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)
2019 const struct lu_buf *buf;
2022 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
2027 #ifdef CONFIG_FS_POSIX_ACL
2028 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2029 /* if dir, then can inherit default ACl */
2030 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2031 XATTR_NAME_ACL_DEFAULT,
2037 if (acl_buf->lb_len > 0) {
2038 rc = mdo_declare_attr_set(env, c, attr, handle);
2042 rc = mdo_declare_xattr_set(env, c, acl_buf,
2043 XATTR_NAME_ACL_ACCESS, 0, handle);
2048 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2052 /* replay case, create LOV EA from client data */
2053 if (spec->no_create ||
2054 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2055 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2056 spec->u.sp_ea.eadatalen);
2057 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2063 if (S_ISLNK(attr->la_mode)) {
2064 const char *target_name = spec->u.sp_symname;
2065 int sym_len = strlen(target_name);
2066 const struct lu_buf *buf;
2068 buf = mdd_buf_get_const(env, target_name, sym_len);
2069 rc = dt_declare_record_write(env, mdd_object_child(c),
2075 if (spec->sp_cr_file_secctx_name != NULL) {
2076 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2077 spec->sp_cr_file_secctx_size);
2078 rc = mdo_declare_xattr_set(env, c, buf,
2079 spec->sp_cr_file_secctx_name, 0,
2088 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2089 struct mdd_object *p, struct mdd_object *c,
2090 const struct lu_name *name,
2091 struct lu_attr *attr,
2092 struct thandle *handle,
2093 const struct md_op_spec *spec,
2094 struct linkea_data *ldata,
2095 struct lu_buf *def_acl_buf,
2096 struct lu_buf *acl_buf,
2097 struct dt_allocation_hint *hint)
2101 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2102 def_acl_buf, acl_buf, hint);
2106 if (S_ISDIR(attr->la_mode)) {
2107 rc = mdo_declare_ref_add(env, p, handle);
2112 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2113 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2117 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2119 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2120 name->ln_name, handle);
2124 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2129 la->la_valid = LA_CTIME | LA_MTIME;
2130 rc = mdo_declare_attr_set(env, p, la, handle);
2134 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2142 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2143 struct lu_attr *la, struct lu_buf *def_acl_buf,
2144 struct lu_buf *acl_buf)
2149 if (S_ISLNK(la->la_mode)) {
2150 acl_buf->lb_len = 0;
2151 def_acl_buf->lb_len = 0;
2155 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2156 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2157 XATTR_NAME_ACL_DEFAULT);
2158 mdd_read_unlock(env, pobj);
2160 /* If there are default ACL, fix mode/ACL by default ACL */
2161 def_acl_buf->lb_len = rc;
2162 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2163 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2164 acl_buf->lb_len = rc;
2165 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2168 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2169 /* If there are no default ACL, fix mode by mask */
2170 struct lu_ucred *uc = lu_ucred(env);
2172 /* The create triggered by MDT internal events, such as
2173 * LFSCK reset, will not contain valid "uc". */
2174 if (unlikely(uc != NULL))
2175 la->la_mode &= ~uc->uc_umask;
2177 acl_buf->lb_len = 0;
2178 def_acl_buf->lb_len = 0;
2185 * Create a metadata object and initialize it, set acl, xattr.
2187 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2188 struct mdd_object *son, struct lu_attr *attr,
2189 struct md_op_spec *spec, struct lu_buf *acl_buf,
2190 struct lu_buf *def_acl_buf,
2191 struct dt_allocation_hint *hint,
2192 struct thandle *handle)
2194 const struct lu_buf *buf;
2197 mdd_write_lock(env, son, MOR_TGT_CHILD);
2198 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2203 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2204 * created in declare phase, they also needs to be added to master
2205 * object as sub-directory entry. So it has to initialize the master
2206 * object, then set dir striped EA.(in mdo_xattr_set) */
2207 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2210 GOTO(err_destroy, rc);
2213 * in case of replay we just set LOVEA provided by the client
2214 * XXX: I think it would be interesting to try "old" way where
2215 * MDT calls this xattr_set(LOV) in a different transaction.
2216 * probably this way we code can be made better.
2219 /* During creation, there are only a few cases we need do xattr_set to
2221 * 1. regular file: see comments above.
2222 * 2. dir: inherit default striping or pool settings from parent.
2223 * 3. create striped directory with provided stripeEA.
2224 * 4. create striped directory because inherit default layout from the
2227 if (spec->no_create ||
2228 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2229 S_ISDIR(attr->la_mode)) {
2230 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2231 spec->u.sp_ea.eadatalen);
2232 rc = mdo_xattr_set(env, son, buf,
2233 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2237 GOTO(err_destroy, rc);
2240 #ifdef CONFIG_FS_POSIX_ACL
2241 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2242 S_ISDIR(attr->la_mode)) {
2243 /* set default acl */
2244 rc = mdo_xattr_set(env, son, def_acl_buf,
2245 XATTR_NAME_ACL_DEFAULT, 0,
2248 GOTO(err_destroy, rc);
2250 /* set its own acl */
2251 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2252 rc = mdo_xattr_set(env, son, acl_buf,
2253 XATTR_NAME_ACL_ACCESS,
2256 GOTO(err_destroy, rc);
2260 if (S_ISLNK(attr->la_mode)) {
2261 struct lu_ucred *uc = lu_ucred_assert(env);
2262 struct dt_object *dt = mdd_object_child(son);
2263 const char *target_name = spec->u.sp_symname;
2264 int sym_len = strlen(target_name);
2265 const struct lu_buf *buf;
2268 buf = mdd_buf_get_const(env, target_name, sym_len);
2269 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2271 CFS_CAP_SYS_RESOURCE_MASK);
2276 GOTO(err_initlized, rc = -EFAULT);
2279 if (spec->sp_cr_file_secctx_name != NULL) {
2280 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2281 spec->sp_cr_file_secctx_size);
2282 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2285 GOTO(err_initlized, rc);
2289 if (unlikely(rc != 0)) {
2291 if (S_ISDIR(attr->la_mode)) {
2292 /* Drop the reference, no need to delete "."/"..",
2293 * because the object to be destroied directly. */
2294 rc2 = mdo_ref_del(env, son, handle);
2298 rc2 = mdo_ref_del(env, son, handle);
2302 mdo_destroy(env, son, handle);
2305 mdd_write_unlock(env, son);
2309 static int mdd_index_delete(const struct lu_env *env,
2310 struct mdd_object *mdd_pobj,
2311 struct lu_attr *cattr,
2312 const struct lu_name *lname)
2314 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2315 struct thandle *handle;
2319 handle = mdd_trans_create(env, mdd);
2321 RETURN(PTR_ERR(handle));
2323 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2328 if (S_ISDIR(cattr->la_mode)) {
2329 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2334 /* Since this will only be used in the error handler path,
2335 * Let's set the thandle to be local and not mess the transno */
2336 handle->th_local = 1;
2337 rc = mdd_trans_start(env, mdd, handle);
2341 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2342 S_ISDIR(cattr->la_mode), handle);
2346 mdd_trans_stop(env, mdd, rc, handle);
2351 * Create object and insert it into namespace.
2353 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2354 const struct lu_name *lname, struct md_object *child,
2355 struct md_op_spec *spec, struct md_attr* ma)
2357 struct mdd_thread_info *info = mdd_env_info(env);
2358 struct lu_attr *la = &info->mti_la_for_fix;
2359 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2360 struct mdd_object *son = md2mdd_obj(child);
2361 struct mdd_device *mdd = mdo2mdd(pobj);
2362 struct lu_attr *attr = &ma->ma_attr;
2363 struct thandle *handle;
2364 struct lu_attr *pattr = &info->mti_pattr;
2365 struct lu_buf acl_buf;
2366 struct lu_buf def_acl_buf;
2367 struct linkea_data *ldata = &info->mti_link_data;
2368 const char *name = lname->ln_name;
2369 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2375 * Two operations have to be performed:
2377 * - an allocation of a new object (->do_create()), and
2379 * - an insertion into a parent index (->dio_insert()).
2381 * Due to locking, operation order is not important, when both are
2382 * successful, *but* error handling cases are quite different:
2384 * - if insertion is done first, and following object creation fails,
2385 * insertion has to be rolled back, but this operation might fail
2386 * also leaving us with dangling index entry.
2388 * - if creation is done first, is has to be undone if insertion
2389 * fails, leaving us with leaked space, which is neither good, nor
2392 * It seems that creation-first is simplest solution, but it is
2393 * sub-optimal in the frequent
2398 * case, because second mkdir is bound to create object, only to
2399 * destroy it immediately.
2401 * To avoid this follow local file systems that do double lookup:
2403 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2405 * 1. create (mdd_object_create_internal())
2407 * 2. insert (__mdd_index_insert(), lookup again)
2410 rc = mdd_la_get(env, mdd_pobj, pattr);
2414 /* Sanity checks before big job. */
2415 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2419 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2420 GOTO(out_free, rc = -EINPROGRESS);
2422 handle = mdd_trans_create(env, mdd);
2424 GOTO(out_free, rc = PTR_ERR(handle));
2426 acl_buf.lb_buf = info->mti_xattr_buf;
2427 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2428 def_acl_buf.lb_buf = info->mti_key;
2429 def_acl_buf.lb_len = sizeof(info->mti_key);
2430 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2434 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2436 memset(ldata, 0, sizeof(*ldata));
2437 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2438 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2441 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2442 &tfid, lname, 1, 0, ldata);
2444 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2445 mdd_object_fid(mdd_pobj),
2446 lname, 1, 0, ldata);
2449 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2450 handle, spec, ldata, &def_acl_buf, &acl_buf,
2455 rc = mdd_trans_start(env, mdd, handle);
2459 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2460 &def_acl_buf, hint, handle);
2464 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2465 mdd_write_lock(env, son, MOR_TGT_CHILD);
2466 rc = __mdd_orphan_add(env, son, handle);
2467 GOTO(out_volatile, rc);
2469 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2470 attr->la_mode, name, handle);
2472 GOTO(err_created, rc);
2474 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2477 /* update parent directory mtime/ctime */
2479 la->la_valid = LA_CTIME | LA_MTIME;
2480 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2482 GOTO(err_insert, rc);
2490 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2491 rc2 = __mdd_orphan_del(env, son, handle);
2493 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2494 S_ISDIR(attr->la_mode),
2500 mdd_write_lock(env, son, MOR_TGT_CHILD);
2501 if (S_ISDIR(attr->la_mode)) {
2502 /* Drop the reference, no need to delete "."/"..",
2503 * because the object is to be destroyed directly. */
2504 rc2 = mdo_ref_del(env, son, handle);
2506 mdd_write_unlock(env, son);
2511 /* For volatile files drop one link immediately, since there is
2512 * no filename in the namespace, and save any error returned. */
2513 rc2 = mdo_ref_del(env, son, handle);
2515 mdd_write_unlock(env, son);
2516 if (unlikely(rc == 0))
2521 /* Don't destroy the volatile object on success */
2522 if (likely(rc != 0))
2523 mdo_destroy(env, son, handle);
2524 mdd_write_unlock(env, son);
2527 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2528 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2529 rc = mdd_changelog_ns_store(env, mdd,
2530 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2531 S_ISREG(attr->la_mode) ? CL_CREATE :
2532 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2533 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2536 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2538 /* If creation fails, it is most likely due to the remote update
2539 * failure, because local transaction will mostly succeed at
2540 * this stage. There is no easy way to rollback all of previous
2541 * updates, so let's remove the object from namespace, and
2542 * LFSCK should handle the orphan object. */
2543 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2544 mdd_index_delete(env, mdd_pobj, attr, lname);
2548 if (is_vmalloc_addr(ldata->ld_buf))
2549 /* if we vmalloced a large buffer drop it */
2550 lu_buf_free(ldata->ld_buf);
2552 /* The child object shouldn't be cached anymore */
2554 set_bit(LU_OBJECT_HEARD_BANSHEE,
2555 &child->mo_lu.lo_header->loh_flags);
2560 * Get locks on parents in proper order
2561 * RETURN: < 0 - error, rename_order if successful
2569 static int mdd_rename_order(const struct lu_env *env,
2570 struct mdd_device *mdd,
2571 struct mdd_object *src_pobj,
2572 const struct lu_attr *pattr,
2573 struct mdd_object *tgt_pobj)
2575 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2579 if (src_pobj == tgt_pobj)
2580 RETURN(MDD_RN_SAME);
2582 /* compared the parent child relationship of src_p&tgt_p */
2583 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2585 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2588 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2602 /* has not mdd_write{read}_lock on any obj yet. */
2603 static int mdd_rename_sanity_check(const struct lu_env *env,
2604 struct mdd_object *src_pobj,
2605 const struct lu_attr *pattr,
2606 struct mdd_object *tgt_pobj,
2607 const struct lu_attr *tpattr,
2608 struct mdd_object *sobj,
2609 const struct lu_attr *cattr,
2610 struct mdd_object *tobj,
2611 const struct lu_attr *tattr)
2616 /* XXX: when get here, sobj must NOT be NULL,
2617 * the other case has been processed in cld_rename
2618 * before mdd_rename and enable MDS_PERM_BYPASS. */
2621 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2625 /* XXX: when get here, "tobj == NULL" means tobj must
2626 * NOT exist (neither on remote MDS, such case has been
2627 * processed in cld_rename before mdd_rename and enable
2629 * So check may_create, but not check may_unlink. */
2631 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2632 (src_pobj != tgt_pobj));
2634 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2635 (src_pobj != tgt_pobj), 1);
2637 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2638 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2643 static int mdd_declare_rename(const struct lu_env *env,
2644 struct mdd_device *mdd,
2645 struct mdd_object *mdd_spobj,
2646 struct mdd_object *mdd_tpobj,
2647 struct mdd_object *mdd_sobj,
2648 struct mdd_object *mdd_tobj,
2649 const struct lu_name *sname,
2650 const struct lu_name *tname,
2652 struct linkea_data *ldata,
2653 struct thandle *handle)
2655 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2658 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2659 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2665 /* name from source dir */
2666 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2670 /* .. from source child */
2671 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2672 /* source child can be directory,
2673 * counted by source dir's nlink */
2674 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2677 if (mdd_spobj != mdd_tpobj) {
2678 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2683 rc = mdo_declare_index_insert(env, mdd_sobj,
2685 S_IFDIR, dotdot, handle);
2690 /* new target child can be directory,
2691 * counted by target dir's nlink */
2692 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2697 la->la_valid = LA_CTIME | LA_MTIME;
2698 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2702 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2706 la->la_valid = LA_CTIME;
2707 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2711 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
2712 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2717 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2718 mdd_object_type(mdd_sobj),
2719 tname->ln_name, handle);
2723 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2724 /* delete target child in target parent directory */
2725 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2730 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2734 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2735 /* target child can be directory,
2736 * delete "." reference in target child directory */
2737 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2741 /* delete ".." reference in target parent directory */
2742 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2747 la->la_valid = LA_CTIME;
2748 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2752 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2757 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2764 /* src object can be remote that is why we use only fid and type of object */
2765 static int mdd_rename(const struct lu_env *env,
2766 struct md_object *src_pobj, struct md_object *tgt_pobj,
2767 const struct lu_fid *lf, const struct lu_name *lsname,
2768 struct md_object *tobj, const struct lu_name *ltname,
2771 const char *sname = lsname->ln_name;
2772 const char *tname = ltname->ln_name;
2773 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2774 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2775 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2776 struct mdd_device *mdd = mdo2mdd(src_pobj);
2777 struct mdd_object *mdd_sobj = NULL; /* source object */
2778 struct mdd_object *mdd_tobj = NULL;
2779 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2780 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2781 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2782 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2783 struct thandle *handle;
2784 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2785 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2786 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2789 bool tobj_locked = 0;
2790 unsigned cl_flags = 0;
2795 mdd_tobj = md2mdd_obj(tobj);
2797 mdd_sobj = mdd_object_find(env, mdd, lf);
2798 if (IS_ERR(mdd_sobj))
2799 RETURN(PTR_ERR(mdd_sobj));
2801 rc = mdd_la_get(env, mdd_sobj, cattr);
2803 GOTO(out_pending, rc);
2805 rc = mdd_la_get(env, mdd_spobj, pattr);
2807 GOTO(out_pending, rc);
2810 rc = mdd_la_get(env, mdd_tobj, tattr);
2812 GOTO(out_pending, rc);
2813 /* search for an existing archive.
2814 * we should check ahead as the object
2815 * can be destroyed in this transaction */
2816 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2817 cl_flags |= CLF_RENAME_LAST_EXISTS;
2820 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2822 GOTO(out_pending, rc);
2824 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2825 mdd_sobj, cattr, mdd_tobj, tattr);
2827 GOTO(out_pending, rc);
2829 rc = mdd_name_check(mdd, ltname);
2831 GOTO(out_pending, rc);
2833 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2834 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2836 GOTO(out_pending, rc);
2838 handle = mdd_trans_create(env, mdd);
2840 GOTO(out_pending, rc = PTR_ERR(handle));
2842 memset(ldata, 0, sizeof(*ldata));
2843 mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname,
2844 mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata);
2845 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2846 mdd_tobj, lsname, ltname, ma, ldata, handle);
2850 rc = mdd_trans_start(env, mdd, handle);
2854 is_dir = S_ISDIR(cattr->la_mode);
2856 /* Remove source name from source directory */
2857 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2861 /* "mv dir1 dir2" needs "dir1/.." link update */
2862 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2863 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2865 GOTO(fixup_spobj2, rc);
2867 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2870 GOTO(fixup_spobj, rc);
2873 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2874 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2876 /* tname might been renamed to something else */
2877 GOTO(fixup_spobj, rc);
2880 /* Insert new fid with target name into target dir */
2881 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2884 GOTO(fixup_tpobj, rc);
2886 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2887 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2889 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2890 la->la_valid = LA_CTIME;
2891 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2893 GOTO(fixup_tpobj, rc);
2895 /* Update the linkEA for the source object */
2896 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2897 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2898 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2901 /* Old files might not have EA entry */
2902 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2903 lsname, handle, NULL, 0);
2904 mdd_write_unlock(env, mdd_sobj);
2905 /* We don't fail the transaction if the link ea can't be
2906 updated -- fid2path will use alternate lookup method. */
2909 /* Remove old target object
2910 * For tobj is remote case cmm layer has processed
2911 * and set tobj to NULL then. So when tobj is NOT NULL,
2912 * it must be local one.
2914 if (tobj && mdd_object_exists(mdd_tobj)) {
2915 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2917 if (mdd_is_dead_obj(mdd_tobj)) {
2918 /* shld not be dead, something is wrong */
2919 CERROR("tobj is dead, something is wrong\n");
2923 mdo_ref_del(env, mdd_tobj, handle);
2925 /* Remove dot reference. */
2926 if (S_ISDIR(tattr->la_mode))
2927 mdo_ref_del(env, mdd_tobj, handle);
2930 /* fetch updated nlink */
2931 rc = mdd_la_get(env, mdd_tobj, tattr);
2933 CERROR("%s: Failed to get nlink for tobj "
2935 mdd2obd_dev(mdd)->obd_name,
2936 PFID(tpobj_fid), rc);
2937 GOTO(fixup_tpobj, rc);
2940 la->la_valid = LA_CTIME;
2941 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2943 CERROR("%s: Failed to set ctime for tobj "
2945 mdd2obd_dev(mdd)->obd_name,
2946 PFID(tpobj_fid), rc);
2947 GOTO(fixup_tpobj, rc);
2950 /* XXX: this transfer to ma will be removed with LOD/OSP */
2951 ma->ma_attr = *tattr;
2952 ma->ma_valid |= MA_INODE;
2953 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2956 CERROR("%s: Failed to unlink tobj "
2958 mdd2obd_dev(mdd)->obd_name,
2959 PFID(tpobj_fid), rc);
2960 GOTO(fixup_tpobj, rc);
2963 /* fetch updated nlink */
2964 rc = mdd_la_get(env, mdd_tobj, tattr);
2965 if (rc == -ENOENT) {
2966 /* the object got removed, let's
2967 * return the latest known attributes */
2968 tattr->la_nlink = 0;
2970 } else if (rc != 0) {
2971 CERROR("%s: Failed to get nlink for tobj "
2973 mdd2obd_dev(mdd)->obd_name,
2974 PFID(tpobj_fid), rc);
2975 GOTO(fixup_tpobj, rc);
2977 /* XXX: this transfer to ma will be removed with LOD/OSP */
2978 ma->ma_attr = *tattr;
2979 ma->ma_valid |= MA_INODE;
2981 if (tattr->la_nlink == 0)
2982 cl_flags |= CLF_RENAME_LAST;
2984 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
2987 la->la_valid = LA_CTIME | LA_MTIME;
2988 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2990 GOTO(fixup_tpobj, rc);
2992 if (mdd_spobj != mdd_tpobj) {
2993 la->la_valid = LA_CTIME | LA_MTIME;
2994 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2996 GOTO(fixup_tpobj, rc);
3003 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3005 CWARN("tp obj fix error %d\n",rc2);
3007 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3008 !mdd_is_dead_obj(mdd_tobj)) {
3010 mdo_ref_add(env, mdd_tobj, handle);
3012 mdo_ref_add(env, mdd_tobj, handle);
3015 rc2 = __mdd_index_insert(env, mdd_tpobj,
3017 mdd_object_type(mdd_tobj),
3020 CWARN("tp obj fix error: rc = %d\n", rc2);
3025 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3026 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3028 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3029 mdd2obd_dev(mdd)->obd_name, rc2);
3032 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3035 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3036 mdd2obd_dev(mdd)->obd_name, rc2);
3041 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3042 mdd_object_type(mdd_sobj), sname,
3045 CWARN("sp obj fix error: rc = %d\n", rc2);
3050 mdd_write_unlock(env, mdd_tobj);
3053 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3054 mdd_tobj, tpobj_fid, lf, spobj_fid,
3055 ltname, lsname, handle);
3058 mdd_trans_stop(env, mdd, rc, handle);
3061 mdd_object_put(env, mdd_sobj);
3066 * During migration once the parent FID has been changed,
3067 * we need update the parent FID in linkea.
3069 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3070 struct mdd_object *parent,
3071 struct mdd_object *newparent,
3072 struct mdd_object *child,
3073 const char *name, int namelen,
3074 struct thandle *handle,
3077 struct mdd_thread_info *info = mdd_env_info(env);
3078 struct linkea_data ldata = { NULL };
3079 struct lu_buf *buf = &info->mti_link_buf;
3085 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3086 if (buf->lb_buf == NULL)
3090 rc = mdd_links_read(env, child, &ldata);
3092 if (rc == -ENOENT || rc == -ENODATA)
3097 LASSERT(ldata.ld_leh != NULL);
3098 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3099 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3100 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3101 struct lu_name lname;
3104 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3107 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3108 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3109 ldata.ld_lee = (struct link_ea_entry *)
3110 ((char *)ldata.ld_lee +
3115 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3116 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3117 lname.ln_namelen, lname.ln_name,
3118 PFID(mdd_object_fid(newparent)));
3119 /* update to the new parent fid */
3120 linkea_entry_pack(ldata.ld_lee, &lname,
3121 mdd_object_fid(newparent));
3123 rc = mdd_declare_links_add(env, child, handle, &ldata,
3126 rc = mdd_links_write(env, child, &ldata, handle);
3132 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3133 struct mdd_object *parent,
3134 struct mdd_object *newparent,
3135 struct mdd_object *child,
3136 const char *name, int namelen,
3137 struct thandle *handle)
3139 return mdd_linkea_update_child_internal(env, parent, newparent,
3141 namelen, handle, true);
3144 static int mdd_linkea_update_child(const struct lu_env *env,
3145 struct mdd_object *parent,
3146 struct mdd_object *newparent,
3147 struct mdd_object *child,
3148 const char *name, int namelen,
3149 struct thandle *handle)
3151 return mdd_linkea_update_child_internal(env, parent, newparent,
3153 namelen, handle, false);
3156 static int mdd_update_linkea_internal(const struct lu_env *env,
3157 struct mdd_object *mdd_pobj,
3158 struct mdd_object *mdd_sobj,
3159 struct mdd_object *mdd_tobj,
3160 const struct lu_name *child_name,
3161 struct linkea_data *ldata,
3162 struct thandle *handle,
3165 struct mdd_thread_info *info = mdd_env_info(env);
3170 LASSERT(ldata->ld_buf != NULL);
3173 /* If it is mulitple links file, we need update the name entry for
3175 LASSERT(ldata->ld_leh != NULL);
3176 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3177 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3178 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3179 struct mdd_object *pobj;
3180 struct lu_name lname;
3183 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3185 pobj = mdd_object_find(env, mdd, &fid);
3187 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3188 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3190 linkea_del_buf(ldata, &lname);
3194 if (!mdd_object_exists(pobj)) {
3195 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3196 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3197 linkea_del_buf(ldata, &lname);
3198 mdd_object_put(env, pobj);
3202 if (pobj == mdd_pobj &&
3203 lname.ln_namelen == child_name->ln_namelen &&
3204 strncmp(lname.ln_name, child_name->ln_name,
3205 lname.ln_namelen) == 0) {
3206 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3207 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3209 linkea_del_buf(ldata, &lname);
3210 mdd_object_put(env, pobj);
3214 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3215 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3216 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3219 /* Remove source name from source directory */
3220 /* Insert new fid with target name into target dir */
3221 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3226 rc = mdo_declare_index_insert(env, pobj,
3227 mdd_object_fid(mdd_tobj),
3228 mdd_object_type(mdd_tobj),
3229 lname.ln_name, handle);
3233 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3237 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3241 char *tmp_name = info->mti_key;
3243 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3244 /* lnamelen is too big(> NAME_MAX + 16),
3245 * something wrong about this linkea, let's
3247 linkea_del_buf(ldata, &lname);
3248 mdd_object_put(env, pobj);
3252 /* Note: lname might be without \0 at the end, see
3253 * linkea_entry_unpack(), let's add extra \0 by
3255 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3256 lname.ln_namelen, lname.ln_name);
3257 lname.ln_name = tmp_name;
3259 /* Let's check if this linkEA still valid, before
3260 * it might be packed into the RPC buffer. */
3261 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3262 &info->mti_fid, NULL);
3264 !lu_fid_eq(&info->mti_fid,
3265 mdd_object_fid(mdd_sobj))) {
3266 /* skip invalid linkea entry */
3267 linkea_del_buf(ldata, &lname);
3268 mdd_object_put(env, pobj);
3272 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3276 rc = __mdd_index_insert(env, pobj,
3277 mdd_object_fid(mdd_tobj),
3278 mdd_object_type(mdd_tobj),
3283 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3284 rc = mdo_ref_add(env, mdd_tobj, handle);
3285 mdd_write_unlock(env, mdd_tobj);
3289 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3290 mdo_ref_del(env, mdd_sobj, handle);
3291 mdd_write_unlock(env, mdd_sobj);
3294 mdd_object_put(env, pobj);
3298 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3305 static int mdd_migrate_xattrs(const struct lu_env *env,
3306 struct mdd_object *mdd_sobj,
3307 struct mdd_object *mdd_tobj)
3309 struct mdd_thread_info *info = mdd_env_info(env);
3310 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3312 struct thandle *handle;
3318 struct lu_buf list_xbuf;
3322 /* retrieve xattr list from the old object */
3323 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3324 if (list_xsize == -ENODATA)
3330 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3331 if (info->mti_big_buf.lb_buf == NULL)
3334 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3335 list_xbuf.lb_len = list_xsize;
3336 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3341 xname = list_xbuf.lb_buf;
3343 xlen = strnlen(xname, rem - 1) + 1;
3344 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3345 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3346 strcmp(XATTR_NAME_LMV, xname) == 0)
3349 /* For directory, if there are default layout, migrate here */
3350 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3351 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3354 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3355 if (xsize == -ENODATA)
3360 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3361 if (info->mti_link_buf.lb_buf == NULL)
3362 GOTO(out, rc = -ENOMEM);
3364 xbuf.lb_len = xsize;
3365 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3366 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3372 handle = mdd_trans_create(env, mdd);
3374 GOTO(out, rc = PTR_ERR(handle));
3376 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3379 GOTO(stop_trans, rc);
3380 /* Note: this transaction is part of migration, and it is not
3381 * the last step of migration, so we set th_local = 1 to avoid
3382 * update last rcvd for this transaction */
3383 handle->th_local = 1;
3384 rc = mdd_trans_start(env, mdd, handle);
3386 GOTO(stop_trans, rc);
3388 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3390 GOTO(stop_trans, rc = 0);
3393 GOTO(stop_trans, rc);
3395 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3402 memmove(xname, xname + xlen, rem);
3408 static int mdd_declare_migrate_create(const struct lu_env *env,
3409 struct mdd_object *mdd_pobj,
3410 struct mdd_object *mdd_sobj,
3411 struct mdd_object *mdd_tobj,
3412 struct md_op_spec *spec,
3414 union lmv_mds_md *mgr_ea,
3415 struct linkea_data *ldata,
3416 struct thandle *handle)
3418 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3419 const struct lu_buf *buf;
3423 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3424 handle, spec, NULL);
3428 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3433 if (S_ISLNK(la->la_mode)) {
3434 const char *target_name = spec->u.sp_symname;
3435 int sym_len = strlen(target_name);
3436 const struct lu_buf *buf;
3438 buf = mdd_buf_get_const(env, target_name, sym_len);
3439 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3443 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3444 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3450 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3451 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3452 spec->u.sp_ea.eadatalen);
3453 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3459 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3460 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3461 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3466 la_flag->la_valid = LA_FLAGS;
3467 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3468 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3473 static int mdd_migrate_create(const struct lu_env *env,
3474 struct mdd_object *mdd_pobj,
3475 struct mdd_object *mdd_sobj,
3476 struct mdd_object *mdd_tobj,
3477 const struct lu_name *lname,
3480 struct mdd_thread_info *info = mdd_env_info(env);
3481 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3482 struct md_op_spec *spec = &info->mti_spec;
3483 struct lu_buf lmm_buf = { NULL };
3484 struct lu_buf link_buf = { NULL };
3485 const struct lu_buf *buf;
3486 struct thandle *handle;
3487 struct lmv_mds_md_v1 *mgr_ea;
3488 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3489 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3491 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3495 /* prepare spec for create */
3496 memset(spec, 0, sizeof(*spec));
3497 spec->sp_cr_lookup = 0;
3498 spec->sp_feat = &dt_directory_features;
3499 if (S_ISLNK(la->la_mode)) {
3500 buf = lu_buf_check_and_alloc(
3501 &mdd_env_info(env)->mti_big_buf,
3504 link_buf.lb_len = la->la_size + 1;
3505 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3506 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3508 rc = rc != 0 ? rc : -EFAULT;
3509 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3510 mdd2obd_dev(mdd)->obd_name,
3511 PFID(mdd_object_fid(mdd_sobj)), rc);
3514 spec->u.sp_symname = link_buf.lb_buf;
3515 } else if (S_ISREG(la->la_mode)) {
3516 /* retrieve lov of the old object */
3517 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3518 if (rc != 0 && rc != -ENODATA)
3520 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3521 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3522 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3523 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3525 } else if (S_ISDIR(la->la_mode)) {
3526 rc = mdd_links_read(env, mdd_sobj, ldata);
3527 if (rc == -ENODATA) {
3528 /* ignore the non-linkEA error */
3536 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3537 memset(mgr_ea, 0, sizeof(*mgr_ea));
3538 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3539 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3540 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3541 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3542 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3543 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3545 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3547 handle = mdd_trans_create(env, mdd);
3549 GOTO(out_free, rc = PTR_ERR(handle));
3551 /* Note: this transaction is part of migration, and it is not
3552 * the last step of migration, so we set th_local = 1 to avoid
3553 * update last rcvd for this transaction */
3554 handle->th_local = 1;
3555 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3557 (union lmv_mds_md *)info->mti_xattr_buf,
3560 GOTO(stop_trans, rc);
3562 rc = mdd_trans_start(env, mdd, handle);
3564 GOTO(stop_trans, rc);
3566 /* don't set nlink from the original object */
3567 la->la_valid &= ~LA_NLINK;
3569 /* create the target object */
3570 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3573 GOTO(stop_trans, rc);
3575 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3576 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3578 GOTO(stop_trans, rc);
3581 /* Set MIGRATE EA on the source inode, so once the migration needs
3582 * to be re-done during failover, the re-do process can locate the
3583 * target object which is already being created. */
3584 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3585 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3586 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
3588 GOTO(stop_trans, rc);
3590 /* Set immutable flag, so any modification is disabled until
3591 * the migration is done. Once the migration is interrupted,
3592 * if the resume process find the migrating object has both
3593 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3594 * flag and approve the migration */
3595 la_flag->la_valid = LA_FLAGS;
3596 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3597 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3599 if (handle != NULL) {
3602 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3607 if (lmm_buf.lb_buf != NULL)
3608 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3612 static int mdd_migrate_entries(const struct lu_env *env,
3613 struct mdd_object *mdd_sobj,
3614 struct mdd_object *mdd_tobj)
3616 struct dt_object *next = mdd_object_child(mdd_sobj);
3617 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3618 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3619 struct thandle *handle;
3621 const struct dt_it_ops *iops;
3624 struct lu_dirent *ent;
3627 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3631 if (!dt_try_as_dir(env, next))
3632 GOTO(out_ent, rc = -ENOTDIR);
3634 * iterate directories
3636 iops = &next->do_index_ops->dio_it;
3637 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3639 GOTO(out_ent, rc = PTR_ERR(it));
3641 rc = iops->load(env, it, 0);
3643 rc = iops->next(env, it);
3647 * At this point and across for-loop:
3649 * rc == 0 -> ok, proceed.
3650 * rc > 0 -> end of directory.
3654 struct mdd_object *child;
3655 char *name = mdd_env_info(env)->mti_key;
3659 bool target_exist = false;
3662 len = iops->key_size(env, it);
3666 result = iops->rec(env, it, (struct dt_rec *)ent,
3667 LUDA_FID | LUDA_TYPE);
3668 if (result == -ESTALE)
3675 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3676 recsize = le16_to_cpu(ent->lde_reclen);
3678 /* Insert new fid with target name into target dir */
3679 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3680 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3681 ent->lde_name[1] == '.'))
3684 child = mdd_object_find(env, mdd, &ent->lde_fid);
3686 GOTO(out, rc = PTR_ERR(child));
3688 mdd_write_lock(env, child, MOR_SRC_CHILD);
3689 is_dir = S_ISDIR(mdd_object_type(child));
3691 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3693 /* Check whether the name has been inserted to the target */
3694 if (dt_try_as_dir(env, dt_tobj)) {
3695 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3697 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3698 (struct dt_key *)name);
3699 if (unlikely(rc == 0))
3700 target_exist = true;
3703 handle = mdd_trans_create(env, mdd);
3705 GOTO(out_put, rc = PTR_ERR(handle));
3707 /* Note: this transaction is part of migration, and it is not
3708 * the last step of migration, so we set th_local = 1 to avoid
3709 * updating last rcvd for this transaction */
3710 handle->th_local = 1;
3711 if (likely(!target_exist)) {
3712 rc = mdo_declare_index_insert(env, mdd_tobj,
3714 mdd_object_type(child),
3720 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3726 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3731 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3735 /* Update .. for child */
3736 rc = mdo_declare_index_delete(env, child, dotdot,
3741 rc = mdo_declare_index_insert(env, child,
3742 mdd_object_fid(mdd_tobj),
3743 S_IFDIR, dotdot, handle);
3748 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3755 rc = mdd_trans_start(env, mdd, handle);
3757 CERROR("%s: transaction start failed: rc = %d\n",
3758 mdd2obd_dev(mdd)->obd_name, rc);
3762 if (likely(!target_exist)) {
3763 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3764 mdd_object_type(child),
3770 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3775 rc = __mdd_index_delete_only(env, child, dotdot,
3780 rc = __mdd_index_insert_only(env, child,
3781 mdd_object_fid(mdd_tobj), S_IFDIR,
3787 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3789 strlen(name), handle);
3792 mdd_write_unlock(env, child);
3793 mdd_object_put(env, child);
3794 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3801 result = iops->next(env, it);
3802 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3803 GOTO(out, rc = -EINTR);
3805 if (result == -ESTALE)
3807 } while (result == 0);
3810 iops->fini(env, it);
3812 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3816 static int mdd_declare_update_linkea(const struct lu_env *env,
3817 struct mdd_object *mdd_pobj,
3818 struct mdd_object *mdd_sobj,
3819 struct mdd_object *mdd_tobj,
3820 const struct lu_name *child_name,
3821 struct linkea_data *ldata,
3822 struct thandle *handle)
3824 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3825 child_name, ldata, handle, 1);
3828 static int mdd_update_linkea(const struct lu_env *env,
3829 struct mdd_object *mdd_pobj,
3830 struct mdd_object *mdd_sobj,
3831 struct mdd_object *mdd_tobj,
3832 const struct lu_name *child_name,
3833 struct linkea_data *ldata,
3834 struct thandle *handle)
3836 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3837 child_name, ldata, handle, 0);
3840 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3841 struct mdd_object *mdd_pobj,
3842 struct mdd_object *mdd_sobj,
3843 struct mdd_object *mdd_tobj,
3844 const struct lu_name *lname,
3846 struct lu_attr *parent_la,
3847 struct linkea_data *ldata,
3848 struct thandle *handle)
3850 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3851 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3854 /* Revert IMMUTABLE flag */
3855 la_flag->la_valid = LA_FLAGS;
3856 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3857 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3861 /* delete entry from source dir */
3862 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3866 if (ldata->ld_buf != NULL) {
3867 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3868 mdd_tobj, lname, ldata, handle);
3873 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3874 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3879 handle->th_complex = 1;
3880 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3882 LU_XATTR_REPLACE, handle);
3887 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3888 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3894 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3895 mdd_object_type(mdd_tobj),
3896 lname->ln_name, handle);
3900 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE);
3904 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3905 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3910 /* delete old object */
3911 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3915 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3916 /* delete old object */
3917 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3920 /* set nlink to 0 */
3921 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3926 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3930 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3934 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3939 static int mdd_migrate_update_name(const struct lu_env *env,
3940 struct mdd_object *mdd_pobj,
3941 struct mdd_object *mdd_sobj,
3942 struct mdd_object *mdd_tobj,
3943 const struct lu_name *lname,
3946 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3947 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3948 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3949 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3950 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3951 struct thandle *handle;
3952 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3953 const char *name = lname->ln_name;
3957 /* update time for parent */
3958 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3959 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3960 p_la->la_valid = LA_CTIME;
3962 rc = mdd_la_get(env, mdd_sobj, so_attr);
3966 ldata->ld_buf = NULL;
3967 rc = mdd_links_read(env, mdd_sobj, ldata);
3968 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
3971 handle = mdd_trans_create(env, mdd);
3973 RETURN(PTR_ERR(handle));
3975 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3976 lname, so_attr, p_la, ldata,
3979 /* If the migration can not be fit in one transaction, just
3980 * leave it in the original MDT */
3982 GOTO(stop_trans, rc = 0);
3984 GOTO(stop_trans, rc);
3987 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3988 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3989 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3990 PFID(mdd_object_fid(mdd_tobj)));
3992 rc = mdd_trans_start(env, mdd, handle);
3994 GOTO(stop_trans, rc);
3996 /* Revert IMMUTABLE flag */
3997 la_flag->la_valid = LA_FLAGS;
3998 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3999 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4001 GOTO(stop_trans, rc);
4003 /* Remove source name from source directory */
4004 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4006 GOTO(stop_trans, rc);
4008 if (ldata->ld_buf != NULL) {
4009 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4010 lname, ldata, handle);
4012 GOTO(stop_trans, rc);
4014 /* linkea update might decrease the source object
4015 * nlink, let's get the attr again after ref_del */
4016 rc = mdd_la_get(env, mdd_sobj, so_attr);
4018 GOTO(stop_trans, rc);
4021 if (S_ISREG(so_attr->la_mode)) {
4022 if (so_attr->la_nlink == 1) {
4023 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4025 if (rc != 0 && rc != -ENODATA)
4026 GOTO(stop_trans, rc);
4028 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4030 LU_XATTR_REPLACE, handle);
4032 GOTO(stop_trans, rc);
4036 /* Insert new fid with target name into target dir */
4037 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4038 mdd_object_type(mdd_tobj), name, handle);
4040 GOTO(stop_trans, rc);
4042 linkea_add_buf(ldata, lname, mdd_object_fid(mdd_pobj));
4043 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
4046 GOTO(stop_trans, rc);
4048 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4050 mdd_sobj->mod_flags |= DEAD_OBJ;
4051 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4053 GOTO(out_unlock, rc);
4055 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4057 GOTO(out_unlock, rc);
4059 mdo_ref_del(env, mdd_sobj, handle);
4061 mdo_ref_del(env, mdd_sobj, handle);
4063 /* Get the attr again after ref_del */
4064 rc = mdd_la_get(env, mdd_sobj, so_attr);
4066 GOTO(out_unlock, rc);
4068 ma->ma_attr = *so_attr;
4069 ma->ma_valid |= MA_INODE;
4071 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4073 GOTO(out_unlock, rc);
4075 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4076 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4077 mdo2fid(mdd_pobj), lname, lname, handle);
4079 CWARN("%s: changelog for migrate %s "DFID
4080 "under "DFID" failed: rc = %d\n",
4081 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4082 PFID(mdd_object_fid(mdd_sobj)),
4083 PFID(mdd_object_fid(mdd_pobj)), rc);
4084 /* Sigh, there are no easy way to migrate back the object, so
4085 * let's reset the result to 0 for now XXX */
4089 mdd_write_unlock(env, mdd_sobj);
4092 mdd_trans_stop(env, mdd, rc, handle);
4097 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4098 const struct lu_fid *fid, __u32 *mdt_index)
4100 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4101 struct seq_server_site *ss;
4104 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4106 range->lsr_flags = LU_SEQ_RANGE_MDT;
4107 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4111 *mdt_index = range->lsr_index;
4116 * Check whether we should migrate the file/dir
4118 * < 0 permission check failed or other error.
4119 * = 0 the file can be migrated.
4120 * > 0 the file does not need to be migrated, mostly
4121 * for multiple link file
4123 static int mdd_migrate_sanity_check(const struct lu_env *env,
4124 struct mdd_object *pobj,
4125 const struct lu_attr *pattr,
4126 struct mdd_object *sobj,
4127 struct lu_attr *sattr)
4129 struct mdd_thread_info *info = mdd_env_info(env);
4130 struct linkea_data *ldata = &info->mti_link_data;
4131 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4133 struct lu_buf *mgr_buf;
4139 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4140 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4141 if (mgr_buf->lb_buf == NULL)
4144 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4146 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4148 /* If the object has migrateEA, it means IMMUTE flag
4149 * is being set by previous migration process, so it
4150 * needs to override the IMMUTE flag, otherwise the
4151 * following sanity check will fail */
4152 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4153 LMV_HASH_FLAG_MIGRATION) {
4154 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4156 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4157 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4158 mdd2obd_dev(mdd)->obd_name,
4159 PFID(mdd_object_fid(sobj)));
4163 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4164 sobj, sattr, NULL, NULL);
4168 /* Then it will check if the file should be migrated. If the file
4169 * has mulitple links, we only need migrate the file if all of its
4170 * entries has been migrated to the remote MDT */
4171 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4174 rc = mdd_links_read(env, sobj, ldata);
4176 /* For multiple links files, if there are no linkEA data at all,
4177 * means the file might be created before linkEA is enabled, and
4178 * all of its links should not be migrated yet, otherwise it
4179 * should have some linkEA there */
4180 if (rc == -ENOENT || rc == -ENODATA)
4185 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4186 /* If there are still links locally, then the file will not be
4188 LASSERT(ldata->ld_leh != NULL);
4189 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4190 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4191 struct lu_name lname;
4193 __u32 parent_mdt_index;
4195 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4197 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4200 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4204 /* Migrate the object only if none of its parents are on the
4206 if (parent_mdt_index != mdt_index)
4209 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4210 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4211 lname.ln_name, PFID(&fid));
4219 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4220 struct md_object *sobj, const struct lu_name *lname,
4221 struct md_object *tobj, struct md_attr *ma)
4223 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4224 struct mdd_device *mdd = mdo2mdd(pobj);
4225 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4226 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4227 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4228 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4229 bool created = false;
4233 /* If the file will being migrated, it will check whether
4234 * the file is being opened by someone else right now */
4235 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4236 if (mdd_sobj->mod_count > 0) {
4238 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4239 mdd2obd_dev(mdd)->obd_name,
4240 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4241 mdd_sobj->mod_count, -EBUSY);
4242 mdd_read_unlock(env, mdd_sobj);
4243 GOTO(put, rc = -EBUSY);
4245 mdd_read_unlock(env, mdd_sobj);
4247 rc = mdd_la_get(env, mdd_sobj, so_attr);
4251 rc = mdd_la_get(env, mdd_pobj, pattr);
4255 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4262 /* Sigh, it is impossible to finish all of migration in a single
4263 * transaction, for example migrating big directory entries to the
4264 * new MDT, it needs insert all of name entries of children in the
4267 * So migration will be done in multiple steps and transactions.
4269 * 1. create an orphan object on the remote MDT in one transaction.
4270 * 2. migrate extend attributes to the new target file/directory.
4271 * 3. For directory, migrate the entries to the new MDT and update
4272 * linkEA of each children. Because we can not migrate all entries
4273 * in a single transaction, so the migrating directory will become
4274 * a striped directory during migration, so once the process is
4275 * interrupted, the directory is still accessible. (During lookup,
4276 * client will locate the name by searching both original and target
4278 * 4. Finally, update the name/FID to point to the new file/directory
4279 * in a separate transaction.
4282 /* step 1: Check whether the orphan object has been created, and create
4283 * orphan object on the remote MDT if needed */
4284 if (!mdd_object_exists(mdd_tobj)) {
4285 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4292 LASSERT(mdd_object_exists(mdd_tobj));
4293 /* step 2: migrate xattr */
4294 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4298 /* step 3: migrate name entries to the orphan object */
4299 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4300 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4303 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4304 OBD_FAIL_MDS_REINT_NET_REP)))
4307 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4310 LASSERT(mdd_object_exists(mdd_tobj));
4311 /* step 4: update name entry to the new object */
4312 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4317 /* newly created target was not locked, don't cache its attributes */
4319 mdd_invalidate(env, tobj);
4324 const struct md_dir_operations mdd_dir_ops = {
4325 .mdo_is_subdir = mdd_is_subdir,
4326 .mdo_lookup = mdd_lookup,
4327 .mdo_create = mdd_create,
4328 .mdo_rename = mdd_rename,
4329 .mdo_link = mdd_link,
4330 .mdo_unlink = mdd_unlink,
4331 .mdo_create_data = mdd_create_data,
4332 .mdo_migrate = mdd_migrate,