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)
1259 rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
1260 name->ln_name, handle);
1264 rc = mdo_declare_ref_add(env, c, handle);
1268 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1269 rc = mdo_declare_ref_add(env, c, handle);
1274 la->la_valid = LA_CTIME | LA_MTIME;
1275 rc = mdo_declare_attr_set(env, p, la, handle);
1279 la->la_valid = LA_CTIME;
1280 rc = mdo_declare_attr_set(env, c, la, handle);
1284 rc = mdd_declare_links_add(env, c, handle, data,
1285 S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE);
1289 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1294 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1295 struct md_object *src_obj, const struct lu_name *lname,
1298 const char *name = lname->ln_name;
1299 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1300 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1301 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1302 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1303 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1304 struct mdd_device *mdd = mdo2mdd(src_obj);
1305 struct thandle *handle;
1306 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1310 rc = mdd_la_get(env, mdd_sobj, cattr);
1314 rc = mdd_la_get(env, mdd_tobj, tattr);
1318 handle = mdd_trans_create(env, mdd);
1320 GOTO(out_pending, rc = PTR_ERR(handle));
1322 memset(ldata, 0, sizeof(*ldata));
1324 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1325 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1327 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1332 rc = mdd_trans_start(env, mdd, handle);
1336 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1337 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1340 GOTO(out_unlock, rc);
1342 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1343 rc = mdo_ref_add(env, mdd_sobj, handle);
1345 GOTO(out_unlock, rc);
1348 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1349 rc = mdo_ref_add(env, mdd_sobj, handle);
1351 GOTO(out_unlock, rc);
1354 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) {
1355 struct lu_fid tfid = *mdo2fid(mdd_sobj);
1358 rc = __mdd_index_insert_only(env, mdd_tobj, &tfid,
1359 mdd_object_type(mdd_sobj),
1362 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
1363 mdd_object_type(mdd_sobj),
1368 mdo_ref_del(env, mdd_sobj, handle);
1369 GOTO(out_unlock, rc);
1372 la->la_valid = LA_CTIME | LA_MTIME;
1373 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1375 GOTO(out_unlock, rc);
1377 la->la_valid = LA_CTIME;
1378 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1380 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1381 mdo2fid(mdd_tobj), lname, 0, 0,
1384 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1385 lname, handle, ldata, 0);
1386 /* The failure of links_add should not cause the link
1387 * failure, reset rc here */
1392 mdd_write_unlock(env, mdd_sobj);
1394 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1395 mdo2fid(mdd_tobj), NULL, NULL,
1396 lname, NULL, handle);
1398 mdd_trans_stop(env, mdd, rc, handle);
1400 if (is_vmalloc_addr(ldata->ld_buf))
1401 /* if we vmalloced a large buffer drop it */
1402 lu_buf_free(ldata->ld_buf);
1407 static int mdd_mark_orphan_object(const struct lu_env *env,
1408 struct mdd_object *obj, struct thandle *handle,
1411 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1414 if (!S_ISDIR(mdd_object_type(obj)))
1417 attr->la_valid = LA_FLAGS;
1418 attr->la_flags = LUSTRE_ORPHAN_FL;
1421 rc = mdo_declare_attr_set(env, obj, attr, handle);
1423 rc = mdo_attr_set(env, obj, attr, handle);
1428 static int mdd_declare_finish_unlink(const struct lu_env *env,
1429 struct mdd_object *obj,
1430 struct thandle *handle)
1434 /* Sigh, we do not know if the unlink object will become orphan in
1435 * declare phase, but fortunately the flags here does not matter
1436 * in current declare implementation */
1437 rc = mdd_mark_orphan_object(env, obj, handle, true);
1441 rc = mdo_declare_destroy(env, obj, handle);
1445 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1449 return mdd_declare_links_del(env, obj, handle);
1452 /* caller should take a lock before calling */
1453 int mdd_finish_unlink(const struct lu_env *env,
1454 struct mdd_object *obj, struct md_attr *ma,
1455 const struct mdd_object *pobj,
1456 const struct lu_name *lname,
1460 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1463 LASSERT(mdd_write_locked(env, obj) != 0);
1465 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1466 /* add new orphan and the object
1467 * will be deleted during mdd_close() */
1468 obj->mod_flags |= DEAD_OBJ;
1469 if (obj->mod_count) {
1470 rc = __mdd_orphan_add(env, obj, th);
1472 CDEBUG(D_HA, "Object "DFID" is inserted into "
1473 "orphan list, open count = %d\n",
1474 PFID(mdd_object_fid(obj)),
1477 CERROR("Object "DFID" fail to be an orphan, "
1478 "open count = %d, maybe cause failed "
1480 PFID(mdd_object_fid(obj)),
1483 /* mark object as an orphan here, not
1484 * before __mdd_orphan_add() as racing
1485 * mdd_la_get() may propagate ORPHAN_OBJ
1486 * causing the asserition */
1487 rc = mdd_mark_orphan_object(env, obj, th, false);
1489 rc = mdo_destroy(env, obj, th);
1491 } else if (!is_dir) {
1492 /* old files may not have link ea; ignore errors */
1493 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1501 * has mdd_write_lock on cobj already, but not on pobj yet
1503 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1504 const struct lu_attr *pattr,
1505 struct mdd_object *cobj,
1506 const struct lu_attr *cattr)
1511 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1516 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1517 struct mdd_object *p, struct mdd_object *c,
1518 const struct lu_name *name, struct md_attr *ma,
1519 struct thandle *handle, int no_name, int is_dir)
1521 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1524 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1525 if (likely(no_name == 0)) {
1526 rc = mdo_declare_index_delete(env, p, name->ln_name,
1533 rc = mdo_declare_ref_del(env, p, handle);
1539 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1540 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1541 la->la_valid = LA_CTIME | LA_MTIME;
1542 rc = mdo_declare_attr_set(env, p, la, handle);
1547 rc = mdo_declare_ref_del(env, c, handle);
1551 rc = mdo_declare_ref_del(env, c, handle);
1555 la->la_valid = LA_CTIME;
1556 rc = mdo_declare_attr_set(env, c, la, handle);
1560 rc = mdd_declare_finish_unlink(env, c, handle);
1564 /* FIXME: need changelog for remove entry */
1565 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1572 * test if a file has an HSM archive
1573 * if HSM attributes are not found in ma update them from
1576 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1577 struct mdd_object *obj,
1582 if (!(ma->ma_valid & MA_HSM)) {
1583 /* no HSM MD provided, read xattr */
1584 struct lu_buf *hsm_buf;
1585 const size_t buflen = sizeof(struct hsm_attrs);
1588 hsm_buf = mdd_buf_get(env, NULL, 0);
1589 lu_buf_alloc(hsm_buf, buflen);
1590 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1591 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1592 lu_buf_free(hsm_buf);
1596 ma->ma_valid |= MA_HSM;
1598 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1604 * Delete name entry and the object.
1605 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1606 * does not exist for this object, and it could only happen during resending
1607 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1608 * is also needed in this case(needed by changelog), so we have to add another
1609 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1610 * the ENOENT failure should be able to be fixed by redo mechanism.
1612 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1613 struct md_object *cobj, const struct lu_name *lname,
1614 struct md_attr *ma, int no_name)
1616 const char *name = lname->ln_name;
1617 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1618 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1619 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1620 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1621 struct mdd_object *mdd_cobj = NULL;
1622 struct mdd_device *mdd = mdo2mdd(pobj);
1623 struct thandle *handle;
1624 int rc, is_dir = 0, cl_flags = 0;
1627 /* cobj == NULL means only delete name entry */
1628 if (likely(cobj != NULL)) {
1629 mdd_cobj = md2mdd_obj(cobj);
1630 if (mdd_object_exists(mdd_cobj) == 0)
1634 rc = mdd_la_get(env, mdd_pobj, pattr);
1638 if (likely(mdd_cobj != NULL)) {
1640 rc = mdd_la_get(env, mdd_cobj, cattr);
1644 is_dir = S_ISDIR(cattr->la_mode);
1645 /* search for an existing archive.
1646 * we should check ahead as the object
1647 * can be destroyed in this transaction */
1648 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1649 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1652 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1656 handle = mdd_trans_create(env, mdd);
1658 RETURN(PTR_ERR(handle));
1660 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1661 lname, ma, handle, no_name, is_dir);
1665 rc = mdd_trans_start(env, mdd, handle);
1669 if (likely(mdd_cobj != NULL))
1670 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1672 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1673 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1678 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1679 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1680 GOTO(cleanup, rc = 0);
1682 if (likely(mdd_cobj != NULL)) {
1683 rc = mdo_ref_del(env, mdd_cobj, handle);
1685 __mdd_index_insert_only(env, mdd_pobj,
1687 mdd_object_type(mdd_cobj),
1694 mdo_ref_del(env, mdd_cobj, handle);
1696 /* fetch updated nlink */
1697 rc = mdd_la_get(env, mdd_cobj, cattr);
1702 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1703 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1705 la->la_valid = LA_CTIME | LA_MTIME;
1706 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1710 /* Enough for only unlink the entry */
1711 if (unlikely(mdd_cobj == NULL))
1714 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1715 /* update ctime of an unlinked file only if it is still
1716 * opened or a link still exists */
1717 la->la_valid = LA_CTIME;
1718 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1723 /* XXX: this transfer to ma will be removed with LOD/OSP */
1724 ma->ma_attr = *cattr;
1725 ma->ma_valid |= MA_INODE;
1726 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1730 /* fetch updated nlink */
1731 rc = mdd_la_get(env, mdd_cobj, cattr);
1732 /* if object is removed then we can't get its attrs,
1734 if (rc == -ENOENT) {
1735 cattr->la_nlink = 0;
1739 if (cattr->la_nlink == 0) {
1740 ma->ma_attr = *cattr;
1741 ma->ma_valid |= MA_INODE;
1746 if (likely(mdd_cobj != NULL))
1747 mdd_write_unlock(env, mdd_cobj);
1750 if (cattr->la_nlink == 0)
1751 cl_flags |= CLF_UNLINK_LAST;
1753 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1755 rc = mdd_changelog_ns_store(env, mdd,
1756 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1757 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1762 mdd_trans_stop(env, mdd, rc, handle);
1768 * The permission has been checked when obj created, no need check again.
1770 static int mdd_cd_sanity_check(const struct lu_env *env,
1771 struct mdd_object *obj)
1776 if (!obj || mdd_is_dead_obj(obj))
1782 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1783 struct md_object *cobj, const struct md_op_spec *spec,
1786 struct mdd_device *mdd = mdo2mdd(cobj);
1787 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1788 struct mdd_object *son = md2mdd_obj(cobj);
1789 struct thandle *handle;
1790 const struct lu_buf *buf;
1791 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1792 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1796 rc = mdd_cd_sanity_check(env, son);
1800 if (!md_should_create(spec->sp_cr_flags))
1804 * there are following use cases for this function:
1805 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1806 * striping can be specified or not
1809 rc = mdd_la_get(env, son, attr);
1813 /* calling ->ah_make_hint() is used to transfer information from parent */
1814 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1816 handle = mdd_trans_create(env, mdd);
1818 GOTO(out_free, rc = PTR_ERR(handle));
1821 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1822 * Should this be fixed?
1824 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1825 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1826 spec->sp_cr_flags, spec->no_create);
1828 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1829 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1830 spec->u.sp_ea.eadatalen);
1835 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1836 XATTR_NAME_LOV, 0, handle);
1840 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1844 rc = mdd_trans_start(env, mdd, handle);
1848 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1854 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1857 mdd_trans_stop(env, mdd, rc, handle);
1862 static int mdd_declare_object_initialize(const struct lu_env *env,
1863 struct mdd_object *parent,
1864 struct mdd_object *child,
1865 const struct lu_attr *attr,
1866 struct thandle *handle)
1871 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1872 if (!S_ISDIR(attr->la_mode))
1875 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1880 rc = mdo_declare_ref_add(env, child, handle);
1884 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1890 static int mdd_object_initialize(const struct lu_env *env,
1891 const struct lu_fid *pfid,
1892 struct mdd_object *child,
1893 struct lu_attr *attr, struct thandle *handle,
1894 const struct md_op_spec *spec)
1899 if (S_ISDIR(attr->la_mode)) {
1900 /* Add "." and ".." for newly created dir */
1901 mdo_ref_add(env, child, handle);
1902 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1903 S_IFDIR, dot, handle);
1905 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1908 mdo_ref_del(env, child, handle);
1915 * This function checks whether it can create a file/dir under the
1916 * directory(@pobj). The directory(@pobj) is not being locked by
1919 * \param[in] env execution environment
1920 * \param[in] pobj the directory to create files
1921 * \param[in] pattr the attributes of the directory
1922 * \param[in] lname the name of the created file/dir
1923 * \param[in] cattr the attributes of the file/dir
1924 * \param[in] spec create specification
1926 * \retval = 0 it is allowed to create file/dir under
1928 * \retval negative error not allowed to create file/dir
1929 * under the directory
1931 static int mdd_create_sanity_check(const struct lu_env *env,
1932 struct md_object *pobj,
1933 const struct lu_attr *pattr,
1934 const struct lu_name *lname,
1935 struct lu_attr *cattr,
1936 struct md_op_spec *spec)
1938 struct mdd_thread_info *info = mdd_env_info(env);
1939 struct lu_fid *fid = &info->mti_fid;
1940 struct mdd_object *obj = md2mdd_obj(pobj);
1941 struct mdd_device *m = mdo2mdd(pobj);
1942 bool check_perm = true;
1947 if (mdd_is_dead_obj(obj))
1951 * In some cases this lookup is not needed - we know before if name
1952 * exists or not because MDT performs lookup for it.
1953 * name length check is done in lookup.
1955 if (spec->sp_cr_lookup) {
1957 * Check if the name already exist, though it will be checked in
1958 * _index_insert also, for avoiding rolling back if exists
1961 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1962 MAY_WRITE | MAY_EXEC);
1964 RETURN(rc ? : -EEXIST);
1966 /* Permission is already being checked in mdd_lookup */
1970 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1975 if (pattr->la_mode & S_ISGID) {
1976 cattr->la_gid = pattr->la_gid;
1977 if (S_ISDIR(cattr->la_mode)) {
1978 cattr->la_mode |= S_ISGID;
1979 cattr->la_valid |= LA_MODE;
1983 rc = mdd_name_check(m, lname);
1987 switch (cattr->la_mode & S_IFMT) {
1989 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1991 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
1992 RETURN(-ENAMETOOLONG);
2011 static int mdd_declare_object_create(const struct lu_env *env,
2012 struct mdd_device *mdd,
2013 struct mdd_object *p, struct mdd_object *c,
2014 struct lu_attr *attr,
2015 struct thandle *handle,
2016 const struct md_op_spec *spec,
2017 struct lu_buf *def_acl_buf,
2018 struct lu_buf *acl_buf,
2019 struct dt_allocation_hint *hint)
2023 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
2028 #ifdef CONFIG_FS_POSIX_ACL
2029 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2030 /* if dir, then can inherit default ACl */
2031 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2032 XATTR_NAME_ACL_DEFAULT,
2038 if (acl_buf->lb_len > 0) {
2039 rc = mdo_declare_attr_set(env, c, attr, handle);
2043 rc = mdo_declare_xattr_set(env, c, acl_buf,
2044 XATTR_NAME_ACL_ACCESS, 0, handle);
2049 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2053 /* replay case, create LOV EA from client data */
2054 if (spec->no_create ||
2055 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2056 const struct lu_buf *buf;
2058 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2059 spec->u.sp_ea.eadatalen);
2060 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2066 if (S_ISLNK(attr->la_mode)) {
2067 const char *target_name = spec->u.sp_symname;
2068 int sym_len = strlen(target_name);
2069 const struct lu_buf *buf;
2071 buf = mdd_buf_get_const(env, target_name, sym_len);
2072 rc = dt_declare_record_write(env, mdd_object_child(c),
2081 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2082 struct mdd_object *p, struct mdd_object *c,
2083 const struct lu_name *name,
2084 struct lu_attr *attr,
2085 struct thandle *handle,
2086 const struct md_op_spec *spec,
2087 struct linkea_data *ldata,
2088 struct lu_buf *def_acl_buf,
2089 struct lu_buf *acl_buf,
2090 struct dt_allocation_hint *hint)
2094 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2095 def_acl_buf, acl_buf, hint);
2099 if (S_ISDIR(attr->la_mode)) {
2100 rc = mdo_declare_ref_add(env, p, handle);
2105 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2106 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2110 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2112 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2113 name->ln_name, handle);
2117 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2122 la->la_valid = LA_CTIME | LA_MTIME;
2123 rc = mdo_declare_attr_set(env, p, la, handle);
2127 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2135 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2136 struct lu_attr *la, struct lu_buf *def_acl_buf,
2137 struct lu_buf *acl_buf)
2142 if (S_ISLNK(la->la_mode)) {
2143 acl_buf->lb_len = 0;
2144 def_acl_buf->lb_len = 0;
2148 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2149 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2150 XATTR_NAME_ACL_DEFAULT);
2151 mdd_read_unlock(env, pobj);
2153 /* If there are default ACL, fix mode/ACL by default ACL */
2154 def_acl_buf->lb_len = rc;
2155 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2156 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2157 acl_buf->lb_len = rc;
2158 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2161 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2162 /* If there are no default ACL, fix mode by mask */
2163 struct lu_ucred *uc = lu_ucred(env);
2165 /* The create triggered by MDT internal events, such as
2166 * LFSCK reset, will not contain valid "uc". */
2167 if (unlikely(uc != NULL))
2168 la->la_mode &= ~uc->uc_umask;
2170 acl_buf->lb_len = 0;
2171 def_acl_buf->lb_len = 0;
2178 * Create a metadata object and initialize it, set acl, xattr.
2180 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2181 struct mdd_object *son, struct lu_attr *attr,
2182 struct md_op_spec *spec, struct lu_buf *acl_buf,
2183 struct lu_buf *def_acl_buf,
2184 struct dt_allocation_hint *hint,
2185 struct thandle *handle)
2189 mdd_write_lock(env, son, MOR_TGT_CHILD);
2190 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2195 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2196 * created in declare phase, they also needs to be added to master
2197 * object as sub-directory entry. So it has to initialize the master
2198 * object, then set dir striped EA.(in mdo_xattr_set) */
2199 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2202 GOTO(err_destroy, rc);
2205 * in case of replay we just set LOVEA provided by the client
2206 * XXX: I think it would be interesting to try "old" way where
2207 * MDT calls this xattr_set(LOV) in a different transaction.
2208 * probably this way we code can be made better.
2211 /* During creation, there are only a few cases we need do xattr_set to
2213 * 1. regular file: see comments above.
2214 * 2. dir: inherit default striping or pool settings from parent.
2215 * 3. create striped directory with provided stripeEA.
2216 * 4. create striped directory because inherit default layout from the
2219 if (spec->no_create ||
2220 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2221 S_ISDIR(attr->la_mode)) {
2222 const struct lu_buf *buf;
2224 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2225 spec->u.sp_ea.eadatalen);
2226 rc = mdo_xattr_set(env, son, buf,
2227 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2231 GOTO(err_destroy, rc);
2234 #ifdef CONFIG_FS_POSIX_ACL
2235 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2236 S_ISDIR(attr->la_mode)) {
2237 /* set default acl */
2238 rc = mdo_xattr_set(env, son, def_acl_buf,
2239 XATTR_NAME_ACL_DEFAULT, 0,
2242 GOTO(err_destroy, rc);
2244 /* set its own acl */
2245 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2246 rc = mdo_xattr_set(env, son, acl_buf,
2247 XATTR_NAME_ACL_ACCESS,
2250 GOTO(err_destroy, rc);
2254 if (S_ISLNK(attr->la_mode)) {
2255 struct lu_ucred *uc = lu_ucred_assert(env);
2256 struct dt_object *dt = mdd_object_child(son);
2257 const char *target_name = spec->u.sp_symname;
2258 int sym_len = strlen(target_name);
2259 const struct lu_buf *buf;
2262 buf = mdd_buf_get_const(env, target_name, sym_len);
2263 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2265 CFS_CAP_SYS_RESOURCE_MASK);
2270 GOTO(err_initlized, rc = -EFAULT);
2274 if (unlikely(rc != 0)) {
2276 if (S_ISDIR(attr->la_mode)) {
2277 /* Drop the reference, no need to delete "."/"..",
2278 * because the object to be destroied directly. */
2279 rc2 = mdo_ref_del(env, son, handle);
2283 rc2 = mdo_ref_del(env, son, handle);
2287 mdo_destroy(env, son, handle);
2290 mdd_write_unlock(env, son);
2294 static int mdd_index_delete(const struct lu_env *env,
2295 struct mdd_object *mdd_pobj,
2296 struct lu_attr *cattr,
2297 const struct lu_name *lname)
2299 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2300 struct thandle *handle;
2304 handle = mdd_trans_create(env, mdd);
2306 RETURN(PTR_ERR(handle));
2308 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2313 if (S_ISDIR(cattr->la_mode)) {
2314 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2319 /* Since this will only be used in the error handler path,
2320 * Let's set the thandle to be local and not mess the transno */
2321 handle->th_local = 1;
2322 rc = mdd_trans_start(env, mdd, handle);
2326 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2327 S_ISDIR(cattr->la_mode), handle);
2331 mdd_trans_stop(env, mdd, rc, handle);
2336 * Create object and insert it into namespace.
2338 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2339 const struct lu_name *lname, struct md_object *child,
2340 struct md_op_spec *spec, struct md_attr* ma)
2342 struct mdd_thread_info *info = mdd_env_info(env);
2343 struct lu_attr *la = &info->mti_la_for_fix;
2344 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2345 struct mdd_object *son = md2mdd_obj(child);
2346 struct mdd_device *mdd = mdo2mdd(pobj);
2347 struct lu_attr *attr = &ma->ma_attr;
2348 struct thandle *handle;
2349 struct lu_attr *pattr = &info->mti_pattr;
2350 struct lu_buf acl_buf;
2351 struct lu_buf def_acl_buf;
2352 struct linkea_data *ldata = &info->mti_link_data;
2353 const char *name = lname->ln_name;
2354 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2360 * Two operations have to be performed:
2362 * - an allocation of a new object (->do_create()), and
2364 * - an insertion into a parent index (->dio_insert()).
2366 * Due to locking, operation order is not important, when both are
2367 * successful, *but* error handling cases are quite different:
2369 * - if insertion is done first, and following object creation fails,
2370 * insertion has to be rolled back, but this operation might fail
2371 * also leaving us with dangling index entry.
2373 * - if creation is done first, is has to be undone if insertion
2374 * fails, leaving us with leaked space, which is neither good, nor
2377 * It seems that creation-first is simplest solution, but it is
2378 * sub-optimal in the frequent
2383 * case, because second mkdir is bound to create object, only to
2384 * destroy it immediately.
2386 * To avoid this follow local file systems that do double lookup:
2388 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2390 * 1. create (mdd_object_create_internal())
2392 * 2. insert (__mdd_index_insert(), lookup again)
2395 rc = mdd_la_get(env, mdd_pobj, pattr);
2399 /* Sanity checks before big job. */
2400 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2404 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2405 GOTO(out_free, rc = -EINPROGRESS);
2407 handle = mdd_trans_create(env, mdd);
2409 GOTO(out_free, rc = PTR_ERR(handle));
2411 acl_buf.lb_buf = info->mti_xattr_buf;
2412 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2413 def_acl_buf.lb_buf = info->mti_key;
2414 def_acl_buf.lb_len = sizeof(info->mti_key);
2415 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2419 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2421 memset(ldata, 0, sizeof(*ldata));
2422 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2423 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2426 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2427 &tfid, lname, 1, 0, ldata);
2429 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2430 mdd_object_fid(mdd_pobj),
2431 lname, 1, 0, ldata);
2434 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2435 handle, spec, ldata, &def_acl_buf, &acl_buf,
2440 rc = mdd_trans_start(env, mdd, handle);
2444 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2445 &def_acl_buf, hint, handle);
2449 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2450 mdd_write_lock(env, son, MOR_TGT_CHILD);
2451 rc = __mdd_orphan_add(env, son, handle);
2452 GOTO(out_volatile, rc);
2454 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2455 attr->la_mode, name, handle);
2457 GOTO(err_created, rc);
2459 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2462 /* update parent directory mtime/ctime */
2464 la->la_valid = LA_CTIME | LA_MTIME;
2465 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2467 GOTO(err_insert, rc);
2475 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2476 rc2 = __mdd_orphan_del(env, son, handle);
2478 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2479 S_ISDIR(attr->la_mode),
2485 mdd_write_lock(env, son, MOR_TGT_CHILD);
2486 if (S_ISDIR(attr->la_mode)) {
2487 /* Drop the reference, no need to delete "."/"..",
2488 * because the object is to be destroyed directly. */
2489 rc2 = mdo_ref_del(env, son, handle);
2491 mdd_write_unlock(env, son);
2496 /* For volatile files drop one link immediately, since there is
2497 * no filename in the namespace, and save any error returned. */
2498 rc2 = mdo_ref_del(env, son, handle);
2500 mdd_write_unlock(env, son);
2501 if (unlikely(rc == 0))
2506 /* Don't destroy the volatile object on success */
2507 if (likely(rc != 0))
2508 mdo_destroy(env, son, handle);
2509 mdd_write_unlock(env, son);
2512 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2513 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2514 rc = mdd_changelog_ns_store(env, mdd,
2515 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2516 S_ISREG(attr->la_mode) ? CL_CREATE :
2517 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2518 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2521 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2523 /* If creation fails, it is most likely due to the remote update
2524 * failure, because local transaction will mostly succeed at
2525 * this stage. There is no easy way to rollback all of previous
2526 * updates, so let's remove the object from namespace, and
2527 * LFSCK should handle the orphan object. */
2528 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2529 mdd_index_delete(env, mdd_pobj, attr, lname);
2533 if (is_vmalloc_addr(ldata->ld_buf))
2534 /* if we vmalloced a large buffer drop it */
2535 lu_buf_free(ldata->ld_buf);
2537 /* The child object shouldn't be cached anymore */
2539 set_bit(LU_OBJECT_HEARD_BANSHEE,
2540 &child->mo_lu.lo_header->loh_flags);
2545 * Get locks on parents in proper order
2546 * RETURN: < 0 - error, rename_order if successful
2554 static int mdd_rename_order(const struct lu_env *env,
2555 struct mdd_device *mdd,
2556 struct mdd_object *src_pobj,
2557 const struct lu_attr *pattr,
2558 struct mdd_object *tgt_pobj)
2560 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2564 if (src_pobj == tgt_pobj)
2565 RETURN(MDD_RN_SAME);
2567 /* compared the parent child relationship of src_p&tgt_p */
2568 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2570 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2573 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2587 /* has not mdd_write{read}_lock on any obj yet. */
2588 static int mdd_rename_sanity_check(const struct lu_env *env,
2589 struct mdd_object *src_pobj,
2590 const struct lu_attr *pattr,
2591 struct mdd_object *tgt_pobj,
2592 const struct lu_attr *tpattr,
2593 struct mdd_object *sobj,
2594 const struct lu_attr *cattr,
2595 struct mdd_object *tobj,
2596 const struct lu_attr *tattr)
2601 /* XXX: when get here, sobj must NOT be NULL,
2602 * the other case has been processed in cld_rename
2603 * before mdd_rename and enable MDS_PERM_BYPASS. */
2606 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2610 /* XXX: when get here, "tobj == NULL" means tobj must
2611 * NOT exist (neither on remote MDS, such case has been
2612 * processed in cld_rename before mdd_rename and enable
2614 * So check may_create, but not check may_unlink. */
2616 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2617 (src_pobj != tgt_pobj));
2619 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2620 (src_pobj != tgt_pobj), 1);
2622 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2623 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2628 static int mdd_declare_rename(const struct lu_env *env,
2629 struct mdd_device *mdd,
2630 struct mdd_object *mdd_spobj,
2631 struct mdd_object *mdd_tpobj,
2632 struct mdd_object *mdd_sobj,
2633 struct mdd_object *mdd_tobj,
2634 const struct lu_name *sname,
2635 const struct lu_name *tname,
2637 struct linkea_data *ldata,
2638 struct thandle *handle)
2640 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2643 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2644 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2650 /* name from source dir */
2651 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2655 /* .. from source child */
2656 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2657 /* source child can be directory,
2658 * counted by source dir's nlink */
2659 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2662 if (mdd_spobj != mdd_tpobj) {
2663 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2668 rc = mdo_declare_index_insert(env, mdd_sobj,
2670 S_IFDIR, dotdot, handle);
2675 /* new target child can be directory,
2676 * counted by target dir's nlink */
2677 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2682 la->la_valid = LA_CTIME | LA_MTIME;
2683 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2687 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2691 la->la_valid = LA_CTIME;
2692 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2696 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
2697 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2702 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2703 mdd_object_type(mdd_sobj),
2704 tname->ln_name, handle);
2708 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2709 /* delete target child in target parent directory */
2710 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2715 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2719 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2720 /* target child can be directory,
2721 * delete "." reference in target child directory */
2722 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2726 /* delete ".." reference in target parent directory */
2727 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2732 la->la_valid = LA_CTIME;
2733 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2737 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2742 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2749 /* src object can be remote that is why we use only fid and type of object */
2750 static int mdd_rename(const struct lu_env *env,
2751 struct md_object *src_pobj, struct md_object *tgt_pobj,
2752 const struct lu_fid *lf, const struct lu_name *lsname,
2753 struct md_object *tobj, const struct lu_name *ltname,
2756 const char *sname = lsname->ln_name;
2757 const char *tname = ltname->ln_name;
2758 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2759 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2760 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2761 struct mdd_device *mdd = mdo2mdd(src_pobj);
2762 struct mdd_object *mdd_sobj = NULL; /* source object */
2763 struct mdd_object *mdd_tobj = NULL;
2764 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2765 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2766 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2767 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2768 struct thandle *handle;
2769 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2770 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2771 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2774 bool tobj_locked = 0;
2775 unsigned cl_flags = 0;
2780 mdd_tobj = md2mdd_obj(tobj);
2782 mdd_sobj = mdd_object_find(env, mdd, lf);
2783 if (IS_ERR(mdd_sobj))
2784 RETURN(PTR_ERR(mdd_sobj));
2786 rc = mdd_la_get(env, mdd_sobj, cattr);
2788 GOTO(out_pending, rc);
2790 rc = mdd_la_get(env, mdd_spobj, pattr);
2792 GOTO(out_pending, rc);
2795 rc = mdd_la_get(env, mdd_tobj, tattr);
2797 GOTO(out_pending, rc);
2798 /* search for an existing archive.
2799 * we should check ahead as the object
2800 * can be destroyed in this transaction */
2801 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2802 cl_flags |= CLF_RENAME_LAST_EXISTS;
2805 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2807 GOTO(out_pending, rc);
2809 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2810 mdd_sobj, cattr, mdd_tobj, tattr);
2812 GOTO(out_pending, rc);
2814 rc = mdd_name_check(mdd, ltname);
2816 GOTO(out_pending, rc);
2818 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2819 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2821 GOTO(out_pending, rc);
2823 handle = mdd_trans_create(env, mdd);
2825 GOTO(out_pending, rc = PTR_ERR(handle));
2827 memset(ldata, 0, sizeof(*ldata));
2828 mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname,
2829 mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata);
2830 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2831 mdd_tobj, lsname, ltname, ma, ldata, handle);
2835 rc = mdd_trans_start(env, mdd, handle);
2839 is_dir = S_ISDIR(cattr->la_mode);
2841 /* Remove source name from source directory */
2842 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2846 /* "mv dir1 dir2" needs "dir1/.." link update */
2847 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2848 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2850 GOTO(fixup_spobj2, rc);
2852 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2855 GOTO(fixup_spobj, rc);
2858 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2859 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2861 /* tname might been renamed to something else */
2862 GOTO(fixup_spobj, rc);
2865 /* Insert new fid with target name into target dir */
2866 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2869 GOTO(fixup_tpobj, rc);
2871 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2872 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2874 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2875 la->la_valid = LA_CTIME;
2876 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2878 GOTO(fixup_tpobj, rc);
2880 /* Update the linkEA for the source object */
2881 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2882 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2883 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2886 /* Old files might not have EA entry */
2887 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2888 lsname, handle, NULL, 0);
2889 mdd_write_unlock(env, mdd_sobj);
2890 /* We don't fail the transaction if the link ea can't be
2891 updated -- fid2path will use alternate lookup method. */
2894 /* Remove old target object
2895 * For tobj is remote case cmm layer has processed
2896 * and set tobj to NULL then. So when tobj is NOT NULL,
2897 * it must be local one.
2899 if (tobj && mdd_object_exists(mdd_tobj)) {
2900 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2902 if (mdd_is_dead_obj(mdd_tobj)) {
2903 /* shld not be dead, something is wrong */
2904 CERROR("tobj is dead, something is wrong\n");
2908 mdo_ref_del(env, mdd_tobj, handle);
2910 /* Remove dot reference. */
2911 if (S_ISDIR(tattr->la_mode))
2912 mdo_ref_del(env, mdd_tobj, handle);
2915 /* fetch updated nlink */
2916 rc = mdd_la_get(env, mdd_tobj, tattr);
2918 CERROR("%s: Failed to get nlink for tobj "
2920 mdd2obd_dev(mdd)->obd_name,
2921 PFID(tpobj_fid), rc);
2922 GOTO(fixup_tpobj, rc);
2925 la->la_valid = LA_CTIME;
2926 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2928 CERROR("%s: Failed to set ctime for tobj "
2930 mdd2obd_dev(mdd)->obd_name,
2931 PFID(tpobj_fid), rc);
2932 GOTO(fixup_tpobj, rc);
2935 /* XXX: this transfer to ma will be removed with LOD/OSP */
2936 ma->ma_attr = *tattr;
2937 ma->ma_valid |= MA_INODE;
2938 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2941 CERROR("%s: Failed to unlink tobj "
2943 mdd2obd_dev(mdd)->obd_name,
2944 PFID(tpobj_fid), rc);
2945 GOTO(fixup_tpobj, rc);
2948 /* fetch updated nlink */
2949 rc = mdd_la_get(env, mdd_tobj, tattr);
2950 if (rc == -ENOENT) {
2951 /* the object got removed, let's
2952 * return the latest known attributes */
2953 tattr->la_nlink = 0;
2955 } else if (rc != 0) {
2956 CERROR("%s: Failed to get nlink for tobj "
2958 mdd2obd_dev(mdd)->obd_name,
2959 PFID(tpobj_fid), rc);
2960 GOTO(fixup_tpobj, rc);
2962 /* XXX: this transfer to ma will be removed with LOD/OSP */
2963 ma->ma_attr = *tattr;
2964 ma->ma_valid |= MA_INODE;
2966 if (tattr->la_nlink == 0)
2967 cl_flags |= CLF_RENAME_LAST;
2969 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
2972 la->la_valid = LA_CTIME | LA_MTIME;
2973 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2975 GOTO(fixup_tpobj, rc);
2977 if (mdd_spobj != mdd_tpobj) {
2978 la->la_valid = LA_CTIME | LA_MTIME;
2979 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2981 GOTO(fixup_tpobj, rc);
2988 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2990 CWARN("tp obj fix error %d\n",rc2);
2992 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2993 !mdd_is_dead_obj(mdd_tobj)) {
2995 mdo_ref_add(env, mdd_tobj, handle);
2997 mdo_ref_add(env, mdd_tobj, handle);
3000 rc2 = __mdd_index_insert(env, mdd_tpobj,
3002 mdd_object_type(mdd_tobj),
3005 CWARN("tp obj fix error: rc = %d\n", rc2);
3010 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3011 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3013 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3014 mdd2obd_dev(mdd)->obd_name, rc2);
3017 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3020 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3021 mdd2obd_dev(mdd)->obd_name, rc2);
3026 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3027 mdd_object_type(mdd_sobj), sname,
3030 CWARN("sp obj fix error: rc = %d\n", rc2);
3035 mdd_write_unlock(env, mdd_tobj);
3038 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3039 mdd_tobj, tpobj_fid, lf, spobj_fid,
3040 ltname, lsname, handle);
3043 mdd_trans_stop(env, mdd, rc, handle);
3046 mdd_object_put(env, mdd_sobj);
3051 * During migration once the parent FID has been changed,
3052 * we need update the parent FID in linkea.
3054 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3055 struct mdd_object *parent,
3056 struct mdd_object *newparent,
3057 struct mdd_object *child,
3058 const char *name, int namelen,
3059 struct thandle *handle,
3062 struct mdd_thread_info *info = mdd_env_info(env);
3063 struct linkea_data ldata = { NULL };
3064 struct lu_buf *buf = &info->mti_link_buf;
3070 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3071 if (buf->lb_buf == NULL)
3075 rc = mdd_links_read(env, child, &ldata);
3077 if (rc == -ENOENT || rc == -ENODATA)
3082 LASSERT(ldata.ld_leh != NULL);
3083 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3084 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3085 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3086 struct lu_name lname;
3089 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3092 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3093 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3094 ldata.ld_lee = (struct link_ea_entry *)
3095 ((char *)ldata.ld_lee +
3100 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3101 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3102 lname.ln_namelen, lname.ln_name,
3103 PFID(mdd_object_fid(newparent)));
3104 /* update to the new parent fid */
3105 linkea_entry_pack(ldata.ld_lee, &lname,
3106 mdd_object_fid(newparent));
3108 rc = mdd_declare_links_add(env, child, handle, &ldata,
3111 rc = mdd_links_write(env, child, &ldata, handle);
3117 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3118 struct mdd_object *parent,
3119 struct mdd_object *newparent,
3120 struct mdd_object *child,
3121 const char *name, int namelen,
3122 struct thandle *handle)
3124 return mdd_linkea_update_child_internal(env, parent, newparent,
3126 namelen, handle, true);
3129 static int mdd_linkea_update_child(const struct lu_env *env,
3130 struct mdd_object *parent,
3131 struct mdd_object *newparent,
3132 struct mdd_object *child,
3133 const char *name, int namelen,
3134 struct thandle *handle)
3136 return mdd_linkea_update_child_internal(env, parent, newparent,
3138 namelen, handle, false);
3141 static int mdd_update_linkea_internal(const struct lu_env *env,
3142 struct mdd_object *mdd_pobj,
3143 struct mdd_object *mdd_sobj,
3144 struct mdd_object *mdd_tobj,
3145 const struct lu_name *child_name,
3146 struct linkea_data *ldata,
3147 struct thandle *handle,
3150 struct mdd_thread_info *info = mdd_env_info(env);
3155 LASSERT(ldata->ld_buf != NULL);
3158 /* If it is mulitple links file, we need update the name entry for
3160 LASSERT(ldata->ld_leh != NULL);
3161 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3162 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3163 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3164 struct mdd_object *pobj;
3165 struct lu_name lname;
3168 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3170 pobj = mdd_object_find(env, mdd, &fid);
3172 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3173 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3175 linkea_del_buf(ldata, &lname);
3179 if (!mdd_object_exists(pobj)) {
3180 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3181 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3182 linkea_del_buf(ldata, &lname);
3183 mdd_object_put(env, pobj);
3187 if (pobj == mdd_pobj &&
3188 lname.ln_namelen == child_name->ln_namelen &&
3189 strncmp(lname.ln_name, child_name->ln_name,
3190 lname.ln_namelen) == 0) {
3191 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3192 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3194 linkea_del_buf(ldata, &lname);
3195 mdd_object_put(env, pobj);
3199 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3200 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3201 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3204 /* Remove source name from source directory */
3205 /* Insert new fid with target name into target dir */
3206 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3211 rc = mdo_declare_index_insert(env, pobj,
3212 mdd_object_fid(mdd_tobj),
3213 mdd_object_type(mdd_tobj),
3214 lname.ln_name, handle);
3218 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3222 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3226 char *tmp_name = info->mti_key;
3228 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3229 /* lnamelen is too big(> NAME_MAX + 16),
3230 * something wrong about this linkea, let's
3232 linkea_del_buf(ldata, &lname);
3233 mdd_object_put(env, pobj);
3237 /* Note: lname might be without \0 at the end, see
3238 * linkea_entry_unpack(), let's add extra \0 by
3240 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3241 lname.ln_namelen, lname.ln_name);
3242 lname.ln_name = tmp_name;
3244 /* Let's check if this linkEA still valid, before
3245 * it might be packed into the RPC buffer. */
3246 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3247 &info->mti_fid, NULL);
3249 !lu_fid_eq(&info->mti_fid,
3250 mdd_object_fid(mdd_sobj))) {
3251 /* skip invalid linkea entry */
3252 linkea_del_buf(ldata, &lname);
3253 mdd_object_put(env, pobj);
3257 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3261 rc = __mdd_index_insert(env, pobj,
3262 mdd_object_fid(mdd_tobj),
3263 mdd_object_type(mdd_tobj),
3268 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3269 rc = mdo_ref_add(env, mdd_tobj, handle);
3270 mdd_write_unlock(env, mdd_tobj);
3274 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3275 mdo_ref_del(env, mdd_sobj, handle);
3276 mdd_write_unlock(env, mdd_sobj);
3279 mdd_object_put(env, pobj);
3283 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3290 static int mdd_migrate_xattrs(const struct lu_env *env,
3291 struct mdd_object *mdd_sobj,
3292 struct mdd_object *mdd_tobj)
3294 struct mdd_thread_info *info = mdd_env_info(env);
3295 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3297 struct thandle *handle;
3303 struct lu_buf list_xbuf;
3307 /* retrieve xattr list from the old object */
3308 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3309 if (list_xsize == -ENODATA)
3315 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3316 if (info->mti_big_buf.lb_buf == NULL)
3319 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3320 list_xbuf.lb_len = list_xsize;
3321 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3326 xname = list_xbuf.lb_buf;
3328 xlen = strnlen(xname, rem - 1) + 1;
3329 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3330 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3331 strcmp(XATTR_NAME_LMV, xname) == 0)
3334 /* For directory, if there are default layout, migrate here */
3335 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3336 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3339 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3340 if (xsize == -ENODATA)
3345 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3346 if (info->mti_link_buf.lb_buf == NULL)
3347 GOTO(out, rc = -ENOMEM);
3349 xbuf.lb_len = xsize;
3350 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3351 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3357 handle = mdd_trans_create(env, mdd);
3359 GOTO(out, rc = PTR_ERR(handle));
3361 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3364 GOTO(stop_trans, rc);
3365 /* Note: this transaction is part of migration, and it is not
3366 * the last step of migration, so we set th_local = 1 to avoid
3367 * update last rcvd for this transaction */
3368 handle->th_local = 1;
3369 rc = mdd_trans_start(env, mdd, handle);
3371 GOTO(stop_trans, rc);
3373 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3375 GOTO(stop_trans, rc = 0);
3378 GOTO(stop_trans, rc);
3380 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3387 memmove(xname, xname + xlen, rem);
3393 static int mdd_declare_migrate_create(const struct lu_env *env,
3394 struct mdd_object *mdd_pobj,
3395 struct mdd_object *mdd_sobj,
3396 struct mdd_object *mdd_tobj,
3397 struct md_op_spec *spec,
3399 union lmv_mds_md *mgr_ea,
3400 struct linkea_data *ldata,
3401 struct thandle *handle)
3403 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3404 const struct lu_buf *buf;
3408 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3409 handle, spec, NULL);
3413 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3418 if (S_ISLNK(la->la_mode)) {
3419 const char *target_name = spec->u.sp_symname;
3420 int sym_len = strlen(target_name);
3421 const struct lu_buf *buf;
3423 buf = mdd_buf_get_const(env, target_name, sym_len);
3424 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3428 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3429 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3435 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3436 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3437 spec->u.sp_ea.eadatalen);
3438 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3444 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3445 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3446 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3451 la_flag->la_valid = LA_FLAGS;
3452 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3453 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3458 static int mdd_migrate_create(const struct lu_env *env,
3459 struct mdd_object *mdd_pobj,
3460 struct mdd_object *mdd_sobj,
3461 struct mdd_object *mdd_tobj,
3462 const struct lu_name *lname,
3465 struct mdd_thread_info *info = mdd_env_info(env);
3466 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3467 struct md_op_spec *spec = &info->mti_spec;
3468 struct lu_buf lmm_buf = { NULL };
3469 struct lu_buf link_buf = { NULL };
3470 const struct lu_buf *buf;
3471 struct thandle *handle;
3472 struct lmv_mds_md_v1 *mgr_ea;
3473 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3474 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3476 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3480 /* prepare spec for create */
3481 memset(spec, 0, sizeof(*spec));
3482 spec->sp_cr_lookup = 0;
3483 spec->sp_feat = &dt_directory_features;
3484 if (S_ISLNK(la->la_mode)) {
3485 buf = lu_buf_check_and_alloc(
3486 &mdd_env_info(env)->mti_big_buf,
3489 link_buf.lb_len = la->la_size + 1;
3490 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3491 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3493 rc = rc != 0 ? rc : -EFAULT;
3494 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3495 mdd2obd_dev(mdd)->obd_name,
3496 PFID(mdd_object_fid(mdd_sobj)), rc);
3499 spec->u.sp_symname = link_buf.lb_buf;
3500 } else if (S_ISREG(la->la_mode)) {
3501 /* retrieve lov of the old object */
3502 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3503 if (rc != 0 && rc != -ENODATA)
3505 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3506 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3507 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3508 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3510 } else if (S_ISDIR(la->la_mode)) {
3511 rc = mdd_links_read(env, mdd_sobj, ldata);
3512 if (rc == -ENODATA) {
3513 /* ignore the non-linkEA error */
3521 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3522 memset(mgr_ea, 0, sizeof(*mgr_ea));
3523 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3524 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3525 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3526 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3527 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3528 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3530 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3532 handle = mdd_trans_create(env, mdd);
3534 GOTO(out_free, rc = PTR_ERR(handle));
3536 /* Note: this transaction is part of migration, and it is not
3537 * the last step of migration, so we set th_local = 1 to avoid
3538 * update last rcvd for this transaction */
3539 handle->th_local = 1;
3540 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3542 (union lmv_mds_md *)info->mti_xattr_buf,
3545 GOTO(stop_trans, rc);
3547 rc = mdd_trans_start(env, mdd, handle);
3549 GOTO(stop_trans, rc);
3551 /* don't set nlink from the original object */
3552 la->la_valid &= ~LA_NLINK;
3554 /* create the target object */
3555 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3558 GOTO(stop_trans, rc);
3560 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3561 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3563 GOTO(stop_trans, rc);
3566 /* Set MIGRATE EA on the source inode, so once the migration needs
3567 * to be re-done during failover, the re-do process can locate the
3568 * target object which is already being created. */
3569 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3570 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3571 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
3573 GOTO(stop_trans, rc);
3575 /* Set immutable flag, so any modification is disabled until
3576 * the migration is done. Once the migration is interrupted,
3577 * if the resume process find the migrating object has both
3578 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3579 * flag and approve the migration */
3580 la_flag->la_valid = LA_FLAGS;
3581 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3582 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3584 if (handle != NULL) {
3587 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3592 if (lmm_buf.lb_buf != NULL)
3593 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3597 static int mdd_migrate_entries(const struct lu_env *env,
3598 struct mdd_object *mdd_sobj,
3599 struct mdd_object *mdd_tobj)
3601 struct dt_object *next = mdd_object_child(mdd_sobj);
3602 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3603 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3604 struct thandle *handle;
3606 const struct dt_it_ops *iops;
3609 struct lu_dirent *ent;
3612 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3616 if (!dt_try_as_dir(env, next))
3617 GOTO(out_ent, rc = -ENOTDIR);
3619 * iterate directories
3621 iops = &next->do_index_ops->dio_it;
3622 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3624 GOTO(out_ent, rc = PTR_ERR(it));
3626 rc = iops->load(env, it, 0);
3628 rc = iops->next(env, it);
3632 * At this point and across for-loop:
3634 * rc == 0 -> ok, proceed.
3635 * rc > 0 -> end of directory.
3639 struct mdd_object *child;
3640 char *name = mdd_env_info(env)->mti_key;
3644 bool target_exist = false;
3647 len = iops->key_size(env, it);
3651 result = iops->rec(env, it, (struct dt_rec *)ent,
3652 LUDA_FID | LUDA_TYPE);
3653 if (result == -ESTALE)
3660 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3661 recsize = le16_to_cpu(ent->lde_reclen);
3663 /* Insert new fid with target name into target dir */
3664 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3665 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3666 ent->lde_name[1] == '.'))
3669 child = mdd_object_find(env, mdd, &ent->lde_fid);
3671 GOTO(out, rc = PTR_ERR(child));
3673 mdd_write_lock(env, child, MOR_SRC_CHILD);
3674 is_dir = S_ISDIR(mdd_object_type(child));
3676 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3678 /* Check whether the name has been inserted to the target */
3679 if (dt_try_as_dir(env, dt_tobj)) {
3680 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3682 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3683 (struct dt_key *)name);
3684 if (unlikely(rc == 0))
3685 target_exist = true;
3688 handle = mdd_trans_create(env, mdd);
3690 GOTO(out_put, rc = PTR_ERR(handle));
3692 /* Note: this transaction is part of migration, and it is not
3693 * the last step of migration, so we set th_local = 1 to avoid
3694 * updating last rcvd for this transaction */
3695 handle->th_local = 1;
3696 if (likely(!target_exist)) {
3697 rc = mdo_declare_index_insert(env, mdd_tobj,
3699 mdd_object_type(child),
3705 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3711 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3716 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3720 /* Update .. for child */
3721 rc = mdo_declare_index_delete(env, child, dotdot,
3726 rc = mdo_declare_index_insert(env, child,
3727 mdd_object_fid(mdd_tobj),
3728 S_IFDIR, dotdot, handle);
3733 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3740 rc = mdd_trans_start(env, mdd, handle);
3742 CERROR("%s: transaction start failed: rc = %d\n",
3743 mdd2obd_dev(mdd)->obd_name, rc);
3747 if (likely(!target_exist)) {
3748 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3749 mdd_object_type(child),
3755 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3760 rc = __mdd_index_delete_only(env, child, dotdot,
3765 rc = __mdd_index_insert_only(env, child,
3766 mdd_object_fid(mdd_tobj), S_IFDIR,
3772 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3774 strlen(name), handle);
3777 mdd_write_unlock(env, child);
3778 mdd_object_put(env, child);
3779 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3786 result = iops->next(env, it);
3787 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3788 GOTO(out, rc = -EINTR);
3790 if (result == -ESTALE)
3792 } while (result == 0);
3795 iops->fini(env, it);
3797 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3801 static int mdd_declare_update_linkea(const struct lu_env *env,
3802 struct mdd_object *mdd_pobj,
3803 struct mdd_object *mdd_sobj,
3804 struct mdd_object *mdd_tobj,
3805 const struct lu_name *child_name,
3806 struct linkea_data *ldata,
3807 struct thandle *handle)
3809 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3810 child_name, ldata, handle, 1);
3813 static int mdd_update_linkea(const struct lu_env *env,
3814 struct mdd_object *mdd_pobj,
3815 struct mdd_object *mdd_sobj,
3816 struct mdd_object *mdd_tobj,
3817 const struct lu_name *child_name,
3818 struct linkea_data *ldata,
3819 struct thandle *handle)
3821 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3822 child_name, ldata, handle, 0);
3825 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3826 struct mdd_object *mdd_pobj,
3827 struct mdd_object *mdd_sobj,
3828 struct mdd_object *mdd_tobj,
3829 const struct lu_name *lname,
3831 struct lu_attr *parent_la,
3832 struct linkea_data *ldata,
3833 struct thandle *handle)
3835 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3836 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3839 /* Revert IMMUTABLE flag */
3840 la_flag->la_valid = LA_FLAGS;
3841 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3842 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3846 /* delete entry from source dir */
3847 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3851 if (ldata->ld_buf != NULL) {
3852 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3853 mdd_tobj, lname, ldata, handle);
3858 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3859 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3864 handle->th_complex = 1;
3865 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3867 LU_XATTR_REPLACE, handle);
3872 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3873 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3879 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3880 mdd_object_type(mdd_tobj),
3881 lname->ln_name, handle);
3885 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE);
3889 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3890 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3895 /* delete old object */
3896 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3900 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3901 /* delete old object */
3902 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3905 /* set nlink to 0 */
3906 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3911 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3915 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3919 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3924 static int mdd_migrate_update_name(const struct lu_env *env,
3925 struct mdd_object *mdd_pobj,
3926 struct mdd_object *mdd_sobj,
3927 struct mdd_object *mdd_tobj,
3928 const struct lu_name *lname,
3931 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3932 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3933 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3934 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3935 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3936 struct thandle *handle;
3937 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3938 const char *name = lname->ln_name;
3942 /* update time for parent */
3943 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3944 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3945 p_la->la_valid = LA_CTIME;
3947 rc = mdd_la_get(env, mdd_sobj, so_attr);
3951 ldata->ld_buf = NULL;
3952 rc = mdd_links_read(env, mdd_sobj, ldata);
3953 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
3956 handle = mdd_trans_create(env, mdd);
3958 RETURN(PTR_ERR(handle));
3960 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3961 lname, so_attr, p_la, ldata,
3964 /* If the migration can not be fit in one transaction, just
3965 * leave it in the original MDT */
3967 GOTO(stop_trans, rc = 0);
3969 GOTO(stop_trans, rc);
3972 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3973 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3974 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3975 PFID(mdd_object_fid(mdd_tobj)));
3977 rc = mdd_trans_start(env, mdd, handle);
3979 GOTO(stop_trans, rc);
3981 /* Revert IMMUTABLE flag */
3982 la_flag->la_valid = LA_FLAGS;
3983 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3984 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3986 GOTO(stop_trans, rc);
3988 /* Remove source name from source directory */
3989 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
3991 GOTO(stop_trans, rc);
3993 if (ldata->ld_buf != NULL) {
3994 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3995 lname, ldata, handle);
3997 GOTO(stop_trans, rc);
3999 /* linkea update might decrease the source object
4000 * nlink, let's get the attr again after ref_del */
4001 rc = mdd_la_get(env, mdd_sobj, so_attr);
4003 GOTO(stop_trans, rc);
4006 if (S_ISREG(so_attr->la_mode)) {
4007 if (so_attr->la_nlink == 1) {
4008 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4010 if (rc != 0 && rc != -ENODATA)
4011 GOTO(stop_trans, rc);
4013 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4015 LU_XATTR_REPLACE, handle);
4017 GOTO(stop_trans, rc);
4021 /* Insert new fid with target name into target dir */
4022 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4023 mdd_object_type(mdd_tobj), name, handle);
4025 GOTO(stop_trans, rc);
4027 linkea_add_buf(ldata, lname, mdd_object_fid(mdd_pobj));
4028 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
4031 GOTO(stop_trans, rc);
4033 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4035 mdd_sobj->mod_flags |= DEAD_OBJ;
4036 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4038 GOTO(out_unlock, rc);
4040 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4042 GOTO(out_unlock, rc);
4044 mdo_ref_del(env, mdd_sobj, handle);
4046 mdo_ref_del(env, mdd_sobj, handle);
4048 /* Get the attr again after ref_del */
4049 rc = mdd_la_get(env, mdd_sobj, so_attr);
4051 GOTO(out_unlock, rc);
4053 ma->ma_attr = *so_attr;
4054 ma->ma_valid |= MA_INODE;
4056 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4058 GOTO(out_unlock, rc);
4060 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4061 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4062 mdo2fid(mdd_pobj), lname, lname, handle);
4064 CWARN("%s: changelog for migrate %s "DFID
4065 "under "DFID" failed: rc = %d\n",
4066 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4067 PFID(mdd_object_fid(mdd_sobj)),
4068 PFID(mdd_object_fid(mdd_pobj)), rc);
4069 /* Sigh, there are no easy way to migrate back the object, so
4070 * let's reset the result to 0 for now XXX */
4074 mdd_write_unlock(env, mdd_sobj);
4077 mdd_trans_stop(env, mdd, rc, handle);
4082 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4083 const struct lu_fid *fid, __u32 *mdt_index)
4085 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4086 struct seq_server_site *ss;
4089 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4091 range->lsr_flags = LU_SEQ_RANGE_MDT;
4092 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4096 *mdt_index = range->lsr_index;
4101 * Check whether we should migrate the file/dir
4103 * < 0 permission check failed or other error.
4104 * = 0 the file can be migrated.
4105 * > 0 the file does not need to be migrated, mostly
4106 * for multiple link file
4108 static int mdd_migrate_sanity_check(const struct lu_env *env,
4109 struct mdd_object *pobj,
4110 const struct lu_attr *pattr,
4111 struct mdd_object *sobj,
4112 struct lu_attr *sattr)
4114 struct mdd_thread_info *info = mdd_env_info(env);
4115 struct linkea_data *ldata = &info->mti_link_data;
4116 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4118 struct lu_buf *mgr_buf;
4124 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4125 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4126 if (mgr_buf->lb_buf == NULL)
4129 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4131 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4133 /* If the object has migrateEA, it means IMMUTE flag
4134 * is being set by previous migration process, so it
4135 * needs to override the IMMUTE flag, otherwise the
4136 * following sanity check will fail */
4137 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4138 LMV_HASH_FLAG_MIGRATION) {
4139 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4141 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4142 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4143 mdd2obd_dev(mdd)->obd_name,
4144 PFID(mdd_object_fid(sobj)));
4148 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4149 sobj, sattr, NULL, NULL);
4153 /* Then it will check if the file should be migrated. If the file
4154 * has mulitple links, we only need migrate the file if all of its
4155 * entries has been migrated to the remote MDT */
4156 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4159 rc = mdd_links_read(env, sobj, ldata);
4161 /* For multiple links files, if there are no linkEA data at all,
4162 * means the file might be created before linkEA is enabled, and
4163 * all of its links should not be migrated yet, otherwise it
4164 * should have some linkEA there */
4165 if (rc == -ENOENT || rc == -ENODATA)
4170 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4171 /* If there are still links locally, then the file will not be
4173 LASSERT(ldata->ld_leh != NULL);
4174 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4175 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4176 struct lu_name lname;
4178 __u32 parent_mdt_index;
4180 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4182 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4185 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4189 /* Migrate the object only if none of its parents are on the
4191 if (parent_mdt_index != mdt_index)
4194 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4195 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4196 lname.ln_name, PFID(&fid));
4204 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4205 struct md_object *sobj, const struct lu_name *lname,
4206 struct md_object *tobj, struct md_attr *ma)
4208 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4209 struct mdd_device *mdd = mdo2mdd(pobj);
4210 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4211 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4212 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4213 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4217 /* If the file will being migrated, it will check whether
4218 * the file is being opened by someone else right now */
4219 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4220 if (mdd_sobj->mod_count > 0) {
4221 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
4222 mdd2obd_dev(mdd)->obd_name,
4223 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4224 mdd_sobj->mod_count, -EBUSY);
4225 mdd_read_unlock(env, mdd_sobj);
4226 GOTO(put, rc = -EBUSY);
4228 mdd_read_unlock(env, mdd_sobj);
4230 rc = mdd_la_get(env, mdd_sobj, so_attr);
4234 rc = mdd_la_get(env, mdd_pobj, pattr);
4238 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4245 /* Sigh, it is impossible to finish all of migration in a single
4246 * transaction, for example migrating big directory entries to the
4247 * new MDT, it needs insert all of name entries of children in the
4250 * So migration will be done in multiple steps and transactions.
4252 * 1. create an orphan object on the remote MDT in one transaction.
4253 * 2. migrate extend attributes to the new target file/directory.
4254 * 3. For directory, migrate the entries to the new MDT and update
4255 * linkEA of each children. Because we can not migrate all entries
4256 * in a single transaction, so the migrating directory will become
4257 * a striped directory during migration, so once the process is
4258 * interrupted, the directory is still accessible. (During lookup,
4259 * client will locate the name by searching both original and target
4261 * 4. Finally, update the name/FID to point to the new file/directory
4262 * in a separate transaction.
4265 /* step 1: Check whether the orphan object has been created, and create
4266 * orphan object on the remote MDT if needed */
4267 if (!mdd_object_exists(mdd_tobj)) {
4268 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4274 LASSERT(mdd_object_exists(mdd_tobj));
4275 /* step 2: migrate xattr */
4276 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4280 /* step 3: migrate name entries to the orphan object */
4281 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4282 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4285 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4286 OBD_FAIL_MDS_REINT_NET_REP)))
4289 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4292 LASSERT(mdd_object_exists(mdd_tobj));
4293 /* step 4: update name entry to the new object */
4294 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4302 const struct md_dir_operations mdd_dir_ops = {
4303 .mdo_is_subdir = mdd_is_subdir,
4304 .mdo_lookup = mdd_lookup,
4305 .mdo_create = mdd_create,
4306 .mdo_rename = mdd_rename,
4307 .mdo_link = mdd_link,
4308 .mdo_unlink = mdd_unlink,
4309 .mdo_create_data = mdd_create_data,
4310 .mdo_migrate = mdd_migrate,