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 struct dt_key *)"");
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);
299 * Determine if the target object can be hard linked, and right now it only
300 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
301 * directory nlink count might exceed the maximum link count(see
302 * osd_object_ref_add), so it only check nlink for non-directories.
304 * \param[in] env thread environment
305 * \param[in] obj object being linked to
306 * \param[in] la attributes of \a obj
308 * \retval 0 if \a obj can be hard linked
309 * \retval negative error if \a obj is a directory or has too
312 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
313 const struct lu_attr *la)
315 struct mdd_device *m = mdd_obj2mdd_dev(obj);
320 /* Subdir count limitation can be broken through
321 * (see osd_object_ref_add), so only check non-directory here. */
322 if (!S_ISDIR(la->la_mode) &&
323 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
330 * Check whether it may create the cobj under the pobj.
332 * \param[in] env execution environment
333 * \param[in] pobj the parent directory
334 * \param[in] pattr the attribute of the parent directory
335 * \param[in] cobj the child to be created
336 * \param[in] check_perm if check WRITE|EXEC permission for parent
338 * \retval = 0 create the child under this dir is allowed
339 * \retval negative errno create the child under this dir is
342 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
343 const struct lu_attr *pattr, struct mdd_object *cobj,
346 struct mdd_thread_info *info = mdd_env_info(env);
351 if (cobj && mdd_object_exists(cobj))
354 if (mdd_is_dead_obj(pobj))
357 /* If the parent is a sub-stripe, check whether it is dead */
358 xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key));
359 rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV,
360 mdd_object_capa(env, pobj));
361 if (unlikely(rc > 0)) {
362 struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf;
364 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE &&
365 le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD)
371 rc = mdd_permission_internal_locked(env, pobj, pattr,
372 MAY_WRITE | MAY_EXEC,
378 * Check whether can unlink from the pobj in the case of "cobj == NULL".
380 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
381 const struct lu_attr *pattr, const struct lu_attr *attr)
386 if (mdd_is_dead_obj(pobj))
389 if ((attr->la_valid & LA_FLAGS) &&
390 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
393 rc = mdd_permission_internal_locked(env, pobj, pattr,
394 MAY_WRITE | MAY_EXEC,
399 if (mdd_is_append(pobj))
406 * pobj == NULL is remote ops case, under such case, pobj's
407 * VTX feature has been checked already, no need check again.
409 static inline int mdd_is_sticky(const struct lu_env *env,
410 struct mdd_object *pobj,
411 const struct lu_attr *pattr,
412 struct mdd_object *cobj,
413 const struct lu_attr *cattr)
415 struct lu_ucred *uc = lu_ucred_assert(env);
418 LASSERT(pattr != NULL);
419 if (!(pattr->la_mode & S_ISVTX) ||
420 (pattr->la_uid == uc->uc_fsuid))
424 LASSERT(cattr != NULL);
425 if (cattr->la_uid == uc->uc_fsuid)
428 return !md_capable(uc, CFS_CAP_FOWNER);
431 static int mdd_may_delete_entry(const struct lu_env *env,
432 struct mdd_object *pobj,
433 const struct lu_attr *pattr,
438 LASSERT(pobj != NULL);
439 if (!mdd_object_exists(pobj))
442 if (mdd_is_dead_obj(pobj))
447 rc = mdd_permission_internal_locked(env, pobj, pattr,
448 MAY_WRITE | MAY_EXEC,
454 if (mdd_is_append(pobj))
461 * Check whether it may delete the cobj from the pobj.
464 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
465 const struct lu_attr *tpattr, struct mdd_object *tobj,
466 const struct lu_attr *tattr, const struct lu_attr *cattr,
467 int check_perm, int check_empty)
473 LASSERT(tpattr != NULL);
474 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
482 if (!mdd_object_exists(tobj))
485 if (mdd_is_dead_obj(tobj))
488 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
491 if (mdd_is_immutable(tobj) || mdd_is_append(tobj))
494 if ((tattr->la_valid & LA_FLAGS) &&
495 (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
498 /* additional check the rename case */
500 if (S_ISDIR(cattr->la_mode)) {
501 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
503 if (!S_ISDIR(tattr->la_mode))
506 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
508 } else if (S_ISDIR(tattr->la_mode))
512 if (S_ISDIR(tattr->la_mode) && check_empty)
513 rc = mdd_dir_is_empty(env, tobj);
519 * Check whether it can create the link file(linked to @src_obj) under
520 * the target directory(@tgt_obj), and src_obj has been locked by
523 * \param[in] env execution environment
524 * \param[in] tgt_obj the target directory
525 * \param[in] tattr attributes of target directory
526 * \param[in] lname the link name
527 * \param[in] src_obj source object for link
528 * \param[in] cattr attributes for source object
530 * \retval = 0 it is allowed to create the link file under tgt_obj
531 * \retval negative error not allowed to create the link file
533 static int mdd_link_sanity_check(const struct lu_env *env,
534 struct mdd_object *tgt_obj,
535 const struct lu_attr *tattr,
536 const struct lu_name *lname,
537 struct mdd_object *src_obj,
538 const struct lu_attr *cattr)
540 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
544 if (!mdd_object_exists(src_obj))
547 if (mdd_is_dead_obj(src_obj))
550 /* Local ops, no lookup before link, check filename length here. */
551 rc = mdd_name_check(m, lname);
555 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
558 if (S_ISDIR(mdd_object_type(src_obj)))
561 LASSERT(src_obj != tgt_obj);
562 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
566 rc = __mdd_may_link(env, src_obj, cattr);
571 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
572 const char *name, struct thandle *handle,
573 struct lustre_capa *capa)
575 struct dt_object *next = mdd_object_child(pobj);
579 if (dt_try_as_dir(env, next)) {
580 rc = next->do_index_ops->dio_delete(env, next,
581 (struct dt_key *)name,
589 static int __mdd_index_insert_only(const struct lu_env *env,
590 struct mdd_object *pobj,
591 const struct lu_fid *lf, __u32 type,
593 struct thandle *handle,
594 struct lustre_capa *capa)
596 struct dt_object *next = mdd_object_child(pobj);
600 if (dt_try_as_dir(env, next)) {
601 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
602 struct lu_ucred *uc = lu_ucred_check(env);
606 rec->rec_type = type;
607 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
608 rc = next->do_index_ops->dio_insert(env, next,
609 (const struct dt_rec *)rec,
610 (const struct dt_key *)name,
611 handle, capa, ignore_quota);
618 /* insert named index, add reference if isdir */
619 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
620 const struct lu_fid *lf, __u32 type,
621 const char *name, struct thandle *handle,
622 struct lustre_capa *capa)
627 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle, capa);
628 if (rc == 0 && S_ISDIR(type)) {
629 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
630 mdo_ref_add(env, pobj, handle);
631 mdd_write_unlock(env, pobj);
637 /* delete named index, drop reference if isdir */
638 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
639 const char *name, int is_dir, struct thandle *handle,
640 struct lustre_capa *capa)
645 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
646 if (rc == 0 && is_dir) {
647 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
648 mdo_ref_del(env, pobj, handle);
649 mdd_write_unlock(env, pobj);
655 int mdd_declare_changelog_store(const struct lu_env *env,
656 struct mdd_device *mdd,
657 const struct lu_name *fname,
658 struct thandle *handle)
660 struct obd_device *obd = mdd2obd_dev(mdd);
661 struct llog_ctxt *ctxt;
662 struct llog_changelog_rec *rec;
668 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
671 reclen = llog_data_len(sizeof(*rec) +
672 (fname != NULL ? fname->ln_namelen : 0));
673 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
674 if (buf->lb_buf == NULL)
678 rec->cr_hdr.lrh_len = reclen;
679 rec->cr_hdr.lrh_type = CHANGELOG_REC;
681 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
685 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
691 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
692 struct mdd_device *mdd,
693 const struct lu_name *tname,
694 const struct lu_name *sname,
695 struct thandle *handle)
697 struct obd_device *obd = mdd2obd_dev(mdd);
698 struct llog_ctxt *ctxt;
699 struct llog_changelog_ext_rec *rec;
705 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
708 reclen = llog_data_len(sizeof(*rec) +
709 (tname != NULL ? tname->ln_namelen : 0) +
710 (sname != NULL ? 1 + sname->ln_namelen : 0));
711 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
712 if (buf->lb_buf == NULL)
716 rec->cr_hdr.lrh_len = reclen;
717 rec->cr_hdr.lrh_type = CHANGELOG_REC;
719 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
723 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
729 /** Add a changelog entry \a rec to the changelog llog
732 * \param handle - currently ignored since llogs start their own transaction;
733 * this will hopefully be fixed in llog rewrite
736 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
737 struct llog_changelog_rec *rec, struct thandle *th)
739 struct obd_device *obd = mdd2obd_dev(mdd);
740 struct llog_ctxt *ctxt;
743 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
744 rec->cr_hdr.lrh_type = CHANGELOG_REC;
745 rec->cr.cr_time = cl_time();
747 spin_lock(&mdd->mdd_cl.mc_lock);
748 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
749 * but as long as the MDD transactions are ordered correctly for e.g.
750 * rename conflicts, I don't think this should matter. */
751 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
752 spin_unlock(&mdd->mdd_cl.mc_lock);
754 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
758 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
765 /** Add a changelog_ext entry \a rec to the changelog llog
768 * \param handle - currently ignored since llogs start their own transaction;
769 * this will hopefully be fixed in llog rewrite
773 mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
774 struct llog_changelog_ext_rec *rec, struct thandle *th)
776 struct obd_device *obd = mdd2obd_dev(mdd);
777 struct llog_ctxt *ctxt;
780 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
781 /* llog_lvfs_write_rec sets the llog tail len */
782 rec->cr_hdr.lrh_type = CHANGELOG_REC;
783 rec->cr.cr_time = cl_time();
785 spin_lock(&mdd->mdd_cl.mc_lock);
786 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
787 * but as long as the MDD transactions are ordered correctly for e.g.
788 * rename conflicts, I don't think this should matter. */
789 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
790 spin_unlock(&mdd->mdd_cl.mc_lock);
792 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
796 /* nested journal transaction */
797 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
805 /** Store a namespace change changelog record
806 * If this fails, we must fail the whole transaction; we don't
807 * want the change to commit without the log entry.
808 * \param target - mdd_object of change
809 * \param parent - parent dir/object
810 * \param tname - target name string
811 * \param handle - transacion handle
813 int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd,
814 enum changelog_rec_type type, unsigned flags,
815 struct mdd_object *target, struct mdd_object *parent,
816 const struct lu_name *tname, struct thandle *handle)
818 struct llog_changelog_rec *rec;
825 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
827 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
830 LASSERT(target != NULL);
831 LASSERT(parent != NULL);
832 LASSERT(tname != NULL);
833 LASSERT(handle != NULL);
835 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
836 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
837 if (buf->lb_buf == NULL)
841 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
842 rec->cr.cr_type = (__u32)type;
843 rec->cr.cr_tfid = *mdo2fid(target);
844 rec->cr.cr_pfid = *mdo2fid(parent);
845 rec->cr.cr_namelen = tname->ln_namelen;
846 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
848 target->mod_cltime = cfs_time_current_64();
850 rc = mdd_changelog_store(env, mdd, rec, handle);
852 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
853 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
854 PFID(&rec->cr.cr_pfid));
862 /** Store a namespace change changelog record
863 * If this fails, we must fail the whole transaction; we don't
864 * want the change to commit without the log entry.
865 * \param target - mdd_object of change
866 * \param tpfid - target parent dir/object fid
867 * \param sfid - source object fid
868 * \param spfid - source parent fid
869 * \param tname - target name string
870 * \param sname - source name string
871 * \param handle - transacion handle
873 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
874 struct mdd_device *mdd,
875 enum changelog_rec_type type,
877 struct mdd_object *target,
878 const struct lu_fid *tpfid,
879 const struct lu_fid *sfid,
880 const struct lu_fid *spfid,
881 const struct lu_name *tname,
882 const struct lu_name *sname,
883 struct thandle *handle)
885 struct llog_changelog_ext_rec *rec;
892 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
894 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
897 LASSERT(sfid != NULL);
898 LASSERT(tpfid != NULL);
899 LASSERT(tname != NULL);
900 LASSERT(handle != NULL);
902 reclen = llog_data_len(sizeof(*rec) +
903 sname != NULL ? 1 + sname->ln_namelen : 0);
904 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
905 if (buf->lb_buf == NULL)
909 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
910 rec->cr.cr_type = (__u32)type;
911 rec->cr.cr_pfid = *tpfid;
912 rec->cr.cr_sfid = *sfid;
913 rec->cr.cr_spfid = *spfid;
914 rec->cr.cr_namelen = tname->ln_namelen;
915 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
917 rec->cr.cr_name[tname->ln_namelen] = '\0';
918 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
920 rec->cr.cr_namelen += 1 + sname->ln_namelen;
923 if (likely(target != NULL)) {
924 rec->cr.cr_tfid = *mdo2fid(target);
925 target->mod_cltime = cfs_time_current_64();
927 fid_zero(&rec->cr.cr_tfid);
930 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
932 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
933 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
940 static int __mdd_links_add(const struct lu_env *env,
941 struct mdd_object *mdd_obj,
942 struct linkea_data *ldata,
943 const struct lu_name *lname,
944 const struct lu_fid *pfid,
945 int first, int check)
949 if (ldata->ld_leh == NULL) {
950 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
954 rc = linkea_data_new(ldata,
955 &mdd_env_info(env)->mti_link_buf);
962 rc = linkea_links_find(ldata, lname, pfid);
963 if (rc && rc != -ENOENT)
969 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
970 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
974 linkea_add_buf(ldata, lname, tfid);
977 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
978 linkea_add_buf(ldata, lname, pfid);
980 return linkea_add_buf(ldata, lname, pfid);
983 static int __mdd_links_del(const struct lu_env *env,
984 struct mdd_object *mdd_obj,
985 struct linkea_data *ldata,
986 const struct lu_name *lname,
987 const struct lu_fid *pfid)
991 if (ldata->ld_leh == NULL) {
992 rc = mdd_links_read(env, mdd_obj, ldata);
997 rc = linkea_links_find(ldata, lname, pfid);
1001 linkea_del_buf(ldata, lname);
1005 static int mdd_linkea_prepare(const struct lu_env *env,
1006 struct mdd_object *mdd_obj,
1007 const struct lu_fid *oldpfid,
1008 const struct lu_name *oldlname,
1009 const struct lu_fid *newpfid,
1010 const struct lu_name *newlname,
1011 int first, int check,
1012 struct linkea_data *ldata)
1018 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1021 LASSERT(oldpfid != NULL || newpfid != NULL);
1023 if (mdd_obj->mod_flags & DEAD_OBJ) {
1024 /* Prevent linkea to be updated which is NOT necessary. */
1025 ldata->ld_reclen = 0;
1026 /* No more links, don't bother */
1030 if (oldpfid != NULL) {
1031 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1034 (rc != -ENODATA && rc != -ENOENT))
1036 /* No changes done. */
1041 /* If renaming, add the new record */
1042 if (newpfid != NULL) {
1043 /* even if the add fails, we still delete the out-of-date
1045 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1049 rc = rc != 0 ? rc : rc2;
1054 int mdd_links_rename(const struct lu_env *env,
1055 struct mdd_object *mdd_obj,
1056 const struct lu_fid *oldpfid,
1057 const struct lu_name *oldlname,
1058 const struct lu_fid *newpfid,
1059 const struct lu_name *newlname,
1060 struct thandle *handle,
1061 struct linkea_data *ldata,
1062 int first, int check)
1067 if (ldata == NULL) {
1068 ldata = &mdd_env_info(env)->mti_link_data;
1069 memset(ldata, 0, sizeof(*ldata));
1070 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1071 newpfid, newlname, first, check,
1077 if (ldata->ld_reclen != 0)
1078 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1083 if (rc == -EOVERFLOW || rc == -ENOSPC)
1085 if (oldpfid == NULL)
1086 CDEBUG(error ? D_ERROR : D_OTHER,
1087 "link_ea add '%.*s' failed %d "DFID"\n",
1088 newlname->ln_namelen, newlname->ln_name,
1089 rc, PFID(mdd_object_fid(mdd_obj)));
1090 else if (newpfid == NULL)
1091 CDEBUG(error ? D_ERROR : D_OTHER,
1092 "link_ea del '%.*s' failed %d "DFID"\n",
1093 oldlname->ln_namelen, oldlname->ln_name,
1094 rc, PFID(mdd_object_fid(mdd_obj)));
1096 CDEBUG(error ? D_ERROR : D_OTHER,
1097 "link_ea rename '%.*s'->'%.*s' failed %d "
1099 oldlname->ln_namelen, oldlname->ln_name,
1100 newlname->ln_namelen, newlname->ln_name,
1101 rc, PFID(mdd_object_fid(mdd_obj)));
1104 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1105 /* if we vmalloced a large buffer drop it */
1106 lu_buf_free(ldata->ld_buf);
1111 static inline int mdd_links_add(const struct lu_env *env,
1112 struct mdd_object *mdd_obj,
1113 const struct lu_fid *pfid,
1114 const struct lu_name *lname,
1115 struct thandle *handle,
1116 struct linkea_data *ldata, int first)
1118 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1119 pfid, lname, handle, ldata, first, 0);
1122 static inline int mdd_links_del(const struct lu_env *env,
1123 struct mdd_object *mdd_obj,
1124 const struct lu_fid *pfid,
1125 const struct lu_name *lname,
1126 struct thandle *handle)
1128 return mdd_links_rename(env, mdd_obj, pfid, lname,
1129 NULL, NULL, handle, NULL, 0, 0);
1132 /** Read the link EA into a temp buffer.
1133 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1134 * A pointer to the buffer is stored in \a ldata::ld_buf.
1136 * \retval 0 or error
1138 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1139 struct linkea_data *ldata)
1143 if (!mdd_object_exists(mdd_obj))
1146 /* First try a small buf */
1147 LASSERT(env != NULL);
1148 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1150 if (ldata->ld_buf->lb_buf == NULL)
1153 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1155 if (rc == -ERANGE) {
1156 /* Buf was too small, figure out what we need. */
1157 lu_buf_free(ldata->ld_buf);
1158 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1159 XATTR_NAME_LINK, BYPASS_CAPA);
1162 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1163 if (ldata->ld_buf->lb_buf == NULL)
1165 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1166 XATTR_NAME_LINK, BYPASS_CAPA);
1169 lu_buf_free(ldata->ld_buf);
1170 ldata->ld_buf = NULL;
1174 return linkea_init(ldata);
1177 /** Read the link EA into a temp buffer.
1178 * Uses the name_buf since it is generally large.
1179 * \retval IS_ERR err
1180 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1182 struct lu_buf *mdd_links_get(const struct lu_env *env,
1183 struct mdd_object *mdd_obj)
1185 struct linkea_data ldata = { 0 };
1188 rc = mdd_links_read(env, mdd_obj, &ldata);
1189 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1192 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1193 struct linkea_data *ldata, struct thandle *handle)
1195 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1196 ldata->ld_leh->leh_len);
1197 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1198 mdd_object_capa(env, mdd_obj));
1201 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1202 struct thandle *handle, struct linkea_data *ldata)
1208 if (ldata != NULL && ldata->ld_leh != NULL) {
1209 ea_len = ldata->ld_leh->leh_len;
1210 linkea = ldata->ld_buf->lb_buf;
1212 ea_len = DEFAULT_LINKEA_SIZE;
1216 /* XXX: max size? */
1217 rc = mdo_declare_xattr_set(env, mdd_obj,
1218 mdd_buf_get_const(env, linkea, ea_len),
1219 XATTR_NAME_LINK, 0, handle);
1223 static inline int mdd_declare_links_del(const struct lu_env *env,
1224 struct mdd_object *c,
1225 struct thandle *handle)
1229 /* For directory, the linkEA will be removed together
1230 * with the object. */
1231 if (!S_ISDIR(mdd_object_type(c)))
1232 rc = mdd_declare_links_add(env, c, handle, NULL);
1237 static int mdd_declare_link(const struct lu_env *env,
1238 struct mdd_device *mdd,
1239 struct mdd_object *p,
1240 struct mdd_object *c,
1241 const struct lu_name *name,
1242 struct thandle *handle,
1244 struct linkea_data *data)
1248 rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
1249 name->ln_name, handle);
1253 rc = mdo_declare_ref_add(env, c, handle);
1257 la->la_valid = LA_CTIME | LA_MTIME;
1258 rc = mdo_declare_attr_set(env, p, la, handle);
1262 la->la_valid = LA_CTIME;
1263 rc = mdo_declare_attr_set(env, c, la, handle);
1267 rc = mdd_declare_links_add(env, c, handle, data);
1271 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1276 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1277 struct md_object *src_obj, const struct lu_name *lname,
1280 const char *name = lname->ln_name;
1281 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1282 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1283 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1284 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1285 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1286 struct mdd_device *mdd = mdo2mdd(src_obj);
1287 struct thandle *handle;
1288 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1292 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
1296 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
1300 handle = mdd_trans_create(env, mdd);
1302 GOTO(out_pending, rc = PTR_ERR(handle));
1304 memset(ldata, 0, sizeof(*ldata));
1306 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1307 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1309 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1314 rc = mdd_trans_start(env, mdd, handle);
1318 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1319 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1322 GOTO(out_unlock, rc);
1324 rc = mdo_ref_add(env, mdd_sobj, handle);
1326 GOTO(out_unlock, rc);
1329 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1330 mdd_object_type(mdd_sobj), name, handle,
1331 mdd_object_capa(env, mdd_tobj));
1333 mdo_ref_del(env, mdd_sobj, handle);
1334 GOTO(out_unlock, rc);
1337 la->la_valid = LA_CTIME | LA_MTIME;
1338 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1340 GOTO(out_unlock, rc);
1342 la->la_valid = LA_CTIME;
1343 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1345 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1346 mdo2fid(mdd_tobj), lname, 0, 0,
1349 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1350 lname, handle, ldata, 0);
1351 /* The failure of links_add should not cause the link
1352 * failure, reset rc here */
1357 mdd_write_unlock(env, mdd_sobj);
1359 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1360 mdd_tobj, lname, handle);
1362 mdd_trans_stop(env, mdd, rc, handle);
1364 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1365 /* if we vmalloced a large buffer drop it */
1366 lu_buf_free(ldata->ld_buf);
1371 static int mdd_mark_dead_object(const struct lu_env *env,
1372 struct mdd_object *obj, struct thandle *handle,
1375 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1379 obj->mod_flags |= DEAD_OBJ;
1381 if (!S_ISDIR(mdd_object_type(obj)))
1384 attr->la_valid = LA_FLAGS;
1385 attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
1388 rc = mdo_declare_attr_set(env, obj, attr, handle);
1390 rc = mdo_attr_set(env, obj, attr, handle,
1391 mdd_object_capa(env, obj));
1396 static int mdd_declare_finish_unlink(const struct lu_env *env,
1397 struct mdd_object *obj,
1398 struct thandle *handle)
1402 rc = mdd_mark_dead_object(env, obj, handle, true);
1406 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1410 rc = mdo_declare_destroy(env, obj, handle);
1414 return mdd_declare_links_del(env, obj, handle);
1417 /* caller should take a lock before calling */
1418 int mdd_finish_unlink(const struct lu_env *env,
1419 struct mdd_object *obj, struct md_attr *ma,
1420 const struct mdd_object *pobj,
1421 const struct lu_name *lname,
1425 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1428 LASSERT(mdd_write_locked(env, obj) != 0);
1430 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1431 rc = mdd_mark_dead_object(env, obj, th, false);
1435 /* add new orphan and the object
1436 * will be deleted during mdd_close() */
1437 if (obj->mod_count) {
1438 rc = __mdd_orphan_add(env, obj, th);
1440 CDEBUG(D_HA, "Object "DFID" is inserted into "
1441 "orphan list, open count = %d\n",
1442 PFID(mdd_object_fid(obj)),
1445 CERROR("Object "DFID" fail to be an orphan, "
1446 "open count = %d, maybe cause failed "
1448 PFID(mdd_object_fid(obj)),
1451 rc = mdo_destroy(env, obj, th);
1453 } else if (!is_dir) {
1454 /* old files may not have link ea; ignore errors */
1455 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1463 * has mdd_write_lock on cobj already, but not on pobj yet
1465 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1466 const struct lu_attr *pattr,
1467 struct mdd_object *cobj,
1468 const struct lu_attr *cattr)
1473 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1478 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1479 struct mdd_object *p, struct mdd_object *c,
1480 const struct lu_name *name, struct md_attr *ma,
1481 struct thandle *handle, int no_name)
1483 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1486 if (likely(no_name == 0)) {
1487 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1492 rc = mdo_declare_ref_del(env, p, handle);
1496 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1497 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1498 la->la_valid = LA_CTIME | LA_MTIME;
1499 rc = mdo_declare_attr_set(env, p, la, handle);
1504 rc = mdo_declare_ref_del(env, c, handle);
1508 rc = mdo_declare_ref_del(env, c, handle);
1512 la->la_valid = LA_CTIME;
1513 rc = mdo_declare_attr_set(env, c, la, handle);
1517 rc = mdd_declare_finish_unlink(env, c, handle);
1521 /* FIXME: need changelog for remove entry */
1522 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1529 * test if a file has an HSM archive
1530 * if HSM attributes are not found in ma update them from
1533 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1534 struct mdd_object *obj,
1539 if (!(ma->ma_valid & MA_HSM)) {
1540 /* no HSM MD provided, read xattr */
1541 struct lu_buf *hsm_buf;
1542 const size_t buflen = sizeof(struct hsm_attrs);
1545 hsm_buf = mdd_buf_get(env, NULL, 0);
1546 lu_buf_alloc(hsm_buf, buflen);
1547 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM,
1548 mdd_object_capa(env, obj));
1549 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1550 lu_buf_free(hsm_buf);
1554 ma->ma_valid = MA_HSM;
1556 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1562 * Delete name entry and the object.
1563 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1564 * does not exist for this object, and it could only happen during resending
1565 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1566 * is also needed in this case(needed by changelog), so we have to add another
1567 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1568 * the ENOENT failure should be able to be fixed by redo mechanism.
1570 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1571 struct md_object *cobj, const struct lu_name *lname,
1572 struct md_attr *ma, int no_name)
1574 const char *name = lname->ln_name;
1575 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1576 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1577 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1578 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1579 struct mdd_object *mdd_cobj = NULL;
1580 struct mdd_device *mdd = mdo2mdd(pobj);
1581 struct thandle *handle;
1585 /* cobj == NULL means only delete name entry */
1586 if (likely(cobj != NULL)) {
1587 mdd_cobj = md2mdd_obj(cobj);
1588 if (mdd_object_exists(mdd_cobj) == 0)
1592 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1596 if (likely(mdd_cobj != NULL)) {
1598 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1602 is_dir = S_ISDIR(cattr->la_mode);
1605 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1609 handle = mdd_trans_create(env, mdd);
1611 RETURN(PTR_ERR(handle));
1613 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1614 lname, ma, handle, no_name);
1618 rc = mdd_trans_start(env, mdd, handle);
1622 if (likely(mdd_cobj != NULL))
1623 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1625 if (likely(no_name == 0)) {
1626 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1627 mdd_object_capa(env, mdd_pobj));
1632 if (likely(mdd_cobj != NULL)) {
1633 rc = mdo_ref_del(env, mdd_cobj, handle);
1635 __mdd_index_insert_only(env, mdd_pobj,
1637 mdd_object_type(mdd_cobj),
1639 mdd_object_capa(env, mdd_pobj));
1645 mdo_ref_del(env, mdd_cobj, handle);
1647 /* fetch updated nlink */
1648 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1653 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1654 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1656 la->la_valid = LA_CTIME | LA_MTIME;
1657 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1661 /* Enough for only unlink the entry */
1662 if (unlikely(mdd_cobj == NULL))
1665 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1666 /* update ctime of an unlinked file only if it is still
1667 * opened or a link still exists */
1668 la->la_valid = LA_CTIME;
1669 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1674 /* XXX: this transfer to ma will be removed with LOD/OSP */
1675 ma->ma_attr = *cattr;
1676 ma->ma_valid |= MA_INODE;
1677 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1679 /* fetch updated nlink */
1681 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1683 /* if object is removed then we can't get its attrs, use last get */
1684 if (cattr->la_nlink == 0) {
1685 ma->ma_attr = *cattr;
1686 ma->ma_valid |= MA_INODE;
1690 mdd_write_unlock(env, mdd_cobj);
1694 if (cattr->la_nlink == 0) {
1695 cl_flags |= CLF_UNLINK_LAST;
1696 /* search for an existing archive */
1697 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1698 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1701 rc = mdd_changelog_ns_store(env, mdd,
1702 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1703 mdd_cobj, mdd_pobj, lname, handle);
1707 mdd_trans_stop(env, mdd, rc, handle);
1713 * The permission has been checked when obj created, no need check again.
1715 static int mdd_cd_sanity_check(const struct lu_env *env,
1716 struct mdd_object *obj)
1721 if (!obj || mdd_is_dead_obj(obj))
1727 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1728 struct md_object *cobj, const struct md_op_spec *spec,
1731 struct mdd_device *mdd = mdo2mdd(cobj);
1732 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1733 struct mdd_object *son = md2mdd_obj(cobj);
1734 struct thandle *handle;
1735 const struct lu_buf *buf;
1736 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1737 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1741 rc = mdd_cd_sanity_check(env, son);
1745 if (!md_should_create(spec->sp_cr_flags))
1749 * there are following use cases for this function:
1750 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1751 * striping can be specified or not
1754 rc = mdd_la_get(env, son, attr, BYPASS_CAPA);
1758 /* calling ->ah_make_hint() is used to transfer information from parent */
1759 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1761 handle = mdd_trans_create(env, mdd);
1763 GOTO(out_free, rc = PTR_ERR(handle));
1766 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1767 * Should this be fixed?
1769 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1770 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1771 spec->sp_cr_flags, spec->no_create);
1773 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1774 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1775 spec->u.sp_ea.eadatalen);
1780 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1781 XATTR_NAME_LOV, 0, handle);
1785 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1789 rc = mdd_trans_start(env, mdd, handle);
1793 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1794 0, handle, mdd_object_capa(env, son));
1799 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1802 mdd_trans_stop(env, mdd, rc, handle);
1807 static int mdd_declare_object_initialize(const struct lu_env *env,
1808 struct mdd_object *parent,
1809 struct mdd_object *child,
1810 struct lu_attr *attr,
1811 struct thandle *handle)
1817 * inode mode has been set in creation time, and it's based on umask,
1818 * la_mode and acl, don't set here again! (which will go wrong
1819 * because below function doesn't consider umask).
1820 * I'd suggest set all object attributes in creation time, see above.
1822 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1823 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1824 rc = mdo_declare_attr_set(env, child, attr, handle);
1825 attr->la_valid |= LA_MODE | LA_TYPE;
1826 if (rc != 0 || !S_ISDIR(attr->la_mode))
1829 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1834 rc = mdo_declare_ref_add(env, child, handle);
1838 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1844 static int mdd_object_initialize(const struct lu_env *env,
1845 const struct lu_fid *pfid,
1846 struct mdd_object *child,
1847 struct lu_attr *attr, struct thandle *handle,
1848 const struct md_op_spec *spec)
1854 * Update attributes for child.
1857 * (1) the valid bits should be converted between Lustre and Linux;
1858 * (2) maybe, the child attributes should be set in OSD when creation.
1861 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1862 /* arguments are supposed to stay the same */
1863 if (S_ISDIR(attr->la_mode)) {
1864 /* Add "." and ".." for newly created dir */
1865 mdo_ref_add(env, child, handle);
1866 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1867 S_IFDIR, dot, handle, BYPASS_CAPA);
1869 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1873 mdo_ref_del(env, child, handle);
1880 * This function checks whether it can create a file/dir under the
1881 * directory(@pobj). The directory(@pobj) is not being locked by
1884 * \param[in] env execution environment
1885 * \param[in] pobj the directory to create files
1886 * \param[in] pattr the attributes of the directory
1887 * \param[in] lname the name of the created file/dir
1888 * \param[in] cattr the attributes of the file/dir
1889 * \param[in] spec create specification
1891 * \retval = 0 it is allowed to create file/dir under
1893 * \retval negative error not allowed to create file/dir
1894 * under the directory
1896 static int mdd_create_sanity_check(const struct lu_env *env,
1897 struct md_object *pobj,
1898 const struct lu_attr *pattr,
1899 const struct lu_name *lname,
1900 struct lu_attr *cattr,
1901 struct md_op_spec *spec)
1903 struct mdd_thread_info *info = mdd_env_info(env);
1904 struct lu_fid *fid = &info->mti_fid;
1905 struct mdd_object *obj = md2mdd_obj(pobj);
1906 struct mdd_device *m = mdo2mdd(pobj);
1907 bool check_perm = true;
1912 if (mdd_is_dead_obj(obj))
1916 * In some cases this lookup is not needed - we know before if name
1917 * exists or not because MDT performs lookup for it.
1918 * name length check is done in lookup.
1920 if (spec->sp_cr_lookup) {
1922 * Check if the name already exist, though it will be checked in
1923 * _index_insert also, for avoiding rolling back if exists
1926 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1927 MAY_WRITE | MAY_EXEC);
1929 RETURN(rc ? : -EEXIST);
1931 /* Permission is already being checked in mdd_lookup */
1935 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1940 if (pattr->la_mode & S_ISGID) {
1941 cattr->la_gid = pattr->la_gid;
1942 if (S_ISDIR(cattr->la_mode)) {
1943 cattr->la_mode |= S_ISGID;
1944 cattr->la_valid |= LA_MODE;
1948 rc = mdd_name_check(m, lname);
1952 switch (cattr->la_mode & S_IFMT) {
1954 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1956 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1957 RETURN(-ENAMETOOLONG);
1976 static int mdd_declare_object_create(const struct lu_env *env,
1977 struct mdd_device *mdd,
1978 struct mdd_object *p, struct mdd_object *c,
1979 struct lu_attr *attr,
1980 struct thandle *handle,
1981 const struct md_op_spec *spec,
1982 struct lu_buf *def_acl_buf,
1983 struct lu_buf *acl_buf,
1984 struct dt_allocation_hint *hint)
1988 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
1993 #ifdef CONFIG_FS_POSIX_ACL
1994 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1995 /* if dir, then can inherit default ACl */
1996 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1997 XATTR_NAME_ACL_DEFAULT,
2003 if (acl_buf->lb_len > 0) {
2004 rc = mdo_declare_attr_set(env, c, attr, handle);
2008 rc = mdo_declare_xattr_set(env, c, acl_buf,
2009 XATTR_NAME_ACL_ACCESS, 0, handle);
2014 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2018 /* replay case, create LOV EA from client data */
2019 if (spec->no_create ||
2020 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2021 const struct lu_buf *buf;
2023 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2024 spec->u.sp_ea.eadatalen);
2025 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2031 if (S_ISLNK(attr->la_mode)) {
2032 const char *target_name = spec->u.sp_symname;
2033 int sym_len = strlen(target_name);
2034 const struct lu_buf *buf;
2036 buf = mdd_buf_get_const(env, target_name, sym_len);
2037 rc = dt_declare_record_write(env, mdd_object_child(c),
2046 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2047 struct mdd_object *p, struct mdd_object *c,
2048 const struct lu_name *name,
2049 struct lu_attr *attr,
2050 struct thandle *handle,
2051 const struct md_op_spec *spec,
2052 struct linkea_data *ldata,
2053 struct lu_buf *def_acl_buf,
2054 struct lu_buf *acl_buf,
2055 struct dt_allocation_hint *hint)
2059 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2060 def_acl_buf, acl_buf, hint);
2064 if (S_ISDIR(attr->la_mode)) {
2065 rc = mdo_declare_ref_add(env, p, handle);
2070 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2071 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2075 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2077 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2078 name->ln_name, handle);
2082 rc = mdd_declare_links_add(env, c, handle, ldata);
2087 la->la_valid = LA_CTIME | LA_MTIME;
2088 rc = mdo_declare_attr_set(env, p, la, handle);
2092 rc = mdd_declare_changelog_store(env, mdd, name, handle);
2097 /* XXX: For remote create, it should indicate the remote RPC
2098 * will be sent after local transaction is finished, which
2099 * is not very nice, but it will be removed once we fully support
2101 if (mdd_object_remote(p) && handle->th_update != NULL)
2102 handle->th_update->tu_sent_after_local_trans = 1;
2107 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2108 struct lu_attr *la, struct lu_buf *def_acl_buf,
2109 struct lu_buf *acl_buf)
2114 if (S_ISLNK(la->la_mode)) {
2115 acl_buf->lb_len = 0;
2116 def_acl_buf->lb_len = 0;
2120 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2121 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2122 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
2123 mdd_read_unlock(env, pobj);
2125 /* If there are default ACL, fix mode/ACL by default ACL */
2126 def_acl_buf->lb_len = rc;
2127 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2128 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2129 acl_buf->lb_len = rc;
2130 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2133 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2134 /* If there are no default ACL, fix mode by mask */
2135 struct lu_ucred *uc = lu_ucred(env);
2137 /* The create triggered by MDT internal events, such as
2138 * LFSCK reset, will not contain valid "uc". */
2139 if (unlikely(uc != NULL))
2140 la->la_mode &= ~uc->uc_umask;
2142 acl_buf->lb_len = 0;
2143 def_acl_buf->lb_len = 0;
2150 * Create a metadata object and initialize it, set acl, xattr.
2152 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2153 struct mdd_object *son, struct lu_attr *attr,
2154 struct md_op_spec *spec, struct lu_buf *acl_buf,
2155 struct lu_buf *def_acl_buf,
2156 struct dt_allocation_hint *hint,
2157 struct thandle *handle)
2161 mdd_write_lock(env, son, MOR_TGT_CHILD);
2162 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2167 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2168 * created in declare phase, they also needs to be added to master
2169 * object as sub-directory entry. So it has to initialize the master
2170 * object, then set dir striped EA.(in mdo_xattr_set) */
2171 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2174 GOTO(err_destroy, rc);
2177 * in case of replay we just set LOVEA provided by the client
2178 * XXX: I think it would be interesting to try "old" way where
2179 * MDT calls this xattr_set(LOV) in a different transaction.
2180 * probably this way we code can be made better.
2183 /* During creation, there are only a few cases we need do xattr_set to
2185 * 1. regular file: see comments above.
2186 * 2. create striped directory with provided stripeEA.
2187 * 3. create striped directory because inherit default layout from the
2189 if (spec->no_create ||
2190 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2191 S_ISDIR(attr->la_mode)) {
2192 const struct lu_buf *buf;
2194 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2195 spec->u.sp_ea.eadatalen);
2196 rc = mdo_xattr_set(env, son, buf,
2197 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2199 handle, BYPASS_CAPA);
2201 GOTO(err_destroy, rc);
2204 #ifdef CONFIG_FS_POSIX_ACL
2205 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2206 S_ISDIR(attr->la_mode)) {
2207 /* set default acl */
2208 rc = mdo_xattr_set(env, son, def_acl_buf,
2209 XATTR_NAME_ACL_DEFAULT, 0,
2210 handle, BYPASS_CAPA);
2212 GOTO(err_destroy, rc);
2214 /* set its own acl */
2215 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2216 rc = mdo_xattr_set(env, son, acl_buf,
2217 XATTR_NAME_ACL_ACCESS,
2218 0, handle, BYPASS_CAPA);
2220 GOTO(err_destroy, rc);
2224 if (S_ISLNK(attr->la_mode)) {
2225 struct lu_ucred *uc = lu_ucred_assert(env);
2226 struct dt_object *dt = mdd_object_child(son);
2227 const char *target_name = spec->u.sp_symname;
2228 int sym_len = strlen(target_name);
2229 const struct lu_buf *buf;
2232 buf = mdd_buf_get_const(env, target_name, sym_len);
2233 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2234 mdd_object_capa(env, son),
2236 CFS_CAP_SYS_RESOURCE_MASK);
2241 GOTO(err_initlized, rc = -EFAULT);
2245 if (unlikely(rc != 0)) {
2247 if (S_ISDIR(attr->la_mode)) {
2248 /* Drop the reference, no need to delete "."/"..",
2249 * because the object to be destroied directly. */
2250 rc2 = mdo_ref_del(env, son, handle);
2254 rc2 = mdo_ref_del(env, son, handle);
2258 mdo_destroy(env, son, handle);
2261 mdd_write_unlock(env, son);
2266 * Create object and insert it into namespace.
2268 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2269 const struct lu_name *lname, struct md_object *child,
2270 struct md_op_spec *spec, struct md_attr* ma)
2272 struct mdd_thread_info *info = mdd_env_info(env);
2273 struct lu_attr *la = &info->mti_la_for_fix;
2274 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2275 struct mdd_object *son = md2mdd_obj(child);
2276 struct mdd_device *mdd = mdo2mdd(pobj);
2277 struct lu_attr *attr = &ma->ma_attr;
2278 struct thandle *handle;
2279 struct lu_attr *pattr = &info->mti_pattr;
2280 struct lu_buf acl_buf;
2281 struct lu_buf def_acl_buf;
2282 struct linkea_data *ldata = &info->mti_link_data;
2283 const char *name = lname->ln_name;
2284 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2290 * Two operations have to be performed:
2292 * - an allocation of a new object (->do_create()), and
2294 * - an insertion into a parent index (->dio_insert()).
2296 * Due to locking, operation order is not important, when both are
2297 * successful, *but* error handling cases are quite different:
2299 * - if insertion is done first, and following object creation fails,
2300 * insertion has to be rolled back, but this operation might fail
2301 * also leaving us with dangling index entry.
2303 * - if creation is done first, is has to be undone if insertion
2304 * fails, leaving us with leaked space, which is neither good, nor
2307 * It seems that creation-first is simplest solution, but it is
2308 * sub-optimal in the frequent
2313 * case, because second mkdir is bound to create object, only to
2314 * destroy it immediately.
2316 * To avoid this follow local file systems that do double lookup:
2318 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2320 * 1. create (mdd_object_create_internal())
2322 * 2. insert (__mdd_index_insert(), lookup again)
2325 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2329 /* Sanity checks before big job. */
2330 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2334 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2335 GOTO(out_free, rc = -EINPROGRESS);
2337 handle = mdd_trans_create(env, mdd);
2339 GOTO(out_free, rc = PTR_ERR(handle));
2341 acl_buf.lb_buf = info->mti_xattr_buf;
2342 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2343 def_acl_buf.lb_buf = info->mti_key;
2344 def_acl_buf.lb_len = sizeof(info->mti_key);
2345 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2349 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2351 memset(ldata, 0, sizeof(*ldata));
2352 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2353 mdd_object_fid(mdd_pobj),
2354 lname, 1, 0, ldata);
2356 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2357 handle, spec, ldata, &def_acl_buf, &acl_buf,
2362 rc = mdd_trans_start(env, mdd, handle);
2366 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2367 &def_acl_buf, hint, handle);
2371 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2372 mdd_write_lock(env, son, MOR_TGT_CHILD);
2373 rc = __mdd_orphan_add(env, son, handle);
2374 GOTO(out_volatile, rc);
2376 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2377 attr->la_mode, name, handle,
2378 mdd_object_capa(env, mdd_pobj));
2380 GOTO(err_created, rc);
2382 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2385 /* update parent directory mtime/ctime */
2387 la->la_valid = LA_CTIME | LA_MTIME;
2388 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2390 GOTO(err_insert, rc);
2398 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2399 rc2 = __mdd_orphan_del(env, son, handle);
2401 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2402 S_ISDIR(attr->la_mode),
2403 handle, BYPASS_CAPA);
2408 mdd_write_lock(env, son, MOR_TGT_CHILD);
2409 if (S_ISDIR(attr->la_mode)) {
2410 /* Drop the reference, no need to delete "."/"..",
2411 * because the object is to be destroyed directly. */
2412 rc2 = mdo_ref_del(env, son, handle);
2414 mdd_write_unlock(env, son);
2419 /* For volatile files drop one link immediately, since there is
2420 * no filename in the namespace, and save any error returned. */
2421 rc2 = mdo_ref_del(env, son, handle);
2423 mdd_write_unlock(env, son);
2424 if (unlikely(rc == 0))
2429 /* Don't destroy the volatile object on success */
2430 if (likely(rc != 0))
2431 mdo_destroy(env, son, handle);
2432 mdd_write_unlock(env, son);
2435 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2436 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2437 rc = mdd_changelog_ns_store(env, mdd,
2438 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2439 S_ISREG(attr->la_mode) ? CL_CREATE :
2440 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2441 0, son, mdd_pobj, lname, handle);
2443 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2447 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2448 /* if we vmalloced a large buffer drop it */
2449 lu_buf_free(ldata->ld_buf);
2451 /* The child object shouldn't be cached anymore */
2453 set_bit(LU_OBJECT_HEARD_BANSHEE,
2454 &child->mo_lu.lo_header->loh_flags);
2459 * Get locks on parents in proper order
2460 * RETURN: < 0 - error, rename_order if successful
2468 static int mdd_rename_order(const struct lu_env *env,
2469 struct mdd_device *mdd,
2470 struct mdd_object *src_pobj,
2471 const struct lu_attr *pattr,
2472 struct mdd_object *tgt_pobj)
2474 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2478 if (src_pobj == tgt_pobj)
2479 RETURN(MDD_RN_SAME);
2481 /* compared the parent child relationship of src_p&tgt_p */
2482 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2484 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2487 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2501 /* has not mdd_write{read}_lock on any obj yet. */
2502 static int mdd_rename_sanity_check(const struct lu_env *env,
2503 struct mdd_object *src_pobj,
2504 const struct lu_attr *pattr,
2505 struct mdd_object *tgt_pobj,
2506 const struct lu_attr *tpattr,
2507 struct mdd_object *sobj,
2508 const struct lu_attr *cattr,
2509 struct mdd_object *tobj,
2510 const struct lu_attr *tattr)
2515 /* XXX: when get here, sobj must NOT be NULL,
2516 * the other case has been processed in cld_rename
2517 * before mdd_rename and enable MDS_PERM_BYPASS. */
2520 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2524 /* XXX: when get here, "tobj == NULL" means tobj must
2525 * NOT exist (neither on remote MDS, such case has been
2526 * processed in cld_rename before mdd_rename and enable
2528 * So check may_create, but not check may_unlink. */
2530 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2531 (src_pobj != tgt_pobj));
2533 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2534 (src_pobj != tgt_pobj), 1);
2536 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2537 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2542 static int mdd_declare_rename(const struct lu_env *env,
2543 struct mdd_device *mdd,
2544 struct mdd_object *mdd_spobj,
2545 struct mdd_object *mdd_tpobj,
2546 struct mdd_object *mdd_sobj,
2547 struct mdd_object *mdd_tobj,
2548 const struct lu_name *tname,
2549 const struct lu_name *sname,
2551 struct linkea_data *ldata,
2552 struct thandle *handle)
2554 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2557 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2558 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2564 /* name from source dir */
2565 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2569 /* .. from source child */
2570 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2571 /* source child can be directory,
2572 * counted by source dir's nlink */
2573 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2576 if (mdd_spobj != mdd_tpobj) {
2577 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2582 rc = mdo_declare_index_insert(env, mdd_sobj,
2584 S_IFDIR, dotdot, handle);
2589 /* new target child can be directory,
2590 * counted by target dir's nlink */
2591 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2596 la->la_valid = LA_CTIME | LA_MTIME;
2597 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2601 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2605 la->la_valid = LA_CTIME;
2606 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2610 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2615 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2616 mdd_object_type(mdd_sobj),
2617 tname->ln_name, handle);
2621 /* name from target dir (old name), we declare it unconditionally
2622 * as mdd_rename() calls delete unconditionally as well. so just
2623 * to balance declarations vs calls to change ... */
2624 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2628 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2629 /* delete target child in target parent directory */
2630 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2634 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2635 /* target child can be directory,
2636 * delete "." reference in target child directory */
2637 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2641 /* delete ".." reference in target parent directory */
2642 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2647 la->la_valid = LA_CTIME;
2648 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2652 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2657 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2664 /* src object can be remote that is why we use only fid and type of object */
2665 static int mdd_rename(const struct lu_env *env,
2666 struct md_object *src_pobj, struct md_object *tgt_pobj,
2667 const struct lu_fid *lf, const struct lu_name *lsname,
2668 struct md_object *tobj, const struct lu_name *ltname,
2671 const char *sname = lsname->ln_name;
2672 const char *tname = ltname->ln_name;
2673 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2674 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2675 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2676 struct mdd_device *mdd = mdo2mdd(src_pobj);
2677 struct mdd_object *mdd_sobj = NULL; /* source object */
2678 struct mdd_object *mdd_tobj = NULL;
2679 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2680 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2681 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2682 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2683 struct thandle *handle;
2684 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2685 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2686 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2689 bool tobj_locked = 0;
2690 unsigned cl_flags = 0;
2695 mdd_tobj = md2mdd_obj(tobj);
2697 mdd_sobj = mdd_object_find(env, mdd, lf);
2699 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
2701 GOTO(out_pending, rc);
2703 rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA);
2705 GOTO(out_pending, rc);
2708 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2710 GOTO(out_pending, rc);
2713 rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA);
2715 GOTO(out_pending, rc);
2717 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2718 mdd_sobj, cattr, mdd_tobj, tattr);
2720 GOTO(out_pending, rc);
2722 rc = mdd_name_check(mdd, ltname);
2724 GOTO(out_pending, rc);
2726 handle = mdd_trans_create(env, mdd);
2728 GOTO(out_pending, rc = PTR_ERR(handle));
2730 memset(ldata, 0, sizeof(*ldata));
2731 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2732 ltname, 1, 0, ldata);
2733 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2734 mdd_tobj, lsname, ltname, ma, ldata, handle);
2738 rc = mdd_trans_start(env, mdd, handle);
2742 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2743 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2745 GOTO(cleanup_unlocked, rc);
2747 is_dir = S_ISDIR(cattr->la_mode);
2749 /* Remove source name from source directory */
2750 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2751 mdd_object_capa(env, mdd_spobj));
2755 /* "mv dir1 dir2" needs "dir1/.." link update */
2756 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2757 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2758 mdd_object_capa(env, mdd_sobj));
2760 GOTO(fixup_spobj2, rc);
2762 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2764 mdd_object_capa(env, mdd_sobj));
2766 GOTO(fixup_spobj, rc);
2769 /* Remove target name from target directory
2770 * Here tobj can be remote one, so we do index_delete unconditionally
2771 * and -ENOENT is allowed.
2773 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2774 mdd_object_capa(env, mdd_tpobj));
2777 /* tname might been renamed to something else */
2778 GOTO(fixup_spobj, rc);
2781 GOTO(fixup_spobj, rc);
2784 /* Insert new fid with target name into target dir */
2785 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2786 tname, handle, mdd_object_capa(env, mdd_tpobj));
2788 GOTO(fixup_tpobj, rc);
2790 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2791 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2793 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2795 la->la_valid = LA_CTIME;
2796 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2798 GOTO(fixup_tpobj, rc);
2801 /* Remove old target object
2802 * For tobj is remote case cmm layer has processed
2803 * and set tobj to NULL then. So when tobj is NOT NULL,
2804 * it must be local one.
2806 if (tobj && mdd_object_exists(mdd_tobj)) {
2807 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2809 if (mdd_is_dead_obj(mdd_tobj)) {
2810 /* shld not be dead, something is wrong */
2811 CERROR("tobj is dead, something is wrong\n");
2815 mdo_ref_del(env, mdd_tobj, handle);
2817 /* Remove dot reference. */
2818 if (S_ISDIR(tattr->la_mode))
2819 mdo_ref_del(env, mdd_tobj, handle);
2822 /* fetch updated nlink */
2823 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2825 CERROR("%s: Failed to get nlink for tobj "
2827 mdd2obd_dev(mdd)->obd_name,
2828 PFID(tpobj_fid), rc);
2829 GOTO(fixup_tpobj, rc);
2832 la->la_valid = LA_CTIME;
2833 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2835 CERROR("%s: Failed to set ctime for tobj "
2837 mdd2obd_dev(mdd)->obd_name,
2838 PFID(tpobj_fid), rc);
2839 GOTO(fixup_tpobj, rc);
2842 /* XXX: this transfer to ma will be removed with LOD/OSP */
2843 ma->ma_attr = *tattr;
2844 ma->ma_valid |= MA_INODE;
2845 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2848 CERROR("%s: Failed to unlink tobj "
2850 mdd2obd_dev(mdd)->obd_name,
2851 PFID(tpobj_fid), rc);
2852 GOTO(fixup_tpobj, rc);
2855 /* fetch updated nlink */
2856 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2858 CERROR("%s: Failed to get nlink for tobj "
2860 mdd2obd_dev(mdd)->obd_name,
2861 PFID(tpobj_fid), rc);
2862 GOTO(fixup_tpobj, rc);
2864 /* XXX: this transfer to ma will be removed with LOD/OSP */
2865 ma->ma_attr = *tattr;
2866 ma->ma_valid |= MA_INODE;
2868 if (tattr->la_nlink == 0) {
2869 cl_flags |= CLF_RENAME_LAST;
2870 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2871 cl_flags |= CLF_RENAME_LAST_EXISTS;
2875 la->la_valid = LA_CTIME | LA_MTIME;
2876 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2878 GOTO(fixup_tpobj, rc);
2880 if (mdd_spobj != mdd_tpobj) {
2881 la->la_valid = LA_CTIME | LA_MTIME;
2882 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2885 if (rc == 0 && mdd_sobj) {
2886 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2887 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2888 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2891 /* Old files might not have EA entry */
2892 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2893 lsname, handle, NULL, 0);
2894 mdd_write_unlock(env, mdd_sobj);
2895 /* We don't fail the transaction if the link ea can't be
2896 updated -- fid2path will use alternate lookup method. */
2904 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2907 CWARN("tp obj fix error %d\n",rc2);
2909 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2910 !mdd_is_dead_obj(mdd_tobj)) {
2912 mdo_ref_add(env, mdd_tobj, handle);
2914 mdo_ref_add(env, mdd_tobj, handle);
2917 rc2 = __mdd_index_insert(env, mdd_tpobj,
2919 mdd_object_type(mdd_tobj),
2920 tname, handle, BYPASS_CAPA);
2922 CWARN("tp obj fix error: rc = %d\n", rc2);
2927 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2928 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2932 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2933 mdd2obd_dev(mdd)->obd_name, rc2);
2936 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
2937 dotdot, handle, BYPASS_CAPA);
2939 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2940 mdd2obd_dev(mdd)->obd_name, rc2);
2945 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
2946 mdd_object_type(mdd_sobj), sname,
2947 handle, BYPASS_CAPA);
2949 CWARN("sp obj fix error: rc = %d\n", rc2);
2954 mdd_write_unlock(env, mdd_tobj);
2958 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2959 mdd_tobj, tpobj_fid, lf,
2960 spobj_fid, ltname, lsname,
2964 mdd_trans_stop(env, mdd, rc, handle);
2967 mdd_object_put(env, mdd_sobj);
2972 * During migration once the parent FID has been changed,
2973 * we need update the parent FID in linkea.
2975 static int mdd_linkea_update_child_internal(const struct lu_env *env,
2976 struct mdd_object *parent,
2977 struct mdd_object *child,
2978 const char *name, int namelen,
2979 struct thandle *handle,
2982 struct mdd_thread_info *info = mdd_env_info(env);
2983 struct linkea_data ldata = {0};
2984 struct lu_buf *buf = &info->mti_link_buf;
2990 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2991 if (buf->lb_buf == NULL)
2995 rc = mdd_links_read(env, child, &ldata);
2997 if (rc == -ENOENT || rc == -ENODATA)
3002 LASSERT(ldata.ld_leh != NULL);
3003 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3004 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3005 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3006 struct lu_name lname;
3009 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3012 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3013 lu_fid_eq(&fid, mdd_object_fid(parent))) {
3014 ldata.ld_lee = (struct link_ea_entry *)
3015 ((char *)ldata.ld_lee +
3020 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3021 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3022 lname.ln_namelen, lname.ln_name,
3023 PFID(mdd_object_fid(parent)));
3024 /* update to the new parent fid */
3025 linkea_entry_pack(ldata.ld_lee, &lname,
3026 mdd_object_fid(parent));
3028 rc = mdd_declare_links_add(env, child, handle, &ldata);
3030 rc = mdd_links_write(env, child, &ldata, handle);
3036 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3037 struct mdd_object *parent,
3038 struct mdd_object *child,
3039 const char *name, int namelen,
3040 struct thandle *handle)
3042 return mdd_linkea_update_child_internal(env, parent, child, name,
3043 namelen, handle, true);
3046 static int mdd_linkea_update_child(const struct lu_env *env,
3047 struct mdd_object *parent,
3048 struct mdd_object *child,
3049 const char *name, int namelen,
3050 struct thandle *handle)
3052 return mdd_linkea_update_child_internal(env, parent, child, name,
3053 namelen, handle, false);
3056 static int mdd_update_linkea_internal(const struct lu_env *env,
3057 struct mdd_object *mdd_pobj,
3058 struct mdd_object *mdd_sobj,
3059 struct mdd_object *mdd_tobj,
3060 const struct lu_name *child_name,
3061 struct thandle *handle,
3064 struct mdd_thread_info *info = mdd_env_info(env);
3065 struct linkea_data *ldata = &info->mti_link_data;
3070 rc = mdd_links_read(env, mdd_sobj, ldata);
3072 if (rc == -ENOENT || rc == -ENODATA)
3078 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3080 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3085 /* If it is mulitple links file, we need update the name entry for
3087 LASSERT(ldata->ld_leh != NULL);
3088 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3089 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3090 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3091 struct mdd_object *pobj;
3092 struct lu_name lname;
3095 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3097 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3099 pobj = mdd_object_find(env, mdd, &fid);
3101 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3102 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3107 if (!mdd_object_exists(pobj)) {
3108 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3109 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3113 if (pobj == mdd_pobj &&
3114 lname.ln_namelen == child_name->ln_namelen &&
3115 strncmp(lname.ln_name, child_name->ln_name,
3116 lname.ln_namelen) == 0) {
3117 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3118 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3123 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3124 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3125 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3128 /* Remove source name from source directory */
3129 /* Insert new fid with target name into target dir */
3130 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3135 rc = mdo_declare_index_insert(env, pobj,
3136 mdd_object_fid(mdd_tobj),
3137 mdd_object_type(mdd_tobj),
3138 lname.ln_name, handle);
3142 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3146 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3150 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3152 mdd_object_capa(env, pobj));
3156 rc = __mdd_index_insert(env, pobj,
3157 mdd_object_fid(mdd_tobj),
3158 mdd_object_type(mdd_tobj),
3159 lname.ln_name, handle,
3160 mdd_object_capa(env, pobj));
3164 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3165 rc = mdo_ref_add(env, mdd_tobj, handle);
3166 mdd_write_unlock(env, mdd_tobj);
3170 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3171 mdo_ref_del(env, mdd_sobj, handle);
3172 mdd_write_unlock(env, mdd_sobj);
3175 mdd_object_put(env, pobj);
3183 static int mdd_migrate_xattrs(const struct lu_env *env,
3184 struct mdd_object *mdd_sobj,
3185 struct mdd_object *mdd_tobj)
3187 struct mdd_thread_info *info = mdd_env_info(env);
3188 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3190 struct thandle *handle;
3196 struct lu_buf list_xbuf;
3199 /* retrieve xattr list from the old object */
3200 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL,
3201 mdd_object_capa(env, mdd_sobj));
3202 if (list_xsize == -ENODATA)
3208 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3209 if (info->mti_big_buf.lb_buf == NULL)
3212 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3213 list_xbuf.lb_len = list_xsize;
3214 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf,
3215 mdd_object_capa(env, mdd_sobj));
3220 xname = list_xbuf.lb_buf;
3222 xlen = strnlen(xname, rem - 1) + 1;
3223 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3224 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3225 strcmp(XATTR_NAME_LMV, xname) == 0)
3228 /* For directory, if there are default layout, migrate here */
3229 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3230 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3233 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL,
3235 mdd_object_capa(env, mdd_sobj));
3236 if (xsize == -ENODATA)
3241 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3242 if (info->mti_link_buf.lb_buf == NULL)
3243 GOTO(out, rc = -ENOMEM);
3245 xbuf.lb_len = xsize;
3246 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3247 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname,
3248 mdd_object_capa(env, mdd_sobj));
3254 handle = mdd_trans_create(env, mdd);
3256 GOTO(out, rc = PTR_ERR(handle));
3258 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3261 GOTO(stop_trans, rc);
3262 /* Note: this transaction is part of migration, and it is not
3263 * the last step of migration, so we set th_local = 1 to avoid
3264 * update last rcvd for this transaction */
3265 handle->th_local = 1;
3266 rc = mdd_trans_start(env, mdd, handle);
3268 GOTO(stop_trans, rc);
3270 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle,
3271 mdd_object_capa(env, mdd_sobj));
3273 GOTO(stop_trans, rc = 0);
3276 GOTO(stop_trans, rc);
3278 mdd_trans_stop(env, mdd, rc, handle);
3283 memmove(xname, xname + xlen, rem);
3289 static int mdd_declare_migrate_create(const struct lu_env *env,
3290 struct mdd_object *mdd_pobj,
3291 struct mdd_object *mdd_sobj,
3292 struct mdd_object *mdd_tobj,
3293 struct md_op_spec *spec,
3295 union lmv_mds_md *mgr_ea,
3296 struct thandle *handle)
3298 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3299 const struct lu_buf *buf;
3303 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3304 handle, spec, NULL);
3308 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3313 if (S_ISLNK(la->la_mode)) {
3314 const char *target_name = spec->u.sp_symname;
3315 int sym_len = strlen(target_name);
3316 const struct lu_buf *buf;
3318 buf = mdd_buf_get_const(env, target_name, sym_len);
3319 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3325 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3326 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3327 spec->u.sp_ea.eadatalen);
3328 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3334 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3335 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3336 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3341 la_flag->la_valid = LA_FLAGS;
3342 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3343 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3344 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3349 static int mdd_migrate_create(const struct lu_env *env,
3350 struct mdd_object *mdd_pobj,
3351 struct mdd_object *mdd_sobj,
3352 struct mdd_object *mdd_tobj,
3355 struct mdd_thread_info *info = mdd_env_info(env);
3356 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3357 struct md_op_spec *spec = &info->mti_spec;
3358 struct lu_buf lmm_buf = { 0 };
3359 struct lu_buf link_buf = { 0 };
3360 const struct lu_buf *buf;
3361 struct thandle *handle;
3362 struct lmv_mds_md_v1 *mgr_ea;
3363 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3364 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3369 /* prepare spec for create */
3370 memset(spec, 0, sizeof(*spec));
3371 spec->sp_cr_lookup = 0;
3372 spec->sp_feat = &dt_directory_features;
3373 if (S_ISLNK(la->la_mode)) {
3374 buf = lu_buf_check_and_alloc(
3375 &mdd_env_info(env)->mti_big_buf,
3378 link_buf.lb_len = la->la_size + 1;
3379 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3381 rc = rc != 0 ? rc : -EFAULT;
3382 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3383 mdd2obd_dev(mdd)->obd_name,
3384 PFID(mdd_object_fid(mdd_sobj)), rc);
3387 spec->u.sp_symname = link_buf.lb_buf;
3388 } else if S_ISREG(la->la_mode) {
3389 /* retrieve lov of the old object */
3390 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3391 if (rc != 0 && rc != -ENODATA)
3393 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3394 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3395 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3396 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3400 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3401 memset(mgr_ea, 0, sizeof(*mgr_ea));
3402 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3403 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3404 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3405 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3406 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3407 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3409 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3411 handle = mdd_trans_create(env, mdd);
3413 GOTO(out_free, rc = PTR_ERR(handle));
3415 /* Note: this transaction is part of migration, and it is not
3416 * the last step of migration, so we set th_local = 1 to avoid
3417 * update last rcvd for this transaction */
3418 handle->th_local = 1;
3419 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3421 (union lmv_mds_md *)info->mti_xattr_buf,
3424 GOTO(stop_trans, rc);
3426 rc = mdd_trans_start(env, mdd, handle);
3428 GOTO(stop_trans, rc);
3430 /* create the target object */
3431 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3434 GOTO(stop_trans, rc);
3436 /* Set MIGRATE EA on the source inode, so once the migration needs
3437 * to be re-done during failover, the re-do process can locate the
3438 * target object which is already being created. */
3439 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3440 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3441 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0,
3442 handle, mdd_object_capa(env, mdd_sobj));
3444 GOTO(stop_trans, rc);
3446 /* Set immutable flag, so any modification is disabled until
3447 * the migration is done. Once the migration is interrupted,
3448 * if the resume process find the migrating object has both
3449 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3450 * flag and approve the migration */
3451 la_flag->la_valid = LA_FLAGS;
3452 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3453 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3454 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3455 mdd_object_capa(env, mdd_sobj));
3458 mdd_trans_stop(env, mdd, rc, handle);
3460 if (lmm_buf.lb_buf != NULL)
3461 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3465 static int mdd_migrate_entries(const struct lu_env *env,
3466 struct mdd_object *mdd_sobj,
3467 struct mdd_object *mdd_tobj)
3469 struct dt_object *next = mdd_object_child(mdd_sobj);
3470 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3471 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3472 struct thandle *handle;
3474 const struct dt_it_ops *iops;
3477 struct lu_dirent *ent;
3480 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3484 if (!dt_try_as_dir(env, next))
3485 GOTO(out_ent, rc = -ENOTDIR);
3487 * iterate directories
3489 iops = &next->do_index_ops->dio_it;
3490 it = iops->init(env, next, LUDA_FID | LUDA_TYPE,
3491 mdd_object_capa(env, mdd_sobj));
3493 GOTO(out_ent, rc = PTR_ERR(it));
3495 rc = iops->load(env, it, 0);
3497 rc = iops->next(env, it);
3501 * At this point and across for-loop:
3503 * rc == 0 -> ok, proceed.
3504 * rc > 0 -> end of directory.
3508 struct mdd_object *child;
3509 char *name = mdd_env_info(env)->mti_key;
3513 bool target_exist = false;
3515 len = iops->key_size(env, it);
3519 result = iops->rec(env, it, (struct dt_rec *)ent,
3520 LUDA_FID | LUDA_TYPE);
3521 if (result == -ESTALE)
3528 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3529 recsize = le16_to_cpu(ent->lde_reclen);
3531 /* Insert new fid with target name into target dir */
3532 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3533 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3534 ent->lde_name[1] == '.'))
3537 child = mdd_object_find(env, mdd, &ent->lde_fid);
3539 GOTO(out, rc = PTR_ERR(child));
3541 is_dir = S_ISDIR(mdd_object_type(child));
3543 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3545 /* Check whether the name has been inserted to the target */
3546 if (dt_try_as_dir(env, dt_tobj)) {
3547 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3549 rc = dt_tobj->do_index_ops->dio_lookup(env, dt_tobj,
3550 (struct dt_rec *)fid,
3551 (struct dt_key *)name,
3552 mdd_object_capa(env, mdd_tobj));
3553 if (unlikely(rc == 0))
3554 target_exist = true;
3557 handle = mdd_trans_create(env, mdd);
3559 GOTO(out, rc = PTR_ERR(handle));
3561 /* Note: this transaction is part of migration, and it is not
3562 * the last step of migration, so we set th_local = 1 to avoid
3563 * updating last rcvd for this transaction */
3564 handle->th_local = 1;
3565 if (likely(!target_exist)) {
3566 rc = mdo_declare_index_insert(env, mdd_tobj,
3568 mdd_object_type(child),
3574 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3580 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3585 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3589 /* Update .. for child */
3590 rc = mdo_declare_index_delete(env, child, dotdot,
3595 rc = mdo_declare_index_insert(env, child,
3596 mdd_object_fid(mdd_tobj),
3597 S_IFDIR, dotdot, handle);
3602 rc = mdd_linkea_declare_update_child(env, mdd_tobj,
3609 rc = mdd_trans_start(env, mdd, handle);
3611 CERROR("%s: transaction start failed: rc = %d\n",
3612 mdd2obd_dev(mdd)->obd_name, rc);
3616 if (likely(!target_exist)) {
3617 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3618 mdd_object_type(child),
3620 mdd_object_capa(env, mdd_tobj));
3625 rc = mdo_ref_add(env, mdd_tobj, handle);
3632 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle,
3633 mdd_object_capa(env, mdd_sobj));
3638 rc = __mdd_index_delete_only(env, child, dotdot, handle,
3639 mdd_object_capa(env, child));
3643 rc = __mdd_index_insert_only(env, child,
3644 mdd_object_fid(mdd_tobj), S_IFDIR,
3646 mdd_object_capa(env, child));
3651 rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
3652 strlen(name), handle);
3655 mdd_object_put(env, child);
3656 mdd_trans_stop(env, mdd, rc, handle);
3660 result = iops->next(env, it);
3661 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3662 GOTO(out, rc = -EINTR);
3664 if (result == -ESTALE)
3666 } while (result == 0);
3669 iops->fini(env, it);
3671 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3675 static int mdd_declare_update_linkea(const struct lu_env *env,
3676 struct mdd_object *mdd_pobj,
3677 struct mdd_object *mdd_sobj,
3678 struct mdd_object *mdd_tobj,
3679 struct thandle *handle,
3680 const struct lu_name *child_name)
3682 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3683 child_name, handle, 1);
3686 static int mdd_update_linkea(const struct lu_env *env,
3687 struct mdd_object *mdd_pobj,
3688 struct mdd_object *mdd_sobj,
3689 struct mdd_object *mdd_tobj,
3690 struct thandle *handle,
3691 const struct lu_name *child_name)
3693 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3694 child_name, handle, 0);
3697 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3698 struct mdd_object *mdd_pobj,
3699 struct mdd_object *mdd_sobj,
3700 struct mdd_object *mdd_tobj,
3701 const struct lu_name *lname,
3703 struct lu_attr *parent_la,
3704 struct thandle *handle)
3706 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3709 /* Revert IMMUTABLE flag */
3710 la_flag->la_valid = LA_FLAGS;
3711 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3712 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3713 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3717 /* delete entry from source dir */
3718 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3722 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3723 mdd_tobj, handle, lname);
3727 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3728 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3734 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3735 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3741 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3742 mdd_object_type(mdd_tobj),
3743 lname->ln_name, handle);
3747 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3748 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3753 /* delete old object */
3754 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3758 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3759 /* delete old object */
3760 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3763 /* set nlink to 0 */
3764 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3769 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3773 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3778 static int mdd_migrate_update_name(const struct lu_env *env,
3779 struct mdd_object *mdd_pobj,
3780 struct mdd_object *mdd_sobj,
3781 struct mdd_object *mdd_tobj,
3782 const struct lu_name *lname,
3785 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3786 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3787 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3788 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3789 struct thandle *handle;
3790 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3791 const char *name = lname->ln_name;
3795 /* update time for parent */
3796 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3797 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3798 p_la->la_valid = LA_CTIME;
3800 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3804 handle = mdd_trans_create(env, mdd);
3806 RETURN(PTR_ERR(handle));
3808 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3809 lname, so_attr, p_la, handle);
3811 /* If the migration can not be fit in one transaction, just
3812 * leave it in the original MDT */
3814 GOTO(stop_trans, rc = 0);
3816 GOTO(stop_trans, rc);
3819 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3820 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3821 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3822 PFID(mdd_object_fid(mdd_tobj)));
3824 rc = mdd_trans_start(env, mdd, handle);
3826 GOTO(stop_trans, rc);
3828 /* Revert IMMUTABLE flag */
3829 la_flag->la_valid = LA_FLAGS;
3830 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3831 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3832 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3833 mdd_object_capa(env, mdd_pobj));
3835 GOTO(stop_trans, rc);
3837 /* Remove source name from source directory */
3838 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
3839 mdd_object_capa(env, mdd_pobj));
3841 GOTO(stop_trans, rc);
3843 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3846 GOTO(stop_trans, rc);
3848 if (S_ISREG(so_attr->la_mode)) {
3849 if (so_attr->la_nlink == 1) {
3850 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3852 mdd_object_capa(env, mdd_sobj));
3853 if (rc != 0 && rc != -ENODATA)
3854 GOTO(stop_trans, rc);
3858 /* Insert new fid with target name into target dir */
3859 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
3860 mdd_object_type(mdd_tobj), name,
3861 handle, mdd_object_capa(env, mdd_pobj));
3863 GOTO(stop_trans, rc);
3865 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3868 GOTO(stop_trans, rc);
3870 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3871 mdo_ref_del(env, mdd_sobj, handle);
3873 mdo_ref_del(env, mdd_sobj, handle);
3875 /* Get the attr again after ref_del */
3876 rc = mdd_la_get(env, mdd_sobj, so_attr,
3877 mdd_object_capa(env, mdd_sobj));
3879 GOTO(stop_trans, rc);
3880 ma->ma_attr = *so_attr;
3881 ma->ma_valid |= MA_INODE;
3882 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3884 GOTO(stop_trans, rc);
3886 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3888 GOTO(stop_trans, rc);
3890 mdd_write_unlock(env, mdd_sobj);
3893 mdd_trans_stop(env, mdd, rc, handle);
3899 * Check whether we should migrate the file/dir
3901 * < 0 permission check failed or other error.
3902 * = 0 the file can be migrated.
3903 * > 0 the file does not need to be migrated, mostly
3904 * for multiple link file
3906 static int mdd_migrate_sanity_check(const struct lu_env *env,
3907 struct mdd_object *pobj,
3908 const struct lu_attr *pattr,
3909 struct mdd_object *sobj,
3910 struct lu_attr *sattr)
3912 struct mdd_thread_info *info = mdd_env_info(env);
3913 struct linkea_data *ldata = &info->mti_link_data;
3915 struct lu_buf *mgr_buf;
3921 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3922 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3923 if (mgr_buf->lb_buf == NULL)
3926 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV,
3927 mdd_object_capa(env, sobj));
3929 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3931 /* If the object has migrateEA, it means IMMUTE flag
3932 * is being set by previous migration process, so it
3933 * needs to override the IMMUTE flag, otherwise the
3934 * following sanity check will fail */
3935 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
3936 LMV_HASH_FLAG_MIGRATION) {
3937 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3939 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3940 sobj->mod_flags &= ~IMMUTE_OBJ;
3941 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3942 mdd2obd_dev(mdd)->obd_name,
3943 PFID(mdd_object_fid(sobj)));
3947 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3948 sobj, sattr, NULL, NULL);
3952 /* Then it will check if the file should be migrated. If the file
3953 * has mulitple links, we only need migrate the file if all of its
3954 * entries has been migrated to the remote MDT */
3955 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3958 rc = mdd_links_read(env, sobj, ldata);
3960 /* For multiple links files, if there are no linkEA data at all,
3961 * means the file might be created before linkEA is enabled, and
3962 * all all of its links should not be migrated yet, otherwise
3963 * it should have some linkEA there */
3964 if (rc == -ENOENT || rc == -ENODATA)
3969 /* If it is mulitple links file, we need update the name entry for
3971 LASSERT(ldata->ld_leh != NULL);
3972 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3973 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3974 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3975 struct mdd_object *lpobj;
3976 struct lu_name lname;
3979 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3981 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3983 lpobj = mdd_object_find(env, mdd, &fid);
3984 if (IS_ERR(lpobj)) {
3985 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3986 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3991 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
3992 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
3993 PFID(&fid), lname.ln_namelen, lname.ln_name);
3994 mdd_object_put(env, lpobj);
3998 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
3999 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4000 lname.ln_name, PFID(&fid));
4001 mdd_object_put(env, lpobj);
4009 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4010 struct md_object *sobj, const struct lu_name *lname,
4011 struct md_object *tobj, struct md_attr *ma)
4013 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4014 struct mdd_device *mdd = mdo2mdd(pobj);
4015 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4016 struct mdd_object *mdd_tobj = NULL;
4017 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4018 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4022 /* If the file will being migrated, it will check whether
4023 * the file is being opened by someone else right now */
4024 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4025 if (mdd_sobj->mod_count >= 1) {
4026 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
4027 mdd2obd_dev(mdd)->obd_name,
4028 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4029 mdd_sobj->mod_count, -EBUSY);
4030 mdd_read_unlock(env, mdd_sobj);
4031 GOTO(put, rc = -EBUSY);
4033 mdd_read_unlock(env, mdd_sobj);
4035 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
4039 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
4043 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4050 /* Sigh, it is impossible to finish all of migration in a single
4051 * transaction, for example migrating big directory entries to the
4052 * new MDT, it needs insert all of name entries of children in the
4055 * So migration will be done in multiple steps and transactions.
4057 * 1. create an orphan object on the remote MDT in one transaction.
4058 * 2. migrate extend attributes to the new target file/directory.
4059 * 3. For directory, migrate the entries to the new MDT and update
4060 * linkEA of each children. Because we can not migrate all entries
4061 * in a single transaction, so the migrating directory will become
4062 * a striped directory during migration, so once the process is
4063 * interrupted, the directory is still accessible. (During lookup,
4064 * client will locate the name by searching both original and target
4066 * 4. Finally, update the name/FID to point to the new file/directory
4067 * in a separate transaction.
4070 /* step 1: Check whether the orphan object has been created, and create
4071 * orphan object on the remote MDT if needed */
4072 mdd_tobj = md2mdd_obj(tobj);
4073 if (!mdd_object_exists(mdd_tobj)) {
4074 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4080 /* step 2: migrate xattr */
4081 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4085 /* step 3: migrate name entries to the orphan object */
4086 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4087 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4090 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4091 OBD_FAIL_MDS_REINT_NET_REP)))
4095 /* step 4: update name entry to the new object */
4096 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4104 const struct md_dir_operations mdd_dir_ops = {
4105 .mdo_is_subdir = mdd_is_subdir,
4106 .mdo_lookup = mdd_lookup,
4107 .mdo_create = mdd_create,
4108 .mdo_rename = mdd_rename,
4109 .mdo_link = mdd_link,
4110 .mdo_unlink = mdd_unlink,
4111 .mdo_create_data = mdd_create_data,
4112 .mdo_migrate = mdd_migrate,