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)))
102 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key,
103 mdd_object_capa(env, mdd_obj));
110 int mdd_lookup(const struct lu_env *env,
111 struct md_object *pobj, const struct lu_name *lname,
112 struct lu_fid *fid, struct md_op_spec *spec)
114 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
118 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr, BYPASS_CAPA);
122 rc = __mdd_lookup(env, pobj, pattr, lname, fid, MAY_EXEC);
126 static inline int mdd_parent_fid(const struct lu_env *env,
127 struct mdd_object *obj,
128 const struct lu_attr *attr,
131 return __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
135 * For root fid use special function, which does not compare version component
136 * of fid. Version component is different for root fids on all MDTs.
138 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
140 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
141 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
145 * return 1: if lf is the fid of the ancestor of p1;
148 * return -EREMOTE: if remote object is found, in this
149 * case fid of remote object is saved to @pf;
151 * otherwise: values < 0, errors.
153 static int mdd_is_parent(const struct lu_env *env,
154 struct mdd_device *mdd,
155 struct mdd_object *p1,
156 const struct lu_attr *attr,
157 const struct lu_fid *lf,
160 struct mdd_object *parent = NULL;
165 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
166 pfid = &mdd_env_info(env)->mti_fid;
168 /* Check for root first. */
169 if (mdd_is_root(mdd, mdo2fid(p1)))
173 /* this is done recursively, bypass capa for each obj */
174 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
175 rc = mdd_parent_fid(env, p1, attr, pfid);
178 if (mdd_is_root(mdd, pfid))
180 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
182 if (lu_fid_eq(pfid, lf))
185 mdd_object_put(env, parent);
187 parent = mdd_object_find(env, mdd, pfid);
188 if (IS_ERR(parent)) {
189 GOTO(out, rc = PTR_ERR(parent));
190 } else if (mdd_object_remote(parent)) {
191 /*FIXME: Because of the restriction of rename in Phase I.
192 * If the parent is remote, we just assumed lf is not the
193 * parent of P1 for now */
200 if (parent && !IS_ERR(parent))
201 mdd_object_put(env, parent);
206 * No permission check is needed.
208 * returns 1: if fid is ancestor of @mo;
209 * returns 0: if fid is not a ancestor of @mo;
211 * returns EREMOTE if remote object is found, fid of remote object is saved to
214 * returns < 0: if error
216 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
217 const struct lu_fid *fid, struct lu_fid *sfid)
219 struct mdd_device *mdd = mdo2mdd(mo);
220 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
224 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
227 rc = mdd_la_get(env, md2mdd_obj(mo), attr, BYPASS_CAPA);
231 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
235 } else if (rc == 1) {
236 /* found @fid is parent */
244 * Check that @dir contains no entries except (possibly) dot and dotdot.
249 * -ENOTDIR not a directory object
250 * -ENOTEMPTY not empty
254 static int mdd_dir_is_empty(const struct lu_env *env,
255 struct mdd_object *dir)
258 struct dt_object *obj;
259 const struct dt_it_ops *iops;
263 obj = mdd_object_child(dir);
264 if (!dt_try_as_dir(env, obj))
267 iops = &obj->do_index_ops->dio_it;
268 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
270 result = iops->get(env, it, (const struct dt_key *)"");
273 for (result = 0, i = 0; result == 0 && i < 3; ++i)
274 result = iops->next(env, it);
277 else if (result == 1)
279 } else if (result == 0)
281 * Huh? Index contains no zero key?
288 result = PTR_ERR(it);
293 * Determine if the target object can be hard linked, and right now it only
294 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
295 * directory nlink count might exceed the maximum link count(see
296 * osd_object_ref_add), so it only check nlink for non-directories.
298 * \param[in] env thread environment
299 * \param[in] obj object being linked to
300 * \param[in] la attributes of \a obj
302 * \retval 0 if \a obj can be hard linked
303 * \retval negative error if \a obj is a directory or has too
306 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
307 const struct lu_attr *la)
309 struct mdd_device *m = mdd_obj2mdd_dev(obj);
314 /* Subdir count limitation can be broken through
315 * (see osd_object_ref_add), so only check non-directory here. */
316 if (!S_ISDIR(la->la_mode) &&
317 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
324 * Check whether it may create the cobj under the pobj.
326 * \param[in] env execution environment
327 * \param[in] pobj the parent directory
328 * \param[in] pattr the attribute of the parent directory
329 * \param[in] cobj the child to be created
330 * \param[in] check_perm if check WRITE|EXEC permission for parent
332 * \retval = 0 create the child under this dir is allowed
333 * \retval negative errno create the child under this dir is
336 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
337 const struct lu_attr *pattr, struct mdd_object *cobj,
340 struct mdd_thread_info *info = mdd_env_info(env);
345 if (cobj && mdd_object_exists(cobj))
348 if (mdd_is_dead_obj(pobj))
351 /* If the parent is a sub-stripe, check whether it is dead */
352 xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key));
353 rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV,
354 mdd_object_capa(env, pobj));
355 if (unlikely(rc > 0)) {
356 struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf;
358 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE &&
359 le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD)
365 rc = mdd_permission_internal_locked(env, pobj, pattr,
366 MAY_WRITE | MAY_EXEC,
372 * Check whether can unlink from the pobj in the case of "cobj == NULL".
374 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
375 const struct lu_attr *pattr, const struct lu_attr *attr)
380 if (mdd_is_dead_obj(pobj))
383 if ((attr->la_valid & LA_FLAGS) &&
384 (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
387 rc = mdd_permission_internal_locked(env, pobj, pattr,
388 MAY_WRITE | MAY_EXEC,
393 if (mdd_is_append(pobj))
400 * pobj == NULL is remote ops case, under such case, pobj's
401 * VTX feature has been checked already, no need check again.
403 static inline int mdd_is_sticky(const struct lu_env *env,
404 struct mdd_object *pobj,
405 const struct lu_attr *pattr,
406 struct mdd_object *cobj,
407 const struct lu_attr *cattr)
409 struct lu_ucred *uc = lu_ucred_assert(env);
412 LASSERT(pattr != NULL);
413 if (!(pattr->la_mode & S_ISVTX) ||
414 (pattr->la_uid == uc->uc_fsuid))
418 LASSERT(cattr != NULL);
419 if (cattr->la_uid == uc->uc_fsuid)
422 return !md_capable(uc, CFS_CAP_FOWNER);
425 static int mdd_may_delete_entry(const struct lu_env *env,
426 struct mdd_object *pobj,
427 const struct lu_attr *pattr,
432 LASSERT(pobj != NULL);
433 if (!mdd_object_exists(pobj))
436 if (mdd_is_dead_obj(pobj))
441 rc = mdd_permission_internal_locked(env, pobj, pattr,
442 MAY_WRITE | MAY_EXEC,
448 if (mdd_is_append(pobj))
455 * Check whether it may delete the cobj from the pobj.
458 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
459 const struct lu_attr *tpattr, struct mdd_object *tobj,
460 const struct lu_attr *tattr, const struct lu_attr *cattr,
461 int check_perm, int check_empty)
467 LASSERT(tpattr != NULL);
468 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
476 if (!mdd_object_exists(tobj))
479 if (mdd_is_dead_obj(tobj))
482 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
485 if (mdd_is_immutable(tobj) || mdd_is_append(tobj))
488 if ((tattr->la_valid & LA_FLAGS) &&
489 (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
492 /* additional check the rename case */
494 if (S_ISDIR(cattr->la_mode)) {
495 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
497 if (!S_ISDIR(tattr->la_mode))
500 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
502 } else if (S_ISDIR(tattr->la_mode))
506 if (S_ISDIR(tattr->la_mode) && check_empty)
507 rc = mdd_dir_is_empty(env, tobj);
513 * Check whether it can create the link file(linked to @src_obj) under
514 * the target directory(@tgt_obj), and src_obj has been locked by
517 * \param[in] env execution environment
518 * \param[in] tgt_obj the target directory
519 * \param[in] tattr attributes of target directory
520 * \param[in] lname the link name
521 * \param[in] src_obj source object for link
522 * \param[in] cattr attributes for source object
524 * \retval = 0 it is allowed to create the link file under tgt_obj
525 * \retval negative error not allowed to create the link file
527 static int mdd_link_sanity_check(const struct lu_env *env,
528 struct mdd_object *tgt_obj,
529 const struct lu_attr *tattr,
530 const struct lu_name *lname,
531 struct mdd_object *src_obj,
532 const struct lu_attr *cattr)
534 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
538 if (!mdd_object_exists(src_obj))
541 if (mdd_is_dead_obj(src_obj))
544 /* Local ops, no lookup before link, check filename length here. */
545 rc = mdd_name_check(m, lname);
549 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
552 if (S_ISDIR(mdd_object_type(src_obj)))
555 LASSERT(src_obj != tgt_obj);
556 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
560 rc = __mdd_may_link(env, src_obj, cattr);
565 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
566 const char *name, struct thandle *handle,
567 struct lustre_capa *capa)
569 struct dt_object *next = mdd_object_child(pobj);
573 if (dt_try_as_dir(env, next))
574 rc = dt_delete(env, next, (struct dt_key *)name, handle, capa);
581 static int __mdd_index_insert_only(const struct lu_env *env,
582 struct mdd_object *pobj,
583 const struct lu_fid *lf, __u32 type,
585 struct thandle *handle,
586 struct lustre_capa *capa)
588 struct dt_object *next = mdd_object_child(pobj);
592 if (dt_try_as_dir(env, next)) {
593 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
594 struct lu_ucred *uc = lu_ucred_check(env);
598 rec->rec_type = type;
599 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
600 rc = dt_insert(env, next, (const struct dt_rec *)rec,
601 (const struct dt_key *)name, handle, capa,
609 /* insert named index, add reference if isdir */
610 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
611 const struct lu_fid *lf, __u32 type,
612 const char *name, struct thandle *handle,
613 struct lustre_capa *capa)
618 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle, capa);
619 if (rc == 0 && S_ISDIR(type)) {
620 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
621 mdo_ref_add(env, pobj, handle);
622 mdd_write_unlock(env, pobj);
628 /* delete named index, drop reference if isdir */
629 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
630 const char *name, int is_dir, struct thandle *handle,
631 struct lustre_capa *capa)
636 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
637 if (rc == 0 && is_dir) {
638 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
639 mdo_ref_del(env, pobj, handle);
640 mdd_write_unlock(env, pobj);
646 static int mdd_llog_record_calc_size(const struct lu_env *env,
647 const struct lu_name *tname,
648 const struct lu_name *sname)
650 const struct lu_ucred *uc = lu_ucred(env);
651 enum changelog_rec_flags crf = 0;
652 size_t hdr_size = sizeof(struct llog_changelog_rec) -
653 sizeof(struct changelog_rec);
658 if (uc != NULL && uc->uc_jobid[0] != '\0')
661 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
662 (tname != NULL ? tname->ln_namelen : 0) +
663 (sname != NULL ? 1 + sname->ln_namelen : 0));
666 int mdd_declare_changelog_store(const struct lu_env *env,
667 struct mdd_device *mdd,
668 const struct lu_name *tname,
669 const struct lu_name *sname,
670 struct thandle *handle)
672 struct obd_device *obd = mdd2obd_dev(mdd);
673 struct llog_ctxt *ctxt;
674 struct llog_changelog_rec *rec;
680 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
683 reclen = mdd_llog_record_calc_size(env, tname, sname);
684 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
685 if (buf->lb_buf == NULL)
689 rec->cr_hdr.lrh_len = reclen;
690 rec->cr_hdr.lrh_type = CHANGELOG_REC;
692 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
696 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
702 /** Add a changelog entry \a rec to the changelog llog
705 * \param handle - currently ignored since llogs start their own transaction;
706 * this will hopefully be fixed in llog rewrite
709 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
710 struct llog_changelog_rec *rec, struct thandle *th)
712 struct obd_device *obd = mdd2obd_dev(mdd);
713 struct llog_ctxt *ctxt;
716 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
717 changelog_rec_varsize(&rec->cr));
719 /* llog_lvfs_write_rec sets the llog tail len */
720 rec->cr_hdr.lrh_type = CHANGELOG_REC;
721 rec->cr.cr_time = cl_time();
723 spin_lock(&mdd->mdd_cl.mc_lock);
724 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
725 * but as long as the MDD transactions are ordered correctly for e.g.
726 * rename conflicts, I don't think this should matter. */
727 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
728 spin_unlock(&mdd->mdd_cl.mc_lock);
730 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
734 /* nested journal transaction */
735 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
743 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
744 const struct lu_fid *sfid,
745 const struct lu_fid *spfid,
746 const struct lu_name *sname)
748 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
749 size_t extsize = sname->ln_namelen + 1;
751 LASSERT(sfid != NULL);
752 LASSERT(spfid != NULL);
753 LASSERT(sname != NULL);
755 rnm->cr_sfid = *sfid;
756 rnm->cr_spfid = *spfid;
758 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
759 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
760 rec->cr_namelen += extsize;
763 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
765 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
767 if (jobid == NULL || jobid[0] == '\0')
770 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
773 /** Store a namespace change changelog record
774 * If this fails, we must fail the whole transaction; we don't
775 * want the change to commit without the log entry.
776 * \param target - mdd_object of change
777 * \param tpfid - target parent dir/object fid
778 * \param sfid - source object fid
779 * \param spfid - source parent fid
780 * \param tname - target name string
781 * \param sname - source name string
782 * \param handle - transacion handle
784 int mdd_changelog_ns_store(const struct lu_env *env,
785 struct mdd_device *mdd,
786 enum changelog_rec_type type,
787 enum changelog_rec_flags crf,
788 struct mdd_object *target,
789 const struct lu_fid *tpfid,
790 const struct lu_fid *sfid,
791 const struct lu_fid *spfid,
792 const struct lu_name *tname,
793 const struct lu_name *sname,
794 struct thandle *handle)
796 const struct lu_ucred *uc = lu_ucred(env);
797 struct llog_changelog_rec *rec;
804 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
807 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
810 LASSERT(tpfid != NULL);
811 LASSERT(tname != NULL);
812 LASSERT(handle != NULL);
814 reclen = mdd_llog_record_calc_size(env, tname, sname);
815 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
816 if (buf->lb_buf == NULL)
820 crf = (crf & CLF_FLAGMASK);
822 if (uc->uc_jobid[0] != '\0')
830 rec->cr.cr_flags = crf;
831 rec->cr.cr_type = (__u32)type;
832 rec->cr.cr_pfid = *tpfid;
833 rec->cr.cr_namelen = tname->ln_namelen;
834 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
836 if (crf & CLF_RENAME)
837 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
840 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
842 if (likely(target != NULL)) {
843 rec->cr.cr_tfid = *mdo2fid(target);
844 target->mod_cltime = cfs_time_current_64();
846 fid_zero(&rec->cr.cr_tfid);
849 rc = mdd_changelog_store(env, mdd, rec, handle);
851 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
852 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
859 static int __mdd_links_add(const struct lu_env *env,
860 struct mdd_object *mdd_obj,
861 struct linkea_data *ldata,
862 const struct lu_name *lname,
863 const struct lu_fid *pfid,
864 int first, int check)
868 if (ldata->ld_leh == NULL) {
869 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
873 rc = linkea_data_new(ldata,
874 &mdd_env_info(env)->mti_link_buf);
881 rc = linkea_links_find(ldata, lname, pfid);
882 if (rc && rc != -ENOENT)
888 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
889 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
893 linkea_add_buf(ldata, lname, tfid);
896 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
897 linkea_add_buf(ldata, lname, pfid);
899 return linkea_add_buf(ldata, lname, pfid);
902 static int __mdd_links_del(const struct lu_env *env,
903 struct mdd_object *mdd_obj,
904 struct linkea_data *ldata,
905 const struct lu_name *lname,
906 const struct lu_fid *pfid)
910 if (ldata->ld_leh == NULL) {
911 rc = mdd_links_read(env, mdd_obj, ldata);
916 rc = linkea_links_find(ldata, lname, pfid);
920 linkea_del_buf(ldata, lname);
924 static int mdd_linkea_prepare(const struct lu_env *env,
925 struct mdd_object *mdd_obj,
926 const struct lu_fid *oldpfid,
927 const struct lu_name *oldlname,
928 const struct lu_fid *newpfid,
929 const struct lu_name *newlname,
930 int first, int check,
931 struct linkea_data *ldata)
937 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
940 LASSERT(oldpfid != NULL || newpfid != NULL);
942 if (mdd_obj->mod_flags & DEAD_OBJ) {
943 /* Prevent linkea to be updated which is NOT necessary. */
944 ldata->ld_reclen = 0;
945 /* No more links, don't bother */
949 if (oldpfid != NULL) {
950 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
953 (rc != -ENODATA && rc != -ENOENT))
955 /* No changes done. */
960 /* If renaming, add the new record */
961 if (newpfid != NULL) {
962 /* even if the add fails, we still delete the out-of-date
964 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
968 rc = rc != 0 ? rc : rc2;
973 int mdd_links_rename(const struct lu_env *env,
974 struct mdd_object *mdd_obj,
975 const struct lu_fid *oldpfid,
976 const struct lu_name *oldlname,
977 const struct lu_fid *newpfid,
978 const struct lu_name *newlname,
979 struct thandle *handle,
980 struct linkea_data *ldata,
981 int first, int check)
987 ldata = &mdd_env_info(env)->mti_link_data;
988 memset(ldata, 0, sizeof(*ldata));
989 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
990 newpfid, newlname, first, check,
996 if (ldata->ld_reclen != 0)
997 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1002 if (rc == -EOVERFLOW || rc == -ENOSPC)
1004 if (oldpfid == NULL)
1005 CDEBUG(error ? D_ERROR : D_OTHER,
1006 "link_ea add '%.*s' failed %d "DFID"\n",
1007 newlname->ln_namelen, newlname->ln_name,
1008 rc, PFID(mdd_object_fid(mdd_obj)));
1009 else if (newpfid == NULL)
1010 CDEBUG(error ? D_ERROR : D_OTHER,
1011 "link_ea del '%.*s' failed %d "DFID"\n",
1012 oldlname->ln_namelen, oldlname->ln_name,
1013 rc, PFID(mdd_object_fid(mdd_obj)));
1015 CDEBUG(error ? D_ERROR : D_OTHER,
1016 "link_ea rename '%.*s'->'%.*s' failed %d "
1018 oldlname->ln_namelen, oldlname->ln_name,
1019 newlname->ln_namelen, newlname->ln_name,
1020 rc, PFID(mdd_object_fid(mdd_obj)));
1023 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1024 /* if we vmalloced a large buffer drop it */
1025 lu_buf_free(ldata->ld_buf);
1030 static inline int mdd_links_add(const struct lu_env *env,
1031 struct mdd_object *mdd_obj,
1032 const struct lu_fid *pfid,
1033 const struct lu_name *lname,
1034 struct thandle *handle,
1035 struct linkea_data *ldata, int first)
1037 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1038 pfid, lname, handle, ldata, first, 0);
1041 static inline int mdd_links_del(const struct lu_env *env,
1042 struct mdd_object *mdd_obj,
1043 const struct lu_fid *pfid,
1044 const struct lu_name *lname,
1045 struct thandle *handle)
1047 return mdd_links_rename(env, mdd_obj, pfid, lname,
1048 NULL, NULL, handle, NULL, 0, 0);
1051 /** Read the link EA into a temp buffer.
1052 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1053 * A pointer to the buffer is stored in \a ldata::ld_buf.
1055 * \retval 0 or error
1057 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1058 struct linkea_data *ldata)
1062 if (!mdd_object_exists(mdd_obj))
1065 /* First try a small buf */
1066 LASSERT(env != NULL);
1067 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1069 if (ldata->ld_buf->lb_buf == NULL)
1072 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1074 if (rc == -ERANGE) {
1075 /* Buf was too small, figure out what we need. */
1076 lu_buf_free(ldata->ld_buf);
1077 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1078 XATTR_NAME_LINK, BYPASS_CAPA);
1081 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1082 if (ldata->ld_buf->lb_buf == NULL)
1084 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1085 XATTR_NAME_LINK, BYPASS_CAPA);
1088 lu_buf_free(ldata->ld_buf);
1089 ldata->ld_buf = NULL;
1093 return linkea_init(ldata);
1096 /** Read the link EA into a temp buffer.
1097 * Uses the name_buf since it is generally large.
1098 * \retval IS_ERR err
1099 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1101 struct lu_buf *mdd_links_get(const struct lu_env *env,
1102 struct mdd_object *mdd_obj)
1104 struct linkea_data ldata = { 0 };
1107 rc = mdd_links_read(env, mdd_obj, &ldata);
1108 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1111 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1112 struct linkea_data *ldata, struct thandle *handle)
1114 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1115 ldata->ld_leh->leh_len);
1117 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1120 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1121 mdd_object_capa(env, mdd_obj));
1124 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1125 struct thandle *handle, struct linkea_data *ldata)
1131 if (ldata != NULL && ldata->ld_leh != NULL) {
1132 ea_len = ldata->ld_leh->leh_len;
1133 linkea = ldata->ld_buf->lb_buf;
1135 ea_len = DEFAULT_LINKEA_SIZE;
1139 /* XXX: max size? */
1140 rc = mdo_declare_xattr_set(env, mdd_obj,
1141 mdd_buf_get_const(env, linkea, ea_len),
1142 XATTR_NAME_LINK, 0, handle);
1146 static inline int mdd_declare_links_del(const struct lu_env *env,
1147 struct mdd_object *c,
1148 struct thandle *handle)
1152 /* For directory, the linkEA will be removed together
1153 * with the object. */
1154 if (!S_ISDIR(mdd_object_type(c)))
1155 rc = mdd_declare_links_add(env, c, handle, NULL);
1160 static int mdd_declare_link(const struct lu_env *env,
1161 struct mdd_device *mdd,
1162 struct mdd_object *p,
1163 struct mdd_object *c,
1164 const struct lu_name *name,
1165 struct thandle *handle,
1167 struct linkea_data *data)
1171 rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
1172 name->ln_name, handle);
1176 rc = mdo_declare_ref_add(env, c, handle);
1180 la->la_valid = LA_CTIME | LA_MTIME;
1181 rc = mdo_declare_attr_set(env, p, la, handle);
1185 la->la_valid = LA_CTIME;
1186 rc = mdo_declare_attr_set(env, c, la, handle);
1190 rc = mdd_declare_links_add(env, c, handle, data);
1194 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1199 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1200 struct md_object *src_obj, const struct lu_name *lname,
1203 const char *name = lname->ln_name;
1204 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1205 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1206 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1207 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1208 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1209 struct mdd_device *mdd = mdo2mdd(src_obj);
1210 struct thandle *handle;
1211 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1215 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
1219 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
1223 handle = mdd_trans_create(env, mdd);
1225 GOTO(out_pending, rc = PTR_ERR(handle));
1227 memset(ldata, 0, sizeof(*ldata));
1229 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1230 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1232 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1237 rc = mdd_trans_start(env, mdd, handle);
1241 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1242 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1245 GOTO(out_unlock, rc);
1247 rc = mdo_ref_add(env, mdd_sobj, handle);
1249 GOTO(out_unlock, rc);
1252 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) {
1253 struct lu_fid tfid = *mdo2fid(mdd_sobj);
1256 rc = __mdd_index_insert_only(env, mdd_tobj, &tfid,
1257 mdd_object_type(mdd_sobj),
1259 mdd_object_capa(env, mdd_tobj));
1261 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1262 mdd_object_type(mdd_sobj),
1264 mdd_object_capa(env, mdd_tobj));
1268 mdo_ref_del(env, mdd_sobj, handle);
1269 GOTO(out_unlock, rc);
1272 la->la_valid = LA_CTIME | LA_MTIME;
1273 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1275 GOTO(out_unlock, rc);
1277 la->la_valid = LA_CTIME;
1278 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1280 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1281 mdo2fid(mdd_tobj), lname, 0, 0,
1284 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1285 lname, handle, ldata, 0);
1286 /* The failure of links_add should not cause the link
1287 * failure, reset rc here */
1292 mdd_write_unlock(env, mdd_sobj);
1294 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1295 mdo2fid(mdd_tobj), NULL, NULL,
1296 lname, NULL, handle);
1298 mdd_trans_stop(env, mdd, rc, handle);
1300 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1301 /* if we vmalloced a large buffer drop it */
1302 lu_buf_free(ldata->ld_buf);
1307 static int mdd_mark_dead_object(const struct lu_env *env,
1308 struct mdd_object *obj, struct thandle *handle,
1311 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1315 obj->mod_flags |= DEAD_OBJ;
1317 if (!S_ISDIR(mdd_object_type(obj)))
1320 attr->la_valid = LA_FLAGS;
1321 attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
1324 rc = mdo_declare_attr_set(env, obj, attr, handle);
1326 rc = mdo_attr_set(env, obj, attr, handle,
1327 mdd_object_capa(env, obj));
1332 static int mdd_declare_finish_unlink(const struct lu_env *env,
1333 struct mdd_object *obj,
1334 struct thandle *handle)
1338 rc = mdd_mark_dead_object(env, obj, handle, true);
1342 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1346 rc = mdo_declare_destroy(env, obj, handle);
1350 return mdd_declare_links_del(env, obj, handle);
1353 /* caller should take a lock before calling */
1354 int mdd_finish_unlink(const struct lu_env *env,
1355 struct mdd_object *obj, struct md_attr *ma,
1356 const struct mdd_object *pobj,
1357 const struct lu_name *lname,
1361 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1364 LASSERT(mdd_write_locked(env, obj) != 0);
1366 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1367 rc = mdd_mark_dead_object(env, obj, th, false);
1371 /* add new orphan and the object
1372 * will be deleted during mdd_close() */
1373 if (obj->mod_count) {
1374 rc = __mdd_orphan_add(env, obj, th);
1376 CDEBUG(D_HA, "Object "DFID" is inserted into "
1377 "orphan list, open count = %d\n",
1378 PFID(mdd_object_fid(obj)),
1381 CERROR("Object "DFID" fail to be an orphan, "
1382 "open count = %d, maybe cause failed "
1384 PFID(mdd_object_fid(obj)),
1387 rc = mdo_destroy(env, obj, th);
1389 } else if (!is_dir) {
1390 /* old files may not have link ea; ignore errors */
1391 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1399 * has mdd_write_lock on cobj already, but not on pobj yet
1401 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1402 const struct lu_attr *pattr,
1403 struct mdd_object *cobj,
1404 const struct lu_attr *cattr)
1409 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1414 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1415 struct mdd_object *p, struct mdd_object *c,
1416 const struct lu_name *name, struct md_attr *ma,
1417 struct thandle *handle, int no_name, int is_dir)
1419 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1422 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1423 if (likely(no_name == 0)) {
1424 rc = mdo_declare_index_delete(env, p, name->ln_name,
1431 rc = mdo_declare_ref_del(env, p, handle);
1437 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1438 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1439 la->la_valid = LA_CTIME | LA_MTIME;
1440 rc = mdo_declare_attr_set(env, p, la, handle);
1445 rc = mdo_declare_ref_del(env, c, handle);
1449 rc = mdo_declare_ref_del(env, c, handle);
1453 la->la_valid = LA_CTIME;
1454 rc = mdo_declare_attr_set(env, c, la, handle);
1458 rc = mdd_declare_finish_unlink(env, c, handle);
1462 /* FIXME: need changelog for remove entry */
1463 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1470 * test if a file has an HSM archive
1471 * if HSM attributes are not found in ma update them from
1474 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1475 struct mdd_object *obj,
1480 if (!(ma->ma_valid & MA_HSM)) {
1481 /* no HSM MD provided, read xattr */
1482 struct lu_buf *hsm_buf;
1483 const size_t buflen = sizeof(struct hsm_attrs);
1486 hsm_buf = mdd_buf_get(env, NULL, 0);
1487 lu_buf_alloc(hsm_buf, buflen);
1488 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM,
1489 mdd_object_capa(env, obj));
1490 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1491 lu_buf_free(hsm_buf);
1495 ma->ma_valid = MA_HSM;
1497 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1503 * Delete name entry and the object.
1504 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1505 * does not exist for this object, and it could only happen during resending
1506 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1507 * is also needed in this case(needed by changelog), so we have to add another
1508 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1509 * the ENOENT failure should be able to be fixed by redo mechanism.
1511 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1512 struct md_object *cobj, const struct lu_name *lname,
1513 struct md_attr *ma, int no_name)
1515 const char *name = lname->ln_name;
1516 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1517 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1518 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1519 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1520 struct mdd_object *mdd_cobj = NULL;
1521 struct mdd_device *mdd = mdo2mdd(pobj);
1522 struct thandle *handle;
1526 /* cobj == NULL means only delete name entry */
1527 if (likely(cobj != NULL)) {
1528 mdd_cobj = md2mdd_obj(cobj);
1529 if (mdd_object_exists(mdd_cobj) == 0)
1533 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1537 if (likely(mdd_cobj != NULL)) {
1539 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1543 is_dir = S_ISDIR(cattr->la_mode);
1546 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1550 handle = mdd_trans_create(env, mdd);
1552 RETURN(PTR_ERR(handle));
1554 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1555 lname, ma, handle, no_name, is_dir);
1559 rc = mdd_trans_start(env, mdd, handle);
1563 if (likely(mdd_cobj != NULL))
1564 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1566 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1567 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1568 mdd_object_capa(env, mdd_pobj));
1573 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1574 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1575 GOTO(cleanup, rc = 0);
1577 if (likely(mdd_cobj != NULL)) {
1578 rc = mdo_ref_del(env, mdd_cobj, handle);
1580 __mdd_index_insert_only(env, mdd_pobj,
1582 mdd_object_type(mdd_cobj),
1584 mdd_object_capa(env, mdd_pobj));
1590 mdo_ref_del(env, mdd_cobj, handle);
1592 /* fetch updated nlink */
1593 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1598 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1599 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1601 la->la_valid = LA_CTIME | LA_MTIME;
1602 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1606 /* Enough for only unlink the entry */
1607 if (unlikely(mdd_cobj == NULL))
1610 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1611 /* update ctime of an unlinked file only if it is still
1612 * opened or a link still exists */
1613 la->la_valid = LA_CTIME;
1614 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1619 /* XXX: this transfer to ma will be removed with LOD/OSP */
1620 ma->ma_attr = *cattr;
1621 ma->ma_valid |= MA_INODE;
1622 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1624 /* fetch updated nlink */
1626 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1628 /* if object is removed then we can't get its attrs, use last get */
1629 if (cattr->la_nlink == 0) {
1630 ma->ma_attr = *cattr;
1631 ma->ma_valid |= MA_INODE;
1635 if (likely(mdd_cobj != NULL))
1636 mdd_write_unlock(env, mdd_cobj);
1641 if (cattr->la_nlink == 0) {
1642 cl_flags |= CLF_UNLINK_LAST;
1643 /* search for an existing archive */
1644 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1645 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1648 rc = mdd_changelog_ns_store(env, mdd,
1649 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1650 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1655 mdd_trans_stop(env, mdd, rc, handle);
1661 * The permission has been checked when obj created, no need check again.
1663 static int mdd_cd_sanity_check(const struct lu_env *env,
1664 struct mdd_object *obj)
1669 if (!obj || mdd_is_dead_obj(obj))
1675 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1676 struct md_object *cobj, const struct md_op_spec *spec,
1679 struct mdd_device *mdd = mdo2mdd(cobj);
1680 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1681 struct mdd_object *son = md2mdd_obj(cobj);
1682 struct thandle *handle;
1683 const struct lu_buf *buf;
1684 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1685 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1689 rc = mdd_cd_sanity_check(env, son);
1693 if (!md_should_create(spec->sp_cr_flags))
1697 * there are following use cases for this function:
1698 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1699 * striping can be specified or not
1702 rc = mdd_la_get(env, son, attr, BYPASS_CAPA);
1706 /* calling ->ah_make_hint() is used to transfer information from parent */
1707 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1709 handle = mdd_trans_create(env, mdd);
1711 GOTO(out_free, rc = PTR_ERR(handle));
1714 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1715 * Should this be fixed?
1717 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1718 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1719 spec->sp_cr_flags, spec->no_create);
1721 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1722 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1723 spec->u.sp_ea.eadatalen);
1728 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1729 XATTR_NAME_LOV, 0, handle);
1733 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1737 rc = mdd_trans_start(env, mdd, handle);
1741 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1742 0, handle, mdd_object_capa(env, son));
1747 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1750 mdd_trans_stop(env, mdd, rc, handle);
1755 static int mdd_declare_object_initialize(const struct lu_env *env,
1756 struct mdd_object *parent,
1757 struct mdd_object *child,
1758 struct lu_attr *attr,
1759 struct thandle *handle)
1765 * inode mode has been set in creation time, and it's based on umask,
1766 * la_mode and acl, don't set here again! (which will go wrong
1767 * because below function doesn't consider umask).
1768 * I'd suggest set all object attributes in creation time, see above.
1770 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1771 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1772 rc = mdo_declare_attr_set(env, child, attr, handle);
1773 attr->la_valid |= LA_MODE | LA_TYPE;
1774 if (rc != 0 || !S_ISDIR(attr->la_mode))
1777 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1782 rc = mdo_declare_ref_add(env, child, handle);
1786 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1792 static int mdd_object_initialize(const struct lu_env *env,
1793 const struct lu_fid *pfid,
1794 struct mdd_object *child,
1795 struct lu_attr *attr, struct thandle *handle,
1796 const struct md_op_spec *spec)
1802 * Update attributes for child.
1805 * (1) the valid bits should be converted between Lustre and Linux;
1806 * (2) maybe, the child attributes should be set in OSD when creation.
1809 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1810 /* arguments are supposed to stay the same */
1811 if (S_ISDIR(attr->la_mode)) {
1812 /* Add "." and ".." for newly created dir */
1813 mdo_ref_add(env, child, handle);
1814 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1815 S_IFDIR, dot, handle, BYPASS_CAPA);
1817 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1821 mdo_ref_del(env, child, handle);
1828 * This function checks whether it can create a file/dir under the
1829 * directory(@pobj). The directory(@pobj) is not being locked by
1832 * \param[in] env execution environment
1833 * \param[in] pobj the directory to create files
1834 * \param[in] pattr the attributes of the directory
1835 * \param[in] lname the name of the created file/dir
1836 * \param[in] cattr the attributes of the file/dir
1837 * \param[in] spec create specification
1839 * \retval = 0 it is allowed to create file/dir under
1841 * \retval negative error not allowed to create file/dir
1842 * under the directory
1844 static int mdd_create_sanity_check(const struct lu_env *env,
1845 struct md_object *pobj,
1846 const struct lu_attr *pattr,
1847 const struct lu_name *lname,
1848 struct lu_attr *cattr,
1849 struct md_op_spec *spec)
1851 struct mdd_thread_info *info = mdd_env_info(env);
1852 struct lu_fid *fid = &info->mti_fid;
1853 struct mdd_object *obj = md2mdd_obj(pobj);
1854 struct mdd_device *m = mdo2mdd(pobj);
1855 bool check_perm = true;
1860 if (mdd_is_dead_obj(obj))
1864 * In some cases this lookup is not needed - we know before if name
1865 * exists or not because MDT performs lookup for it.
1866 * name length check is done in lookup.
1868 if (spec->sp_cr_lookup) {
1870 * Check if the name already exist, though it will be checked in
1871 * _index_insert also, for avoiding rolling back if exists
1874 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1875 MAY_WRITE | MAY_EXEC);
1877 RETURN(rc ? : -EEXIST);
1879 /* Permission is already being checked in mdd_lookup */
1883 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1888 if (pattr->la_mode & S_ISGID) {
1889 cattr->la_gid = pattr->la_gid;
1890 if (S_ISDIR(cattr->la_mode)) {
1891 cattr->la_mode |= S_ISGID;
1892 cattr->la_valid |= LA_MODE;
1896 rc = mdd_name_check(m, lname);
1900 switch (cattr->la_mode & S_IFMT) {
1902 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1904 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1905 RETURN(-ENAMETOOLONG);
1924 static int mdd_declare_object_create(const struct lu_env *env,
1925 struct mdd_device *mdd,
1926 struct mdd_object *p, struct mdd_object *c,
1927 struct lu_attr *attr,
1928 struct thandle *handle,
1929 const struct md_op_spec *spec,
1930 struct lu_buf *def_acl_buf,
1931 struct lu_buf *acl_buf,
1932 struct dt_allocation_hint *hint)
1936 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
1941 #ifdef CONFIG_FS_POSIX_ACL
1942 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1943 /* if dir, then can inherit default ACl */
1944 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1945 XATTR_NAME_ACL_DEFAULT,
1951 if (acl_buf->lb_len > 0) {
1952 rc = mdo_declare_attr_set(env, c, attr, handle);
1956 rc = mdo_declare_xattr_set(env, c, acl_buf,
1957 XATTR_NAME_ACL_ACCESS, 0, handle);
1962 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
1966 /* replay case, create LOV EA from client data */
1967 if (spec->no_create ||
1968 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
1969 const struct lu_buf *buf;
1971 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1972 spec->u.sp_ea.eadatalen);
1973 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
1979 if (S_ISLNK(attr->la_mode)) {
1980 const char *target_name = spec->u.sp_symname;
1981 int sym_len = strlen(target_name);
1982 const struct lu_buf *buf;
1984 buf = mdd_buf_get_const(env, target_name, sym_len);
1985 rc = dt_declare_record_write(env, mdd_object_child(c),
1994 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
1995 struct mdd_object *p, struct mdd_object *c,
1996 const struct lu_name *name,
1997 struct lu_attr *attr,
1998 struct thandle *handle,
1999 const struct md_op_spec *spec,
2000 struct linkea_data *ldata,
2001 struct lu_buf *def_acl_buf,
2002 struct lu_buf *acl_buf,
2003 struct dt_allocation_hint *hint)
2007 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2008 def_acl_buf, acl_buf, hint);
2012 if (S_ISDIR(attr->la_mode)) {
2013 rc = mdo_declare_ref_add(env, p, handle);
2018 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2019 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2023 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2025 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2026 name->ln_name, handle);
2030 rc = mdd_declare_links_add(env, c, handle, ldata);
2035 la->la_valid = LA_CTIME | LA_MTIME;
2036 rc = mdo_declare_attr_set(env, p, la, handle);
2040 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2045 /* XXX: For remote create, it should indicate the remote RPC
2046 * will be sent after local transaction is finished, which
2047 * is not very nice, but it will be removed once we fully support
2049 if (mdd_object_remote(p) && handle->th_update != NULL)
2050 handle->th_update->tu_sent_after_local_trans = 1;
2055 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2056 struct lu_attr *la, struct lu_buf *def_acl_buf,
2057 struct lu_buf *acl_buf)
2062 if (S_ISLNK(la->la_mode)) {
2063 acl_buf->lb_len = 0;
2064 def_acl_buf->lb_len = 0;
2068 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2069 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2070 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
2071 mdd_read_unlock(env, pobj);
2073 /* If there are default ACL, fix mode/ACL by default ACL */
2074 def_acl_buf->lb_len = rc;
2075 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2076 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2077 acl_buf->lb_len = rc;
2078 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2081 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2082 /* If there are no default ACL, fix mode by mask */
2083 struct lu_ucred *uc = lu_ucred(env);
2085 /* The create triggered by MDT internal events, such as
2086 * LFSCK reset, will not contain valid "uc". */
2087 if (unlikely(uc != NULL))
2088 la->la_mode &= ~uc->uc_umask;
2090 acl_buf->lb_len = 0;
2091 def_acl_buf->lb_len = 0;
2098 * Create a metadata object and initialize it, set acl, xattr.
2100 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2101 struct mdd_object *son, struct lu_attr *attr,
2102 struct md_op_spec *spec, struct lu_buf *acl_buf,
2103 struct lu_buf *def_acl_buf,
2104 struct dt_allocation_hint *hint,
2105 struct thandle *handle)
2109 mdd_write_lock(env, son, MOR_TGT_CHILD);
2110 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2115 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2116 * created in declare phase, they also needs to be added to master
2117 * object as sub-directory entry. So it has to initialize the master
2118 * object, then set dir striped EA.(in mdo_xattr_set) */
2119 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2122 GOTO(err_destroy, rc);
2125 * in case of replay we just set LOVEA provided by the client
2126 * XXX: I think it would be interesting to try "old" way where
2127 * MDT calls this xattr_set(LOV) in a different transaction.
2128 * probably this way we code can be made better.
2131 /* During creation, there are only a few cases we need do xattr_set to
2133 * 1. regular file: see comments above.
2134 * 2. create striped directory with provided stripeEA.
2135 * 3. create striped directory because inherit default layout from the
2137 if (spec->no_create ||
2138 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2139 S_ISDIR(attr->la_mode)) {
2140 const struct lu_buf *buf;
2142 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2143 spec->u.sp_ea.eadatalen);
2144 rc = mdo_xattr_set(env, son, buf,
2145 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2147 handle, BYPASS_CAPA);
2149 GOTO(err_destroy, rc);
2152 #ifdef CONFIG_FS_POSIX_ACL
2153 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2154 S_ISDIR(attr->la_mode)) {
2155 /* set default acl */
2156 rc = mdo_xattr_set(env, son, def_acl_buf,
2157 XATTR_NAME_ACL_DEFAULT, 0,
2158 handle, BYPASS_CAPA);
2160 GOTO(err_destroy, rc);
2162 /* set its own acl */
2163 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2164 rc = mdo_xattr_set(env, son, acl_buf,
2165 XATTR_NAME_ACL_ACCESS,
2166 0, handle, BYPASS_CAPA);
2168 GOTO(err_destroy, rc);
2172 if (S_ISLNK(attr->la_mode)) {
2173 struct lu_ucred *uc = lu_ucred_assert(env);
2174 struct dt_object *dt = mdd_object_child(son);
2175 const char *target_name = spec->u.sp_symname;
2176 int sym_len = strlen(target_name);
2177 const struct lu_buf *buf;
2180 buf = mdd_buf_get_const(env, target_name, sym_len);
2181 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2182 mdd_object_capa(env, son),
2184 CFS_CAP_SYS_RESOURCE_MASK);
2189 GOTO(err_initlized, rc = -EFAULT);
2193 if (unlikely(rc != 0)) {
2195 if (S_ISDIR(attr->la_mode)) {
2196 /* Drop the reference, no need to delete "."/"..",
2197 * because the object to be destroied directly. */
2198 rc2 = mdo_ref_del(env, son, handle);
2202 rc2 = mdo_ref_del(env, son, handle);
2206 mdo_destroy(env, son, handle);
2209 mdd_write_unlock(env, son);
2214 * Create object and insert it into namespace.
2216 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2217 const struct lu_name *lname, struct md_object *child,
2218 struct md_op_spec *spec, struct md_attr* ma)
2220 struct mdd_thread_info *info = mdd_env_info(env);
2221 struct lu_attr *la = &info->mti_la_for_fix;
2222 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2223 struct mdd_object *son = md2mdd_obj(child);
2224 struct mdd_device *mdd = mdo2mdd(pobj);
2225 struct lu_attr *attr = &ma->ma_attr;
2226 struct thandle *handle;
2227 struct lu_attr *pattr = &info->mti_pattr;
2228 struct lu_buf acl_buf;
2229 struct lu_buf def_acl_buf;
2230 struct linkea_data *ldata = &info->mti_link_data;
2231 const char *name = lname->ln_name;
2232 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2238 * Two operations have to be performed:
2240 * - an allocation of a new object (->do_create()), and
2242 * - an insertion into a parent index (->dio_insert()).
2244 * Due to locking, operation order is not important, when both are
2245 * successful, *but* error handling cases are quite different:
2247 * - if insertion is done first, and following object creation fails,
2248 * insertion has to be rolled back, but this operation might fail
2249 * also leaving us with dangling index entry.
2251 * - if creation is done first, is has to be undone if insertion
2252 * fails, leaving us with leaked space, which is neither good, nor
2255 * It seems that creation-first is simplest solution, but it is
2256 * sub-optimal in the frequent
2261 * case, because second mkdir is bound to create object, only to
2262 * destroy it immediately.
2264 * To avoid this follow local file systems that do double lookup:
2266 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2268 * 1. create (mdd_object_create_internal())
2270 * 2. insert (__mdd_index_insert(), lookup again)
2273 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2277 /* Sanity checks before big job. */
2278 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2282 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2283 GOTO(out_free, rc = -EINPROGRESS);
2285 handle = mdd_trans_create(env, mdd);
2287 GOTO(out_free, rc = PTR_ERR(handle));
2289 acl_buf.lb_buf = info->mti_xattr_buf;
2290 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2291 def_acl_buf.lb_buf = info->mti_key;
2292 def_acl_buf.lb_len = sizeof(info->mti_key);
2293 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2297 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2299 memset(ldata, 0, sizeof(*ldata));
2300 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT2)) {
2301 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2304 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2305 &tfid, lname, 1, 0, ldata);
2307 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2308 mdd_object_fid(mdd_pobj),
2309 lname, 1, 0, ldata);
2312 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2313 handle, spec, ldata, &def_acl_buf, &acl_buf,
2318 rc = mdd_trans_start(env, mdd, handle);
2322 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2323 &def_acl_buf, hint, handle);
2327 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2328 mdd_write_lock(env, son, MOR_TGT_CHILD);
2329 rc = __mdd_orphan_add(env, son, handle);
2330 GOTO(out_volatile, rc);
2332 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2333 attr->la_mode, name, handle,
2334 mdd_object_capa(env, mdd_pobj));
2336 GOTO(err_created, rc);
2338 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2341 /* update parent directory mtime/ctime */
2343 la->la_valid = LA_CTIME | LA_MTIME;
2344 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2346 GOTO(err_insert, rc);
2354 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2355 rc2 = __mdd_orphan_del(env, son, handle);
2357 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2358 S_ISDIR(attr->la_mode),
2359 handle, BYPASS_CAPA);
2364 mdd_write_lock(env, son, MOR_TGT_CHILD);
2365 if (S_ISDIR(attr->la_mode)) {
2366 /* Drop the reference, no need to delete "."/"..",
2367 * because the object is to be destroyed directly. */
2368 rc2 = mdo_ref_del(env, son, handle);
2370 mdd_write_unlock(env, son);
2375 /* For volatile files drop one link immediately, since there is
2376 * no filename in the namespace, and save any error returned. */
2377 rc2 = mdo_ref_del(env, son, handle);
2379 mdd_write_unlock(env, son);
2380 if (unlikely(rc == 0))
2385 /* Don't destroy the volatile object on success */
2386 if (likely(rc != 0))
2387 mdo_destroy(env, son, handle);
2388 mdd_write_unlock(env, son);
2391 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2392 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2393 rc = mdd_changelog_ns_store(env, mdd,
2394 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2395 S_ISREG(attr->la_mode) ? CL_CREATE :
2396 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2397 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2400 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2404 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2405 /* if we vmalloced a large buffer drop it */
2406 lu_buf_free(ldata->ld_buf);
2408 /* The child object shouldn't be cached anymore */
2410 set_bit(LU_OBJECT_HEARD_BANSHEE,
2411 &child->mo_lu.lo_header->loh_flags);
2416 * Get locks on parents in proper order
2417 * RETURN: < 0 - error, rename_order if successful
2425 static int mdd_rename_order(const struct lu_env *env,
2426 struct mdd_device *mdd,
2427 struct mdd_object *src_pobj,
2428 const struct lu_attr *pattr,
2429 struct mdd_object *tgt_pobj)
2431 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2435 if (src_pobj == tgt_pobj)
2436 RETURN(MDD_RN_SAME);
2438 /* compared the parent child relationship of src_p&tgt_p */
2439 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2441 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2444 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2458 /* has not mdd_write{read}_lock on any obj yet. */
2459 static int mdd_rename_sanity_check(const struct lu_env *env,
2460 struct mdd_object *src_pobj,
2461 const struct lu_attr *pattr,
2462 struct mdd_object *tgt_pobj,
2463 const struct lu_attr *tpattr,
2464 struct mdd_object *sobj,
2465 const struct lu_attr *cattr,
2466 struct mdd_object *tobj,
2467 const struct lu_attr *tattr)
2472 /* XXX: when get here, sobj must NOT be NULL,
2473 * the other case has been processed in cld_rename
2474 * before mdd_rename and enable MDS_PERM_BYPASS. */
2477 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2481 /* XXX: when get here, "tobj == NULL" means tobj must
2482 * NOT exist (neither on remote MDS, such case has been
2483 * processed in cld_rename before mdd_rename and enable
2485 * So check may_create, but not check may_unlink. */
2487 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2488 (src_pobj != tgt_pobj));
2490 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2491 (src_pobj != tgt_pobj), 1);
2493 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2494 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2499 static int mdd_declare_rename(const struct lu_env *env,
2500 struct mdd_device *mdd,
2501 struct mdd_object *mdd_spobj,
2502 struct mdd_object *mdd_tpobj,
2503 struct mdd_object *mdd_sobj,
2504 struct mdd_object *mdd_tobj,
2505 const struct lu_name *tname,
2506 const struct lu_name *sname,
2508 struct linkea_data *ldata,
2509 struct thandle *handle)
2511 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2514 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2515 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2521 /* name from source dir */
2522 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2526 /* .. from source child */
2527 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2528 /* source child can be directory,
2529 * counted by source dir's nlink */
2530 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2533 if (mdd_spobj != mdd_tpobj) {
2534 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2539 rc = mdo_declare_index_insert(env, mdd_sobj,
2541 S_IFDIR, dotdot, handle);
2546 /* new target child can be directory,
2547 * counted by target dir's nlink */
2548 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2553 la->la_valid = LA_CTIME | LA_MTIME;
2554 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2558 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2562 la->la_valid = LA_CTIME;
2563 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2567 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2572 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2573 mdd_object_type(mdd_sobj),
2574 tname->ln_name, handle);
2578 /* name from target dir (old name), we declare it unconditionally
2579 * as mdd_rename() calls delete unconditionally as well. so just
2580 * to balance declarations vs calls to change ... */
2581 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2585 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2586 /* delete target child in target parent directory */
2587 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2591 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2592 /* target child can be directory,
2593 * delete "." reference in target child directory */
2594 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2598 /* delete ".." reference in target parent directory */
2599 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2604 la->la_valid = LA_CTIME;
2605 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2609 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2614 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2621 /* src object can be remote that is why we use only fid and type of object */
2622 static int mdd_rename(const struct lu_env *env,
2623 struct md_object *src_pobj, struct md_object *tgt_pobj,
2624 const struct lu_fid *lf, const struct lu_name *lsname,
2625 struct md_object *tobj, const struct lu_name *ltname,
2628 const char *sname = lsname->ln_name;
2629 const char *tname = ltname->ln_name;
2630 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2631 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2632 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2633 struct mdd_device *mdd = mdo2mdd(src_pobj);
2634 struct mdd_object *mdd_sobj = NULL; /* source object */
2635 struct mdd_object *mdd_tobj = NULL;
2636 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2637 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2638 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2639 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2640 struct thandle *handle;
2641 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2642 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2643 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2646 bool tobj_locked = 0;
2647 unsigned cl_flags = 0;
2652 mdd_tobj = md2mdd_obj(tobj);
2654 mdd_sobj = mdd_object_find(env, mdd, lf);
2656 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
2658 GOTO(out_pending, rc);
2660 rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA);
2662 GOTO(out_pending, rc);
2665 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2667 GOTO(out_pending, rc);
2670 rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA);
2672 GOTO(out_pending, rc);
2674 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2675 mdd_sobj, cattr, mdd_tobj, tattr);
2677 GOTO(out_pending, rc);
2679 rc = mdd_name_check(mdd, ltname);
2681 GOTO(out_pending, rc);
2683 handle = mdd_trans_create(env, mdd);
2685 GOTO(out_pending, rc = PTR_ERR(handle));
2687 memset(ldata, 0, sizeof(*ldata));
2688 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2689 ltname, 1, 0, ldata);
2690 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2691 mdd_tobj, lsname, ltname, ma, ldata, handle);
2695 rc = mdd_trans_start(env, mdd, handle);
2699 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2700 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2702 GOTO(cleanup_unlocked, rc);
2704 is_dir = S_ISDIR(cattr->la_mode);
2706 /* Remove source name from source directory */
2707 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2708 mdd_object_capa(env, mdd_spobj));
2712 /* "mv dir1 dir2" needs "dir1/.." link update */
2713 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2714 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2715 mdd_object_capa(env, mdd_sobj));
2717 GOTO(fixup_spobj2, rc);
2719 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2721 mdd_object_capa(env, mdd_sobj));
2723 GOTO(fixup_spobj, rc);
2726 /* Remove target name from target directory
2727 * Here tobj can be remote one, so we do index_delete unconditionally
2728 * and -ENOENT is allowed.
2730 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2731 mdd_object_capa(env, mdd_tpobj));
2734 /* tname might been renamed to something else */
2735 GOTO(fixup_spobj, rc);
2738 GOTO(fixup_spobj, rc);
2741 /* Insert new fid with target name into target dir */
2742 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2743 tname, handle, mdd_object_capa(env, mdd_tpobj));
2745 GOTO(fixup_tpobj, rc);
2747 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2748 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2750 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2752 la->la_valid = LA_CTIME;
2753 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2755 GOTO(fixup_tpobj, rc);
2758 /* Remove old target object
2759 * For tobj is remote case cmm layer has processed
2760 * and set tobj to NULL then. So when tobj is NOT NULL,
2761 * it must be local one.
2763 if (tobj && mdd_object_exists(mdd_tobj)) {
2764 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2766 if (mdd_is_dead_obj(mdd_tobj)) {
2767 /* shld not be dead, something is wrong */
2768 CERROR("tobj is dead, something is wrong\n");
2772 mdo_ref_del(env, mdd_tobj, handle);
2774 /* Remove dot reference. */
2775 if (S_ISDIR(tattr->la_mode))
2776 mdo_ref_del(env, mdd_tobj, handle);
2779 /* fetch updated nlink */
2780 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2782 CERROR("%s: Failed to get nlink for tobj "
2784 mdd2obd_dev(mdd)->obd_name,
2785 PFID(tpobj_fid), rc);
2786 GOTO(fixup_tpobj, rc);
2789 la->la_valid = LA_CTIME;
2790 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2792 CERROR("%s: Failed to set ctime for tobj "
2794 mdd2obd_dev(mdd)->obd_name,
2795 PFID(tpobj_fid), rc);
2796 GOTO(fixup_tpobj, rc);
2799 /* XXX: this transfer to ma will be removed with LOD/OSP */
2800 ma->ma_attr = *tattr;
2801 ma->ma_valid |= MA_INODE;
2802 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2805 CERROR("%s: Failed to unlink tobj "
2807 mdd2obd_dev(mdd)->obd_name,
2808 PFID(tpobj_fid), rc);
2809 GOTO(fixup_tpobj, rc);
2812 /* fetch updated nlink */
2813 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2815 CERROR("%s: Failed to get nlink for tobj "
2817 mdd2obd_dev(mdd)->obd_name,
2818 PFID(tpobj_fid), rc);
2819 GOTO(fixup_tpobj, rc);
2821 /* XXX: this transfer to ma will be removed with LOD/OSP */
2822 ma->ma_attr = *tattr;
2823 ma->ma_valid |= MA_INODE;
2825 if (tattr->la_nlink == 0) {
2826 cl_flags |= CLF_RENAME_LAST;
2827 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2828 cl_flags |= CLF_RENAME_LAST_EXISTS;
2832 la->la_valid = LA_CTIME | LA_MTIME;
2833 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2835 GOTO(fixup_tpobj, rc);
2837 if (mdd_spobj != mdd_tpobj) {
2838 la->la_valid = LA_CTIME | LA_MTIME;
2839 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2842 if (rc == 0 && mdd_sobj) {
2843 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2844 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2845 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2848 /* Old files might not have EA entry */
2849 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2850 lsname, handle, NULL, 0);
2851 mdd_write_unlock(env, mdd_sobj);
2852 /* We don't fail the transaction if the link ea can't be
2853 updated -- fid2path will use alternate lookup method. */
2861 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2864 CWARN("tp obj fix error %d\n",rc2);
2866 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2867 !mdd_is_dead_obj(mdd_tobj)) {
2869 mdo_ref_add(env, mdd_tobj, handle);
2871 mdo_ref_add(env, mdd_tobj, handle);
2874 rc2 = __mdd_index_insert(env, mdd_tpobj,
2876 mdd_object_type(mdd_tobj),
2877 tname, handle, BYPASS_CAPA);
2879 CWARN("tp obj fix error: rc = %d\n", rc2);
2884 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2885 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2889 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2890 mdd2obd_dev(mdd)->obd_name, rc2);
2893 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
2894 dotdot, handle, BYPASS_CAPA);
2896 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2897 mdd2obd_dev(mdd)->obd_name, rc2);
2902 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
2903 mdd_object_type(mdd_sobj), sname,
2904 handle, BYPASS_CAPA);
2906 CWARN("sp obj fix error: rc = %d\n", rc2);
2911 mdd_write_unlock(env, mdd_tobj);
2915 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
2916 mdd_tobj, tpobj_fid, lf, spobj_fid,
2917 ltname, lsname, handle);
2920 mdd_trans_stop(env, mdd, rc, handle);
2923 mdd_object_put(env, mdd_sobj);
2928 * During migration once the parent FID has been changed,
2929 * we need update the parent FID in linkea.
2931 static int mdd_linkea_update_child_internal(const struct lu_env *env,
2932 struct mdd_object *parent,
2933 struct mdd_object *child,
2934 const char *name, int namelen,
2935 struct thandle *handle,
2938 struct mdd_thread_info *info = mdd_env_info(env);
2939 struct linkea_data ldata = {0};
2940 struct lu_buf *buf = &info->mti_link_buf;
2946 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2947 if (buf->lb_buf == NULL)
2951 rc = mdd_links_read(env, child, &ldata);
2953 if (rc == -ENOENT || rc == -ENODATA)
2958 LASSERT(ldata.ld_leh != NULL);
2959 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
2960 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
2961 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
2962 struct lu_name lname;
2965 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
2968 if (strncmp(lname.ln_name, name, namelen) != 0 ||
2969 lu_fid_eq(&fid, mdd_object_fid(parent))) {
2970 ldata.ld_lee = (struct link_ea_entry *)
2971 ((char *)ldata.ld_lee +
2976 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
2977 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
2978 lname.ln_namelen, lname.ln_name,
2979 PFID(mdd_object_fid(parent)));
2980 /* update to the new parent fid */
2981 linkea_entry_pack(ldata.ld_lee, &lname,
2982 mdd_object_fid(parent));
2984 rc = mdd_declare_links_add(env, child, handle, &ldata);
2986 rc = mdd_links_write(env, child, &ldata, handle);
2992 static int mdd_linkea_declare_update_child(const struct lu_env *env,
2993 struct mdd_object *parent,
2994 struct mdd_object *child,
2995 const char *name, int namelen,
2996 struct thandle *handle)
2998 return mdd_linkea_update_child_internal(env, parent, child, name,
2999 namelen, handle, true);
3002 static int mdd_linkea_update_child(const struct lu_env *env,
3003 struct mdd_object *parent,
3004 struct mdd_object *child,
3005 const char *name, int namelen,
3006 struct thandle *handle)
3008 return mdd_linkea_update_child_internal(env, parent, child, name,
3009 namelen, handle, false);
3012 static int mdd_update_linkea_internal(const struct lu_env *env,
3013 struct mdd_object *mdd_pobj,
3014 struct mdd_object *mdd_sobj,
3015 struct mdd_object *mdd_tobj,
3016 const struct lu_name *child_name,
3017 struct thandle *handle,
3020 struct mdd_thread_info *info = mdd_env_info(env);
3021 struct linkea_data *ldata = &info->mti_link_data;
3026 rc = mdd_links_read(env, mdd_sobj, ldata);
3028 if (rc == -ENOENT || rc == -ENODATA)
3034 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3036 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3041 /* If it is mulitple links file, we need update the name entry for
3043 LASSERT(ldata->ld_leh != NULL);
3044 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3045 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3046 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3047 struct mdd_object *pobj;
3048 struct lu_name lname;
3051 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3053 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3055 pobj = mdd_object_find(env, mdd, &fid);
3057 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3058 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3063 if (!mdd_object_exists(pobj)) {
3064 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3065 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3069 if (pobj == mdd_pobj &&
3070 lname.ln_namelen == child_name->ln_namelen &&
3071 strncmp(lname.ln_name, child_name->ln_name,
3072 lname.ln_namelen) == 0) {
3073 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3074 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3079 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3080 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3081 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3084 /* Remove source name from source directory */
3085 /* Insert new fid with target name into target dir */
3086 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3091 rc = mdo_declare_index_insert(env, pobj,
3092 mdd_object_fid(mdd_tobj),
3093 mdd_object_type(mdd_tobj),
3094 lname.ln_name, handle);
3098 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3102 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3106 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3108 mdd_object_capa(env, pobj));
3112 rc = __mdd_index_insert(env, pobj,
3113 mdd_object_fid(mdd_tobj),
3114 mdd_object_type(mdd_tobj),
3115 lname.ln_name, handle,
3116 mdd_object_capa(env, pobj));
3120 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3121 rc = mdo_ref_add(env, mdd_tobj, handle);
3122 mdd_write_unlock(env, mdd_tobj);
3126 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3127 mdo_ref_del(env, mdd_sobj, handle);
3128 mdd_write_unlock(env, mdd_sobj);
3131 mdd_object_put(env, pobj);
3139 static int mdd_migrate_xattrs(const struct lu_env *env,
3140 struct mdd_object *mdd_sobj,
3141 struct mdd_object *mdd_tobj)
3143 struct mdd_thread_info *info = mdd_env_info(env);
3144 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3146 struct thandle *handle;
3152 struct lu_buf list_xbuf;
3155 /* retrieve xattr list from the old object */
3156 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL,
3157 mdd_object_capa(env, mdd_sobj));
3158 if (list_xsize == -ENODATA)
3164 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3165 if (info->mti_big_buf.lb_buf == NULL)
3168 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3169 list_xbuf.lb_len = list_xsize;
3170 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf,
3171 mdd_object_capa(env, mdd_sobj));
3176 xname = list_xbuf.lb_buf;
3178 xlen = strnlen(xname, rem - 1) + 1;
3179 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3180 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3181 strcmp(XATTR_NAME_LMV, xname) == 0)
3184 /* For directory, if there are default layout, migrate here */
3185 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3186 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3189 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL,
3191 mdd_object_capa(env, mdd_sobj));
3192 if (xsize == -ENODATA)
3197 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3198 if (info->mti_link_buf.lb_buf == NULL)
3199 GOTO(out, rc = -ENOMEM);
3201 xbuf.lb_len = xsize;
3202 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3203 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname,
3204 mdd_object_capa(env, mdd_sobj));
3210 handle = mdd_trans_create(env, mdd);
3212 GOTO(out, rc = PTR_ERR(handle));
3214 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3217 GOTO(stop_trans, rc);
3218 /* Note: this transaction is part of migration, and it is not
3219 * the last step of migration, so we set th_local = 1 to avoid
3220 * update last rcvd for this transaction */
3221 handle->th_local = 1;
3222 rc = mdd_trans_start(env, mdd, handle);
3224 GOTO(stop_trans, rc);
3226 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle,
3227 mdd_object_capa(env, mdd_sobj));
3229 GOTO(stop_trans, rc = 0);
3232 GOTO(stop_trans, rc);
3234 mdd_trans_stop(env, mdd, rc, handle);
3239 memmove(xname, xname + xlen, rem);
3245 static int mdd_declare_migrate_create(const struct lu_env *env,
3246 struct mdd_object *mdd_pobj,
3247 struct mdd_object *mdd_sobj,
3248 struct mdd_object *mdd_tobj,
3249 struct md_op_spec *spec,
3251 union lmv_mds_md *mgr_ea,
3252 struct thandle *handle)
3254 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3255 const struct lu_buf *buf;
3259 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3260 handle, spec, NULL);
3264 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3269 if (S_ISLNK(la->la_mode)) {
3270 const char *target_name = spec->u.sp_symname;
3271 int sym_len = strlen(target_name);
3272 const struct lu_buf *buf;
3274 buf = mdd_buf_get_const(env, target_name, sym_len);
3275 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3281 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3282 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3283 spec->u.sp_ea.eadatalen);
3284 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3290 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3291 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3292 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3297 la_flag->la_valid = LA_FLAGS;
3298 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3299 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3300 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3305 static int mdd_migrate_create(const struct lu_env *env,
3306 struct mdd_object *mdd_pobj,
3307 struct mdd_object *mdd_sobj,
3308 struct mdd_object *mdd_tobj,
3311 struct mdd_thread_info *info = mdd_env_info(env);
3312 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3313 struct md_op_spec *spec = &info->mti_spec;
3314 struct lu_buf lmm_buf = { 0 };
3315 struct lu_buf link_buf = { 0 };
3316 const struct lu_buf *buf;
3317 struct thandle *handle;
3318 struct lmv_mds_md_v1 *mgr_ea;
3319 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3320 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3325 /* prepare spec for create */
3326 memset(spec, 0, sizeof(*spec));
3327 spec->sp_cr_lookup = 0;
3328 spec->sp_feat = &dt_directory_features;
3329 if (S_ISLNK(la->la_mode)) {
3330 buf = lu_buf_check_and_alloc(
3331 &mdd_env_info(env)->mti_big_buf,
3334 link_buf.lb_len = la->la_size + 1;
3335 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3337 rc = rc != 0 ? rc : -EFAULT;
3338 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3339 mdd2obd_dev(mdd)->obd_name,
3340 PFID(mdd_object_fid(mdd_sobj)), rc);
3343 spec->u.sp_symname = link_buf.lb_buf;
3344 } else if S_ISREG(la->la_mode) {
3345 /* retrieve lov of the old object */
3346 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3347 if (rc != 0 && rc != -ENODATA)
3349 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3350 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3351 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3352 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3356 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3357 memset(mgr_ea, 0, sizeof(*mgr_ea));
3358 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3359 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3360 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3361 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3362 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3363 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3365 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3367 handle = mdd_trans_create(env, mdd);
3369 GOTO(out_free, rc = PTR_ERR(handle));
3371 /* Note: this transaction is part of migration, and it is not
3372 * the last step of migration, so we set th_local = 1 to avoid
3373 * update last rcvd for this transaction */
3374 handle->th_local = 1;
3375 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3377 (union lmv_mds_md *)info->mti_xattr_buf,
3380 GOTO(stop_trans, rc);
3382 rc = mdd_trans_start(env, mdd, handle);
3384 GOTO(stop_trans, rc);
3386 /* create the target object */
3387 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3390 GOTO(stop_trans, rc);
3392 /* Set MIGRATE EA on the source inode, so once the migration needs
3393 * to be re-done during failover, the re-do process can locate the
3394 * target object which is already being created. */
3395 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3396 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3397 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0,
3398 handle, mdd_object_capa(env, mdd_sobj));
3400 GOTO(stop_trans, rc);
3402 /* Set immutable flag, so any modification is disabled until
3403 * the migration is done. Once the migration is interrupted,
3404 * if the resume process find the migrating object has both
3405 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3406 * flag and approve the migration */
3407 la_flag->la_valid = LA_FLAGS;
3408 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3409 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3410 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3411 mdd_object_capa(env, mdd_sobj));
3414 mdd_trans_stop(env, mdd, rc, handle);
3416 if (lmm_buf.lb_buf != NULL)
3417 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3421 static int mdd_migrate_entries(const struct lu_env *env,
3422 struct mdd_object *mdd_sobj,
3423 struct mdd_object *mdd_tobj)
3425 struct dt_object *next = mdd_object_child(mdd_sobj);
3426 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3427 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3428 struct thandle *handle;
3430 const struct dt_it_ops *iops;
3433 struct lu_dirent *ent;
3436 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3440 if (!dt_try_as_dir(env, next))
3441 GOTO(out_ent, rc = -ENOTDIR);
3443 * iterate directories
3445 iops = &next->do_index_ops->dio_it;
3446 it = iops->init(env, next, LUDA_FID | LUDA_TYPE,
3447 mdd_object_capa(env, mdd_sobj));
3449 GOTO(out_ent, rc = PTR_ERR(it));
3451 rc = iops->load(env, it, 0);
3453 rc = iops->next(env, it);
3457 * At this point and across for-loop:
3459 * rc == 0 -> ok, proceed.
3460 * rc > 0 -> end of directory.
3464 struct mdd_object *child;
3465 char *name = mdd_env_info(env)->mti_key;
3469 bool target_exist = false;
3471 len = iops->key_size(env, it);
3475 result = iops->rec(env, it, (struct dt_rec *)ent,
3476 LUDA_FID | LUDA_TYPE);
3477 if (result == -ESTALE)
3484 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3485 recsize = le16_to_cpu(ent->lde_reclen);
3487 /* Insert new fid with target name into target dir */
3488 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3489 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3490 ent->lde_name[1] == '.'))
3493 child = mdd_object_find(env, mdd, &ent->lde_fid);
3495 GOTO(out, rc = PTR_ERR(child));
3497 is_dir = S_ISDIR(mdd_object_type(child));
3499 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3501 /* Check whether the name has been inserted to the target */
3502 if (dt_try_as_dir(env, dt_tobj)) {
3503 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3505 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3506 (struct dt_key *)name,
3507 mdd_object_capa(env, mdd_tobj));
3508 if (unlikely(rc == 0))
3509 target_exist = true;
3512 handle = mdd_trans_create(env, mdd);
3514 GOTO(out, rc = PTR_ERR(handle));
3516 /* Note: this transaction is part of migration, and it is not
3517 * the last step of migration, so we set th_local = 1 to avoid
3518 * updating last rcvd for this transaction */
3519 handle->th_local = 1;
3520 if (likely(!target_exist)) {
3521 rc = mdo_declare_index_insert(env, mdd_tobj,
3523 mdd_object_type(child),
3529 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3535 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3540 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3544 /* Update .. for child */
3545 rc = mdo_declare_index_delete(env, child, dotdot,
3550 rc = mdo_declare_index_insert(env, child,
3551 mdd_object_fid(mdd_tobj),
3552 S_IFDIR, dotdot, handle);
3557 rc = mdd_linkea_declare_update_child(env, mdd_tobj,
3564 rc = mdd_trans_start(env, mdd, handle);
3566 CERROR("%s: transaction start failed: rc = %d\n",
3567 mdd2obd_dev(mdd)->obd_name, rc);
3571 if (likely(!target_exist)) {
3572 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3573 mdd_object_type(child),
3575 mdd_object_capa(env, mdd_tobj));
3580 rc = mdo_ref_add(env, mdd_tobj, handle);
3587 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle,
3588 mdd_object_capa(env, mdd_sobj));
3593 rc = __mdd_index_delete_only(env, child, dotdot, handle,
3594 mdd_object_capa(env, child));
3598 rc = __mdd_index_insert_only(env, child,
3599 mdd_object_fid(mdd_tobj), S_IFDIR,
3601 mdd_object_capa(env, child));
3606 rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
3607 strlen(name), handle);
3610 mdd_object_put(env, child);
3611 mdd_trans_stop(env, mdd, rc, handle);
3615 result = iops->next(env, it);
3616 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3617 GOTO(out, rc = -EINTR);
3619 if (result == -ESTALE)
3621 } while (result == 0);
3624 iops->fini(env, it);
3626 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3630 static int mdd_declare_update_linkea(const struct lu_env *env,
3631 struct mdd_object *mdd_pobj,
3632 struct mdd_object *mdd_sobj,
3633 struct mdd_object *mdd_tobj,
3634 struct thandle *handle,
3635 const struct lu_name *child_name)
3637 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3638 child_name, handle, 1);
3641 static int mdd_update_linkea(const struct lu_env *env,
3642 struct mdd_object *mdd_pobj,
3643 struct mdd_object *mdd_sobj,
3644 struct mdd_object *mdd_tobj,
3645 struct thandle *handle,
3646 const struct lu_name *child_name)
3648 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3649 child_name, handle, 0);
3652 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3653 struct mdd_object *mdd_pobj,
3654 struct mdd_object *mdd_sobj,
3655 struct mdd_object *mdd_tobj,
3656 const struct lu_name *lname,
3658 struct lu_attr *parent_la,
3659 struct thandle *handle)
3661 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3664 /* Revert IMMUTABLE flag */
3665 la_flag->la_valid = LA_FLAGS;
3666 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3667 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3668 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3672 /* delete entry from source dir */
3673 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3677 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3678 mdd_tobj, handle, lname);
3682 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3683 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3689 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3690 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3696 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3697 mdd_object_type(mdd_tobj),
3698 lname->ln_name, handle);
3702 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3703 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3708 /* delete old object */
3709 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3713 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3714 /* delete old object */
3715 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3718 /* set nlink to 0 */
3719 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3724 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3728 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3733 static int mdd_migrate_update_name(const struct lu_env *env,
3734 struct mdd_object *mdd_pobj,
3735 struct mdd_object *mdd_sobj,
3736 struct mdd_object *mdd_tobj,
3737 const struct lu_name *lname,
3740 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3741 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3742 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3743 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3744 struct thandle *handle;
3745 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3746 const char *name = lname->ln_name;
3750 /* update time for parent */
3751 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3752 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3753 p_la->la_valid = LA_CTIME;
3755 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3759 handle = mdd_trans_create(env, mdd);
3761 RETURN(PTR_ERR(handle));
3763 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3764 lname, so_attr, p_la, handle);
3766 /* If the migration can not be fit in one transaction, just
3767 * leave it in the original MDT */
3769 GOTO(stop_trans, rc = 0);
3771 GOTO(stop_trans, rc);
3774 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3775 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3776 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3777 PFID(mdd_object_fid(mdd_tobj)));
3779 rc = mdd_trans_start(env, mdd, handle);
3781 GOTO(stop_trans, rc);
3783 /* Revert IMMUTABLE flag */
3784 la_flag->la_valid = LA_FLAGS;
3785 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3786 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3787 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3788 mdd_object_capa(env, mdd_pobj));
3790 GOTO(stop_trans, rc);
3792 /* Remove source name from source directory */
3793 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
3794 mdd_object_capa(env, mdd_pobj));
3796 GOTO(stop_trans, rc);
3798 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3801 GOTO(stop_trans, rc);
3803 if (S_ISREG(so_attr->la_mode)) {
3804 if (so_attr->la_nlink == 1) {
3805 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3807 mdd_object_capa(env, mdd_sobj));
3808 if (rc != 0 && rc != -ENODATA)
3809 GOTO(stop_trans, rc);
3813 /* Insert new fid with target name into target dir */
3814 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
3815 mdd_object_type(mdd_tobj), name,
3816 handle, mdd_object_capa(env, mdd_pobj));
3818 GOTO(stop_trans, rc);
3820 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3823 GOTO(stop_trans, rc);
3825 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3826 mdo_ref_del(env, mdd_sobj, handle);
3828 mdo_ref_del(env, mdd_sobj, handle);
3830 /* Get the attr again after ref_del */
3831 rc = mdd_la_get(env, mdd_sobj, so_attr,
3832 mdd_object_capa(env, mdd_sobj));
3834 GOTO(out_unlock, rc);
3836 ma->ma_attr = *so_attr;
3837 ma->ma_valid |= MA_INODE;
3838 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3840 GOTO(out_unlock, rc);
3842 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3844 GOTO(out_unlock, rc);
3847 mdd_write_unlock(env, mdd_sobj);
3850 mdd_trans_stop(env, mdd, rc, handle);
3856 * Check whether we should migrate the file/dir
3858 * < 0 permission check failed or other error.
3859 * = 0 the file can be migrated.
3860 * > 0 the file does not need to be migrated, mostly
3861 * for multiple link file
3863 static int mdd_migrate_sanity_check(const struct lu_env *env,
3864 struct mdd_object *pobj,
3865 const struct lu_attr *pattr,
3866 struct mdd_object *sobj,
3867 struct lu_attr *sattr)
3869 struct mdd_thread_info *info = mdd_env_info(env);
3870 struct linkea_data *ldata = &info->mti_link_data;
3872 struct lu_buf *mgr_buf;
3878 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3879 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3880 if (mgr_buf->lb_buf == NULL)
3883 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV,
3884 mdd_object_capa(env, sobj));
3886 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3888 /* If the object has migrateEA, it means IMMUTE flag
3889 * is being set by previous migration process, so it
3890 * needs to override the IMMUTE flag, otherwise the
3891 * following sanity check will fail */
3892 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
3893 LMV_HASH_FLAG_MIGRATION) {
3894 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3896 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3897 sobj->mod_flags &= ~IMMUTE_OBJ;
3898 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3899 mdd2obd_dev(mdd)->obd_name,
3900 PFID(mdd_object_fid(sobj)));
3904 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3905 sobj, sattr, NULL, NULL);
3909 /* Then it will check if the file should be migrated. If the file
3910 * has mulitple links, we only need migrate the file if all of its
3911 * entries has been migrated to the remote MDT */
3912 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3915 rc = mdd_links_read(env, sobj, ldata);
3917 /* For multiple links files, if there are no linkEA data at all,
3918 * means the file might be created before linkEA is enabled, and
3919 * all all of its links should not be migrated yet, otherwise
3920 * it should have some linkEA there */
3921 if (rc == -ENOENT || rc == -ENODATA)
3926 /* If it is mulitple links file, we need update the name entry for
3928 LASSERT(ldata->ld_leh != NULL);
3929 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3930 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3931 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3932 struct mdd_object *lpobj;
3933 struct lu_name lname;
3936 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3938 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3940 lpobj = mdd_object_find(env, mdd, &fid);
3941 if (IS_ERR(lpobj)) {
3942 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3943 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3948 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
3949 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
3950 PFID(&fid), lname.ln_namelen, lname.ln_name);
3951 mdd_object_put(env, lpobj);
3955 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
3956 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
3957 lname.ln_name, PFID(&fid));
3958 mdd_object_put(env, lpobj);
3966 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
3967 struct md_object *sobj, const struct lu_name *lname,
3968 struct md_object *tobj, struct md_attr *ma)
3970 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
3971 struct mdd_device *mdd = mdo2mdd(pobj);
3972 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
3973 struct mdd_object *mdd_tobj = NULL;
3974 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3975 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3979 /* If the file will being migrated, it will check whether
3980 * the file is being opened by someone else right now */
3981 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
3982 if (mdd_sobj->mod_count >= 1) {
3983 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
3984 mdd2obd_dev(mdd)->obd_name,
3985 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3986 mdd_sobj->mod_count, -EBUSY);
3987 mdd_read_unlock(env, mdd_sobj);
3988 GOTO(put, rc = -EBUSY);
3990 mdd_read_unlock(env, mdd_sobj);
3992 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3996 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
4000 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4007 /* Sigh, it is impossible to finish all of migration in a single
4008 * transaction, for example migrating big directory entries to the
4009 * new MDT, it needs insert all of name entries of children in the
4012 * So migration will be done in multiple steps and transactions.
4014 * 1. create an orphan object on the remote MDT in one transaction.
4015 * 2. migrate extend attributes to the new target file/directory.
4016 * 3. For directory, migrate the entries to the new MDT and update
4017 * linkEA of each children. Because we can not migrate all entries
4018 * in a single transaction, so the migrating directory will become
4019 * a striped directory during migration, so once the process is
4020 * interrupted, the directory is still accessible. (During lookup,
4021 * client will locate the name by searching both original and target
4023 * 4. Finally, update the name/FID to point to the new file/directory
4024 * in a separate transaction.
4027 /* step 1: Check whether the orphan object has been created, and create
4028 * orphan object on the remote MDT if needed */
4029 mdd_tobj = md2mdd_obj(tobj);
4030 if (!mdd_object_exists(mdd_tobj)) {
4031 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4037 /* step 2: migrate xattr */
4038 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4042 /* step 3: migrate name entries to the orphan object */
4043 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4044 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4047 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4048 OBD_FAIL_MDS_REINT_NET_REP)))
4052 /* step 4: update name entry to the new object */
4053 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4061 const struct md_dir_operations mdd_dir_ops = {
4062 .mdo_is_subdir = mdd_is_subdir,
4063 .mdo_lookup = mdd_lookup,
4064 .mdo_create = mdd_create,
4065 .mdo_rename = mdd_rename,
4066 .mdo_link = mdd_link,
4067 .mdo_unlink = mdd_unlink,
4068 .mdo_create_data = mdd_create_data,
4069 .mdo_migrate = mdd_migrate,