4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_dir.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@intel.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <obd_class.h>
46 #include <obd_support.h>
47 #include <lustre_mds.h>
48 #include <lustre_fid.h>
50 #include "mdd_internal.h"
52 static const char dot[] = ".";
53 static const char dotdot[] = "..";
55 static struct lu_name lname_dotdot = {
61 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
63 if (!lu_name_is_valid(ln))
65 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
71 /* Get FID from name and parent */
73 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
74 const struct lu_attr *pattr, const struct lu_name *lname,
75 struct lu_fid* fid, int mask)
77 const char *name = lname->ln_name;
78 const struct dt_key *key = (const struct dt_key *)name;
79 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
80 struct mdd_device *m = mdo2mdd(pobj);
81 struct dt_object *dir = mdd_object_child(mdd_obj);
85 if (unlikely(mdd_is_dead_obj(mdd_obj)))
88 if (mdd_object_remote(mdd_obj)) {
89 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
90 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
91 } else if (!mdd_object_exists(mdd_obj)) {
95 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
100 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
101 dt_try_as_dir(env, dir)))
102 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
109 int mdd_lookup(const struct lu_env *env,
110 struct md_object *pobj, const struct lu_name *lname,
111 struct lu_fid *fid, struct md_op_spec *spec)
113 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
117 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
121 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
122 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
127 * Get parent FID of the directory
129 * Read parent FID from linkEA, if that fails, then do lookup
130 * dotdot to get the parent FID.
132 * \param[in] env execution environment
133 * \param[in] obj object from which to find the parent FID
134 * \param[in] attr attribute of the object
135 * \param[out] fid fid to get the parent FID
137 * \retval 0 if getting the parent FID succeeds.
138 * \retval negative errno if getting the parent FID fails.
140 static inline int mdd_parent_fid(const struct lu_env *env,
141 struct mdd_object *obj,
142 const struct lu_attr *attr,
145 struct mdd_thread_info *info = mdd_env_info(env);
146 struct linkea_data ldata = { NULL };
147 struct lu_buf *buf = &info->mti_link_buf;
148 struct lu_name lname;
153 LASSERT(S_ISDIR(mdd_object_type(obj)));
155 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
156 if (buf->lb_buf == NULL)
157 GOTO(lookup, rc = 0);
160 rc = mdd_links_read(env, obj, &ldata);
164 LASSERT(ldata.ld_leh != NULL);
165 /* Directory should only have 1 parent */
166 if (ldata.ld_leh->leh_reccount > 1)
169 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
171 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
172 if (likely(fid_is_sane(fid)))
175 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
180 * For root fid use special function, which does not compare version component
181 * of fid. Version component is different for root fids on all MDTs.
183 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
185 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
186 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
190 * return 1: if lf is the fid of the ancestor of p1;
193 * return -EREMOTE: if remote object is found, in this
194 * case fid of remote object is saved to @pf;
196 * otherwise: values < 0, errors.
198 static int mdd_is_parent(const struct lu_env *env,
199 struct mdd_device *mdd,
200 struct mdd_object *p1,
201 const struct lu_attr *attr,
202 const struct lu_fid *lf,
205 struct mdd_object *parent = NULL;
210 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
211 pfid = &mdd_env_info(env)->mti_fid;
213 /* Check for root first. */
214 if (mdd_is_root(mdd, mdo2fid(p1)))
218 /* this is done recursively */
219 rc = mdd_parent_fid(env, p1, attr, pfid);
222 if (mdd_is_root(mdd, pfid))
224 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
226 if (lu_fid_eq(pfid, lf))
229 mdd_object_put(env, parent);
231 parent = mdd_object_find(env, mdd, pfid);
233 GOTO(out, rc = PTR_ERR(parent));
238 if (parent && !IS_ERR(parent))
239 mdd_object_put(env, parent);
244 * No permission check is needed.
246 * returns 1: if fid is ancestor of @mo;
247 * returns 0: if fid is not an ancestor of @mo;
249 * returns EREMOTE if remote object is found, fid of remote object is saved to
252 * returns < 0: if error
254 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
255 const struct lu_fid *fid, struct lu_fid *sfid)
257 struct mdd_device *mdd = mdo2mdd(mo);
258 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
262 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
265 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
269 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
273 } else if (rc == 1) {
274 /* found @fid is parent */
282 * Check that @dir contains no entries except (possibly) dot and dotdot.
287 * -ENOTDIR not a directory object
288 * -ENOTEMPTY not empty
292 static int mdd_dir_is_empty(const struct lu_env *env,
293 struct mdd_object *dir)
296 struct dt_object *obj;
297 const struct dt_it_ops *iops;
301 obj = mdd_object_child(dir);
302 if (!dt_try_as_dir(env, obj))
305 iops = &obj->do_index_ops->dio_it;
306 it = iops->init(env, obj, LUDA_64BITHASH);
308 result = iops->get(env, it, (const struct dt_key *)"");
311 for (result = 0, i = 0; result == 0 && i < 3; ++i)
312 result = iops->next(env, it);
315 else if (result == 1)
317 } else if (result == 0)
319 * Huh? Index contains no zero key?
326 result = PTR_ERR(it);
331 * Determine if the target object can be hard linked, and right now it only
332 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
333 * directory nlink count might exceed the maximum link count(see
334 * osd_object_ref_add), so it only check nlink for non-directories.
336 * \param[in] env thread environment
337 * \param[in] obj object being linked to
338 * \param[in] la attributes of \a obj
340 * \retval 0 if \a obj can be hard linked
341 * \retval negative error if \a obj is a directory or has too
344 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
345 const struct lu_attr *la)
347 struct mdd_device *m = mdd_obj2mdd_dev(obj);
352 /* Subdir count limitation can be broken through
353 * (see osd_object_ref_add), so only check non-directory here. */
354 if (!S_ISDIR(la->la_mode) &&
355 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
362 * Check whether it may create the cobj under the pobj.
364 * \param[in] env execution environment
365 * \param[in] pobj the parent directory
366 * \param[in] pattr the attribute of the parent directory
367 * \param[in] cobj the child to be created
368 * \param[in] check_perm if check WRITE|EXEC permission for parent
370 * \retval = 0 create the child under this dir is allowed
371 * \retval negative errno create the child under this dir is
374 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
375 const struct lu_attr *pattr, struct mdd_object *cobj,
378 struct mdd_thread_info *info = mdd_env_info(env);
383 if (cobj && mdd_object_exists(cobj))
386 if (mdd_is_dead_obj(pobj))
389 /* If the parent is a sub-stripe, check whether it is dead */
390 xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key));
391 rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV);
392 if (unlikely(rc > 0)) {
393 struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf;
395 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE &&
396 le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD)
402 rc = mdd_permission_internal_locked(env, pobj, pattr,
403 MAY_WRITE | MAY_EXEC,
409 * Check whether can unlink from the pobj in the case of "cobj == NULL".
411 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
412 const struct lu_attr *pattr, const struct lu_attr *attr)
417 if (mdd_is_dead_obj(pobj))
420 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
423 rc = mdd_permission_internal_locked(env, pobj, pattr,
424 MAY_WRITE | MAY_EXEC,
429 if (pattr->la_flags & LUSTRE_APPEND_FL)
436 * pobj == NULL is remote ops case, under such case, pobj's
437 * VTX feature has been checked already, no need check again.
439 static inline int mdd_is_sticky(const struct lu_env *env,
440 struct mdd_object *pobj,
441 const struct lu_attr *pattr,
442 struct mdd_object *cobj,
443 const struct lu_attr *cattr)
445 struct lu_ucred *uc = lu_ucred_assert(env);
448 LASSERT(pattr != NULL);
449 if (!(pattr->la_mode & S_ISVTX) ||
450 (pattr->la_uid == uc->uc_fsuid))
454 LASSERT(cattr != NULL);
455 if (cattr->la_uid == uc->uc_fsuid)
458 return !md_capable(uc, CFS_CAP_FOWNER);
461 static int mdd_may_delete_entry(const struct lu_env *env,
462 struct mdd_object *pobj,
463 const struct lu_attr *pattr,
468 LASSERT(pobj != NULL);
469 if (!mdd_object_exists(pobj))
472 if (mdd_is_dead_obj(pobj))
477 rc = mdd_permission_internal_locked(env, pobj, pattr,
478 MAY_WRITE | MAY_EXEC,
484 if (pattr->la_flags & LUSTRE_APPEND_FL)
491 * Check whether it may delete the cobj from the pobj.
494 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
495 const struct lu_attr *tpattr, struct mdd_object *tobj,
496 const struct lu_attr *tattr, const struct lu_attr *cattr,
497 int check_perm, int check_empty)
503 LASSERT(tpattr != NULL);
504 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
512 if (!mdd_object_exists(tobj))
515 if (mdd_is_dead_obj(tobj))
518 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
521 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
524 /* additional check the rename case */
526 if (S_ISDIR(cattr->la_mode)) {
527 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
529 if (!S_ISDIR(tattr->la_mode))
532 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
534 } else if (S_ISDIR(tattr->la_mode))
538 if (S_ISDIR(tattr->la_mode) && check_empty)
539 rc = mdd_dir_is_empty(env, tobj);
545 * Check whether it can create the link file(linked to @src_obj) under
546 * the target directory(@tgt_obj), and src_obj has been locked by
549 * \param[in] env execution environment
550 * \param[in] tgt_obj the target directory
551 * \param[in] tattr attributes of target directory
552 * \param[in] lname the link name
553 * \param[in] src_obj source object for link
554 * \param[in] cattr attributes for source object
556 * \retval = 0 it is allowed to create the link file under tgt_obj
557 * \retval negative error not allowed to create the link file
559 static int mdd_link_sanity_check(const struct lu_env *env,
560 struct mdd_object *tgt_obj,
561 const struct lu_attr *tattr,
562 const struct lu_name *lname,
563 struct mdd_object *src_obj,
564 const struct lu_attr *cattr)
566 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
570 if (!mdd_object_exists(src_obj))
573 if (mdd_is_dead_obj(src_obj))
576 /* Local ops, no lookup before link, check filename length here. */
577 rc = mdd_name_check(m, lname);
581 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
584 if (S_ISDIR(mdd_object_type(src_obj)))
587 LASSERT(src_obj != tgt_obj);
588 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
592 rc = __mdd_may_link(env, src_obj, cattr);
597 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
598 const char *name, struct thandle *handle)
600 struct dt_object *next = mdd_object_child(pobj);
604 if (dt_try_as_dir(env, next))
605 rc = dt_delete(env, next, (struct dt_key *)name, handle);
612 static int __mdd_index_insert_only(const struct lu_env *env,
613 struct mdd_object *pobj,
614 const struct lu_fid *lf, __u32 type,
616 struct thandle *handle)
618 struct dt_object *next = mdd_object_child(pobj);
622 if (dt_try_as_dir(env, next)) {
623 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
624 struct lu_ucred *uc = lu_ucred_check(env);
628 rec->rec_type = type;
629 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
630 rc = dt_insert(env, next, (const struct dt_rec *)rec,
631 (const struct dt_key *)name, handle,
639 /* insert named index, add reference if isdir */
640 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
641 const struct lu_fid *lf, __u32 type,
642 const char *name, struct thandle *handle)
647 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
648 if (rc == 0 && S_ISDIR(type)) {
649 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
650 mdo_ref_add(env, pobj, handle);
651 mdd_write_unlock(env, pobj);
657 /* delete named index, drop reference if isdir */
658 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
659 const char *name, int is_dir,
660 struct thandle *handle)
665 rc = __mdd_index_delete_only(env, pobj, name, handle);
666 if (rc == 0 && is_dir) {
667 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
668 mdo_ref_del(env, pobj, handle);
669 mdd_write_unlock(env, pobj);
675 static int mdd_llog_record_calc_size(const struct lu_env *env,
676 const struct lu_name *tname,
677 const struct lu_name *sname)
679 const struct lu_ucred *uc = lu_ucred(env);
680 enum changelog_rec_flags crf = 0;
681 size_t hdr_size = sizeof(struct llog_changelog_rec) -
682 sizeof(struct changelog_rec);
687 if (uc != NULL && uc->uc_jobid[0] != '\0')
690 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
691 (tname != NULL ? tname->ln_namelen : 0) +
692 (sname != NULL ? 1 + sname->ln_namelen : 0));
695 int mdd_declare_changelog_store(const struct lu_env *env,
696 struct mdd_device *mdd,
697 const struct lu_name *tname,
698 const struct lu_name *sname,
699 struct thandle *handle)
701 struct obd_device *obd = mdd2obd_dev(mdd);
702 struct llog_ctxt *ctxt;
703 struct llog_changelog_rec *rec;
705 struct thandle *llog_th;
710 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
713 reclen = mdd_llog_record_calc_size(env, tname, sname);
714 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
715 if (buf->lb_buf == NULL)
719 rec->cr_hdr.lrh_len = reclen;
720 rec->cr_hdr.lrh_type = CHANGELOG_REC;
722 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
726 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
728 GOTO(out_put, rc = PTR_ERR(llog_th));
730 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
738 /** Add a changelog entry \a rec to the changelog llog
741 * \param handle - currently ignored since llogs start their own transaction;
742 * this will hopefully be fixed in llog rewrite
745 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
746 struct llog_changelog_rec *rec, struct thandle *th)
748 struct obd_device *obd = mdd2obd_dev(mdd);
749 struct llog_ctxt *ctxt;
750 struct thandle *llog_th;
753 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
754 changelog_rec_varsize(&rec->cr));
756 /* llog_lvfs_write_rec sets the llog tail len */
757 rec->cr_hdr.lrh_type = CHANGELOG_REC;
758 rec->cr.cr_time = cl_time();
760 spin_lock(&mdd->mdd_cl.mc_lock);
761 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
762 * but as long as the MDD transactions are ordered correctly for e.g.
763 * rename conflicts, I don't think this should matter. */
764 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
765 spin_unlock(&mdd->mdd_cl.mc_lock);
767 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
771 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
773 GOTO(out_put, rc = PTR_ERR(llog_th));
775 /* nested journal transaction */
776 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
785 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
786 const struct lu_fid *sfid,
787 const struct lu_fid *spfid,
788 const struct lu_name *sname)
790 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
791 size_t extsize = sname->ln_namelen + 1;
793 LASSERT(sfid != NULL);
794 LASSERT(spfid != NULL);
795 LASSERT(sname != NULL);
797 rnm->cr_sfid = *sfid;
798 rnm->cr_spfid = *spfid;
800 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
801 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
802 rec->cr_namelen += extsize;
805 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
807 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
809 if (jobid == NULL || jobid[0] == '\0')
812 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
815 /** Store a namespace change changelog record
816 * If this fails, we must fail the whole transaction; we don't
817 * want the change to commit without the log entry.
818 * \param target - mdd_object of change
819 * \param tpfid - target parent dir/object fid
820 * \param sfid - source object fid
821 * \param spfid - source parent fid
822 * \param tname - target name string
823 * \param sname - source name string
824 * \param handle - transaction handle
826 int mdd_changelog_ns_store(const struct lu_env *env,
827 struct mdd_device *mdd,
828 enum changelog_rec_type type,
829 enum changelog_rec_flags crf,
830 struct mdd_object *target,
831 const struct lu_fid *tpfid,
832 const struct lu_fid *sfid,
833 const struct lu_fid *spfid,
834 const struct lu_name *tname,
835 const struct lu_name *sname,
836 struct thandle *handle)
838 const struct lu_ucred *uc = lu_ucred(env);
839 struct llog_changelog_rec *rec;
846 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
849 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
852 LASSERT(tpfid != NULL);
853 LASSERT(tname != NULL);
854 LASSERT(handle != NULL);
856 reclen = mdd_llog_record_calc_size(env, tname, sname);
857 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
858 if (buf->lb_buf == NULL)
864 if (uc != NULL && uc->uc_jobid[0] != '\0')
872 rec->cr.cr_flags = crf;
873 rec->cr.cr_type = (__u32)type;
874 rec->cr.cr_pfid = *tpfid;
875 rec->cr.cr_namelen = tname->ln_namelen;
876 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
878 if (crf & CLF_RENAME)
879 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
882 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
884 if (likely(target != NULL)) {
885 rec->cr.cr_tfid = *mdo2fid(target);
886 target->mod_cltime = cfs_time_current_64();
888 fid_zero(&rec->cr.cr_tfid);
891 rc = mdd_changelog_store(env, mdd, rec, handle);
893 CERROR("%s: cannot store changelog record: type = %d, "
894 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
895 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
896 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
903 static int __mdd_links_add(const struct lu_env *env,
904 struct mdd_object *mdd_obj,
905 struct linkea_data *ldata,
906 const struct lu_name *lname,
907 const struct lu_fid *pfid,
908 int first, int check)
912 if (ldata->ld_leh == NULL) {
913 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
917 rc = linkea_data_new(ldata,
918 &mdd_env_info(env)->mti_link_buf);
925 rc = linkea_links_find(ldata, lname, pfid);
926 if (rc && rc != -ENOENT)
932 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
933 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
937 linkea_add_buf(ldata, lname, tfid);
940 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
941 linkea_add_buf(ldata, lname, pfid);
943 return linkea_add_buf(ldata, lname, pfid);
946 static int __mdd_links_del(const struct lu_env *env,
947 struct mdd_object *mdd_obj,
948 struct linkea_data *ldata,
949 const struct lu_name *lname,
950 const struct lu_fid *pfid)
954 if (ldata->ld_leh == NULL) {
955 rc = mdd_links_read(env, mdd_obj, ldata);
960 rc = linkea_links_find(ldata, lname, pfid);
964 linkea_del_buf(ldata, lname);
968 static int mdd_linkea_prepare(const struct lu_env *env,
969 struct mdd_object *mdd_obj,
970 const struct lu_fid *oldpfid,
971 const struct lu_name *oldlname,
972 const struct lu_fid *newpfid,
973 const struct lu_name *newlname,
974 int first, int check,
975 struct linkea_data *ldata)
981 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
984 LASSERT(oldpfid != NULL || newpfid != NULL);
986 if (mdd_obj->mod_flags & DEAD_OBJ) {
987 /* Prevent linkea to be updated which is NOT necessary. */
988 ldata->ld_reclen = 0;
989 /* No more links, don't bother */
993 if (oldpfid != NULL) {
994 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
997 (rc != -ENODATA && rc != -ENOENT))
999 /* No changes done. */
1004 /* If renaming, add the new record */
1005 if (newpfid != NULL) {
1006 /* even if the add fails, we still delete the out-of-date
1008 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1012 rc = rc != 0 ? rc : rc2;
1017 int mdd_links_rename(const struct lu_env *env,
1018 struct mdd_object *mdd_obj,
1019 const struct lu_fid *oldpfid,
1020 const struct lu_name *oldlname,
1021 const struct lu_fid *newpfid,
1022 const struct lu_name *newlname,
1023 struct thandle *handle,
1024 struct linkea_data *ldata,
1025 int first, int check)
1030 if (ldata == NULL) {
1031 ldata = &mdd_env_info(env)->mti_link_data;
1032 memset(ldata, 0, sizeof(*ldata));
1033 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1034 newpfid, newlname, first, check,
1040 if (ldata->ld_reclen != 0)
1041 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1046 if (rc == -EOVERFLOW || rc == -ENOSPC)
1048 if (oldpfid == NULL)
1049 CDEBUG(error ? D_ERROR : D_OTHER,
1050 "link_ea add '%.*s' failed %d "DFID"\n",
1051 newlname->ln_namelen, newlname->ln_name,
1052 rc, PFID(mdd_object_fid(mdd_obj)));
1053 else if (newpfid == NULL)
1054 CDEBUG(error ? D_ERROR : D_OTHER,
1055 "link_ea del '%.*s' failed %d "DFID"\n",
1056 oldlname->ln_namelen, oldlname->ln_name,
1057 rc, PFID(mdd_object_fid(mdd_obj)));
1059 CDEBUG(error ? D_ERROR : D_OTHER,
1060 "link_ea rename '%.*s'->'%.*s' failed %d "
1062 oldlname->ln_namelen, oldlname->ln_name,
1063 newlname->ln_namelen, newlname->ln_name,
1064 rc, PFID(mdd_object_fid(mdd_obj)));
1067 if (is_vmalloc_addr(ldata->ld_buf))
1068 /* if we vmalloced a large buffer drop it */
1069 lu_buf_free(ldata->ld_buf);
1074 static inline int mdd_links_add(const struct lu_env *env,
1075 struct mdd_object *mdd_obj,
1076 const struct lu_fid *pfid,
1077 const struct lu_name *lname,
1078 struct thandle *handle,
1079 struct linkea_data *ldata, int first)
1081 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1082 pfid, lname, handle, ldata, first, 0);
1085 static inline int mdd_links_del(const struct lu_env *env,
1086 struct mdd_object *mdd_obj,
1087 const struct lu_fid *pfid,
1088 const struct lu_name *lname,
1089 struct thandle *handle)
1091 return mdd_links_rename(env, mdd_obj, pfid, lname,
1092 NULL, NULL, handle, NULL, 0, 0);
1095 /** Read the link EA into a temp buffer.
1096 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1097 * A pointer to the buffer is stored in \a ldata::ld_buf.
1099 * \retval 0 or error
1101 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1102 struct linkea_data *ldata)
1106 if (!mdd_object_exists(mdd_obj))
1109 /* First try a small buf */
1110 LASSERT(env != NULL);
1111 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1113 if (ldata->ld_buf->lb_buf == NULL)
1116 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
1117 if (rc == -ERANGE) {
1118 /* Buf was too small, figure out what we need. */
1119 lu_buf_free(ldata->ld_buf);
1120 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1124 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1125 if (ldata->ld_buf->lb_buf == NULL)
1127 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1131 lu_buf_free(ldata->ld_buf);
1132 ldata->ld_buf = NULL;
1136 return linkea_init(ldata);
1139 /** Read the link EA into a temp buffer.
1140 * Uses the name_buf since it is generally large.
1141 * \retval IS_ERR err
1142 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1144 struct lu_buf *mdd_links_get(const struct lu_env *env,
1145 struct mdd_object *mdd_obj)
1147 struct linkea_data ldata = { NULL };
1150 rc = mdd_links_read(env, mdd_obj, &ldata);
1151 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1154 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1155 struct linkea_data *ldata, struct thandle *handle)
1157 const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1158 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_dead_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);
1415 obj->mod_flags |= DEAD_OBJ;
1417 if (!S_ISDIR(mdd_object_type(obj)))
1420 attr->la_valid = LA_FLAGS;
1421 attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
1424 rc = mdo_declare_attr_set(env, obj, attr, handle);
1426 rc = mdo_attr_set(env, obj, attr, handle);
1431 static int mdd_declare_finish_unlink(const struct lu_env *env,
1432 struct mdd_object *obj,
1433 struct thandle *handle)
1437 rc = mdd_mark_dead_object(env, obj, handle, true);
1441 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1445 rc = mdo_declare_destroy(env, 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 rc = mdd_mark_dead_object(env, obj, th, false);
1470 /* add new orphan and the object
1471 * will be deleted during mdd_close() */
1472 if (obj->mod_count) {
1473 rc = __mdd_orphan_add(env, obj, th);
1475 CDEBUG(D_HA, "Object "DFID" is inserted into "
1476 "orphan list, open count = %d\n",
1477 PFID(mdd_object_fid(obj)),
1480 CERROR("Object "DFID" fail to be an orphan, "
1481 "open count = %d, maybe cause failed "
1483 PFID(mdd_object_fid(obj)),
1486 rc = mdo_destroy(env, obj, th);
1488 } else if (!is_dir) {
1489 /* old files may not have link ea; ignore errors */
1490 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1498 * has mdd_write_lock on cobj already, but not on pobj yet
1500 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1501 const struct lu_attr *pattr,
1502 struct mdd_object *cobj,
1503 const struct lu_attr *cattr)
1508 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1513 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1514 struct mdd_object *p, struct mdd_object *c,
1515 const struct lu_name *name, struct md_attr *ma,
1516 struct thandle *handle, int no_name, int is_dir)
1518 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1521 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1522 if (likely(no_name == 0)) {
1523 rc = mdo_declare_index_delete(env, p, name->ln_name,
1530 rc = mdo_declare_ref_del(env, p, handle);
1536 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1537 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1538 la->la_valid = LA_CTIME | LA_MTIME;
1539 rc = mdo_declare_attr_set(env, p, la, handle);
1544 rc = mdo_declare_ref_del(env, c, handle);
1548 rc = mdo_declare_ref_del(env, c, handle);
1552 la->la_valid = LA_CTIME;
1553 rc = mdo_declare_attr_set(env, c, la, handle);
1557 rc = mdd_declare_finish_unlink(env, c, handle);
1561 /* FIXME: need changelog for remove entry */
1562 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1569 * test if a file has an HSM archive
1570 * if HSM attributes are not found in ma update them from
1573 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1574 struct mdd_object *obj,
1579 if (!(ma->ma_valid & MA_HSM)) {
1580 /* no HSM MD provided, read xattr */
1581 struct lu_buf *hsm_buf;
1582 const size_t buflen = sizeof(struct hsm_attrs);
1585 hsm_buf = mdd_buf_get(env, NULL, 0);
1586 lu_buf_alloc(hsm_buf, buflen);
1587 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1588 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1589 lu_buf_free(hsm_buf);
1593 ma->ma_valid = MA_HSM;
1595 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1601 * Delete name entry and the object.
1602 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1603 * does not exist for this object, and it could only happen during resending
1604 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1605 * is also needed in this case(needed by changelog), so we have to add another
1606 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1607 * the ENOENT failure should be able to be fixed by redo mechanism.
1609 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1610 struct md_object *cobj, const struct lu_name *lname,
1611 struct md_attr *ma, int no_name)
1613 const char *name = lname->ln_name;
1614 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1615 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1616 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1617 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1618 struct mdd_object *mdd_cobj = NULL;
1619 struct mdd_device *mdd = mdo2mdd(pobj);
1620 struct thandle *handle;
1624 /* cobj == NULL means only delete name entry */
1625 if (likely(cobj != NULL)) {
1626 mdd_cobj = md2mdd_obj(cobj);
1627 if (mdd_object_exists(mdd_cobj) == 0)
1631 rc = mdd_la_get(env, mdd_pobj, pattr);
1635 if (likely(mdd_cobj != NULL)) {
1637 rc = mdd_la_get(env, mdd_cobj, cattr);
1641 is_dir = S_ISDIR(cattr->la_mode);
1644 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1648 handle = mdd_trans_create(env, mdd);
1650 RETURN(PTR_ERR(handle));
1652 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1653 lname, ma, handle, no_name, is_dir);
1657 rc = mdd_trans_start(env, mdd, handle);
1661 if (likely(mdd_cobj != NULL))
1662 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1664 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1665 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1670 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1671 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1672 GOTO(cleanup, rc = 0);
1674 if (likely(mdd_cobj != NULL)) {
1675 rc = mdo_ref_del(env, mdd_cobj, handle);
1677 __mdd_index_insert_only(env, mdd_pobj,
1679 mdd_object_type(mdd_cobj),
1686 mdo_ref_del(env, mdd_cobj, handle);
1688 /* fetch updated nlink */
1689 rc = mdd_la_get(env, mdd_cobj, cattr);
1694 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1695 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1697 la->la_valid = LA_CTIME | LA_MTIME;
1698 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1702 /* Enough for only unlink the entry */
1703 if (unlikely(mdd_cobj == NULL))
1706 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1707 /* update ctime of an unlinked file only if it is still
1708 * opened or a link still exists */
1709 la->la_valid = LA_CTIME;
1710 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1715 /* XXX: this transfer to ma will be removed with LOD/OSP */
1716 ma->ma_attr = *cattr;
1717 ma->ma_valid |= MA_INODE;
1718 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1720 /* fetch updated nlink */
1722 rc = mdd_la_get(env, mdd_cobj, cattr);
1724 /* if object is removed then we can't get its attrs, use last get */
1725 if (cattr->la_nlink == 0) {
1726 ma->ma_attr = *cattr;
1727 ma->ma_valid |= MA_INODE;
1731 if (likely(mdd_cobj != NULL))
1732 mdd_write_unlock(env, mdd_cobj);
1737 if (cattr->la_nlink == 0) {
1738 cl_flags |= CLF_UNLINK_LAST;
1739 /* search for an existing archive */
1740 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1741 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1744 rc = mdd_changelog_ns_store(env, mdd,
1745 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1746 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1751 mdd_trans_stop(env, mdd, rc, handle);
1757 * The permission has been checked when obj created, no need check again.
1759 static int mdd_cd_sanity_check(const struct lu_env *env,
1760 struct mdd_object *obj)
1765 if (!obj || mdd_is_dead_obj(obj))
1771 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1772 struct md_object *cobj, const struct md_op_spec *spec,
1775 struct mdd_device *mdd = mdo2mdd(cobj);
1776 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1777 struct mdd_object *son = md2mdd_obj(cobj);
1778 struct thandle *handle;
1779 const struct lu_buf *buf;
1780 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1781 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1785 rc = mdd_cd_sanity_check(env, son);
1789 if (!md_should_create(spec->sp_cr_flags))
1793 * there are following use cases for this function:
1794 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1795 * striping can be specified or not
1798 rc = mdd_la_get(env, son, attr);
1802 /* calling ->ah_make_hint() is used to transfer information from parent */
1803 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1805 handle = mdd_trans_create(env, mdd);
1807 GOTO(out_free, rc = PTR_ERR(handle));
1810 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1811 * Should this be fixed?
1813 CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
1814 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1815 spec->sp_cr_flags, spec->no_create);
1817 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1818 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1819 spec->u.sp_ea.eadatalen);
1824 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1825 XATTR_NAME_LOV, 0, handle);
1829 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1833 rc = mdd_trans_start(env, mdd, handle);
1837 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1843 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1846 mdd_trans_stop(env, mdd, rc, handle);
1851 static int mdd_declare_object_initialize(const struct lu_env *env,
1852 struct mdd_object *parent,
1853 struct mdd_object *child,
1854 struct lu_attr *attr,
1855 struct thandle *handle)
1861 * inode mode has been set in creation time, and it's based on umask,
1862 * la_mode and acl, don't set here again! (which will go wrong
1863 * because below function doesn't consider umask).
1864 * I'd suggest set all object attributes in creation time, see above.
1866 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1867 attr->la_valid &= ~(LA_MODE | LA_TYPE);
1868 rc = mdo_declare_attr_set(env, child, attr, handle);
1869 attr->la_valid |= LA_MODE | LA_TYPE;
1870 if (rc != 0 || !S_ISDIR(attr->la_mode))
1873 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1878 rc = mdo_declare_ref_add(env, child, handle);
1882 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1888 static int mdd_object_initialize(const struct lu_env *env,
1889 const struct lu_fid *pfid,
1890 struct mdd_object *child,
1891 struct lu_attr *attr, struct thandle *handle,
1892 const struct md_op_spec *spec)
1897 if (S_ISDIR(attr->la_mode)) {
1898 /* Add "." and ".." for newly created dir */
1899 mdo_ref_add(env, child, handle);
1900 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1901 S_IFDIR, dot, handle);
1903 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1906 mdo_ref_del(env, child, handle);
1913 * This function checks whether it can create a file/dir under the
1914 * directory(@pobj). The directory(@pobj) is not being locked by
1917 * \param[in] env execution environment
1918 * \param[in] pobj the directory to create files
1919 * \param[in] pattr the attributes of the directory
1920 * \param[in] lname the name of the created file/dir
1921 * \param[in] cattr the attributes of the file/dir
1922 * \param[in] spec create specification
1924 * \retval = 0 it is allowed to create file/dir under
1926 * \retval negative error not allowed to create file/dir
1927 * under the directory
1929 static int mdd_create_sanity_check(const struct lu_env *env,
1930 struct md_object *pobj,
1931 const struct lu_attr *pattr,
1932 const struct lu_name *lname,
1933 struct lu_attr *cattr,
1934 struct md_op_spec *spec)
1936 struct mdd_thread_info *info = mdd_env_info(env);
1937 struct lu_fid *fid = &info->mti_fid;
1938 struct mdd_object *obj = md2mdd_obj(pobj);
1939 struct mdd_device *m = mdo2mdd(pobj);
1940 bool check_perm = true;
1945 if (mdd_is_dead_obj(obj))
1949 * In some cases this lookup is not needed - we know before if name
1950 * exists or not because MDT performs lookup for it.
1951 * name length check is done in lookup.
1953 if (spec->sp_cr_lookup) {
1955 * Check if the name already exist, though it will be checked in
1956 * _index_insert also, for avoiding rolling back if exists
1959 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1960 MAY_WRITE | MAY_EXEC);
1962 RETURN(rc ? : -EEXIST);
1964 /* Permission is already being checked in mdd_lookup */
1968 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1973 if (pattr->la_mode & S_ISGID) {
1974 cattr->la_gid = pattr->la_gid;
1975 if (S_ISDIR(cattr->la_mode)) {
1976 cattr->la_mode |= S_ISGID;
1977 cattr->la_valid |= LA_MODE;
1981 rc = mdd_name_check(m, lname);
1985 switch (cattr->la_mode & S_IFMT) {
1987 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1989 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1990 RETURN(-ENAMETOOLONG);
2009 static int mdd_declare_object_create(const struct lu_env *env,
2010 struct mdd_device *mdd,
2011 struct mdd_object *p, struct mdd_object *c,
2012 struct lu_attr *attr,
2013 struct thandle *handle,
2014 const struct md_op_spec *spec,
2015 struct lu_buf *def_acl_buf,
2016 struct lu_buf *acl_buf,
2017 struct dt_allocation_hint *hint)
2021 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
2026 #ifdef CONFIG_FS_POSIX_ACL
2027 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2028 /* if dir, then can inherit default ACl */
2029 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2030 XATTR_NAME_ACL_DEFAULT,
2036 if (acl_buf->lb_len > 0) {
2037 rc = mdo_declare_attr_set(env, c, attr, handle);
2041 rc = mdo_declare_xattr_set(env, c, acl_buf,
2042 XATTR_NAME_ACL_ACCESS, 0, handle);
2047 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2051 /* replay case, create LOV EA from client data */
2052 if (spec->no_create ||
2053 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2054 const struct lu_buf *buf;
2056 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2057 spec->u.sp_ea.eadatalen);
2058 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2064 if (S_ISLNK(attr->la_mode)) {
2065 const char *target_name = spec->u.sp_symname;
2066 int sym_len = strlen(target_name);
2067 const struct lu_buf *buf;
2069 buf = mdd_buf_get_const(env, target_name, sym_len);
2070 rc = dt_declare_record_write(env, mdd_object_child(c),
2079 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2080 struct mdd_object *p, struct mdd_object *c,
2081 const struct lu_name *name,
2082 struct lu_attr *attr,
2083 struct thandle *handle,
2084 const struct md_op_spec *spec,
2085 struct linkea_data *ldata,
2086 struct lu_buf *def_acl_buf,
2087 struct lu_buf *acl_buf,
2088 struct dt_allocation_hint *hint)
2092 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2093 def_acl_buf, acl_buf, hint);
2097 if (S_ISDIR(attr->la_mode)) {
2098 rc = mdo_declare_ref_add(env, p, handle);
2103 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2104 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2108 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2110 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2111 name->ln_name, handle);
2115 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2120 la->la_valid = LA_CTIME | LA_MTIME;
2121 rc = mdo_declare_attr_set(env, p, la, handle);
2125 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2133 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2134 struct lu_attr *la, struct lu_buf *def_acl_buf,
2135 struct lu_buf *acl_buf)
2140 if (S_ISLNK(la->la_mode)) {
2141 acl_buf->lb_len = 0;
2142 def_acl_buf->lb_len = 0;
2146 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2147 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2148 XATTR_NAME_ACL_DEFAULT);
2149 mdd_read_unlock(env, pobj);
2151 /* If there are default ACL, fix mode/ACL by default ACL */
2152 def_acl_buf->lb_len = rc;
2153 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2154 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2155 acl_buf->lb_len = rc;
2156 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2159 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2160 /* If there are no default ACL, fix mode by mask */
2161 struct lu_ucred *uc = lu_ucred(env);
2163 /* The create triggered by MDT internal events, such as
2164 * LFSCK reset, will not contain valid "uc". */
2165 if (unlikely(uc != NULL))
2166 la->la_mode &= ~uc->uc_umask;
2168 acl_buf->lb_len = 0;
2169 def_acl_buf->lb_len = 0;
2176 * Create a metadata object and initialize it, set acl, xattr.
2178 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2179 struct mdd_object *son, struct lu_attr *attr,
2180 struct md_op_spec *spec, struct lu_buf *acl_buf,
2181 struct lu_buf *def_acl_buf,
2182 struct dt_allocation_hint *hint,
2183 struct thandle *handle)
2187 mdd_write_lock(env, son, MOR_TGT_CHILD);
2188 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2193 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2194 * created in declare phase, they also needs to be added to master
2195 * object as sub-directory entry. So it has to initialize the master
2196 * object, then set dir striped EA.(in mdo_xattr_set) */
2197 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2200 GOTO(err_destroy, rc);
2203 * in case of replay we just set LOVEA provided by the client
2204 * XXX: I think it would be interesting to try "old" way where
2205 * MDT calls this xattr_set(LOV) in a different transaction.
2206 * probably this way we code can be made better.
2209 /* During creation, there are only a few cases we need do xattr_set to
2211 * 1. regular file: see comments above.
2212 * 2. dir: inherit default striping or pool settings from parent.
2213 * 3. create striped directory with provided stripeEA.
2214 * 4. create striped directory because inherit default layout from the
2217 if (spec->no_create ||
2218 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2219 S_ISDIR(attr->la_mode)) {
2220 const struct lu_buf *buf;
2222 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2223 spec->u.sp_ea.eadatalen);
2224 rc = mdo_xattr_set(env, son, buf,
2225 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2229 GOTO(err_destroy, rc);
2232 #ifdef CONFIG_FS_POSIX_ACL
2233 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2234 S_ISDIR(attr->la_mode)) {
2235 /* set default acl */
2236 rc = mdo_xattr_set(env, son, def_acl_buf,
2237 XATTR_NAME_ACL_DEFAULT, 0,
2240 GOTO(err_destroy, rc);
2242 /* set its own acl */
2243 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2244 rc = mdo_xattr_set(env, son, acl_buf,
2245 XATTR_NAME_ACL_ACCESS,
2248 GOTO(err_destroy, rc);
2252 if (S_ISLNK(attr->la_mode)) {
2253 struct lu_ucred *uc = lu_ucred_assert(env);
2254 struct dt_object *dt = mdd_object_child(son);
2255 const char *target_name = spec->u.sp_symname;
2256 int sym_len = strlen(target_name);
2257 const struct lu_buf *buf;
2260 buf = mdd_buf_get_const(env, target_name, sym_len);
2261 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2263 CFS_CAP_SYS_RESOURCE_MASK);
2268 GOTO(err_initlized, rc = -EFAULT);
2272 if (unlikely(rc != 0)) {
2274 if (S_ISDIR(attr->la_mode)) {
2275 /* Drop the reference, no need to delete "."/"..",
2276 * because the object to be destroied directly. */
2277 rc2 = mdo_ref_del(env, son, handle);
2281 rc2 = mdo_ref_del(env, son, handle);
2285 mdo_destroy(env, son, handle);
2288 mdd_write_unlock(env, son);
2293 * Create object and insert it into namespace.
2295 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2296 const struct lu_name *lname, struct md_object *child,
2297 struct md_op_spec *spec, struct md_attr* ma)
2299 struct mdd_thread_info *info = mdd_env_info(env);
2300 struct lu_attr *la = &info->mti_la_for_fix;
2301 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2302 struct mdd_object *son = md2mdd_obj(child);
2303 struct mdd_device *mdd = mdo2mdd(pobj);
2304 struct lu_attr *attr = &ma->ma_attr;
2305 struct thandle *handle;
2306 struct lu_attr *pattr = &info->mti_pattr;
2307 struct lu_buf acl_buf;
2308 struct lu_buf def_acl_buf;
2309 struct linkea_data *ldata = &info->mti_link_data;
2310 const char *name = lname->ln_name;
2311 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2317 * Two operations have to be performed:
2319 * - an allocation of a new object (->do_create()), and
2321 * - an insertion into a parent index (->dio_insert()).
2323 * Due to locking, operation order is not important, when both are
2324 * successful, *but* error handling cases are quite different:
2326 * - if insertion is done first, and following object creation fails,
2327 * insertion has to be rolled back, but this operation might fail
2328 * also leaving us with dangling index entry.
2330 * - if creation is done first, is has to be undone if insertion
2331 * fails, leaving us with leaked space, which is neither good, nor
2334 * It seems that creation-first is simplest solution, but it is
2335 * sub-optimal in the frequent
2340 * case, because second mkdir is bound to create object, only to
2341 * destroy it immediately.
2343 * To avoid this follow local file systems that do double lookup:
2345 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2347 * 1. create (mdd_object_create_internal())
2349 * 2. insert (__mdd_index_insert(), lookup again)
2352 rc = mdd_la_get(env, mdd_pobj, pattr);
2356 /* Sanity checks before big job. */
2357 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2361 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2362 GOTO(out_free, rc = -EINPROGRESS);
2364 handle = mdd_trans_create(env, mdd);
2366 GOTO(out_free, rc = PTR_ERR(handle));
2368 acl_buf.lb_buf = info->mti_xattr_buf;
2369 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2370 def_acl_buf.lb_buf = info->mti_key;
2371 def_acl_buf.lb_len = sizeof(info->mti_key);
2372 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2376 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2378 memset(ldata, 0, sizeof(*ldata));
2379 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2380 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2383 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2384 &tfid, lname, 1, 0, ldata);
2386 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2387 mdd_object_fid(mdd_pobj),
2388 lname, 1, 0, ldata);
2391 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2392 handle, spec, ldata, &def_acl_buf, &acl_buf,
2397 rc = mdd_trans_start(env, mdd, handle);
2401 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2402 &def_acl_buf, hint, handle);
2406 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2407 mdd_write_lock(env, son, MOR_TGT_CHILD);
2408 rc = __mdd_orphan_add(env, son, handle);
2409 GOTO(out_volatile, rc);
2411 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2412 attr->la_mode, name, handle);
2414 GOTO(err_created, rc);
2416 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2419 /* update parent directory mtime/ctime */
2421 la->la_valid = LA_CTIME | LA_MTIME;
2422 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2424 GOTO(err_insert, rc);
2432 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2433 rc2 = __mdd_orphan_del(env, son, handle);
2435 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2436 S_ISDIR(attr->la_mode),
2442 mdd_write_lock(env, son, MOR_TGT_CHILD);
2443 if (S_ISDIR(attr->la_mode)) {
2444 /* Drop the reference, no need to delete "."/"..",
2445 * because the object is to be destroyed directly. */
2446 rc2 = mdo_ref_del(env, son, handle);
2448 mdd_write_unlock(env, son);
2453 /* For volatile files drop one link immediately, since there is
2454 * no filename in the namespace, and save any error returned. */
2455 rc2 = mdo_ref_del(env, son, handle);
2457 mdd_write_unlock(env, son);
2458 if (unlikely(rc == 0))
2463 /* Don't destroy the volatile object on success */
2464 if (likely(rc != 0))
2465 mdo_destroy(env, son, handle);
2466 mdd_write_unlock(env, son);
2469 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2470 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2471 rc = mdd_changelog_ns_store(env, mdd,
2472 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2473 S_ISREG(attr->la_mode) ? CL_CREATE :
2474 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2475 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2478 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2482 if (is_vmalloc_addr(ldata->ld_buf))
2483 /* if we vmalloced a large buffer drop it */
2484 lu_buf_free(ldata->ld_buf);
2486 /* The child object shouldn't be cached anymore */
2488 set_bit(LU_OBJECT_HEARD_BANSHEE,
2489 &child->mo_lu.lo_header->loh_flags);
2494 * Get locks on parents in proper order
2495 * RETURN: < 0 - error, rename_order if successful
2503 static int mdd_rename_order(const struct lu_env *env,
2504 struct mdd_device *mdd,
2505 struct mdd_object *src_pobj,
2506 const struct lu_attr *pattr,
2507 struct mdd_object *tgt_pobj)
2509 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2513 if (src_pobj == tgt_pobj)
2514 RETURN(MDD_RN_SAME);
2516 /* compared the parent child relationship of src_p&tgt_p */
2517 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2519 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2522 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2536 /* has not mdd_write{read}_lock on any obj yet. */
2537 static int mdd_rename_sanity_check(const struct lu_env *env,
2538 struct mdd_object *src_pobj,
2539 const struct lu_attr *pattr,
2540 struct mdd_object *tgt_pobj,
2541 const struct lu_attr *tpattr,
2542 struct mdd_object *sobj,
2543 const struct lu_attr *cattr,
2544 struct mdd_object *tobj,
2545 const struct lu_attr *tattr)
2550 /* XXX: when get here, sobj must NOT be NULL,
2551 * the other case has been processed in cld_rename
2552 * before mdd_rename and enable MDS_PERM_BYPASS. */
2555 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2559 /* XXX: when get here, "tobj == NULL" means tobj must
2560 * NOT exist (neither on remote MDS, such case has been
2561 * processed in cld_rename before mdd_rename and enable
2563 * So check may_create, but not check may_unlink. */
2565 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2566 (src_pobj != tgt_pobj));
2568 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2569 (src_pobj != tgt_pobj), 1);
2571 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2572 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2577 static int mdd_declare_rename(const struct lu_env *env,
2578 struct mdd_device *mdd,
2579 struct mdd_object *mdd_spobj,
2580 struct mdd_object *mdd_tpobj,
2581 struct mdd_object *mdd_sobj,
2582 struct mdd_object *mdd_tobj,
2583 const struct lu_name *tname,
2584 const struct lu_name *sname,
2586 struct linkea_data *ldata,
2587 struct thandle *handle)
2589 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2592 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2593 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2599 /* name from source dir */
2600 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2604 /* .. from source child */
2605 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2606 /* source child can be directory,
2607 * counted by source dir's nlink */
2608 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2611 if (mdd_spobj != mdd_tpobj) {
2612 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2617 rc = mdo_declare_index_insert(env, mdd_sobj,
2619 S_IFDIR, dotdot, handle);
2624 /* new target child can be directory,
2625 * counted by target dir's nlink */
2626 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2631 la->la_valid = LA_CTIME | LA_MTIME;
2632 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2636 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2640 la->la_valid = LA_CTIME;
2641 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2645 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
2646 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2651 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2652 mdd_object_type(mdd_sobj),
2653 tname->ln_name, handle);
2657 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2658 /* delete target child in target parent directory */
2659 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2664 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2668 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2669 /* target child can be directory,
2670 * delete "." reference in target child directory */
2671 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2675 /* delete ".." reference in target parent directory */
2676 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2681 la->la_valid = LA_CTIME;
2682 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2686 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2691 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2698 /* src object can be remote that is why we use only fid and type of object */
2699 static int mdd_rename(const struct lu_env *env,
2700 struct md_object *src_pobj, struct md_object *tgt_pobj,
2701 const struct lu_fid *lf, const struct lu_name *lsname,
2702 struct md_object *tobj, const struct lu_name *ltname,
2705 const char *sname = lsname->ln_name;
2706 const char *tname = ltname->ln_name;
2707 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2708 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2709 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2710 struct mdd_device *mdd = mdo2mdd(src_pobj);
2711 struct mdd_object *mdd_sobj = NULL; /* source object */
2712 struct mdd_object *mdd_tobj = NULL;
2713 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2714 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2715 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2716 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2717 struct thandle *handle;
2718 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2719 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2720 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2723 bool tobj_locked = 0;
2724 unsigned cl_flags = 0;
2729 mdd_tobj = md2mdd_obj(tobj);
2731 mdd_sobj = mdd_object_find(env, mdd, lf);
2732 if (IS_ERR(mdd_sobj))
2733 RETURN(PTR_ERR(mdd_sobj));
2735 rc = mdd_la_get(env, mdd_sobj, cattr);
2737 GOTO(out_pending, rc);
2739 rc = mdd_la_get(env, mdd_spobj, pattr);
2741 GOTO(out_pending, rc);
2744 rc = mdd_la_get(env, mdd_tobj, tattr);
2746 GOTO(out_pending, rc);
2749 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2751 GOTO(out_pending, rc);
2753 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2754 mdd_sobj, cattr, mdd_tobj, tattr);
2756 GOTO(out_pending, rc);
2758 rc = mdd_name_check(mdd, ltname);
2760 GOTO(out_pending, rc);
2762 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2763 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2765 GOTO(out_pending, rc);
2767 handle = mdd_trans_create(env, mdd);
2769 GOTO(out_pending, rc = PTR_ERR(handle));
2771 memset(ldata, 0, sizeof(*ldata));
2772 mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname,
2773 mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata);
2774 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2775 mdd_tobj, lsname, ltname, ma, ldata, handle);
2779 rc = mdd_trans_start(env, mdd, handle);
2783 is_dir = S_ISDIR(cattr->la_mode);
2785 /* Remove source name from source directory */
2786 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2790 /* "mv dir1 dir2" needs "dir1/.." link update */
2791 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2792 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2794 GOTO(fixup_spobj2, rc);
2796 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2799 GOTO(fixup_spobj, rc);
2802 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2803 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2805 /* tname might been renamed to something else */
2806 GOTO(fixup_spobj, rc);
2809 /* Insert new fid with target name into target dir */
2810 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2813 GOTO(fixup_tpobj, rc);
2815 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2816 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2818 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2819 la->la_valid = LA_CTIME;
2820 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2822 GOTO(fixup_tpobj, rc);
2824 /* Update the linkEA for the source object */
2825 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2826 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2827 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2830 /* Old files might not have EA entry */
2831 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2832 lsname, handle, NULL, 0);
2833 mdd_write_unlock(env, mdd_sobj);
2834 /* We don't fail the transaction if the link ea can't be
2835 updated -- fid2path will use alternate lookup method. */
2838 /* Remove old target object
2839 * For tobj is remote case cmm layer has processed
2840 * and set tobj to NULL then. So when tobj is NOT NULL,
2841 * it must be local one.
2843 if (tobj && mdd_object_exists(mdd_tobj)) {
2844 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2846 if (mdd_is_dead_obj(mdd_tobj)) {
2847 /* shld not be dead, something is wrong */
2848 CERROR("tobj is dead, something is wrong\n");
2852 mdo_ref_del(env, mdd_tobj, handle);
2854 /* Remove dot reference. */
2855 if (S_ISDIR(tattr->la_mode))
2856 mdo_ref_del(env, mdd_tobj, handle);
2859 /* fetch updated nlink */
2860 rc = mdd_la_get(env, mdd_tobj, tattr);
2862 CERROR("%s: Failed to get nlink for tobj "
2864 mdd2obd_dev(mdd)->obd_name,
2865 PFID(tpobj_fid), rc);
2866 GOTO(fixup_tpobj, rc);
2869 la->la_valid = LA_CTIME;
2870 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2872 CERROR("%s: Failed to set ctime for tobj "
2874 mdd2obd_dev(mdd)->obd_name,
2875 PFID(tpobj_fid), rc);
2876 GOTO(fixup_tpobj, rc);
2879 /* XXX: this transfer to ma will be removed with LOD/OSP */
2880 ma->ma_attr = *tattr;
2881 ma->ma_valid |= MA_INODE;
2882 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2885 CERROR("%s: Failed to unlink tobj "
2887 mdd2obd_dev(mdd)->obd_name,
2888 PFID(tpobj_fid), rc);
2889 GOTO(fixup_tpobj, rc);
2892 /* fetch updated nlink */
2893 rc = mdd_la_get(env, mdd_tobj, tattr);
2895 CERROR("%s: Failed to get nlink for tobj "
2897 mdd2obd_dev(mdd)->obd_name,
2898 PFID(tpobj_fid), rc);
2899 GOTO(fixup_tpobj, rc);
2901 /* XXX: this transfer to ma will be removed with LOD/OSP */
2902 ma->ma_attr = *tattr;
2903 ma->ma_valid |= MA_INODE;
2905 if (tattr->la_nlink == 0) {
2906 cl_flags |= CLF_RENAME_LAST;
2907 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2908 cl_flags |= CLF_RENAME_LAST_EXISTS;
2912 la->la_valid = LA_CTIME | LA_MTIME;
2913 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2915 GOTO(fixup_tpobj, rc);
2917 if (mdd_spobj != mdd_tpobj) {
2918 la->la_valid = LA_CTIME | LA_MTIME;
2919 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2921 GOTO(fixup_tpobj, rc);
2928 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2930 CWARN("tp obj fix error %d\n",rc2);
2932 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2933 !mdd_is_dead_obj(mdd_tobj)) {
2935 mdo_ref_add(env, mdd_tobj, handle);
2937 mdo_ref_add(env, mdd_tobj, handle);
2940 rc2 = __mdd_index_insert(env, mdd_tpobj,
2942 mdd_object_type(mdd_tobj),
2945 CWARN("tp obj fix error: rc = %d\n", rc2);
2950 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
2951 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2953 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
2954 mdd2obd_dev(mdd)->obd_name, rc2);
2957 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
2960 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
2961 mdd2obd_dev(mdd)->obd_name, rc2);
2966 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
2967 mdd_object_type(mdd_sobj), sname,
2970 CWARN("sp obj fix error: rc = %d\n", rc2);
2975 mdd_write_unlock(env, mdd_tobj);
2978 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
2979 mdd_tobj, tpobj_fid, lf, spobj_fid,
2980 ltname, lsname, handle);
2983 mdd_trans_stop(env, mdd, rc, handle);
2986 mdd_object_put(env, mdd_sobj);
2991 * During migration once the parent FID has been changed,
2992 * we need update the parent FID in linkea.
2994 static int mdd_linkea_update_child_internal(const struct lu_env *env,
2995 struct mdd_object *parent,
2996 struct mdd_object *child,
2997 const char *name, int namelen,
2998 struct thandle *handle,
3001 struct mdd_thread_info *info = mdd_env_info(env);
3002 struct linkea_data ldata = { NULL };
3003 struct lu_buf *buf = &info->mti_link_buf;
3009 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3010 if (buf->lb_buf == NULL)
3014 rc = mdd_links_read(env, child, &ldata);
3016 if (rc == -ENOENT || rc == -ENODATA)
3021 LASSERT(ldata.ld_leh != NULL);
3022 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3023 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3024 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3025 struct lu_name lname;
3028 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3031 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3032 lu_fid_eq(&fid, mdd_object_fid(parent))) {
3033 ldata.ld_lee = (struct link_ea_entry *)
3034 ((char *)ldata.ld_lee +
3039 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3040 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3041 lname.ln_namelen, lname.ln_name,
3042 PFID(mdd_object_fid(parent)));
3043 /* update to the new parent fid */
3044 linkea_entry_pack(ldata.ld_lee, &lname,
3045 mdd_object_fid(parent));
3047 rc = mdd_declare_links_add(env, child, handle, &ldata,
3050 rc = mdd_links_write(env, child, &ldata, handle);
3056 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3057 struct mdd_object *parent,
3058 struct mdd_object *child,
3059 const char *name, int namelen,
3060 struct thandle *handle)
3062 return mdd_linkea_update_child_internal(env, parent, child, name,
3063 namelen, handle, true);
3066 static int mdd_linkea_update_child(const struct lu_env *env,
3067 struct mdd_object *parent,
3068 struct mdd_object *child,
3069 const char *name, int namelen,
3070 struct thandle *handle)
3072 return mdd_linkea_update_child_internal(env, parent, child, name,
3073 namelen, handle, false);
3076 static int mdd_update_linkea_internal(const struct lu_env *env,
3077 struct mdd_object *mdd_pobj,
3078 struct mdd_object *mdd_sobj,
3079 struct mdd_object *mdd_tobj,
3080 const struct lu_name *child_name,
3081 struct thandle *handle,
3084 struct mdd_thread_info *info = mdd_env_info(env);
3085 struct linkea_data *ldata = &info->mti_link_data;
3090 rc = mdd_links_read(env, mdd_sobj, ldata);
3092 if (rc == -ENOENT || rc == -ENODATA)
3097 /* If it is mulitple links file, we need update the name entry for
3099 LASSERT(ldata->ld_leh != NULL);
3100 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3101 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3102 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3103 struct mdd_object *pobj;
3104 struct lu_name lname;
3107 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3109 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3111 pobj = mdd_object_find(env, mdd, &fid);
3113 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3114 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3119 if (!mdd_object_exists(pobj)) {
3120 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3121 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3125 if (pobj == mdd_pobj &&
3126 lname.ln_namelen == child_name->ln_namelen &&
3127 strncmp(lname.ln_name, child_name->ln_name,
3128 lname.ln_namelen) == 0) {
3129 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3130 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3135 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3136 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3137 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3140 /* Remove source name from source directory */
3141 /* Insert new fid with target name into target dir */
3142 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3147 rc = mdo_declare_index_insert(env, pobj,
3148 mdd_object_fid(mdd_tobj),
3149 mdd_object_type(mdd_tobj),
3150 lname.ln_name, handle);
3154 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3158 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3162 rc = __mdd_index_delete(env, pobj, lname.ln_name,
3167 rc = __mdd_index_insert(env, pobj,
3168 mdd_object_fid(mdd_tobj),
3169 mdd_object_type(mdd_tobj),
3170 lname.ln_name, handle);
3174 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3175 rc = mdo_ref_add(env, mdd_tobj, handle);
3176 mdd_write_unlock(env, mdd_tobj);
3180 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3181 mdo_ref_del(env, mdd_sobj, handle);
3182 mdd_write_unlock(env, mdd_sobj);
3185 mdd_object_put(env, pobj);
3193 static int mdd_migrate_xattrs(const struct lu_env *env,
3194 struct mdd_object *mdd_sobj,
3195 struct mdd_object *mdd_tobj)
3197 struct mdd_thread_info *info = mdd_env_info(env);
3198 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3200 struct thandle *handle;
3206 struct lu_buf list_xbuf;
3210 /* retrieve xattr list from the old object */
3211 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3212 if (list_xsize == -ENODATA)
3218 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3219 if (info->mti_big_buf.lb_buf == NULL)
3222 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3223 list_xbuf.lb_len = list_xsize;
3224 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3229 xname = list_xbuf.lb_buf;
3231 xlen = strnlen(xname, rem - 1) + 1;
3232 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3233 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3234 strcmp(XATTR_NAME_LMV, xname) == 0)
3237 /* For directory, if there are default layout, migrate here */
3238 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3239 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3242 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3243 if (xsize == -ENODATA)
3248 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3249 if (info->mti_link_buf.lb_buf == NULL)
3250 GOTO(out, rc = -ENOMEM);
3252 xbuf.lb_len = xsize;
3253 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3254 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3260 handle = mdd_trans_create(env, mdd);
3262 GOTO(out, rc = PTR_ERR(handle));
3264 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3267 GOTO(stop_trans, rc);
3268 /* Note: this transaction is part of migration, and it is not
3269 * the last step of migration, so we set th_local = 1 to avoid
3270 * update last rcvd for this transaction */
3271 handle->th_local = 1;
3272 rc = mdd_trans_start(env, mdd, handle);
3274 GOTO(stop_trans, rc);
3276 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3278 GOTO(stop_trans, rc = 0);
3281 GOTO(stop_trans, rc);
3283 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3290 memmove(xname, xname + xlen, rem);
3296 static int mdd_declare_migrate_create(const struct lu_env *env,
3297 struct mdd_object *mdd_pobj,
3298 struct mdd_object *mdd_sobj,
3299 struct mdd_object *mdd_tobj,
3300 struct md_op_spec *spec,
3302 union lmv_mds_md *mgr_ea,
3303 struct linkea_data *ldata,
3304 struct thandle *handle)
3306 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3307 const struct lu_buf *buf;
3311 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3312 handle, spec, NULL);
3316 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3321 if (S_ISLNK(la->la_mode)) {
3322 const char *target_name = spec->u.sp_symname;
3323 int sym_len = strlen(target_name);
3324 const struct lu_buf *buf;
3326 buf = mdd_buf_get_const(env, target_name, sym_len);
3327 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3331 } else if (S_ISDIR(la->la_mode)) {
3332 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3338 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3339 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3340 spec->u.sp_ea.eadatalen);
3341 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3347 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3348 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3349 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3354 la_flag->la_valid = LA_FLAGS;
3355 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3356 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3361 static int mdd_migrate_create(const struct lu_env *env,
3362 struct mdd_object *mdd_pobj,
3363 struct mdd_object *mdd_sobj,
3364 struct mdd_object *mdd_tobj,
3365 const struct lu_name *lname,
3368 struct mdd_thread_info *info = mdd_env_info(env);
3369 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3370 struct md_op_spec *spec = &info->mti_spec;
3371 struct lu_buf lmm_buf = { NULL };
3372 struct lu_buf link_buf = { NULL };
3373 const struct lu_buf *buf;
3374 struct thandle *handle;
3375 struct lmv_mds_md_v1 *mgr_ea;
3376 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3377 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3379 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3383 /* prepare spec for create */
3384 memset(spec, 0, sizeof(*spec));
3385 spec->sp_cr_lookup = 0;
3386 spec->sp_feat = &dt_directory_features;
3387 if (S_ISLNK(la->la_mode)) {
3388 buf = lu_buf_check_and_alloc(
3389 &mdd_env_info(env)->mti_big_buf,
3392 link_buf.lb_len = la->la_size + 1;
3393 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3394 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3396 rc = rc != 0 ? rc : -EFAULT;
3397 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3398 mdd2obd_dev(mdd)->obd_name,
3399 PFID(mdd_object_fid(mdd_sobj)), rc);
3402 spec->u.sp_symname = link_buf.lb_buf;
3403 } else if (S_ISREG(la->la_mode)) {
3404 /* retrieve lov of the old object */
3405 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3406 if (rc != 0 && rc != -ENODATA)
3408 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3409 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3410 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3411 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3413 } else if (S_ISDIR(la->la_mode)) {
3414 rc = mdd_links_read(env, mdd_sobj, ldata);
3415 if (rc < 0 && rc != -ENODATA)
3419 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3420 memset(mgr_ea, 0, sizeof(*mgr_ea));
3421 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3422 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3423 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3424 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3425 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3426 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3428 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3430 handle = mdd_trans_create(env, mdd);
3432 GOTO(out_free, rc = PTR_ERR(handle));
3434 /* Note: this transaction is part of migration, and it is not
3435 * the last step of migration, so we set th_local = 1 to avoid
3436 * update last rcvd for this transaction */
3437 handle->th_local = 1;
3438 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3440 (union lmv_mds_md *)info->mti_xattr_buf,
3443 GOTO(stop_trans, rc);
3445 rc = mdd_trans_start(env, mdd, handle);
3447 GOTO(stop_trans, rc);
3449 /* create the target object */
3450 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3453 GOTO(stop_trans, rc);
3455 if (S_ISDIR(la->la_mode)) {
3456 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3458 GOTO(stop_trans, rc);
3461 /* Set MIGRATE EA on the source inode, so once the migration needs
3462 * to be re-done during failover, the re-do process can locate the
3463 * target object which is already being created. */
3464 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3465 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3466 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
3468 GOTO(stop_trans, rc);
3470 /* Set immutable flag, so any modification is disabled until
3471 * the migration is done. Once the migration is interrupted,
3472 * if the resume process find the migrating object has both
3473 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3474 * flag and approve the migration */
3475 la_flag->la_valid = LA_FLAGS;
3476 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3477 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3479 if (handle != NULL) {
3482 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3487 if (lmm_buf.lb_buf != NULL)
3488 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3492 static int mdd_migrate_entries(const struct lu_env *env,
3493 struct mdd_object *mdd_sobj,
3494 struct mdd_object *mdd_tobj)
3496 struct dt_object *next = mdd_object_child(mdd_sobj);
3497 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3498 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3499 struct thandle *handle;
3501 const struct dt_it_ops *iops;
3504 struct lu_dirent *ent;
3507 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3511 if (!dt_try_as_dir(env, next))
3512 GOTO(out_ent, rc = -ENOTDIR);
3514 * iterate directories
3516 iops = &next->do_index_ops->dio_it;
3517 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3519 GOTO(out_ent, rc = PTR_ERR(it));
3521 rc = iops->load(env, it, 0);
3523 rc = iops->next(env, it);
3527 * At this point and across for-loop:
3529 * rc == 0 -> ok, proceed.
3530 * rc > 0 -> end of directory.
3534 struct mdd_object *child;
3535 char *name = mdd_env_info(env)->mti_key;
3539 bool target_exist = false;
3542 len = iops->key_size(env, it);
3546 result = iops->rec(env, it, (struct dt_rec *)ent,
3547 LUDA_FID | LUDA_TYPE);
3548 if (result == -ESTALE)
3555 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3556 recsize = le16_to_cpu(ent->lde_reclen);
3558 /* Insert new fid with target name into target dir */
3559 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3560 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3561 ent->lde_name[1] == '.'))
3564 child = mdd_object_find(env, mdd, &ent->lde_fid);
3566 GOTO(out, rc = PTR_ERR(child));
3568 mdd_write_lock(env, child, MOR_SRC_CHILD);
3569 is_dir = S_ISDIR(mdd_object_type(child));
3571 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3573 /* Check whether the name has been inserted to the target */
3574 if (dt_try_as_dir(env, dt_tobj)) {
3575 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3577 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3578 (struct dt_key *)name);
3579 if (unlikely(rc == 0))
3580 target_exist = true;
3583 handle = mdd_trans_create(env, mdd);
3585 GOTO(out, rc = PTR_ERR(handle));
3587 /* Note: this transaction is part of migration, and it is not
3588 * the last step of migration, so we set th_local = 1 to avoid
3589 * updating last rcvd for this transaction */
3590 handle->th_local = 1;
3591 if (likely(!target_exist)) {
3592 rc = mdo_declare_index_insert(env, mdd_tobj,
3594 mdd_object_type(child),
3600 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3606 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3611 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3615 /* Update .. for child */
3616 rc = mdo_declare_index_delete(env, child, dotdot,
3621 rc = mdo_declare_index_insert(env, child,
3622 mdd_object_fid(mdd_tobj),
3623 S_IFDIR, dotdot, handle);
3628 rc = mdd_linkea_declare_update_child(env, mdd_tobj,
3635 rc = mdd_trans_start(env, mdd, handle);
3637 CERROR("%s: transaction start failed: rc = %d\n",
3638 mdd2obd_dev(mdd)->obd_name, rc);
3642 if (likely(!target_exist)) {
3643 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3644 mdd_object_type(child),
3650 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3655 rc = __mdd_index_delete_only(env, child, dotdot,
3660 rc = __mdd_index_insert_only(env, child,
3661 mdd_object_fid(mdd_tobj), S_IFDIR,
3667 rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
3668 strlen(name), handle);
3671 mdd_write_unlock(env, child);
3672 mdd_object_put(env, child);
3673 rc1 = mdd_trans_stop(env, mdd, rc, handle);
3680 result = iops->next(env, it);
3681 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3682 GOTO(out, rc = -EINTR);
3684 if (result == -ESTALE)
3686 } while (result == 0);
3689 iops->fini(env, it);
3691 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3695 static int mdd_declare_update_linkea(const struct lu_env *env,
3696 struct mdd_object *mdd_pobj,
3697 struct mdd_object *mdd_sobj,
3698 struct mdd_object *mdd_tobj,
3699 struct thandle *handle,
3700 const struct lu_name *child_name)
3702 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3703 child_name, handle, 1);
3706 static int mdd_update_linkea(const struct lu_env *env,
3707 struct mdd_object *mdd_pobj,
3708 struct mdd_object *mdd_sobj,
3709 struct mdd_object *mdd_tobj,
3710 struct thandle *handle,
3711 const struct lu_name *child_name)
3713 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3714 child_name, handle, 0);
3717 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3718 struct mdd_object *mdd_pobj,
3719 struct mdd_object *mdd_sobj,
3720 struct mdd_object *mdd_tobj,
3721 const struct lu_name *lname,
3723 struct lu_attr *parent_la,
3724 struct thandle *handle)
3726 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3729 /* Revert IMMUTABLE flag */
3730 la_flag->la_valid = LA_FLAGS;
3731 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3732 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3736 /* delete entry from source dir */
3737 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3741 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3742 mdd_tobj, handle, lname);
3746 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3747 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3752 handle->th_complex = 1;
3753 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3755 LU_XATTR_REPLACE, handle);
3760 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3761 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3767 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3768 mdd_object_type(mdd_tobj),
3769 lname->ln_name, handle);
3773 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE);
3777 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3778 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3783 /* delete old object */
3784 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3788 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3789 /* delete old object */
3790 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3793 /* set nlink to 0 */
3794 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3799 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3803 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3808 static int mdd_migrate_update_name(const struct lu_env *env,
3809 struct mdd_object *mdd_pobj,
3810 struct mdd_object *mdd_sobj,
3811 struct mdd_object *mdd_tobj,
3812 const struct lu_name *lname,
3815 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3816 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3817 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3818 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3819 struct thandle *handle;
3820 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3821 const char *name = lname->ln_name;
3825 /* update time for parent */
3826 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3827 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3828 p_la->la_valid = LA_CTIME;
3830 rc = mdd_la_get(env, mdd_sobj, so_attr);
3834 handle = mdd_trans_create(env, mdd);
3836 RETURN(PTR_ERR(handle));
3838 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3839 lname, so_attr, p_la, handle);
3841 /* If the migration can not be fit in one transaction, just
3842 * leave it in the original MDT */
3844 GOTO(stop_trans, rc = 0);
3846 GOTO(stop_trans, rc);
3849 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3850 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3851 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3852 PFID(mdd_object_fid(mdd_tobj)));
3854 rc = mdd_trans_start(env, mdd, handle);
3856 GOTO(stop_trans, rc);
3858 /* Revert IMMUTABLE flag */
3859 la_flag->la_valid = LA_FLAGS;
3860 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3861 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3863 GOTO(stop_trans, rc);
3865 /* Remove source name from source directory */
3866 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
3868 GOTO(stop_trans, rc);
3870 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3873 GOTO(stop_trans, rc);
3875 if (S_ISREG(so_attr->la_mode)) {
3876 if (so_attr->la_nlink == 1) {
3877 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3879 if (rc != 0 && rc != -ENODATA)
3880 GOTO(stop_trans, rc);
3882 rc = mdo_xattr_set(env, mdd_tobj, NULL,
3884 LU_XATTR_REPLACE, handle);
3886 GOTO(stop_trans, rc);
3890 /* Insert new fid with target name into target dir */
3891 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
3892 mdd_object_type(mdd_tobj), name, handle);
3894 GOTO(stop_trans, rc);
3896 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
3899 GOTO(stop_trans, rc);
3901 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3903 /* Increase mod_count to add the source object to the orphan list,
3904 * so if other clients still send RPC to the old object, then these
3905 * objects can help the request to find the new object, see
3906 * mdt_reint_open() */
3907 mdd_sobj->mod_count++;
3908 rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
3909 mdd_sobj->mod_count--;
3911 GOTO(out_unlock, rc);
3913 mdo_ref_del(env, mdd_sobj, handle);
3915 mdo_ref_del(env, mdd_sobj, handle);
3917 /* Get the attr again after ref_del */
3918 rc = mdd_la_get(env, mdd_sobj, so_attr);
3920 GOTO(out_unlock, rc);
3922 ma->ma_attr = *so_attr;
3923 ma->ma_valid |= MA_INODE;
3925 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
3927 GOTO(out_unlock, rc);
3930 mdd_write_unlock(env, mdd_sobj);
3933 mdd_trans_stop(env, mdd, rc, handle);
3939 * Check whether we should migrate the file/dir
3941 * < 0 permission check failed or other error.
3942 * = 0 the file can be migrated.
3943 * > 0 the file does not need to be migrated, mostly
3944 * for multiple link file
3946 static int mdd_migrate_sanity_check(const struct lu_env *env,
3947 struct mdd_object *pobj,
3948 const struct lu_attr *pattr,
3949 struct mdd_object *sobj,
3950 struct lu_attr *sattr)
3952 struct mdd_thread_info *info = mdd_env_info(env);
3953 struct linkea_data *ldata = &info->mti_link_data;
3955 struct lu_buf *mgr_buf;
3961 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3962 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
3963 if (mgr_buf->lb_buf == NULL)
3966 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
3968 union lmv_mds_md *lmm = mgr_buf->lb_buf;
3970 /* If the object has migrateEA, it means IMMUTE flag
3971 * is being set by previous migration process, so it
3972 * needs to override the IMMUTE flag, otherwise the
3973 * following sanity check will fail */
3974 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
3975 LMV_HASH_FLAG_MIGRATION) {
3976 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3978 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
3979 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
3980 mdd2obd_dev(mdd)->obd_name,
3981 PFID(mdd_object_fid(sobj)));
3985 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
3986 sobj, sattr, NULL, NULL);
3990 /* Then it will check if the file should be migrated. If the file
3991 * has mulitple links, we only need migrate the file if all of its
3992 * entries has been migrated to the remote MDT */
3993 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
3996 rc = mdd_links_read(env, sobj, ldata);
3998 /* For multiple links files, if there are no linkEA data at all,
3999 * means the file might be created before linkEA is enabled, and
4000 * all of its links should not be migrated yet, otherwise it
4001 * should have some linkEA there */
4002 if (rc == -ENOENT || rc == -ENODATA)
4007 /* If there are still links locally, then the file will not be
4009 LASSERT(ldata->ld_leh != NULL);
4010 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4011 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4012 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4013 struct mdd_object *lpobj;
4014 struct lu_name lname;
4017 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4019 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4021 lpobj = mdd_object_find(env, mdd, &fid);
4022 if (IS_ERR(lpobj)) {
4023 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
4024 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
4029 if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) {
4030 CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n",
4031 PFID(&fid), lname.ln_namelen, lname.ln_name);
4032 mdd_object_put(env, lpobj);
4036 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4037 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4038 lname.ln_name, PFID(&fid));
4039 mdd_object_put(env, lpobj);
4047 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4048 struct md_object *sobj, const struct lu_name *lname,
4049 struct md_object *tobj, struct md_attr *ma)
4051 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4052 struct mdd_device *mdd = mdo2mdd(pobj);
4053 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4054 struct mdd_object *mdd_tobj = NULL;
4055 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4056 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4060 /* If the file will being migrated, it will check whether
4061 * the file is being opened by someone else right now */
4062 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4063 if (mdd_sobj->mod_count > 0) {
4064 CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
4065 mdd2obd_dev(mdd)->obd_name,
4066 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4067 mdd_sobj->mod_count, -EBUSY);
4068 mdd_read_unlock(env, mdd_sobj);
4069 GOTO(put, rc = -EBUSY);
4071 mdd_read_unlock(env, mdd_sobj);
4073 rc = mdd_la_get(env, mdd_sobj, so_attr);
4077 rc = mdd_la_get(env, mdd_pobj, pattr);
4081 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4088 /* Sigh, it is impossible to finish all of migration in a single
4089 * transaction, for example migrating big directory entries to the
4090 * new MDT, it needs insert all of name entries of children in the
4093 * So migration will be done in multiple steps and transactions.
4095 * 1. create an orphan object on the remote MDT in one transaction.
4096 * 2. migrate extend attributes to the new target file/directory.
4097 * 3. For directory, migrate the entries to the new MDT and update
4098 * linkEA of each children. Because we can not migrate all entries
4099 * in a single transaction, so the migrating directory will become
4100 * a striped directory during migration, so once the process is
4101 * interrupted, the directory is still accessible. (During lookup,
4102 * client will locate the name by searching both original and target
4104 * 4. Finally, update the name/FID to point to the new file/directory
4105 * in a separate transaction.
4108 /* step 1: Check whether the orphan object has been created, and create
4109 * orphan object on the remote MDT if needed */
4110 mdd_tobj = md2mdd_obj(tobj);
4111 if (!mdd_object_exists(mdd_tobj)) {
4112 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4118 LASSERT(mdd_object_exists(mdd_tobj));
4119 /* step 2: migrate xattr */
4120 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4124 /* step 3: migrate name entries to the orphan object */
4125 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4126 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4129 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4130 OBD_FAIL_MDS_REINT_NET_REP)))
4133 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4136 LASSERT(mdd_object_exists(mdd_tobj));
4137 /* step 4: update name entry to the new object */
4138 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4146 const struct md_dir_operations mdd_dir_ops = {
4147 .mdo_is_subdir = mdd_is_subdir,
4148 .mdo_lookup = mdd_lookup,
4149 .mdo_create = mdd_create,
4150 .mdo_rename = mdd_rename,
4151 .mdo_link = mdd_link,
4152 .mdo_unlink = mdd_unlink,
4153 .mdo_create_data = mdd_create_data,
4154 .mdo_migrate = mdd_migrate,