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, 2014, 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,
123 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
127 static inline int mdd_parent_fid(const struct lu_env *env,
128 struct mdd_object *obj,
129 const struct lu_attr *attr,
132 return __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
136 * For root fid use special function, which does not compare version component
137 * of fid. Version component is different for root fids on all MDTs.
139 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
141 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
142 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
146 * return 1: if lf is the fid of the ancestor of p1;
149 * return -EREMOTE: if remote object is found, in this
150 * case fid of remote object is saved to @pf;
152 * otherwise: values < 0, errors.
154 static int mdd_is_parent(const struct lu_env *env,
155 struct mdd_device *mdd,
156 struct mdd_object *p1,
157 const struct lu_attr *attr,
158 const struct lu_fid *lf,
161 struct mdd_object *parent = NULL;
166 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
167 pfid = &mdd_env_info(env)->mti_fid;
169 /* Check for root first. */
170 if (mdd_is_root(mdd, mdo2fid(p1)))
174 /* this is done recursively, bypass capa for each obj */
175 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
176 rc = mdd_parent_fid(env, p1, attr, pfid);
179 if (mdd_is_root(mdd, pfid))
181 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
183 if (lu_fid_eq(pfid, lf))
186 mdd_object_put(env, parent);
188 parent = mdd_object_find(env, mdd, pfid);
189 if (IS_ERR(parent)) {
190 GOTO(out, rc = PTR_ERR(parent));
191 } else if (mdd_object_remote(parent)) {
192 /*FIXME: Because of the restriction of rename in Phase I.
193 * If the parent is remote, we just assumed lf is not the
194 * parent of P1 for now */
201 if (parent && !IS_ERR(parent))
202 mdd_object_put(env, parent);
207 * No permission check is needed.
209 * returns 1: if fid is ancestor of @mo;
210 * returns 0: if fid is not a ancestor of @mo;
212 * returns EREMOTE if remote object is found, fid of remote object is saved to
215 * returns < 0: if error
217 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
218 const struct lu_fid *fid, struct lu_fid *sfid)
220 struct mdd_device *mdd = mdo2mdd(mo);
221 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
225 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
228 rc = mdd_la_get(env, md2mdd_obj(mo), attr, BYPASS_CAPA);
232 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
236 } else if (rc == 1) {
237 /* found @fid is parent */
245 * Check that @dir contains no entries except (possibly) dot and dotdot.
250 * -ENOTDIR not a directory object
251 * -ENOTEMPTY not empty
255 static int mdd_dir_is_empty(const struct lu_env *env,
256 struct mdd_object *dir)
259 struct dt_object *obj;
260 const struct dt_it_ops *iops;
264 obj = mdd_object_child(dir);
265 if (!dt_try_as_dir(env, obj))
268 iops = &obj->do_index_ops->dio_it;
269 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
271 result = iops->get(env, it, (const struct dt_key *)"");
274 for (result = 0, i = 0; result == 0 && i < 3; ++i)
275 result = iops->next(env, it);
278 else if (result == 1)
280 } else if (result == 0)
282 * Huh? Index contains no zero key?
289 result = PTR_ERR(it);
294 * Determine if the target object can be hard linked, and right now it only
295 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
296 * directory nlink count might exceed the maximum link count(see
297 * osd_object_ref_add), so it only check nlink for non-directories.
299 * \param[in] env thread environment
300 * \param[in] obj object being linked to
301 * \param[in] la attributes of \a obj
303 * \retval 0 if \a obj can be hard linked
304 * \retval negative error if \a obj is a directory or has too
307 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
308 const struct lu_attr *la)
310 struct mdd_device *m = mdd_obj2mdd_dev(obj);
315 /* Subdir count limitation can be broken through
316 * (see osd_object_ref_add), so only check non-directory here. */
317 if (!S_ISDIR(la->la_mode) &&
318 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
325 * Check whether it may create the cobj under the pobj.
327 * \param[in] env execution environment
328 * \param[in] pobj the parent directory
329 * \param[in] pattr the attribute of the parent directory
330 * \param[in] cobj the child to be created
331 * \param[in] check_perm if check WRITE|EXEC permission for parent
333 * \retval = 0 create the child under this dir is allowed
334 * \retval negative errno create the child under this dir is
337 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
338 const struct lu_attr *pattr, struct mdd_object *cobj,
341 struct mdd_thread_info *info = mdd_env_info(env);
346 if (cobj && mdd_object_exists(cobj))
349 if (mdd_is_dead_obj(pobj))
352 /* If the parent is a sub-stripe, check whether it is dead */
353 xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key));
354 rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV,
355 mdd_object_capa(env, pobj));
356 if (unlikely(rc > 0)) {
357 struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf;
359 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE &&
360 le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD)
366 rc = mdd_permission_internal_locked(env, pobj, pattr,
367 MAY_WRITE | MAY_EXEC,
373 * Check whether can unlink from the pobj in the case of "cobj == NULL".
375 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
376 const struct lu_attr *pattr, const struct lu_attr *attr)
381 if (mdd_is_dead_obj(pobj))
384 if (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 (pattr->la_flags & LUSTRE_APPEND_FL)
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 (pattr->la_flags & LUSTRE_APPEND_FL)
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 (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
488 /* additional check the rename case */
490 if (S_ISDIR(cattr->la_mode)) {
491 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
493 if (!S_ISDIR(tattr->la_mode))
496 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
498 } else if (S_ISDIR(tattr->la_mode))
502 if (S_ISDIR(tattr->la_mode) && check_empty)
503 rc = mdd_dir_is_empty(env, tobj);
509 * Check whether it can create the link file(linked to @src_obj) under
510 * the target directory(@tgt_obj), and src_obj has been locked by
513 * \param[in] env execution environment
514 * \param[in] tgt_obj the target directory
515 * \param[in] tattr attributes of target directory
516 * \param[in] lname the link name
517 * \param[in] src_obj source object for link
518 * \param[in] cattr attributes for source object
520 * \retval = 0 it is allowed to create the link file under tgt_obj
521 * \retval negative error not allowed to create the link file
523 static int mdd_link_sanity_check(const struct lu_env *env,
524 struct mdd_object *tgt_obj,
525 const struct lu_attr *tattr,
526 const struct lu_name *lname,
527 struct mdd_object *src_obj,
528 const struct lu_attr *cattr)
530 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
534 if (!mdd_object_exists(src_obj))
537 if (mdd_is_dead_obj(src_obj))
540 /* Local ops, no lookup before link, check filename length here. */
541 rc = mdd_name_check(m, lname);
545 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
548 if (S_ISDIR(mdd_object_type(src_obj)))
551 LASSERT(src_obj != tgt_obj);
552 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
556 rc = __mdd_may_link(env, src_obj, cattr);
561 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
562 const char *name, struct thandle *handle,
563 struct lustre_capa *capa)
565 struct dt_object *next = mdd_object_child(pobj);
569 if (dt_try_as_dir(env, next))
570 rc = dt_delete(env, next, (struct dt_key *)name, handle, capa);
577 static int __mdd_index_insert_only(const struct lu_env *env,
578 struct mdd_object *pobj,
579 const struct lu_fid *lf, __u32 type,
581 struct thandle *handle,
582 struct lustre_capa *capa)
584 struct dt_object *next = mdd_object_child(pobj);
588 if (dt_try_as_dir(env, next)) {
589 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
590 struct lu_ucred *uc = lu_ucred_check(env);
594 rec->rec_type = type;
595 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
596 rc = dt_insert(env, next, (const struct dt_rec *)rec,
597 (const struct dt_key *)name, handle, capa,
605 /* insert named index, add reference if isdir */
606 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
607 const struct lu_fid *lf, __u32 type,
608 const char *name, struct thandle *handle,
609 struct lustre_capa *capa)
614 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle, capa);
615 if (rc == 0 && S_ISDIR(type)) {
616 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
617 mdo_ref_add(env, pobj, handle);
618 mdd_write_unlock(env, pobj);
624 /* delete named index, drop reference if isdir */
625 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
626 const char *name, int is_dir, struct thandle *handle,
627 struct lustre_capa *capa)
632 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
633 if (rc == 0 && is_dir) {
634 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
635 mdo_ref_del(env, pobj, handle);
636 mdd_write_unlock(env, pobj);
642 static int mdd_llog_record_calc_size(const struct lu_env *env,
643 const struct lu_name *tname,
644 const struct lu_name *sname)
646 const struct lu_ucred *uc = lu_ucred(env);
647 enum changelog_rec_flags crf = 0;
648 size_t hdr_size = sizeof(struct llog_changelog_rec) -
649 sizeof(struct changelog_rec);
654 if (uc != NULL && uc->uc_jobid[0] != '\0')
657 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
658 (tname != NULL ? tname->ln_namelen : 0) +
659 (sname != NULL ? 1 + sname->ln_namelen : 0));
662 int mdd_declare_changelog_store(const struct lu_env *env,
663 struct mdd_device *mdd,
664 const struct lu_name *tname,
665 const struct lu_name *sname,
666 struct thandle *handle)
668 struct obd_device *obd = mdd2obd_dev(mdd);
669 struct llog_ctxt *ctxt;
670 struct llog_changelog_rec *rec;
676 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
679 reclen = mdd_llog_record_calc_size(env, tname, sname);
680 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
681 if (buf->lb_buf == NULL)
685 rec->cr_hdr.lrh_len = reclen;
686 rec->cr_hdr.lrh_type = CHANGELOG_REC;
688 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
692 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle);
698 /** Add a changelog entry \a rec to the changelog llog
701 * \param handle - currently ignored since llogs start their own transaction;
702 * this will hopefully be fixed in llog rewrite
705 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
706 struct llog_changelog_rec *rec, struct thandle *th)
708 struct obd_device *obd = mdd2obd_dev(mdd);
709 struct llog_ctxt *ctxt;
712 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
713 changelog_rec_varsize(&rec->cr));
715 /* llog_lvfs_write_rec sets the llog tail len */
716 rec->cr_hdr.lrh_type = CHANGELOG_REC;
717 rec->cr.cr_time = cl_time();
719 spin_lock(&mdd->mdd_cl.mc_lock);
720 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
721 * but as long as the MDD transactions are ordered correctly for e.g.
722 * rename conflicts, I don't think this should matter. */
723 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
724 spin_unlock(&mdd->mdd_cl.mc_lock);
726 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
730 /* nested journal transaction */
731 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th);
739 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
740 const struct lu_fid *sfid,
741 const struct lu_fid *spfid,
742 const struct lu_name *sname)
744 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
745 size_t extsize = sname->ln_namelen + 1;
747 LASSERT(sfid != NULL);
748 LASSERT(spfid != NULL);
749 LASSERT(sname != NULL);
751 rnm->cr_sfid = *sfid;
752 rnm->cr_spfid = *spfid;
754 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
755 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
756 rec->cr_namelen += extsize;
759 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
761 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
763 if (jobid == NULL || jobid[0] == '\0')
766 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
769 /** Store a namespace change changelog record
770 * If this fails, we must fail the whole transaction; we don't
771 * want the change to commit without the log entry.
772 * \param target - mdd_object of change
773 * \param tpfid - target parent dir/object fid
774 * \param sfid - source object fid
775 * \param spfid - source parent fid
776 * \param tname - target name string
777 * \param sname - source name string
778 * \param handle - transacion handle
780 int mdd_changelog_ns_store(const struct lu_env *env,
781 struct mdd_device *mdd,
782 enum changelog_rec_type type,
783 enum changelog_rec_flags crf,
784 struct mdd_object *target,
785 const struct lu_fid *tpfid,
786 const struct lu_fid *sfid,
787 const struct lu_fid *spfid,
788 const struct lu_name *tname,
789 const struct lu_name *sname,
790 struct thandle *handle)
792 const struct lu_ucred *uc = lu_ucred(env);
793 struct llog_changelog_rec *rec;
800 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
803 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
806 LASSERT(tpfid != NULL);
807 LASSERT(tname != NULL);
808 LASSERT(handle != NULL);
810 reclen = mdd_llog_record_calc_size(env, tname, sname);
811 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
812 if (buf->lb_buf == NULL)
816 crf = (crf & CLF_FLAGMASK);
818 if (uc->uc_jobid[0] != '\0')
826 rec->cr.cr_flags = crf;
827 rec->cr.cr_type = (__u32)type;
828 rec->cr.cr_pfid = *tpfid;
829 rec->cr.cr_namelen = tname->ln_namelen;
830 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
832 if (crf & CLF_RENAME)
833 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
836 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
838 if (likely(target != NULL)) {
839 rec->cr.cr_tfid = *mdo2fid(target);
840 target->mod_cltime = cfs_time_current_64();
842 fid_zero(&rec->cr.cr_tfid);
845 rc = mdd_changelog_store(env, mdd, rec, handle);
847 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
848 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
855 static int __mdd_links_add(const struct lu_env *env,
856 struct mdd_object *mdd_obj,
857 struct linkea_data *ldata,
858 const struct lu_name *lname,
859 const struct lu_fid *pfid,
860 int first, int check)
864 if (ldata->ld_leh == NULL) {
865 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
869 rc = linkea_data_new(ldata,
870 &mdd_env_info(env)->mti_link_buf);
877 rc = linkea_links_find(ldata, lname, pfid);
878 if (rc && rc != -ENOENT)
884 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
885 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
889 linkea_add_buf(ldata, lname, tfid);
892 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
893 linkea_add_buf(ldata, lname, pfid);
895 return linkea_add_buf(ldata, lname, pfid);
898 static int __mdd_links_del(const struct lu_env *env,
899 struct mdd_object *mdd_obj,
900 struct linkea_data *ldata,
901 const struct lu_name *lname,
902 const struct lu_fid *pfid)
906 if (ldata->ld_leh == NULL) {
907 rc = mdd_links_read(env, mdd_obj, ldata);
912 rc = linkea_links_find(ldata, lname, pfid);
916 linkea_del_buf(ldata, lname);
920 static int mdd_linkea_prepare(const struct lu_env *env,
921 struct mdd_object *mdd_obj,
922 const struct lu_fid *oldpfid,
923 const struct lu_name *oldlname,
924 const struct lu_fid *newpfid,
925 const struct lu_name *newlname,
926 int first, int check,
927 struct linkea_data *ldata)
933 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
936 LASSERT(oldpfid != NULL || newpfid != NULL);
938 if (mdd_obj->mod_flags & DEAD_OBJ) {
939 /* Prevent linkea to be updated which is NOT necessary. */
940 ldata->ld_reclen = 0;
941 /* No more links, don't bother */
945 if (oldpfid != NULL) {
946 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
949 (rc != -ENODATA && rc != -ENOENT))
951 /* No changes done. */
956 /* If renaming, add the new record */
957 if (newpfid != NULL) {
958 /* even if the add fails, we still delete the out-of-date
960 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
964 rc = rc != 0 ? rc : rc2;
969 int mdd_links_rename(const struct lu_env *env,
970 struct mdd_object *mdd_obj,
971 const struct lu_fid *oldpfid,
972 const struct lu_name *oldlname,
973 const struct lu_fid *newpfid,
974 const struct lu_name *newlname,
975 struct thandle *handle,
976 struct linkea_data *ldata,
977 int first, int check)
983 ldata = &mdd_env_info(env)->mti_link_data;
984 memset(ldata, 0, sizeof(*ldata));
985 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
986 newpfid, newlname, first, check,
992 if (ldata->ld_reclen != 0)
993 rc = mdd_links_write(env, mdd_obj, ldata, handle);
998 if (rc == -EOVERFLOW || rc == -ENOSPC)
1000 if (oldpfid == NULL)
1001 CDEBUG(error ? D_ERROR : D_OTHER,
1002 "link_ea add '%.*s' failed %d "DFID"\n",
1003 newlname->ln_namelen, newlname->ln_name,
1004 rc, PFID(mdd_object_fid(mdd_obj)));
1005 else if (newpfid == NULL)
1006 CDEBUG(error ? D_ERROR : D_OTHER,
1007 "link_ea del '%.*s' failed %d "DFID"\n",
1008 oldlname->ln_namelen, oldlname->ln_name,
1009 rc, PFID(mdd_object_fid(mdd_obj)));
1011 CDEBUG(error ? D_ERROR : D_OTHER,
1012 "link_ea rename '%.*s'->'%.*s' failed %d "
1014 oldlname->ln_namelen, oldlname->ln_name,
1015 newlname->ln_namelen, newlname->ln_name,
1016 rc, PFID(mdd_object_fid(mdd_obj)));
1019 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1020 /* if we vmalloced a large buffer drop it */
1021 lu_buf_free(ldata->ld_buf);
1026 static inline int mdd_links_add(const struct lu_env *env,
1027 struct mdd_object *mdd_obj,
1028 const struct lu_fid *pfid,
1029 const struct lu_name *lname,
1030 struct thandle *handle,
1031 struct linkea_data *ldata, int first)
1033 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1034 pfid, lname, handle, ldata, first, 0);
1037 static inline int mdd_links_del(const struct lu_env *env,
1038 struct mdd_object *mdd_obj,
1039 const struct lu_fid *pfid,
1040 const struct lu_name *lname,
1041 struct thandle *handle)
1043 return mdd_links_rename(env, mdd_obj, pfid, lname,
1044 NULL, NULL, handle, NULL, 0, 0);
1047 /** Read the link EA into a temp buffer.
1048 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1049 * A pointer to the buffer is stored in \a ldata::ld_buf.
1051 * \retval 0 or error
1053 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1054 struct linkea_data *ldata)
1058 if (!mdd_object_exists(mdd_obj))
1061 /* First try a small buf */
1062 LASSERT(env != NULL);
1063 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1065 if (ldata->ld_buf->lb_buf == NULL)
1068 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK,
1070 if (rc == -ERANGE) {
1071 /* Buf was too small, figure out what we need. */
1072 lu_buf_free(ldata->ld_buf);
1073 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1074 XATTR_NAME_LINK, BYPASS_CAPA);
1077 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1078 if (ldata->ld_buf->lb_buf == NULL)
1080 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1081 XATTR_NAME_LINK, BYPASS_CAPA);
1084 lu_buf_free(ldata->ld_buf);
1085 ldata->ld_buf = NULL;
1089 return linkea_init(ldata);
1092 /** Read the link EA into a temp buffer.
1093 * Uses the name_buf since it is generally large.
1094 * \retval IS_ERR err
1095 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1097 struct lu_buf *mdd_links_get(const struct lu_env *env,
1098 struct mdd_object *mdd_obj)
1100 struct linkea_data ldata = { NULL };
1103 rc = mdd_links_read(env, mdd_obj, &ldata);
1104 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1107 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1108 struct linkea_data *ldata, struct thandle *handle)
1110 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1111 ldata->ld_leh->leh_len);
1114 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1117 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle,
1118 mdd_object_capa(env, mdd_obj));
1119 if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) &&
1120 mdd_object_remote(mdd_obj) == 0) {
1121 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1123 /* XXX: If the linkEA is overflow, then we need to notify the
1124 * namespace LFSCK to skip "nlink" attribute verification
1125 * on this object to avoid the "nlink" to be shrinked by
1126 * wrong. It may be not good an interaction with LFSCK
1127 * like this. We will consider to replace it with other
1128 * mechanism in future. LU-5802. */
1129 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK,
1130 LFSCK_TYPE_NAMESPACE);
1131 lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1138 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1139 struct thandle *handle, struct linkea_data *ldata,
1140 enum mdd_links_add_overflow overflow)
1146 if (ldata != NULL && ldata->ld_leh != NULL) {
1147 ea_len = ldata->ld_leh->leh_len;
1148 linkea = ldata->ld_buf->lb_buf;
1150 ea_len = DEFAULT_LINKEA_SIZE;
1154 /* XXX: max size? */
1155 rc = mdo_declare_xattr_set(env, mdd_obj,
1156 mdd_buf_get_const(env, linkea, ea_len),
1157 XATTR_NAME_LINK, 0, handle);
1161 if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) {
1162 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1164 /* XXX: If the linkEA is overflow, then we need to notify the
1165 * namespace LFSCK to skip "nlink" attribute verification
1166 * on this object to avoid the "nlink" to be shrinked by
1167 * wrong. It may be not good an interaction with LFSCK
1168 * like this. We will consider to replace it with other
1169 * mechanism in future. LU-5802. */
1170 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE,
1171 LFSCK_TYPE_NAMESPACE);
1172 rc = lfsck_in_notify(env,
1173 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1180 static inline int mdd_declare_links_del(const struct lu_env *env,
1181 struct mdd_object *c,
1182 struct thandle *handle)
1186 /* For directory, the linkEA will be removed together
1187 * with the object. */
1188 if (!S_ISDIR(mdd_object_type(c)))
1189 rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE);
1194 static int mdd_declare_link(const struct lu_env *env,
1195 struct mdd_device *mdd,
1196 struct mdd_object *p,
1197 struct mdd_object *c,
1198 const struct lu_name *name,
1199 struct thandle *handle,
1201 struct linkea_data *data)
1205 rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
1206 name->ln_name, handle);
1210 rc = mdo_declare_ref_add(env, c, handle);
1214 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1215 rc = mdo_declare_ref_add(env, c, handle);
1220 la->la_valid = LA_CTIME | LA_MTIME;
1221 rc = mdo_declare_attr_set(env, p, la, handle);
1225 la->la_valid = LA_CTIME;
1226 rc = mdo_declare_attr_set(env, c, la, handle);
1230 rc = mdd_declare_links_add(env, c, handle, data,
1231 S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE);
1235 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1240 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1241 struct md_object *src_obj, const struct lu_name *lname,
1244 const char *name = lname->ln_name;
1245 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1246 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1247 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1248 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1249 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1250 struct mdd_device *mdd = mdo2mdd(src_obj);
1251 struct thandle *handle;
1252 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1256 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
1260 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
1264 handle = mdd_trans_create(env, mdd);
1266 GOTO(out_pending, rc = PTR_ERR(handle));
1268 memset(ldata, 0, sizeof(*ldata));
1270 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1271 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1273 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1278 rc = mdd_trans_start(env, mdd, handle);
1282 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1283 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1286 GOTO(out_unlock, rc);
1288 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1289 rc = mdo_ref_add(env, mdd_sobj, handle);
1291 GOTO(out_unlock, rc);
1294 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1295 rc = mdo_ref_add(env, mdd_sobj, handle);
1297 GOTO(out_unlock, rc);
1300 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) {
1301 struct lu_fid tfid = *mdo2fid(mdd_sobj);
1304 rc = __mdd_index_insert_only(env, mdd_tobj, &tfid,
1305 mdd_object_type(mdd_sobj),
1307 mdd_object_capa(env, mdd_tobj));
1309 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1310 mdd_object_type(mdd_sobj),
1312 mdd_object_capa(env, mdd_tobj));
1316 mdo_ref_del(env, mdd_sobj, handle);
1317 GOTO(out_unlock, rc);
1320 la->la_valid = LA_CTIME | LA_MTIME;
1321 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1323 GOTO(out_unlock, rc);
1325 la->la_valid = LA_CTIME;
1326 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1328 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1329 mdo2fid(mdd_tobj), lname, 0, 0,
1332 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1333 lname, handle, ldata, 0);
1334 /* The failure of links_add should not cause the link
1335 * failure, reset rc here */
1340 mdd_write_unlock(env, mdd_sobj);
1342 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1343 mdo2fid(mdd_tobj), NULL, NULL,
1344 lname, NULL, handle);
1346 mdd_trans_stop(env, mdd, rc, handle);
1348 if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG)
1349 /* if we vmalloced a large buffer drop it */
1350 lu_buf_free(ldata->ld_buf);
1355 static int mdd_mark_dead_object(const struct lu_env *env,
1356 struct mdd_object *obj, struct thandle *handle,
1359 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1363 obj->mod_flags |= DEAD_OBJ;
1365 if (!S_ISDIR(mdd_object_type(obj)))
1368 attr->la_valid = LA_FLAGS;
1369 attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
1372 rc = mdo_declare_attr_set(env, obj, attr, handle);
1374 rc = mdo_attr_set(env, obj, attr, handle,
1375 mdd_object_capa(env, obj));
1380 static int mdd_declare_finish_unlink(const struct lu_env *env,
1381 struct mdd_object *obj,
1382 struct thandle *handle)
1386 rc = mdd_mark_dead_object(env, obj, handle, true);
1390 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1394 rc = mdo_declare_destroy(env, obj, handle);
1398 return mdd_declare_links_del(env, obj, handle);
1401 /* caller should take a lock before calling */
1402 int mdd_finish_unlink(const struct lu_env *env,
1403 struct mdd_object *obj, struct md_attr *ma,
1404 const struct mdd_object *pobj,
1405 const struct lu_name *lname,
1409 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1412 LASSERT(mdd_write_locked(env, obj) != 0);
1414 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1415 rc = mdd_mark_dead_object(env, obj, th, false);
1419 /* add new orphan and the object
1420 * will be deleted during mdd_close() */
1421 if (obj->mod_count) {
1422 rc = __mdd_orphan_add(env, obj, th);
1424 CDEBUG(D_HA, "Object "DFID" is inserted into "
1425 "orphan list, open count = %d\n",
1426 PFID(mdd_object_fid(obj)),
1429 CERROR("Object "DFID" fail to be an orphan, "
1430 "open count = %d, maybe cause failed "
1432 PFID(mdd_object_fid(obj)),
1435 rc = mdo_destroy(env, obj, th);
1437 } else if (!is_dir) {
1438 /* old files may not have link ea; ignore errors */
1439 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1447 * has mdd_write_lock on cobj already, but not on pobj yet
1449 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1450 const struct lu_attr *pattr,
1451 struct mdd_object *cobj,
1452 const struct lu_attr *cattr)
1457 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1462 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1463 struct mdd_object *p, struct mdd_object *c,
1464 const struct lu_name *name, struct md_attr *ma,
1465 struct thandle *handle, int no_name, int is_dir)
1467 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1470 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1471 if (likely(no_name == 0)) {
1472 rc = mdo_declare_index_delete(env, p, name->ln_name,
1479 rc = mdo_declare_ref_del(env, p, handle);
1485 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1486 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1487 la->la_valid = LA_CTIME | LA_MTIME;
1488 rc = mdo_declare_attr_set(env, p, la, handle);
1493 rc = mdo_declare_ref_del(env, c, handle);
1497 rc = mdo_declare_ref_del(env, c, handle);
1501 la->la_valid = LA_CTIME;
1502 rc = mdo_declare_attr_set(env, c, la, handle);
1506 rc = mdd_declare_finish_unlink(env, c, handle);
1510 /* FIXME: need changelog for remove entry */
1511 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1518 * test if a file has an HSM archive
1519 * if HSM attributes are not found in ma update them from
1522 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1523 struct mdd_object *obj,
1528 if (!(ma->ma_valid & MA_HSM)) {
1529 /* no HSM MD provided, read xattr */
1530 struct lu_buf *hsm_buf;
1531 const size_t buflen = sizeof(struct hsm_attrs);
1534 hsm_buf = mdd_buf_get(env, NULL, 0);
1535 lu_buf_alloc(hsm_buf, buflen);
1536 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM,
1537 mdd_object_capa(env, obj));
1538 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1539 lu_buf_free(hsm_buf);
1543 ma->ma_valid = MA_HSM;
1545 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1551 * Delete name entry and the object.
1552 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1553 * does not exist for this object, and it could only happen during resending
1554 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1555 * is also needed in this case(needed by changelog), so we have to add another
1556 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1557 * the ENOENT failure should be able to be fixed by redo mechanism.
1559 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1560 struct md_object *cobj, const struct lu_name *lname,
1561 struct md_attr *ma, int no_name)
1563 const char *name = lname->ln_name;
1564 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1565 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1566 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1567 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1568 struct mdd_object *mdd_cobj = NULL;
1569 struct mdd_device *mdd = mdo2mdd(pobj);
1570 struct thandle *handle;
1574 /* cobj == NULL means only delete name entry */
1575 if (likely(cobj != NULL)) {
1576 mdd_cobj = md2mdd_obj(cobj);
1577 if (mdd_object_exists(mdd_cobj) == 0)
1581 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
1585 if (likely(mdd_cobj != NULL)) {
1587 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1591 is_dir = S_ISDIR(cattr->la_mode);
1594 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1598 handle = mdd_trans_create(env, mdd);
1600 RETURN(PTR_ERR(handle));
1602 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1603 lname, ma, handle, no_name, is_dir);
1607 rc = mdd_trans_start(env, mdd, handle);
1611 if (likely(mdd_cobj != NULL))
1612 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1614 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1615 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1616 mdd_object_capa(env, mdd_pobj));
1621 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1622 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1623 GOTO(cleanup, rc = 0);
1625 if (likely(mdd_cobj != NULL)) {
1626 rc = mdo_ref_del(env, mdd_cobj, handle);
1628 __mdd_index_insert_only(env, mdd_pobj,
1630 mdd_object_type(mdd_cobj),
1632 mdd_object_capa(env, mdd_pobj));
1638 mdo_ref_del(env, mdd_cobj, handle);
1640 /* fetch updated nlink */
1641 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1646 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1647 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1649 la->la_valid = LA_CTIME | LA_MTIME;
1650 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1654 /* Enough for only unlink the entry */
1655 if (unlikely(mdd_cobj == NULL))
1658 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1659 /* update ctime of an unlinked file only if it is still
1660 * opened or a link still exists */
1661 la->la_valid = LA_CTIME;
1662 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1667 /* XXX: this transfer to ma will be removed with LOD/OSP */
1668 ma->ma_attr = *cattr;
1669 ma->ma_valid |= MA_INODE;
1670 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1672 /* fetch updated nlink */
1674 rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA);
1676 /* if object is removed then we can't get its attrs, use last get */
1677 if (cattr->la_nlink == 0) {
1678 ma->ma_attr = *cattr;
1679 ma->ma_valid |= MA_INODE;
1683 if (likely(mdd_cobj != NULL))
1684 mdd_write_unlock(env, mdd_cobj);
1689 if (cattr->la_nlink == 0) {
1690 cl_flags |= CLF_UNLINK_LAST;
1691 /* search for an existing archive */
1692 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1693 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1696 rc = mdd_changelog_ns_store(env, mdd,
1697 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1698 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1703 mdd_trans_stop(env, mdd, rc, handle);
1709 * The permission has been checked when obj created, no need check again.
1711 static int mdd_cd_sanity_check(const struct lu_env *env,
1712 struct mdd_object *obj)
1717 if (!obj || mdd_is_dead_obj(obj))
1723 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1724 struct md_object *cobj, const struct md_op_spec *spec,
1727 struct mdd_device *mdd = mdo2mdd(cobj);
1728 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1729 struct mdd_object *son = md2mdd_obj(cobj);
1730 struct thandle *handle;
1731 const struct lu_buf *buf;
1732 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1733 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1737 rc = mdd_cd_sanity_check(env, son);
1741 if (!md_should_create(spec->sp_cr_flags))
1745 * there are following use cases for this function:
1746 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1747 * striping can be specified or not
1750 rc = mdd_la_get(env, son, attr, BYPASS_CAPA);
1754 /* calling ->ah_make_hint() is used to transfer information from parent */
1755 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1757 handle = mdd_trans_create(env, mdd);
1759 GOTO(out_free, rc = PTR_ERR(handle));
1762 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1763 * Should this be fixed?
1765 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1766 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1767 spec->sp_cr_flags, spec->no_create);
1769 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1770 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1771 spec->u.sp_ea.eadatalen);
1776 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1777 XATTR_NAME_LOV, 0, handle);
1781 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1785 rc = mdd_trans_start(env, mdd, handle);
1789 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1790 0, handle, mdd_object_capa(env, son));
1795 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1798 mdd_trans_stop(env, mdd, rc, handle);
1803 static int mdd_declare_object_initialize(const struct lu_env *env,
1804 struct mdd_object *parent,
1805 struct mdd_object *child,
1806 struct lu_attr *attr,
1807 struct thandle *handle)
1813 * inode mode has been set in creation time, and it's based on umask,
1814 * la_mode and acl, don't set here again! (which will go wrong
1815 * because below function doesn't consider umask).
1816 * I'd suggest set all object attributes in creation time, see above.
1818 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1819 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1820 rc = mdo_declare_attr_set(env, child, attr, handle);
1821 attr->la_valid |= LA_MODE | LA_TYPE;
1822 if (rc != 0 || !S_ISDIR(attr->la_mode))
1825 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1830 rc = mdo_declare_ref_add(env, child, handle);
1834 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1840 static int mdd_object_initialize(const struct lu_env *env,
1841 const struct lu_fid *pfid,
1842 struct mdd_object *child,
1843 struct lu_attr *attr, struct thandle *handle,
1844 const struct md_op_spec *spec)
1849 if (S_ISDIR(attr->la_mode)) {
1850 /* Add "." and ".." for newly created dir */
1851 mdo_ref_add(env, child, handle);
1852 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1853 S_IFDIR, dot, handle, BYPASS_CAPA);
1855 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1859 mdo_ref_del(env, child, handle);
1866 * This function checks whether it can create a file/dir under the
1867 * directory(@pobj). The directory(@pobj) is not being locked by
1870 * \param[in] env execution environment
1871 * \param[in] pobj the directory to create files
1872 * \param[in] pattr the attributes of the directory
1873 * \param[in] lname the name of the created file/dir
1874 * \param[in] cattr the attributes of the file/dir
1875 * \param[in] spec create specification
1877 * \retval = 0 it is allowed to create file/dir under
1879 * \retval negative error not allowed to create file/dir
1880 * under the directory
1882 static int mdd_create_sanity_check(const struct lu_env *env,
1883 struct md_object *pobj,
1884 const struct lu_attr *pattr,
1885 const struct lu_name *lname,
1886 struct lu_attr *cattr,
1887 struct md_op_spec *spec)
1889 struct mdd_thread_info *info = mdd_env_info(env);
1890 struct lu_fid *fid = &info->mti_fid;
1891 struct mdd_object *obj = md2mdd_obj(pobj);
1892 struct mdd_device *m = mdo2mdd(pobj);
1893 bool check_perm = true;
1898 if (mdd_is_dead_obj(obj))
1902 * In some cases this lookup is not needed - we know before if name
1903 * exists or not because MDT performs lookup for it.
1904 * name length check is done in lookup.
1906 if (spec->sp_cr_lookup) {
1908 * Check if the name already exist, though it will be checked in
1909 * _index_insert also, for avoiding rolling back if exists
1912 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1913 MAY_WRITE | MAY_EXEC);
1915 RETURN(rc ? : -EEXIST);
1917 /* Permission is already being checked in mdd_lookup */
1921 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1926 if (pattr->la_mode & S_ISGID) {
1927 cattr->la_gid = pattr->la_gid;
1928 if (S_ISDIR(cattr->la_mode)) {
1929 cattr->la_mode |= S_ISGID;
1930 cattr->la_valid |= LA_MODE;
1934 rc = mdd_name_check(m, lname);
1938 switch (cattr->la_mode & S_IFMT) {
1940 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1942 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1943 RETURN(-ENAMETOOLONG);
1962 static int mdd_declare_object_create(const struct lu_env *env,
1963 struct mdd_device *mdd,
1964 struct mdd_object *p, struct mdd_object *c,
1965 struct lu_attr *attr,
1966 struct thandle *handle,
1967 const struct md_op_spec *spec,
1968 struct lu_buf *def_acl_buf,
1969 struct lu_buf *acl_buf,
1970 struct dt_allocation_hint *hint)
1974 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
1979 #ifdef CONFIG_FS_POSIX_ACL
1980 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
1981 /* if dir, then can inherit default ACl */
1982 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
1983 XATTR_NAME_ACL_DEFAULT,
1989 if (acl_buf->lb_len > 0) {
1990 rc = mdo_declare_attr_set(env, c, attr, handle);
1994 rc = mdo_declare_xattr_set(env, c, acl_buf,
1995 XATTR_NAME_ACL_ACCESS, 0, handle);
2000 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2004 /* replay case, create LOV EA from client data */
2005 if (spec->no_create ||
2006 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2007 const struct lu_buf *buf;
2009 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2010 spec->u.sp_ea.eadatalen);
2011 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2017 if (S_ISLNK(attr->la_mode)) {
2018 const char *target_name = spec->u.sp_symname;
2019 int sym_len = strlen(target_name);
2020 const struct lu_buf *buf;
2022 buf = mdd_buf_get_const(env, target_name, sym_len);
2023 rc = dt_declare_record_write(env, mdd_object_child(c),
2032 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2033 struct mdd_object *p, struct mdd_object *c,
2034 const struct lu_name *name,
2035 struct lu_attr *attr,
2036 struct thandle *handle,
2037 const struct md_op_spec *spec,
2038 struct linkea_data *ldata,
2039 struct lu_buf *def_acl_buf,
2040 struct lu_buf *acl_buf,
2041 struct dt_allocation_hint *hint)
2045 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2046 def_acl_buf, acl_buf, hint);
2050 if (S_ISDIR(attr->la_mode)) {
2051 rc = mdo_declare_ref_add(env, p, handle);
2056 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2057 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2061 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2063 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2064 name->ln_name, handle);
2068 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2073 la->la_valid = LA_CTIME | LA_MTIME;
2074 rc = mdo_declare_attr_set(env, p, la, handle);
2078 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2083 /* XXX: For remote create, it should indicate the remote RPC
2084 * will be sent after local transaction is finished, which
2085 * is not very nice, but it will be removed once we fully support
2087 if (mdd_object_remote(p) && handle->th_update != NULL)
2088 handle->th_update->tu_sent_after_local_trans = 1;
2093 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2094 struct lu_attr *la, struct lu_buf *def_acl_buf,
2095 struct lu_buf *acl_buf)
2100 if (S_ISLNK(la->la_mode)) {
2101 acl_buf->lb_len = 0;
2102 def_acl_buf->lb_len = 0;
2106 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2107 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2108 XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA);
2109 mdd_read_unlock(env, pobj);
2111 /* If there are default ACL, fix mode/ACL by default ACL */
2112 def_acl_buf->lb_len = rc;
2113 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2114 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2115 acl_buf->lb_len = rc;
2116 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2119 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2120 /* If there are no default ACL, fix mode by mask */
2121 struct lu_ucred *uc = lu_ucred(env);
2123 /* The create triggered by MDT internal events, such as
2124 * LFSCK reset, will not contain valid "uc". */
2125 if (unlikely(uc != NULL))
2126 la->la_mode &= ~uc->uc_umask;
2128 acl_buf->lb_len = 0;
2129 def_acl_buf->lb_len = 0;
2136 * Create a metadata object and initialize it, set acl, xattr.
2138 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2139 struct mdd_object *son, struct lu_attr *attr,
2140 struct md_op_spec *spec, struct lu_buf *acl_buf,
2141 struct lu_buf *def_acl_buf,
2142 struct dt_allocation_hint *hint,
2143 struct thandle *handle)
2147 mdd_write_lock(env, son, MOR_TGT_CHILD);
2148 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2153 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2154 * created in declare phase, they also needs to be added to master
2155 * object as sub-directory entry. So it has to initialize the master
2156 * object, then set dir striped EA.(in mdo_xattr_set) */
2157 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2160 GOTO(err_destroy, rc);
2163 * in case of replay we just set LOVEA provided by the client
2164 * XXX: I think it would be interesting to try "old" way where
2165 * MDT calls this xattr_set(LOV) in a different transaction.
2166 * probably this way we code can be made better.
2169 /* During creation, there are only a few cases we need do xattr_set to
2171 * 1. regular file: see comments above.
2172 * 2. dir: inherit default striping or pool settings from parent.
2173 * 3. create striped directory with provided stripeEA.
2174 * 4. create striped directory because inherit default layout from the
2177 if (spec->no_create ||
2178 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2179 S_ISDIR(attr->la_mode)) {
2180 const struct lu_buf *buf;
2182 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2183 spec->u.sp_ea.eadatalen);
2184 rc = mdo_xattr_set(env, son, buf,
2185 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2187 handle, BYPASS_CAPA);
2189 GOTO(err_destroy, rc);
2192 #ifdef CONFIG_FS_POSIX_ACL
2193 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2194 S_ISDIR(attr->la_mode)) {
2195 /* set default acl */
2196 rc = mdo_xattr_set(env, son, def_acl_buf,
2197 XATTR_NAME_ACL_DEFAULT, 0,
2198 handle, BYPASS_CAPA);
2200 GOTO(err_destroy, rc);
2202 /* set its own acl */
2203 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2204 rc = mdo_xattr_set(env, son, acl_buf,
2205 XATTR_NAME_ACL_ACCESS,
2206 0, handle, BYPASS_CAPA);
2208 GOTO(err_destroy, rc);
2212 if (S_ISLNK(attr->la_mode)) {
2213 struct lu_ucred *uc = lu_ucred_assert(env);
2214 struct dt_object *dt = mdd_object_child(son);
2215 const char *target_name = spec->u.sp_symname;
2216 int sym_len = strlen(target_name);
2217 const struct lu_buf *buf;
2220 buf = mdd_buf_get_const(env, target_name, sym_len);
2221 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2222 mdd_object_capa(env, son),
2224 CFS_CAP_SYS_RESOURCE_MASK);
2229 GOTO(err_initlized, rc = -EFAULT);
2233 if (unlikely(rc != 0)) {
2235 if (S_ISDIR(attr->la_mode)) {
2236 /* Drop the reference, no need to delete "."/"..",
2237 * because the object to be destroied directly. */
2238 rc2 = mdo_ref_del(env, son, handle);
2242 rc2 = mdo_ref_del(env, son, handle);
2246 mdo_destroy(env, son, handle);
2249 mdd_write_unlock(env, son);
2254 * Create object and insert it into namespace.
2256 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2257 const struct lu_name *lname, struct md_object *child,
2258 struct md_op_spec *spec, struct md_attr* ma)
2260 struct mdd_thread_info *info = mdd_env_info(env);
2261 struct lu_attr *la = &info->mti_la_for_fix;
2262 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2263 struct mdd_object *son = md2mdd_obj(child);
2264 struct mdd_device *mdd = mdo2mdd(pobj);
2265 struct lu_attr *attr = &ma->ma_attr;
2266 struct thandle *handle;
2267 struct lu_attr *pattr = &info->mti_pattr;
2268 struct lu_buf acl_buf;
2269 struct lu_buf def_acl_buf;
2270 struct linkea_data *ldata = &info->mti_link_data;
2271 const char *name = lname->ln_name;
2272 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2278 * Two operations have to be performed:
2280 * - an allocation of a new object (->do_create()), and
2282 * - an insertion into a parent index (->dio_insert()).
2284 * Due to locking, operation order is not important, when both are
2285 * successful, *but* error handling cases are quite different:
2287 * - if insertion is done first, and following object creation fails,
2288 * insertion has to be rolled back, but this operation might fail
2289 * also leaving us with dangling index entry.
2291 * - if creation is done first, is has to be undone if insertion
2292 * fails, leaving us with leaked space, which is neither good, nor
2295 * It seems that creation-first is simplest solution, but it is
2296 * sub-optimal in the frequent
2301 * case, because second mkdir is bound to create object, only to
2302 * destroy it immediately.
2304 * To avoid this follow local file systems that do double lookup:
2306 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2308 * 1. create (mdd_object_create_internal())
2310 * 2. insert (__mdd_index_insert(), lookup again)
2313 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
2317 /* Sanity checks before big job. */
2318 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2322 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2323 GOTO(out_free, rc = -EINPROGRESS);
2325 handle = mdd_trans_create(env, mdd);
2327 GOTO(out_free, rc = PTR_ERR(handle));
2329 acl_buf.lb_buf = info->mti_xattr_buf;
2330 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2331 def_acl_buf.lb_buf = info->mti_key;
2332 def_acl_buf.lb_len = sizeof(info->mti_key);
2333 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2337 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2339 memset(ldata, 0, sizeof(*ldata));
2340 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2341 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2344 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2345 &tfid, lname, 1, 0, ldata);
2347 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2348 mdd_object_fid(mdd_pobj),
2349 lname, 1, 0, ldata);
2352 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2353 handle, spec, ldata, &def_acl_buf, &acl_buf,
2358 rc = mdd_trans_start(env, mdd, handle);
2362 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2363 &def_acl_buf, hint, handle);
2367 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2368 mdd_write_lock(env, son, MOR_TGT_CHILD);
2369 rc = __mdd_orphan_add(env, son, handle);
2370 GOTO(out_volatile, rc);
2372 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2373 attr->la_mode, name, handle,
2374 mdd_object_capa(env, mdd_pobj));
2376 GOTO(err_created, rc);
2378 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2381 /* update parent directory mtime/ctime */
2383 la->la_valid = LA_CTIME | LA_MTIME;
2384 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2386 GOTO(err_insert, rc);
2394 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2395 rc2 = __mdd_orphan_del(env, son, handle);
2397 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2398 S_ISDIR(attr->la_mode),
2399 handle, BYPASS_CAPA);
2404 mdd_write_lock(env, son, MOR_TGT_CHILD);
2405 if (S_ISDIR(attr->la_mode)) {
2406 /* Drop the reference, no need to delete "."/"..",
2407 * because the object is to be destroyed directly. */
2408 rc2 = mdo_ref_del(env, son, handle);
2410 mdd_write_unlock(env, son);
2415 /* For volatile files drop one link immediately, since there is
2416 * no filename in the namespace, and save any error returned. */
2417 rc2 = mdo_ref_del(env, son, handle);
2419 mdd_write_unlock(env, son);
2420 if (unlikely(rc == 0))
2425 /* Don't destroy the volatile object on success */
2426 if (likely(rc != 0))
2427 mdo_destroy(env, son, handle);
2428 mdd_write_unlock(env, son);
2431 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2432 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2433 rc = mdd_changelog_ns_store(env, mdd,
2434 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2435 S_ISREG(attr->la_mode) ? CL_CREATE :
2436 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2437 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
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,
2608 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2613 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2614 mdd_object_type(mdd_sobj),
2615 tname->ln_name, handle);
2619 /* name from target dir (old name), we declare it unconditionally
2620 * as mdd_rename() calls delete unconditionally as well. so just
2621 * to balance declarations vs calls to change ... */
2622 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2626 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2627 /* delete target child in target parent directory */
2628 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2632 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2633 /* target child can be directory,
2634 * delete "." reference in target child directory */
2635 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2639 /* delete ".." reference in target parent directory */
2640 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2645 la->la_valid = LA_CTIME;
2646 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2650 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2655 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2662 /* src object can be remote that is why we use only fid and type of object */
2663 static int mdd_rename(const struct lu_env *env,
2664 struct md_object *src_pobj, struct md_object *tgt_pobj,
2665 const struct lu_fid *lf, const struct lu_name *lsname,
2666 struct md_object *tobj, const struct lu_name *ltname,
2669 const char *sname = lsname->ln_name;
2670 const char *tname = ltname->ln_name;
2671 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2672 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2673 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2674 struct mdd_device *mdd = mdo2mdd(src_pobj);
2675 struct mdd_object *mdd_sobj = NULL; /* source object */
2676 struct mdd_object *mdd_tobj = NULL;
2677 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2678 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2679 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2680 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2681 struct thandle *handle;
2682 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2683 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2684 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2687 bool tobj_locked = 0;
2688 unsigned cl_flags = 0;
2693 mdd_tobj = md2mdd_obj(tobj);
2695 mdd_sobj = mdd_object_find(env, mdd, lf);
2697 rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA);
2699 GOTO(out_pending, rc);
2701 rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA);
2703 GOTO(out_pending, rc);
2706 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2708 GOTO(out_pending, rc);
2711 rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA);
2713 GOTO(out_pending, rc);
2715 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2716 mdd_sobj, cattr, mdd_tobj, tattr);
2718 GOTO(out_pending, rc);
2720 rc = mdd_name_check(mdd, ltname);
2722 GOTO(out_pending, rc);
2724 handle = mdd_trans_create(env, mdd);
2726 GOTO(out_pending, rc = PTR_ERR(handle));
2728 memset(ldata, 0, sizeof(*ldata));
2729 mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj),
2730 ltname, 1, 0, ldata);
2731 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2732 mdd_tobj, lsname, ltname, ma, ldata, handle);
2736 rc = mdd_trans_start(env, mdd, handle);
2740 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2741 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2743 GOTO(cleanup_unlocked, rc);
2745 is_dir = S_ISDIR(cattr->la_mode);
2747 /* Remove source name from source directory */
2748 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2749 mdd_object_capa(env, mdd_spobj));
2753 /* "mv dir1 dir2" needs "dir1/.." link update */
2754 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2755 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2756 mdd_object_capa(env, mdd_sobj));
2758 GOTO(fixup_spobj2, rc);
2760 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2762 mdd_object_capa(env, mdd_sobj));
2764 GOTO(fixup_spobj, rc);
2767 /* Remove target name from target directory
2768 * Here tobj can be remote one, so we do index_delete unconditionally
2769 * and -ENOENT is allowed.
2771 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2772 mdd_object_capa(env, mdd_tpobj));
2775 /* tname might been renamed to something else */
2776 GOTO(fixup_spobj, rc);
2779 GOTO(fixup_spobj, rc);
2782 /* Insert new fid with target name into target dir */
2783 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2784 tname, handle, mdd_object_capa(env, mdd_tpobj));
2786 GOTO(fixup_tpobj, rc);
2788 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2789 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2791 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2793 la->la_valid = LA_CTIME;
2794 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2796 GOTO(fixup_tpobj, rc);
2799 /* Remove old target object
2800 * For tobj is remote case cmm layer has processed
2801 * and set tobj to NULL then. So when tobj is NOT NULL,
2802 * it must be local one.
2804 if (tobj && mdd_object_exists(mdd_tobj)) {
2805 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2807 if (mdd_is_dead_obj(mdd_tobj)) {
2808 /* shld not be dead, something is wrong */
2809 CERROR("tobj is dead, something is wrong\n");
2813 mdo_ref_del(env, mdd_tobj, handle);
2815 /* Remove dot reference. */
2816 if (S_ISDIR(tattr->la_mode))
2817 mdo_ref_del(env, mdd_tobj, handle);
2820 /* fetch updated nlink */
2821 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2823 CERROR("%s: Failed to get nlink for tobj "
2825 mdd2obd_dev(mdd)->obd_name,
2826 PFID(tpobj_fid), rc);
2827 GOTO(fixup_tpobj, rc);
2830 la->la_valid = LA_CTIME;
2831 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2833 CERROR("%s: Failed to set ctime for tobj "
2835 mdd2obd_dev(mdd)->obd_name,
2836 PFID(tpobj_fid), rc);
2837 GOTO(fixup_tpobj, rc);
2840 /* XXX: this transfer to ma will be removed with LOD/OSP */
2841 ma->ma_attr = *tattr;
2842 ma->ma_valid |= MA_INODE;
2843 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2846 CERROR("%s: Failed to unlink tobj "
2848 mdd2obd_dev(mdd)->obd_name,
2849 PFID(tpobj_fid), rc);
2850 GOTO(fixup_tpobj, rc);
2853 /* fetch updated nlink */
2854 rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA);
2856 CERROR("%s: Failed to get nlink for tobj "
2858 mdd2obd_dev(mdd)->obd_name,
2859 PFID(tpobj_fid), rc);
2860 GOTO(fixup_tpobj, rc);
2862 /* XXX: this transfer to ma will be removed with LOD/OSP */
2863 ma->ma_attr = *tattr;
2864 ma->ma_valid |= MA_INODE;
2866 if (tattr->la_nlink == 0) {
2867 cl_flags |= CLF_RENAME_LAST;
2868 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2869 cl_flags |= CLF_RENAME_LAST_EXISTS;
2873 la->la_valid = LA_CTIME | LA_MTIME;
2874 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2876 GOTO(fixup_tpobj, rc);
2878 if (mdd_spobj != mdd_tpobj) {
2879 la->la_valid = LA_CTIME | LA_MTIME;
2880 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2883 if (rc == 0 && mdd_sobj) {
2884 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2885 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2886 mdo2fid(mdd_tpobj), ltname, handle, NULL,
2889 /* Old files might not have EA entry */
2890 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2891 lsname, handle, NULL, 0);
2892 mdd_write_unlock(env, mdd_sobj);
2893 /* We don't fail the transaction if the link ea can't be
2894 updated -- fid2path will use alternate lookup method. */
2902 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2905 CWARN("tp obj fix error %d\n",rc2);
2907 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2908 !mdd_is_dead_obj(mdd_tobj)) {
2910 mdo_ref_add(env, mdd_tobj, handle);
2912 mdo_ref_add(env, mdd_tobj, handle);
2915 rc2 = __mdd_index_insert(env, mdd_tpobj,
2917 mdd_object_type(mdd_tobj),
2918 tname, handle, BYPASS_CAPA);
2920 CWARN("tp obj fix error: rc = %d\n", rc2);
2925 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2926 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2930 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2931 mdd2obd_dev(mdd)->obd_name, rc2);
2934 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
2935 dotdot, handle, BYPASS_CAPA);
2937 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2938 mdd2obd_dev(mdd)->obd_name, rc2);
2943 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
2944 mdd_object_type(mdd_sobj), sname,
2945 handle, BYPASS_CAPA);
2947 CWARN("sp obj fix error: rc = %d\n", rc2);
2952 mdd_write_unlock(env, mdd_tobj);
2956 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
2957 mdd_tobj, tpobj_fid, lf, spobj_fid,
2958 ltname, lsname, handle);
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 = { NULL };
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,
3028 rc = mdd_links_write(env, child, &ldata, handle);
3034 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3035 struct mdd_object *parent,
3036 struct mdd_object *child,
3037 const char *name, int namelen,
3038 struct thandle *handle)
3040 return mdd_linkea_update_child_internal(env, parent, child, name,
3041 namelen, handle, true);
3044 static int mdd_linkea_update_child(const struct lu_env *env,
3045 struct mdd_object *parent,
3046 struct mdd_object *child,
3047 const char *name, int namelen,
3048 struct thandle *handle)
3050 return mdd_linkea_update_child_internal(env, parent, child, name,
3051 namelen, handle, false);
3054 static int mdd_update_linkea_internal(const struct lu_env *env,
3055 struct mdd_object *mdd_pobj,
3056 struct mdd_object *mdd_sobj,
3057 struct mdd_object *mdd_tobj,
3058 const struct lu_name *child_name,
3059 struct thandle *handle,
3062 struct mdd_thread_info *info = mdd_env_info(env);
3063 struct linkea_data *ldata = &info->mti_link_data;
3068 rc = mdd_links_read(env, mdd_sobj, ldata);
3070 if (rc == -ENOENT || rc == -ENODATA)
3076 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3079 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3084 /* If it is mulitple links file, we need update the name entry for
3086 LASSERT(ldata->ld_leh != NULL);
3087 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3088 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3089 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3090 struct mdd_object *pobj;
3091 struct lu_name lname;
3094 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3096 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3098 pobj = mdd_object_find(env, mdd, &fid);
3100 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3101 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3106 if (!mdd_object_exists(pobj)) {
3107 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3108 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3112 if (pobj == mdd_pobj &&
3113 lname.ln_namelen == child_name->ln_namelen &&
3114 strncmp(lname.ln_name, child_name->ln_name,
3115 lname.ln_namelen) == 0) {
3116 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3117 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3122 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3123 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3124 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3127 /* Remove source name from source directory */
3128 /* Insert new fid with target name into target dir */
3129 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3134 rc = mdo_declare_index_insert(env, pobj,
3135 mdd_object_fid(mdd_tobj),
3136 mdd_object_type(mdd_tobj),
3137 lname.ln_name, handle);
3141 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3145 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3149 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3151 mdd_object_capa(env, pobj));
3155 rc = __mdd_index_insert(env, pobj,
3156 mdd_object_fid(mdd_tobj),
3157 mdd_object_type(mdd_tobj),
3158 lname.ln_name, handle,
3159 mdd_object_capa(env, pobj));
3163 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3164 rc = mdo_ref_add(env, mdd_tobj, handle);
3165 mdd_write_unlock(env, mdd_tobj);
3169 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3170 mdo_ref_del(env, mdd_sobj, handle);
3171 mdd_write_unlock(env, mdd_sobj);
3174 mdd_object_put(env, pobj);
3182 static int mdd_migrate_xattrs(const struct lu_env *env,
3183 struct mdd_object *mdd_sobj,
3184 struct mdd_object *mdd_tobj)
3186 struct mdd_thread_info *info = mdd_env_info(env);
3187 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3189 struct thandle *handle;
3195 struct lu_buf list_xbuf;
3198 /* retrieve xattr list from the old object */
3199 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL,
3200 mdd_object_capa(env, mdd_sobj));
3201 if (list_xsize == -ENODATA)
3207 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3208 if (info->mti_big_buf.lb_buf == NULL)
3211 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3212 list_xbuf.lb_len = list_xsize;
3213 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf,
3214 mdd_object_capa(env, mdd_sobj));
3219 xname = list_xbuf.lb_buf;
3221 xlen = strnlen(xname, rem - 1) + 1;
3222 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3223 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3224 strcmp(XATTR_NAME_LMV, xname) == 0)
3227 /* For directory, if there are default layout, migrate here */
3228 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3229 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3232 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL,
3234 mdd_object_capa(env, mdd_sobj));
3235 if (xsize == -ENODATA)
3240 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3241 if (info->mti_link_buf.lb_buf == NULL)
3242 GOTO(out, rc = -ENOMEM);
3244 xbuf.lb_len = xsize;
3245 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3246 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname,
3247 mdd_object_capa(env, mdd_sobj));
3253 handle = mdd_trans_create(env, mdd);
3255 GOTO(out, rc = PTR_ERR(handle));
3257 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3260 GOTO(stop_trans, rc);
3261 /* Note: this transaction is part of migration, and it is not
3262 * the last step of migration, so we set th_local = 1 to avoid
3263 * update last rcvd for this transaction */
3264 handle->th_local = 1;
3265 rc = mdd_trans_start(env, mdd, handle);
3267 GOTO(stop_trans, rc);
3269 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle,
3270 mdd_object_capa(env, mdd_sobj));
3272 GOTO(stop_trans, rc = 0);
3275 GOTO(stop_trans, rc);
3277 mdd_trans_stop(env, mdd, rc, handle);
3282 memmove(xname, xname + xlen, rem);
3288 static int mdd_declare_migrate_create(const struct lu_env *env,
3289 struct mdd_object *mdd_pobj,
3290 struct mdd_object *mdd_sobj,
3291 struct mdd_object *mdd_tobj,
3292 struct md_op_spec *spec,
3294 union lmv_mds_md *mgr_ea,
3295 struct thandle *handle)
3297 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3298 const struct lu_buf *buf;
3302 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3303 handle, spec, NULL);
3307 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3312 if (S_ISLNK(la->la_mode)) {
3313 const char *target_name = spec->u.sp_symname;
3314 int sym_len = strlen(target_name);
3315 const struct lu_buf *buf;
3317 buf = mdd_buf_get_const(env, target_name, sym_len);
3318 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3324 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3325 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3326 spec->u.sp_ea.eadatalen);
3327 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3333 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3334 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3335 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3340 la_flag->la_valid = LA_FLAGS;
3341 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3342 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3347 static int mdd_migrate_create(const struct lu_env *env,
3348 struct mdd_object *mdd_pobj,
3349 struct mdd_object *mdd_sobj,
3350 struct mdd_object *mdd_tobj,
3353 struct mdd_thread_info *info = mdd_env_info(env);
3354 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3355 struct md_op_spec *spec = &info->mti_spec;
3356 struct lu_buf lmm_buf = { NULL };
3357 struct lu_buf link_buf = { NULL };
3358 const struct lu_buf *buf;
3359 struct thandle *handle;
3360 struct lmv_mds_md_v1 *mgr_ea;
3361 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3362 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3367 /* prepare spec for create */
3368 memset(spec, 0, sizeof(*spec));
3369 spec->sp_cr_lookup = 0;
3370 spec->sp_feat = &dt_directory_features;
3371 if (S_ISLNK(la->la_mode)) {
3372 buf = lu_buf_check_and_alloc(
3373 &mdd_env_info(env)->mti_big_buf,
3376 link_buf.lb_len = la->la_size + 1;
3377 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3379 rc = rc != 0 ? rc : -EFAULT;
3380 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3381 mdd2obd_dev(mdd)->obd_name,
3382 PFID(mdd_object_fid(mdd_sobj)), rc);
3385 spec->u.sp_symname = link_buf.lb_buf;
3386 } else if S_ISREG(la->la_mode) {
3387 /* retrieve lov of the old object */
3388 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3389 if (rc != 0 && rc != -ENODATA)
3391 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3392 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3393 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3394 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3398 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3399 memset(mgr_ea, 0, sizeof(*mgr_ea));
3400 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3401 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3402 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3403 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3404 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3405 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3407 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3409 handle = mdd_trans_create(env, mdd);
3411 GOTO(out_free, rc = PTR_ERR(handle));
3413 /* Note: this transaction is part of migration, and it is not
3414 * the last step of migration, so we set th_local = 1 to avoid
3415 * update last rcvd for this transaction */
3416 handle->th_local = 1;
3417 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3419 (union lmv_mds_md *)info->mti_xattr_buf,
3422 GOTO(stop_trans, rc);
3424 rc = mdd_trans_start(env, mdd, handle);
3426 GOTO(stop_trans, rc);
3428 /* create the target object */
3429 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3432 GOTO(stop_trans, rc);
3434 /* Set MIGRATE EA on the source inode, so once the migration needs
3435 * to be re-done during failover, the re-do process can locate the
3436 * target object which is already being created. */
3437 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3438 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3439 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0,
3440 handle, mdd_object_capa(env, mdd_sobj));
3442 GOTO(stop_trans, rc);
3444 /* Set immutable flag, so any modification is disabled until
3445 * the migration is done. Once the migration is interrupted,
3446 * if the resume process find the migrating object has both
3447 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3448 * flag and approve the migration */
3449 la_flag->la_valid = LA_FLAGS;
3450 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
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 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3712 /* delete entry from source dir */
3713 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3717 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3718 mdd_tobj, handle, lname);
3722 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3723 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3729 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3730 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3736 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3737 mdd_object_type(mdd_tobj),
3738 lname->ln_name, handle);
3742 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3743 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3748 /* delete old object */
3749 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3753 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3754 /* delete old object */
3755 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3758 /* set nlink to 0 */
3759 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3764 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3768 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3773 static int mdd_migrate_update_name(const struct lu_env *env,
3774 struct mdd_object *mdd_pobj,
3775 struct mdd_object *mdd_sobj,
3776 struct mdd_object *mdd_tobj,
3777 const struct lu_name *lname,
3780 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3781 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3782 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3783 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3784 struct thandle *handle;
3785 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3786 const char *name = lname->ln_name;
3790 /* update time for parent */
3791 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3792 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3793 p_la->la_valid = LA_CTIME;
3795 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
3799 handle = mdd_trans_create(env, mdd);
3801 RETURN(PTR_ERR(handle));
3803 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3804 lname, so_attr, p_la, handle);
3806 /* If the migration can not be fit in one transaction, just
3807 * leave it in the original MDT */
3809 GOTO(stop_trans, rc = 0);
3811 GOTO(stop_trans, rc);
3814 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3815 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3816 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3817 PFID(mdd_object_fid(mdd_tobj)));
3819 rc = mdd_trans_start(env, mdd, handle);
3821 GOTO(stop_trans, rc);
3823 /* Revert IMMUTABLE flag */
3824 la_flag->la_valid = LA_FLAGS;
3825 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3826 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle,
3827 mdd_object_capa(env, mdd_pobj));
3829 GOTO(stop_trans, rc);
3831 /* Remove source name from source directory */
3832 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
3833 mdd_object_capa(env, mdd_pobj));
3835 GOTO(stop_trans, rc);
3837 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3840 GOTO(stop_trans, rc);
3842 if (S_ISREG(so_attr->la_mode)) {
3843 if (so_attr->la_nlink == 1) {
3844 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3846 mdd_object_capa(env, mdd_sobj));
3847 if (rc != 0 && rc != -ENODATA)
3848 GOTO(stop_trans, rc);
3852 /* Insert new fid with target name into target dir */
3853 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
3854 mdd_object_type(mdd_tobj), name,
3855 handle, mdd_object_capa(env, mdd_pobj));
3857 GOTO(stop_trans, rc);
3859 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3862 GOTO(stop_trans, rc);
3864 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3865 mdo_ref_del(env, mdd_sobj, handle);
3867 mdo_ref_del(env, mdd_sobj, handle);
3869 /* Get the attr again after ref_del */
3870 rc = mdd_la_get(env, mdd_sobj, so_attr,
3871 mdd_object_capa(env, mdd_sobj));
3873 GOTO(out_unlock, rc);
3875 ma->ma_attr = *so_attr;
3876 ma->ma_valid |= MA_INODE;
3877 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3879 GOTO(out_unlock, rc);
3881 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3883 GOTO(out_unlock, rc);
3886 mdd_write_unlock(env, mdd_sobj);
3889 mdd_trans_stop(env, mdd, rc, handle);
3895 * Check whether we should migrate the file/dir
3897 * < 0 permission check failed or other error.
3898 * = 0 the file can be migrated.
3899 * > 0 the file does not need to be migrated, mostly
3900 * for multiple link file
3902 static int mdd_migrate_sanity_check(const struct lu_env *env,
3903 struct mdd_object *pobj,
3904 const struct lu_attr *pattr,
3905 struct mdd_object *sobj,
3906 struct lu_attr *sattr)
3908 struct mdd_thread_info *info = mdd_env_info(env);
3909 struct linkea_data *ldata = &info->mti_link_data;
3911 struct lu_buf *mgr_buf;
3917 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3918 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3919 if (mgr_buf->lb_buf == NULL)
3922 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV,
3923 mdd_object_capa(env, sobj));
3925 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3927 /* If the object has migrateEA, it means IMMUTE flag
3928 * is being set by previous migration process, so it
3929 * needs to override the IMMUTE flag, otherwise the
3930 * following sanity check will fail */
3931 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
3932 LMV_HASH_FLAG_MIGRATION) {
3933 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3935 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3936 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3937 mdd2obd_dev(mdd)->obd_name,
3938 PFID(mdd_object_fid(sobj)));
3942 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3943 sobj, sattr, NULL, NULL);
3947 /* Then it will check if the file should be migrated. If the file
3948 * has mulitple links, we only need migrate the file if all of its
3949 * entries has been migrated to the remote MDT */
3950 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3953 rc = mdd_links_read(env, sobj, ldata);
3955 /* For multiple links files, if there are no linkEA data at all,
3956 * means the file might be created before linkEA is enabled, and
3957 * all all of its links should not be migrated yet, otherwise
3958 * it should have some linkEA there */
3959 if (rc == -ENOENT || rc == -ENODATA)
3964 /* If it is mulitple links file, we need update the name entry for
3966 LASSERT(ldata->ld_leh != NULL);
3967 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3968 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3969 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3970 struct mdd_object *lpobj;
3971 struct lu_name lname;
3974 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3976 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3978 lpobj = mdd_object_find(env, mdd, &fid);
3979 if (IS_ERR(lpobj)) {
3980 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3981 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3986 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
3987 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
3988 PFID(&fid), lname.ln_namelen, lname.ln_name);
3989 mdd_object_put(env, lpobj);
3993 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
3994 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
3995 lname.ln_name, PFID(&fid));
3996 mdd_object_put(env, lpobj);
4004 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4005 struct md_object *sobj, const struct lu_name *lname,
4006 struct md_object *tobj, struct md_attr *ma)
4008 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4009 struct mdd_device *mdd = mdo2mdd(pobj);
4010 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4011 struct mdd_object *mdd_tobj = NULL;
4012 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4013 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4017 /* If the file will being migrated, it will check whether
4018 * the file is being opened by someone else right now */
4019 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4020 if (mdd_sobj->mod_count >= 1) {
4021 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
4022 mdd2obd_dev(mdd)->obd_name,
4023 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4024 mdd_sobj->mod_count, -EBUSY);
4025 mdd_read_unlock(env, mdd_sobj);
4026 GOTO(put, rc = -EBUSY);
4028 mdd_read_unlock(env, mdd_sobj);
4030 rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj));
4034 rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA);
4038 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4045 /* Sigh, it is impossible to finish all of migration in a single
4046 * transaction, for example migrating big directory entries to the
4047 * new MDT, it needs insert all of name entries of children in the
4050 * So migration will be done in multiple steps and transactions.
4052 * 1. create an orphan object on the remote MDT in one transaction.
4053 * 2. migrate extend attributes to the new target file/directory.
4054 * 3. For directory, migrate the entries to the new MDT and update
4055 * linkEA of each children. Because we can not migrate all entries
4056 * in a single transaction, so the migrating directory will become
4057 * a striped directory during migration, so once the process is
4058 * interrupted, the directory is still accessible. (During lookup,
4059 * client will locate the name by searching both original and target
4061 * 4. Finally, update the name/FID to point to the new file/directory
4062 * in a separate transaction.
4065 /* step 1: Check whether the orphan object has been created, and create
4066 * orphan object on the remote MDT if needed */
4067 mdd_tobj = md2mdd_obj(tobj);
4068 if (!mdd_object_exists(mdd_tobj)) {
4069 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4075 /* step 2: migrate xattr */
4076 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4080 /* step 3: migrate name entries to the orphan object */
4081 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4082 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4085 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4086 OBD_FAIL_MDS_REINT_NET_REP)))
4090 /* step 4: update name entry to the new object */
4091 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4099 const struct md_dir_operations mdd_dir_ops = {
4100 .mdo_is_subdir = mdd_is_subdir,
4101 .mdo_lookup = mdd_lookup,
4102 .mdo_create = mdd_create,
4103 .mdo_rename = mdd_rename,
4104 .mdo_link = mdd_link,
4105 .mdo_unlink = mdd_unlink,
4106 .mdo_create_data = mdd_create_data,
4107 .mdo_migrate = mdd_migrate,