_rec_type type,
777 struct mdd_object *target,
778 struct mdd_object *parent,
779 const struct lu_name *tname,
780 struct thandle *handle)
782 struct llog_changelog_rec *rec;
789 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
791 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
794 LASSERT(target != NULL);
795 LASSERT(parent != NULL);
796 LASSERT(tname != NULL);
797 LASSERT(handle != NULL);
799 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
800 buf = mdd_buf_alloc(env, reclen);
801 if (buf->lb_buf == NULL)
805 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
806 rec->cr.cr_type = (__u32)type;
807 rec->cr.cr_tfid = *mdo2fid(target);
808 rec->cr.cr_pfid = *mdo2fid(parent);
809 rec->cr.cr_namelen = tname->ln_namelen;
810 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
812 target->mod_cltime = cfs_time_current_64();
814 rc = mdd_changelog_store(env, mdd, rec, handle);
816 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
817 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
818 PFID(&rec->cr.cr_pfid));
826 /** Store a namespace change changelog record
827 * If this fails, we must fail the whole transaction; we don't
828 * want the change to commit without the log entry.
829 * \param target - mdd_object of change
830 * \param tpfid - target parent dir/object fid
831 * \param sfid - source object fid
832 * \param spfid - source parent fid
833 * \param tname - target name string
834 * \param sname - source name string
835 * \param handle - transacion handle
837 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
838 struct mdd_device *mdd,
839 enum changelog_rec_type type,
841 struct mdd_object *target,
842 const struct lu_fid *tpfid,
843 const struct lu_fid *sfid,
844 const struct lu_fid *spfid,
845 const struct lu_name *tname,
846 const struct lu_name *sname,
847 struct thandle *handle)
849 struct llog_changelog_ext_rec *rec;
856 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
858 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
861 LASSERT(sfid != NULL);
862 LASSERT(tpfid != NULL);
863 LASSERT(tname != NULL);
864 LASSERT(handle != NULL);
866 reclen = llog_data_len(sizeof(*rec) +
867 sname != NULL ? 1 + sname->ln_namelen : 0);
868 buf = mdd_buf_alloc(env, reclen);
869 if (buf->lb_buf == NULL)
873 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
874 rec->cr.cr_type = (__u32)type;
875 rec->cr.cr_pfid = *tpfid;
876 rec->cr.cr_sfid = *sfid;
877 rec->cr.cr_spfid = *spfid;
878 rec->cr.cr_namelen = tname->ln_namelen;
879 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
881 rec->cr.cr_name[tname->ln_namelen] = '\0';
882 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
884 rec->cr.cr_namelen += 1 + sname->ln_namelen;
887 if (likely(target != NULL)) {
888 rec->cr.cr_tfid = *mdo2fid(target);
889 target->mod_cltime = cfs_time_current_64();
891 fid_zero(&rec->cr.cr_tfid);
894 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
896 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
897 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
904 static int mdd_declare_link(const struct lu_env *env,
905 struct mdd_device *mdd,
906 struct mdd_object *p,
907 struct mdd_object *c,
908 const struct lu_name *name,
909 struct thandle *handle)
913 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
917 rc = mdo_declare_ref_add(env, c, handle);
921 rc = mdo_declare_attr_set(env, p, NULL, handle);
925 rc = mdo_declare_attr_set(env, c, NULL, handle);
929 rc = mdd_declare_links_add(env, c, handle);
933 rc = mdd_declare_changelog_store(env, mdd, name, handle);
938 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
939 struct md_object *src_obj, const struct lu_name *lname,
942 const char *name = lname->ln_name;
943 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
944 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
945 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
946 struct mdd_device *mdd = mdo2mdd(src_obj);
947 struct dynlock_handle *dlh;
948 struct thandle *handle;
952 handle = mdd_trans_create(env, mdd);
954 GOTO(out_pending, rc = PTR_ERR(handle));
956 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
960 rc = mdd_trans_start(env, mdd, handle);
964 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
966 GOTO(out_trans, rc = -ENOMEM);
967 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
969 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
971 GOTO(out_unlock, rc);
973 rc = mdo_ref_add(env, mdd_sobj, handle);
975 GOTO(out_unlock, rc);
978 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
980 mdd_object_capa(env, mdd_tobj));
982 mdo_ref_del(env, mdd_sobj, handle);
983 GOTO(out_unlock, rc);
986 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
987 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
989 la->la_valid = LA_CTIME | LA_MTIME;
990 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
992 GOTO(out_unlock, rc);
994 la->la_valid = LA_CTIME;
995 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
997 mdd_links_add(env, mdd_sobj,
998 mdo2fid(mdd_tobj), lname, handle, 0);
1003 mdd_write_unlock(env, mdd_sobj);
1004 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
1007 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1008 mdd_tobj, lname, handle);
1010 mdd_trans_stop(env, mdd, rc, handle);
1015 int mdd_declare_finish_unlink(const struct lu_env *env,
1016 struct mdd_object *obj,
1018 struct thandle *handle)
1022 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1026 return mdo_declare_destroy(env, obj, handle);
1029 /* caller should take a lock before calling */
1030 int mdd_finish_unlink(const struct lu_env *env,
1031 struct mdd_object *obj, struct md_attr *ma,
1035 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1038 LASSERT(mdd_write_locked(env, obj) != 0);
1040 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
1041 obj->mod_flags |= DEAD_OBJ;
1042 /* add new orphan and the object
1043 * will be deleted during mdd_close() */
1044 if (obj->mod_count) {
1045 rc = __mdd_orphan_add(env, obj, th);
1047 CDEBUG(D_HA, "Object "DFID" is inserted into "
1048 "orphan list, open count = %d\n",
1049 PFID(mdd_object_fid(obj)),
1052 CERROR("Object "DFID" fail to be an orphan, "
1053 "open count = %d, maybe cause failed "
1055 PFID(mdd_object_fid(obj)),
1058 rc = mdo_destroy(env, obj, th);
1067 * has mdd_write_lock on cobj already, but not on pobj yet
1069 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1070 struct mdd_object *cobj, struct lu_attr *cattr)
1075 rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1);
1080 static inline int mdd_declare_links_del(const struct lu_env *env,
1081 struct mdd_object *c,
1082 struct thandle *handle)
1086 /* For directory, the linkEA will be removed together with the object. */
1087 if (!S_ISDIR(mdd_object_type(c)))
1088 rc = mdd_declare_links_add(env, c, handle);
1093 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1094 struct mdd_object *p, struct mdd_object *c,
1095 const struct lu_name *name, struct md_attr *ma,
1096 struct thandle *handle)
1100 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1104 rc = mdo_declare_ref_del(env, p, handle);
1108 rc = mdo_declare_ref_del(env, c, handle);
1112 rc = mdo_declare_ref_del(env, c, handle);
1116 rc = mdo_declare_attr_set(env, p, NULL, handle);
1120 rc = mdo_declare_attr_set(env, c, NULL, handle);
1124 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1128 rc = mdd_declare_links_del(env, c, handle);
1132 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1137 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1138 struct md_object *cobj, const struct lu_name *lname,
1141 const char *name = lname->ln_name;
1142 struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr;
1143 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1144 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1145 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
1146 struct mdd_device *mdd = mdo2mdd(pobj);
1147 struct dynlock_handle *dlh;
1148 struct thandle *handle;
1152 if (mdd_object_exists(mdd_cobj) <= 0)
1155 handle = mdd_trans_create(env, mdd);
1157 RETURN(PTR_ERR(handle));
1159 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1164 rc = mdd_trans_start(env, mdd, handle);
1168 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1170 GOTO(stop, rc = -ENOMEM);
1171 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1174 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1178 is_dir = S_ISDIR(cattr->la_mode);
1180 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr);
1184 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1185 mdd_object_capa(env, mdd_pobj));
1189 rc = mdo_ref_del(env, mdd_cobj, handle);
1191 __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj),
1193 mdd_object_capa(env, mdd_pobj));
1199 mdo_ref_del(env, mdd_cobj, handle);
1201 /* fetch updated nlink */
1202 rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj));
1206 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1207 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1209 la->la_valid = LA_CTIME | LA_MTIME;
1210 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1214 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1215 /* update ctime of an unlinked file only if it is still
1216 * opened or a link still exists */
1217 la->la_valid = LA_CTIME;
1218 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1223 /* XXX: this transfer to ma will be removed with LOD/OSP */
1224 ma->ma_attr = *cattr;
1225 ma->ma_valid |= MA_INODE;
1226 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1228 /* fetch updated nlink */
1230 rc = mdd_la_get(env, mdd_cobj, cattr,
1231 mdd_object_capa(env, mdd_cobj));
1234 /* old files may not have link ea; ignore errors */
1235 mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle);
1237 /* if object is removed then we can't get its attrs, use last get */
1238 if (cattr->la_nlink == 0) {
1239 ma->ma_attr = *cattr;
1240 ma->ma_valid |= MA_INODE;
1244 mdd_write_unlock(env, mdd_cobj);
1245 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1249 cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1250 if ((ma->ma_valid & MA_HSM) &&
1251 (ma->ma_hsm.mh_flags & HS_EXISTS))
1252 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1254 rc = mdd_changelog_ns_store(env, mdd,
1255 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1256 mdd_cobj, mdd_pobj, lname, handle);
1260 mdd_trans_stop(env, mdd, rc, handle);
1266 * The permission has been checked when obj created, no need check again.
1268 static int mdd_cd_sanity_check(const struct lu_env *env,
1269 struct mdd_object *obj)
1274 if (!obj || mdd_is_dead_obj(obj))
1281 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1282 struct md_object *cobj, const struct md_op_spec *spec,
1285 struct mdd_device *mdd = mdo2mdd(cobj);
1286 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1287 struct mdd_object *son = md2mdd_obj(cobj);
1288 struct thandle *handle;
1289 const struct lu_buf *buf;
1290 struct lu_attr *attr = &mdd_env_info(env)->mti_cattr;
1294 /* do not let users to create stripes via .lustre/
1295 * mdd_obf_setup() sets IMMUTE_OBJ on this directory */
1296 if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ)
1299 rc = mdd_cd_sanity_check(env, son);
1303 if (!md_should_create(spec->sp_cr_flags))
1307 * there are following use cases for this function:
1308 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1309 * striping can be specified or not
1312 rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son));
1316 /* calling ->ah_make_hint() is used to transfer information from parent */
1317 mdd_object_make_hint(env, mdd_pobj, son, attr);
1319 handle = mdd_trans_create(env, mdd);
1321 GOTO(out_free, rc = PTR_ERR(handle));
1324 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1325 * Should this be fixed?
1327 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, no_create %u\n",
1328 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1329 spec->sp_cr_flags, spec->no_create);
1331 if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) {
1332 /* replay case or lfs setstripe */
1333 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1334 spec->u.sp_ea.eadatalen);
1339 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1340 XATTR_NAME_LOV, 0, handle);
1344 rc = mdd_trans_start(env, mdd, handle);
1348 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1349 0, handle, mdd_object_capa(env, son));
1351 mdd_trans_stop(env, mdd, rc, handle);
1356 /* Get fid from name and parent */
1358 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1359 const struct lu_name *lname, struct lu_fid* fid, int mask)
1361 const char *name = lname->ln_name;
1362 const struct dt_key *key = (const struct dt_key *)name;
1363 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1364 struct mdd_device *m = mdo2mdd(pobj);
1365 struct dt_object *dir = mdd_object_child(mdd_obj);
1369 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1372 rc = mdd_object_exists(mdd_obj);
1373 if (unlikely(rc == 0))
1375 else if (unlikely(rc < 0)) {
1376 CERROR("Object "DFID" locates on remote server\n",
1377 PFID(mdo2fid(mdd_obj)));
1381 /* The common filename length check. */
1382 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1383 RETURN(-ENAMETOOLONG);
1385 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1390 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1391 dt_try_as_dir(env, dir))) {
1393 rc = dir->do_index_ops->dio_lookup(env, dir,
1394 (struct dt_rec *)fid, key,
1395 mdd_object_capa(env, mdd_obj));
1406 int mdd_declare_object_initialize(const struct lu_env *env,
1407 struct mdd_object *child,
1408 struct lu_attr *attr,
1409 struct thandle *handle)
1414 * inode mode has been set in creation time, and it's based on umask,
1415 * la_mode and acl, don't set here again! (which will go wrong
1416 * because below function doesn't consider umask).
1417 * I'd suggest set all object attributes in creation time, see above.
1419 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1420 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1421 rc = mdo_declare_attr_set(env, child, attr, handle);
1422 attr->la_valid |= LA_MODE | LA_TYPE;
1423 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1424 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1427 rc = mdo_declare_ref_add(env, child, handle);
1431 mdd_declare_links_add(env, child, handle);
1436 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1437 const struct lu_name *lname, struct mdd_object *child,
1438 struct lu_attr *attr, struct thandle *handle,
1439 const struct md_op_spec *spec)
1445 * Update attributes for child.
1448 * (1) the valid bits should be converted between Lustre and Linux;
1449 * (2) maybe, the child attributes should be set in OSD when creation.
1453 * inode mode has been set in creation time, and it's based on umask,
1454 * la_mode and acl, don't set here again! (which will go wrong
1455 * because below function doesn't consider umask).
1456 * I'd suggest set all object attributes in creation time, see above.
1458 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1459 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1460 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1461 /* arguments are supposed to stay the same */
1462 attr->la_valid |= LA_MODE | LA_TYPE;
1466 if (S_ISDIR(attr->la_mode)) {
1467 /* Add "." and ".." for newly created dir */
1468 mdo_ref_add(env, child, handle);
1469 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1470 dot, handle, BYPASS_CAPA);
1472 rc = __mdd_index_insert_only(env, child, pfid,
1476 mdo_ref_del(env, child, handle);
1479 mdd_links_add(env, child, pfid, lname, handle, 1);
1484 /* has not lock on pobj yet */
1485 static int mdd_create_sanity_check(const struct lu_env *env,
1486 struct md_object *pobj,
1487 struct lu_attr *pattr,
1488 const struct lu_name *lname,
1489 struct lu_attr *cattr,
1490 struct md_op_spec *spec)
1492 struct mdd_thread_info *info = mdd_env_info(env);
1493 struct lu_fid *fid = &info->mti_fid;
1494 struct mdd_object *obj = md2mdd_obj(pobj);
1495 struct mdd_device *m = mdo2mdd(pobj);
1500 if (mdd_is_dead_obj(obj))
1504 * In some cases this lookup is not needed - we know before if name
1505 * exists or not because MDT performs lookup for it.
1506 * name length check is done in lookup.
1508 if (spec->sp_cr_lookup) {
1510 * Check if the name already exist, though it will be checked in
1511 * _index_insert also, for avoiding rolling back if exists
1514 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1515 MAY_WRITE | MAY_EXEC);
1517 RETURN(rc ? : -EEXIST);
1520 * Check WRITE permission for the parent.
1521 * EXEC permission have been checked
1522 * when lookup before create already.
1524 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1531 if (pattr->la_mode & S_ISGID) {
1532 cattr->la_gid = pattr->la_gid;
1533 if (S_ISDIR(cattr->la_mode)) {
1534 cattr->la_mode |= S_ISGID;
1535 cattr->la_valid |= LA_MODE;
1539 switch (cattr->la_mode & S_IFMT) {
1541 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1543 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1544 RETURN(-ENAMETOOLONG);
1563 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1564 struct mdd_object *p, struct mdd_object *c,
1565 const struct lu_name *name,
1566 struct lu_attr *attr,
1568 struct thandle *handle,
1569 const struct md_op_spec *spec)
1573 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec);
1577 #ifdef CONFIG_FS_POSIX_ACL
1578 if (got_def_acl > 0) {
1579 struct lu_buf *acl_buf;
1581 acl_buf = mdd_buf_get(env, NULL, got_def_acl);
1582 /* if dir, then can inherit default ACl */
1583 if (S_ISDIR(attr->la_mode)) {
1584 rc = mdo_declare_xattr_set(env, c, acl_buf,
1585 XATTR_NAME_ACL_DEFAULT,
1591 rc = mdo_declare_attr_set(env, c, attr, handle);
1595 rc = mdo_declare_xattr_set(env, c, acl_buf,
1596 XATTR_NAME_ACL_ACCESS, 0, handle);
1602 if (S_ISDIR(attr->la_mode)) {
1603 rc = mdo_declare_ref_add(env, p, handle);
1608 rc = mdd_declare_object_initialize(env, c, attr, handle);
1612 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1613 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
1615 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1616 name->ln_name, handle);
1620 /* replay case, create LOV EA from client data */
1621 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1622 const struct lu_buf *buf;
1624 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1625 spec->u.sp_ea.eadatalen);
1626 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1632 if (S_ISLNK(attr->la_mode)) {
1633 rc = dt_declare_record_write(env, mdd_object_child(c),
1634 strlen(spec->u.sp_symname), 0,
1640 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
1641 rc = mdo_declare_attr_set(env, p, attr, handle);
1646 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1656 * Create object and insert it into namespace.
1658 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
1659 const struct lu_name *lname, struct md_object *child,
1660 struct md_op_spec *spec, struct md_attr* ma)
1662 struct mdd_thread_info *info = mdd_env_info(env);
1663 struct lu_attr *la = &info->mti_la_for_fix;
1664 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1665 struct mdd_object *son = md2mdd_obj(child);
1666 struct mdd_device *mdd = mdo2mdd(pobj);
1667 struct lu_attr *attr = &ma->ma_attr;
1668 struct thandle *handle;
1669 struct lu_attr *pattr = &info->mti_pattr;
1670 struct dynlock_handle *dlh;
1671 const char *name = lname->ln_name;
1672 int rc, created = 0, initialized = 0, inserted = 0;
1673 int got_def_acl = 0;
1677 * Two operations have to be performed:
1679 * - an allocation of a new object (->do_create()), and
1681 * - an insertion into a parent index (->dio_insert()).
1683 * Due to locking, operation order is not important, when both are
1684 * successful, *but* error handling cases are quite different:
1686 * - if insertion is done first, and following object creation fails,
1687 * insertion has to be rolled back, but this operation might fail
1688 * also leaving us with dangling index entry.
1690 * - if creation is done first, is has to be undone if insertion
1691 * fails, leaving us with leaked space, which is neither good, nor
1694 * It seems that creation-first is simplest solution, but it is
1695 * sub-optimal in the frequent
1700 * case, because second mkdir is bound to create object, only to
1701 * destroy it immediately.
1703 * To avoid this follow local file systems that do double lookup:
1705 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1707 * 1. create (mdd_object_create_internal())
1709 * 2. insert (__mdd_index_insert(), lookup again)
1712 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1716 /* Sanity checks before big job. */
1717 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
1721 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1722 GOTO(out_free, rc = -EINPROGRESS);
1724 if (!S_ISLNK(attr->la_mode)) {
1725 struct lu_buf *acl_buf;
1727 acl_buf = mdd_buf_get(env, info->mti_xattr_buf,
1728 sizeof(info->mti_xattr_buf));
1729 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1730 rc = mdo_xattr_get(env, mdd_pobj, acl_buf,
1731 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
1732 mdd_read_unlock(env, mdd_pobj);
1735 else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA)
1739 mdd_object_make_hint(env, mdd_pobj, son, attr);
1741 handle = mdd_trans_create(env, mdd);
1743 GOTO(out_free, rc = PTR_ERR(handle));
1745 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
1746 got_def_acl, handle, spec);
1750 rc = mdd_trans_start(env, mdd, handle);
1754 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1756 GOTO(out_trans, rc = -ENOMEM);
1758 mdd_write_lock(env, son, MOR_TGT_CHILD);
1759 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec);
1761 mdd_write_unlock(env, son);
1767 #ifdef CONFIG_FS_POSIX_ACL
1769 struct lu_buf *acl_buf;
1771 acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl);
1772 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1774 mdd_write_unlock(env, son);
1780 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1781 son, attr, handle, spec);
1784 * in case of replay we just set LOVEA provided by the client
1785 * XXX: I think it would be interesting to try "old" way where
1786 * MDT calls this xattr_set(LOV) in a different transaction.
1787 * probably this way we code can be made better.
1789 if (rc == 0 && (spec->no_create ||
1790 (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) {
1791 const struct lu_buf *buf;
1793 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1794 spec->u.sp_ea.eadatalen);
1795 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
1799 if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1800 rc = __mdd_orphan_add(env, son, handle);
1802 mdd_write_unlock(env, son);
1806 * Object has no links, so it will be destroyed when last
1807 * reference is released. (XXX not now.)
1813 if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE))
1814 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1815 name, S_ISDIR(attr->la_mode), handle,
1816 mdd_object_capa(env, mdd_pobj));
1823 if (S_ISLNK(attr->la_mode)) {
1824 struct lu_ucred *uc = lu_ucred_assert(env);
1825 struct dt_object *dt = mdd_object_child(son);
1826 const char *target_name = spec->u.sp_symname;
1827 int sym_len = strlen(target_name);
1828 const struct lu_buf *buf;
1831 buf = mdd_buf_get_const(env, target_name, sym_len);
1832 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1833 mdd_object_capa(env, son),
1835 CFS_CAP_SYS_RESOURCE_MASK);
1840 GOTO(cleanup, rc = -EFAULT);
1843 /* volatile file creation does not update parent directory times */
1844 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1845 GOTO(cleanup, rc = 0);
1847 /* update parent directory mtime/ctime */
1849 la->la_valid = LA_CTIME | LA_MTIME;
1850 rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0);
1856 if (rc != 0 && created != 0) {
1859 if (inserted != 0) {
1860 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
1861 rc2 = __mdd_orphan_del(env, son, handle);
1863 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1864 S_ISDIR(attr->la_mode),
1865 handle, BYPASS_CAPA);
1870 mdd_write_lock(env, son, MOR_TGT_CHILD);
1871 if (initialized != 0 && S_ISDIR(attr->la_mode)) {
1872 /* Drop the reference, no need to delete "."/"..",
1873 * because the object to be destroied directly. */
1874 rc2 = mdo_ref_del(env, son, handle);
1876 mdd_write_unlock(env, son);
1881 rc2 = mdo_ref_del(env, son, handle);
1883 mdd_write_unlock(env, son);
1887 mdo_destroy(env, son, handle);
1888 mdd_write_unlock(env, son);
1891 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1893 if (rc == 0 && fid_is_client_mdt_visible(mdo2fid(son)))
1894 rc = mdd_changelog_ns_store(env, mdd,
1895 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1896 S_ISREG(attr->la_mode) ? CL_CREATE :
1897 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1898 0, son, mdd_pobj, lname, handle);
1900 mdd_trans_stop(env, mdd, rc, handle);
1902 /* The child object shouldn't be cached anymore */
1904 set_bit(LU_OBJECT_HEARD_BANSHEE,
1905 &child->mo_lu.lo_header->loh_flags);
1910 * Get locks on parents in proper order
1911 * RETURN: < 0 - error, rename_order if successful
1919 static int mdd_rename_order(const struct lu_env *env,
1920 struct mdd_device *mdd,
1921 struct mdd_object *src_pobj,
1922 struct mdd_object *tgt_pobj)
1924 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1928 if (src_pobj == tgt_pobj)
1929 RETURN(MDD_RN_SAME);
1931 /* compared the parent child relationship of src_p&tgt_p */
1932 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1934 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1937 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1950 /* has not mdd_write{read}_lock on any obj yet. */
1951 static int mdd_rename_sanity_check(const struct lu_env *env,
1952 struct mdd_object *src_pobj,
1953 struct mdd_object *tgt_pobj,
1954 struct mdd_object *sobj,
1955 struct mdd_object *tobj,
1956 struct lu_attr *so_attr,
1957 struct lu_attr *tg_attr)
1962 /* XXX: when get here, sobj must NOT be NULL,
1963 * the other case has been processed in cml_rename
1964 * before mdd_rename and enable MDS_PERM_BYPASS. */
1967 rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0);
1971 /* XXX: when get here, "tobj == NULL" means tobj must
1972 * NOT exist (neither on remote MDS, such case has been
1973 * processed in cml_rename before mdd_rename and enable
1975 * So check may_create, but not check may_unlink. */
1977 rc = mdd_may_create(env, tgt_pobj, NULL,
1978 (src_pobj != tgt_pobj), 0);
1980 rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr,
1981 (src_pobj != tgt_pobj), 1);
1983 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1984 S_ISDIR(so_attr->la_mode))
1985 rc = __mdd_may_link(env, tgt_pobj);
1990 static int mdd_declare_rename(const struct lu_env *env,
1991 struct mdd_device *mdd,
1992 struct mdd_object *mdd_spobj,
1993 struct mdd_object *mdd_tpobj,
1994 struct mdd_object *mdd_sobj,
1995 struct mdd_object *mdd_tobj,
1996 const struct lu_name *tname,
1997 const struct lu_name *sname,
1999 struct thandle *handle)
2007 /* name from source dir */
2008 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2012 /* .. from source child */
2013 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2014 /* source child can be directory,
2015 * counted by source dir's nlink */
2016 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2020 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2024 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2029 /* new target child can be directory,
2030 * counted by target dir's nlink */
2031 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2037 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2041 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2044 mdd_declare_links_add(env, mdd_sobj, handle);
2048 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2053 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2054 tname->ln_name, handle);
2058 /* name from target dir (old name), we declare it unconditionally
2059 * as mdd_rename() calls delete unconditionally as well. so just
2060 * to balance declarations vs calls to change ... */
2061 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2065 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2066 /* delete target child in target parent directory */
2067 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2071 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2072 /* target child can be directory,
2073 * delete "." reference in target child directory */
2074 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2078 /* delete ".." reference in target parent directory */
2079 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2084 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2088 mdd_declare_links_add(env, mdd_tobj, handle);
2092 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2097 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2104 /* src object can be remote that is why we use only fid and type of object */
2105 static int mdd_rename(const struct lu_env *env,
2106 struct md_object *src_pobj, struct md_object *tgt_pobj,
2107 const struct lu_fid *lf, const struct lu_name *lsname,
2108 struct md_object *tobj, const struct lu_name *ltname,
2111 const char *sname = lsname->ln_name;
2112 const char *tname = ltname->ln_name;
2113 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2114 struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr;
2115 struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr;
2116 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2117 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2118 struct mdd_device *mdd = mdo2mdd(src_pobj);
2119 struct mdd_object *mdd_sobj = NULL; /* source object */
2120 struct mdd_object *mdd_tobj = NULL;
2121 struct dynlock_handle *sdlh, *tdlh;
2122 struct thandle *handle;
2123 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2124 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2127 bool tobj_locked = 0;
2128 unsigned cl_flags = 0;
2133 mdd_tobj = md2mdd_obj(tobj);
2135 mdd_sobj = mdd_object_find(env, mdd, lf);
2137 handle = mdd_trans_create(env, mdd);
2139 GOTO(out_pending, rc = PTR_ERR(handle));
2141 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2142 mdd_tobj, lsname, ltname, ma, handle);
2146 rc = mdd_trans_start(env, mdd, handle);
2150 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2151 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2153 GOTO(cleanup_unlocked, rc);
2155 /* Get locks in determined order */
2156 if (rc == MDD_RN_SAME) {
2157 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2158 sname, MOR_SRC_PARENT);
2159 /* check hashes to determine do we need one lock or two */
2160 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2161 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2165 } else if (rc == MDD_RN_SRCTGT) {
2166 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2167 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2169 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2170 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2172 if (sdlh == NULL || tdlh == NULL)
2173 GOTO(cleanup, rc = -ENOMEM);
2175 rc = mdd_la_get(env, mdd_sobj, so_attr,
2176 mdd_object_capa(env, mdd_sobj));
2181 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2182 mdd_object_capa(env, mdd_tobj));
2187 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj,
2188 mdd_tobj, so_attr, tg_attr);
2192 is_dir = S_ISDIR(so_attr->la_mode);
2194 /* Remove source name from source directory */
2195 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2196 mdd_object_capa(env, mdd_spobj));
2200 /* "mv dir1 dir2" needs "dir1/.." link update */
2201 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2202 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2203 mdd_object_capa(env, mdd_sobj));
2205 GOTO(fixup_spobj2, rc);
2207 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2208 handle, mdd_object_capa(env, mdd_sobj));
2210 GOTO(fixup_spobj, rc);
2213 /* Remove target name from target directory
2214 * Here tobj can be remote one, so we do index_delete unconditionally
2215 * and -ENOENT is allowed.
2217 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2218 mdd_object_capa(env, mdd_tpobj));
2221 /* tname might been renamed to something else */
2222 GOTO(fixup_spobj, rc);
2225 GOTO(fixup_spobj, rc);
2228 /* Insert new fid with target name into target dir */
2229 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2230 mdd_object_capa(env, mdd_tpobj));
2232 GOTO(fixup_tpobj, rc);
2234 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2235 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2237 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2239 la->la_valid = LA_CTIME;
2240 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
2242 GOTO(fixup_tpobj, rc);
2245 /* Remove old target object
2246 * For tobj is remote case cmm layer has processed
2247 * and set tobj to NULL then. So when tobj is NOT NULL,
2248 * it must be local one.
2250 if (tobj && mdd_object_exists(mdd_tobj)) {
2251 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2253 if (mdd_is_dead_obj(mdd_tobj)) {
2254 /* shld not be dead, something is wrong */
2255 CERROR("tobj is dead, something is wrong\n");
2259 mdo_ref_del(env, mdd_tobj, handle);
2261 /* Remove dot reference. */
2262 if (S_ISDIR(tg_attr->la_mode))
2263 mdo_ref_del(env, mdd_tobj, handle);
2266 /* fetch updated nlink */
2267 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2268 mdd_object_capa(env, mdd_tobj));
2270 CERROR("%s: Failed to get nlink for tobj "
2272 mdd2obd_dev(mdd)->obd_name,
2273 PFID(tpobj_fid), rc);
2274 GOTO(fixup_tpobj, rc);
2277 la->la_valid = LA_CTIME;
2278 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2280 CERROR("%s: Failed to set ctime for tobj "
2282 mdd2obd_dev(mdd)->obd_name,
2283 PFID(tpobj_fid), rc);
2284 GOTO(fixup_tpobj, rc);
2287 /* XXX: this transfer to ma will be removed with LOD/OSP */
2288 ma->ma_attr = *tg_attr;
2289 ma->ma_valid |= MA_INODE;
2290 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2292 CERROR("%s: Failed to unlink tobj "
2294 mdd2obd_dev(mdd)->obd_name,
2295 PFID(tpobj_fid), rc);
2296 GOTO(fixup_tpobj, rc);
2299 /* fetch updated nlink */
2300 rc = mdd_la_get(env, mdd_tobj, tg_attr,
2301 mdd_object_capa(env, mdd_tobj));
2303 CERROR("%s: Failed to get nlink for tobj "
2305 mdd2obd_dev(mdd)->obd_name,
2306 PFID(tpobj_fid), rc);
2307 GOTO(fixup_tpobj, rc);
2309 /* XXX: this transfer to ma will be removed with LOD/OSP */
2310 ma->ma_attr = *tg_attr;
2311 ma->ma_valid |= MA_INODE;
2313 if (so_attr->la_nlink == 0)
2314 cl_flags |= CLF_RENAME_LAST;
2317 la->la_valid = LA_CTIME | LA_MTIME;
2318 rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0);
2320 GOTO(fixup_tpobj, rc);
2322 if (mdd_spobj != mdd_tpobj) {
2323 la->la_valid = LA_CTIME | LA_MTIME;
2324 rc = mdd_attr_check_set_internal(env, mdd_tpobj, la,
2328 if (rc == 0 && mdd_sobj) {
2329 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2330 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2331 mdo2fid(mdd_tpobj), ltname, handle, 0, 0);
2333 /* Old files might not have EA entry */
2334 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2336 mdd_write_unlock(env, mdd_sobj);
2337 /* We don't fail the transaction if the link ea can't be
2338 updated -- fid2path will use alternate lookup method. */
2346 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2349 CWARN("tp obj fix error %d\n",rc2);
2351 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2352 !mdd_is_dead_obj(mdd_tobj)) {
2354 mdo_ref_add(env, mdd_tobj, handle);
2356 mdo_ref_add(env, mdd_tobj, handle);
2359 rc2 = __mdd_index_insert(env, mdd_tpobj,
2360 mdo2fid(mdd_tobj), tname,
2365 CWARN("tp obj fix error %d\n",rc2);
2370 if (rc && is_dir && mdd_sobj) {
2371 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2375 CWARN("sp obj dotdot delete error %d\n",rc2);
2378 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2379 dotdot, handle, BYPASS_CAPA);
2381 CWARN("sp obj dotdot insert error %d\n",rc2);
2386 rc2 = __mdd_index_insert(env, mdd_spobj,
2387 lf, sname, is_dir, handle, BYPASS_CAPA);
2389 CWARN("sp obj fix error %d\n",rc2);
2393 mdd_write_unlock(env, mdd_tobj);
2394 if (likely(tdlh) && sdlh != tdlh)
2395 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2397 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2400 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2401 mdd_tobj, tpobj_fid, lf,
2402 spobj_fid, ltname, lsname,
2406 mdd_trans_stop(env, mdd, rc, handle);
2408 mdd_object_put(env, mdd_sobj);
2414 * The data that link search is done on.
2416 struct mdd_link_data {
2418 * Buffer to keep link EA body.
2420 struct lu_buf *ml_buf;
2422 * The matched header, entry and its lenght in the EA
2424 struct link_ea_header *ml_leh;
2425 struct link_ea_entry *ml_lee;
2429 static int mdd_links_new(const struct lu_env *env,
2430 struct mdd_link_data *ldata)
2432 ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2433 if (ldata->ml_buf->lb_buf == NULL)
2435 ldata->ml_leh = ldata->ml_buf->lb_buf;
2436 ldata->ml_leh->leh_magic = LINK_EA_MAGIC;
2437 ldata->ml_leh->leh_len = sizeof(struct link_ea_header);
2438 ldata->ml_leh->leh_reccount = 0;
2442 /** Read the link EA into a temp buffer.
2443 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
2444 * A pointer to the buffer is stored in \a ldata::ml_buf.
2446 * \retval 0 or error
2448 int mdd_links_read(const struct lu_env *env,
2449 struct mdd_object *mdd_obj,
2450 struct mdd_link_data *ldata)
2452 struct lustre_capa *capa;
2453 struct link_ea_header *leh;
2456 /* First try a small buf */
2457 LASSERT(env != NULL);
2458 ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2459 if (ldata->ml_buf->lb_buf == NULL)
2462 if (!mdd_object_exists(mdd_obj))
2465 capa = mdd_object_capa(env, mdd_obj);
2466 rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
2467 XATTR_NAME_LINK, capa);
2468 if (rc == -ERANGE) {
2469 /* Buf was too small, figure out what we need. */
2470 mdd_buf_put(ldata->ml_buf);
2471 rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
2472 XATTR_NAME_LINK, capa);
2475 ldata->ml_buf = mdd_buf_alloc(env, rc);
2476 if (ldata->ml_buf->lb_buf == NULL)
2478 rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf,
2479 XATTR_NAME_LINK, capa);
2484 leh = ldata->ml_buf->lb_buf;
2485 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2486 leh->leh_magic = LINK_EA_MAGIC;
2487 leh->leh_reccount = __swab32(leh->leh_reccount);
2488 leh->leh_len = __swab64(leh->leh_len);
2489 /* entries are swabbed by mdd_lee_unpack */
2491 if (leh->leh_magic != LINK_EA_MAGIC)
2493 if (leh->leh_reccount == 0)
2496 ldata->ml_leh = leh;
2500 /** Read the link EA into a temp buffer.
2501 * Uses the name_buf since it is generally large.
2502 * \retval IS_ERR err
2503 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2505 struct lu_buf *mdd_links_get(const struct lu_env *env,
2506 struct mdd_object *mdd_obj)
2508 struct mdd_link_data ldata = { 0 };
2511 rc = mdd_links_read(env, mdd_obj, &ldata);
2512 return rc ? ERR_PTR(rc) : ldata.ml_buf;
2515 static int mdd_links_write(const struct lu_env *env,
2516 struct mdd_object *mdd_obj,
2517 struct mdd_link_data *ldata,
2518 struct thandle *handle)
2520 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ml_buf->lb_buf,
2521 ldata->ml_leh->leh_len);
2522 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
2523 mdd_object_capa(env, mdd_obj));
2526 /** Pack a link_ea_entry.
2527 * All elements are stored as chars to avoid alignment issues.
2528 * Numbers are always big-endian
2529 * \retval record length
2531 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2532 const struct lu_fid *pfid)
2534 struct lu_fid tmpfid;
2537 fid_cpu_to_be(&tmpfid, pfid);
2538 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2539 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2540 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2542 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2543 lee->lee_reclen[1] = reclen & 0xff;
2547 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2548 struct lu_name *lname, struct lu_fid *pfid)
2550 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2551 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2552 fid_be_to_cpu(pfid, pfid);
2553 lname->ln_name = lee->lee_name;
2554 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2557 static int mdd_declare_links_add(const struct lu_env *env,
2558 struct mdd_object *mdd_obj,
2559 struct thandle *handle)
2563 /* XXX: max size? */
2564 rc = mdo_declare_xattr_set(env, mdd_obj,
2565 mdd_buf_get_const(env, NULL, 4096),
2566 XATTR_NAME_LINK, 0, handle);
2571 /* For pathologic linkers, we don't want to spend lots of time scanning the
2572 * link ea. Limit ourseleves to something reasonable; links not in the EA
2573 * can be looked up via (slower) parent lookup.
2575 #define LINKEA_MAX_COUNT 128
2577 /** Add a record to the end of link ea buf */
2578 static int mdd_links_add_buf(const struct lu_env *env,
2579 struct mdd_link_data *ldata,
2580 const struct lu_name *lname,
2581 const struct lu_fid *pfid)
2583 LASSERT(ldata->ml_leh != NULL);
2585 if (lname == NULL || pfid == NULL)
2588 /* Make sure our buf is big enough for the new one */
2589 if (ldata->ml_leh->leh_reccount > LINKEA_MAX_COUNT)
2592 ldata->ml_reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2593 if (ldata->ml_leh->leh_len + ldata->ml_reclen >
2594 ldata->ml_buf->lb_len) {
2595 if (mdd_buf_grow(env, ldata->ml_leh->leh_len +
2596 ldata->ml_reclen) < 0)
2600 ldata->ml_leh = ldata->ml_buf->lb_buf;
2601 ldata->ml_lee = ldata->ml_buf->lb_buf + ldata->ml_leh->leh_len;
2602 ldata->ml_reclen = mdd_lee_pack(ldata->ml_lee, lname, pfid);
2603 ldata->ml_leh->leh_len += ldata->ml_reclen;
2604 ldata->ml_leh->leh_reccount++;
2605 CDEBUG(D_INODE, "New link_ea name '%.*s' is added\n",
2606 lname->ln_namelen, lname->ln_name);
2610 /** Del the current record from the link ea buf */
2611 static void mdd_links_del_buf(const struct lu_env *env,
2612 struct mdd_link_data *ldata,
2613 const struct lu_name *lname)
2615 LASSERT(ldata->ml_leh != NULL);
2617 ldata->ml_leh->leh_reccount--;
2618 ldata->ml_leh->leh_len -= ldata->ml_reclen;
2619 memmove(ldata->ml_lee, (char *)ldata->ml_lee + ldata->ml_reclen,
2620 (char *)ldata->ml_leh + ldata->ml_leh->leh_len -
2621 (char *)ldata->ml_lee);
2622 CDEBUG(D_INODE, "Old link_ea name '%.*s' is removed\n",
2623 lname->ln_namelen, lname->ln_name);
2628 * Check if such a link exists in linkEA.
2630 * \param mdd_obj object being handled
2631 * \param pfid parent fid the link to be found for
2632 * \param lname name in the parent's directory entry pointing to this object
2633 * \param ldata link data the search to be done on
2636 * \retval -ENOENT link does not exist
2637 * \retval -ve on error
2639 static int mdd_links_find(const struct lu_env *env,
2640 struct mdd_object *mdd_obj,
2641 struct mdd_link_data *ldata,
2642 const struct lu_name *lname,
2643 const struct lu_fid *pfid)
2645 struct lu_name *tmpname = &mdd_env_info(env)->mti_name2;
2646 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2649 LASSERT(ldata->ml_leh != NULL);
2652 ldata->ml_lee = (struct link_ea_entry *)(ldata->ml_leh + 1);
2654 for (count = 0; count < ldata->ml_leh->leh_reccount; count++) {
2655 mdd_lee_unpack(ldata->ml_lee, &ldata->ml_reclen,
2657 if (tmpname->ln_namelen == lname->ln_namelen &&
2658 lu_fid_eq(tmpfid, pfid) &&
2659 (strncmp(tmpname->ln_name, lname->ln_name,
2660 tmpname->ln_namelen) == 0))
2662 ldata->ml_lee = (struct link_ea_entry *)((char *)ldata->ml_lee +
2666 if (count == ldata->ml_leh->leh_reccount) {
2667 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2668 lname->ln_namelen, lname->ln_name);
2674 static int __mdd_links_add(const struct lu_env *env,
2675 struct mdd_object *mdd_obj,
2676 struct mdd_link_data *ldata,
2677 const struct lu_name *lname,
2678 const struct lu_fid *pfid,
2679 int first, int check)
2683 if (ldata->ml_leh == NULL) {
2684 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
2688 rc = mdd_links_new(env, ldata);
2695 rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid);
2696 if (rc && rc != -ENOENT)
2702 return mdd_links_add_buf(env, ldata, lname, pfid);
2705 static int __mdd_links_del(const struct lu_env *env,
2706 struct mdd_object *mdd_obj,
2707 struct mdd_link_data *ldata,
2708 const struct lu_name *lname,
2709 const struct lu_fid *pfid)
2713 if (ldata->ml_leh == NULL) {
2714 rc = mdd_links_read(env, mdd_obj, ldata);
2719 rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid);
2723 mdd_links_del_buf(env, ldata, lname);
2727 static int mdd_links_rename(const struct lu_env *env,
2728 struct mdd_object *mdd_obj,
2729 const struct lu_fid *oldpfid,
2730 const struct lu_name *oldlname,
2731 const struct lu_fid *newpfid,
2732 const struct lu_name *newlname,
2733 struct thandle *handle,
2734 int first, int check)
2736 struct mdd_link_data ldata = { 0 };
2742 LASSERT(oldpfid != NULL || newpfid != NULL);
2744 if (mdd_obj->mod_flags & DEAD_OBJ)
2745 /* No more links, don't bother */
2748 if (oldpfid != NULL) {
2749 rc = __mdd_links_del(env, mdd_obj, &ldata,
2753 (rc != -ENODATA && rc != -ENOENT))
2755 /* No changes done. */
2762 /* If renaming, add the new record */
2763 if (newpfid != NULL) {
2764 /* even if the add fails, we still delete the out-of-date
2766 rc2 = __mdd_links_add(env, mdd_obj, &ldata,
2767 newlname, newpfid, first, check);
2775 rc = mdd_links_write(env, mdd_obj, &ldata, handle);
2782 if (rc == -EOVERFLOW || rc == - ENOENT)
2784 if (oldpfid == NULL)
2785 CDEBUG(error ? D_ERROR : D_OTHER,
2786 "link_ea add '%.*s' failed %d "DFID"\n",
2787 newlname->ln_namelen, newlname->ln_name,
2788 rc, PFID(mdd_object_fid(mdd_obj)));
2789 else if (newpfid == NULL)
2790 CDEBUG(error ? D_ERROR : D_OTHER,
2791 "link_ea del '%.*s' failed %d "DFID"\n",
2792 oldlname->ln_namelen, oldlname->ln_name,
2793 rc, PFID(mdd_object_fid(mdd_obj)));
2795 CDEBUG(error ? D_ERROR : D_OTHER,
2796 "link_ea rename '%.*s'->'%.*s' failed %d "
2798 oldlname->ln_namelen, oldlname->ln_name,
2799 newlname->ln_namelen, newlname->ln_name,
2800 rc, PFID(mdd_object_fid(mdd_obj)));
2803 if (ldata.ml_buf && ldata.ml_buf->lb_len > OBD_ALLOC_BIG)
2804 /* if we vmalloced a large buffer drop it */
2805 mdd_buf_put(ldata.ml_buf);
2810 static inline int mdd_links_add(const struct lu_env *env,
2811 struct mdd_object *mdd_obj,
2812 const struct lu_fid *pfid,
2813 const struct lu_name *lname,
2814 struct thandle *handle, int first)
2816 return mdd_links_rename(env, mdd_obj, NULL, NULL,
2817 pfid, lname, handle, first, 0);
2820 static inline int mdd_links_del(const struct lu_env *env,
2821 struct mdd_object *mdd_obj,
2822 const struct lu_fid *pfid,
2823 const struct lu_name *lname,
2824 struct thandle *handle)
2826 return mdd_links_rename(env, mdd_obj, pfid, lname,
2827 NULL, NULL, handle, 0, 0);
2830 const struct md_dir_operations mdd_dir_ops = {
2831 .mdo_is_subdir = mdd_is_subdir,
2832 .mdo_lookup = mdd_lookup,
2833 .mdo_create = mdd_create,
2834 .mdo_rename = mdd_rename,
2835 .mdo_link = mdd_link,
2836 .mdo_unlink = mdd_unlink,
2837 .mdo_create_data = mdd_create_data,