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, 2013, 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@intel.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 = {
61 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
63 if (!lu_name_is_valid(ln))
65 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
71 /* Get FID from name and parent */
73 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
74 const struct lu_attr *pattr, const struct lu_name *lname,
75 struct lu_fid* fid, int mask)
77 const char *name = lname->ln_name;
78 const struct dt_key *key = (const struct dt_key *)name;
79 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
80 struct mdd_device *m = mdo2mdd(pobj);
81 struct dt_object *dir = mdd_object_child(mdd_obj);
85 if (unlikely(mdd_is_dead_obj(mdd_obj)))
88 if (mdd_object_remote(mdd_obj)) {
89 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
90 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
91 } else if (!mdd_object_exists(mdd_obj)) {
95 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
100 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
101 dt_try_as_dir(env, dir))) {
103 rc = dir->do_index_ops->dio_lookup(env, dir,
104 (struct dt_rec *)fid, key,
105 mdd_object_capa(env, mdd_obj));
116 int mdd_lookup(const struct lu_env *env,
117 struct md_object *pobj, const struct lu_name *lname,
118 struct lu_fid *fid, struct md_op_spec *spec)
120 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
124 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr, BYPASS_CAPA);
128 rc = __mdd_lookup(env, pobj, pattr, lname, fid, MAY_EXEC);
132 static inline int mdd_parent_fid(const struct lu_env *env,
133 struct mdd_object *obj,
134 const struct lu_attr *attr,
137 return __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
141 * For root fid use special function, which does not compare version component
142 * of fid. Version component is different for root fids on all MDTs.
144 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
146 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
147 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
151 * return 1: if lf is the fid of the ancestor of p1;
154 * return -EREMOTE: if remote object is found, in this
155 * case fid of remote object is saved to @pf;
157 * otherwise: values < 0, errors.
159 static int mdd_is_parent(const struct lu_env *env,
160 struct mdd_device *mdd,
161 struct mdd_object *p1,
162 const struct lu_attr *attr,
163 const struct lu_fid *lf,
166 struct mdd_object *parent = NULL;
171 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
172 pfid = &mdd_env_info(env)->mti_fid;
174 /* Check for root first. */
175 if (mdd_is_root(mdd, mdo2fid(p1)))
179 /* this is done recursively, bypass capa for each obj */
180 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
181 rc = mdd_parent_fid(env, p1, attr, pfid);
184 if (mdd_is_root(mdd, pfid))
186 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
188 if (lu_fid_eq(pfid, lf))
191 mdd_object_put(env, parent);
193 parent = mdd_object_find(env, mdd, pfid);
194 if (IS_ERR(parent)) {
195 GOTO(out, rc = PTR_ERR(parent));
196 } else if (mdd_object_remote(parent)) {
197 /*FIXME: Because of the restriction of rename in Phase I.
198 * If the parent is remote, we just assumed lf is not the
199 * parent of P1 for now */
206 if (parent && !IS_ERR(parent))
207 mdd_object_put(env, parent);
212 * No permission check is needed.
214 * returns 1: if fid is ancestor of @mo;
215 * returns 0: if fid is not a ancestor of @mo;
217 * returns EREMOTE if remote object is found, fid of remote object is saved to
220 * returns < 0: if error
222 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
223 const struct lu_fid *fid, struct lu_fid *sfid)
225 struct mdd_device *mdd = mdo2mdd(mo);
226 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
230 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
233 rc = mdd_la_get(env, md2mdd_obj(mo), attr, BYPASS_CAPA);
237 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
241 } else if (rc == 1) {
242 /* found @fid is parent */
250 * Check that @dir contains no entries except (possibly) dot and dotdot.
255 * -ENOTDIR not a directory object
256 * -ENOTEMPTY not empty
260 static int mdd_dir_is_empty(const struct lu_env *env,
261 struct mdd_object *dir)
264 struct dt_object *obj;
265 const struct dt_it_ops *iops;
269 obj = mdd_object_child(dir);
270 if (!dt_try_as_dir(env, obj))
273 iops = &obj->do_index_ops->dio_it;
274 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
276 result = iops->get(env, it, (const void *)"");
279 for (result = 0, i = 0; result == 0 && i < 3; ++i)
280 result = iops->next(env, it);
283 else if (result == +1)
285 } else if (result == 0)
287 * Huh? Index contains no zero key?
294 result = PTR_ERR(it);
298 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
299 const struct lu_attr *la)
301 struct mdd_device *m = mdd_obj2mdd_dev(obj);
306 if (!S_ISDIR(la->la_mode))
310 * Subdir count limitation can be broken through.
312 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
319 * Check whether it may create the cobj under the pobj.
322 int mdd_may_create(const struct lu_env *env,
323 struct mdd_object *pobj, const struct lu_attr *pattr,
324 struct mdd_object *cobj, int check_perm, int check_nlink)
329 if (cobj && mdd_object_exists(cobj))
332 if (mdd_is_dead_obj(pobj))
336 rc = mdd_permission_internal_locked(env, pobj, pattr,
337 MAY_WRITE | MAY_EXEC,
339 if (!rc && check_nlink)
340 rc = __mdd_may_link(env, pobj, pattr);
346 * Check whether can unlink from the pobj in the case of "cobj == NULL".
348 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
349 const struct lu_attr *pattr, const struct lu_attr *attr)
354 if (mdd_is_dead_obj(pobj))
357 if ((attr->la_valid & LA_FLAGS) &&
358 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
361 rc = mdd_permission_internal_locked(env, pobj, pattr,
362 MAY_WRITE | MAY_EXEC,
367 if (mdd_is_append(pobj))
374 * pobj == NULL is remote ops case, under such case, pobj's
375 * VTX feature has been checked already, no need check again.
377 static inline int mdd_is_sticky(const struct lu_env *env,
378 struct mdd_object *pobj,
379 const struct lu_attr *pattr,
380 struct mdd_object *cobj,
381 const struct lu_attr *cattr)
383 struct lu_ucred *uc = lu_ucred_assert(env);
386 LASSERT(pattr != NULL);
387 if (!(pattr->la_mode & S_ISVTX) ||
388 (pattr->la_uid == uc->uc_fsuid))
392 LASSERT(cattr != NULL);
393 if (cattr->la_uid == uc->uc_fsuid)
396 return !md_capable(uc, CFS_CAP_FOWNER);
399 static int mdd_may_delete_entry(const struct lu_env *env,
400 struct mdd_object *pobj,
401 const struct lu_attr *pattr,
406 LASSERT(pobj != NULL);
407 if (!mdd_object_exists(pobj))
410 if (mdd_is_dead_obj(pobj))
415 rc = mdd_permission_internal_locked(env, pobj, pattr,
416 MAY_WRITE | MAY_EXEC,
422 if (mdd_is_append(pobj))
429 * Check whether it may delete the cobj from the pobj.
432 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
433 const struct lu_attr *tpattr, struct mdd_object *tobj,
434 const struct lu_attr *tattr, const struct lu_attr *cattr,
435 int check_perm, int check_empty)
441 LASSERT(tpattr != NULL);
442 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
450 if (!mdd_object_exists(tobj))
453 if (mdd_is_dead_obj(tobj))
456 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
459 if (mdd_is_immutable(tobj) || mdd_is_append(tobj))
462 if ((tattr->la_valid & LA_FLAGS) &&
463 (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
466 /* additional check the rename case */
468 if (S_ISDIR(cattr->la_mode)) {
469 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
471 if (!S_ISDIR(tattr->la_mode))
474 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
476 } else if (S_ISDIR(tattr->la_mode))
480 if (S_ISDIR(tattr->la_mode) && check_empty)
481 rc = mdd_dir_is_empty(env, tobj);
488 * has mdd_write_lock on src already, but not on tgt yet
490 static int mdd_link_sanity_check(const struct lu_env *env,
491 struct mdd_object *tgt_obj,
492 const struct lu_attr *tattr,
493 const struct lu_name *lname,
494 struct mdd_object *src_obj,
495 const struct lu_attr *cattr)
497 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
501 if (!mdd_object_exists(src_obj))
504 if (mdd_is_dead_obj(src_obj))
507 /* Local ops, no lookup before link, check filename length here. */
508 rc = mdd_name_check(m, lname);
512 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
515 if (S_ISDIR(mdd_object_type(src_obj)))
518 LASSERT(src_obj != tgt_obj);
520 rc = mdd_may_create(env, tgt_obj, tattr, NULL, 1, 0);
525 rc = __mdd_may_link(env, src_obj, cattr);
530 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
531 const char *name, struct thandle *handle,
532 struct lustre_capa *capa)
534 struct dt_object *next = mdd_object_child(pobj);
538 if (dt_try_as_dir(env, next)) {
539 rc = next->do_index_ops->dio_delete(env, next,
540 (struct dt_key *)name,
548 static int __mdd_index_insert_only(const struct lu_env *env,
549 struct mdd_object *pobj,
550 const struct lu_fid *lf, const char *name,
551 struct thandle *handle,
552 struct lustre_capa *capa)
554 struct dt_object *next = mdd_object_child(pobj);
558 if (dt_try_as_dir(env, next)) {
559 struct lu_ucred *uc = lu_ucred_check(env);
562 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
563 rc = next->do_index_ops->dio_insert(env, next,
565 (const struct dt_key *)name,
566 handle, capa, ignore_quota);
573 /* insert named index, add reference if isdir */
574 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
575 const struct lu_fid *lf, const char *name, int is_dir,
576 struct thandle *handle, struct lustre_capa *capa)
581 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
582 if (rc == 0 && is_dir) {
583 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
584 mdo_ref_add(env, pobj, handle);
585 mdd_write_unlock(env, pobj);
590 /* delete named index, drop reference if isdir */
591 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
592 const char *name, int is_dir, struct thandle *handle,
593 struct lustre_capa *capa)
598 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
599 if (rc == 0 && is_dir) {
600 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
601 mdo_ref_del(env, pobj, handle);
602 mdd_write_unlock(env, pobj);
608 int mdd_declare_changelog_store(const struct lu_env *env,
609 struct mdd_device *mdd,
610 const struct lu_name *fname,
611 struct thandle *handle)
613 struct obd_device *obd = mdd2obd_dev(mdd);
614 struct llog_ctxt *ctxt;
615 struct llog_changelog_rec *rec;
621 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
624 reclen = llog_data_len(sizeof(*rec) +
625 (fname != NULL ? fname->ln_namelen : 0));
626 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
627 if (buf->lb_buf == NULL)
631 rec->cr_hdr.lrh_len = reclen;
632 rec->cr_hdr.lrh_type = CHANGELOG_REC;
634 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
638 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
644 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
645 struct mdd_device *mdd,
646 const struct lu_name *tname,
647 const struct lu_name *sname,
648 struct thandle *handle)
650 struct obd_device *obd = mdd2obd_dev(mdd);
651 struct llog_ctxt *ctxt;
652 struct llog_changelog_ext_rec *rec;
658 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
661 reclen = llog_data_len(sizeof(*rec) +
662 (tname != NULL ? tname->ln_namelen : 0) +
663 (sname != NULL ? 1 + sname->ln_namelen : 0));
664 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
665 if (buf->lb_buf == NULL)
669 rec->cr_hdr.lrh_len = reclen;
670 rec->cr_hdr.lrh_type = CHANGELOG_REC;
672 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
676 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
682 /** Add a changelog entry \a rec to the changelog llog
685 * \param handle - currently ignored since llogs start their own transaction;
686 * this will hopefully be fixed in llog rewrite
689 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
690 struct llog_changelog_rec *rec, struct thandle *th)
692 struct obd_device *obd = mdd2obd_dev(mdd);
693 struct llog_ctxt *ctxt;
696 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
697 rec->cr_hdr.lrh_type = CHANGELOG_REC;
698 rec->cr.cr_time = cl_time();
700 spin_lock(&mdd->mdd_cl.mc_lock);
701 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
702 * but as long as the MDD transactions are ordered correctly for e.g.
703 * rename conflicts, I don't think this should matter. */
704 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
705 spin_unlock(&mdd->mdd_cl.mc_lock);
707 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
711 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
718 /** Add a changelog_ext entry \a rec to the changelog llog
721 * \param handle - currently ignored since llogs start their own transaction;
722 * this will hopefully be fixed in llog rewrite
726 mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
727 struct llog_changelog_ext_rec *rec, struct thandle *th)
729 struct obd_device *obd = mdd2obd_dev(mdd);
730 struct llog_ctxt *ctxt;
733 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
734 /* llog_lvfs_write_rec sets the llog tail len */
735 rec->cr_hdr.lrh_type = CHANGELOG_REC;
736 rec->cr.cr_time = cl_time();
738 spin_lock(&mdd->mdd_cl.mc_lock);
739 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
740 * but as long as the MDD transactions are ordered correctly for e.g.
741 * rename conflicts, I don't think this should matter. */
742 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
743 spin_unlock(&mdd->mdd_cl.mc_lock);
745 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
749 /* nested journal transaction */
750 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th);
758 /** Store a namespace change changelog record
759 * If this fails, we must fail the whole transaction; we don't
760 * want the change to commit without the log entry.
761 * \param target - mdd_object of change
762 * \param parent - parent dir/object
763 * \param tname - target name string
764 * \param handle - transacion handle
766 int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd,
767 enum changelog_rec_type type, unsigned flags,
768 struct mdd_object *target, struct mdd_object *parent,
769 const struct lu_name *tname, struct thandle *handle)
771 struct llog_changelog_rec *rec;
778 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
780 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
783 LASSERT(target != NULL);
784 LASSERT(parent != NULL);
785 LASSERT(tname != NULL);
786 LASSERT(handle != NULL);
788 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
789 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
790 if (buf->lb_buf == NULL)
794 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
795 rec->cr.cr_type = (__u32)type;
796 rec->cr.cr_tfid = *mdo2fid(target);
797 rec->cr.cr_pfid = *mdo2fid(parent);
798 rec->cr.cr_namelen = tname->ln_namelen;
799 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
801 target->mod_cltime = cfs_time_current_64();
803 rc = mdd_changelog_store(env, mdd, rec, handle);
805 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
806 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
807 PFID(&rec->cr.cr_pfid));
815 /** Store a namespace change changelog record
816 * If this fails, we must fail the whole transaction; we don't
817 * want the change to commit without the log entry.
818 * \param target - mdd_object of change
819 * \param tpfid - target parent dir/object fid
820 * \param sfid - source object fid
821 * \param spfid - source parent fid
822 * \param tname - target name string
823 * \param sname - source name string
824 * \param handle - transacion handle
826 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
827 struct mdd_device *mdd,
828 enum changelog_rec_type type,
830 struct mdd_object *target,
831 const struct lu_fid *tpfid,
832 const struct lu_fid *sfid,
833 const struct lu_fid *spfid,
834 const struct lu_name *tname,
835 const struct lu_name *sname,
836 struct thandle *handle)
838 struct llog_changelog_ext_rec *rec;
845 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
847 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
850 LASSERT(sfid != NULL);
851 LASSERT(tpfid != NULL);
852 LASSERT(tname != NULL);
853 LASSERT(handle != NULL);
855 reclen = llog_data_len(sizeof(*rec) +
856 sname != NULL ? 1 + sname->ln_namelen : 0);
857 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
858 if (buf->lb_buf == NULL)
862 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
863 rec->cr.cr_type = (__u32)type;
864 rec->cr.cr_pfid = *tpfid;
865 rec->cr.cr_sfid = *sfid;
866 rec->cr.cr_spfid = *spfid;
867 rec->cr.cr_namelen = tname->ln_namelen;
868 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
870 rec->cr.cr_name[tname->ln_namelen] = '\0';
871 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
873 rec->cr.cr_namelen += 1 + sname->ln_namelen;
876 if (likely(target != NULL)) {
877 rec->cr.cr_tfid = *mdo2fid(target);
878 target->mod_cltime = cfs_time_current_64();
880 fid_zero(&rec->cr.cr_tfid);
883 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
885 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
886 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
893 static int __mdd_links_add(const struct lu_env *env,
894 struct mdd_object *mdd_obj,
895 struct linkea_data *ldata,
896 const struct lu_name *lname,
897 const struct lu_fid *pfid,
898 int first, int check)
902 if (ldata->ld_leh == NULL) {
903 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
907 rc = linkea_data_new(ldata,
908 &mdd_env_info(env)->mti_link_buf);
915 rc = linkea_links_find(ldata, lname, pfid);
916 if (rc && rc != -ENOENT)
922 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
923 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
927 linkea_add_buf(ldata, lname, tfid);
930 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
931 linkea_add_buf(ldata, lname, pfid);
933 return linkea_add_buf(ldata, lname, pfid);
936 static int __mdd_links_del(const struct lu_env *env,
937 struct mdd_object *mdd_obj,
938 struct linkea_data *ldata,
939 const struct lu_name *lname,
940 const struct lu_fid *pfid)
944 if (ldata->ld_leh == NULL) {
945 rc = mdd_links_read(env, mdd_obj, ldata);
950 rc = linkea_links_find(ldata, lname, pfid);
954 linkea_del_buf(ldata, lname);
958 static int mdd_linkea_prepare(const struct lu_env *env,
959 struct mdd_object *mdd_obj,
960 const struct lu_fid *oldpfid,
961 const struct lu_name *oldlname,
962 const struct lu_fid *newpfid,
963 const struct lu_name *newlname,
964 int first, int check,
965 struct linkea_data *ldata)
971 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
974 LASSERT(oldpfid != NULL || newpfid != NULL);
976 if (mdd_obj->mod_flags & DEAD_OBJ) {
977 /* Prevent linkea to be updated which is NOT necessary. */
978 ldata->ld_reclen = 0;
979 /* No more links, don't bother */
983 if (oldpfid != NULL) {
984 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
987 (rc != -ENODATA && rc != -ENOENT))
989 /* No changes done. */
994 /* If renaming, add the new record */
995 if (newpfid != NULL) {
996 /* even if the add fails, we still delete the out-of-date
998 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1002 rc = rc != 0 ? rc : rc2;
1007 int mdd_links_rename(const struct lu_env *env,
1008 struct mdd_object *mdd_obj,
1009 const struct lu_fid *oldpfid,
1010 const struct lu_name *oldlname,
1011 const struct lu_fid *newpfid,
1012 const struct lu_name *newlname,
1013 struct thandle *handle,
1014 struct linkea_data *ldata,
1015 int first, int check)
1021 if (ldata == NULL) {
1022 ldata = &mdd_env_info(env)->mti_link_data;
1023 memset(ldata, 0, sizeof(*ldata));
1024 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1025 newpfid, newlname, first, check,
1031 if (ldata->ld_reclen != 0)
1032 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1039 if (rc == -EOVERFLOW || rc == -ENOSPC)
1041 if (oldpfid == NULL)
1042 CDEBUG(error ? D_ERROR : D_OTHER,
1043 "link_ea add '%.*s' failed %d "DFID"\n",
1044 newlname->ln_namelen, newlname->ln_name,
1045 rc, PFID(mdd_object_fid(mdd_obj)));
1046 else if (newpfid == NULL)
1047 CDEBUG(error ? D_ERROR : D_OTHER,
1048 "link_ea del '%.*s' failed %d "DFID"\n",
1049 oldlname->ln_namelen, oldlname->ln_name,
1050 rc, PFID(mdd_object_fid(mdd_obj)));
1052 CDEBUG(error ? D_ERROR : D_OTHER,
1053 "link_ea rename '%.*s'->'%.*s' failed %d "
1055 oldlname->ln_namelen, oldlname->ln_name,
1056 newlname->ln_namelen, newlname->ln_name,
1057 rc, PFID(mdd_object_fid(mdd_obj)));
1060 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1061 /* if we vmalloced a large buffer drop it */
1062 lu_buf_free(ldata->ld_buf);
1067 static inline int mdd_links_add(const struct lu_env *env,
1068 struct mdd_object *mdd_obj,
1069 const struct lu_fid *pfid,
1070 const struct lu_name *lname,
1071 struct thandle *handle,
1072 struct linkea_data *data, int first)
1074 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1075 pfid, lname, handle, data, first, 0);
1078 static inline int mdd_links_del(const struct lu_env *env,
1079 struct mdd_object *mdd_obj,
1080 const struct lu_fid *pfid,
1081 const struct lu_name *lname,
1082 struct thandle *handle)
1084 return mdd_links_rename(env, mdd_obj, pfid, lname,
1085 NULL, NULL, handle, NULL, 0, 0);
1088 /** Read the link EA into a temp buffer.
1089 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1090 * A pointer to the buffer is stored in \a ldata::ld_buf.
1092 * \retval 0 or error
1094 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1095 struct linkea_data *ldata)
1099 if (!mdd_object_exists(mdd_obj))
1102 /* First try a small buf */
1103 LASSERT(env != NULL);
1104 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1106 if (ldata->ld_buf->lb_buf == NULL)
1109 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1111 if (rc == -ERANGE) {
1112 /* Buf was too small, figure out what we need. */
1113 lu_buf_free(ldata->ld_buf);
1114 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1115 XATTR_NAME_LINK, BYPASS_CAPA);
1118 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1119 if (ldata->ld_buf->lb_buf == NULL)
1121 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1122 XATTR_NAME_LINK, BYPASS_CAPA);
1125 lu_buf_free(ldata->ld_buf);
1126 ldata->ld_buf = NULL;
1130 return linkea_init(ldata);
1133 /** Read the link EA into a temp buffer.
1134 * Uses the name_buf since it is generally large.
1135 * \retval IS_ERR err
1136 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1138 struct lu_buf *mdd_links_get(const struct lu_env *env,
1139 struct mdd_object *mdd_obj)
1141 struct linkea_data ldata = { 0 };
1144 rc = mdd_links_read(env, mdd_obj, &ldata);
1145 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1148 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1149 struct linkea_data *ldata, struct thandle *handle)
1151 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1152 ldata->ld_leh->leh_len);
1153 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1154 mdd_object_capa(env, mdd_obj));
1157 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1158 struct thandle *handle, struct linkea_data *ldata)
1164 if (ldata != NULL && ldata->ld_leh != NULL) {
1165 ea_len = ldata->ld_leh->leh_len;
1166 linkea = ldata->ld_buf->lb_buf;
1168 ea_len = DEFAULT_LINKEA_SIZE;
1172 /* XXX: max size? */
1173 rc = mdo_declare_xattr_set(env, mdd_obj,
1174 mdd_buf_get_const(env, linkea, ea_len),
1175 XATTR_NAME_LINK, 0, handle);
1179 static inline int mdd_declare_links_del(const struct lu_env *env,
1180 struct mdd_object *c,
1181 struct thandle *handle)
1185 /* For directory, the linkEA will be removed together
1186 * with the object. */
1187 if (!S_ISDIR(mdd_object_type(c)))
1188 rc = mdd_declare_links_add(env, c, handle, NULL);
1193 static int mdd_declare_link(const struct lu_env *env,
1194 struct mdd_device *mdd,
1195 struct mdd_object *p,
1196 struct mdd_object *c,
1197 const struct lu_name *name,
1198 struct thandle *handle,
1200 struct linkea_data *data)
1204 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
1208 rc = mdo_declare_ref_add(env, c, handle);
1212 la->la_valid = LA_CTIME | LA_MTIME;
1213 rc = mdo_declare_attr_set(env, p, la, handle);
1217 la->la_valid = LA_CTIME;
1218 rc = mdo_declare_attr_set(env, c, la, handle);
1222 rc = mdd_declare_links_add(env, c, handle, data);
1226 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1231 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1232 struct md_object *src_obj, const struct lu_name *lname,
1235 const char *name = lname->ln_name;
1236 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1237 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1238 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1239 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1240 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1241 struct mdd_device *mdd = mdo2mdd(src_obj);
1242 struct thandle *handle;
1243 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1247 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
1251 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
1255 handle = mdd_trans_create(env, mdd);
1257 GOTO(out_pending, rc = PTR_ERR(handle));
1259 memset(ldata, 0, sizeof(*ldata));
1261 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1262 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1264 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1269 rc = mdd_trans_start(env, mdd, handle);
1273 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1274 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1277 GOTO(out_unlock, rc);
1279 rc = mdo_ref_add(env, mdd_sobj, handle);
1281 GOTO(out_unlock, rc);
1284 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1286 mdd_object_capa(env, mdd_tobj));
1288 mdo_ref_del(env, mdd_sobj, handle);
1289 GOTO(out_unlock, rc);
1292 la->la_valid = LA_CTIME | LA_MTIME;
1293 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1295 GOTO(out_unlock, rc);
1297 la->la_valid = LA_CTIME;
1298 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1300 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1301 mdo2fid(mdd_tobj), lname, 0, 0,
1304 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1305 lname, handle, ldata, 0);
1306 /* The failure of links_add should not cause the link
1307 * failure, reset rc here */
1312 mdd_write_unlock(env, mdd_sobj);
1314 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1315 mdd_tobj, lname, handle);
1317 mdd_trans_stop(env, mdd, rc, handle);
1319 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1320 /* if we vmalloced a large buffer drop it */
1321 lu_buf_free(ldata->ld_buf);
1326 static int mdd_declare_finish_unlink(const struct lu_env *env,
1327 struct mdd_object *obj,
1328 struct thandle *handle)
1332 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1336 rc = mdo_declare_destroy(env, obj, handle);
1340 return mdd_declare_links_del(env, obj, handle);
1343 /* caller should take a lock before calling */
1344 int mdd_finish_unlink(const struct lu_env *env,
1345 struct mdd_object *obj, struct md_attr *ma,
1346 const struct mdd_object *pobj,
1347 const struct lu_name *lname,
1351 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1354 LASSERT(mdd_write_locked(env, obj) != 0);
1356 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1357 obj->mod_flags |= DEAD_OBJ;
1359 /* add new orphan and the object
1360 * will be deleted during mdd_close() */
1361 if (obj->mod_count) {
1362 rc = __mdd_orphan_add(env, obj, th);
1364 CDEBUG(D_HA, "Object "DFID" is inserted into "
1365 "orphan list, open count = %d\n",
1366 PFID(mdd_object_fid(obj)),
1369 CERROR("Object "DFID" fail to be an orphan, "
1370 "open count = %d, maybe cause failed "
1372 PFID(mdd_object_fid(obj)),
1375 rc = mdo_destroy(env, obj, th);
1377 } else if (!is_dir) {
1378 /* old files may not have link ea; ignore errors */
1379 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1387 * has mdd_write_lock on cobj already, but not on pobj yet
1389 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1390 const struct lu_attr *pattr,
1391 struct mdd_object *cobj,
1392 const struct lu_attr *cattr)
1397 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1402 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1403 struct mdd_object *p, struct mdd_object *c,
1404 const struct lu_name *name, struct md_attr *ma,
1405 struct thandle *handle, int no_name)
1407 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1410 if (likely(no_name == 0)) {
1411 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1416 rc = mdo_declare_ref_del(env, p, handle);
1420 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1421 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1422 la->la_valid = LA_CTIME | LA_MTIME;
1423 rc = mdo_declare_attr_set(env, p, la, handle);
1428 rc = mdo_declare_ref_del(env, c, handle);
1432 rc = mdo_declare_ref_del(env, c, handle);
1436 la->la_valid = LA_CTIME;
1437 rc = mdo_declare_attr_set(env, c, la, handle);
1441 rc = mdd_declare_finish_unlink(env, c, handle);
1445 /* FIXME: need changelog for remove entry */
1446 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1453 * test if a file has an HSM archive
1454 * if HSM attributes are not found in ma update them from
1457 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1458 struct mdd_object *obj,
1463 if (!(ma->ma_valid & MA_HSM)) {
1464 /* no HSM MD provided, read xattr */
1465 struct lu_buf *hsm_buf;
1466 const size_t buflen = sizeof(struct hsm_attrs);
1469 hsm_buf = mdd_buf_get(env, NULL, 0);
1470 lu_buf_alloc(hsm_buf, buflen);
1471 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM,
1472 mdd_object_capa(env, obj));
1473 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1474 lu_buf_free(hsm_buf);
1478 ma->ma_valid = MA_HSM;
1480 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1486 * Delete name entry and the object.
1487 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1488 * does not exist for this object, and it could only happen during resending
1489 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1490 * is also needed in this case(needed by changelog), so we have to add another
1491 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1492 * the ENOENT failure should be able to be fixed by redo mechanism.
1494 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1495 struct md_object *cobj, const struct lu_name *lname,
1496 struct md_attr *ma, int no_name)
1498 const char *name = lname->ln_name;
1499 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1500 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1501 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1502 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1503 struct mdd_object *mdd_cobj = NULL;
1504 struct mdd_device *mdd = mdo2mdd(pobj);
1505 struct thandle *handle;
1509 /* cobj == NULL means only delete name entry */
1510 if (likely(cobj != NULL)) {
1511 mdd_cobj = md2mdd_obj(cobj);
1512 if (mdd_object_exists(mdd_cobj) == 0)
1516 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1520 if (likely(mdd_cobj != NULL)) {
1522 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1526 is_dir = S_ISDIR(cattr->la_mode);
1529 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1533 handle = mdd_trans_create(env, mdd);
1535 RETURN(PTR_ERR(handle));
1537 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1538 lname, ma, handle, no_name);
1542 rc = mdd_trans_start(env, mdd, handle);
1546 if (likely(mdd_cobj != NULL))
1547 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1549 if (likely(no_name == 0)) {
1550 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1551 mdd_object_capa(env, mdd_pobj));
1556 if (likely(mdd_cobj != NULL)) {
1557 rc = mdo_ref_del(env, mdd_cobj, handle);
1559 __mdd_index_insert_only(env, mdd_pobj,
1562 mdd_object_capa(env, mdd_pobj));
1568 mdo_ref_del(env, mdd_cobj, handle);
1570 /* fetch updated nlink */
1571 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1576 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1577 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1579 la->la_valid = LA_CTIME | LA_MTIME;
1580 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1584 /* Enough for only unlink the entry */
1585 if (unlikely(mdd_cobj == NULL))
1588 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1589 /* update ctime of an unlinked file only if it is still
1590 * opened or a link still exists */
1591 la->la_valid = LA_CTIME;
1592 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1597 /* XXX: this transfer to ma will be removed with LOD/OSP */
1598 ma->ma_attr = *cattr;
1599 ma->ma_valid |= MA_INODE;
1600 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1602 /* fetch updated nlink */
1604 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1606 /* if object is removed then we can't get its attrs, use last get */
1607 if (cattr->la_nlink == 0) {
1608 ma->ma_attr = *cattr;
1609 ma->ma_valid |= MA_INODE;
1613 mdd_write_unlock(env, mdd_cobj);
1617 if (cattr->la_nlink == 0) {
1618 cl_flags |= CLF_UNLINK_LAST;
1619 /* search for an existing archive */
1620 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1621 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1624 rc = mdd_changelog_ns_store(env, mdd,
1625 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1626 mdd_cobj, mdd_pobj, lname, handle);
1630 mdd_trans_stop(env, mdd, rc, handle);
1636 * The permission has been checked when obj created, no need check again.
1638 static int mdd_cd_sanity_check(const struct lu_env *env,
1639 struct mdd_object *obj)
1644 if (!obj || mdd_is_dead_obj(obj))
1650 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1651 struct md_object *cobj, const struct md_op_spec *spec,
1654 struct mdd_device *mdd = mdo2mdd(cobj);
1655 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1656 struct mdd_object *son = md2mdd_obj(cobj);
1657 struct thandle *handle;
1658 const struct lu_buf *buf;
1659 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1660 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1664 rc = mdd_cd_sanity_check(env, son);
1668 if (!md_should_create(spec->sp_cr_flags))
1672 * there are following use cases for this function:
1673 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1674 * striping can be specified or not
1677 rc = mdd_la_get(env, son, attr, BYPASS_CAPA);
1681 /* calling ->ah_make_hint() is used to transfer information from parent */
1682 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1684 handle = mdd_trans_create(env, mdd);
1686 GOTO(out_free, rc = PTR_ERR(handle));
1689 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1690 * Should this be fixed?
1692 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1693 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1694 spec->sp_cr_flags, spec->no_create);
1696 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1697 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1698 spec->u.sp_ea.eadatalen);
1703 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1704 XATTR_NAME_LOV, 0, handle);
1708 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1712 rc = mdd_trans_start(env, mdd, handle);
1716 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1717 0, handle, mdd_object_capa(env, son));
1722 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1725 mdd_trans_stop(env, mdd, rc, handle);
1730 static int mdd_declare_object_initialize(const struct lu_env *env,
1731 struct mdd_object *parent,
1732 struct mdd_object *child,
1733 struct lu_attr *attr,
1734 struct thandle *handle)
1740 * inode mode has been set in creation time, and it's based on umask,
1741 * la_mode and acl, don't set here again! (which will go wrong
1742 * because below function doesn't consider umask).
1743 * I'd suggest set all object attributes in creation time, see above.
1745 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1746 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1747 rc = mdo_declare_attr_set(env, child, attr, handle);
1748 attr->la_valid |= LA_MODE | LA_TYPE;
1749 if (rc == 0 && S_ISDIR(attr->la_mode)) {
1750 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1753 rc = mdo_declare_ref_add(env, child, handle);
1755 rc = mdo_declare_index_insert(env, child, mdo2fid(parent),
1762 static int mdd_object_initialize(const struct lu_env *env,
1763 const struct lu_fid *pfid,
1764 struct mdd_object *child,
1765 struct lu_attr *attr, struct thandle *handle,
1766 const struct md_op_spec *spec)
1772 * Update attributes for child.
1775 * (1) the valid bits should be converted between Lustre and Linux;
1776 * (2) maybe, the child attributes should be set in OSD when creation.
1779 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1780 /* arguments are supposed to stay the same */
1781 if (S_ISDIR(attr->la_mode)) {
1782 /* Add "." and ".." for newly created dir */
1783 mdo_ref_add(env, child, handle);
1784 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1785 dot, handle, BYPASS_CAPA);
1787 rc = __mdd_index_insert_only(env, child, pfid,
1791 mdo_ref_del(env, child, handle);
1797 /* has not lock on pobj yet */
1798 static int mdd_create_sanity_check(const struct lu_env *env,
1799 struct md_object *pobj,
1800 const struct lu_attr *pattr,
1801 const struct lu_name *lname,
1802 struct lu_attr *cattr,
1803 struct md_op_spec *spec)
1805 struct mdd_thread_info *info = mdd_env_info(env);
1806 struct lu_fid *fid = &info->mti_fid;
1807 struct mdd_object *obj = md2mdd_obj(pobj);
1808 struct mdd_device *m = mdo2mdd(pobj);
1813 if (mdd_is_dead_obj(obj))
1817 * In some cases this lookup is not needed - we know before if name
1818 * exists or not because MDT performs lookup for it.
1819 * name length check is done in lookup.
1821 if (spec->sp_cr_lookup) {
1823 * Check if the name already exist, though it will be checked in
1824 * _index_insert also, for avoiding rolling back if exists
1827 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1828 MAY_WRITE | MAY_EXEC);
1830 RETURN(rc ? : -EEXIST);
1833 * Check WRITE permission for the parent.
1834 * EXEC permission have been checked
1835 * when lookup before create already.
1837 rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE,
1844 if (pattr->la_mode & S_ISGID) {
1845 cattr->la_gid = pattr->la_gid;
1846 if (S_ISDIR(cattr->la_mode)) {
1847 cattr->la_mode |= S_ISGID;
1848 cattr->la_valid |= LA_MODE;
1852 rc = mdd_name_check(m, lname);
1856 switch (cattr->la_mode & S_IFMT) {
1858 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1860 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1861 RETURN(-ENAMETOOLONG);
1880 static int mdd_declare_object_create(const struct lu_env *env,
1881 struct mdd_device *mdd,
1882 struct mdd_object *p, struct mdd_object *c,
1883 struct lu_attr *attr,
1884 struct thandle *handle,
1885 const struct md_op_spec *spec,
1886 struct lu_buf *def_acl_buf,
1887 struct lu_buf *acl_buf,
1888 struct dt_allocation_hint *hint)
1892 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
1897 #ifdef CONFIG_FS_POSIX_ACL
1898 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1899 /* if dir, then can inherit default ACl */
1900 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1901 XATTR_NAME_ACL_DEFAULT,
1907 if (acl_buf->lb_len > 0) {
1908 rc = mdo_declare_attr_set(env, c, attr, handle);
1912 rc = mdo_declare_xattr_set(env, c, acl_buf,
1913 XATTR_NAME_ACL_ACCESS, 0, handle);
1918 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
1922 /* replay case, create LOV EA from client data */
1923 if (spec->no_create ||
1924 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
1925 const struct lu_buf *buf;
1927 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1928 spec->u.sp_ea.eadatalen);
1929 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
1935 if (S_ISLNK(attr->la_mode)) {
1936 const char *target_name = spec->u.sp_symname;
1937 int sym_len = strlen(target_name);
1938 const struct lu_buf *buf;
1940 buf = mdd_buf_get_const(env, target_name, sym_len);
1941 rc = dt_declare_record_write(env, mdd_object_child(c),
1950 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1951 struct mdd_object *p, struct mdd_object *c,
1952 const struct lu_name *name,
1953 struct lu_attr *attr,
1954 struct thandle *handle,
1955 const struct md_op_spec *spec,
1956 struct linkea_data *ldata,
1957 struct lu_buf *def_acl_buf,
1958 struct lu_buf *acl_buf,
1959 struct dt_allocation_hint *hint)
1963 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
1964 def_acl_buf, acl_buf, hint);
1968 if (S_ISDIR(attr->la_mode)) {
1969 rc = mdo_declare_ref_add(env, p, handle);
1974 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) {
1975 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
1979 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1981 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,
1985 rc = mdd_declare_links_add(env, c, handle, ldata);
1990 la->la_valid = LA_CTIME | LA_MTIME;
1991 rc = mdo_declare_attr_set(env, p, la, handle);
1996 rc = mdd_declare_changelog_store(env, mdd, name, handle);
2000 /* XXX: For remote create, it should indicate the remote RPC
2001 * will be sent after local transaction is finished, which
2002 * is not very nice, but it will be removed once we fully support
2004 if (mdd_object_remote(p) && handle->th_update != NULL)
2005 handle->th_update->tu_sent_after_local_trans = 1;
2010 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2011 struct lu_attr *la, struct lu_buf *def_acl_buf,
2012 struct lu_buf *acl_buf)
2017 if (S_ISLNK(la->la_mode)) {
2018 acl_buf->lb_len = 0;
2019 def_acl_buf->lb_len = 0;
2023 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2024 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2025 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
2026 mdd_read_unlock(env, pobj);
2028 /* If there are default ACL, fix mode/ACL by default ACL */
2029 def_acl_buf->lb_len = rc;
2030 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2031 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2032 acl_buf->lb_len = rc;
2033 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2036 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2037 /* If there are no default ACL, fix mode by mask */
2038 struct lu_ucred *uc = lu_ucred(env);
2040 /* The create triggered by MDT internal events, such as
2041 * LFSCK reset, will not contain valid "uc". */
2042 if (unlikely(uc != NULL))
2043 la->la_mode &= ~uc->uc_umask;
2045 acl_buf->lb_len = 0;
2046 def_acl_buf->lb_len = 0;
2053 * Create a metadata object and initialize it, set acl, xattr.
2055 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2056 struct mdd_object *son, struct lu_attr *attr,
2057 struct md_op_spec *spec, struct lu_buf *acl_buf,
2058 struct lu_buf *def_acl_buf,
2059 struct dt_allocation_hint *hint,
2060 struct thandle *handle)
2064 mdd_write_lock(env, son, MOR_TGT_CHILD);
2065 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2070 #ifdef CONFIG_FS_POSIX_ACL
2071 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2072 S_ISDIR(attr->la_mode)) {
2073 /* set default acl */
2074 rc = mdo_xattr_set(env, son, def_acl_buf,
2075 XATTR_NAME_ACL_DEFAULT, 0,
2076 handle, BYPASS_CAPA);
2078 GOTO(err_destroy, rc);
2080 /* set its own acl */
2081 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2082 rc = mdo_xattr_set(env, son, acl_buf,
2083 XATTR_NAME_ACL_ACCESS,
2084 0, handle, BYPASS_CAPA);
2086 GOTO(err_destroy, rc);
2090 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2093 GOTO(err_destroy, rc);
2096 * in case of replay we just set LOVEA provided by the client
2097 * XXX: I think it would be interesting to try "old" way where
2098 * MDT calls this xattr_set(LOV) in a different transaction.
2099 * probably this way we code can be made better.
2101 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA &&
2102 S_ISREG(attr->la_mode))) {
2103 const struct lu_buf *buf;
2105 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2106 spec->u.sp_ea.eadatalen);
2107 rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle,
2110 GOTO(err_destroy, rc);
2113 if (S_ISLNK(attr->la_mode)) {
2114 struct lu_ucred *uc = lu_ucred_assert(env);
2115 struct dt_object *dt = mdd_object_child(son);
2116 const char *target_name = spec->u.sp_symname;
2117 int sym_len = strlen(target_name);
2118 const struct lu_buf *buf;
2121 buf = mdd_buf_get_const(env, target_name, sym_len);
2122 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2123 mdd_object_capa(env, son),
2125 CFS_CAP_SYS_RESOURCE_MASK);
2130 GOTO(err_initlized, rc = -EFAULT);
2134 if (unlikely(rc != 0)) {
2136 if (S_ISDIR(attr->la_mode)) {
2137 /* Drop the reference, no need to delete "."/"..",
2138 * because the object to be destroied directly. */
2139 rc2 = mdo_ref_del(env, son, handle);
2143 rc2 = mdo_ref_del(env, son, handle);
2147 mdo_destroy(env, son, handle);
2150 mdd_write_unlock(env, son);
2155 * Create object and insert it into namespace.
2157 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2158 const struct lu_name *lname, struct md_object *child,
2159 struct md_op_spec *spec, struct md_attr* ma)
2161 struct mdd_thread_info *info = mdd_env_info(env);
2162 struct lu_attr *la = &info->mti_la_for_fix;
2163 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2164 struct mdd_object *son = md2mdd_obj(child);
2165 struct mdd_device *mdd = mdo2mdd(pobj);
2166 struct lu_attr *attr = &ma->ma_attr;
2167 struct thandle *handle;
2168 struct lu_attr *pattr = &info->mti_pattr;
2169 struct lu_buf acl_buf;
2170 struct lu_buf def_acl_buf;
2171 struct linkea_data *ldata = &info->mti_link_data;
2172 const char *name = lname->ln_name;
2173 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2179 * Two operations have to be performed:
2181 * - an allocation of a new object (->do_create()), and
2183 * - an insertion into a parent index (->dio_insert()).
2185 * Due to locking, operation order is not important, when both are
2186 * successful, *but* error handling cases are quite different:
2188 * - if insertion is done first, and following object creation fails,
2189 * insertion has to be rolled back, but this operation might fail
2190 * also leaving us with dangling index entry.
2192 * - if creation is done first, is has to be undone if insertion
2193 * fails, leaving us with leaked space, which is neither good, nor
2196 * It seems that creation-first is simplest solution, but it is
2197 * sub-optimal in the frequent
2202 * case, because second mkdir is bound to create object, only to
2203 * destroy it immediately.
2205 * To avoid this follow local file systems that do double lookup:
2207 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2209 * 1. create (mdd_object_create_internal())
2211 * 2. insert (__mdd_index_insert(), lookup again)
2214 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2218 /* Sanity checks before big job. */
2219 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2223 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2224 GOTO(out_free, rc = -EINPROGRESS);
2226 handle = mdd_trans_create(env, mdd);
2228 GOTO(out_free, rc = PTR_ERR(handle));
2230 acl_buf.lb_buf = info->mti_xattr_buf;
2231 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2232 def_acl_buf.lb_buf = info->mti_key;
2233 def_acl_buf.lb_len = sizeof(info->mti_key);
2234 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2238 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2240 memset(ldata, 0, sizeof(*ldata));
2241 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2242 mdd_object_fid(mdd_pobj),
2243 lname, 1, 0, ldata);
2245 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2246 handle, spec, ldata, &def_acl_buf, &acl_buf,
2251 rc = mdd_trans_start(env, mdd, handle);
2255 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2256 &def_acl_buf, hint, handle);
2260 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2261 mdd_write_lock(env, son, MOR_TGT_CHILD);
2262 rc = __mdd_orphan_add(env, son, handle);
2263 mdd_write_unlock(env, son);
2265 GOTO(err_created, rc);
2267 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2268 name, S_ISDIR(attr->la_mode), handle,
2269 mdd_object_capa(env, mdd_pobj));
2271 GOTO(err_created, rc);
2273 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2276 /* update parent directory mtime/ctime */
2278 la->la_valid = LA_CTIME | LA_MTIME;
2279 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2281 GOTO(err_insert, rc);
2289 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2290 rc2 = __mdd_orphan_del(env, son, handle);
2292 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2293 S_ISDIR(attr->la_mode),
2294 handle, BYPASS_CAPA);
2299 mdd_write_lock(env, son, MOR_TGT_CHILD);
2300 if (S_ISDIR(attr->la_mode)) {
2301 /* Drop the reference, no need to delete "."/"..",
2302 * because the object to be destroied directly. */
2303 rc2 = mdo_ref_del(env, son, handle);
2305 mdd_write_unlock(env, son);
2309 rc2 = mdo_ref_del(env, son, handle);
2311 mdd_write_unlock(env, son);
2315 mdo_destroy(env, son, handle);
2316 mdd_write_unlock(env, son);
2319 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)))
2320 rc = mdd_changelog_ns_store(env, mdd,
2321 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2322 S_ISREG(attr->la_mode) ? CL_CREATE :
2323 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2324 0, son, mdd_pobj, lname, handle);
2326 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2330 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2331 /* if we vmalloced a large buffer drop it */
2332 lu_buf_free(ldata->ld_buf);
2334 /* The child object shouldn't be cached anymore */
2336 set_bit(LU_OBJECT_HEARD_BANSHEE,
2337 &child->mo_lu.lo_header->loh_flags);
2342 * Get locks on parents in proper order
2343 * RETURN: < 0 - error, rename_order if successful
2351 static int mdd_rename_order(const struct lu_env *env,
2352 struct mdd_device *mdd,
2353 struct mdd_object *src_pobj,
2354 const struct lu_attr *pattr,
2355 struct mdd_object *tgt_pobj)
2357 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2361 if (src_pobj == tgt_pobj)
2362 RETURN(MDD_RN_SAME);
2364 /* compared the parent child relationship of src_p&tgt_p */
2365 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2367 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2370 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2384 /* has not mdd_write{read}_lock on any obj yet. */
2385 static int mdd_rename_sanity_check(const struct lu_env *env,
2386 struct mdd_object *src_pobj,
2387 const struct lu_attr *pattr,
2388 struct mdd_object *tgt_pobj,
2389 const struct lu_attr *tpattr,
2390 struct mdd_object *sobj,
2391 const struct lu_attr *cattr,
2392 struct mdd_object *tobj,
2393 const struct lu_attr *tattr)
2398 /* XXX: when get here, sobj must NOT be NULL,
2399 * the other case has been processed in cld_rename
2400 * before mdd_rename and enable MDS_PERM_BYPASS. */
2403 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2407 /* XXX: when get here, "tobj == NULL" means tobj must
2408 * NOT exist (neither on remote MDS, such case has been
2409 * processed in cld_rename before mdd_rename and enable
2411 * So check may_create, but not check may_unlink. */
2413 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2414 (src_pobj != tgt_pobj), 0);
2416 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2417 (src_pobj != tgt_pobj), 1);
2419 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2420 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2425 static int mdd_declare_rename(const struct lu_env *env,
2426 struct mdd_device *mdd,
2427 struct mdd_object *mdd_spobj,
2428 struct mdd_object *mdd_tpobj,
2429 struct mdd_object *mdd_sobj,
2430 struct mdd_object *mdd_tobj,
2431 const struct lu_name *tname,
2432 const struct lu_name *sname,
2434 struct linkea_data *ldata,
2435 struct thandle *handle)
2437 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2440 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2441 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2447 /* name from source dir */
2448 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2452 /* .. from source child */
2453 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2454 /* source child can be directory,
2455 * counted by source dir's nlink */
2456 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2459 if (mdd_spobj != mdd_tpobj) {
2460 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2465 rc = mdo_declare_index_insert(env, mdd_sobj,
2471 /* new target child can be directory,
2472 * counted by target dir's nlink */
2473 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2478 la->la_valid = LA_CTIME | LA_MTIME;
2479 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2483 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2487 la->la_valid = LA_CTIME;
2488 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2492 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2497 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2498 tname->ln_name, handle);
2502 /* name from target dir (old name), we declare it unconditionally
2503 * as mdd_rename() calls delete unconditionally as well. so just
2504 * to balance declarations vs calls to change ... */
2505 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2509 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2510 /* delete target child in target parent directory */
2511 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2515 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2516 /* target child can be directory,
2517 * delete "." reference in target child directory */
2518 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2522 /* delete ".." reference in target parent directory */
2523 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2528 la->la_valid = LA_CTIME;
2529 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2533 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2538 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2545 /* src object can be remote that is why we use only fid and type of object */
2546 static int mdd_rename(const struct lu_env *env,
2547 struct md_object *src_pobj, struct md_object *tgt_pobj,
2548 const struct lu_fid *lf, const struct lu_name *lsname,
2549 struct md_object *tobj, const struct lu_name *ltname,
2552 const char *sname = lsname->ln_name;
2553 const char *tname = ltname->ln_name;
2554 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2555 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2556 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2557 struct mdd_device *mdd = mdo2mdd(src_pobj);
2558 struct mdd_object *mdd_sobj = NULL; /* source object */
2559 struct mdd_object *mdd_tobj = NULL;
2560 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2561 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2562 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2563 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2564 struct thandle *handle;
2565 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2566 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2567 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2570 bool tobj_locked = 0;
2571 unsigned cl_flags = 0;
2576 mdd_tobj = md2mdd_obj(tobj);
2578 mdd_sobj = mdd_object_find(env, mdd, lf);
2580 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
2582 GOTO(out_pending, rc);
2584 rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA);
2586 GOTO(out_pending, rc);
2589 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2591 GOTO(out_pending, rc);
2594 rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA);
2596 GOTO(out_pending, rc);
2598 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2599 mdd_sobj, cattr, mdd_tobj, tattr);
2601 GOTO(out_pending, rc);
2603 rc = mdd_name_check(mdd, ltname);
2605 GOTO(out_pending, rc);
2607 handle = mdd_trans_create(env, mdd);
2609 GOTO(out_pending, rc = PTR_ERR(handle));
2611 memset(ldata, 0, sizeof(*ldata));
2612 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2613 ltname, 1, 0, ldata);
2614 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2615 mdd_tobj, lsname, ltname, ma, ldata, handle);
2619 rc = mdd_trans_start(env, mdd, handle);
2623 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2624 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2626 GOTO(cleanup_unlocked, rc);
2628 is_dir = S_ISDIR(cattr->la_mode);
2630 /* Remove source name from source directory */
2631 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2632 mdd_object_capa(env, mdd_spobj));
2636 /* "mv dir1 dir2" needs "dir1/.." link update */
2637 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2638 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2639 mdd_object_capa(env, mdd_sobj));
2641 GOTO(fixup_spobj2, rc);
2643 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2644 handle, mdd_object_capa(env, mdd_sobj));
2646 GOTO(fixup_spobj, rc);
2649 /* Remove target name from target directory
2650 * Here tobj can be remote one, so we do index_delete unconditionally
2651 * and -ENOENT is allowed.
2653 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2654 mdd_object_capa(env, mdd_tpobj));
2657 /* tname might been renamed to something else */
2658 GOTO(fixup_spobj, rc);
2661 GOTO(fixup_spobj, rc);
2664 /* Insert new fid with target name into target dir */
2665 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2666 mdd_object_capa(env, mdd_tpobj));
2668 GOTO(fixup_tpobj, rc);
2670 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2671 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2673 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2675 la->la_valid = LA_CTIME;
2676 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2678 GOTO(fixup_tpobj, rc);
2681 /* Remove old target object
2682 * For tobj is remote case cmm layer has processed
2683 * and set tobj to NULL then. So when tobj is NOT NULL,
2684 * it must be local one.
2686 if (tobj && mdd_object_exists(mdd_tobj)) {
2687 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2689 if (mdd_is_dead_obj(mdd_tobj)) {
2690 /* shld not be dead, something is wrong */
2691 CERROR("tobj is dead, something is wrong\n");
2695 mdo_ref_del(env, mdd_tobj, handle);
2697 /* Remove dot reference. */
2698 if (S_ISDIR(tattr->la_mode))
2699 mdo_ref_del(env, mdd_tobj, handle);
2702 /* fetch updated nlink */
2703 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2705 CERROR("%s: Failed to get nlink for tobj "
2707 mdd2obd_dev(mdd)->obd_name,
2708 PFID(tpobj_fid), rc);
2709 GOTO(fixup_tpobj, rc);
2712 la->la_valid = LA_CTIME;
2713 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2715 CERROR("%s: Failed to set ctime for tobj "
2717 mdd2obd_dev(mdd)->obd_name,
2718 PFID(tpobj_fid), rc);
2719 GOTO(fixup_tpobj, rc);
2722 /* XXX: this transfer to ma will be removed with LOD/OSP */
2723 ma->ma_attr = *tattr;
2724 ma->ma_valid |= MA_INODE;
2725 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2728 CERROR("%s: Failed to unlink tobj "
2730 mdd2obd_dev(mdd)->obd_name,
2731 PFID(tpobj_fid), rc);
2732 GOTO(fixup_tpobj, rc);
2735 /* fetch updated nlink */
2736 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2738 CERROR("%s: Failed to get nlink for tobj "
2740 mdd2obd_dev(mdd)->obd_name,
2741 PFID(tpobj_fid), rc);
2742 GOTO(fixup_tpobj, rc);
2744 /* XXX: this transfer to ma will be removed with LOD/OSP */
2745 ma->ma_attr = *tattr;
2746 ma->ma_valid |= MA_INODE;
2748 if (tattr->la_nlink == 0) {
2749 cl_flags |= CLF_RENAME_LAST;
2750 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2751 cl_flags |= CLF_RENAME_LAST_EXISTS;
2755 la->la_valid = LA_CTIME | LA_MTIME;
2756 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2758 GOTO(fixup_tpobj, rc);
2760 if (mdd_spobj != mdd_tpobj) {
2761 la->la_valid = LA_CTIME | LA_MTIME;
2762 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2765 if (rc == 0 && mdd_sobj) {
2766 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2767 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2768 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2771 /* Old files might not have EA entry */
2772 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2773 lsname, handle, NULL, 0);
2774 mdd_write_unlock(env, mdd_sobj);
2775 /* We don't fail the transaction if the link ea can't be
2776 updated -- fid2path will use alternate lookup method. */
2784 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2787 CWARN("tp obj fix error %d\n",rc2);
2789 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2790 !mdd_is_dead_obj(mdd_tobj)) {
2792 mdo_ref_add(env, mdd_tobj, handle);
2794 mdo_ref_add(env, mdd_tobj, handle);
2797 rc2 = __mdd_index_insert(env, mdd_tpobj,
2798 mdo2fid(mdd_tobj), tname,
2803 CWARN("tp obj fix error %d\n",rc2);
2808 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2809 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2813 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2814 mdd2obd_dev(mdd)->obd_name, rc2);
2817 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2818 dotdot, handle, BYPASS_CAPA);
2820 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2821 mdd2obd_dev(mdd)->obd_name, rc2);
2826 rc2 = __mdd_index_insert(env, mdd_spobj,
2827 lf, sname, is_dir, handle, BYPASS_CAPA);
2829 CWARN("sp obj fix error %d\n",rc2);
2833 mdd_write_unlock(env, mdd_tobj);
2836 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2837 mdd_tobj, tpobj_fid, lf,
2838 spobj_fid, ltname, lsname,
2842 mdd_trans_stop(env, mdd, rc, handle);
2844 mdd_object_put(env, mdd_sobj);
2849 * During migration once the parent FID has been changed,
2850 * we need update the parent FID in linkea.
2852 static int mdd_linkea_update_child_internal(const struct lu_env *env,
2853 struct mdd_object *parent,
2854 struct mdd_object *child,
2855 const char *name, int namelen,
2856 struct thandle *handle,
2859 struct mdd_thread_info *info = mdd_env_info(env);
2860 struct linkea_data ldata = {0};
2861 struct lu_buf *buf = &info->mti_link_buf;
2867 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2868 if (buf->lb_buf == NULL)
2872 rc = mdd_links_read(env, child, &ldata);
2874 if (rc == -ENOENT || rc == -ENODATA)
2879 LASSERT(ldata.ld_leh != NULL);
2880 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
2881 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
2882 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
2883 struct lu_name lname;
2886 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
2889 if (strncmp(lname.ln_name, name, namelen) != 0 ||
2890 lu_fid_eq(&fid, mdd_object_fid(parent))) {
2891 ldata.ld_lee = (struct link_ea_entry *)
2892 ((char *)ldata.ld_lee +
2897 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
2898 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
2899 lname.ln_namelen, lname.ln_name,
2900 PFID(mdd_object_fid(parent)));
2901 /* update to the new parent fid */
2902 linkea_entry_pack(ldata.ld_lee, &lname,
2903 mdd_object_fid(parent));
2905 rc = mdd_declare_links_add(env, child, handle, &ldata);
2907 rc = mdd_links_write(env, child, &ldata, handle);
2913 static int mdd_linkea_declare_update_child(const struct lu_env *env,
2914 struct mdd_object *parent,
2915 struct mdd_object *child,
2916 const char *name, int namelen,
2917 struct thandle *handle)
2919 return mdd_linkea_update_child_internal(env, parent, child, name,
2920 namelen, handle, true);
2923 static int mdd_linkea_update_child(const struct lu_env *env,
2924 struct mdd_object *parent,
2925 struct mdd_object *child,
2926 const char *name, int namelen,
2927 struct thandle *handle)
2929 return mdd_linkea_update_child_internal(env, parent, child, name,
2930 namelen, handle, false);
2933 static int mdd_update_linkea_internal(const struct lu_env *env,
2934 struct mdd_object *mdd_pobj,
2935 struct mdd_object *mdd_sobj,
2936 struct mdd_object *mdd_tobj,
2937 const struct lu_name *child_name,
2938 struct thandle *handle,
2941 struct mdd_thread_info *info = mdd_env_info(env);
2942 struct linkea_data *ldata = &info->mti_link_data;
2947 rc = mdd_links_read(env, mdd_sobj, ldata);
2949 if (rc == -ENOENT || rc == -ENODATA)
2955 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
2957 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
2962 /* If it is mulitple links file, we need update the name entry for
2964 LASSERT(ldata->ld_leh != NULL);
2965 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
2966 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
2967 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
2968 struct mdd_object *pobj;
2969 struct lu_name lname;
2972 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
2974 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
2976 pobj = mdd_object_find(env, mdd, &fid);
2978 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
2979 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
2984 if (!mdd_object_exists(pobj)) {
2985 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
2986 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
2990 if (pobj == mdd_pobj &&
2991 lname.ln_namelen == child_name->ln_namelen &&
2992 strncmp(lname.ln_name, child_name->ln_name,
2993 lname.ln_namelen) == 0) {
2994 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
2995 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3000 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3001 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3002 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3005 /* Remove source name from source directory */
3006 /* Insert new fid with target name into target dir */
3007 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3012 rc = mdo_declare_index_insert(env, pobj,
3013 mdd_object_fid(mdd_tobj),
3014 lname.ln_name, handle);
3018 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3022 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3026 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3028 mdd_object_capa(env, pobj));
3032 rc = __mdd_index_insert(env, pobj,
3033 mdd_object_fid(mdd_tobj),
3034 lname.ln_name, 0, handle,
3035 mdd_object_capa(env, pobj));
3039 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3040 rc = mdo_ref_add(env, mdd_tobj, handle);
3041 mdd_write_unlock(env, mdd_tobj);
3045 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3046 mdo_ref_del(env, mdd_sobj, handle);
3047 mdd_write_unlock(env, mdd_sobj);
3050 mdd_object_put(env, pobj);
3058 static int mdd_migrate_xattrs(const struct lu_env *env,
3059 struct mdd_object *mdd_sobj,
3060 struct mdd_object *mdd_tobj)
3062 struct mdd_thread_info *info = mdd_env_info(env);
3063 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3065 struct thandle *handle;
3071 struct lu_buf list_xbuf;
3074 /* retrieve xattr list from the old object */
3075 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL,
3076 mdd_object_capa(env, mdd_sobj));
3077 if (list_xsize == -ENODATA)
3083 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3084 if (info->mti_big_buf.lb_buf == NULL)
3087 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3088 list_xbuf.lb_len = list_xsize;
3089 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf,
3090 mdd_object_capa(env, mdd_sobj));
3095 xname = list_xbuf.lb_buf;
3097 xlen = strnlen(xname, rem - 1) + 1;
3098 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3099 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3100 strcmp(XATTR_NAME_LMV, xname) == 0)
3103 /* For directory, if there are default layout, migrate here */
3104 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3105 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3108 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL,
3110 mdd_object_capa(env, mdd_sobj));
3111 if (xsize == -ENODATA)
3116 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3117 if (info->mti_link_buf.lb_buf == NULL)
3118 GOTO(out, rc = -ENOMEM);
3120 xbuf.lb_len = xsize;
3121 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3122 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname,
3123 mdd_object_capa(env, mdd_sobj));
3129 handle = mdd_trans_create(env, mdd);
3131 GOTO(out, rc = PTR_ERR(handle));
3133 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3136 GOTO(stop_trans, rc);
3137 /* Note: this transaction is part of migration, and it is not
3138 * the last step of migration, so we set th_local = 1 to avoid
3139 * update last rcvd for this transaction */
3140 handle->th_local = 1;
3141 rc = mdd_trans_start(env, mdd, handle);
3143 GOTO(stop_trans, rc);
3145 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle,
3146 mdd_object_capa(env, mdd_sobj));
3148 GOTO(stop_trans, rc = 0);
3151 GOTO(stop_trans, rc);
3153 mdd_trans_stop(env, mdd, rc, handle);
3158 memmove(xname, xname + xlen, rem);
3164 static int mdd_declare_migrate_create(const struct lu_env *env,
3165 struct mdd_object *mdd_pobj,
3166 struct mdd_object *mdd_sobj,
3167 struct mdd_object *mdd_tobj,
3168 struct md_op_spec *spec,
3170 union lmv_mds_md *mgr_ea,
3171 struct thandle *handle)
3173 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3174 const struct lu_buf *buf;
3178 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3179 handle, spec, NULL);
3183 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3188 if (S_ISLNK(la->la_mode)) {
3189 const char *target_name = spec->u.sp_symname;
3190 int sym_len = strlen(target_name);
3191 const struct lu_buf *buf;
3193 buf = mdd_buf_get_const(env, target_name, sym_len);
3194 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3200 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3201 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3202 spec->u.sp_ea.eadatalen);
3203 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3209 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_MIGRATE);
3210 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3211 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3216 la_flag->la_valid = LA_FLAGS;
3217 la_flag->la_flags = LUSTRE_IMMUTABLE_FL;
3218 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3219 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3224 static int mdd_migrate_create(const struct lu_env *env,
3225 struct mdd_object *mdd_pobj,
3226 struct mdd_object *mdd_sobj,
3227 struct mdd_object *mdd_tobj,
3230 struct mdd_thread_info *info = mdd_env_info(env);
3231 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3232 struct md_op_spec *spec = &info->mti_spec;
3233 struct lu_buf lmm_buf = { 0 };
3234 struct lu_buf link_buf = { 0 };
3235 const struct lu_buf *buf;
3236 struct thandle *handle;
3237 struct lmv_mds_md_v1 *mgr_ea;
3238 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3243 /* prepare spec for create */
3244 memset(spec, 0, sizeof(*spec));
3245 spec->sp_cr_lookup = 0;
3246 spec->sp_feat = &dt_directory_features;
3247 if (S_ISLNK(la->la_mode)) {
3248 buf = lu_buf_check_and_alloc(
3249 &mdd_env_info(env)->mti_big_buf,
3252 link_buf.lb_len = la->la_size + 1;
3253 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3255 rc = rc != 0 ? rc : -EFAULT;
3256 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3257 mdd2obd_dev(mdd)->obd_name,
3258 PFID(mdd_object_fid(mdd_sobj)), rc);
3261 spec->u.sp_symname = link_buf.lb_buf;
3263 /* retrieve lov of the old object */
3264 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3265 if (rc != 0 && rc != -ENODATA)
3267 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3268 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3269 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3270 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3274 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3275 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_MIGRATE);
3276 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3277 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3278 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_MIGRATION);
3279 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3280 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3282 handle = mdd_trans_create(env, mdd);
3284 GOTO(out_free, rc = PTR_ERR(handle));
3286 /* Note: this transaction is part of migration, and it is not
3287 * the last step of migration, so we set th_local = 1 to avoid
3288 * update last rcvd for this transaction */
3289 handle->th_local = 1;
3290 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3292 (union lmv_mds_md *)info->mti_xattr_buf,
3295 GOTO(stop_trans, rc);
3297 rc = mdd_trans_start(env, mdd, handle);
3299 GOTO(stop_trans, rc);
3301 /* create the target object */
3302 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3305 GOTO(stop_trans, rc);
3307 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3308 buf = mdd_buf_get_const(env, lmm_buf.lb_buf, lmm_buf.lb_len);
3309 rc = mdo_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3310 0, handle, mdd_object_capa(env, mdd_sobj));
3312 GOTO(stop_trans, rc);
3315 /* Set MIGRATE EA on the source inode, so once the migration needs
3316 * to be re-done during failover, the re-do process can locate the
3317 * target object which is already being created. */
3318 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_MIGRATE);
3319 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3320 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0,
3321 handle, mdd_object_capa(env, mdd_sobj));
3323 GOTO(stop_trans, rc);
3325 /* Set immutable flag, so any modification is disabled until
3326 * the migration is done. Once the migration is interrupted,
3327 * if the resume process find the migrating object has both
3328 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3329 * flag and approve the migration */
3330 la_flag->la_valid = LA_FLAGS;
3331 la_flag->la_flags = LUSTRE_IMMUTABLE_FL;
3332 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3333 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3334 mdd_object_capa(env, mdd_sobj));
3337 mdd_trans_stop(env, mdd, rc, handle);
3339 if (lmm_buf.lb_buf != NULL)
3340 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3344 static int mdd_migrate_entries(const struct lu_env *env,
3345 struct mdd_object *mdd_sobj,
3346 struct mdd_object *mdd_tobj)
3348 struct dt_object *next = mdd_object_child(mdd_sobj);
3349 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3350 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3351 struct thandle *handle;
3353 const struct dt_it_ops *iops;
3356 struct lu_dirent *ent;
3359 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3363 if (!dt_try_as_dir(env, next))
3364 GOTO(out_ent, rc = -ENOTDIR);
3366 * iterate directories
3368 iops = &next->do_index_ops->dio_it;
3369 it = iops->init(env, next, LUDA_FID | LUDA_TYPE,
3370 mdd_object_capa(env, mdd_sobj));
3372 GOTO(out_ent, rc = PTR_ERR(it));
3374 rc = iops->load(env, it, 0);
3376 rc = iops->next(env, it);
3380 * At this point and across for-loop:
3382 * rc == 0 -> ok, proceed.
3383 * rc > 0 -> end of directory.
3387 struct mdd_object *child;
3388 char *name = mdd_env_info(env)->mti_key;
3392 bool target_exist = false;
3394 len = iops->key_size(env, it);
3398 result = iops->rec(env, it, (struct dt_rec *)ent,
3399 LUDA_FID | LUDA_TYPE);
3400 if (result == -ESTALE)
3407 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3408 recsize = le16_to_cpu(ent->lde_reclen);
3410 /* Insert new fid with target name into target dir */
3411 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3412 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3413 ent->lde_name[1] == '.'))
3416 child = mdd_object_find(env, mdd, &ent->lde_fid);
3418 GOTO(out, rc = PTR_ERR(child));
3420 is_dir = S_ISDIR(lu_object_attr(&child->mod_obj.mo_lu));
3422 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3424 /* Check whether the name has been inserted to the target */
3425 if (dt_try_as_dir(env, dt_tobj)) {
3426 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3428 rc = dt_tobj->do_index_ops->dio_lookup(env, dt_tobj,
3429 (struct dt_rec *)fid,
3430 (struct dt_key *)name,
3431 mdd_object_capa(env, mdd_tobj));
3432 if (unlikely(rc == 0))
3433 target_exist = true;
3436 handle = mdd_trans_create(env, mdd);
3438 GOTO(out, rc = PTR_ERR(handle));
3440 /* Note: this transaction is part of migration, and it is not
3441 * the last step of migration, so we set th_local = 1 to avoid
3442 * updating last rcvd for this transaction */
3443 handle->th_local = 1;
3444 if (likely(!target_exist)) {
3445 rc = mdo_declare_index_insert(env, mdd_tobj,
3452 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3458 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3463 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3467 /* Update .. for child */
3468 rc = mdo_declare_index_delete(env, child, dotdot,
3473 rc = mdo_declare_index_insert(env, child,
3474 mdd_object_fid(mdd_tobj),
3480 rc = mdd_linkea_declare_update_child(env, mdd_tobj,
3487 rc = mdd_trans_start(env, mdd, handle);
3489 CERROR("%s: transaction start failed: rc = %d\n",
3490 mdd2obd_dev(mdd)->obd_name, rc);
3494 if (likely(!target_exist)) {
3495 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3496 name, is_dir, handle,
3497 mdd_object_capa(env, mdd_tobj));
3502 rc = mdo_ref_add(env, mdd_tobj, handle);
3509 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle,
3510 mdd_object_capa(env, mdd_sobj));
3515 rc = __mdd_index_delete_only(env, child, dotdot, handle,
3516 mdd_object_capa(env, child));
3520 rc = __mdd_index_insert_only(env, child,
3521 mdd_object_fid(mdd_tobj),
3523 mdd_object_capa(env, child));
3528 rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
3529 strlen(name), handle);
3532 mdd_object_put(env, child);
3533 mdd_trans_stop(env, mdd, rc, handle);
3537 result = iops->next(env, it);
3538 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3539 GOTO(out, rc = -EINTR);
3541 if (result == -ESTALE)
3543 } while (result == 0);
3546 iops->fini(env, it);
3548 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3552 static int mdd_declare_update_linkea(const struct lu_env *env,
3553 struct mdd_object *mdd_pobj,
3554 struct mdd_object *mdd_sobj,
3555 struct mdd_object *mdd_tobj,
3556 struct thandle *handle,
3557 const struct lu_name *child_name)
3559 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3560 child_name, handle, 1);
3563 static int mdd_update_linkea(const struct lu_env *env,
3564 struct mdd_object *mdd_pobj,
3565 struct mdd_object *mdd_sobj,
3566 struct mdd_object *mdd_tobj,
3567 struct thandle *handle,
3568 const struct lu_name *child_name)
3570 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3571 child_name, handle, 0);
3574 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3575 struct mdd_object *mdd_pobj,
3576 struct mdd_object *mdd_sobj,
3577 struct mdd_object *mdd_tobj,
3578 const struct lu_name *lname,
3580 struct lu_attr *parent_la,
3581 struct thandle *handle)
3583 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3586 /* Revert IMMUTABLE flag */
3587 la_flag->la_valid = LA_FLAGS;
3588 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3589 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3590 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3594 /* delete entry from source dir */
3595 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3599 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3600 mdd_tobj, handle, lname);
3604 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3605 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3611 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3612 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3618 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3619 lname->ln_name, handle);
3623 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3624 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3629 /* delete old object */
3630 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3634 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3635 /* delete old object */
3636 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3639 /* set nlink to 0 */
3640 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3645 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3649 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3654 static int mdd_migrate_update_name(const struct lu_env *env,
3655 struct mdd_object *mdd_pobj,
3656 struct mdd_object *mdd_sobj,
3657 struct mdd_object *mdd_tobj,
3658 const struct lu_name *lname,
3661 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3662 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3663 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3664 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3665 struct thandle *handle;
3666 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3667 const char *name = lname->ln_name;
3671 /* update time for parent */
3672 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3673 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3674 p_la->la_valid = LA_CTIME;
3676 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3680 handle = mdd_trans_create(env, mdd);
3682 RETURN(PTR_ERR(handle));
3684 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3685 lname, so_attr, p_la, handle);
3687 /* If the migration can not be fit in one transaction, just
3688 * leave it in the original MDT */
3690 GOTO(stop_trans, rc = 0);
3692 GOTO(stop_trans, rc);
3695 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3696 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3697 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3698 PFID(mdd_object_fid(mdd_tobj)));
3700 rc = mdd_trans_start(env, mdd, handle);
3702 GOTO(stop_trans, rc);
3704 /* Revert IMMUTABLE flag */
3705 la_flag->la_valid = LA_FLAGS;
3706 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3707 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3708 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3709 mdd_object_capa(env, mdd_pobj));
3711 GOTO(stop_trans, rc);
3713 /* Remove source name from source directory */
3714 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
3715 mdd_object_capa(env, mdd_pobj));
3717 GOTO(stop_trans, rc);
3719 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3722 GOTO(stop_trans, rc);
3724 if (S_ISREG(so_attr->la_mode)) {
3725 if (so_attr->la_nlink == 1) {
3726 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3728 mdd_object_capa(env, mdd_sobj));
3729 if (rc != 0 && rc != -ENODATA)
3730 GOTO(stop_trans, rc);
3734 /* Insert new fid with target name into target dir */
3735 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj), name,
3736 is_dir, handle, mdd_object_capa(env, mdd_pobj));
3738 GOTO(stop_trans, rc);
3740 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3743 GOTO(stop_trans, rc);
3745 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3746 mdo_ref_del(env, mdd_sobj, handle);
3748 mdo_ref_del(env, mdd_sobj, handle);
3750 /* Get the attr again after ref_del */
3751 rc = mdd_la_get(env, mdd_sobj, so_attr,
3752 mdd_object_capa(env, mdd_sobj));
3754 GOTO(stop_trans, rc);
3755 ma->ma_attr = *so_attr;
3756 ma->ma_valid |= MA_INODE;
3757 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3759 GOTO(stop_trans, rc);
3761 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3763 GOTO(stop_trans, rc);
3765 mdd_write_unlock(env, mdd_sobj);
3768 mdd_trans_stop(env, mdd, rc, handle);
3774 * Check whether we should migrate the file/dir
3776 * < 0 permission check failed or other error.
3777 * = 0 the file can be migrated.
3778 * > 0 the file does not need to be migrated, mostly
3779 * for multiple link file
3781 static int mdd_migrate_sanity_check(const struct lu_env *env,
3782 struct mdd_object *pobj,
3783 const struct lu_attr *pattr,
3784 struct mdd_object *sobj,
3785 struct lu_attr *sattr)
3787 struct mdd_thread_info *info = mdd_env_info(env);
3788 struct linkea_data *ldata = &info->mti_link_data;
3790 struct lu_buf *mgr_buf;
3796 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_MIGRATE);
3797 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3798 if (mgr_buf->lb_buf == NULL)
3801 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV,
3802 mdd_object_capa(env, sobj));
3804 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3806 /* If the object has migrateEA, it means IMMUTE flag
3807 * is being set by previous migration process, so it
3808 * needs to override the IMMUTE flag, otherwise the
3809 * following sanity check will fail */
3810 if (le32_to_cpu(lmm->lmv_md_v1.lmv_magic) ==
3811 LMV_MAGIC_MIGRATE) {
3812 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3814 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3815 sobj->mod_flags &= ~IMMUTE_OBJ;
3816 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3817 mdd2obd_dev(mdd)->obd_name,
3818 PFID(mdd_object_fid(sobj)));
3822 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3823 sobj, sattr, NULL, NULL);
3827 /* Then it will check if the file should be migrated. If the file
3828 * has mulitple links, we only need migrate the file if all of its
3829 * entries has been migrated to the remote MDT */
3830 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3833 rc = mdd_links_read(env, sobj, ldata);
3835 /* For multiple links files, if there are no linkEA data at all,
3836 * means the file might be created before linkEA is enabled, and
3837 * all all of its links should not be migrated yet, otherwise
3838 * it should have some linkEA there */
3839 if (rc == -ENOENT || rc == -ENODATA)
3844 /* If it is mulitple links file, we need update the name entry for
3846 LASSERT(ldata->ld_leh != NULL);
3847 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3848 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3849 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3850 struct mdd_object *lpobj;
3851 struct lu_name lname;
3854 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3856 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3858 lpobj = mdd_object_find(env, mdd, &fid);
3859 if (IS_ERR(lpobj)) {
3860 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3861 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3866 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
3867 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
3868 PFID(&fid), lname.ln_namelen, lname.ln_name);
3869 mdd_object_put(env, lpobj);
3873 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
3874 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
3875 lname.ln_name, PFID(&fid));
3876 mdd_object_put(env, lpobj);
3884 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
3885 struct md_object *sobj, const struct lu_name *lname,
3886 struct md_object *tobj, struct md_attr *ma)
3888 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
3889 struct mdd_device *mdd = mdo2mdd(pobj);
3890 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
3891 struct mdd_object *mdd_tobj = NULL;
3892 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3893 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3897 /* If the file will being migrated, it will check whether
3898 * the file is being opened by someone else right now */
3899 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
3900 if (mdd_sobj->mod_count >= 1) {
3901 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
3902 mdd2obd_dev(mdd)->obd_name,
3903 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3904 mdd_sobj->mod_count, -EBUSY);
3905 mdd_read_unlock(env, mdd_sobj);
3906 GOTO(put, rc = -EBUSY);
3908 mdd_read_unlock(env, mdd_sobj);
3910 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3914 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
3918 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
3925 /* Sigh, it is impossible to finish all of migration in a single
3926 * transaction, for example migrating big directory entries to the
3927 * new MDT, it needs insert all of name entries of children in the
3930 * So migration will be done in multiple steps and transactions.
3932 * 1. create an orphan object on the remote MDT in one transaction.
3933 * 2. migrate extend attributes to the new target file/directory.
3934 * 3. For directory, migrate the entries to the new MDT and update
3935 * linkEA of each children. Because we can not migrate all entries
3936 * in a single transaction, so the migrating directory will become
3937 * a striped directory during migration, so once the process is
3938 * interrupted, the directory is still accessible. (During lookup,
3939 * client will locate the name by searching both original and target
3941 * 4. Finally, update the name/FID to point to the new file/directory
3942 * in a separate transaction.
3945 /* step 1: Check whether the orphan object has been created, and create
3946 * orphan object on the remote MDT if needed */
3947 mdd_tobj = md2mdd_obj(tobj);
3948 if (!mdd_object_exists(mdd_tobj)) {
3949 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3955 /* step 2: migrate xattr */
3956 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
3960 /* step 3: migrate name entries to the orphan object */
3961 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
3962 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
3965 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
3966 OBD_FAIL_MDS_REINT_NET_REP)))
3970 /* step 4: update name entry to the new object */
3971 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
3979 const struct md_dir_operations mdd_dir_ops = {
3980 .mdo_is_subdir = mdd_is_subdir,
3981 .mdo_lookup = mdd_lookup,
3982 .mdo_create = mdd_create,
3983 .mdo_rename = mdd_rename,
3984 .mdo_link = mdd_link,
3985 .mdo_unlink = mdd_unlink,
3986 .mdo_create_data = mdd_create_data,
3987 .mdo_migrate = mdd_migrate,