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, 2015, 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_exists(mdd_obj))
91 if (mdd_object_remote(mdd_obj)) {
92 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
93 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
96 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
101 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
102 dt_try_as_dir(env, dir)))
103 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
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);
122 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
123 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
128 * Get parent FID of the directory
130 * Read parent FID from linkEA, if that fails, then do lookup
131 * dotdot to get the parent FID.
133 * \param[in] env execution environment
134 * \param[in] obj object from which to find the parent FID
135 * \param[in] attr attribute of the object
136 * \param[out] fid fid to get the parent FID
138 * \retval 0 if getting the parent FID succeeds.
139 * \retval negative errno if getting the parent FID fails.
141 static inline int mdd_parent_fid(const struct lu_env *env,
142 struct mdd_object *obj,
143 const struct lu_attr *attr,
146 struct mdd_thread_info *info = mdd_env_info(env);
147 struct linkea_data ldata = { NULL };
148 struct lu_buf *buf = &info->mti_link_buf;
149 struct lu_name lname;
154 LASSERT(S_ISDIR(mdd_object_type(obj)));
156 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
157 if (buf->lb_buf == NULL)
158 GOTO(lookup, rc = 0);
161 rc = mdd_links_read(env, obj, &ldata);
165 LASSERT(ldata.ld_leh != NULL);
166 /* Directory should only have 1 parent */
167 if (ldata.ld_leh->leh_reccount > 1)
170 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
172 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
173 if (likely(fid_is_sane(fid)))
176 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
181 * For root fid use special function, which does not compare version component
182 * of fid. Version component is different for root fids on all MDTs.
184 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
186 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
187 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
191 * return 1: if lf is the fid of the ancestor of p1;
194 * return -EREMOTE: if remote object is found, in this
195 * case fid of remote object is saved to @pf;
197 * otherwise: values < 0, errors.
199 static int mdd_is_parent(const struct lu_env *env,
200 struct mdd_device *mdd,
201 struct mdd_object *p1,
202 const struct lu_attr *attr,
203 const struct lu_fid *lf,
206 struct mdd_object *parent = NULL;
211 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
212 pfid = &mdd_env_info(env)->mti_fid;
214 /* Check for root first. */
215 if (mdd_is_root(mdd, mdo2fid(p1)))
219 /* this is done recursively */
220 rc = mdd_parent_fid(env, p1, attr, pfid);
223 if (mdd_is_root(mdd, pfid))
225 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
227 if (lu_fid_eq(pfid, lf))
230 mdd_object_put(env, parent);
232 parent = mdd_object_find(env, mdd, pfid);
234 GOTO(out, rc = PTR_ERR(parent));
236 if (!mdd_object_exists(parent))
237 GOTO(out, rc = -EINVAL);
243 if (parent && !IS_ERR(parent))
244 mdd_object_put(env, parent);
249 * No permission check is needed.
251 * returns 1: if fid is ancestor of @mo;
252 * returns 0: if fid is not an ancestor of @mo;
254 * returns EREMOTE if remote object is found, fid of remote object is saved to
257 * returns < 0: if error
259 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
260 const struct lu_fid *fid, struct lu_fid *sfid)
262 struct mdd_device *mdd = mdo2mdd(mo);
263 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
267 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
270 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
274 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
278 } else if (rc == 1) {
279 /* found @fid is parent */
287 * Check that @dir contains no entries except (possibly) dot and dotdot.
292 * -ENOTDIR not a directory object
293 * -ENOTEMPTY not empty
297 static int mdd_dir_is_empty(const struct lu_env *env,
298 struct mdd_object *dir)
301 struct dt_object *obj;
302 const struct dt_it_ops *iops;
306 obj = mdd_object_child(dir);
307 if (!dt_try_as_dir(env, obj))
310 iops = &obj->do_index_ops->dio_it;
311 it = iops->init(env, obj, LUDA_64BITHASH);
313 result = iops->get(env, it, (const struct dt_key *)"");
316 for (result = 0, i = 0; result == 0 && i < 3; ++i)
317 result = iops->next(env, it);
320 else if (result == 1)
322 } else if (result == 0)
324 * Huh? Index contains no zero key?
331 result = PTR_ERR(it);
336 * Determine if the target object can be hard linked, and right now it only
337 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
338 * directory nlink count might exceed the maximum link count(see
339 * osd_object_ref_add), so it only check nlink for non-directories.
341 * \param[in] env thread environment
342 * \param[in] obj object being linked to
343 * \param[in] la attributes of \a obj
345 * \retval 0 if \a obj can be hard linked
346 * \retval negative error if \a obj is a directory or has too
349 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
350 const struct lu_attr *la)
352 struct mdd_device *m = mdd_obj2mdd_dev(obj);
357 /* Subdir count limitation can be broken through
358 * (see osd_object_ref_add), so only check non-directory here. */
359 if (!S_ISDIR(la->la_mode) &&
360 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
367 * Check whether it may create the cobj under the pobj.
369 * \param[in] env execution environment
370 * \param[in] pobj the parent directory
371 * \param[in] pattr the attribute of the parent directory
372 * \param[in] cobj the child to be created
373 * \param[in] check_perm if check WRITE|EXEC permission for parent
375 * \retval = 0 create the child under this dir is allowed
376 * \retval negative errno create the child under this dir is
379 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
380 const struct lu_attr *pattr, struct mdd_object *cobj,
386 if (cobj && mdd_object_exists(cobj))
389 if (mdd_is_dead_obj(pobj))
393 rc = mdd_permission_internal_locked(env, pobj, pattr,
394 MAY_WRITE | MAY_EXEC,
400 * Check whether can unlink from the pobj in the case of "cobj == NULL".
402 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
403 const struct lu_attr *pattr, const struct lu_attr *attr)
408 if (mdd_is_dead_obj(pobj))
411 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
414 rc = mdd_permission_internal_locked(env, pobj, pattr,
415 MAY_WRITE | MAY_EXEC,
420 if (pattr->la_flags & LUSTRE_APPEND_FL)
427 * pobj == NULL is remote ops case, under such case, pobj's
428 * VTX feature has been checked already, no need check again.
430 static inline int mdd_is_sticky(const struct lu_env *env,
431 struct mdd_object *pobj,
432 const struct lu_attr *pattr,
433 struct mdd_object *cobj,
434 const struct lu_attr *cattr)
436 struct lu_ucred *uc = lu_ucred_assert(env);
439 LASSERT(pattr != NULL);
440 if (!(pattr->la_mode & S_ISVTX) ||
441 (pattr->la_uid == uc->uc_fsuid))
445 LASSERT(cattr != NULL);
446 if (cattr->la_uid == uc->uc_fsuid)
449 return !md_capable(uc, CFS_CAP_FOWNER);
452 static int mdd_may_delete_entry(const struct lu_env *env,
453 struct mdd_object *pobj,
454 const struct lu_attr *pattr,
459 LASSERT(pobj != NULL);
460 if (!mdd_object_exists(pobj))
463 if (mdd_is_dead_obj(pobj))
468 rc = mdd_permission_internal_locked(env, pobj, pattr,
469 MAY_WRITE | MAY_EXEC,
475 if (pattr->la_flags & LUSTRE_APPEND_FL)
482 * Check whether it may delete the cobj from the pobj.
485 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
486 const struct lu_attr *tpattr, struct mdd_object *tobj,
487 const struct lu_attr *tattr, const struct lu_attr *cattr,
488 int check_perm, int check_empty)
494 LASSERT(tpattr != NULL);
495 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
503 if (!mdd_object_exists(tobj))
506 if (mdd_is_dead_obj(tobj))
509 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
512 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
515 /* additional check the rename case */
517 if (S_ISDIR(cattr->la_mode)) {
518 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
520 if (!S_ISDIR(tattr->la_mode))
523 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
525 } else if (S_ISDIR(tattr->la_mode))
529 if (S_ISDIR(tattr->la_mode) && check_empty)
530 rc = mdd_dir_is_empty(env, tobj);
536 * Check whether it can create the link file(linked to @src_obj) under
537 * the target directory(@tgt_obj), and src_obj has been locked by
540 * \param[in] env execution environment
541 * \param[in] tgt_obj the target directory
542 * \param[in] tattr attributes of target directory
543 * \param[in] lname the link name
544 * \param[in] src_obj source object for link
545 * \param[in] cattr attributes for source object
547 * \retval = 0 it is allowed to create the link file under tgt_obj
548 * \retval negative error not allowed to create the link file
550 static int mdd_link_sanity_check(const struct lu_env *env,
551 struct mdd_object *tgt_obj,
552 const struct lu_attr *tattr,
553 const struct lu_name *lname,
554 struct mdd_object *src_obj,
555 const struct lu_attr *cattr)
557 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
561 if (!mdd_object_exists(src_obj))
564 if (mdd_is_dead_obj(src_obj))
567 /* Local ops, no lookup before link, check filename length here. */
568 rc = mdd_name_check(m, lname);
572 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
575 if (S_ISDIR(mdd_object_type(src_obj)))
578 LASSERT(src_obj != tgt_obj);
579 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
583 rc = __mdd_may_link(env, src_obj, cattr);
588 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
589 const char *name, struct thandle *handle)
591 struct dt_object *next = mdd_object_child(pobj);
595 if (dt_try_as_dir(env, next))
596 rc = dt_delete(env, next, (struct dt_key *)name, handle);
603 static int __mdd_index_insert_only(const struct lu_env *env,
604 struct mdd_object *pobj,
605 const struct lu_fid *lf, __u32 type,
607 struct thandle *handle)
609 struct dt_object *next = mdd_object_child(pobj);
613 if (dt_try_as_dir(env, next)) {
614 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
615 struct lu_ucred *uc = lu_ucred_check(env);
619 rec->rec_type = type;
620 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
621 rc = dt_insert(env, next, (const struct dt_rec *)rec,
622 (const struct dt_key *)name, handle,
630 /* insert named index, add reference if isdir */
631 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
632 const struct lu_fid *lf, __u32 type,
633 const char *name, struct thandle *handle)
638 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
639 if (rc == 0 && S_ISDIR(type)) {
640 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
641 mdo_ref_add(env, pobj, handle);
642 mdd_write_unlock(env, pobj);
648 /* delete named index, drop reference if isdir */
649 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
650 const char *name, int is_dir,
651 struct thandle *handle)
656 rc = __mdd_index_delete_only(env, pobj, name, handle);
657 if (rc == 0 && is_dir) {
658 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
659 mdo_ref_del(env, pobj, handle);
660 mdd_write_unlock(env, pobj);
666 static int mdd_llog_record_calc_size(const struct lu_env *env,
667 const struct lu_name *tname,
668 const struct lu_name *sname)
670 const struct lu_ucred *uc = lu_ucred(env);
671 enum changelog_rec_flags crf = 0;
672 size_t hdr_size = sizeof(struct llog_changelog_rec) -
673 sizeof(struct changelog_rec);
678 if (uc != NULL && uc->uc_jobid[0] != '\0')
681 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
682 (tname != NULL ? tname->ln_namelen : 0) +
683 (sname != NULL ? 1 + sname->ln_namelen : 0));
686 int mdd_declare_changelog_store(const struct lu_env *env,
687 struct mdd_device *mdd,
688 const struct lu_name *tname,
689 const struct lu_name *sname,
690 struct thandle *handle)
692 struct obd_device *obd = mdd2obd_dev(mdd);
693 struct llog_ctxt *ctxt;
694 struct llog_changelog_rec *rec;
696 struct thandle *llog_th;
701 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
704 reclen = mdd_llog_record_calc_size(env, tname, sname);
705 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
706 if (buf->lb_buf == NULL)
710 rec->cr_hdr.lrh_len = reclen;
711 rec->cr_hdr.lrh_type = CHANGELOG_REC;
713 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
717 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
719 GOTO(out_put, rc = PTR_ERR(llog_th));
721 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
729 /** Add a changelog entry \a rec to the changelog llog
732 * \param handle - currently ignored since llogs start their own transaction;
733 * this will hopefully be fixed in llog rewrite
736 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
737 struct llog_changelog_rec *rec, struct thandle *th)
739 struct obd_device *obd = mdd2obd_dev(mdd);
740 struct llog_ctxt *ctxt;
741 struct thandle *llog_th;
744 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
745 changelog_rec_varsize(&rec->cr));
747 /* llog_lvfs_write_rec sets the llog tail len */
748 rec->cr_hdr.lrh_type = CHANGELOG_REC;
749 rec->cr.cr_time = cl_time();
751 spin_lock(&mdd->mdd_cl.mc_lock);
752 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
753 * but as long as the MDD transactions are ordered correctly for e.g.
754 * rename conflicts, I don't think this should matter. */
755 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
756 spin_unlock(&mdd->mdd_cl.mc_lock);
758 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
762 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
764 GOTO(out_put, rc = PTR_ERR(llog_th));
766 /* nested journal transaction */
767 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
776 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
777 const struct lu_fid *sfid,
778 const struct lu_fid *spfid,
779 const struct lu_name *sname)
781 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
782 size_t extsize = sname->ln_namelen + 1;
784 LASSERT(sfid != NULL);
785 LASSERT(spfid != NULL);
786 LASSERT(sname != NULL);
788 rnm->cr_sfid = *sfid;
789 rnm->cr_spfid = *spfid;
791 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
792 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
793 rec->cr_namelen += extsize;
796 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
798 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
800 if (jobid == NULL || jobid[0] == '\0')
803 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
806 /** Store a namespace change changelog record
807 * If this fails, we must fail the whole transaction; we don't
808 * want the change to commit without the log entry.
809 * \param target - mdd_object of change
810 * \param tpfid - target parent dir/object fid
811 * \param sfid - source object fid
812 * \param spfid - source parent fid
813 * \param tname - target name string
814 * \param sname - source name string
815 * \param handle - transaction handle
817 int mdd_changelog_ns_store(const struct lu_env *env,
818 struct mdd_device *mdd,
819 enum changelog_rec_type type,
820 enum changelog_rec_flags crf,
821 struct mdd_object *target,
822 const struct lu_fid *tpfid,
823 const struct lu_fid *sfid,
824 const struct lu_fid *spfid,
825 const struct lu_name *tname,
826 const struct lu_name *sname,
827 struct thandle *handle)
829 const struct lu_ucred *uc = lu_ucred(env);
830 struct llog_changelog_rec *rec;
837 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
840 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
843 LASSERT(tpfid != NULL);
844 LASSERT(tname != NULL);
845 LASSERT(handle != NULL);
847 reclen = mdd_llog_record_calc_size(env, tname, sname);
848 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
849 if (buf->lb_buf == NULL)
855 if (uc != NULL && uc->uc_jobid[0] != '\0')
863 rec->cr.cr_flags = crf;
864 rec->cr.cr_type = (__u32)type;
865 rec->cr.cr_pfid = *tpfid;
866 rec->cr.cr_namelen = tname->ln_namelen;
867 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
869 if (crf & CLF_RENAME)
870 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
873 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
875 if (likely(target != NULL)) {
876 rec->cr.cr_tfid = *mdo2fid(target);
877 target->mod_cltime = cfs_time_current_64();
879 fid_zero(&rec->cr.cr_tfid);
882 rc = mdd_changelog_store(env, mdd, rec, handle);
884 CERROR("%s: cannot store changelog record: type = %d, "
885 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
886 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
887 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
894 static int __mdd_links_add(const struct lu_env *env,
895 struct mdd_object *mdd_obj,
896 struct linkea_data *ldata,
897 const struct lu_name *lname,
898 const struct lu_fid *pfid,
899 int first, int check)
903 if (ldata->ld_leh == NULL) {
904 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
908 rc = linkea_data_new(ldata,
909 &mdd_env_info(env)->mti_link_buf);
916 rc = linkea_links_find(ldata, lname, pfid);
917 if (rc && rc != -ENOENT)
923 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
924 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
928 linkea_add_buf(ldata, lname, tfid);
931 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
932 linkea_add_buf(ldata, lname, pfid);
934 return linkea_add_buf(ldata, lname, pfid);
937 static int __mdd_links_del(const struct lu_env *env,
938 struct mdd_object *mdd_obj,
939 struct linkea_data *ldata,
940 const struct lu_name *lname,
941 const struct lu_fid *pfid)
945 if (ldata->ld_leh == NULL) {
946 rc = mdd_links_read(env, mdd_obj, ldata);
951 rc = linkea_links_find(ldata, lname, pfid);
955 linkea_del_buf(ldata, lname);
959 static int mdd_linkea_prepare(const struct lu_env *env,
960 struct mdd_object *mdd_obj,
961 const struct lu_fid *oldpfid,
962 const struct lu_name *oldlname,
963 const struct lu_fid *newpfid,
964 const struct lu_name *newlname,
965 int first, int check,
966 struct linkea_data *ldata)
972 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
975 LASSERT(oldpfid != NULL || newpfid != NULL);
977 if (mdd_obj->mod_flags & DEAD_OBJ) {
978 /* Prevent linkea to be updated which is NOT necessary. */
979 ldata->ld_reclen = 0;
980 /* No more links, don't bother */
984 if (oldpfid != NULL) {
985 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
988 (rc != -ENODATA && rc != -ENOENT))
990 /* No changes done. */
995 /* If renaming, add the new record */
996 if (newpfid != NULL) {
997 /* even if the add fails, we still delete the out-of-date
999 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1003 rc = rc != 0 ? rc : rc2;
1008 int mdd_links_rename(const struct lu_env *env,
1009 struct mdd_object *mdd_obj,
1010 const struct lu_fid *oldpfid,
1011 const struct lu_name *oldlname,
1012 const struct lu_fid *newpfid,
1013 const struct lu_name *newlname,
1014 struct thandle *handle,
1015 struct linkea_data *ldata,
1016 int first, int check)
1021 if (ldata == NULL) {
1022 ldata = &mdd_env_info(env)->mti_link_data;
1023 memset(ldata, 0, sizeof(*ldata));
1024 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1025 newpfid, newlname, first, check,
1031 if (ldata->ld_reclen != 0)
1032 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1037 if (rc == -EOVERFLOW || rc == -ENOSPC)
1039 if (newlname == NULL)
1040 CDEBUG(error ? D_ERROR : D_OTHER,
1041 "link_ea add failed %d "DFID"\n",
1042 rc, PFID(mdd_object_fid(mdd_obj)));
1043 else if (oldpfid == NULL)
1044 CDEBUG(error ? D_ERROR : D_OTHER,
1045 "link_ea add '%.*s' failed %d "DFID"\n",
1046 newlname->ln_namelen, newlname->ln_name,
1047 rc, PFID(mdd_object_fid(mdd_obj)));
1048 else if (newpfid == NULL)
1049 CDEBUG(error ? D_ERROR : D_OTHER,
1050 "link_ea del '%.*s' failed %d "DFID"\n",
1051 oldlname->ln_namelen, oldlname->ln_name,
1052 rc, PFID(mdd_object_fid(mdd_obj)));
1054 CDEBUG(error ? D_ERROR : D_OTHER,
1055 "link_ea rename '%.*s'->'%.*s' failed %d "
1057 oldlname->ln_namelen, oldlname->ln_name,
1058 newlname->ln_namelen, newlname->ln_name,
1059 rc, PFID(mdd_object_fid(mdd_obj)));
1062 if (is_vmalloc_addr(ldata->ld_buf))
1063 /* if we vmalloced a large buffer drop it */
1064 lu_buf_free(ldata->ld_buf);
1069 static inline int mdd_links_add(const struct lu_env *env,
1070 struct mdd_object *mdd_obj,
1071 const struct lu_fid *pfid,
1072 const struct lu_name *lname,
1073 struct thandle *handle,
1074 struct linkea_data *ldata, int first)
1076 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1077 pfid, lname, handle, ldata, first, 0);
1080 static inline int mdd_links_del(const struct lu_env *env,
1081 struct mdd_object *mdd_obj,
1082 const struct lu_fid *pfid,
1083 const struct lu_name *lname,
1084 struct thandle *handle)
1086 return mdd_links_rename(env, mdd_obj, pfid, lname,
1087 NULL, NULL, handle, NULL, 0, 0);
1090 /** Read the link EA into a temp buffer.
1091 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1092 * A pointer to the buffer is stored in \a ldata::ld_buf.
1094 * \retval 0 or error
1096 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1097 struct linkea_data *ldata)
1101 if (!mdd_object_exists(mdd_obj))
1104 /* First try a small buf */
1105 LASSERT(env != NULL);
1106 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1108 if (ldata->ld_buf->lb_buf == NULL)
1111 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
1112 if (rc == -ERANGE) {
1113 /* Buf was too small, figure out what we need. */
1114 lu_buf_free(ldata->ld_buf);
1115 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1119 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1120 if (ldata->ld_buf->lb_buf == NULL)
1122 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1126 lu_buf_free(ldata->ld_buf);
1127 ldata->ld_buf = NULL;
1131 return linkea_init(ldata);
1134 /** Read the link EA into a temp buffer.
1135 * Uses the name_buf since it is generally large.
1136 * \retval IS_ERR err
1137 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1139 struct lu_buf *mdd_links_get(const struct lu_env *env,
1140 struct mdd_object *mdd_obj)
1142 struct linkea_data ldata = { NULL };
1145 rc = mdd_links_read(env, mdd_obj, &ldata);
1146 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1149 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1150 struct linkea_data *ldata, struct thandle *handle)
1152 const struct lu_buf *buf;
1155 if (ldata == NULL || ldata->ld_buf == NULL ||
1156 ldata->ld_leh == NULL)
1159 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1160 ldata->ld_leh->leh_len);
1161 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1164 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1165 if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) &&
1166 mdd_object_remote(mdd_obj) == 0) {
1167 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1168 struct thandle *sub_th;
1170 /* XXX: If the linkEA is overflow, then we need to notify the
1171 * namespace LFSCK to skip "nlink" attribute verification
1172 * on this object to avoid the "nlink" to be shrinked by
1173 * wrong. It may be not good an interaction with LFSCK
1174 * like this. We will consider to replace it with other
1175 * mechanism in future. LU-5802. */
1176 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK,
1177 LFSCK_TYPE_NAMESPACE);
1179 sub_th = thandle_get_sub_by_dt(env, handle,
1180 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom);
1181 lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1188 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1189 struct thandle *handle, struct linkea_data *ldata,
1190 enum mdd_links_add_overflow overflow)
1196 if (ldata != NULL && ldata->ld_leh != NULL) {
1197 ea_len = ldata->ld_leh->leh_len;
1198 linkea = ldata->ld_buf->lb_buf;
1200 ea_len = DEFAULT_LINKEA_SIZE;
1204 /* XXX: max size? */
1205 rc = mdo_declare_xattr_set(env, mdd_obj,
1206 mdd_buf_get_const(env, linkea, ea_len),
1207 XATTR_NAME_LINK, 0, handle);
1211 if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) {
1212 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1213 struct thandle *sub_th;
1215 /* XXX: If the linkEA is overflow, then we need to notify the
1216 * namespace LFSCK to skip "nlink" attribute verification
1217 * on this object to avoid the "nlink" to be shrinked by
1218 * wrong. It may be not good an interaction with LFSCK
1219 * like this. We will consider to replace it with other
1220 * mechanism in future. LU-5802. */
1221 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE,
1222 LFSCK_TYPE_NAMESPACE);
1224 sub_th = thandle_get_sub_by_dt(env, handle,
1225 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom);
1226 rc = lfsck_in_notify(env,
1227 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1234 static inline int mdd_declare_links_del(const struct lu_env *env,
1235 struct mdd_object *c,
1236 struct thandle *handle)
1240 /* For directory, the linkEA will be removed together
1241 * with the object. */
1242 if (!S_ISDIR(mdd_object_type(c)))
1243 rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE);
1248 static int mdd_declare_link(const struct lu_env *env,
1249 struct mdd_device *mdd,
1250 struct mdd_object *p,
1251 struct mdd_object *c,
1252 const struct lu_name *name,
1253 struct thandle *handle,
1255 struct linkea_data *data)
1257 struct lu_fid tfid = *mdo2fid(c);
1260 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1261 tfid.f_oid = cfs_fail_val;
1263 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1264 name->ln_name, handle);
1268 rc = mdo_declare_ref_add(env, c, handle);
1272 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1273 rc = mdo_declare_ref_add(env, c, handle);
1278 la->la_valid = LA_CTIME | LA_MTIME;
1279 rc = mdo_declare_attr_set(env, p, la, handle);
1283 la->la_valid = LA_CTIME;
1284 rc = mdo_declare_attr_set(env, c, la, handle);
1288 rc = mdd_declare_links_add(env, c, handle, data,
1289 S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE);
1293 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1298 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1299 struct md_object *src_obj, const struct lu_name *lname,
1302 const char *name = lname->ln_name;
1303 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1304 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1305 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1306 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1307 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1308 struct mdd_device *mdd = mdo2mdd(src_obj);
1309 struct thandle *handle;
1310 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1311 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1315 rc = mdd_la_get(env, mdd_sobj, cattr);
1319 rc = mdd_la_get(env, mdd_tobj, tattr);
1323 handle = mdd_trans_create(env, mdd);
1325 GOTO(out_pending, rc = PTR_ERR(handle));
1327 memset(ldata, 0, sizeof(*ldata));
1329 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1330 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1332 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1337 rc = mdd_trans_start(env, mdd, handle);
1341 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1342 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1345 GOTO(out_unlock, rc);
1347 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1348 rc = mdo_ref_add(env, mdd_sobj, handle);
1350 GOTO(out_unlock, rc);
1353 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1354 rc = mdo_ref_add(env, mdd_sobj, handle);
1356 GOTO(out_unlock, rc);
1359 *tfid = *mdo2fid(mdd_sobj);
1360 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1361 tfid->f_oid = cfs_fail_val;
1363 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1364 mdd_object_type(mdd_sobj), name, handle);
1366 mdo_ref_del(env, mdd_sobj, handle);
1367 GOTO(out_unlock, rc);
1370 la->la_valid = LA_CTIME | LA_MTIME;
1371 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1373 GOTO(out_unlock, rc);
1375 la->la_valid = LA_CTIME;
1376 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1378 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1379 mdo2fid(mdd_tobj), lname, 0, 0,
1382 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1383 lname, handle, ldata, 0);
1384 /* The failure of links_add should not cause the link
1385 * failure, reset rc here */
1390 mdd_write_unlock(env, mdd_sobj);
1392 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1393 mdo2fid(mdd_tobj), NULL, NULL,
1394 lname, NULL, handle);
1396 mdd_trans_stop(env, mdd, rc, handle);
1398 if (is_vmalloc_addr(ldata->ld_buf))
1399 /* if we vmalloced a large buffer drop it */
1400 lu_buf_free(ldata->ld_buf);
1405 static int mdd_mark_orphan_object(const struct lu_env *env,
1406 struct mdd_object *obj, struct thandle *handle,
1409 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1412 if (!S_ISDIR(mdd_object_type(obj)))
1415 attr->la_valid = LA_FLAGS;
1416 attr->la_flags = LUSTRE_ORPHAN_FL;
1419 rc = mdo_declare_attr_set(env, obj, attr, handle);
1421 rc = mdo_attr_set(env, obj, attr, handle);
1426 static int mdd_declare_finish_unlink(const struct lu_env *env,
1427 struct mdd_object *obj,
1428 struct thandle *handle)
1432 /* Sigh, we do not know if the unlink object will become orphan in
1433 * declare phase, but fortunately the flags here does not matter
1434 * in current declare implementation */
1435 rc = mdd_mark_orphan_object(env, obj, handle, true);
1439 rc = mdo_declare_destroy(env, obj, handle);
1443 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1447 return mdd_declare_links_del(env, obj, handle);
1450 /* caller should take a lock before calling */
1451 int mdd_finish_unlink(const struct lu_env *env,
1452 struct mdd_object *obj, struct md_attr *ma,
1453 const struct mdd_object *pobj,
1454 const struct lu_name *lname,
1458 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1461 LASSERT(mdd_write_locked(env, obj) != 0);
1463 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1464 /* add new orphan and the object
1465 * will be deleted during mdd_close() */
1466 obj->mod_flags |= DEAD_OBJ;
1467 if (obj->mod_count) {
1468 rc = __mdd_orphan_add(env, obj, th);
1470 CDEBUG(D_HA, "Object "DFID" is inserted into "
1471 "orphan list, open count = %d\n",
1472 PFID(mdd_object_fid(obj)),
1475 CERROR("Object "DFID" fail to be an orphan, "
1476 "open count = %d, maybe cause failed "
1478 PFID(mdd_object_fid(obj)),
1481 /* mark object as an orphan here, not
1482 * before __mdd_orphan_add() as racing
1483 * mdd_la_get() may propagate ORPHAN_OBJ
1484 * causing the asserition */
1485 rc = mdd_mark_orphan_object(env, obj, th, false);
1487 rc = mdo_destroy(env, obj, th);
1489 } else if (!is_dir) {
1490 /* old files may not have link ea; ignore errors */
1491 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1499 * has mdd_write_lock on cobj already, but not on pobj yet
1501 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1502 const struct lu_attr *pattr,
1503 struct mdd_object *cobj,
1504 const struct lu_attr *cattr)
1509 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1514 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1515 struct mdd_object *p, struct mdd_object *c,
1516 const struct lu_name *name, struct md_attr *ma,
1517 struct thandle *handle, int no_name, int is_dir)
1519 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1522 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1523 if (likely(no_name == 0)) {
1524 rc = mdo_declare_index_delete(env, p, name->ln_name,
1531 rc = mdo_declare_ref_del(env, p, handle);
1537 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1538 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1539 la->la_valid = LA_CTIME | LA_MTIME;
1540 rc = mdo_declare_attr_set(env, p, la, handle);
1545 rc = mdo_declare_ref_del(env, c, handle);
1549 rc = mdo_declare_ref_del(env, c, handle);
1553 la->la_valid = LA_CTIME;
1554 rc = mdo_declare_attr_set(env, c, la, handle);
1558 rc = mdd_declare_finish_unlink(env, c, handle);
1562 /* FIXME: need changelog for remove entry */
1563 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1570 * test if a file has an HSM archive
1571 * if HSM attributes are not found in ma update them from
1574 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1575 struct mdd_object *obj,
1580 if (!(ma->ma_valid & MA_HSM)) {
1581 /* no HSM MD provided, read xattr */
1582 struct lu_buf *hsm_buf;
1583 const size_t buflen = sizeof(struct hsm_attrs);
1586 hsm_buf = mdd_buf_get(env, NULL, 0);
1587 lu_buf_alloc(hsm_buf, buflen);
1588 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1589 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1590 lu_buf_free(hsm_buf);
1594 ma->ma_valid |= MA_HSM;
1596 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1602 * Delete name entry and the object.
1603 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1604 * does not exist for this object, and it could only happen during resending
1605 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1606 * is also needed in this case(needed by changelog), so we have to add another
1607 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1608 * the ENOENT failure should be able to be fixed by redo mechanism.
1610 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1611 struct md_object *cobj, const struct lu_name *lname,
1612 struct md_attr *ma, int no_name)
1614 const char *name = lname->ln_name;
1615 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1616 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1617 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1618 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1619 struct mdd_object *mdd_cobj = NULL;
1620 struct mdd_device *mdd = mdo2mdd(pobj);
1621 struct thandle *handle;
1622 int rc, is_dir = 0, cl_flags = 0;
1625 /* cobj == NULL means only delete name entry */
1626 if (likely(cobj != NULL)) {
1627 mdd_cobj = md2mdd_obj(cobj);
1628 if (mdd_object_exists(mdd_cobj) == 0)
1632 rc = mdd_la_get(env, mdd_pobj, pattr);
1636 if (likely(mdd_cobj != NULL)) {
1638 rc = mdd_la_get(env, mdd_cobj, cattr);
1642 is_dir = S_ISDIR(cattr->la_mode);
1643 /* search for an existing archive.
1644 * we should check ahead as the object
1645 * can be destroyed in this transaction */
1646 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1647 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1650 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1654 handle = mdd_trans_create(env, mdd);
1656 RETURN(PTR_ERR(handle));
1658 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1659 lname, ma, handle, no_name, is_dir);
1663 rc = mdd_trans_start(env, mdd, handle);
1667 if (likely(mdd_cobj != NULL))
1668 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1670 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1671 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1676 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1677 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1678 GOTO(cleanup, rc = 0);
1680 if (likely(mdd_cobj != NULL)) {
1681 rc = mdo_ref_del(env, mdd_cobj, handle);
1683 __mdd_index_insert_only(env, mdd_pobj,
1685 mdd_object_type(mdd_cobj),
1692 mdo_ref_del(env, mdd_cobj, handle);
1694 /* fetch updated nlink */
1695 rc = mdd_la_get(env, mdd_cobj, cattr);
1700 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1701 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1703 la->la_valid = LA_CTIME | LA_MTIME;
1704 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1708 /* Enough for only unlink the entry */
1709 if (unlikely(mdd_cobj == NULL))
1712 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1713 /* update ctime of an unlinked file only if it is still
1714 * opened or a link still exists */
1715 la->la_valid = LA_CTIME;
1716 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1721 /* XXX: this transfer to ma will be removed with LOD/OSP */
1722 ma->ma_attr = *cattr;
1723 ma->ma_valid |= MA_INODE;
1724 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1728 /* fetch updated nlink */
1729 rc = mdd_la_get(env, mdd_cobj, cattr);
1730 /* if object is removed then we can't get its attrs,
1732 if (rc == -ENOENT) {
1733 cattr->la_nlink = 0;
1737 if (cattr->la_nlink == 0) {
1738 ma->ma_attr = *cattr;
1739 ma->ma_valid |= MA_INODE;
1744 if (likely(mdd_cobj != NULL))
1745 mdd_write_unlock(env, mdd_cobj);
1748 if (cattr->la_nlink == 0)
1749 cl_flags |= CLF_UNLINK_LAST;
1751 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1753 rc = mdd_changelog_ns_store(env, mdd,
1754 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1755 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1760 mdd_trans_stop(env, mdd, rc, handle);
1766 * The permission has been checked when obj created, no need check again.
1768 static int mdd_cd_sanity_check(const struct lu_env *env,
1769 struct mdd_object *obj)
1774 if (!obj || mdd_is_dead_obj(obj))
1780 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1781 struct md_object *cobj, const struct md_op_spec *spec,
1784 struct mdd_device *mdd = mdo2mdd(cobj);
1785 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1786 struct mdd_object *son = md2mdd_obj(cobj);
1787 struct thandle *handle;
1788 const struct lu_buf *buf;
1789 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1790 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1794 rc = mdd_cd_sanity_check(env, son);
1798 if (!md_should_create(spec->sp_cr_flags))
1802 * there are following use cases for this function:
1803 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1804 * striping can be specified or not
1807 rc = mdd_la_get(env, son, attr);
1811 /* calling ->ah_make_hint() is used to transfer information from parent */
1812 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1814 handle = mdd_trans_create(env, mdd);
1816 GOTO(out_free, rc = PTR_ERR(handle));
1819 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1820 * Should this be fixed?
1822 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1823 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1824 spec->sp_cr_flags, spec->no_create);
1826 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1827 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1828 spec->u.sp_ea.eadatalen);
1833 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1834 XATTR_NAME_LOV, 0, handle);
1838 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1842 rc = mdd_trans_start(env, mdd, handle);
1846 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1852 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1855 mdd_trans_stop(env, mdd, rc, handle);
1860 static int mdd_declare_object_initialize(const struct lu_env *env,
1861 struct mdd_object *parent,
1862 struct mdd_object *child,
1863 const struct lu_attr *attr,
1864 struct thandle *handle)
1869 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1870 if (!S_ISDIR(attr->la_mode))
1873 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1878 rc = mdo_declare_ref_add(env, child, handle);
1882 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1888 static int mdd_object_initialize(const struct lu_env *env,
1889 const struct lu_fid *pfid,
1890 struct mdd_object *child,
1891 struct lu_attr *attr, struct thandle *handle,
1892 const struct md_op_spec *spec)
1897 if (S_ISDIR(attr->la_mode)) {
1898 /* Add "." and ".." for newly created dir */
1899 mdo_ref_add(env, child, handle);
1900 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1901 S_IFDIR, dot, handle);
1903 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1906 mdo_ref_del(env, child, handle);
1913 * This function checks whether it can create a file/dir under the
1914 * directory(@pobj). The directory(@pobj) is not being locked by
1917 * \param[in] env execution environment
1918 * \param[in] pobj the directory to create files
1919 * \param[in] pattr the attributes of the directory
1920 * \param[in] lname the name of the created file/dir
1921 * \param[in] cattr the attributes of the file/dir
1922 * \param[in] spec create specification
1924 * \retval = 0 it is allowed to create file/dir under
1926 * \retval negative error not allowed to create file/dir
1927 * under the directory
1929 static int mdd_create_sanity_check(const struct lu_env *env,
1930 struct md_object *pobj,
1931 const struct lu_attr *pattr,
1932 const struct lu_name *lname,
1933 struct lu_attr *cattr,
1934 struct md_op_spec *spec)
1936 struct mdd_thread_info *info = mdd_env_info(env);
1937 struct lu_fid *fid = &info->mti_fid;
1938 struct mdd_object *obj = md2mdd_obj(pobj);
1939 struct mdd_device *m = mdo2mdd(pobj);
1940 bool check_perm = true;
1945 if (mdd_is_dead_obj(obj))
1949 * In some cases this lookup is not needed - we know before if name
1950 * exists or not because MDT performs lookup for it.
1951 * name length check is done in lookup.
1953 if (spec->sp_cr_lookup) {
1955 * Check if the name already exist, though it will be checked in
1956 * _index_insert also, for avoiding rolling back if exists
1959 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1960 MAY_WRITE | MAY_EXEC);
1962 RETURN(rc ? : -EEXIST);
1964 /* Permission is already being checked in mdd_lookup */
1968 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1973 if (pattr->la_mode & S_ISGID) {
1974 cattr->la_gid = pattr->la_gid;
1975 if (S_ISDIR(cattr->la_mode)) {
1976 cattr->la_mode |= S_ISGID;
1977 cattr->la_valid |= LA_MODE;
1981 rc = mdd_name_check(m, lname);
1985 switch (cattr->la_mode & S_IFMT) {
1987 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1989 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
1990 RETURN(-ENAMETOOLONG);
2009 static int mdd_declare_object_create(const struct lu_env *env,
2010 struct mdd_device *mdd,
2011 struct mdd_object *p, struct mdd_object *c,
2012 struct lu_attr *attr,
2013 struct thandle *handle,
2014 const struct md_op_spec *spec,
2015 struct lu_buf *def_acl_buf,
2016 struct lu_buf *acl_buf,
2017 struct dt_allocation_hint *hint)
2021 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
2026 #ifdef CONFIG_FS_POSIX_ACL
2027 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2028 /* if dir, then can inherit default ACl */
2029 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2030 XATTR_NAME_ACL_DEFAULT,
2036 if (acl_buf->lb_len > 0) {
2037 rc = mdo_declare_attr_set(env, c, attr, handle);
2041 rc = mdo_declare_xattr_set(env, c, acl_buf,
2042 XATTR_NAME_ACL_ACCESS, 0, handle);
2047 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2051 /* replay case, create LOV EA from client data */
2052 if (spec->no_create ||
2053 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2054 const struct lu_buf *buf;
2056 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2057 spec->u.sp_ea.eadatalen);
2058 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2064 if (S_ISLNK(attr->la_mode)) {
2065 const char *target_name = spec->u.sp_symname;
2066 int sym_len = strlen(target_name);
2067 const struct lu_buf *buf;
2069 buf = mdd_buf_get_const(env, target_name, sym_len);
2070 rc = dt_declare_record_write(env, mdd_object_child(c),
2079 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2080 struct mdd_object *p, struct mdd_object *c,
2081 const struct lu_name *name,
2082 struct lu_attr *attr,
2083 struct thandle *handle,
2084 const struct md_op_spec *spec,
2085 struct linkea_data *ldata,
2086 struct lu_buf *def_acl_buf,
2087 struct lu_buf *acl_buf,
2088 struct dt_allocation_hint *hint)
2092 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2093 def_acl_buf, acl_buf, hint);
2097 if (S_ISDIR(attr->la_mode)) {
2098 rc = mdo_declare_ref_add(env, p, handle);
2103 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2104 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2108 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2110 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2111 name->ln_name, handle);
2115 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2120 la->la_valid = LA_CTIME | LA_MTIME;
2121 rc = mdo_declare_attr_set(env, p, la, handle);
2125 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2133 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2134 struct lu_attr *la, struct lu_buf *def_acl_buf,
2135 struct lu_buf *acl_buf)
2140 if (S_ISLNK(la->la_mode)) {
2141 acl_buf->lb_len = 0;
2142 def_acl_buf->lb_len = 0;
2146 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2147 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2148 XATTR_NAME_ACL_DEFAULT);
2149 mdd_read_unlock(env, pobj);
2151 /* If there are default ACL, fix mode/ACL by default ACL */
2152 def_acl_buf->lb_len = rc;
2153 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2154 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2155 acl_buf->lb_len = rc;
2156 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2159 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2160 /* If there are no default ACL, fix mode by mask */
2161 struct lu_ucred *uc = lu_ucred(env);
2163 /* The create triggered by MDT internal events, such as
2164 * LFSCK reset, will not contain valid "uc". */
2165 if (unlikely(uc != NULL))
2166 la->la_mode &= ~uc->uc_umask;
2168 acl_buf->lb_len = 0;
2169 def_acl_buf->lb_len = 0;
2176 * Create a metadata object and initialize it, set acl, xattr.
2178 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2179 struct mdd_object *son, struct lu_attr *attr,
2180 struct md_op_spec *spec, struct lu_buf *acl_buf,
2181 struct lu_buf *def_acl_buf,
2182 struct dt_allocation_hint *hint,
2183 struct thandle *handle)
2187 mdd_write_lock(env, son, MOR_TGT_CHILD);
2188 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2193 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2194 * created in declare phase, they also needs to be added to master
2195 * object as sub-directory entry. So it has to initialize the master
2196 * object, then set dir striped EA.(in mdo_xattr_set) */
2197 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2200 GOTO(err_destroy, rc);
2203 * in case of replay we just set LOVEA provided by the client
2204 * XXX: I think it would be interesting to try "old" way where
2205 * MDT calls this xattr_set(LOV) in a different transaction.
2206 * probably this way we code can be made better.
2209 /* During creation, there are only a few cases we need do xattr_set to
2211 * 1. regular file: see comments above.
2212 * 2. dir: inherit default striping or pool settings from parent.
2213 * 3. create striped directory with provided stripeEA.
2214 * 4. create striped directory because inherit default layout from the
2217 if (spec->no_create ||
2218 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2219 S_ISDIR(attr->la_mode)) {
2220 const struct lu_buf *buf;
2222 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2223 spec->u.sp_ea.eadatalen);
2224 rc = mdo_xattr_set(env, son, buf,
2225 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2229 GOTO(err_destroy, rc);
2232 #ifdef CONFIG_FS_POSIX_ACL
2233 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2234 S_ISDIR(attr->la_mode)) {
2235 /* set default acl */
2236 rc = mdo_xattr_set(env, son, def_acl_buf,
2237 XATTR_NAME_ACL_DEFAULT, 0,
2240 GOTO(err_destroy, rc);
2242 /* set its own acl */
2243 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2244 rc = mdo_xattr_set(env, son, acl_buf,
2245 XATTR_NAME_ACL_ACCESS,
2248 GOTO(err_destroy, rc);
2252 if (S_ISLNK(attr->la_mode)) {
2253 struct lu_ucred *uc = lu_ucred_assert(env);
2254 struct dt_object *dt = mdd_object_child(son);
2255 const char *target_name = spec->u.sp_symname;
2256 int sym_len = strlen(target_name);
2257 const struct lu_buf *buf;
2260 buf = mdd_buf_get_const(env, target_name, sym_len);
2261 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2263 CFS_CAP_SYS_RESOURCE_MASK);
2268 GOTO(err_initlized, rc = -EFAULT);
2272 if (unlikely(rc != 0)) {
2274 if (S_ISDIR(attr->la_mode)) {
2275 /* Drop the reference, no need to delete "."/"..",
2276 * because the object to be destroied directly. */
2277 rc2 = mdo_ref_del(env, son, handle);
2281 rc2 = mdo_ref_del(env, son, handle);
2285 mdo_destroy(env, son, handle);
2288 mdd_write_unlock(env, son);
2292 static int mdd_index_delete(const struct lu_env *env,
2293 struct mdd_object *mdd_pobj,
2294 struct lu_attr *cattr,
2295 const struct lu_name *lname)
2297 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2298 struct thandle *handle;
2302 handle = mdd_trans_create(env, mdd);
2304 RETURN(PTR_ERR(handle));
2306 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2311 if (S_ISDIR(cattr->la_mode)) {
2312 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2317 /* Since this will only be used in the error handler path,
2318 * Let's set the thandle to be local and not mess the transno */
2319 handle->th_local = 1;
2320 rc = mdd_trans_start(env, mdd, handle);
2324 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2325 S_ISDIR(cattr->la_mode), handle);
2329 mdd_trans_stop(env, mdd, rc, handle);
2334 * Create object and insert it into namespace.
2336 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2337 const struct lu_name *lname, struct md_object *child,
2338 struct md_op_spec *spec, struct md_attr* ma)
2340 struct mdd_thread_info *info = mdd_env_info(env);
2341 struct lu_attr *la = &info->mti_la_for_fix;
2342 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2343 struct mdd_object *son = md2mdd_obj(child);
2344 struct mdd_device *mdd = mdo2mdd(pobj);
2345 struct lu_attr *attr = &ma->ma_attr;
2346 struct thandle *handle;
2347 struct lu_attr *pattr = &info->mti_pattr;
2348 struct lu_buf acl_buf;
2349 struct lu_buf def_acl_buf;
2350 struct linkea_data *ldata = &info->mti_link_data;
2351 const char *name = lname->ln_name;
2352 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2358 * Two operations have to be performed:
2360 * - an allocation of a new object (->do_create()), and
2362 * - an insertion into a parent index (->dio_insert()).
2364 * Due to locking, operation order is not important, when both are
2365 * successful, *but* error handling cases are quite different:
2367 * - if insertion is done first, and following object creation fails,
2368 * insertion has to be rolled back, but this operation might fail
2369 * also leaving us with dangling index entry.
2371 * - if creation is done first, is has to be undone if insertion
2372 * fails, leaving us with leaked space, which is neither good, nor
2375 * It seems that creation-first is simplest solution, but it is
2376 * sub-optimal in the frequent
2381 * case, because second mkdir is bound to create object, only to
2382 * destroy it immediately.
2384 * To avoid this follow local file systems that do double lookup:
2386 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2388 * 1. create (mdd_object_create_internal())
2390 * 2. insert (__mdd_index_insert(), lookup again)
2393 rc = mdd_la_get(env, mdd_pobj, pattr);
2397 /* Sanity checks before big job. */
2398 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2402 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2403 GOTO(out_free, rc = -EINPROGRESS);
2405 handle = mdd_trans_create(env, mdd);
2407 GOTO(out_free, rc = PTR_ERR(handle));
2409 acl_buf.lb_buf = info->mti_xattr_buf;
2410 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2411 def_acl_buf.lb_buf = info->mti_key;
2412 def_acl_buf.lb_len = sizeof(info->mti_key);
2413 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2417 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2419 memset(ldata, 0, sizeof(*ldata));
2420 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2421 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2424 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2425 &tfid, lname, 1, 0, ldata);
2427 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2428 mdd_object_fid(mdd_pobj),
2429 lname, 1, 0, ldata);
2432 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2433 handle, spec, ldata, &def_acl_buf, &acl_buf,
2438 rc = mdd_trans_start(env, mdd, handle);
2442 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2443 &def_acl_buf, hint, handle);
2447 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2448 mdd_write_lock(env, son, MOR_TGT_CHILD);
2449 rc = __mdd_orphan_add(env, son, handle);
2450 GOTO(out_volatile, rc);
2452 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2453 attr->la_mode, name, handle);
2455 GOTO(err_created, rc);
2457 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2460 /* update parent directory mtime/ctime */
2462 la->la_valid = LA_CTIME | LA_MTIME;
2463 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2465 GOTO(err_insert, rc);
2473 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2474 rc2 = __mdd_orphan_del(env, son, handle);
2476 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2477 S_ISDIR(attr->la_mode),
2483 mdd_write_lock(env, son, MOR_TGT_CHILD);
2484 if (S_ISDIR(attr->la_mode)) {
2485 /* Drop the reference, no need to delete "."/"..",
2486 * because the object is to be destroyed directly. */
2487 rc2 = mdo_ref_del(env, son, handle);
2489 mdd_write_unlock(env, son);
2494 /* For volatile files drop one link immediately, since there is
2495 * no filename in the namespace, and save any error returned. */
2496 rc2 = mdo_ref_del(env, son, handle);
2498 mdd_write_unlock(env, son);
2499 if (unlikely(rc == 0))
2504 /* Don't destroy the volatile object on success */
2505 if (likely(rc != 0))
2506 mdo_destroy(env, son, handle);
2507 mdd_write_unlock(env, son);
2510 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2511 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2512 rc = mdd_changelog_ns_store(env, mdd,
2513 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2514 S_ISREG(attr->la_mode) ? CL_CREATE :
2515 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2516 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2519 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2521 /* If creation fails, it is most likely due to the remote update
2522 * failure, because local transaction will mostly succeed at
2523 * this stage. There is no easy way to rollback all of previous
2524 * updates, so let's remove the object from namespace, and
2525 * LFSCK should handle the orphan object. */
2526 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2527 mdd_index_delete(env, mdd_pobj, attr, lname);
2531 if (is_vmalloc_addr(ldata->ld_buf))
2532 /* if we vmalloced a large buffer drop it */
2533 lu_buf_free(ldata->ld_buf);
2535 /* The child object shouldn't be cached anymore */
2537 set_bit(LU_OBJECT_HEARD_BANSHEE,
2538 &child->mo_lu.lo_header->loh_flags);
2543 * Get locks on parents in proper order
2544 * RETURN: < 0 - error, rename_order if successful
2552 static int mdd_rename_order(const struct lu_env *env,
2553 struct mdd_device *mdd,
2554 struct mdd_object *src_pobj,
2555 const struct lu_attr *pattr,
2556 struct mdd_object *tgt_pobj)
2558 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2562 if (src_pobj == tgt_pobj)
2563 RETURN(MDD_RN_SAME);
2565 /* compared the parent child relationship of src_p&tgt_p */
2566 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2568 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2571 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2585 /* has not mdd_write{read}_lock on any obj yet. */
2586 static int mdd_rename_sanity_check(const struct lu_env *env,
2587 struct mdd_object *src_pobj,
2588 const struct lu_attr *pattr,
2589 struct mdd_object *tgt_pobj,
2590 const struct lu_attr *tpattr,
2591 struct mdd_object *sobj,
2592 const struct lu_attr *cattr,
2593 struct mdd_object *tobj,
2594 const struct lu_attr *tattr)
2599 /* XXX: when get here, sobj must NOT be NULL,
2600 * the other case has been processed in cld_rename
2601 * before mdd_rename and enable MDS_PERM_BYPASS. */
2604 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2608 /* XXX: when get here, "tobj == NULL" means tobj must
2609 * NOT exist (neither on remote MDS, such case has been
2610 * processed in cld_rename before mdd_rename and enable
2612 * So check may_create, but not check may_unlink. */
2614 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2615 (src_pobj != tgt_pobj));
2617 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2618 (src_pobj != tgt_pobj), 1);
2620 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2621 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2626 static int mdd_declare_rename(const struct lu_env *env,
2627 struct mdd_device *mdd,
2628 struct mdd_object *mdd_spobj,
2629 struct mdd_object *mdd_tpobj,
2630 struct mdd_object *mdd_sobj,
2631 struct mdd_object *mdd_tobj,
2632 const struct lu_name *sname,
2633 const struct lu_name *tname,
2635 struct linkea_data *ldata,
2636 struct thandle *handle)
2638 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2641 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2642 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2648 /* name from source dir */
2649 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2653 /* .. from source child */
2654 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2655 /* source child can be directory,
2656 * counted by source dir's nlink */
2657 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2660 if (mdd_spobj != mdd_tpobj) {
2661 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2666 rc = mdo_declare_index_insert(env, mdd_sobj,
2668 S_IFDIR, dotdot, handle);
2673 /* new target child can be directory,
2674 * counted by target dir's nlink */
2675 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2680 la->la_valid = LA_CTIME | LA_MTIME;
2681 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2685 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2689 la->la_valid = LA_CTIME;
2690 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2694 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
2695 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2700 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2701 mdd_object_type(mdd_sobj),
2702 tname->ln_name, handle);
2706 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2707 /* delete target child in target parent directory */
2708 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2713 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2717 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2718 /* target child can be directory,
2719 * delete "." reference in target child directory */
2720 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2724 /* delete ".." reference in target parent directory */
2725 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2730 la->la_valid = LA_CTIME;
2731 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2735 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2740 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2747 /* src object can be remote that is why we use only fid and type of object */
2748 static int mdd_rename(const struct lu_env *env,
2749 struct md_object *src_pobj, struct md_object *tgt_pobj,
2750 const struct lu_fid *lf, const struct lu_name *lsname,
2751 struct md_object *tobj, const struct lu_name *ltname,
2754 const char *sname = lsname->ln_name;
2755 const char *tname = ltname->ln_name;
2756 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2757 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2758 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2759 struct mdd_device *mdd = mdo2mdd(src_pobj);
2760 struct mdd_object *mdd_sobj = NULL; /* source object */
2761 struct mdd_object *mdd_tobj = NULL;
2762 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2763 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2764 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2765 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2766 struct thandle *handle;
2767 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2768 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2769 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2772 bool tobj_locked = 0;
2773 unsigned cl_flags = 0;
2778 mdd_tobj = md2mdd_obj(tobj);
2780 mdd_sobj = mdd_object_find(env, mdd, lf);
2781 if (IS_ERR(mdd_sobj))
2782 RETURN(PTR_ERR(mdd_sobj));
2784 rc = mdd_la_get(env, mdd_sobj, cattr);
2786 GOTO(out_pending, rc);
2788 rc = mdd_la_get(env, mdd_spobj, pattr);
2790 GOTO(out_pending, rc);
2793 rc = mdd_la_get(env, mdd_tobj, tattr);
2795 GOTO(out_pending, rc);
2796 /* search for an existing archive.
2797 * we should check ahead as the object
2798 * can be destroyed in this transaction */
2799 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2800 cl_flags |= CLF_RENAME_LAST_EXISTS;
2803 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2805 GOTO(out_pending, rc);
2807 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2808 mdd_sobj, cattr, mdd_tobj, tattr);
2810 GOTO(out_pending, rc);
2812 rc = mdd_name_check(mdd, ltname);
2814 GOTO(out_pending, rc);
2816 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2817 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2819 GOTO(out_pending, rc);
2821 handle = mdd_trans_create(env, mdd);
2823 GOTO(out_pending, rc = PTR_ERR(handle));
2825 memset(ldata, 0, sizeof(*ldata));
2826 mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname,
2827 mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata);
2828 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2829 mdd_tobj, lsname, ltname, ma, ldata, handle);
2833 rc = mdd_trans_start(env, mdd, handle);
2837 is_dir = S_ISDIR(cattr->la_mode);
2839 /* Remove source name from source directory */
2840 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2844 /* "mv dir1 dir2" needs "dir1/.." link update */
2845 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2846 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2848 GOTO(fixup_spobj2, rc);
2850 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2853 GOTO(fixup_spobj, rc);
2856 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2857 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2859 /* tname might been renamed to something else */
2860 GOTO(fixup_spobj, rc);
2863 /* Insert new fid with target name into target dir */
2864 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2867 GOTO(fixup_tpobj, rc);
2869 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2870 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2872 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2873 la->la_valid = LA_CTIME;
2874 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2876 GOTO(fixup_tpobj, rc);
2878 /* Update the linkEA for the source object */
2879 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2880 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2881 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2884 /* Old files might not have EA entry */
2885 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2886 lsname, handle, NULL, 0);
2887 mdd_write_unlock(env, mdd_sobj);
2888 /* We don't fail the transaction if the link ea can't be
2889 updated -- fid2path will use alternate lookup method. */
2892 /* Remove old target object
2893 * For tobj is remote case cmm layer has processed
2894 * and set tobj to NULL then. So when tobj is NOT NULL,
2895 * it must be local one.
2897 if (tobj && mdd_object_exists(mdd_tobj)) {
2898 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2900 if (mdd_is_dead_obj(mdd_tobj)) {
2901 /* shld not be dead, something is wrong */
2902 CERROR("tobj is dead, something is wrong\n");
2906 mdo_ref_del(env, mdd_tobj, handle);
2908 /* Remove dot reference. */
2909 if (S_ISDIR(tattr->la_mode))
2910 mdo_ref_del(env, mdd_tobj, handle);
2913 /* fetch updated nlink */
2914 rc = mdd_la_get(env, mdd_tobj, tattr);
2916 CERROR("%s: Failed to get nlink for tobj "
2918 mdd2obd_dev(mdd)->obd_name,
2919 PFID(tpobj_fid), rc);
2920 GOTO(fixup_tpobj, rc);
2923 la->la_valid = LA_CTIME;
2924 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2926 CERROR("%s: Failed to set ctime for tobj "
2928 mdd2obd_dev(mdd)->obd_name,
2929 PFID(tpobj_fid), rc);
2930 GOTO(fixup_tpobj, rc);
2933 /* XXX: this transfer to ma will be removed with LOD/OSP */
2934 ma->ma_attr = *tattr;
2935 ma->ma_valid |= MA_INODE;
2936 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2939 CERROR("%s: Failed to unlink tobj "
2941 mdd2obd_dev(mdd)->obd_name,
2942 PFID(tpobj_fid), rc);
2943 GOTO(fixup_tpobj, rc);
2946 /* fetch updated nlink */
2947 rc = mdd_la_get(env, mdd_tobj, tattr);
2948 if (rc == -ENOENT) {
2949 /* the object got removed, let's
2950 * return the latest known attributes */
2951 tattr->la_nlink = 0;
2953 } else if (rc != 0) {
2954 CERROR("%s: Failed to get nlink for tobj "
2956 mdd2obd_dev(mdd)->obd_name,
2957 PFID(tpobj_fid), rc);
2958 GOTO(fixup_tpobj, rc);
2960 /* XXX: this transfer to ma will be removed with LOD/OSP */
2961 ma->ma_attr = *tattr;
2962 ma->ma_valid |= MA_INODE;
2964 if (tattr->la_nlink == 0)
2965 cl_flags |= CLF_RENAME_LAST;
2967 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
2970 la->la_valid = LA_CTIME | LA_MTIME;
2971 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2973 GOTO(fixup_tpobj, rc);
2975 if (mdd_spobj != mdd_tpobj) {
2976 la->la_valid = LA_CTIME | LA_MTIME;
2977 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2979 GOTO(fixup_tpobj, rc);
2986 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2988 CWARN("tp obj fix error %d\n",rc2);
2990 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2991 !mdd_is_dead_obj(mdd_tobj)) {
2993 mdo_ref_add(env, mdd_tobj, handle);
2995 mdo_ref_add(env, mdd_tobj, handle);
2998 rc2 = __mdd_index_insert(env, mdd_tpobj,
3000 mdd_object_type(mdd_tobj),
3003 CWARN("tp obj fix error: rc = %d\n", rc2);
3008 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3009 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3011 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3012 mdd2obd_dev(mdd)->obd_name, rc2);
3015 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3018 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3019 mdd2obd_dev(mdd)->obd_name, rc2);
3024 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3025 mdd_object_type(mdd_sobj), sname,
3028 CWARN("sp obj fix error: rc = %d\n", rc2);
3033 mdd_write_unlock(env, mdd_tobj);
3036 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3037 mdd_tobj, tpobj_fid, lf, spobj_fid,
3038 ltname, lsname, handle);
3041 mdd_trans_stop(env, mdd, rc, handle);
3044 mdd_object_put(env, mdd_sobj);
3049 * During migration once the parent FID has been changed,
3050 * we need update the parent FID in linkea.
3052 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3053 struct mdd_object *parent,
3054 struct mdd_object *newparent,
3055 struct mdd_object *child,
3056 const char *name, int namelen,
3057 struct thandle *handle,
3060 struct mdd_thread_info *info = mdd_env_info(env);
3061 struct linkea_data ldata = { NULL };
3062 struct lu_buf *buf = &info->mti_link_buf;
3068 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3069 if (buf->lb_buf == NULL)
3073 rc = mdd_links_read(env, child, &ldata);
3075 if (rc == -ENOENT || rc == -ENODATA)
3080 LASSERT(ldata.ld_leh != NULL);
3081 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3082 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3083 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3084 struct lu_name lname;
3087 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3090 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3091 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3092 ldata.ld_lee = (struct link_ea_entry *)
3093 ((char *)ldata.ld_lee +
3098 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3099 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3100 lname.ln_namelen, lname.ln_name,
3101 PFID(mdd_object_fid(newparent)));
3102 /* update to the new parent fid */
3103 linkea_entry_pack(ldata.ld_lee, &lname,
3104 mdd_object_fid(newparent));
3106 rc = mdd_declare_links_add(env, child, handle, &ldata,
3109 rc = mdd_links_write(env, child, &ldata, handle);
3115 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3116 struct mdd_object *parent,
3117 struct mdd_object *newparent,
3118 struct mdd_object *child,
3119 const char *name, int namelen,
3120 struct thandle *handle)
3122 return mdd_linkea_update_child_internal(env, parent, newparent,
3124 namelen, handle, true);
3127 static int mdd_linkea_update_child(const struct lu_env *env,
3128 struct mdd_object *parent,
3129 struct mdd_object *newparent,
3130 struct mdd_object *child,
3131 const char *name, int namelen,
3132 struct thandle *handle)
3134 return mdd_linkea_update_child_internal(env, parent, newparent,
3136 namelen, handle, false);
3139 static int mdd_update_linkea_internal(const struct lu_env *env,
3140 struct mdd_object *mdd_pobj,
3141 struct mdd_object *mdd_sobj,
3142 struct mdd_object *mdd_tobj,
3143 const struct lu_name *child_name,
3144 struct linkea_data *ldata,
3145 struct thandle *handle,
3148 struct mdd_thread_info *info = mdd_env_info(env);
3153 LASSERT(ldata->ld_buf != NULL);
3156 /* If it is mulitple links file, we need update the name entry for
3158 LASSERT(ldata->ld_leh != NULL);
3159 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3160 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3161 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3162 struct mdd_object *pobj;
3163 struct lu_name lname;
3166 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3168 pobj = mdd_object_find(env, mdd, &fid);
3170 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3171 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3173 linkea_del_buf(ldata, &lname);
3177 if (!mdd_object_exists(pobj)) {
3178 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3179 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3180 linkea_del_buf(ldata, &lname);
3181 mdd_object_put(env, pobj);
3185 if (pobj == mdd_pobj &&
3186 lname.ln_namelen == child_name->ln_namelen &&
3187 strncmp(lname.ln_name, child_name->ln_name,
3188 lname.ln_namelen) == 0) {
3189 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3190 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3192 linkea_del_buf(ldata, &lname);
3193 mdd_object_put(env, pobj);
3197 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3198 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3199 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3202 /* Remove source name from source directory */
3203 /* Insert new fid with target name into target dir */
3204 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3209 rc = mdo_declare_index_insert(env, pobj,
3210 mdd_object_fid(mdd_tobj),
3211 mdd_object_type(mdd_tobj),
3212 lname.ln_name, handle);
3216 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3220 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3224 char *tmp_name = info->mti_key;
3226 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3227 /* lnamelen is too big(> NAME_MAX + 16),
3228 * something wrong about this linkea, let's
3230 linkea_del_buf(ldata, &lname);
3231 mdd_object_put(env, pobj);
3235 /* Note: lname might be without \0 at the end, see
3236 * linkea_entry_unpack(), let's add extra \0 by
3238 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3239 lname.ln_namelen, lname.ln_name);
3240 lname.ln_name = tmp_name;
3242 /* Let's check if this linkEA still valid, before
3243 * it might be packed into the RPC buffer. */
3244 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3245 &info->mti_fid, NULL);
3247 !lu_fid_eq(&info->mti_fid,
3248 mdd_object_fid(mdd_sobj))) {
3249 /* skip invalid linkea entry */
3250 linkea_del_buf(ldata, &lname);
3251 mdd_object_put(env, pobj);
3255 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3259 rc = __mdd_index_insert(env, pobj,
3260 mdd_object_fid(mdd_tobj),
3261 mdd_object_type(mdd_tobj),
3266 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3267 rc = mdo_ref_add(env, mdd_tobj, handle);
3268 mdd_write_unlock(env, mdd_tobj);
3272 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3273 mdo_ref_del(env, mdd_sobj, handle);
3274 mdd_write_unlock(env, mdd_sobj);
3277 mdd_object_put(env, pobj);
3281 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3288 static int mdd_migrate_xattrs(const struct lu_env *env,
3289 struct mdd_object *mdd_sobj,
3290 struct mdd_object *mdd_tobj)
3292 struct mdd_thread_info *info = mdd_env_info(env);
3293 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3295 struct thandle *handle;
3301 struct lu_buf list_xbuf;
3305 /* retrieve xattr list from the old object */
3306 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3307 if (list_xsize == -ENODATA)
3313 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3314 if (info->mti_big_buf.lb_buf == NULL)
3317 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3318 list_xbuf.lb_len = list_xsize;
3319 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3324 xname = list_xbuf.lb_buf;
3326 xlen = strnlen(xname, rem - 1) + 1;
3327 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3328 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3329 strcmp(XATTR_NAME_LMV, xname) == 0)
3332 /* For directory, if there are default layout, migrate here */
3333 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3334 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3337 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3338 if (xsize == -ENODATA)
3343 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3344 if (info->mti_link_buf.lb_buf == NULL)
3345 GOTO(out, rc = -ENOMEM);
3347 xbuf.lb_len = xsize;
3348 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3349 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3355 handle = mdd_trans_create(env, mdd);
3357 GOTO(out, rc = PTR_ERR(handle));
3359 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3362 GOTO(stop_trans, rc);
3363 /* Note: this transaction is part of migration, and it is not
3364 * the last step of migration, so we set th_local = 1 to avoid
3365 * update last rcvd for this transaction */
3366 handle->th_local = 1;
3367 rc = mdd_trans_start(env, mdd, handle);
3369 GOTO(stop_trans, rc);
3371 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3373 GOTO(stop_trans, rc = 0);
3376 GOTO(stop_trans, rc);
3378 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3385 memmove(xname, xname + xlen, rem);
3391 static int mdd_declare_migrate_create(const struct lu_env *env,
3392 struct mdd_object *mdd_pobj,
3393 struct mdd_object *mdd_sobj,
3394 struct mdd_object *mdd_tobj,
3395 struct md_op_spec *spec,
3397 union lmv_mds_md *mgr_ea,
3398 struct linkea_data *ldata,
3399 struct thandle *handle)
3401 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3402 const struct lu_buf *buf;
3406 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3407 handle, spec, NULL);
3411 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3416 if (S_ISLNK(la->la_mode)) {
3417 const char *target_name = spec->u.sp_symname;
3418 int sym_len = strlen(target_name);
3419 const struct lu_buf *buf;
3421 buf = mdd_buf_get_const(env, target_name, sym_len);
3422 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3426 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3427 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3433 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3434 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3435 spec->u.sp_ea.eadatalen);
3436 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3442 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3443 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3444 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3449 la_flag->la_valid = LA_FLAGS;
3450 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3451 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3456 static int mdd_migrate_create(const struct lu_env *env,
3457 struct mdd_object *mdd_pobj,
3458 struct mdd_object *mdd_sobj,
3459 struct mdd_object *mdd_tobj,
3460 const struct lu_name *lname,
3463 struct mdd_thread_info *info = mdd_env_info(env);
3464 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3465 struct md_op_spec *spec = &info->mti_spec;
3466 struct lu_buf lmm_buf = { NULL };
3467 struct lu_buf link_buf = { NULL };
3468 const struct lu_buf *buf;
3469 struct thandle *handle;
3470 struct lmv_mds_md_v1 *mgr_ea;
3471 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3472 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3474 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3478 /* prepare spec for create */
3479 memset(spec, 0, sizeof(*spec));
3480 spec->sp_cr_lookup = 0;
3481 spec->sp_feat = &dt_directory_features;
3482 if (S_ISLNK(la->la_mode)) {
3483 buf = lu_buf_check_and_alloc(
3484 &mdd_env_info(env)->mti_big_buf,
3487 link_buf.lb_len = la->la_size + 1;
3488 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3489 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3491 rc = rc != 0 ? rc : -EFAULT;
3492 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3493 mdd2obd_dev(mdd)->obd_name,
3494 PFID(mdd_object_fid(mdd_sobj)), rc);
3497 spec->u.sp_symname = link_buf.lb_buf;
3498 } else if (S_ISREG(la->la_mode)) {
3499 /* retrieve lov of the old object */
3500 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3501 if (rc != 0 && rc != -ENODATA)
3503 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3504 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3505 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3506 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3508 } else if (S_ISDIR(la->la_mode)) {
3509 rc = mdd_links_read(env, mdd_sobj, ldata);
3510 if (rc == -ENODATA) {
3511 /* ignore the non-linkEA error */
3519 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3520 memset(mgr_ea, 0, sizeof(*mgr_ea));
3521 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3522 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3523 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3524 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3525 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3526 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3528 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3530 handle = mdd_trans_create(env, mdd);
3532 GOTO(out_free, rc = PTR_ERR(handle));
3534 /* Note: this transaction is part of migration, and it is not
3535 * the last step of migration, so we set th_local = 1 to avoid
3536 * update last rcvd for this transaction */
3537 handle->th_local = 1;
3538 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3540 (union lmv_mds_md *)info->mti_xattr_buf,
3543 GOTO(stop_trans, rc);
3545 rc = mdd_trans_start(env, mdd, handle);
3547 GOTO(stop_trans, rc);
3549 /* don't set nlink from the original object */
3550 la->la_valid &= ~LA_NLINK;
3552 /* create the target object */
3553 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3556 GOTO(stop_trans, rc);
3558 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3559 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3561 GOTO(stop_trans, rc);
3564 /* Set MIGRATE EA on the source inode, so once the migration needs
3565 * to be re-done during failover, the re-do process can locate the
3566 * target object which is already being created. */
3567 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3568 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3569 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
3571 GOTO(stop_trans, rc);
3573 /* Set immutable flag, so any modification is disabled until
3574 * the migration is done. Once the migration is interrupted,
3575 * if the resume process find the migrating object has both
3576 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3577 * flag and approve the migration */
3578 la_flag->la_valid = LA_FLAGS;
3579 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3580 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3582 if (handle != NULL) {
3585 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3590 if (lmm_buf.lb_buf != NULL)
3591 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3595 static int mdd_migrate_entries(const struct lu_env *env,
3596 struct mdd_object *mdd_sobj,
3597 struct mdd_object *mdd_tobj)
3599 struct dt_object *next = mdd_object_child(mdd_sobj);
3600 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3601 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3602 struct thandle *handle;
3604 const struct dt_it_ops *iops;
3607 struct lu_dirent *ent;
3610 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3614 if (!dt_try_as_dir(env, next))
3615 GOTO(out_ent, rc = -ENOTDIR);
3617 * iterate directories
3619 iops = &next->do_index_ops->dio_it;
3620 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3622 GOTO(out_ent, rc = PTR_ERR(it));
3624 rc = iops->load(env, it, 0);
3626 rc = iops->next(env, it);
3630 * At this point and across for-loop:
3632 * rc == 0 -> ok, proceed.
3633 * rc > 0 -> end of directory.
3637 struct mdd_object *child;
3638 char *name = mdd_env_info(env)->mti_key;
3642 bool target_exist = false;
3645 len = iops->key_size(env, it);
3649 result = iops->rec(env, it, (struct dt_rec *)ent,
3650 LUDA_FID | LUDA_TYPE);
3651 if (result == -ESTALE)
3658 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3659 recsize = le16_to_cpu(ent->lde_reclen);
3661 /* Insert new fid with target name into target dir */
3662 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3663 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3664 ent->lde_name[1] == '.'))
3667 child = mdd_object_find(env, mdd, &ent->lde_fid);
3669 GOTO(out, rc = PTR_ERR(child));
3671 mdd_write_lock(env, child, MOR_SRC_CHILD);
3672 is_dir = S_ISDIR(mdd_object_type(child));
3674 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3676 /* Check whether the name has been inserted to the target */
3677 if (dt_try_as_dir(env, dt_tobj)) {
3678 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3680 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3681 (struct dt_key *)name);
3682 if (unlikely(rc == 0))
3683 target_exist = true;
3686 handle = mdd_trans_create(env, mdd);
3688 GOTO(out_put, rc = PTR_ERR(handle));
3690 /* Note: this transaction is part of migration, and it is not
3691 * the last step of migration, so we set th_local = 1 to avoid
3692 * updating last rcvd for this transaction */
3693 handle->th_local = 1;
3694 if (likely(!target_exist)) {
3695 rc = mdo_declare_index_insert(env, mdd_tobj,
3697 mdd_object_type(child),
3703 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3709 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3714 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3718 /* Update .. for child */
3719 rc = mdo_declare_index_delete(env, child, dotdot,
3724 rc = mdo_declare_index_insert(env, child,
3725 mdd_object_fid(mdd_tobj),
3726 S_IFDIR, dotdot, handle);
3731 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3738 rc = mdd_trans_start(env, mdd, handle);
3740 CERROR("%s: transaction start failed: rc = %d\n",
3741 mdd2obd_dev(mdd)->obd_name, rc);
3745 if (likely(!target_exist)) {
3746 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3747 mdd_object_type(child),
3753 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3758 rc = __mdd_index_delete_only(env, child, dotdot,
3763 rc = __mdd_index_insert_only(env, child,
3764 mdd_object_fid(mdd_tobj), S_IFDIR,
3770 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3772 strlen(name), handle);
3775 mdd_write_unlock(env, child);
3776 mdd_object_put(env, child);
3777 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3784 result = iops->next(env, it);
3785 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3786 GOTO(out, rc = -EINTR);
3788 if (result == -ESTALE)
3790 } while (result == 0);
3793 iops->fini(env, it);
3795 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3799 static int mdd_declare_update_linkea(const struct lu_env *env,
3800 struct mdd_object *mdd_pobj,
3801 struct mdd_object *mdd_sobj,
3802 struct mdd_object *mdd_tobj,
3803 const struct lu_name *child_name,
3804 struct linkea_data *ldata,
3805 struct thandle *handle)
3807 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3808 child_name, ldata, handle, 1);
3811 static int mdd_update_linkea(const struct lu_env *env,
3812 struct mdd_object *mdd_pobj,
3813 struct mdd_object *mdd_sobj,
3814 struct mdd_object *mdd_tobj,
3815 const struct lu_name *child_name,
3816 struct linkea_data *ldata,
3817 struct thandle *handle)
3819 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3820 child_name, ldata, handle, 0);
3823 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3824 struct mdd_object *mdd_pobj,
3825 struct mdd_object *mdd_sobj,
3826 struct mdd_object *mdd_tobj,
3827 const struct lu_name *lname,
3829 struct lu_attr *parent_la,
3830 struct linkea_data *ldata,
3831 struct thandle *handle)
3833 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3834 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3837 /* Revert IMMUTABLE flag */
3838 la_flag->la_valid = LA_FLAGS;
3839 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3840 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3844 /* delete entry from source dir */
3845 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3849 if (ldata->ld_buf != NULL) {
3850 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3851 mdd_tobj, lname, ldata, handle);
3856 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3857 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3862 handle->th_complex = 1;
3863 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3865 LU_XATTR_REPLACE, handle);
3870 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3871 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3877 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3878 mdd_object_type(mdd_tobj),
3879 lname->ln_name, handle);
3883 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE);
3887 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3888 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3893 /* delete old object */
3894 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3898 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3899 /* delete old object */
3900 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3903 /* set nlink to 0 */
3904 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3909 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3913 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3917 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3922 static int mdd_migrate_update_name(const struct lu_env *env,
3923 struct mdd_object *mdd_pobj,
3924 struct mdd_object *mdd_sobj,
3925 struct mdd_object *mdd_tobj,
3926 const struct lu_name *lname,
3929 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3930 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3931 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3932 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3933 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3934 struct thandle *handle;
3935 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3936 const char *name = lname->ln_name;
3940 /* update time for parent */
3941 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3942 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3943 p_la->la_valid = LA_CTIME;
3945 rc = mdd_la_get(env, mdd_sobj, so_attr);
3949 ldata->ld_buf = NULL;
3950 rc = mdd_links_read(env, mdd_sobj, ldata);
3951 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
3954 handle = mdd_trans_create(env, mdd);
3956 RETURN(PTR_ERR(handle));
3958 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3959 lname, so_attr, p_la, ldata,
3962 /* If the migration can not be fit in one transaction, just
3963 * leave it in the original MDT */
3965 GOTO(stop_trans, rc = 0);
3967 GOTO(stop_trans, rc);
3970 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3971 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3972 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3973 PFID(mdd_object_fid(mdd_tobj)));
3975 rc = mdd_trans_start(env, mdd, handle);
3977 GOTO(stop_trans, rc);
3979 /* Revert IMMUTABLE flag */
3980 la_flag->la_valid = LA_FLAGS;
3981 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3982 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3984 GOTO(stop_trans, rc);
3986 /* Remove source name from source directory */
3987 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
3989 GOTO(stop_trans, rc);
3991 if (ldata->ld_buf != NULL) {
3992 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3993 lname, ldata, handle);
3995 GOTO(stop_trans, rc);
3997 /* linkea update might decrease the source object
3998 * nlink, let's get the attr again after ref_del */
3999 rc = mdd_la_get(env, mdd_sobj, so_attr);
4001 GOTO(stop_trans, rc);
4004 if (S_ISREG(so_attr->la_mode)) {
4005 if (so_attr->la_nlink == 1) {
4006 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4008 if (rc != 0 && rc != -ENODATA)
4009 GOTO(stop_trans, rc);
4011 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4013 LU_XATTR_REPLACE, handle);
4015 GOTO(stop_trans, rc);
4019 /* Insert new fid with target name into target dir */
4020 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4021 mdd_object_type(mdd_tobj), name, handle);
4023 GOTO(stop_trans, rc);
4025 linkea_add_buf(ldata, lname, mdd_object_fid(mdd_pobj));
4026 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
4029 GOTO(stop_trans, rc);
4031 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4033 mdd_sobj->mod_flags |= DEAD_OBJ;
4034 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4036 GOTO(out_unlock, rc);
4038 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4040 GOTO(out_unlock, rc);
4042 mdo_ref_del(env, mdd_sobj, handle);
4044 mdo_ref_del(env, mdd_sobj, handle);
4046 /* Get the attr again after ref_del */
4047 rc = mdd_la_get(env, mdd_sobj, so_attr);
4049 GOTO(out_unlock, rc);
4051 ma->ma_attr = *so_attr;
4052 ma->ma_valid |= MA_INODE;
4054 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4056 GOTO(out_unlock, rc);
4058 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4059 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4060 mdo2fid(mdd_pobj), lname, lname, handle);
4062 CWARN("%s: changelog for migrate %s "DFID
4063 "under "DFID" failed: rc = %d\n",
4064 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4065 PFID(mdd_object_fid(mdd_sobj)),
4066 PFID(mdd_object_fid(mdd_pobj)), rc);
4067 /* Sigh, there are no easy way to migrate back the object, so
4068 * let's reset the result to 0 for now XXX */
4072 mdd_write_unlock(env, mdd_sobj);
4075 mdd_trans_stop(env, mdd, rc, handle);
4080 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4081 const struct lu_fid *fid, __u32 *mdt_index)
4083 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4084 struct seq_server_site *ss;
4087 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4089 range->lsr_flags = LU_SEQ_RANGE_MDT;
4090 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4094 *mdt_index = range->lsr_index;
4099 * Check whether we should migrate the file/dir
4101 * < 0 permission check failed or other error.
4102 * = 0 the file can be migrated.
4103 * > 0 the file does not need to be migrated, mostly
4104 * for multiple link file
4106 static int mdd_migrate_sanity_check(const struct lu_env *env,
4107 struct mdd_object *pobj,
4108 const struct lu_attr *pattr,
4109 struct mdd_object *sobj,
4110 struct lu_attr *sattr)
4112 struct mdd_thread_info *info = mdd_env_info(env);
4113 struct linkea_data *ldata = &info->mti_link_data;
4114 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4116 struct lu_buf *mgr_buf;
4122 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4123 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4124 if (mgr_buf->lb_buf == NULL)
4127 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4129 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4131 /* If the object has migrateEA, it means IMMUTE flag
4132 * is being set by previous migration process, so it
4133 * needs to override the IMMUTE flag, otherwise the
4134 * following sanity check will fail */
4135 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4136 LMV_HASH_FLAG_MIGRATION) {
4137 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4139 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4140 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4141 mdd2obd_dev(mdd)->obd_name,
4142 PFID(mdd_object_fid(sobj)));
4146 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4147 sobj, sattr, NULL, NULL);
4151 /* Then it will check if the file should be migrated. If the file
4152 * has mulitple links, we only need migrate the file if all of its
4153 * entries has been migrated to the remote MDT */
4154 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4157 rc = mdd_links_read(env, sobj, ldata);
4159 /* For multiple links files, if there are no linkEA data at all,
4160 * means the file might be created before linkEA is enabled, and
4161 * all of its links should not be migrated yet, otherwise it
4162 * should have some linkEA there */
4163 if (rc == -ENOENT || rc == -ENODATA)
4168 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4169 /* If there are still links locally, then the file will not be
4171 LASSERT(ldata->ld_leh != NULL);
4172 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4173 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4174 struct lu_name lname;
4176 __u32 parent_mdt_index;
4178 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4180 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4183 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4187 /* Migrate the object only if none of its parents are on the
4189 if (parent_mdt_index != mdt_index)
4192 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4193 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4194 lname.ln_name, PFID(&fid));
4202 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4203 struct md_object *sobj, const struct lu_name *lname,
4204 struct md_object *tobj, struct md_attr *ma)
4206 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4207 struct mdd_device *mdd = mdo2mdd(pobj);
4208 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4209 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4210 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4211 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4212 bool created = false;
4216 /* If the file will being migrated, it will check whether
4217 * the file is being opened by someone else right now */
4218 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4219 if (mdd_sobj->mod_count > 0) {
4221 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4222 mdd2obd_dev(mdd)->obd_name,
4223 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4224 mdd_sobj->mod_count, -EBUSY);
4225 mdd_read_unlock(env, mdd_sobj);
4226 GOTO(put, rc = -EBUSY);
4228 mdd_read_unlock(env, mdd_sobj);
4230 rc = mdd_la_get(env, mdd_sobj, so_attr);
4234 rc = mdd_la_get(env, mdd_pobj, pattr);
4238 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4245 /* Sigh, it is impossible to finish all of migration in a single
4246 * transaction, for example migrating big directory entries to the
4247 * new MDT, it needs insert all of name entries of children in the
4250 * So migration will be done in multiple steps and transactions.
4252 * 1. create an orphan object on the remote MDT in one transaction.
4253 * 2. migrate extend attributes to the new target file/directory.
4254 * 3. For directory, migrate the entries to the new MDT and update
4255 * linkEA of each children. Because we can not migrate all entries
4256 * in a single transaction, so the migrating directory will become
4257 * a striped directory during migration, so once the process is
4258 * interrupted, the directory is still accessible. (During lookup,
4259 * client will locate the name by searching both original and target
4261 * 4. Finally, update the name/FID to point to the new file/directory
4262 * in a separate transaction.
4265 /* step 1: Check whether the orphan object has been created, and create
4266 * orphan object on the remote MDT if needed */
4267 if (!mdd_object_exists(mdd_tobj)) {
4268 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4275 LASSERT(mdd_object_exists(mdd_tobj));
4276 /* step 2: migrate xattr */
4277 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4281 /* step 3: migrate name entries to the orphan object */
4282 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4283 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4286 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4287 OBD_FAIL_MDS_REINT_NET_REP)))
4290 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4293 LASSERT(mdd_object_exists(mdd_tobj));
4294 /* step 4: update name entry to the new object */
4295 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4300 /* newly created target was not locked, don't cache its attributes */
4302 mdd_invalidate(env, tobj);
4307 const struct md_dir_operations mdd_dir_ops = {
4308 .mdo_is_subdir = mdd_is_subdir,
4309 .mdo_lookup = mdd_lookup,
4310 .mdo_create = mdd_create,
4311 .mdo_rename = mdd_rename,
4312 .mdo_link = mdd_link,
4313 .mdo_unlink = mdd_unlink,
4314 .mdo_create_data = mdd_create_data,
4315 .mdo_migrate = mdd_migrate,