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 int mdd_declare_changelog_store(const struct lu_env *env,
647 struct mdd_device *mdd,
648 const struct lu_name *fname,
649 struct thandle *handle)
651 struct obd_device *obd = mdd2obd_dev(mdd);
652 struct llog_ctxt *ctxt;
653 struct llog_changelog_rec *rec;
659 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
662 reclen = llog_data_len(sizeof(*rec) +
663 (fname != NULL ? fname->ln_namelen : 0));
664 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
665 if (buf->lb_buf == NULL)
669 rec->cr_hdr.lrh_len = reclen;
670 rec->cr_hdr.lrh_type = CHANGELOG_REC;
672 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
676 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
682 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
683 struct mdd_device *mdd,
684 const struct lu_name *tname,
685 const struct lu_name *sname,
686 struct thandle *handle)
688 struct obd_device *obd = mdd2obd_dev(mdd);
689 struct llog_ctxt *ctxt;
690 struct llog_changelog_ext_rec *rec;
696 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
699 reclen = llog_data_len(sizeof(*rec) +
700 (tname != NULL ? tname->ln_namelen : 0) +
701 (sname != NULL ? 1 + sname->ln_namelen : 0));
702 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
703 if (buf->lb_buf == NULL)
707 rec->cr_hdr.lrh_len = reclen;
708 rec->cr_hdr.lrh_type = CHANGELOG_REC;
710 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
714 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
720 /** Add a changelog entry \a rec to the changelog llog
723 * \param handle - currently ignored since llogs start their own transaction;
724 * this will hopefully be fixed in llog rewrite
727 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
728 struct llog_changelog_rec *rec, struct thandle *th)
730 struct obd_device *obd = mdd2obd_dev(mdd);
731 struct llog_ctxt *ctxt;
734 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
735 rec->cr_hdr.lrh_type = CHANGELOG_REC;
736 rec->cr.cr_time = cl_time();
738 spin_lock(&mdd->mdd_cl.mc_lock);
739 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
740 * but as long as the MDD transactions are ordered correctly for e.g.
741 * rename conflicts, I don't think this should matter. */
742 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
743 spin_unlock(&mdd->mdd_cl.mc_lock);
745 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
749 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
756 /** Add a changelog_ext entry \a rec to the changelog llog
759 * \param handle - currently ignored since llogs start their own transaction;
760 * this will hopefully be fixed in llog rewrite
764 mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd,
765 struct llog_changelog_ext_rec *rec, struct thandle *th)
767 struct obd_device *obd = mdd2obd_dev(mdd);
768 struct llog_ctxt *ctxt;
771 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
772 /* llog_lvfs_write_rec sets the llog tail len */
773 rec->cr_hdr.lrh_type = CHANGELOG_REC;
774 rec->cr.cr_time = cl_time();
776 spin_lock(&mdd->mdd_cl.mc_lock);
777 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
778 * but as long as the MDD transactions are ordered correctly for e.g.
779 * rename conflicts, I don't think this should matter. */
780 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
781 spin_unlock(&mdd->mdd_cl.mc_lock);
783 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
787 /* nested journal transaction */
788 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
796 /** Store a namespace change changelog record
797 * If this fails, we must fail the whole transaction; we don't
798 * want the change to commit without the log entry.
799 * \param target - mdd_object of change
800 * \param parent - parent dir/object
801 * \param tname - target name string
802 * \param handle - transacion handle
804 int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd,
805 enum changelog_rec_type type, unsigned flags,
806 struct mdd_object *target, struct mdd_object *parent,
807 const struct lu_name *tname, struct thandle *handle)
809 struct llog_changelog_rec *rec;
816 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
818 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
821 LASSERT(target != NULL);
822 LASSERT(parent != NULL);
823 LASSERT(tname != NULL);
824 LASSERT(handle != NULL);
826 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
827 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
828 if (buf->lb_buf == NULL)
832 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
833 rec->cr.cr_type = (__u32)type;
834 rec->cr.cr_tfid = *mdo2fid(target);
835 rec->cr.cr_pfid = *mdo2fid(parent);
836 rec->cr.cr_namelen = tname->ln_namelen;
837 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
839 target->mod_cltime = cfs_time_current_64();
841 rc = mdd_changelog_store(env, mdd, rec, handle);
843 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
844 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
845 PFID(&rec->cr.cr_pfid));
853 /** Store a namespace change changelog record
854 * If this fails, we must fail the whole transaction; we don't
855 * want the change to commit without the log entry.
856 * \param target - mdd_object of change
857 * \param tpfid - target parent dir/object fid
858 * \param sfid - source object fid
859 * \param spfid - source parent fid
860 * \param tname - target name string
861 * \param sname - source name string
862 * \param handle - transacion handle
864 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
865 struct mdd_device *mdd,
866 enum changelog_rec_type type,
868 struct mdd_object *target,
869 const struct lu_fid *tpfid,
870 const struct lu_fid *sfid,
871 const struct lu_fid *spfid,
872 const struct lu_name *tname,
873 const struct lu_name *sname,
874 struct thandle *handle)
876 struct llog_changelog_ext_rec *rec;
883 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
885 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
888 LASSERT(sfid != NULL);
889 LASSERT(tpfid != NULL);
890 LASSERT(tname != NULL);
891 LASSERT(handle != NULL);
893 reclen = llog_data_len(sizeof(*rec) +
894 sname != NULL ? 1 + sname->ln_namelen : 0);
895 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
896 if (buf->lb_buf == NULL)
900 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
901 rec->cr.cr_type = (__u32)type;
902 rec->cr.cr_pfid = *tpfid;
903 rec->cr.cr_sfid = *sfid;
904 rec->cr.cr_spfid = *spfid;
905 rec->cr.cr_namelen = tname->ln_namelen;
906 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
908 rec->cr.cr_name[tname->ln_namelen] = '\0';
909 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
911 rec->cr.cr_namelen += 1 + sname->ln_namelen;
914 if (likely(target != NULL)) {
915 rec->cr.cr_tfid = *mdo2fid(target);
916 target->mod_cltime = cfs_time_current_64();
918 fid_zero(&rec->cr.cr_tfid);
921 rc = mdd_changelog_ext_store(env, mdd, rec, handle);
923 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
924 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
931 static int __mdd_links_add(const struct lu_env *env,
932 struct mdd_object *mdd_obj,
933 struct linkea_data *ldata,
934 const struct lu_name *lname,
935 const struct lu_fid *pfid,
936 int first, int check)
940 if (ldata->ld_leh == NULL) {
941 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
945 rc = linkea_data_new(ldata,
946 &mdd_env_info(env)->mti_link_buf);
953 rc = linkea_links_find(ldata, lname, pfid);
954 if (rc && rc != -ENOENT)
960 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
961 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
965 linkea_add_buf(ldata, lname, tfid);
968 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
969 linkea_add_buf(ldata, lname, pfid);
971 return linkea_add_buf(ldata, lname, pfid);
974 static int __mdd_links_del(const struct lu_env *env,
975 struct mdd_object *mdd_obj,
976 struct linkea_data *ldata,
977 const struct lu_name *lname,
978 const struct lu_fid *pfid)
982 if (ldata->ld_leh == NULL) {
983 rc = mdd_links_read(env, mdd_obj, ldata);
988 rc = linkea_links_find(ldata, lname, pfid);
992 linkea_del_buf(ldata, lname);
996 static int mdd_linkea_prepare(const struct lu_env *env,
997 struct mdd_object *mdd_obj,
998 const struct lu_fid *oldpfid,
999 const struct lu_name *oldlname,
1000 const struct lu_fid *newpfid,
1001 const struct lu_name *newlname,
1002 int first, int check,
1003 struct linkea_data *ldata)
1009 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1012 LASSERT(oldpfid != NULL || newpfid != NULL);
1014 if (mdd_obj->mod_flags & DEAD_OBJ) {
1015 /* Prevent linkea to be updated which is NOT necessary. */
1016 ldata->ld_reclen = 0;
1017 /* No more links, don't bother */
1021 if (oldpfid != NULL) {
1022 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1025 (rc != -ENODATA && rc != -ENOENT))
1027 /* No changes done. */
1032 /* If renaming, add the new record */
1033 if (newpfid != NULL) {
1034 /* even if the add fails, we still delete the out-of-date
1036 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1040 rc = rc != 0 ? rc : rc2;
1045 int mdd_links_rename(const struct lu_env *env,
1046 struct mdd_object *mdd_obj,
1047 const struct lu_fid *oldpfid,
1048 const struct lu_name *oldlname,
1049 const struct lu_fid *newpfid,
1050 const struct lu_name *newlname,
1051 struct thandle *handle,
1052 struct linkea_data *ldata,
1053 int first, int check)
1058 if (ldata == NULL) {
1059 ldata = &mdd_env_info(env)->mti_link_data;
1060 memset(ldata, 0, sizeof(*ldata));
1061 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1062 newpfid, newlname, first, check,
1068 if (ldata->ld_reclen != 0)
1069 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1074 if (rc == -EOVERFLOW || rc == -ENOSPC)
1076 if (oldpfid == NULL)
1077 CDEBUG(error ? D_ERROR : D_OTHER,
1078 "link_ea add '%.*s' failed %d "DFID"\n",
1079 newlname->ln_namelen, newlname->ln_name,
1080 rc, PFID(mdd_object_fid(mdd_obj)));
1081 else if (newpfid == NULL)
1082 CDEBUG(error ? D_ERROR : D_OTHER,
1083 "link_ea del '%.*s' failed %d "DFID"\n",
1084 oldlname->ln_namelen, oldlname->ln_name,
1085 rc, PFID(mdd_object_fid(mdd_obj)));
1087 CDEBUG(error ? D_ERROR : D_OTHER,
1088 "link_ea rename '%.*s'->'%.*s' failed %d "
1090 oldlname->ln_namelen, oldlname->ln_name,
1091 newlname->ln_namelen, newlname->ln_name,
1092 rc, PFID(mdd_object_fid(mdd_obj)));
1095 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1096 /* if we vmalloced a large buffer drop it */
1097 lu_buf_free(ldata->ld_buf);
1102 static inline int mdd_links_add(const struct lu_env *env,
1103 struct mdd_object *mdd_obj,
1104 const struct lu_fid *pfid,
1105 const struct lu_name *lname,
1106 struct thandle *handle,
1107 struct linkea_data *ldata, int first)
1109 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1110 pfid, lname, handle, ldata, first, 0);
1113 static inline int mdd_links_del(const struct lu_env *env,
1114 struct mdd_object *mdd_obj,
1115 const struct lu_fid *pfid,
1116 const struct lu_name *lname,
1117 struct thandle *handle)
1119 return mdd_links_rename(env, mdd_obj, pfid, lname,
1120 NULL, NULL, handle, NULL, 0, 0);
1123 /** Read the link EA into a temp buffer.
1124 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1125 * A pointer to the buffer is stored in \a ldata::ld_buf.
1127 * \retval 0 or error
1129 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1130 struct linkea_data *ldata)
1134 if (!mdd_object_exists(mdd_obj))
1137 /* First try a small buf */
1138 LASSERT(env != NULL);
1139 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1141 if (ldata->ld_buf->lb_buf == NULL)
1144 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1146 if (rc == -ERANGE) {
1147 /* Buf was too small, figure out what we need. */
1148 lu_buf_free(ldata->ld_buf);
1149 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1150 XATTR_NAME_LINK, BYPASS_CAPA);
1153 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1154 if (ldata->ld_buf->lb_buf == NULL)
1156 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1157 XATTR_NAME_LINK, BYPASS_CAPA);
1160 lu_buf_free(ldata->ld_buf);
1161 ldata->ld_buf = NULL;
1165 return linkea_init(ldata);
1168 /** Read the link EA into a temp buffer.
1169 * Uses the name_buf since it is generally large.
1170 * \retval IS_ERR err
1171 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1173 struct lu_buf *mdd_links_get(const struct lu_env *env,
1174 struct mdd_object *mdd_obj)
1176 struct linkea_data ldata = { 0 };
1179 rc = mdd_links_read(env, mdd_obj, &ldata);
1180 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1183 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1184 struct linkea_data *ldata, struct thandle *handle)
1186 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1187 ldata->ld_leh->leh_len);
1189 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1192 return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1193 mdd_object_capa(env, mdd_obj));
1196 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1197 struct thandle *handle, struct linkea_data *ldata)
1203 if (ldata != NULL && ldata->ld_leh != NULL) {
1204 ea_len = ldata->ld_leh->leh_len;
1205 linkea = ldata->ld_buf->lb_buf;
1207 ea_len = DEFAULT_LINKEA_SIZE;
1211 /* XXX: max size? */
1212 rc = mdo_declare_xattr_set(env, mdd_obj,
1213 mdd_buf_get_const(env, linkea, ea_len),
1214 XATTR_NAME_LINK, 0, handle);
1218 static inline int mdd_declare_links_del(const struct lu_env *env,
1219 struct mdd_object *c,
1220 struct thandle *handle)
1224 /* For directory, the linkEA will be removed together
1225 * with the object. */
1226 if (!S_ISDIR(mdd_object_type(c)))
1227 rc = mdd_declare_links_add(env, c, handle, NULL);
1232 static int mdd_declare_link(const struct lu_env *env,
1233 struct mdd_device *mdd,
1234 struct mdd_object *p,
1235 struct mdd_object *c,
1236 const struct lu_name *name,
1237 struct thandle *handle,
1239 struct linkea_data *data)
1243 rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
1244 name->ln_name, handle);
1248 rc = mdo_declare_ref_add(env, c, handle);
1252 la->la_valid = LA_CTIME | LA_MTIME;
1253 rc = mdo_declare_attr_set(env, p, la, handle);
1257 la->la_valid = LA_CTIME;
1258 rc = mdo_declare_attr_set(env, c, la, handle);
1262 rc = mdd_declare_links_add(env, c, handle, data);
1266 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1271 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1272 struct md_object *src_obj, const struct lu_name *lname,
1275 const char *name = lname->ln_name;
1276 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1277 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1278 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1279 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1280 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1281 struct mdd_device *mdd = mdo2mdd(src_obj);
1282 struct thandle *handle;
1283 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1287 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
1291 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
1295 handle = mdd_trans_create(env, mdd);
1297 GOTO(out_pending, rc = PTR_ERR(handle));
1299 memset(ldata, 0, sizeof(*ldata));
1301 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1302 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1304 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1309 rc = mdd_trans_start(env, mdd, handle);
1313 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1314 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1317 GOTO(out_unlock, rc);
1319 rc = mdo_ref_add(env, mdd_sobj, handle);
1321 GOTO(out_unlock, rc);
1324 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1325 mdd_object_type(mdd_sobj), name, handle,
1326 mdd_object_capa(env, mdd_tobj));
1328 mdo_ref_del(env, mdd_sobj, handle);
1329 GOTO(out_unlock, rc);
1332 la->la_valid = LA_CTIME | LA_MTIME;
1333 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1335 GOTO(out_unlock, rc);
1337 la->la_valid = LA_CTIME;
1338 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1340 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1341 mdo2fid(mdd_tobj), lname, 0, 0,
1344 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1345 lname, handle, ldata, 0);
1346 /* The failure of links_add should not cause the link
1347 * failure, reset rc here */
1352 mdd_write_unlock(env, mdd_sobj);
1354 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1355 mdd_tobj, lname, handle);
1357 mdd_trans_stop(env, mdd, rc, handle);
1359 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1360 /* if we vmalloced a large buffer drop it */
1361 lu_buf_free(ldata->ld_buf);
1366 static int mdd_mark_dead_object(const struct lu_env *env,
1367 struct mdd_object *obj, struct thandle *handle,
1370 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1374 obj->mod_flags |= DEAD_OBJ;
1376 if (!S_ISDIR(mdd_object_type(obj)))
1379 attr->la_valid = LA_FLAGS;
1380 attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
1383 rc = mdo_declare_attr_set(env, obj, attr, handle);
1385 rc = mdo_attr_set(env, obj, attr, handle,
1386 mdd_object_capa(env, obj));
1391 static int mdd_declare_finish_unlink(const struct lu_env *env,
1392 struct mdd_object *obj,
1393 struct thandle *handle)
1397 rc = mdd_mark_dead_object(env, obj, handle, true);
1401 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1405 rc = mdo_declare_destroy(env, obj, handle);
1409 return mdd_declare_links_del(env, obj, handle);
1412 /* caller should take a lock before calling */
1413 int mdd_finish_unlink(const struct lu_env *env,
1414 struct mdd_object *obj, struct md_attr *ma,
1415 const struct mdd_object *pobj,
1416 const struct lu_name *lname,
1420 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1423 LASSERT(mdd_write_locked(env, obj) != 0);
1425 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1426 rc = mdd_mark_dead_object(env, obj, th, false);
1430 /* add new orphan and the object
1431 * will be deleted during mdd_close() */
1432 if (obj->mod_count) {
1433 rc = __mdd_orphan_add(env, obj, th);
1435 CDEBUG(D_HA, "Object "DFID" is inserted into "
1436 "orphan list, open count = %d\n",
1437 PFID(mdd_object_fid(obj)),
1440 CERROR("Object "DFID" fail to be an orphan, "
1441 "open count = %d, maybe cause failed "
1443 PFID(mdd_object_fid(obj)),
1446 rc = mdo_destroy(env, obj, th);
1448 } else if (!is_dir) {
1449 /* old files may not have link ea; ignore errors */
1450 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1458 * has mdd_write_lock on cobj already, but not on pobj yet
1460 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1461 const struct lu_attr *pattr,
1462 struct mdd_object *cobj,
1463 const struct lu_attr *cattr)
1468 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1473 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1474 struct mdd_object *p, struct mdd_object *c,
1475 const struct lu_name *name, struct md_attr *ma,
1476 struct thandle *handle, int no_name)
1478 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1481 if (likely(no_name == 0)) {
1482 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1487 rc = mdo_declare_ref_del(env, p, handle);
1491 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1492 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1493 la->la_valid = LA_CTIME | LA_MTIME;
1494 rc = mdo_declare_attr_set(env, p, la, handle);
1499 rc = mdo_declare_ref_del(env, c, handle);
1503 rc = mdo_declare_ref_del(env, c, handle);
1507 la->la_valid = LA_CTIME;
1508 rc = mdo_declare_attr_set(env, c, la, handle);
1512 rc = mdd_declare_finish_unlink(env, c, handle);
1516 /* FIXME: need changelog for remove entry */
1517 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1524 * test if a file has an HSM archive
1525 * if HSM attributes are not found in ma update them from
1528 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1529 struct mdd_object *obj,
1534 if (!(ma->ma_valid & MA_HSM)) {
1535 /* no HSM MD provided, read xattr */
1536 struct lu_buf *hsm_buf;
1537 const size_t buflen = sizeof(struct hsm_attrs);
1540 hsm_buf = mdd_buf_get(env, NULL, 0);
1541 lu_buf_alloc(hsm_buf, buflen);
1542 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM,
1543 mdd_object_capa(env, obj));
1544 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1545 lu_buf_free(hsm_buf);
1549 ma->ma_valid = MA_HSM;
1551 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1557 * Delete name entry and the object.
1558 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1559 * does not exist for this object, and it could only happen during resending
1560 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1561 * is also needed in this case(needed by changelog), so we have to add another
1562 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1563 * the ENOENT failure should be able to be fixed by redo mechanism.
1565 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1566 struct md_object *cobj, const struct lu_name *lname,
1567 struct md_attr *ma, int no_name)
1569 const char *name = lname->ln_name;
1570 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1571 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1572 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1573 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1574 struct mdd_object *mdd_cobj = NULL;
1575 struct mdd_device *mdd = mdo2mdd(pobj);
1576 struct thandle *handle;
1580 /* cobj == NULL means only delete name entry */
1581 if (likely(cobj != NULL)) {
1582 mdd_cobj = md2mdd_obj(cobj);
1583 if (mdd_object_exists(mdd_cobj) == 0)
1587 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1591 if (likely(mdd_cobj != NULL)) {
1593 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1597 is_dir = S_ISDIR(cattr->la_mode);
1600 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1604 handle = mdd_trans_create(env, mdd);
1606 RETURN(PTR_ERR(handle));
1608 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1609 lname, ma, handle, no_name);
1613 rc = mdd_trans_start(env, mdd, handle);
1617 if (likely(mdd_cobj != NULL))
1618 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1620 if (likely(no_name == 0)) {
1621 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1622 mdd_object_capa(env, mdd_pobj));
1627 if (likely(mdd_cobj != NULL)) {
1628 rc = mdo_ref_del(env, mdd_cobj, handle);
1630 __mdd_index_insert_only(env, mdd_pobj,
1632 mdd_object_type(mdd_cobj),
1634 mdd_object_capa(env, mdd_pobj));
1640 mdo_ref_del(env, mdd_cobj, handle);
1642 /* fetch updated nlink */
1643 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1648 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1649 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1651 la->la_valid = LA_CTIME | LA_MTIME;
1652 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1656 /* Enough for only unlink the entry */
1657 if (unlikely(mdd_cobj == NULL))
1660 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1661 /* update ctime of an unlinked file only if it is still
1662 * opened or a link still exists */
1663 la->la_valid = LA_CTIME;
1664 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1669 /* XXX: this transfer to ma will be removed with LOD/OSP */
1670 ma->ma_attr = *cattr;
1671 ma->ma_valid |= MA_INODE;
1672 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1674 /* fetch updated nlink */
1676 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1678 /* if object is removed then we can't get its attrs, use last get */
1679 if (cattr->la_nlink == 0) {
1680 ma->ma_attr = *cattr;
1681 ma->ma_valid |= MA_INODE;
1685 if (likely(mdd_cobj != NULL))
1686 mdd_write_unlock(env, mdd_cobj);
1691 if (cattr->la_nlink == 0) {
1692 cl_flags |= CLF_UNLINK_LAST;
1693 /* search for an existing archive */
1694 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1695 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1698 rc = mdd_changelog_ns_store(env, mdd,
1699 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1700 mdd_cobj, mdd_pobj, lname, handle);
1704 mdd_trans_stop(env, mdd, rc, handle);
1710 * The permission has been checked when obj created, no need check again.
1712 static int mdd_cd_sanity_check(const struct lu_env *env,
1713 struct mdd_object *obj)
1718 if (!obj || mdd_is_dead_obj(obj))
1724 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1725 struct md_object *cobj, const struct md_op_spec *spec,
1728 struct mdd_device *mdd = mdo2mdd(cobj);
1729 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1730 struct mdd_object *son = md2mdd_obj(cobj);
1731 struct thandle *handle;
1732 const struct lu_buf *buf;
1733 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1734 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1738 rc = mdd_cd_sanity_check(env, son);
1742 if (!md_should_create(spec->sp_cr_flags))
1746 * there are following use cases for this function:
1747 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1748 * striping can be specified or not
1751 rc = mdd_la_get(env, son, attr, BYPASS_CAPA);
1755 /* calling ->ah_make_hint() is used to transfer information from parent */
1756 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1758 handle = mdd_trans_create(env, mdd);
1760 GOTO(out_free, rc = PTR_ERR(handle));
1763 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1764 * Should this be fixed?
1766 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1767 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1768 spec->sp_cr_flags, spec->no_create);
1770 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1771 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1772 spec->u.sp_ea.eadatalen);
1777 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1778 XATTR_NAME_LOV, 0, handle);
1782 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1786 rc = mdd_trans_start(env, mdd, handle);
1790 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1791 0, handle, mdd_object_capa(env, son));
1796 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1799 mdd_trans_stop(env, mdd, rc, handle);
1804 static int mdd_declare_object_initialize(const struct lu_env *env,
1805 struct mdd_object *parent,
1806 struct mdd_object *child,
1807 struct lu_attr *attr,
1808 struct thandle *handle)
1814 * inode mode has been set in creation time, and it's based on umask,
1815 * la_mode and acl, don't set here again! (which will go wrong
1816 * because below function doesn't consider umask).
1817 * I'd suggest set all object attributes in creation time, see above.
1819 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1820 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1821 rc = mdo_declare_attr_set(env, child, attr, handle);
1822 attr->la_valid |= LA_MODE | LA_TYPE;
1823 if (rc != 0 || !S_ISDIR(attr->la_mode))
1826 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1831 rc = mdo_declare_ref_add(env, child, handle);
1835 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1841 static int mdd_object_initialize(const struct lu_env *env,
1842 const struct lu_fid *pfid,
1843 struct mdd_object *child,
1844 struct lu_attr *attr, struct thandle *handle,
1845 const struct md_op_spec *spec)
1851 * Update attributes for child.
1854 * (1) the valid bits should be converted between Lustre and Linux;
1855 * (2) maybe, the child attributes should be set in OSD when creation.
1858 rc = mdd_attr_set_internal(env, child, attr, handle, 0);
1859 /* arguments are supposed to stay the same */
1860 if (S_ISDIR(attr->la_mode)) {
1861 /* Add "." and ".." for newly created dir */
1862 mdo_ref_add(env, child, handle);
1863 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1864 S_IFDIR, dot, handle, BYPASS_CAPA);
1866 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1870 mdo_ref_del(env, child, handle);
1877 * This function checks whether it can create a file/dir under the
1878 * directory(@pobj). The directory(@pobj) is not being locked by
1881 * \param[in] env execution environment
1882 * \param[in] pobj the directory to create files
1883 * \param[in] pattr the attributes of the directory
1884 * \param[in] lname the name of the created file/dir
1885 * \param[in] cattr the attributes of the file/dir
1886 * \param[in] spec create specification
1888 * \retval = 0 it is allowed to create file/dir under
1890 * \retval negative error not allowed to create file/dir
1891 * under the directory
1893 static int mdd_create_sanity_check(const struct lu_env *env,
1894 struct md_object *pobj,
1895 const struct lu_attr *pattr,
1896 const struct lu_name *lname,
1897 struct lu_attr *cattr,
1898 struct md_op_spec *spec)
1900 struct mdd_thread_info *info = mdd_env_info(env);
1901 struct lu_fid *fid = &info->mti_fid;
1902 struct mdd_object *obj = md2mdd_obj(pobj);
1903 struct mdd_device *m = mdo2mdd(pobj);
1904 bool check_perm = true;
1909 if (mdd_is_dead_obj(obj))
1913 * In some cases this lookup is not needed - we know before if name
1914 * exists or not because MDT performs lookup for it.
1915 * name length check is done in lookup.
1917 if (spec->sp_cr_lookup) {
1919 * Check if the name already exist, though it will be checked in
1920 * _index_insert also, for avoiding rolling back if exists
1923 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1924 MAY_WRITE | MAY_EXEC);
1926 RETURN(rc ? : -EEXIST);
1928 /* Permission is already being checked in mdd_lookup */
1932 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1937 if (pattr->la_mode & S_ISGID) {
1938 cattr->la_gid = pattr->la_gid;
1939 if (S_ISDIR(cattr->la_mode)) {
1940 cattr->la_mode |= S_ISGID;
1941 cattr->la_valid |= LA_MODE;
1945 rc = mdd_name_check(m, lname);
1949 switch (cattr->la_mode & S_IFMT) {
1951 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1953 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1954 RETURN(-ENAMETOOLONG);
1973 static int mdd_declare_object_create(const struct lu_env *env,
1974 struct mdd_device *mdd,
1975 struct mdd_object *p, struct mdd_object *c,
1976 struct lu_attr *attr,
1977 struct thandle *handle,
1978 const struct md_op_spec *spec,
1979 struct lu_buf *def_acl_buf,
1980 struct lu_buf *acl_buf,
1981 struct dt_allocation_hint *hint)
1985 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
1990 #ifdef CONFIG_FS_POSIX_ACL
1991 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1992 /* if dir, then can inherit default ACl */
1993 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1994 XATTR_NAME_ACL_DEFAULT,
2000 if (acl_buf->lb_len > 0) {
2001 rc = mdo_declare_attr_set(env, c, attr, handle);
2005 rc = mdo_declare_xattr_set(env, c, acl_buf,
2006 XATTR_NAME_ACL_ACCESS, 0, handle);
2011 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2015 /* replay case, create LOV EA from client data */
2016 if (spec->no_create ||
2017 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2018 const struct lu_buf *buf;
2020 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2021 spec->u.sp_ea.eadatalen);
2022 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2028 if (S_ISLNK(attr->la_mode)) {
2029 const char *target_name = spec->u.sp_symname;
2030 int sym_len = strlen(target_name);
2031 const struct lu_buf *buf;
2033 buf = mdd_buf_get_const(env, target_name, sym_len);
2034 rc = dt_declare_record_write(env, mdd_object_child(c),
2043 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2044 struct mdd_object *p, struct mdd_object *c,
2045 const struct lu_name *name,
2046 struct lu_attr *attr,
2047 struct thandle *handle,
2048 const struct md_op_spec *spec,
2049 struct linkea_data *ldata,
2050 struct lu_buf *def_acl_buf,
2051 struct lu_buf *acl_buf,
2052 struct dt_allocation_hint *hint)
2056 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2057 def_acl_buf, acl_buf, hint);
2061 if (S_ISDIR(attr->la_mode)) {
2062 rc = mdo_declare_ref_add(env, p, handle);
2067 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2068 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2072 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2074 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2075 name->ln_name, handle);
2079 rc = mdd_declare_links_add(env, c, handle, ldata);
2084 la->la_valid = LA_CTIME | LA_MTIME;
2085 rc = mdo_declare_attr_set(env, p, la, handle);
2089 rc = mdd_declare_changelog_store(env, mdd, name, handle);
2094 /* XXX: For remote create, it should indicate the remote RPC
2095 * will be sent after local transaction is finished, which
2096 * is not very nice, but it will be removed once we fully support
2098 if (mdd_object_remote(p) && handle->th_update != NULL)
2099 handle->th_update->tu_sent_after_local_trans = 1;
2104 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2105 struct lu_attr *la, struct lu_buf *def_acl_buf,
2106 struct lu_buf *acl_buf)
2111 if (S_ISLNK(la->la_mode)) {
2112 acl_buf->lb_len = 0;
2113 def_acl_buf->lb_len = 0;
2117 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2118 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2119 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
2120 mdd_read_unlock(env, pobj);
2122 /* If there are default ACL, fix mode/ACL by default ACL */
2123 def_acl_buf->lb_len = rc;
2124 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2125 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2126 acl_buf->lb_len = rc;
2127 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2130 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2131 /* If there are no default ACL, fix mode by mask */
2132 struct lu_ucred *uc = lu_ucred(env);
2134 /* The create triggered by MDT internal events, such as
2135 * LFSCK reset, will not contain valid "uc". */
2136 if (unlikely(uc != NULL))
2137 la->la_mode &= ~uc->uc_umask;
2139 acl_buf->lb_len = 0;
2140 def_acl_buf->lb_len = 0;
2147 * Create a metadata object and initialize it, set acl, xattr.
2149 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2150 struct mdd_object *son, struct lu_attr *attr,
2151 struct md_op_spec *spec, struct lu_buf *acl_buf,
2152 struct lu_buf *def_acl_buf,
2153 struct dt_allocation_hint *hint,
2154 struct thandle *handle)
2158 mdd_write_lock(env, son, MOR_TGT_CHILD);
2159 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2164 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2165 * created in declare phase, they also needs to be added to master
2166 * object as sub-directory entry. So it has to initialize the master
2167 * object, then set dir striped EA.(in mdo_xattr_set) */
2168 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2171 GOTO(err_destroy, rc);
2174 * in case of replay we just set LOVEA provided by the client
2175 * XXX: I think it would be interesting to try "old" way where
2176 * MDT calls this xattr_set(LOV) in a different transaction.
2177 * probably this way we code can be made better.
2180 /* During creation, there are only a few cases we need do xattr_set to
2182 * 1. regular file: see comments above.
2183 * 2. create striped directory with provided stripeEA.
2184 * 3. create striped directory because inherit default layout from the
2186 if (spec->no_create ||
2187 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2188 S_ISDIR(attr->la_mode)) {
2189 const struct lu_buf *buf;
2191 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2192 spec->u.sp_ea.eadatalen);
2193 rc = mdo_xattr_set(env, son, buf,
2194 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2196 handle, BYPASS_CAPA);
2198 GOTO(err_destroy, rc);
2201 #ifdef CONFIG_FS_POSIX_ACL
2202 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2203 S_ISDIR(attr->la_mode)) {
2204 /* set default acl */
2205 rc = mdo_xattr_set(env, son, def_acl_buf,
2206 XATTR_NAME_ACL_DEFAULT, 0,
2207 handle, BYPASS_CAPA);
2209 GOTO(err_destroy, rc);
2211 /* set its own acl */
2212 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2213 rc = mdo_xattr_set(env, son, acl_buf,
2214 XATTR_NAME_ACL_ACCESS,
2215 0, handle, BYPASS_CAPA);
2217 GOTO(err_destroy, rc);
2221 if (S_ISLNK(attr->la_mode)) {
2222 struct lu_ucred *uc = lu_ucred_assert(env);
2223 struct dt_object *dt = mdd_object_child(son);
2224 const char *target_name = spec->u.sp_symname;
2225 int sym_len = strlen(target_name);
2226 const struct lu_buf *buf;
2229 buf = mdd_buf_get_const(env, target_name, sym_len);
2230 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2231 mdd_object_capa(env, son),
2233 CFS_CAP_SYS_RESOURCE_MASK);
2238 GOTO(err_initlized, rc = -EFAULT);
2242 if (unlikely(rc != 0)) {
2244 if (S_ISDIR(attr->la_mode)) {
2245 /* Drop the reference, no need to delete "."/"..",
2246 * because the object to be destroied directly. */
2247 rc2 = mdo_ref_del(env, son, handle);
2251 rc2 = mdo_ref_del(env, son, handle);
2255 mdo_destroy(env, son, handle);
2258 mdd_write_unlock(env, son);
2263 * Create object and insert it into namespace.
2265 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2266 const struct lu_name *lname, struct md_object *child,
2267 struct md_op_spec *spec, struct md_attr* ma)
2269 struct mdd_thread_info *info = mdd_env_info(env);
2270 struct lu_attr *la = &info->mti_la_for_fix;
2271 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2272 struct mdd_object *son = md2mdd_obj(child);
2273 struct mdd_device *mdd = mdo2mdd(pobj);
2274 struct lu_attr *attr = &ma->ma_attr;
2275 struct thandle *handle;
2276 struct lu_attr *pattr = &info->mti_pattr;
2277 struct lu_buf acl_buf;
2278 struct lu_buf def_acl_buf;
2279 struct linkea_data *ldata = &info->mti_link_data;
2280 const char *name = lname->ln_name;
2281 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2287 * Two operations have to be performed:
2289 * - an allocation of a new object (->do_create()), and
2291 * - an insertion into a parent index (->dio_insert()).
2293 * Due to locking, operation order is not important, when both are
2294 * successful, *but* error handling cases are quite different:
2296 * - if insertion is done first, and following object creation fails,
2297 * insertion has to be rolled back, but this operation might fail
2298 * also leaving us with dangling index entry.
2300 * - if creation is done first, is has to be undone if insertion
2301 * fails, leaving us with leaked space, which is neither good, nor
2304 * It seems that creation-first is simplest solution, but it is
2305 * sub-optimal in the frequent
2310 * case, because second mkdir is bound to create object, only to
2311 * destroy it immediately.
2313 * To avoid this follow local file systems that do double lookup:
2315 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2317 * 1. create (mdd_object_create_internal())
2319 * 2. insert (__mdd_index_insert(), lookup again)
2322 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2326 /* Sanity checks before big job. */
2327 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2331 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2332 GOTO(out_free, rc = -EINPROGRESS);
2334 handle = mdd_trans_create(env, mdd);
2336 GOTO(out_free, rc = PTR_ERR(handle));
2338 acl_buf.lb_buf = info->mti_xattr_buf;
2339 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2340 def_acl_buf.lb_buf = info->mti_key;
2341 def_acl_buf.lb_len = sizeof(info->mti_key);
2342 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2346 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2348 memset(ldata, 0, sizeof(*ldata));
2349 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2350 mdd_object_fid(mdd_pobj),
2351 lname, 1, 0, ldata);
2353 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2354 handle, spec, ldata, &def_acl_buf, &acl_buf,
2359 rc = mdd_trans_start(env, mdd, handle);
2363 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2364 &def_acl_buf, hint, handle);
2368 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2369 mdd_write_lock(env, son, MOR_TGT_CHILD);
2370 rc = __mdd_orphan_add(env, son, handle);
2371 GOTO(out_volatile, rc);
2373 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2374 attr->la_mode, name, handle,
2375 mdd_object_capa(env, mdd_pobj));
2377 GOTO(err_created, rc);
2379 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2382 /* update parent directory mtime/ctime */
2384 la->la_valid = LA_CTIME | LA_MTIME;
2385 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2387 GOTO(err_insert, rc);
2395 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2396 rc2 = __mdd_orphan_del(env, son, handle);
2398 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2399 S_ISDIR(attr->la_mode),
2400 handle, BYPASS_CAPA);
2405 mdd_write_lock(env, son, MOR_TGT_CHILD);
2406 if (S_ISDIR(attr->la_mode)) {
2407 /* Drop the reference, no need to delete "."/"..",
2408 * because the object is to be destroyed directly. */
2409 rc2 = mdo_ref_del(env, son, handle);
2411 mdd_write_unlock(env, son);
2416 /* For volatile files drop one link immediately, since there is
2417 * no filename in the namespace, and save any error returned. */
2418 rc2 = mdo_ref_del(env, son, handle);
2420 mdd_write_unlock(env, son);
2421 if (unlikely(rc == 0))
2426 /* Don't destroy the volatile object on success */
2427 if (likely(rc != 0))
2428 mdo_destroy(env, son, handle);
2429 mdd_write_unlock(env, son);
2432 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2433 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2434 rc = mdd_changelog_ns_store(env, mdd,
2435 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2436 S_ISREG(attr->la_mode) ? CL_CREATE :
2437 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2438 0, son, mdd_pobj, lname, handle);
2440 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2444 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
2445 /* if we vmalloced a large buffer drop it */
2446 lu_buf_free(ldata->ld_buf);
2448 /* The child object shouldn't be cached anymore */
2450 set_bit(LU_OBJECT_HEARD_BANSHEE,
2451 &child->mo_lu.lo_header->loh_flags);
2456 * Get locks on parents in proper order
2457 * RETURN: < 0 - error, rename_order if successful
2465 static int mdd_rename_order(const struct lu_env *env,
2466 struct mdd_device *mdd,
2467 struct mdd_object *src_pobj,
2468 const struct lu_attr *pattr,
2469 struct mdd_object *tgt_pobj)
2471 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2475 if (src_pobj == tgt_pobj)
2476 RETURN(MDD_RN_SAME);
2478 /* compared the parent child relationship of src_p&tgt_p */
2479 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2481 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2484 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2498 /* has not mdd_write{read}_lock on any obj yet. */
2499 static int mdd_rename_sanity_check(const struct lu_env *env,
2500 struct mdd_object *src_pobj,
2501 const struct lu_attr *pattr,
2502 struct mdd_object *tgt_pobj,
2503 const struct lu_attr *tpattr,
2504 struct mdd_object *sobj,
2505 const struct lu_attr *cattr,
2506 struct mdd_object *tobj,
2507 const struct lu_attr *tattr)
2512 /* XXX: when get here, sobj must NOT be NULL,
2513 * the other case has been processed in cld_rename
2514 * before mdd_rename and enable MDS_PERM_BYPASS. */
2517 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2521 /* XXX: when get here, "tobj == NULL" means tobj must
2522 * NOT exist (neither on remote MDS, such case has been
2523 * processed in cld_rename before mdd_rename and enable
2525 * So check may_create, but not check may_unlink. */
2527 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2528 (src_pobj != tgt_pobj));
2530 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2531 (src_pobj != tgt_pobj), 1);
2533 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2534 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2539 static int mdd_declare_rename(const struct lu_env *env,
2540 struct mdd_device *mdd,
2541 struct mdd_object *mdd_spobj,
2542 struct mdd_object *mdd_tpobj,
2543 struct mdd_object *mdd_sobj,
2544 struct mdd_object *mdd_tobj,
2545 const struct lu_name *tname,
2546 const struct lu_name *sname,
2548 struct linkea_data *ldata,
2549 struct thandle *handle)
2551 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2554 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2555 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2561 /* name from source dir */
2562 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2566 /* .. from source child */
2567 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2568 /* source child can be directory,
2569 * counted by source dir's nlink */
2570 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2573 if (mdd_spobj != mdd_tpobj) {
2574 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2579 rc = mdo_declare_index_insert(env, mdd_sobj,
2581 S_IFDIR, dotdot, handle);
2586 /* new target child can be directory,
2587 * counted by target dir's nlink */
2588 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2593 la->la_valid = LA_CTIME | LA_MTIME;
2594 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2598 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2602 la->la_valid = LA_CTIME;
2603 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2607 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2612 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2613 mdd_object_type(mdd_sobj),
2614 tname->ln_name, handle);
2618 /* name from target dir (old name), we declare it unconditionally
2619 * as mdd_rename() calls delete unconditionally as well. so just
2620 * to balance declarations vs calls to change ... */
2621 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2625 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2626 /* delete target child in target parent directory */
2627 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2631 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2632 /* target child can be directory,
2633 * delete "." reference in target child directory */
2634 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2638 /* delete ".." reference in target parent directory */
2639 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2644 la->la_valid = LA_CTIME;
2645 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2649 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2654 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2661 /* src object can be remote that is why we use only fid and type of object */
2662 static int mdd_rename(const struct lu_env *env,
2663 struct md_object *src_pobj, struct md_object *tgt_pobj,
2664 const struct lu_fid *lf, const struct lu_name *lsname,
2665 struct md_object *tobj, const struct lu_name *ltname,
2668 const char *sname = lsname->ln_name;
2669 const char *tname = ltname->ln_name;
2670 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2671 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2672 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2673 struct mdd_device *mdd = mdo2mdd(src_pobj);
2674 struct mdd_object *mdd_sobj = NULL; /* source object */
2675 struct mdd_object *mdd_tobj = NULL;
2676 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2677 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2678 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2679 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2680 struct thandle *handle;
2681 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2682 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2683 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2686 bool tobj_locked = 0;
2687 unsigned cl_flags = 0;
2692 mdd_tobj = md2mdd_obj(tobj);
2694 mdd_sobj = mdd_object_find(env, mdd, lf);
2696 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
2698 GOTO(out_pending, rc);
2700 rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA);
2702 GOTO(out_pending, rc);
2705 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2707 GOTO(out_pending, rc);
2710 rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA);
2712 GOTO(out_pending, rc);
2714 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2715 mdd_sobj, cattr, mdd_tobj, tattr);
2717 GOTO(out_pending, rc);
2719 rc = mdd_name_check(mdd, ltname);
2721 GOTO(out_pending, rc);
2723 handle = mdd_trans_create(env, mdd);
2725 GOTO(out_pending, rc = PTR_ERR(handle));
2727 memset(ldata, 0, sizeof(*ldata));
2728 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2729 ltname, 1, 0, ldata);
2730 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2731 mdd_tobj, lsname, ltname, ma, ldata, handle);
2735 rc = mdd_trans_start(env, mdd, handle);
2739 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2740 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2742 GOTO(cleanup_unlocked, rc);
2744 is_dir = S_ISDIR(cattr->la_mode);
2746 /* Remove source name from source directory */
2747 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2748 mdd_object_capa(env, mdd_spobj));
2752 /* "mv dir1 dir2" needs "dir1/.." link update */
2753 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2754 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2755 mdd_object_capa(env, mdd_sobj));
2757 GOTO(fixup_spobj2, rc);
2759 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2761 mdd_object_capa(env, mdd_sobj));
2763 GOTO(fixup_spobj, rc);
2766 /* Remove target name from target directory
2767 * Here tobj can be remote one, so we do index_delete unconditionally
2768 * and -ENOENT is allowed.
2770 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2771 mdd_object_capa(env, mdd_tpobj));
2774 /* tname might been renamed to something else */
2775 GOTO(fixup_spobj, rc);
2778 GOTO(fixup_spobj, rc);
2781 /* Insert new fid with target name into target dir */
2782 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2783 tname, handle, mdd_object_capa(env, mdd_tpobj));
2785 GOTO(fixup_tpobj, rc);
2787 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2788 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2790 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2792 la->la_valid = LA_CTIME;
2793 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2795 GOTO(fixup_tpobj, rc);
2798 /* Remove old target object
2799 * For tobj is remote case cmm layer has processed
2800 * and set tobj to NULL then. So when tobj is NOT NULL,
2801 * it must be local one.
2803 if (tobj && mdd_object_exists(mdd_tobj)) {
2804 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2806 if (mdd_is_dead_obj(mdd_tobj)) {
2807 /* shld not be dead, something is wrong */
2808 CERROR("tobj is dead, something is wrong\n");
2812 mdo_ref_del(env, mdd_tobj, handle);
2814 /* Remove dot reference. */
2815 if (S_ISDIR(tattr->la_mode))
2816 mdo_ref_del(env, mdd_tobj, handle);
2819 /* fetch updated nlink */
2820 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2822 CERROR("%s: Failed to get nlink for tobj "
2824 mdd2obd_dev(mdd)->obd_name,
2825 PFID(tpobj_fid), rc);
2826 GOTO(fixup_tpobj, rc);
2829 la->la_valid = LA_CTIME;
2830 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2832 CERROR("%s: Failed to set ctime for tobj "
2834 mdd2obd_dev(mdd)->obd_name,
2835 PFID(tpobj_fid), rc);
2836 GOTO(fixup_tpobj, rc);
2839 /* XXX: this transfer to ma will be removed with LOD/OSP */
2840 ma->ma_attr = *tattr;
2841 ma->ma_valid |= MA_INODE;
2842 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2845 CERROR("%s: Failed to unlink tobj "
2847 mdd2obd_dev(mdd)->obd_name,
2848 PFID(tpobj_fid), rc);
2849 GOTO(fixup_tpobj, rc);
2852 /* fetch updated nlink */
2853 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2855 CERROR("%s: Failed to get nlink for tobj "
2857 mdd2obd_dev(mdd)->obd_name,
2858 PFID(tpobj_fid), rc);
2859 GOTO(fixup_tpobj, rc);
2861 /* XXX: this transfer to ma will be removed with LOD/OSP */
2862 ma->ma_attr = *tattr;
2863 ma->ma_valid |= MA_INODE;
2865 if (tattr->la_nlink == 0) {
2866 cl_flags |= CLF_RENAME_LAST;
2867 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2868 cl_flags |= CLF_RENAME_LAST_EXISTS;
2872 la->la_valid = LA_CTIME | LA_MTIME;
2873 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2875 GOTO(fixup_tpobj, rc);
2877 if (mdd_spobj != mdd_tpobj) {
2878 la->la_valid = LA_CTIME | LA_MTIME;
2879 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2882 if (rc == 0 && mdd_sobj) {
2883 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2884 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2885 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2888 /* Old files might not have EA entry */
2889 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2890 lsname, handle, NULL, 0);
2891 mdd_write_unlock(env, mdd_sobj);
2892 /* We don't fail the transaction if the link ea can't be
2893 updated -- fid2path will use alternate lookup method. */
2901 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2904 CWARN("tp obj fix error %d\n",rc2);
2906 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2907 !mdd_is_dead_obj(mdd_tobj)) {
2909 mdo_ref_add(env, mdd_tobj, handle);
2911 mdo_ref_add(env, mdd_tobj, handle);
2914 rc2 = __mdd_index_insert(env, mdd_tpobj,
2916 mdd_object_type(mdd_tobj),
2917 tname, handle, BYPASS_CAPA);
2919 CWARN("tp obj fix error: rc = %d\n", rc2);
2924 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2925 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2929 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2930 mdd2obd_dev(mdd)->obd_name, rc2);
2933 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
2934 dotdot, handle, BYPASS_CAPA);
2936 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2937 mdd2obd_dev(mdd)->obd_name, rc2);
2942 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
2943 mdd_object_type(mdd_sobj), sname,
2944 handle, BYPASS_CAPA);
2946 CWARN("sp obj fix error: rc = %d\n", rc2);
2951 mdd_write_unlock(env, mdd_tobj);
2955 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2956 mdd_tobj, tpobj_fid, lf,
2957 spobj_fid, ltname, lsname,
2961 mdd_trans_stop(env, mdd, rc, handle);
2964 mdd_object_put(env, mdd_sobj);
2969 * During migration once the parent FID has been changed,
2970 * we need update the parent FID in linkea.
2972 static int mdd_linkea_update_child_internal(const struct lu_env *env,
2973 struct mdd_object *parent,
2974 struct mdd_object *child,
2975 const char *name, int namelen,
2976 struct thandle *handle,
2979 struct mdd_thread_info *info = mdd_env_info(env);
2980 struct linkea_data ldata = {0};
2981 struct lu_buf *buf = &info->mti_link_buf;
2987 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2988 if (buf->lb_buf == NULL)
2992 rc = mdd_links_read(env, child, &ldata);
2994 if (rc == -ENOENT || rc == -ENODATA)
2999 LASSERT(ldata.ld_leh != NULL);
3000 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3001 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3002 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3003 struct lu_name lname;
3006 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3009 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3010 lu_fid_eq(&fid, mdd_object_fid(parent))) {
3011 ldata.ld_lee = (struct link_ea_entry *)
3012 ((char *)ldata.ld_lee +
3017 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3018 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3019 lname.ln_namelen, lname.ln_name,
3020 PFID(mdd_object_fid(parent)));
3021 /* update to the new parent fid */
3022 linkea_entry_pack(ldata.ld_lee, &lname,
3023 mdd_object_fid(parent));
3025 rc = mdd_declare_links_add(env, child, handle, &ldata);
3027 rc = mdd_links_write(env, child, &ldata, handle);
3033 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3034 struct mdd_object *parent,
3035 struct mdd_object *child,
3036 const char *name, int namelen,
3037 struct thandle *handle)
3039 return mdd_linkea_update_child_internal(env, parent, child, name,
3040 namelen, handle, true);
3043 static int mdd_linkea_update_child(const struct lu_env *env,
3044 struct mdd_object *parent,
3045 struct mdd_object *child,
3046 const char *name, int namelen,
3047 struct thandle *handle)
3049 return mdd_linkea_update_child_internal(env, parent, child, name,
3050 namelen, handle, false);
3053 static int mdd_update_linkea_internal(const struct lu_env *env,
3054 struct mdd_object *mdd_pobj,
3055 struct mdd_object *mdd_sobj,
3056 struct mdd_object *mdd_tobj,
3057 const struct lu_name *child_name,
3058 struct thandle *handle,
3061 struct mdd_thread_info *info = mdd_env_info(env);
3062 struct linkea_data *ldata = &info->mti_link_data;
3067 rc = mdd_links_read(env, mdd_sobj, ldata);
3069 if (rc == -ENOENT || rc == -ENODATA)
3075 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3077 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3082 /* If it is mulitple links file, we need update the name entry for
3084 LASSERT(ldata->ld_leh != NULL);
3085 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3086 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3087 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3088 struct mdd_object *pobj;
3089 struct lu_name lname;
3092 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3094 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3096 pobj = mdd_object_find(env, mdd, &fid);
3098 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3099 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3104 if (!mdd_object_exists(pobj)) {
3105 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3106 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3110 if (pobj == mdd_pobj &&
3111 lname.ln_namelen == child_name->ln_namelen &&
3112 strncmp(lname.ln_name, child_name->ln_name,
3113 lname.ln_namelen) == 0) {
3114 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3115 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3120 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3121 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3122 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3125 /* Remove source name from source directory */
3126 /* Insert new fid with target name into target dir */
3127 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3132 rc = mdo_declare_index_insert(env, pobj,
3133 mdd_object_fid(mdd_tobj),
3134 mdd_object_type(mdd_tobj),
3135 lname.ln_name, handle);
3139 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3143 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3147 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3149 mdd_object_capa(env, pobj));
3153 rc = __mdd_index_insert(env, pobj,
3154 mdd_object_fid(mdd_tobj),
3155 mdd_object_type(mdd_tobj),
3156 lname.ln_name, handle,
3157 mdd_object_capa(env, pobj));
3161 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3162 rc = mdo_ref_add(env, mdd_tobj, handle);
3163 mdd_write_unlock(env, mdd_tobj);
3167 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3168 mdo_ref_del(env, mdd_sobj, handle);
3169 mdd_write_unlock(env, mdd_sobj);
3172 mdd_object_put(env, pobj);
3180 static int mdd_migrate_xattrs(const struct lu_env *env,
3181 struct mdd_object *mdd_sobj,
3182 struct mdd_object *mdd_tobj)
3184 struct mdd_thread_info *info = mdd_env_info(env);
3185 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3187 struct thandle *handle;
3193 struct lu_buf list_xbuf;
3196 /* retrieve xattr list from the old object */
3197 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL,
3198 mdd_object_capa(env, mdd_sobj));
3199 if (list_xsize == -ENODATA)
3205 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3206 if (info->mti_big_buf.lb_buf == NULL)
3209 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3210 list_xbuf.lb_len = list_xsize;
3211 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf,
3212 mdd_object_capa(env, mdd_sobj));
3217 xname = list_xbuf.lb_buf;
3219 xlen = strnlen(xname, rem - 1) + 1;
3220 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3221 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3222 strcmp(XATTR_NAME_LMV, xname) == 0)
3225 /* For directory, if there are default layout, migrate here */
3226 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3227 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3230 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL,
3232 mdd_object_capa(env, mdd_sobj));
3233 if (xsize == -ENODATA)
3238 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3239 if (info->mti_link_buf.lb_buf == NULL)
3240 GOTO(out, rc = -ENOMEM);
3242 xbuf.lb_len = xsize;
3243 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3244 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname,
3245 mdd_object_capa(env, mdd_sobj));
3251 handle = mdd_trans_create(env, mdd);
3253 GOTO(out, rc = PTR_ERR(handle));
3255 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3258 GOTO(stop_trans, rc);
3259 /* Note: this transaction is part of migration, and it is not
3260 * the last step of migration, so we set th_local = 1 to avoid
3261 * update last rcvd for this transaction */
3262 handle->th_local = 1;
3263 rc = mdd_trans_start(env, mdd, handle);
3265 GOTO(stop_trans, rc);
3267 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle,
3268 mdd_object_capa(env, mdd_sobj));
3270 GOTO(stop_trans, rc = 0);
3273 GOTO(stop_trans, rc);
3275 mdd_trans_stop(env, mdd, rc, handle);
3280 memmove(xname, xname + xlen, rem);
3286 static int mdd_declare_migrate_create(const struct lu_env *env,
3287 struct mdd_object *mdd_pobj,
3288 struct mdd_object *mdd_sobj,
3289 struct mdd_object *mdd_tobj,
3290 struct md_op_spec *spec,
3292 union lmv_mds_md *mgr_ea,
3293 struct thandle *handle)
3295 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3296 const struct lu_buf *buf;
3300 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3301 handle, spec, NULL);
3305 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3310 if (S_ISLNK(la->la_mode)) {
3311 const char *target_name = spec->u.sp_symname;
3312 int sym_len = strlen(target_name);
3313 const struct lu_buf *buf;
3315 buf = mdd_buf_get_const(env, target_name, sym_len);
3316 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3322 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3323 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3324 spec->u.sp_ea.eadatalen);
3325 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3331 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3332 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3333 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3338 la_flag->la_valid = LA_FLAGS;
3339 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3340 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3341 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3346 static int mdd_migrate_create(const struct lu_env *env,
3347 struct mdd_object *mdd_pobj,
3348 struct mdd_object *mdd_sobj,
3349 struct mdd_object *mdd_tobj,
3352 struct mdd_thread_info *info = mdd_env_info(env);
3353 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3354 struct md_op_spec *spec = &info->mti_spec;
3355 struct lu_buf lmm_buf = { 0 };
3356 struct lu_buf link_buf = { 0 };
3357 const struct lu_buf *buf;
3358 struct thandle *handle;
3359 struct lmv_mds_md_v1 *mgr_ea;
3360 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3361 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3366 /* prepare spec for create */
3367 memset(spec, 0, sizeof(*spec));
3368 spec->sp_cr_lookup = 0;
3369 spec->sp_feat = &dt_directory_features;
3370 if (S_ISLNK(la->la_mode)) {
3371 buf = lu_buf_check_and_alloc(
3372 &mdd_env_info(env)->mti_big_buf,
3375 link_buf.lb_len = la->la_size + 1;
3376 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3378 rc = rc != 0 ? rc : -EFAULT;
3379 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3380 mdd2obd_dev(mdd)->obd_name,
3381 PFID(mdd_object_fid(mdd_sobj)), rc);
3384 spec->u.sp_symname = link_buf.lb_buf;
3385 } else if S_ISREG(la->la_mode) {
3386 /* retrieve lov of the old object */
3387 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3388 if (rc != 0 && rc != -ENODATA)
3390 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3391 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3392 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3393 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3397 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3398 memset(mgr_ea, 0, sizeof(*mgr_ea));
3399 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3400 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3401 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3402 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3403 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3404 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3406 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3408 handle = mdd_trans_create(env, mdd);
3410 GOTO(out_free, rc = PTR_ERR(handle));
3412 /* Note: this transaction is part of migration, and it is not
3413 * the last step of migration, so we set th_local = 1 to avoid
3414 * update last rcvd for this transaction */
3415 handle->th_local = 1;
3416 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3418 (union lmv_mds_md *)info->mti_xattr_buf,
3421 GOTO(stop_trans, rc);
3423 rc = mdd_trans_start(env, mdd, handle);
3425 GOTO(stop_trans, rc);
3427 /* create the target object */
3428 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3431 GOTO(stop_trans, rc);
3433 /* Set MIGRATE EA on the source inode, so once the migration needs
3434 * to be re-done during failover, the re-do process can locate the
3435 * target object which is already being created. */
3436 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3437 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3438 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0,
3439 handle, mdd_object_capa(env, mdd_sobj));
3441 GOTO(stop_trans, rc);
3443 /* Set immutable flag, so any modification is disabled until
3444 * the migration is done. Once the migration is interrupted,
3445 * if the resume process find the migrating object has both
3446 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3447 * flag and approve the migration */
3448 la_flag->la_valid = LA_FLAGS;
3449 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3450 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3451 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3452 mdd_object_capa(env, mdd_sobj));
3455 mdd_trans_stop(env, mdd, rc, handle);
3457 if (lmm_buf.lb_buf != NULL)
3458 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3462 static int mdd_migrate_entries(const struct lu_env *env,
3463 struct mdd_object *mdd_sobj,
3464 struct mdd_object *mdd_tobj)
3466 struct dt_object *next = mdd_object_child(mdd_sobj);
3467 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3468 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3469 struct thandle *handle;
3471 const struct dt_it_ops *iops;
3474 struct lu_dirent *ent;
3477 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3481 if (!dt_try_as_dir(env, next))
3482 GOTO(out_ent, rc = -ENOTDIR);
3484 * iterate directories
3486 iops = &next->do_index_ops->dio_it;
3487 it = iops->init(env, next, LUDA_FID | LUDA_TYPE,
3488 mdd_object_capa(env, mdd_sobj));
3490 GOTO(out_ent, rc = PTR_ERR(it));
3492 rc = iops->load(env, it, 0);
3494 rc = iops->next(env, it);
3498 * At this point and across for-loop:
3500 * rc == 0 -> ok, proceed.
3501 * rc > 0 -> end of directory.
3505 struct mdd_object *child;
3506 char *name = mdd_env_info(env)->mti_key;
3510 bool target_exist = false;
3512 len = iops->key_size(env, it);
3516 result = iops->rec(env, it, (struct dt_rec *)ent,
3517 LUDA_FID | LUDA_TYPE);
3518 if (result == -ESTALE)
3525 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3526 recsize = le16_to_cpu(ent->lde_reclen);
3528 /* Insert new fid with target name into target dir */
3529 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3530 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3531 ent->lde_name[1] == '.'))
3534 child = mdd_object_find(env, mdd, &ent->lde_fid);
3536 GOTO(out, rc = PTR_ERR(child));
3538 is_dir = S_ISDIR(mdd_object_type(child));
3540 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3542 /* Check whether the name has been inserted to the target */
3543 if (dt_try_as_dir(env, dt_tobj)) {
3544 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3546 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3547 (struct dt_key *)name,
3548 mdd_object_capa(env, mdd_tobj));
3549 if (unlikely(rc == 0))
3550 target_exist = true;
3553 handle = mdd_trans_create(env, mdd);
3555 GOTO(out, rc = PTR_ERR(handle));
3557 /* Note: this transaction is part of migration, and it is not
3558 * the last step of migration, so we set th_local = 1 to avoid
3559 * updating last rcvd for this transaction */
3560 handle->th_local = 1;
3561 if (likely(!target_exist)) {
3562 rc = mdo_declare_index_insert(env, mdd_tobj,
3564 mdd_object_type(child),
3570 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3576 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3581 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3585 /* Update .. for child */
3586 rc = mdo_declare_index_delete(env, child, dotdot,
3591 rc = mdo_declare_index_insert(env, child,
3592 mdd_object_fid(mdd_tobj),
3593 S_IFDIR, dotdot, handle);
3598 rc = mdd_linkea_declare_update_child(env, mdd_tobj,
3605 rc = mdd_trans_start(env, mdd, handle);
3607 CERROR("%s: transaction start failed: rc = %d\n",
3608 mdd2obd_dev(mdd)->obd_name, rc);
3612 if (likely(!target_exist)) {
3613 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3614 mdd_object_type(child),
3616 mdd_object_capa(env, mdd_tobj));
3621 rc = mdo_ref_add(env, mdd_tobj, handle);
3628 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle,
3629 mdd_object_capa(env, mdd_sobj));
3634 rc = __mdd_index_delete_only(env, child, dotdot, handle,
3635 mdd_object_capa(env, child));
3639 rc = __mdd_index_insert_only(env, child,
3640 mdd_object_fid(mdd_tobj), S_IFDIR,
3642 mdd_object_capa(env, child));
3647 rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
3648 strlen(name), handle);
3651 mdd_object_put(env, child);
3652 mdd_trans_stop(env, mdd, rc, handle);
3656 result = iops->next(env, it);
3657 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3658 GOTO(out, rc = -EINTR);
3660 if (result == -ESTALE)
3662 } while (result == 0);
3665 iops->fini(env, it);
3667 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3671 static int mdd_declare_update_linkea(const struct lu_env *env,
3672 struct mdd_object *mdd_pobj,
3673 struct mdd_object *mdd_sobj,
3674 struct mdd_object *mdd_tobj,
3675 struct thandle *handle,
3676 const struct lu_name *child_name)
3678 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3679 child_name, handle, 1);
3682 static int mdd_update_linkea(const struct lu_env *env,
3683 struct mdd_object *mdd_pobj,
3684 struct mdd_object *mdd_sobj,
3685 struct mdd_object *mdd_tobj,
3686 struct thandle *handle,
3687 const struct lu_name *child_name)
3689 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3690 child_name, handle, 0);
3693 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3694 struct mdd_object *mdd_pobj,
3695 struct mdd_object *mdd_sobj,
3696 struct mdd_object *mdd_tobj,
3697 const struct lu_name *lname,
3699 struct lu_attr *parent_la,
3700 struct thandle *handle)
3702 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3705 /* Revert IMMUTABLE flag */
3706 la_flag->la_valid = LA_FLAGS;
3707 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3708 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3709 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3713 /* delete entry from source dir */
3714 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3718 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3719 mdd_tobj, handle, lname);
3723 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3724 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3730 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3731 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3737 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3738 mdd_object_type(mdd_tobj),
3739 lname->ln_name, handle);
3743 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3744 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3749 /* delete old object */
3750 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3754 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3755 /* delete old object */
3756 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3759 /* set nlink to 0 */
3760 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3765 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3769 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3774 static int mdd_migrate_update_name(const struct lu_env *env,
3775 struct mdd_object *mdd_pobj,
3776 struct mdd_object *mdd_sobj,
3777 struct mdd_object *mdd_tobj,
3778 const struct lu_name *lname,
3781 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3782 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3783 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3784 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3785 struct thandle *handle;
3786 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3787 const char *name = lname->ln_name;
3791 /* update time for parent */
3792 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3793 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3794 p_la->la_valid = LA_CTIME;
3796 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3800 handle = mdd_trans_create(env, mdd);
3802 RETURN(PTR_ERR(handle));
3804 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3805 lname, so_attr, p_la, handle);
3807 /* If the migration can not be fit in one transaction, just
3808 * leave it in the original MDT */
3810 GOTO(stop_trans, rc = 0);
3812 GOTO(stop_trans, rc);
3815 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3816 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3817 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3818 PFID(mdd_object_fid(mdd_tobj)));
3820 rc = mdd_trans_start(env, mdd, handle);
3822 GOTO(stop_trans, rc);
3824 /* Revert IMMUTABLE flag */
3825 la_flag->la_valid = LA_FLAGS;
3826 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3827 mdd_flags_xlate(mdd_sobj, la_flag->la_flags);
3828 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3829 mdd_object_capa(env, mdd_pobj));
3831 GOTO(stop_trans, rc);
3833 /* Remove source name from source directory */
3834 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
3835 mdd_object_capa(env, mdd_pobj));
3837 GOTO(stop_trans, rc);
3839 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3842 GOTO(stop_trans, rc);
3844 if (S_ISREG(so_attr->la_mode)) {
3845 if (so_attr->la_nlink == 1) {
3846 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3848 mdd_object_capa(env, mdd_sobj));
3849 if (rc != 0 && rc != -ENODATA)
3850 GOTO(stop_trans, rc);
3854 /* Insert new fid with target name into target dir */
3855 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
3856 mdd_object_type(mdd_tobj), name,
3857 handle, mdd_object_capa(env, mdd_pobj));
3859 GOTO(stop_trans, rc);
3861 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3864 GOTO(stop_trans, rc);
3866 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3867 mdo_ref_del(env, mdd_sobj, handle);
3869 mdo_ref_del(env, mdd_sobj, handle);
3871 /* Get the attr again after ref_del */
3872 rc = mdd_la_get(env, mdd_sobj, so_attr,
3873 mdd_object_capa(env, mdd_sobj));
3875 GOTO(out_unlock, rc);
3877 ma->ma_attr = *so_attr;
3878 ma->ma_valid |= MA_INODE;
3879 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3881 GOTO(out_unlock, rc);
3883 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3885 GOTO(out_unlock, rc);
3888 mdd_write_unlock(env, mdd_sobj);
3891 mdd_trans_stop(env, mdd, rc, handle);
3897 * Check whether we should migrate the file/dir
3899 * < 0 permission check failed or other error.
3900 * = 0 the file can be migrated.
3901 * > 0 the file does not need to be migrated, mostly
3902 * for multiple link file
3904 static int mdd_migrate_sanity_check(const struct lu_env *env,
3905 struct mdd_object *pobj,
3906 const struct lu_attr *pattr,
3907 struct mdd_object *sobj,
3908 struct lu_attr *sattr)
3910 struct mdd_thread_info *info = mdd_env_info(env);
3911 struct linkea_data *ldata = &info->mti_link_data;
3913 struct lu_buf *mgr_buf;
3919 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3920 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3921 if (mgr_buf->lb_buf == NULL)
3924 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV,
3925 mdd_object_capa(env, sobj));
3927 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3929 /* If the object has migrateEA, it means IMMUTE flag
3930 * is being set by previous migration process, so it
3931 * needs to override the IMMUTE flag, otherwise the
3932 * following sanity check will fail */
3933 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
3934 LMV_HASH_FLAG_MIGRATION) {
3935 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3937 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3938 sobj->mod_flags &= ~IMMUTE_OBJ;
3939 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3940 mdd2obd_dev(mdd)->obd_name,
3941 PFID(mdd_object_fid(sobj)));
3945 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3946 sobj, sattr, NULL, NULL);
3950 /* Then it will check if the file should be migrated. If the file
3951 * has mulitple links, we only need migrate the file if all of its
3952 * entries has been migrated to the remote MDT */
3953 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3956 rc = mdd_links_read(env, sobj, ldata);
3958 /* For multiple links files, if there are no linkEA data at all,
3959 * means the file might be created before linkEA is enabled, and
3960 * all all of its links should not be migrated yet, otherwise
3961 * it should have some linkEA there */
3962 if (rc == -ENOENT || rc == -ENODATA)
3967 /* If it is mulitple links file, we need update the name entry for
3969 LASSERT(ldata->ld_leh != NULL);
3970 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3971 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3972 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3973 struct mdd_object *lpobj;
3974 struct lu_name lname;
3977 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3979 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3981 lpobj = mdd_object_find(env, mdd, &fid);
3982 if (IS_ERR(lpobj)) {
3983 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3984 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3989 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
3990 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
3991 PFID(&fid), lname.ln_namelen, lname.ln_name);
3992 mdd_object_put(env, lpobj);
3996 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
3997 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
3998 lname.ln_name, PFID(&fid));
3999 mdd_object_put(env, lpobj);
4007 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4008 struct md_object *sobj, const struct lu_name *lname,
4009 struct md_object *tobj, struct md_attr *ma)
4011 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4012 struct mdd_device *mdd = mdo2mdd(pobj);
4013 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4014 struct mdd_object *mdd_tobj = NULL;
4015 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4016 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4020 /* If the file will being migrated, it will check whether
4021 * the file is being opened by someone else right now */
4022 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4023 if (mdd_sobj->mod_count >= 1) {
4024 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
4025 mdd2obd_dev(mdd)->obd_name,
4026 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4027 mdd_sobj->mod_count, -EBUSY);
4028 mdd_read_unlock(env, mdd_sobj);
4029 GOTO(put, rc = -EBUSY);
4031 mdd_read_unlock(env, mdd_sobj);
4033 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
4037 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
4041 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4048 /* Sigh, it is impossible to finish all of migration in a single
4049 * transaction, for example migrating big directory entries to the
4050 * new MDT, it needs insert all of name entries of children in the
4053 * So migration will be done in multiple steps and transactions.
4055 * 1. create an orphan object on the remote MDT in one transaction.
4056 * 2. migrate extend attributes to the new target file/directory.
4057 * 3. For directory, migrate the entries to the new MDT and update
4058 * linkEA of each children. Because we can not migrate all entries
4059 * in a single transaction, so the migrating directory will become
4060 * a striped directory during migration, so once the process is
4061 * interrupted, the directory is still accessible. (During lookup,
4062 * client will locate the name by searching both original and target
4064 * 4. Finally, update the name/FID to point to the new file/directory
4065 * in a separate transaction.
4068 /* step 1: Check whether the orphan object has been created, and create
4069 * orphan object on the remote MDT if needed */
4070 mdd_tobj = md2mdd_obj(tobj);
4071 if (!mdd_object_exists(mdd_tobj)) {
4072 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4078 /* step 2: migrate xattr */
4079 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4083 /* step 3: migrate name entries to the orphan object */
4084 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4085 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4088 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4089 OBD_FAIL_MDS_REINT_NET_REP)))
4093 /* step 4: update name entry to the new object */
4094 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4102 const struct md_dir_operations mdd_dir_ops = {
4103 .mdo_is_subdir = mdd_is_subdir,
4104 .mdo_lookup = mdd_lookup,
4105 .mdo_create = mdd_create,
4106 .mdo_rename = mdd_rename,
4107 .mdo_link = mdd_link,
4108 .mdo_unlink = mdd_unlink,
4109 .mdo_create_data = mdd_create_data,
4110 .mdo_migrate = mdd_migrate,