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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_dir.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@intel.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <obd_class.h>
42 #include <obd_support.h>
43 #include <lustre_mds.h>
44 #include <lustre_fid.h>
45 #include <lustre_lmv.h>
46 #include <lustre_idmap.h>
48 #include "mdd_internal.h"
50 static const char dot[] = ".";
51 static const char dotdot[] = "..";
53 static struct lu_name lname_dotdot = {
54 .ln_name = (char *) dotdot,
55 .ln_namelen = sizeof(dotdot) - 1,
59 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
61 if (!lu_name_is_valid(ln))
63 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
69 /* Get FID from name and parent */
71 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
72 const struct lu_attr *pattr, const struct lu_name *lname,
73 struct lu_fid* fid, int mask)
75 const char *name = lname->ln_name;
76 const struct dt_key *key = (const struct dt_key *)name;
77 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
78 struct mdd_device *m = mdo2mdd(pobj);
79 struct dt_object *dir = mdd_object_child(mdd_obj);
83 if (unlikely(mdd_is_dead_obj(mdd_obj)))
86 if (!mdd_object_exists(mdd_obj))
89 if (mdd_object_remote(mdd_obj)) {
90 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
91 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
94 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
99 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
100 dt_try_as_dir(env, dir)))
101 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
108 int mdd_lookup(const struct lu_env *env,
109 struct md_object *pobj, const struct lu_name *lname,
110 struct lu_fid *fid, struct md_op_spec *spec)
112 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
116 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
120 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
121 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
125 /** Read the link EA into a temp buffer.
126 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
127 * A pointer to the buffer is stored in \a ldata::ld_buf.
131 static int __mdd_links_read(const struct lu_env *env,
132 struct mdd_object *mdd_obj,
133 struct linkea_data *ldata)
137 if (!mdd_object_exists(mdd_obj))
140 /* First try a small buf */
141 LASSERT(env != NULL);
142 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
144 if (ldata->ld_buf->lb_buf == NULL)
147 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
149 /* Buf was too small, figure out what we need. */
150 lu_buf_free(ldata->ld_buf);
151 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
155 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
156 if (ldata->ld_buf->lb_buf == NULL)
158 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
162 lu_buf_free(ldata->ld_buf);
163 ldata->ld_buf = NULL;
167 return linkea_init(ldata);
170 static int mdd_links_read(const struct lu_env *env,
171 struct mdd_object *mdd_obj,
172 struct linkea_data *ldata)
176 rc = __mdd_links_read(env, mdd_obj, ldata);
178 rc = linkea_init(ldata);
183 static int mdd_links_read_with_rec(const struct lu_env *env,
184 struct mdd_object *mdd_obj,
185 struct linkea_data *ldata)
189 rc = __mdd_links_read(env, mdd_obj, ldata);
191 rc = linkea_init_with_rec(ldata);
197 * Get parent FID of the directory
199 * Read parent FID from linkEA, if that fails, then do lookup
200 * dotdot to get the parent FID.
202 * \param[in] env execution environment
203 * \param[in] obj object from which to find the parent FID
204 * \param[in] attr attribute of the object
205 * \param[out] fid fid to get the parent FID
207 * \retval 0 if getting the parent FID succeeds.
208 * \retval negative errno if getting the parent FID fails.
210 static inline int mdd_parent_fid(const struct lu_env *env,
211 struct mdd_object *obj,
212 const struct lu_attr *attr,
215 struct mdd_thread_info *info = mdd_env_info(env);
216 struct linkea_data ldata = { NULL };
217 struct lu_buf *buf = &info->mti_link_buf;
218 struct lu_name lname;
223 LASSERTF(S_ISDIR(mdd_object_type(obj)),
224 "%s: FID "DFID" is not a directory type = %o\n",
225 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)),
226 mdd_object_type(obj));
228 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
229 if (buf->lb_buf == NULL)
230 GOTO(lookup, rc = 0);
233 rc = mdd_links_read_with_rec(env, obj, &ldata);
237 LASSERT(ldata.ld_leh != NULL);
238 /* Directory should only have 1 parent */
239 if (ldata.ld_leh->leh_reccount > 1)
242 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
244 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
245 if (likely(fid_is_sane(fid)))
248 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
253 * For root fid use special function, which does not compare version component
254 * of fid. Version component is different for root fids on all MDTs.
256 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
258 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
259 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
263 * return 1: if \a tfid is the fid of the ancestor of \a mo;
265 * otherwise: values < 0, errors.
267 static int mdd_is_parent(const struct lu_env *env,
268 struct mdd_device *mdd,
269 struct mdd_object *mo,
270 const struct lu_attr *attr,
271 const struct lu_fid *tfid)
273 struct mdd_object *mp;
277 LASSERT(!lu_fid_eq(mdo2fid(mo), tfid));
278 pfid = &mdd_env_info(env)->mti_fid;
280 if (mdd_is_root(mdd, mdo2fid(mo)))
283 if (mdd_is_root(mdd, tfid))
286 rc = mdd_parent_fid(env, mo, attr, pfid);
291 if (lu_fid_eq(pfid, tfid))
294 if (mdd_is_root(mdd, pfid))
297 mp = mdd_object_find(env, mdd, pfid);
301 if (!mdd_object_exists(mp)) {
302 mdd_object_put(env, mp);
306 rc = mdd_parent_fid(env, mp, attr, pfid);
307 mdd_object_put(env, mp);
316 * No permission check is needed.
318 * returns 1: if fid is ancestor of @mo;
319 * returns 0: if fid is not an ancestor of @mo;
320 * returns < 0: if error
322 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
323 const struct lu_fid *fid)
325 struct mdd_device *mdd = mdo2mdd(mo);
326 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
330 if (!mdd_object_exists(md2mdd_obj(mo)))
333 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
336 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
340 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid);
345 * Check that @dir contains no entries except (possibly) dot and dotdot.
350 * -ENOTDIR not a directory object
351 * -ENOTEMPTY not empty
355 static int mdd_dir_is_empty(const struct lu_env *env,
356 struct mdd_object *dir)
359 struct dt_object *obj;
360 const struct dt_it_ops *iops;
364 obj = mdd_object_child(dir);
365 if (!dt_try_as_dir(env, obj))
368 iops = &obj->do_index_ops->dio_it;
369 it = iops->init(env, obj, LUDA_64BITHASH);
371 result = iops->get(env, it, (const struct dt_key *)"");
374 for (result = 0, i = 0; result == 0 && i < 3; ++i)
375 result = iops->next(env, it);
378 else if (result == 1)
380 } else if (result == 0)
382 * Huh? Index contains no zero key?
389 result = PTR_ERR(it);
390 /* -ENODEV means no valid stripe */
391 if (result == -ENODEV)
398 * Determine if the target object can be hard linked, and right now it only
399 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
400 * directory nlink count might exceed the maximum link count(see
401 * osd_object_ref_add), so it only check nlink for non-directories.
403 * \param[in] env thread environment
404 * \param[in] obj object being linked to
405 * \param[in] la attributes of \a obj
407 * \retval 0 if \a obj can be hard linked
408 * \retval negative error if \a obj is a directory or has too
411 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
412 const struct lu_attr *la)
414 struct mdd_device *m = mdd_obj2mdd_dev(obj);
419 /* Subdir count limitation can be broken through
420 * (see osd_object_ref_add), so only check non-directory here. */
421 if (!S_ISDIR(la->la_mode) &&
422 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
429 * Check whether it may create the cobj under the pobj.
431 * \param[in] env execution environment
432 * \param[in] pobj the parent directory
433 * \param[in] pattr the attribute of the parent directory
434 * \param[in] cobj the child to be created
435 * \param[in] check_perm if check WRITE|EXEC permission for parent
437 * \retval = 0 create the child under this dir is allowed
438 * \retval negative errno create the child under this dir is
441 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
442 const struct lu_attr *pattr, struct mdd_object *cobj,
448 if (cobj && mdd_object_exists(cobj))
451 if (mdd_is_dead_obj(pobj))
455 rc = mdd_permission_internal_locked(env, pobj, pattr,
456 MAY_WRITE | MAY_EXEC,
462 * Check whether can unlink from the pobj in the case of "cobj == NULL".
464 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
465 const struct lu_attr *pattr, const struct lu_attr *attr)
470 if (mdd_is_dead_obj(pobj))
473 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
476 rc = mdd_permission_internal_locked(env, pobj, pattr,
477 MAY_WRITE | MAY_EXEC,
482 if (pattr->la_flags & LUSTRE_APPEND_FL)
489 * pobj == NULL is remote ops case, under such case, pobj's
490 * VTX feature has been checked already, no need check again.
492 static inline int mdd_is_sticky(const struct lu_env *env,
493 struct mdd_object *pobj,
494 const struct lu_attr *pattr,
495 struct mdd_object *cobj,
496 const struct lu_attr *cattr)
498 struct lu_ucred *uc = lu_ucred_assert(env);
501 LASSERT(pattr != NULL);
502 if (!(pattr->la_mode & S_ISVTX) ||
503 (pattr->la_uid == uc->uc_fsuid))
507 LASSERT(cattr != NULL);
508 if (cattr->la_uid == uc->uc_fsuid)
511 return !md_capable(uc, CFS_CAP_FOWNER);
514 static int mdd_may_delete_entry(const struct lu_env *env,
515 struct mdd_object *pobj,
516 const struct lu_attr *pattr,
521 LASSERT(pobj != NULL);
522 if (!mdd_object_exists(pobj))
525 if (mdd_is_dead_obj(pobj))
530 rc = mdd_permission_internal_locked(env, pobj, pattr,
531 MAY_WRITE | MAY_EXEC,
537 if (pattr->la_flags & LUSTRE_APPEND_FL)
544 * Check whether it may delete the cobj from the pobj.
547 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
548 const struct lu_attr *tpattr, struct mdd_object *tobj,
549 const struct lu_attr *tattr, const struct lu_attr *cattr,
550 int check_perm, int check_empty)
556 LASSERT(tpattr != NULL);
557 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
565 if (!mdd_object_exists(tobj))
568 if (mdd_is_dead_obj(tobj))
571 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
574 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
577 /* additional check the rename case */
579 if (S_ISDIR(cattr->la_mode)) {
580 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
582 if (!S_ISDIR(tattr->la_mode))
585 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
587 } else if (S_ISDIR(tattr->la_mode))
591 if (S_ISDIR(tattr->la_mode) && check_empty)
592 rc = mdd_dir_is_empty(env, tobj);
598 * Check whether it can create the link file(linked to @src_obj) under
599 * the target directory(@tgt_obj), and src_obj has been locked by
602 * \param[in] env execution environment
603 * \param[in] tgt_obj the target directory
604 * \param[in] tattr attributes of target directory
605 * \param[in] lname the link name
606 * \param[in] src_obj source object for link
607 * \param[in] cattr attributes for source object
609 * \retval = 0 it is allowed to create the link file under tgt_obj
610 * \retval negative error not allowed to create the link file
612 static int mdd_link_sanity_check(const struct lu_env *env,
613 struct mdd_object *tgt_obj,
614 const struct lu_attr *tattr,
615 const struct lu_name *lname,
616 struct mdd_object *src_obj,
617 const struct lu_attr *cattr)
619 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
623 if (!mdd_object_exists(src_obj))
626 if (mdd_is_dead_obj(src_obj))
629 /* Local ops, no lookup before link, check filename length here. */
630 rc = mdd_name_check(m, lname);
634 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
637 if (S_ISDIR(mdd_object_type(src_obj)))
640 LASSERT(src_obj != tgt_obj);
641 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
645 rc = __mdd_may_link(env, src_obj, cattr);
650 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
651 const char *name, struct thandle *handle)
653 struct dt_object *next = mdd_object_child(pobj);
657 if (dt_try_as_dir(env, next))
658 rc = dt_delete(env, next, (struct dt_key *)name, handle);
665 static int __mdd_index_insert_only(const struct lu_env *env,
666 struct mdd_object *pobj,
667 const struct lu_fid *lf, __u32 type,
669 struct thandle *handle)
671 struct dt_object *next = mdd_object_child(pobj);
675 if (dt_try_as_dir(env, next)) {
676 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
679 rec->rec_type = type;
680 rc = dt_insert(env, next, (const struct dt_rec *)rec,
681 (const struct dt_key *)name, handle);
688 /* insert named index, add reference if isdir */
689 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
690 const struct lu_fid *lf, __u32 type,
691 const char *name, struct thandle *handle)
696 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
697 if (rc == 0 && S_ISDIR(type)) {
698 mdd_write_lock(env, pobj, DT_TGT_PARENT);
699 mdo_ref_add(env, pobj, handle);
700 mdd_write_unlock(env, pobj);
706 /* delete named index, drop reference if isdir */
707 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
708 const char *name, int is_dir,
709 struct thandle *handle)
714 rc = __mdd_index_delete_only(env, pobj, name, handle);
715 if (rc == 0 && is_dir) {
716 mdd_write_lock(env, pobj, DT_TGT_PARENT);
717 mdo_ref_del(env, pobj, handle);
718 mdd_write_unlock(env, pobj);
724 static int mdd_llog_record_calc_size(const struct lu_env *env,
725 const struct lu_name *tname,
726 const struct lu_name *sname)
728 const struct lu_ucred *uc = lu_ucred(env);
729 enum changelog_rec_flags clf_flags = CLF_EXTRA_FLAGS;
730 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
733 clf_flags |= CLF_RENAME;
735 if (uc != NULL && uc->uc_jobid[0] != '\0')
736 clf_flags |= CLF_JOBID;
738 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
739 changelog_rec_offset(clf_flags, crfe) +
740 (tname != NULL ? tname->ln_namelen : 0) +
741 (sname != NULL ? 1 + sname->ln_namelen : 0));
744 int mdd_declare_changelog_store(const struct lu_env *env,
745 struct mdd_device *mdd,
746 enum changelog_rec_type type,
747 const struct lu_name *tname,
748 const struct lu_name *sname,
749 struct thandle *handle)
751 struct obd_device *obd = mdd2obd_dev(mdd);
752 struct llog_ctxt *ctxt;
753 struct llog_changelog_rec *rec;
755 struct thandle *llog_th;
759 if (!mdd_changelog_enabled(env, mdd, type))
762 reclen = mdd_llog_record_calc_size(env, tname, sname);
763 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
764 if (buf->lb_buf == NULL)
768 rec->cr_hdr.lrh_len = reclen;
769 rec->cr_hdr.lrh_type = CHANGELOG_REC;
771 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
775 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
777 GOTO(out_put, rc = PTR_ERR(llog_th));
779 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
787 /** Add a changelog entry \a rec to the changelog llog
790 * \param handle - currently ignored since llogs start their own transaction;
791 * this will hopefully be fixed in llog rewrite
794 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
795 struct llog_changelog_rec *rec, struct thandle *th)
797 struct obd_device *obd = mdd2obd_dev(mdd);
798 struct llog_ctxt *ctxt;
799 struct thandle *llog_th;
802 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
803 changelog_rec_varsize(&rec->cr));
805 /* llog_lvfs_write_rec sets the llog tail len */
806 rec->cr_hdr.lrh_type = CHANGELOG_REC;
807 rec->cr.cr_time = cl_time();
809 spin_lock(&mdd->mdd_cl.mc_lock);
810 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
811 * but as long as the MDD transactions are ordered correctly for e.g.
812 * rename conflicts, I don't think this should matter. */
813 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
814 spin_unlock(&mdd->mdd_cl.mc_lock);
816 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
820 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
822 GOTO(out_put, rc = PTR_ERR(llog_th));
824 /* nested journal transaction */
825 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
827 /* time to recover some space ?? */
828 if (likely(!mdd->mdd_changelog_gc ||
829 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
830 mdd->mdd_changelog_min_gc_interval >=
831 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
832 /* save a spin_lock trip */
834 spin_lock(&mdd->mdd_cl.mc_lock);
835 if (likely(mdd->mdd_changelog_gc &&
836 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
837 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
838 mdd->mdd_changelog_min_gc_interval)) {
839 if (unlikely(llog_cat_free_space(ctxt->loc_handle) <=
840 mdd->mdd_changelog_min_free_cat_entries ||
841 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD))) {
842 CWARN("%s:%s low on changelog_catalog free entries, "
843 "starting ChangeLog garbage collection thread\n",
845 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
848 /* indicate further kthread run will occur outside
849 * right after current journal transaction filling has
852 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
854 /* next check in mdd_changelog_min_gc_interval anyway
856 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
858 spin_unlock(&mdd->mdd_cl.mc_lock);
866 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
867 const struct lu_fid *sfid,
868 const struct lu_fid *spfid,
869 const struct lu_name *sname)
871 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
874 LASSERT(sfid != NULL);
875 LASSERT(spfid != NULL);
876 LASSERT(sname != NULL);
878 extsize = sname->ln_namelen + 1;
880 rnm->cr_sfid = *sfid;
881 rnm->cr_spfid = *spfid;
883 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
884 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
885 rec->cr_namelen += extsize;
888 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
890 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
892 if (jobid == NULL || jobid[0] == '\0')
895 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
898 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
900 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
902 ef->cr_extra_flags = eflags;
905 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
906 __u64 uid, __u64 gid)
908 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
910 uidgid->cr_uid = uid;
911 uidgid->cr_gid = gid;
914 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
917 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
922 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
924 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
926 omd->cr_openflags = flags;
929 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
930 const char *xattr_name)
932 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
934 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
937 /** Store a namespace change changelog record
938 * If this fails, we must fail the whole transaction; we don't
939 * want the change to commit without the log entry.
940 * \param target - mdd_object of change
941 * \param tpfid - target parent dir/object fid
942 * \param sfid - source object fid
943 * \param spfid - source parent fid
944 * \param tname - target name string
945 * \param sname - source name string
946 * \param handle - transaction handle
948 int mdd_changelog_ns_store(const struct lu_env *env,
949 struct mdd_device *mdd,
950 enum changelog_rec_type type,
951 enum changelog_rec_flags clf_flags,
952 struct mdd_object *target,
953 const struct lu_fid *tpfid,
954 const struct lu_fid *sfid,
955 const struct lu_fid *spfid,
956 const struct lu_name *tname,
957 const struct lu_name *sname,
958 struct thandle *handle)
960 const struct lu_ucred *uc = lu_ucred(env);
961 struct llog_changelog_rec *rec;
964 __u64 xflags = CLFE_INVALID;
968 if (!mdd_changelog_enabled(env, mdd, type))
971 LASSERT(tpfid != NULL);
972 LASSERT(tname != NULL);
973 LASSERT(handle != NULL);
975 reclen = mdd_llog_record_calc_size(env, tname, sname);
976 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
977 if (buf->lb_buf == NULL)
981 clf_flags &= CLF_FLAGMASK;
982 clf_flags |= CLF_EXTRA_FLAGS;
985 if (uc->uc_jobid[0] != '\0')
986 clf_flags |= CLF_JOBID;
987 xflags |= CLFE_UIDGID;
992 clf_flags |= CLF_RENAME;
994 clf_flags |= CLF_VERSION;
996 rec->cr.cr_flags = clf_flags;
998 if (clf_flags & CLF_EXTRA_FLAGS) {
999 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1000 if (xflags & CLFE_UIDGID)
1001 mdd_changelog_rec_extra_uidgid(&rec->cr,
1002 uc->uc_uid, uc->uc_gid);
1003 if (xflags & CLFE_NID)
1004 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1007 rec->cr.cr_type = (__u32)type;
1008 rec->cr.cr_pfid = *tpfid;
1009 rec->cr.cr_namelen = tname->ln_namelen;
1010 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1012 if (clf_flags & CLF_RENAME)
1013 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1015 if (clf_flags & CLF_JOBID)
1016 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1018 if (likely(target != NULL)) {
1019 rec->cr.cr_tfid = *mdo2fid(target);
1020 target->mod_cltime = ktime_get();
1022 fid_zero(&rec->cr.cr_tfid);
1025 rc = mdd_changelog_store(env, mdd, rec, handle);
1027 CERROR("%s: cannot store changelog record: type = %d, "
1028 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1029 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1030 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1037 static int __mdd_links_add(const struct lu_env *env,
1038 struct mdd_object *mdd_obj,
1039 struct linkea_data *ldata,
1040 const struct lu_name *lname,
1041 const struct lu_fid *pfid,
1042 int first, int check)
1046 if (ldata->ld_leh == NULL) {
1047 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1051 rc = linkea_data_new(ldata,
1052 &mdd_env_info(env)->mti_link_buf);
1059 rc = linkea_links_find(ldata, lname, pfid);
1060 if (rc && rc != -ENOENT)
1066 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1067 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1071 linkea_add_buf(ldata, lname, tfid);
1074 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1075 linkea_add_buf(ldata, lname, pfid);
1077 return linkea_add_buf(ldata, lname, pfid);
1080 static int __mdd_links_del(const struct lu_env *env,
1081 struct mdd_object *mdd_obj,
1082 struct linkea_data *ldata,
1083 const struct lu_name *lname,
1084 const struct lu_fid *pfid)
1088 if (ldata->ld_leh == NULL) {
1089 rc = mdd_links_read(env, mdd_obj, ldata);
1094 rc = linkea_links_find(ldata, lname, pfid);
1098 linkea_del_buf(ldata, lname);
1102 static int mdd_linkea_prepare(const struct lu_env *env,
1103 struct mdd_object *mdd_obj,
1104 const struct lu_fid *oldpfid,
1105 const struct lu_name *oldlname,
1106 const struct lu_fid *newpfid,
1107 const struct lu_name *newlname,
1108 int first, int check,
1109 struct linkea_data *ldata)
1114 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1117 LASSERT(oldpfid != NULL || newpfid != NULL);
1119 if (mdd_obj->mod_flags & DEAD_OBJ)
1120 /* Unnecessary to update linkEA for dead object. */
1123 if (oldpfid != NULL) {
1124 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1126 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1129 /* No changes done. */
1134 /* If renaming, add the new record */
1135 if (newpfid != NULL)
1136 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1142 int mdd_links_rename(const struct lu_env *env,
1143 struct mdd_object *mdd_obj,
1144 const struct lu_fid *oldpfid,
1145 const struct lu_name *oldlname,
1146 const struct lu_fid *newpfid,
1147 const struct lu_name *newlname,
1148 struct thandle *handle,
1149 struct linkea_data *ldata,
1150 int first, int check)
1155 if (ldata == NULL) {
1156 ldata = &mdd_env_info(env)->mti_link_data;
1157 memset(ldata, 0, sizeof(*ldata));
1158 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1159 newpfid, newlname, first, check, ldata);
1164 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1165 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1171 if (newlname == NULL)
1172 CERROR("link_ea add failed %d "DFID"\n",
1173 rc, PFID(mdd_object_fid(mdd_obj)));
1174 else if (oldpfid == NULL)
1175 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1176 newlname->ln_namelen, newlname->ln_name, rc,
1177 PFID(mdd_object_fid(mdd_obj)));
1178 else if (newpfid == NULL)
1179 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1180 oldlname->ln_namelen, oldlname->ln_name, rc,
1181 PFID(mdd_object_fid(mdd_obj)));
1183 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1184 "\n", oldlname->ln_namelen, oldlname->ln_name,
1185 newlname->ln_namelen, newlname->ln_name, rc,
1186 PFID(mdd_object_fid(mdd_obj)));
1189 if (is_vmalloc_addr(ldata->ld_buf))
1190 /* if we vmalloced a large buffer drop it */
1191 lu_buf_free(ldata->ld_buf);
1196 static inline int mdd_links_add(const struct lu_env *env,
1197 struct mdd_object *mdd_obj,
1198 const struct lu_fid *pfid,
1199 const struct lu_name *lname,
1200 struct thandle *handle,
1201 struct linkea_data *ldata, int first)
1203 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1204 pfid, lname, handle, ldata, first, 0);
1207 static inline int mdd_links_del(const struct lu_env *env,
1208 struct mdd_object *mdd_obj,
1209 const struct lu_fid *pfid,
1210 const struct lu_name *lname,
1211 struct thandle *handle)
1213 return mdd_links_rename(env, mdd_obj, pfid, lname,
1214 NULL, NULL, handle, NULL, 0, 0);
1217 /** Read the link EA into a temp buffer.
1218 * Uses the name_buf since it is generally large.
1219 * \retval IS_ERR err
1220 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1222 struct lu_buf *mdd_links_get(const struct lu_env *env,
1223 struct mdd_object *mdd_obj)
1225 struct linkea_data ldata = { NULL };
1228 rc = mdd_links_read(env, mdd_obj, &ldata);
1229 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1232 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1233 struct linkea_data *ldata, struct thandle *handle)
1235 const struct lu_buf *buf;
1238 if (ldata == NULL || ldata->ld_buf == NULL ||
1239 ldata->ld_leh == NULL)
1242 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1246 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1247 ldata->ld_leh->leh_len);
1248 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1249 if (unlikely(rc == -ENOSPC)) {
1250 rc = linkea_overflow_shrink(ldata);
1258 static int mdd_declare_links_add(const struct lu_env *env,
1259 struct mdd_object *mdd_obj,
1260 struct thandle *handle,
1261 struct linkea_data *ldata)
1267 if (ldata != NULL && ldata->ld_leh != NULL) {
1268 ea_len = ldata->ld_leh->leh_len;
1269 linkea = ldata->ld_buf->lb_buf;
1271 ea_len = MAX_LINKEA_SIZE;
1275 rc = mdo_declare_xattr_set(env, mdd_obj,
1276 mdd_buf_get_const(env, linkea, ea_len),
1277 XATTR_NAME_LINK, 0, handle);
1282 static inline int mdd_declare_links_del(const struct lu_env *env,
1283 struct mdd_object *c,
1284 struct thandle *handle)
1288 /* For directory, the linkEA will be removed together
1289 * with the object. */
1290 if (!S_ISDIR(mdd_object_type(c)))
1291 rc = mdd_declare_links_add(env, c, handle, NULL);
1296 static int mdd_declare_link(const struct lu_env *env,
1297 struct mdd_device *mdd,
1298 struct mdd_object *p,
1299 struct mdd_object *c,
1300 const struct lu_name *name,
1301 struct thandle *handle,
1303 struct linkea_data *data)
1305 struct lu_fid tfid = *mdo2fid(c);
1308 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1309 tfid.f_oid = cfs_fail_val;
1311 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1312 name->ln_name, handle);
1316 rc = mdo_declare_ref_add(env, c, handle);
1320 la->la_valid = LA_CTIME | LA_MTIME;
1321 rc = mdo_declare_attr_set(env, p, la, handle);
1325 la->la_valid = LA_CTIME;
1326 rc = mdo_declare_attr_set(env, c, la, handle);
1330 rc = mdd_declare_links_add(env, c, handle, data);
1334 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1340 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1341 struct md_object *src_obj, const struct lu_name *lname,
1344 const char *name = lname->ln_name;
1345 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1346 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1347 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1348 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1349 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1350 struct mdd_device *mdd = mdo2mdd(src_obj);
1351 struct thandle *handle;
1352 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1353 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1357 rc = mdd_la_get(env, mdd_sobj, cattr);
1361 rc = mdd_la_get(env, mdd_tobj, tattr);
1366 * If we are using project inheritance, we only allow hard link
1367 * creation in our tree when the project IDs are the same;
1368 * otherwise the tree quota mechanism could be circumvented.
1370 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1371 (tattr->la_projid != cattr->la_projid))
1374 handle = mdd_trans_create(env, mdd);
1376 GOTO(out_pending, rc = PTR_ERR(handle));
1378 memset(ldata, 0, sizeof(*ldata));
1380 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1381 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1383 /* Note: even this function will change ldata, but it comes from
1384 * thread_info, which is completely temporary and only seen in
1385 * this function, so we do not need reset ldata once it fails.*/
1386 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1387 lname, 0, 0, ldata);
1391 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1396 rc = mdd_trans_start(env, mdd, handle);
1400 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1401 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1404 GOTO(out_unlock, rc);
1406 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1407 rc = mdo_ref_add(env, mdd_sobj, handle);
1409 GOTO(out_unlock, rc);
1412 *tfid = *mdo2fid(mdd_sobj);
1413 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1414 tfid->f_oid = cfs_fail_val;
1416 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1417 mdd_object_type(mdd_sobj), name, handle);
1419 mdo_ref_del(env, mdd_sobj, handle);
1420 GOTO(out_unlock, rc);
1423 la->la_valid = LA_CTIME | LA_MTIME;
1424 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1426 GOTO(out_unlock, rc);
1428 la->la_valid = LA_CTIME;
1429 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1431 /* Note: The failure of links_add should not cause the
1432 * link failure, so do not check return value. */
1433 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1434 lname, handle, ldata, 0);
1438 mdd_write_unlock(env, mdd_sobj);
1440 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1441 mdo2fid(mdd_tobj), NULL, NULL,
1442 lname, NULL, handle);
1444 rc = mdd_trans_stop(env, mdd, rc, handle);
1445 if (is_vmalloc_addr(ldata->ld_buf))
1446 /* if we vmalloced a large buffer drop it */
1447 lu_buf_free(ldata->ld_buf);
1452 static int mdd_mark_orphan_object(const struct lu_env *env,
1453 struct mdd_object *obj, struct thandle *handle,
1456 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1459 if (!S_ISDIR(mdd_object_type(obj)))
1462 attr->la_valid = LA_FLAGS;
1463 attr->la_flags = LUSTRE_ORPHAN_FL;
1466 rc = mdo_declare_attr_set(env, obj, attr, handle);
1468 rc = mdo_attr_set(env, obj, attr, handle);
1473 static int mdd_declare_finish_unlink(const struct lu_env *env,
1474 struct mdd_object *obj,
1475 struct thandle *handle)
1479 /* Sigh, we do not know if the unlink object will become orphan in
1480 * declare phase, but fortunately the flags here does not matter
1481 * in current declare implementation */
1482 rc = mdd_mark_orphan_object(env, obj, handle, true);
1486 rc = mdo_declare_destroy(env, obj, handle);
1490 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1494 return mdd_declare_links_del(env, obj, handle);
1497 /* caller should take a lock before calling */
1498 int mdd_finish_unlink(const struct lu_env *env,
1499 struct mdd_object *obj, struct md_attr *ma,
1500 const struct mdd_object *pobj,
1501 const struct lu_name *lname,
1505 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1508 LASSERT(mdd_write_locked(env, obj) != 0);
1510 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1511 /* add new orphan and the object
1512 * will be deleted during mdd_close() */
1513 obj->mod_flags |= DEAD_OBJ;
1514 if (obj->mod_count) {
1515 rc = mdd_orphan_insert(env, obj, th);
1517 CDEBUG(D_HA, "Object "DFID" is inserted into "
1518 "orphan list, open count = %d\n",
1519 PFID(mdd_object_fid(obj)),
1522 CERROR("Object "DFID" fail to be an orphan, "
1523 "open count = %d, maybe cause failed "
1525 PFID(mdd_object_fid(obj)),
1528 /* mark object as an orphan here, not
1529 * before mdd_orphan_insert() as racing
1530 * mdd_la_get() may propagate ORPHAN_OBJ
1531 * causing the asserition */
1532 rc = mdd_mark_orphan_object(env, obj, th, false);
1534 rc = mdo_destroy(env, obj, th);
1536 } else if (!is_dir) {
1537 /* old files may not have link ea; ignore errors */
1538 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1546 * has mdd_write_lock on cobj already, but not on pobj yet
1548 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1549 const struct lu_attr *pattr,
1550 struct mdd_object *cobj,
1551 const struct lu_attr *cattr)
1556 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1561 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1562 struct mdd_object *p, struct mdd_object *c,
1563 const struct lu_name *name, struct md_attr *ma,
1564 struct thandle *handle, int no_name, int is_dir)
1566 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1569 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1570 if (likely(no_name == 0)) {
1571 rc = mdo_declare_index_delete(env, p, name->ln_name,
1578 rc = mdo_declare_ref_del(env, p, handle);
1584 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1585 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1586 la->la_valid = LA_CTIME | LA_MTIME;
1587 rc = mdo_declare_attr_set(env, p, la, handle);
1592 rc = mdo_declare_ref_del(env, c, handle);
1596 rc = mdo_declare_ref_del(env, c, handle);
1600 la->la_valid = LA_CTIME;
1601 rc = mdo_declare_attr_set(env, c, la, handle);
1605 rc = mdd_declare_finish_unlink(env, c, handle);
1609 /* FIXME: need changelog for remove entry */
1610 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1618 * test if a file has an HSM archive
1619 * if HSM attributes are not found in ma update them from
1622 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1623 struct mdd_object *obj,
1628 if (!(ma->ma_valid & MA_HSM)) {
1629 /* no HSM MD provided, read xattr */
1630 struct lu_buf *hsm_buf;
1631 const size_t buflen = sizeof(struct hsm_attrs);
1634 hsm_buf = mdd_buf_get(env, NULL, 0);
1635 lu_buf_alloc(hsm_buf, buflen);
1636 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1637 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1638 lu_buf_free(hsm_buf);
1642 ma->ma_valid |= MA_HSM;
1644 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1650 * Delete name entry and the object.
1651 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1652 * does not exist for this object, and it could only happen during resending
1653 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1654 * is also needed in this case(needed by changelog), so we have to add another
1655 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1656 * the ENOENT failure should be able to be fixed by redo mechanism.
1658 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1659 struct md_object *cobj, const struct lu_name *lname,
1660 struct md_attr *ma, int no_name)
1662 const char *name = lname->ln_name;
1663 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1664 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1665 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1666 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1667 struct mdd_object *mdd_cobj = NULL;
1668 struct mdd_device *mdd = mdo2mdd(pobj);
1669 struct thandle *handle;
1670 int rc, is_dir = 0, cl_flags = 0;
1673 /* let shutdown to start */
1674 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1676 /* cobj == NULL means only delete name entry */
1677 if (likely(cobj != NULL)) {
1678 mdd_cobj = md2mdd_obj(cobj);
1679 if (mdd_object_exists(mdd_cobj) == 0)
1683 rc = mdd_la_get(env, mdd_pobj, pattr);
1687 if (likely(mdd_cobj != NULL)) {
1689 rc = mdd_la_get(env, mdd_cobj, cattr);
1693 is_dir = S_ISDIR(cattr->la_mode);
1694 /* search for an existing archive.
1695 * we should check ahead as the object
1696 * can be destroyed in this transaction */
1697 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1698 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1701 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1705 handle = mdd_trans_create(env, mdd);
1707 RETURN(PTR_ERR(handle));
1709 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1710 lname, ma, handle, no_name, is_dir);
1714 rc = mdd_trans_start(env, mdd, handle);
1718 if (likely(mdd_cobj != NULL))
1719 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1721 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1722 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1727 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1728 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1729 GOTO(cleanup, rc = 0);
1731 if (likely(mdd_cobj != NULL)) {
1732 rc = mdo_ref_del(env, mdd_cobj, handle);
1734 __mdd_index_insert_only(env, mdd_pobj,
1736 mdd_object_type(mdd_cobj),
1743 mdo_ref_del(env, mdd_cobj, handle);
1745 /* fetch updated nlink */
1746 rc = mdd_la_get(env, mdd_cobj, cattr);
1751 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1752 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1754 la->la_valid = LA_CTIME | LA_MTIME;
1755 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1759 /* Enough for only unlink the entry */
1760 if (unlikely(mdd_cobj == NULL))
1763 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1764 /* update ctime of an unlinked file only if it is still
1765 * opened or a link still exists */
1766 la->la_valid = LA_CTIME;
1767 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1772 /* XXX: this transfer to ma will be removed with LOD/OSP */
1773 ma->ma_attr = *cattr;
1774 ma->ma_valid |= MA_INODE;
1775 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1779 /* fetch updated nlink */
1780 rc = mdd_la_get(env, mdd_cobj, cattr);
1781 /* if object is removed then we can't get its attrs,
1783 if (rc == -ENOENT) {
1784 cattr->la_nlink = 0;
1788 if (cattr->la_nlink == 0) {
1789 ma->ma_attr = *cattr;
1790 ma->ma_valid |= MA_INODE;
1795 if (likely(mdd_cobj != NULL))
1796 mdd_write_unlock(env, mdd_cobj);
1799 if (cattr->la_nlink == 0)
1800 cl_flags |= CLF_UNLINK_LAST;
1802 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1804 rc = mdd_changelog_ns_store(env, mdd,
1805 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1806 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1811 rc = mdd_trans_stop(env, mdd, rc, handle);
1817 * The permission has been checked when obj created, no need check again.
1819 static int mdd_cd_sanity_check(const struct lu_env *env,
1820 struct mdd_object *obj)
1825 if (!obj || mdd_is_dead_obj(obj))
1831 static int mdd_create_data(const struct lu_env *env,
1832 struct md_object *pobj,
1833 struct md_object *cobj,
1834 const struct md_op_spec *spec,
1837 struct mdd_device *mdd = mdo2mdd(cobj);
1838 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1839 struct mdd_object *son = md2mdd_obj(cobj);
1840 struct thandle *handle;
1841 const struct lu_buf *buf;
1842 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1843 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1847 rc = mdd_cd_sanity_check(env, son);
1851 if (!md_should_create(spec->sp_cr_flags))
1855 * there are following use cases for this function:
1856 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1857 * striping can be specified or not
1860 rc = mdd_la_get(env, son, attr);
1864 /* calling ->ah_make_hint() is used to transfer information from parent */
1865 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1867 handle = mdd_trans_create(env, mdd);
1869 GOTO(out_free, rc = PTR_ERR(handle));
1872 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1873 * Should this be fixed?
1875 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1876 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1877 spec->sp_cr_flags, spec->no_create);
1879 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1880 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1881 spec->u.sp_ea.eadatalen);
1886 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1887 XATTR_NAME_LOV, 0, handle);
1891 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1896 rc = mdd_trans_start(env, mdd, handle);
1900 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1906 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1909 rc = mdd_trans_stop(env, mdd, rc, handle);
1915 static int mdd_declare_object_initialize(const struct lu_env *env,
1916 struct mdd_object *parent,
1917 struct mdd_object *child,
1918 const struct lu_attr *attr,
1919 struct thandle *handle)
1924 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1925 if (!S_ISDIR(attr->la_mode))
1928 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1933 rc = mdo_declare_ref_add(env, child, handle);
1937 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1943 static int mdd_object_initialize(const struct lu_env *env,
1944 const struct lu_fid *pfid,
1945 struct mdd_object *child,
1946 struct lu_attr *attr,
1947 struct thandle *handle)
1952 if (S_ISDIR(attr->la_mode)) {
1953 /* Add "." and ".." for newly created dir */
1954 mdo_ref_add(env, child, handle);
1955 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1956 S_IFDIR, dot, handle);
1958 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1961 mdo_ref_del(env, child, handle);
1968 * This function checks whether it can create a file/dir under the
1969 * directory(@pobj). The directory(@pobj) is not being locked by
1972 * \param[in] env execution environment
1973 * \param[in] pobj the directory to create files
1974 * \param[in] pattr the attributes of the directory
1975 * \param[in] lname the name of the created file/dir
1976 * \param[in] cattr the attributes of the file/dir
1977 * \param[in] spec create specification
1979 * \retval = 0 it is allowed to create file/dir under
1981 * \retval negative error not allowed to create file/dir
1982 * under the directory
1984 static int mdd_create_sanity_check(const struct lu_env *env,
1985 struct md_object *pobj,
1986 const struct lu_attr *pattr,
1987 const struct lu_name *lname,
1988 struct lu_attr *cattr,
1989 struct md_op_spec *spec)
1991 struct mdd_thread_info *info = mdd_env_info(env);
1992 struct lu_fid *fid = &info->mti_fid;
1993 struct mdd_object *obj = md2mdd_obj(pobj);
1994 struct mdd_device *m = mdo2mdd(pobj);
1995 bool check_perm = true;
2000 if (mdd_is_dead_obj(obj))
2004 * In some cases this lookup is not needed - we know before if name
2005 * exists or not because MDT performs lookup for it.
2006 * name length check is done in lookup.
2008 if (spec->sp_cr_lookup) {
2010 * Check if the name already exist, though it will be checked in
2011 * _index_insert also, for avoiding rolling back if exists
2014 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2015 MAY_WRITE | MAY_EXEC);
2017 RETURN(rc ? : -EEXIST);
2019 /* Permission is already being checked in mdd_lookup */
2023 if (S_ISDIR(cattr->la_mode) &&
2024 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2025 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2026 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2028 if (!lmv_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2029 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2031 CERROR("%s: invalid lmv_user_md: magic = %x, "
2032 "stripe_offset = %d, stripe_count = %u: "
2033 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2034 le32_to_cpu(lum->lum_magic),
2035 (int)le32_to_cpu(lum->lum_stripe_offset),
2036 le32_to_cpu(lum->lum_stripe_count), rc);
2041 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2046 if (pattr->la_mode & S_ISGID) {
2047 struct lu_ucred *uc = lu_ucred(env);
2049 cattr->la_gid = pattr->la_gid;
2051 /* Directories are special, and always inherit S_ISGID */
2052 if (S_ISDIR(cattr->la_mode)) {
2053 cattr->la_mode |= S_ISGID;
2054 cattr->la_valid |= LA_MODE;
2055 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2056 == (S_ISGID | S_IXGRP) &&
2057 !lustre_in_group_p(uc,
2058 (cattr->la_valid & LA_GID) ?
2059 cattr->la_gid : pattr->la_gid) &&
2060 !md_capable(uc, CFS_CAP_FSETID)) {
2061 cattr->la_mode &= ~S_ISGID;
2062 cattr->la_valid |= LA_MODE;
2066 /* Inherit project ID from parent directory */
2067 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2068 cattr->la_projid = pattr->la_projid;
2069 if (S_ISDIR(cattr->la_mode)) {
2070 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2071 cattr->la_valid |= LA_FLAGS;
2073 cattr->la_valid |= LA_PROJID;
2076 rc = mdd_name_check(m, lname);
2080 switch (cattr->la_mode & S_IFMT) {
2082 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2084 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2085 RETURN(-ENAMETOOLONG);
2104 static int mdd_declare_create_object(const struct lu_env *env,
2105 struct mdd_device *mdd,
2106 struct mdd_object *p, struct mdd_object *c,
2107 struct lu_attr *attr,
2108 struct thandle *handle,
2109 const struct md_op_spec *spec,
2110 struct lu_buf *def_acl_buf,
2111 struct lu_buf *acl_buf,
2112 struct lu_buf *hsm_buf,
2113 struct dt_allocation_hint *hint)
2115 const struct lu_buf *buf;
2118 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2123 #ifdef CONFIG_FS_POSIX_ACL
2124 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2125 /* if dir, then can inherit default ACl */
2126 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2127 XATTR_NAME_ACL_DEFAULT,
2133 if (acl_buf && acl_buf->lb_len > 0) {
2134 rc = mdo_declare_attr_set(env, c, attr, handle);
2138 rc = mdo_declare_xattr_set(env, c, acl_buf,
2139 XATTR_NAME_ACL_ACCESS, 0, handle);
2144 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2148 /* replay case, create LOV EA from client data */
2149 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2150 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2151 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2152 spec->u.sp_ea.eadatalen);
2153 rc = mdo_declare_xattr_set(env, c, buf,
2154 S_ISDIR(attr->la_mode) ?
2155 XATTR_NAME_LMV : XATTR_NAME_LOV,
2160 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2161 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2169 if (S_ISLNK(attr->la_mode)) {
2170 const char *target_name = spec->u.sp_symname;
2171 int sym_len = strlen(target_name);
2172 const struct lu_buf *buf;
2174 buf = mdd_buf_get_const(env, target_name, sym_len);
2175 rc = dt_declare_record_write(env, mdd_object_child(c),
2181 if (spec->sp_cr_file_secctx_name != NULL) {
2182 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2183 spec->sp_cr_file_secctx_size);
2184 rc = mdo_declare_xattr_set(env, c, buf,
2185 spec->sp_cr_file_secctx_name, 0,
2194 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2195 struct mdd_object *p, struct mdd_object *c,
2196 const struct lu_name *name,
2197 struct lu_attr *attr,
2198 struct thandle *handle,
2199 const struct md_op_spec *spec,
2200 struct linkea_data *ldata,
2201 struct lu_buf *def_acl_buf,
2202 struct lu_buf *acl_buf,
2203 struct lu_buf *hsm_buf,
2204 struct dt_allocation_hint *hint)
2208 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2209 def_acl_buf, acl_buf, hsm_buf, hint);
2213 if (S_ISDIR(attr->la_mode)) {
2214 rc = mdo_declare_ref_add(env, p, handle);
2219 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2220 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2224 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2225 enum changelog_rec_type type;
2227 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2228 name->ln_name, handle);
2232 rc = mdd_declare_links_add(env, c, handle, ldata);
2237 la->la_valid = LA_CTIME | LA_MTIME;
2238 rc = mdo_declare_attr_set(env, p, la, handle);
2242 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2243 S_ISREG(attr->la_mode) ? CL_CREATE :
2244 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2246 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2255 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2256 struct lu_attr *la, struct lu_buf *def_acl_buf,
2257 struct lu_buf *acl_buf)
2262 if (S_ISLNK(la->la_mode)) {
2263 acl_buf->lb_len = 0;
2264 def_acl_buf->lb_len = 0;
2268 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2269 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2270 XATTR_NAME_ACL_DEFAULT);
2271 mdd_read_unlock(env, pobj);
2273 /* If there are default ACL, fix mode/ACL by default ACL */
2274 def_acl_buf->lb_len = rc;
2275 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2276 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2277 acl_buf->lb_len = rc;
2278 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2281 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2282 /* If there are no default ACL, fix mode by mask */
2283 struct lu_ucred *uc = lu_ucred(env);
2285 /* The create triggered by MDT internal events, such as
2286 * LFSCK reset, will not contain valid "uc". */
2287 if (unlikely(uc != NULL))
2288 la->la_mode &= ~uc->uc_umask;
2290 acl_buf->lb_len = 0;
2291 def_acl_buf->lb_len = 0;
2298 * Create a metadata object and initialize it, set acl, xattr.
2300 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2301 struct mdd_object *son, struct lu_attr *attr,
2302 struct md_op_spec *spec, struct lu_buf *acl_buf,
2303 struct lu_buf *def_acl_buf,
2304 struct lu_buf *hsm_buf,
2305 struct dt_allocation_hint *hint,
2306 struct thandle *handle)
2308 const struct lu_buf *buf;
2311 mdd_write_lock(env, son, DT_TGT_CHILD);
2312 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2317 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2318 * created in declare phase, they also needs to be added to master
2319 * object as sub-directory entry. So it has to initialize the master
2320 * object, then set dir striped EA.(in mdo_xattr_set) */
2321 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle);
2323 GOTO(err_destroy, rc);
2326 * in case of replay we just set LOVEA provided by the client
2327 * XXX: I think it would be interesting to try "old" way where
2328 * MDT calls this xattr_set(LOV) in a different transaction.
2329 * probably this way we code can be made better.
2332 /* During creation, there are only a few cases we need do xattr_set to
2334 * 1. regular file: see comments above.
2335 * 2. dir: inherit default striping or pool settings from parent.
2336 * 3. create striped directory with provided stripeEA.
2337 * 4. create striped directory because inherit default layout from the
2340 if (spec->no_create ||
2341 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2342 S_ISDIR(attr->la_mode)) {
2343 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2344 spec->u.sp_ea.eadatalen);
2345 rc = mdo_xattr_set(env, son, buf,
2346 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2350 GOTO(err_destroy, rc);
2353 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2356 memset(&mh, 0, sizeof(mh));
2357 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2358 mh.mh_arch_id = spec->sp_archive_id;
2359 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2360 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2363 GOTO(err_destroy, rc);
2366 #ifdef CONFIG_FS_POSIX_ACL
2367 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2368 S_ISDIR(attr->la_mode)) {
2369 /* set default acl */
2370 rc = mdo_xattr_set(env, son, def_acl_buf,
2371 XATTR_NAME_ACL_DEFAULT, 0,
2374 GOTO(err_destroy, rc);
2376 /* set its own acl */
2377 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2378 rc = mdo_xattr_set(env, son, acl_buf,
2379 XATTR_NAME_ACL_ACCESS,
2382 GOTO(err_destroy, rc);
2386 if (S_ISLNK(attr->la_mode)) {
2387 struct dt_object *dt = mdd_object_child(son);
2388 const char *target_name = spec->u.sp_symname;
2389 int sym_len = strlen(target_name);
2392 buf = mdd_buf_get_const(env, target_name, sym_len);
2393 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2397 GOTO(err_initlized, rc = -EFAULT);
2400 if (spec->sp_cr_file_secctx_name != NULL) {
2401 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2402 spec->sp_cr_file_secctx_size);
2403 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2406 GOTO(err_initlized, rc);
2410 if (unlikely(rc != 0)) {
2412 if (S_ISDIR(attr->la_mode)) {
2413 /* Drop the reference, no need to delete "."/"..",
2414 * because the object to be destroied directly. */
2415 rc2 = mdo_ref_del(env, son, handle);
2419 rc2 = mdo_ref_del(env, son, handle);
2423 mdo_destroy(env, son, handle);
2426 mdd_write_unlock(env, son);
2430 static int mdd_index_delete(const struct lu_env *env,
2431 struct mdd_object *mdd_pobj,
2432 struct lu_attr *cattr,
2433 const struct lu_name *lname)
2435 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2436 struct thandle *handle;
2440 handle = mdd_trans_create(env, mdd);
2442 RETURN(PTR_ERR(handle));
2444 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2449 if (S_ISDIR(cattr->la_mode)) {
2450 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2455 /* Since this will only be used in the error handler path,
2456 * Let's set the thandle to be local and not mess the transno */
2457 handle->th_local = 1;
2458 rc = mdd_trans_start(env, mdd, handle);
2462 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2463 S_ISDIR(cattr->la_mode), handle);
2467 rc = mdd_trans_stop(env, mdd, rc, handle);
2473 * Create object and insert it into namespace.
2475 * Two operations have to be performed:
2477 * - an allocation of a new object (->do_create()), and
2478 * - an insertion into a parent index (->dio_insert()).
2480 * Due to locking, operation order is not important, when both are
2481 * successful, *but* error handling cases are quite different:
2483 * - if insertion is done first, and following object creation fails,
2484 * insertion has to be rolled back, but this operation might fail
2485 * also leaving us with dangling index entry.
2487 * - if creation is done first, is has to be undone if insertion fails,
2488 * leaving us with leaked space, which is not good but not fatal.
2490 * It seems that creation-first is simplest solution, but it is sub-optimal
2496 * case, because second mkdir is bound to create object, only to
2497 * destroy it immediately.
2499 * To avoid this follow local file systems that do double lookup:
2501 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2502 * 1. create (mdd_create_object_internal())
2503 * 2. insert (__mdd_index_insert(), lookup again)
2505 * \param[in] pobj parent object
2506 * \param[in] lname name of child being created
2507 * \param[in,out] child child object being created
2508 * \param[in] spec additional create parameters
2509 * \param[in] ma attributes for new child object
2511 * \retval 0 on success
2512 * \retval negative errno on failure
2514 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2515 const struct lu_name *lname, struct md_object *child,
2516 struct md_op_spec *spec, struct md_attr *ma)
2518 struct mdd_thread_info *info = mdd_env_info(env);
2519 struct lu_attr *la = &info->mti_la_for_fix;
2520 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2521 struct mdd_object *son = md2mdd_obj(child);
2522 struct mdd_device *mdd = mdo2mdd(pobj);
2523 struct lu_attr *attr = &ma->ma_attr;
2524 struct thandle *handle;
2525 struct lu_attr *pattr = &info->mti_pattr;
2526 struct lu_buf acl_buf;
2527 struct lu_buf def_acl_buf;
2528 struct lu_buf hsm_buf;
2529 struct linkea_data *ldata = &info->mti_link_data;
2530 const char *name = lname->ln_name;
2531 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2536 rc = mdd_la_get(env, mdd_pobj, pattr);
2540 /* Sanity checks before big job. */
2541 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2545 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2546 GOTO(out_free, rc = -EINPROGRESS);
2548 handle = mdd_trans_create(env, mdd);
2550 GOTO(out_free, rc = PTR_ERR(handle));
2552 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2553 MIN(mdd->mdd_dt_conf.ddp_max_ea_size, XATTR_SIZE_MAX));
2554 acl_buf = info->mti_xattr_buf;
2555 def_acl_buf.lb_buf = info->mti_key;
2556 def_acl_buf.lb_len = sizeof(info->mti_key);
2557 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2561 if (S_ISDIR(attr->la_mode)) {
2562 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2565 * migrate may create 1-stripe directory, so lod_ah_init()
2566 * doesn't adjust stripe count from lmu.
2568 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1)) {
2569 info->mti_lmu = *lmu;
2570 info->mti_lmu.lum_stripe_count = 0;
2571 spec->u.sp_ea.eadata = &info->mti_lmu;
2575 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2577 memset(ldata, 0, sizeof(*ldata));
2578 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2579 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2582 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2583 &tfid, lname, 1, 0, ldata);
2585 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2586 mdd_object_fid(mdd_pobj),
2587 lname, 1, 0, ldata);
2590 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2591 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2593 memset(&hsm_buf, 0, sizeof(hsm_buf));
2594 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2595 if (hsm_buf.lb_buf == NULL)
2596 GOTO(out_stop, rc = -ENOMEM);
2599 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2600 handle, spec, ldata, &def_acl_buf, &acl_buf,
2605 rc = mdd_trans_start(env, mdd, handle);
2609 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2610 &def_acl_buf, &hsm_buf, hint, handle);
2614 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2615 mdd_write_lock(env, son, DT_TGT_CHILD);
2616 son->mod_flags |= VOLATILE_OBJ;
2617 rc = mdd_orphan_insert(env, son, handle);
2618 GOTO(out_volatile, rc);
2620 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2621 attr->la_mode, name, handle);
2623 GOTO(err_created, rc);
2625 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2628 /* update parent directory mtime/ctime */
2630 la->la_valid = LA_CTIME | LA_MTIME;
2631 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2633 GOTO(err_insert, rc);
2639 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2640 rc2 = mdd_orphan_delete(env, son, handle);
2642 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2643 S_ISDIR(attr->la_mode),
2649 mdd_write_lock(env, son, DT_TGT_CHILD);
2650 if (S_ISDIR(attr->la_mode)) {
2651 /* Drop the reference, no need to delete "."/"..",
2652 * because the object is to be destroyed directly. */
2653 rc2 = mdo_ref_del(env, son, handle);
2655 mdd_write_unlock(env, son);
2660 /* For volatile files drop one link immediately, since there is
2661 * no filename in the namespace, and save any error returned. */
2662 rc2 = mdo_ref_del(env, son, handle);
2664 mdd_write_unlock(env, son);
2665 if (unlikely(rc == 0))
2670 /* Don't destroy the volatile object on success */
2671 if (likely(rc != 0))
2672 mdo_destroy(env, son, handle);
2673 mdd_write_unlock(env, son);
2676 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2677 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2678 rc = mdd_changelog_ns_store(env, mdd,
2679 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2680 S_ISREG(attr->la_mode) ? CL_CREATE :
2681 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2682 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2685 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2687 /* If creation fails, it is most likely due to the remote update
2688 * failure, because local transaction will mostly succeed at
2689 * this stage. There is no easy way to rollback all of previous
2690 * updates, so let's remove the object from namespace, and
2691 * LFSCK should handle the orphan object. */
2692 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2693 mdd_index_delete(env, mdd_pobj, attr, lname);
2697 if (is_vmalloc_addr(ldata->ld_buf))
2698 /* if we vmalloced a large buffer drop it */
2699 lu_buf_free(ldata->ld_buf);
2701 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2702 lu_buf_free(&hsm_buf);
2704 /* The child object shouldn't be cached anymore */
2706 set_bit(LU_OBJECT_HEARD_BANSHEE,
2707 &child->mo_lu.lo_header->loh_flags);
2712 * Get locks on parents in proper order
2713 * RETURN: < 0 - error, rename_order if successful
2721 static int mdd_rename_order(const struct lu_env *env,
2722 struct mdd_device *mdd,
2723 struct mdd_object *src_pobj,
2724 const struct lu_attr *pattr,
2725 struct mdd_object *tgt_pobj)
2727 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2731 if (src_pobj == tgt_pobj)
2732 RETURN(MDD_RN_SAME);
2734 /* compared the parent child relationship of src_p&tgt_p */
2735 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2737 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2740 rc = mdd_is_parent(env, mdd, src_pobj, pattr,
2754 /* has not mdd_write{read}_lock on any obj yet. */
2755 static int mdd_rename_sanity_check(const struct lu_env *env,
2756 struct mdd_object *src_pobj,
2757 const struct lu_attr *pattr,
2758 struct mdd_object *tgt_pobj,
2759 const struct lu_attr *tpattr,
2760 struct mdd_object *sobj,
2761 const struct lu_attr *cattr,
2762 struct mdd_object *tobj,
2763 const struct lu_attr *tattr)
2768 /* XXX: when get here, sobj must NOT be NULL,
2769 * the other case has been processed in cld_rename
2770 * before mdd_rename and enable MDS_PERM_BYPASS. */
2774 * If we are using project inheritance, we only allow renames
2775 * into our tree when the project IDs are the same; otherwise
2776 * tree quota mechanism would be circumvented.
2778 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2779 tpattr->la_projid != cattr->la_projid) ||
2780 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2781 (pattr->la_projid != tpattr->la_projid)))
2784 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2788 /* XXX: when get here, "tobj == NULL" means tobj must
2789 * NOT exist (neither on remote MDS, such case has been
2790 * processed in cld_rename before mdd_rename and enable
2792 * So check may_create, but not check may_unlink. */
2794 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2795 (src_pobj != tgt_pobj));
2797 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2798 (src_pobj != tgt_pobj), 1);
2800 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2801 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2806 static int mdd_declare_rename(const struct lu_env *env,
2807 struct mdd_device *mdd,
2808 struct mdd_object *mdd_spobj,
2809 struct mdd_object *mdd_tpobj,
2810 struct mdd_object *mdd_sobj,
2811 struct mdd_object *mdd_tobj,
2812 const struct lu_name *sname,
2813 const struct lu_name *tname,
2815 struct linkea_data *ldata,
2816 struct thandle *handle)
2818 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2821 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2822 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2828 /* name from source dir */
2829 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2833 /* .. from source child */
2834 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2835 /* source child can be directory,
2836 * counted by source dir's nlink */
2837 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2840 if (mdd_spobj != mdd_tpobj) {
2841 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2846 rc = mdo_declare_index_insert(env, mdd_sobj,
2848 S_IFDIR, dotdot, handle);
2853 /* new target child can be directory,
2854 * counted by target dir's nlink */
2855 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2860 la->la_valid = LA_CTIME | LA_MTIME;
2861 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2865 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2869 la->la_valid = LA_CTIME;
2870 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2874 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2879 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2880 mdd_object_type(mdd_sobj),
2881 tname->ln_name, handle);
2885 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2886 /* delete target child in target parent directory */
2887 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2892 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2896 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2897 /* target child can be directory,
2898 * delete "." reference in target child directory */
2899 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2903 /* delete ".." reference in target parent directory */
2904 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2909 la->la_valid = LA_CTIME;
2910 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2914 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2919 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2927 /* src object can be remote that is why we use only fid and type of object */
2928 static int mdd_rename(const struct lu_env *env,
2929 struct md_object *src_pobj, struct md_object *tgt_pobj,
2930 const struct lu_fid *lf, const struct lu_name *lsname,
2931 struct md_object *tobj, const struct lu_name *ltname,
2934 const char *sname = lsname->ln_name;
2935 const char *tname = ltname->ln_name;
2936 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2937 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2938 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2939 struct mdd_device *mdd = mdo2mdd(src_pobj);
2940 struct mdd_object *mdd_sobj = NULL; /* source object */
2941 struct mdd_object *mdd_tobj = NULL;
2942 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2943 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2944 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2945 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2946 struct thandle *handle;
2947 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2948 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2949 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2952 bool tobj_locked = 0;
2953 unsigned cl_flags = 0;
2957 /* let unlink to complete and commit */
2958 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
2961 mdd_tobj = md2mdd_obj(tobj);
2963 mdd_sobj = mdd_object_find(env, mdd, lf);
2964 if (IS_ERR(mdd_sobj))
2965 RETURN(PTR_ERR(mdd_sobj));
2967 rc = mdd_la_get(env, mdd_sobj, cattr);
2969 GOTO(out_pending, rc);
2971 rc = mdd_la_get(env, mdd_spobj, pattr);
2973 GOTO(out_pending, rc);
2976 rc = mdd_la_get(env, mdd_tobj, tattr);
2978 GOTO(out_pending, rc);
2979 /* search for an existing archive.
2980 * we should check ahead as the object
2981 * can be destroyed in this transaction */
2982 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2983 cl_flags |= CLF_RENAME_LAST_EXISTS;
2986 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2988 GOTO(out_pending, rc);
2990 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2991 mdd_sobj, cattr, mdd_tobj, tattr);
2993 GOTO(out_pending, rc);
2995 rc = mdd_name_check(mdd, ltname);
2997 GOTO(out_pending, rc);
2999 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3000 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3002 GOTO(out_pending, rc);
3004 handle = mdd_trans_create(env, mdd);
3006 GOTO(out_pending, rc = PTR_ERR(handle));
3008 memset(ldata, 0, sizeof(*ldata));
3009 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3010 lsname, mdd_object_fid(mdd_tpobj), ltname,
3015 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3016 mdd_tobj, lsname, ltname, ma, ldata, handle);
3020 rc = mdd_trans_start(env, mdd, handle);
3024 is_dir = S_ISDIR(cattr->la_mode);
3026 /* Remove source name from source directory */
3027 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3031 /* "mv dir1 dir2" needs "dir1/.." link update */
3032 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3033 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3035 GOTO(fixup_spobj2, rc);
3037 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3040 GOTO(fixup_spobj, rc);
3043 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3044 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3046 /* tname might been renamed to something else */
3047 GOTO(fixup_spobj, rc);
3050 /* Insert new fid with target name into target dir */
3051 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3054 GOTO(fixup_tpobj, rc);
3056 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3057 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3059 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3060 la->la_valid = LA_CTIME;
3061 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3063 GOTO(fixup_tpobj, rc);
3065 /* Update the linkEA for the source object */
3066 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3067 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3068 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3071 /* Old files might not have EA entry */
3072 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3073 lsname, handle, NULL, 0);
3074 mdd_write_unlock(env, mdd_sobj);
3075 /* We don't fail the transaction if the link ea can't be
3076 updated -- fid2path will use alternate lookup method. */
3079 /* Remove old target object
3080 * For tobj is remote case cmm layer has processed
3081 * and set tobj to NULL then. So when tobj is NOT NULL,
3082 * it must be local one.
3084 if (tobj && mdd_object_exists(mdd_tobj)) {
3085 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3087 if (mdd_is_dead_obj(mdd_tobj)) {
3088 /* shld not be dead, something is wrong */
3089 CERROR("tobj is dead, something is wrong\n");
3093 mdo_ref_del(env, mdd_tobj, handle);
3095 /* Remove dot reference. */
3096 if (S_ISDIR(tattr->la_mode))
3097 mdo_ref_del(env, mdd_tobj, handle);
3100 /* fetch updated nlink */
3101 rc = mdd_la_get(env, mdd_tobj, tattr);
3103 CERROR("%s: Failed to get nlink for tobj "
3105 mdd2obd_dev(mdd)->obd_name,
3106 PFID(tpobj_fid), rc);
3107 GOTO(fixup_tpobj, rc);
3110 la->la_valid = LA_CTIME;
3111 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3113 CERROR("%s: Failed to set ctime for tobj "
3115 mdd2obd_dev(mdd)->obd_name,
3116 PFID(tpobj_fid), rc);
3117 GOTO(fixup_tpobj, rc);
3120 /* XXX: this transfer to ma will be removed with LOD/OSP */
3121 ma->ma_attr = *tattr;
3122 ma->ma_valid |= MA_INODE;
3123 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3126 CERROR("%s: Failed to unlink tobj "
3128 mdd2obd_dev(mdd)->obd_name,
3129 PFID(tpobj_fid), rc);
3130 GOTO(fixup_tpobj, rc);
3133 /* fetch updated nlink */
3134 rc = mdd_la_get(env, mdd_tobj, tattr);
3135 if (rc == -ENOENT) {
3136 /* the object got removed, let's
3137 * return the latest known attributes */
3138 tattr->la_nlink = 0;
3140 } else if (rc != 0) {
3141 CERROR("%s: Failed to get nlink for tobj "
3143 mdd2obd_dev(mdd)->obd_name,
3144 PFID(tpobj_fid), rc);
3145 GOTO(fixup_tpobj, rc);
3147 /* XXX: this transfer to ma will be removed with LOD/OSP */
3148 ma->ma_attr = *tattr;
3149 ma->ma_valid |= MA_INODE;
3151 if (tattr->la_nlink == 0)
3152 cl_flags |= CLF_RENAME_LAST;
3154 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3157 la->la_valid = LA_CTIME | LA_MTIME;
3158 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3160 GOTO(fixup_tpobj, rc);
3162 if (mdd_spobj != mdd_tpobj) {
3163 la->la_valid = LA_CTIME | LA_MTIME;
3164 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3166 GOTO(fixup_tpobj, rc);
3173 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3175 CWARN("tp obj fix error %d\n",rc2);
3177 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3178 !mdd_is_dead_obj(mdd_tobj)) {
3180 mdo_ref_add(env, mdd_tobj, handle);
3182 mdo_ref_add(env, mdd_tobj, handle);
3185 rc2 = __mdd_index_insert(env, mdd_tpobj,
3187 mdd_object_type(mdd_tobj),
3190 CWARN("tp obj fix error: rc = %d\n", rc2);
3195 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3196 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3198 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3199 mdd2obd_dev(mdd)->obd_name, rc2);
3202 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3205 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3206 mdd2obd_dev(mdd)->obd_name, rc2);
3211 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3212 mdd_object_type(mdd_sobj), sname,
3215 CWARN("sp obj fix error: rc = %d\n", rc2);
3220 mdd_write_unlock(env, mdd_tobj);
3223 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3224 mdd_tobj, tpobj_fid, lf, spobj_fid,
3225 ltname, lsname, handle);
3228 rc = mdd_trans_stop(env, mdd, rc, handle);
3231 mdd_object_put(env, mdd_sobj);
3236 * Check whether we should migrate the file/dir
3238 * < 0 permission check failed or other error.
3239 * = 0 the file can be migrated.
3241 static int mdd_migrate_sanity_check(const struct lu_env *env,
3242 struct mdd_device *mdd,
3243 struct mdd_object *spobj,
3244 struct mdd_object *tpobj,
3245 struct mdd_object *sobj,
3246 struct mdd_object *tobj,
3247 const struct lu_attr *spattr,
3248 const struct lu_attr *tpattr,
3249 const struct lu_attr *attr)
3255 if (!mdd_object_remote(sobj)) {
3256 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3257 if (sobj->mod_count > 0) {
3258 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3259 mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(sobj)),
3261 mdd_read_unlock(env, sobj);
3264 mdd_read_unlock(env, sobj);
3267 if (mdd_object_exists(tobj))
3270 rc = mdd_rename_sanity_check(env, spobj, spattr, tpobj, tpattr, sobj,
3275 typedef int (*mdd_dir_stripe_cb)(const struct lu_env *env,
3276 struct mdd_object *obj,
3277 struct mdd_object *stripe,
3278 const struct lu_buf *lmv_buf,
3279 const struct lu_buf *lmu_buf,
3281 struct thandle *handle);
3283 static int mdd_dir_declare_delete_stripe(const struct lu_env *env,
3284 struct mdd_object *obj,
3285 struct mdd_object *stripe,
3286 const struct lu_buf *lmv_buf,
3287 const struct lu_buf *lmu_buf,
3289 struct thandle *handle)
3291 struct mdd_thread_info *info = mdd_env_info(env);
3292 char *stripe_name = info->mti_name;
3293 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3296 if (index < le32_to_cpu(lmu->lum_stripe_count))
3299 rc = mdo_declare_index_delete(env, stripe, dotdot, handle);
3303 snprintf(stripe_name, sizeof(info->mti_name), DFID":%d",
3304 PFID(mdd_object_fid(stripe)), index);
3306 rc = mdo_declare_index_delete(env, obj, stripe_name, handle);
3310 rc = mdo_declare_ref_del(env, obj, handle);
3315 /* delete stripe from its master object namespace */
3316 static int mdd_dir_delete_stripe(const struct lu_env *env,
3317 struct mdd_object *obj,
3318 struct mdd_object *stripe,
3319 const struct lu_buf *lmv_buf,
3320 const struct lu_buf *lmu_buf,
3322 struct thandle *handle)
3324 struct mdd_thread_info *info = mdd_env_info(env);
3325 char *stripe_name = info->mti_name;
3326 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3327 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3328 __u32 del_offset = le32_to_cpu(lmu->lum_stripe_count);
3333 /* local dir will delete via LOD */
3334 LASSERT(mdd_object_remote(obj));
3335 LASSERT(del_offset < le32_to_cpu(lmv->lmv_stripe_count));
3337 if (index < del_offset)
3340 mdd_write_lock(env, stripe, DT_SRC_CHILD);
3341 rc = __mdd_index_delete_only(env, stripe, dotdot, handle);
3345 snprintf(stripe_name, sizeof(info->mti_name), DFID":%d",
3346 PFID(mdd_object_fid(stripe)), index);
3348 rc = __mdd_index_delete_only(env, obj, stripe_name, handle);
3352 rc = mdo_ref_del(env, obj, handle);
3355 mdd_write_unlock(env, stripe);
3360 static int mdd_dir_declare_destroy_stripe(const struct lu_env *env,
3361 struct mdd_object *obj,
3362 struct mdd_object *stripe,
3363 const struct lu_buf *lmv_buf,
3364 const struct lu_buf *lmu_buf,
3366 struct thandle *handle)
3368 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3369 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3372 if (index < shrink_offset) {
3373 if (shrink_offset < 2)
3375 return mdo_declare_xattr_set(env, stripe, lmv_buf,
3376 XATTR_NAME_LMV".set", 0, handle);
3379 rc = mdo_declare_ref_del(env, stripe, handle);
3383 rc = mdo_declare_destroy(env, stripe, handle);
3388 static int mdd_dir_destroy_stripe(const struct lu_env *env,
3389 struct mdd_object *obj,
3390 struct mdd_object *stripe,
3391 const struct lu_buf *lmv_buf,
3392 const struct lu_buf *lmu_buf,
3394 struct thandle *handle)
3396 struct mdd_thread_info *info = mdd_env_info(env);
3397 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3398 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3399 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3404 /* update remaining stripes' LMV */
3405 if (index < shrink_offset) {
3406 struct lmv_mds_md_v1 *slave_lmv;
3407 struct lu_buf slave_buf = {
3408 .lb_buf = &info->mti_lmv.lmv_md_v1,
3409 .lb_len = sizeof(*slave_lmv)
3411 __u32 version = le32_to_cpu(lmv->lmv_layout_version);
3413 /* if dir will be shrunk to 1-stripe, don't update */
3414 if (shrink_offset < 2)
3417 slave_lmv = slave_buf.lb_buf;
3418 memset(slave_lmv, 0, sizeof(*slave_lmv));
3419 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
3420 slave_lmv->lmv_stripe_count = lmu->lum_stripe_count;
3421 slave_lmv->lmv_master_mdt_index = cpu_to_le32(index);
3422 slave_lmv->lmv_hash_type = lmv->lmv_hash_type &
3423 cpu_to_le32(LMV_HASH_TYPE_MASK);
3424 slave_lmv->lmv_layout_version = cpu_to_le32(++version);
3426 rc = mdo_xattr_set(env, stripe, &slave_buf,
3427 XATTR_NAME_LMV".set", 0, handle);
3431 mdd_write_lock(env, stripe, DT_SRC_CHILD);
3432 rc = mdo_ref_del(env, stripe, handle);
3434 rc = mdo_destroy(env, stripe, handle);
3435 mdd_write_unlock(env, stripe);
3440 static int mdd_shrink_stripe_is_empty(const struct lu_env *env,
3441 struct mdd_object *obj,
3442 struct mdd_object *stripe,
3443 const struct lu_buf *lmv_buf,
3444 const struct lu_buf *lmu_buf,
3446 struct thandle *handle)
3448 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3449 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3451 /* the default value is 0, but it means 1 */
3455 if (index < shrink_offset)
3458 return mdd_dir_is_empty(env, stripe);
3462 * iterate stripes of striped directory on remote MDT, local striped directory
3463 * is accessed via LOD.
3465 static int mdd_dir_iterate_stripes(const struct lu_env *env,
3466 struct mdd_object *obj,
3467 const struct lu_buf *lmv_buf,
3468 const struct lu_buf *lmu_buf,
3469 struct thandle *handle,
3470 mdd_dir_stripe_cb cb)
3472 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
3473 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3474 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3475 struct mdd_object *stripe;
3483 for (i = 0; i < le32_to_cpu(lmv->lmv_stripe_count); i++) {
3484 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[i]);
3485 if (!fid_is_sane(fid))
3488 stripe = mdd_object_find(env, mdd, fid);
3490 RETURN(PTR_ERR(stripe));
3492 rc = cb(env, obj, stripe, lmv_buf, lmu_buf, i, handle);
3493 mdd_object_put(env, stripe);
3501 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3502 struct mdd_object *obj,
3503 const struct lu_buf *buf,
3505 int fl, struct thandle *handle);
3507 /* iterate xattrs, but ignore LMA, LMV, and LINKEA if 'skip_linkea' is set. */
3508 static int mdd_iterate_xattrs(const struct lu_env *env,
3509 struct mdd_object *sobj,
3510 struct mdd_object *tobj,
3512 struct thandle *handle,
3515 struct mdd_thread_info *info = mdd_env_info(env);
3517 struct lu_buf list_xbuf;
3518 struct lu_buf xbuf = { NULL };
3527 /* retrieve xattr list from the old object */
3528 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3529 if (list_xsize == -ENODATA)
3535 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3536 if (info->mti_big_buf.lb_buf == NULL)
3539 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3540 list_xbuf.lb_len = list_xsize;
3541 rc = mdo_xattr_list(env, sobj, &list_xbuf);
3547 xname = list_xbuf.lb_buf;
3549 xlen = strnlen(xname, rem - 1) + 1;
3550 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3551 strcmp(XATTR_NAME_LMV, xname) == 0)
3555 strcmp(XATTR_NAME_LINK, xname) == 0)
3558 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3559 if (xsize == -ENODATA)
3562 GOTO(out, rc = xsize);
3564 lu_buf_check_and_alloc(&xbuf, xsize);
3565 if (xbuf.lb_buf == NULL)
3566 GOTO(out, rc = -ENOMEM);
3568 rc = mdo_xattr_get(env, sobj, &xbuf, xname);
3575 rc = cb(env, tobj, &xbuf, xname, 0, handle);
3576 if (unlikely(rc == -ENOSPC &&
3577 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3578 rc = linkea_overflow_shrink(
3579 (struct linkea_data *)(xbuf.lb_buf));
3580 if (likely(rc > 0)) {
3598 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3599 struct mdd_object *sobj,
3600 struct mdd_object *tobj,
3601 const struct lu_name *sname,
3602 const struct lu_fid *sfid,
3603 const struct lu_name *lname,
3604 const struct lu_fid *fid,
3606 struct thandle *handle);
3608 static int mdd_declare_update_link(const struct lu_env *env,
3609 struct mdd_object *sobj,
3610 struct mdd_object *tobj,
3611 const struct lu_name *tname,
3612 const struct lu_fid *tpfid,
3613 const struct lu_name *lname,
3614 const struct lu_fid *fid,
3616 struct thandle *handle)
3618 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3619 struct mdd_object *pobj;
3623 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3624 !strcmp(tname->ln_name, lname->ln_name))
3627 pobj = mdd_object_find(env, mdd, fid);
3629 return PTR_ERR(pobj);
3632 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3634 rc = mdo_declare_index_insert(env, pobj, mdo2fid(tobj),
3635 mdd_object_type(sobj),
3636 lname->ln_name, handle);
3637 mdd_object_put(env, pobj);
3641 rc = mdo_declare_ref_add(env, tobj, handle);
3645 rc = mdo_declare_ref_del(env, sobj, handle);
3649 static int mdd_update_link(const struct lu_env *env,
3650 struct mdd_object *sobj,
3651 struct mdd_object *tobj,
3652 const struct lu_name *tname,
3653 const struct lu_fid *tpfid,
3654 const struct lu_name *lname,
3655 const struct lu_fid *fid,
3657 struct thandle *handle)
3659 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3660 struct mdd_object *pobj;
3665 LASSERT(lu_name_is_valid(lname));
3668 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3669 !strncmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3672 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3673 PFID(fid), PNAME(lname), PFID(mdo2fid(tobj)));
3675 pobj = mdd_object_find(env, mdd, fid);
3677 CWARN("%s: cannot find obj "DFID": %ld\n",
3678 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3679 RETURN(PTR_ERR(pobj));
3682 if (!mdd_object_exists(pobj)) {
3683 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3684 mdd_object_put(env, pobj);
3688 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3689 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3691 rc = __mdd_index_insert_only(env, pobj, mdo2fid(tobj),
3692 mdd_object_type(sobj),
3693 lname->ln_name, handle);
3694 mdd_write_unlock(env, pobj);
3695 mdd_object_put(env, pobj);
3699 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3700 rc = mdo_ref_add(env, tobj, handle);
3701 mdd_write_unlock(env, tobj);
3705 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3706 rc = mdo_ref_del(env, sobj, handle);
3707 mdd_write_unlock(env, sobj);
3712 static inline int mdd_fld_lookup(const struct lu_env *env,
3713 struct mdd_device *mdd,
3714 const struct lu_fid *fid,
3717 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
3718 struct seq_server_site *ss;
3721 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3723 range->lsr_flags = LU_SEQ_RANGE_MDT;
3724 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3728 *mdt_index = range->lsr_index;
3733 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3734 struct mdd_object *sobj,
3735 struct mdd_object *tobj,
3736 const struct lu_name *tname,
3737 const struct lu_fid *tpfid,
3738 const struct lu_name *lname,
3739 const struct lu_fid *fid,
3741 struct thandle *handle)
3743 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3744 __u32 source_mdt_index = *(__u32 *)opaque;
3745 __u32 link_mdt_index;
3751 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3752 !strcmp(tname->ln_name, lname->ln_name))
3755 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3759 RETURN(link_mdt_index == source_mdt_index);
3762 static int mdd_iterate_linkea(const struct lu_env *env,
3763 struct mdd_object *sobj,
3764 struct mdd_object *tobj,
3765 const struct lu_name *tname,
3766 const struct lu_fid *tpfid,
3767 struct linkea_data *ldata,
3769 struct thandle *handle,
3772 struct mdd_thread_info *info = mdd_env_info(env);
3773 char *filename = info->mti_name;
3774 struct lu_name lname;
3781 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3782 linkea_next_entry(ldata)) {
3783 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3786 /* Note: lname might miss \0 at the end */
3787 snprintf(filename, sizeof(info->mti_name), "%.*s",
3788 lname.ln_namelen, lname.ln_name);
3789 lname.ln_name = filename;
3791 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3793 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3801 * Prepare linkea, and check whether file needs migrate: if source still has
3802 * link on source MDT, no need to migrate, just update namespace on source and
3805 * \retval 0 do migrate
3806 * \retval 1 don't migrate
3807 * \retval -errno on failure
3809 static int migrate_linkea_prepare(const struct lu_env *env,
3810 struct mdd_device *mdd,
3811 struct mdd_object *spobj,
3812 struct mdd_object *tpobj,
3813 struct mdd_object *sobj,
3814 const struct lu_name *lname,
3815 const struct lu_attr *attr,
3816 struct linkea_data *ldata)
3818 __u32 source_mdt_index;
3823 memset(ldata, 0, sizeof(*ldata));
3824 rc = mdd_linkea_prepare(env, sobj, mdo2fid(spobj), lname,
3825 mdo2fid(tpobj), lname, 1, 0, ldata);
3830 * Then it will check if the file should be migrated. If the file has
3831 * mulitple links, we only need migrate the file if all of its entries
3832 * has been migrated to the remote MDT.
3834 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3837 /* If there are still links locally, don't migrate this file */
3838 LASSERT(ldata->ld_leh != NULL);
3841 * If linkEA is overflow, it means there are some unknown name entries
3842 * under unknown parents, which will prevent the migration.
3844 if (unlikely(ldata->ld_leh->leh_overflow_time))
3847 rc = mdd_fld_lookup(env, mdd, mdo2fid(sobj), &source_mdt_index);
3851 rc = mdd_iterate_linkea(env, sobj, NULL, lname, mdo2fid(tpobj), ldata,
3852 &source_mdt_index, NULL,
3853 mdd_is_link_on_source_mdt);
3857 static int mdd_dir_declare_layout_delete(const struct lu_env *env,
3858 struct mdd_object *obj,
3859 const struct lu_buf *lmv_buf,
3860 const struct lu_buf *lmu_buf,
3861 struct thandle *handle)
3865 if (!lmv_buf->lb_buf)
3866 rc = mdo_declare_index_delete(env, obj, dotdot, handle);
3867 else if (mdd_object_remote(obj))
3868 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, lmu_buf, handle,
3869 mdd_dir_declare_delete_stripe);
3871 rc = mdo_declare_xattr_set(env, obj, lmu_buf,
3872 XATTR_NAME_LMV".del", 0, handle);
3877 static int mdd_dir_layout_delete(const struct lu_env *env,
3878 struct mdd_object *obj,
3879 const struct lu_buf *lmv_buf,
3880 const struct lu_buf *lmu_buf,
3881 struct thandle *handle)
3887 mdd_write_lock(env, obj, DT_SRC_PARENT);
3888 if (!lmv_buf->lb_buf)
3890 rc = __mdd_index_delete_only(env, obj, dotdot, handle);
3891 else if (mdd_object_remote(obj))
3892 /* striped, but remote */
3893 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, lmu_buf, handle,
3894 mdd_dir_delete_stripe);
3896 rc = mdo_xattr_set(env, obj, lmu_buf, XATTR_NAME_LMV".del", 0,
3898 mdd_write_unlock(env, obj);
3903 static int mdd_declare_migrate_create(const struct lu_env *env,
3904 struct mdd_object *tpobj,
3905 struct mdd_object *sobj,
3906 struct mdd_object *tobj,
3907 const struct lu_name *lname,
3908 struct lu_attr *attr,
3909 struct lu_buf *sbuf,
3910 struct linkea_data *ldata,
3911 struct md_op_spec *spec,
3912 struct dt_allocation_hint *hint,
3913 struct thandle *handle)
3915 struct mdd_thread_info *info = mdd_env_info(env);
3916 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3919 if (S_ISDIR(attr->la_mode)) {
3920 struct lu_buf lmu_buf = { NULL };
3923 struct lmv_user_md *lmu = &info->mti_lmv.lmv_user_md;
3925 lmu->lum_stripe_count = 0;
3926 lmu_buf.lb_buf = lmu;
3927 lmu_buf.lb_len = sizeof(*lmu);
3930 rc = mdd_dir_declare_layout_delete(env, sobj, sbuf, &lmu_buf,
3936 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LMV,
3943 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3944 lname, attr, handle, spec, ldata, NULL, NULL,
3949 if (S_ISDIR(attr->la_mode) && mdd_dir_is_empty(env, sobj) != 0) {
3952 * if sobj is not striped, fake a 1-stripe LMV, which
3953 * will be used to generate a compound LMV for tobj.
3955 LASSERT(sizeof(info->mti_key) >
3956 lmv_mds_md_size(1, LMV_MAGIC_V1));
3957 lmv = (typeof(lmv))info->mti_key;
3958 memset(lmv, 0, sizeof(*lmv));
3959 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3960 lmv->lmv_stripe_count = cpu_to_le32(1);
3961 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdo2fid(sobj));
3963 sbuf->lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3965 rc = mdo_declare_xattr_set(env, tobj, sbuf,
3966 XATTR_NAME_LMV".add", 0,
3968 sbuf->lb_buf = NULL;
3971 rc = mdo_declare_xattr_set(env, tobj, sbuf,
3972 XATTR_NAME_LMV".add", 0,
3980 * tobj mode will be used in lod_declare_xattr_set(), but it's not
3981 * createb yet, copy from sobj.
3983 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
3984 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
3985 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
3987 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle,
3988 mdo_declare_xattr_set);
3992 if (S_ISREG(attr->la_mode)) {
3993 struct lu_buf fid_buf;
3995 handle->th_complex = 1;
3997 /* target may be remote, update PFID via sobj. */
3998 fid_buf.lb_buf = (void *)mdo2fid(tobj);
3999 fid_buf.lb_len = sizeof(struct lu_fid);
4000 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
4005 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4010 if (!S_ISDIR(attr->la_mode)) {
4011 rc = mdd_iterate_linkea(env, sobj, tobj, lname, mdo2fid(tpobj),
4012 ldata, NULL, handle,
4013 mdd_declare_update_link);
4018 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LMV,
4029 * Create target, migrate xattrs and update links.
4031 * Create target according to \a spec, and then migrate xattrs, if it's
4032 * directory, migrate source stripes to target, else update fid to target
4035 * \param[in] env execution environment
4036 * \param[in] tpobj target parent object
4037 * \param[in] sobj source object
4038 * \param[in] tobj target object
4039 * \param[in] lname file name
4040 * \param[in] attr source attributes
4041 * \param[in] sbuf source LMV buf
4042 * \param[in] ldata source linkea
4043 * \param[in] spec migrate create spec
4044 * \param[in] hint target creation hint
4045 * \param[in] handle tranasction handle
4047 * \retval 0 on success
4048 * \retval -errno on failure
4050 static int mdd_migrate_create(const struct lu_env *env,
4051 struct mdd_object *tpobj,
4052 struct mdd_object *sobj,
4053 struct mdd_object *tobj,
4054 const struct lu_name *lname,
4055 struct lu_attr *attr,
4056 const struct lu_buf *sbuf,
4057 struct linkea_data *ldata,
4058 struct md_op_spec *spec,
4059 struct dt_allocation_hint *hint,
4060 struct thandle *handle)
4067 * directory will migrate sobj stripes to tobj:
4068 * 1. delete stripes from sobj.
4069 * 2. add stripes to tobj, see lod_dir_declare_layout_add().
4070 * 3. create/attach stripes for tobj, see lod_xattr_set_lmv().
4072 if (S_ISDIR(attr->la_mode)) {
4073 struct lu_buf lmu_buf = { NULL };
4076 struct mdd_thread_info *info = mdd_env_info(env);
4077 struct lmv_user_md *lmu = &info->mti_lmv.lmv_user_md;
4079 lmu->lum_stripe_count = 0;
4080 lmu_buf.lb_buf = lmu;
4081 lmu_buf.lb_len = sizeof(*lmu);
4084 rc = mdd_dir_layout_delete(env, sobj, sbuf, &lmu_buf, handle);
4089 * delete LMV so that later when destroying sobj it won't delete
4093 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4094 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LMV, handle);
4095 mdd_write_unlock(env, sobj);
4101 /* don't set nlink from sobj */
4102 attr->la_valid &= ~LA_NLINK;
4104 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4109 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4110 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
4111 mdd_write_unlock(env, tobj);
4115 /* for regular file, update OST objects XATTR_NAME_FID */
4116 if (S_ISREG(attr->la_mode)) {
4117 struct lu_buf fid_buf;
4119 /* target may be remote, update PFID via sobj. */
4120 fid_buf.lb_buf = (void *)mdo2fid(tobj);
4121 fid_buf.lb_len = sizeof(struct lu_fid);
4122 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4127 /* delete LOV to avoid deleting OST objs when destroying sobj */
4128 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4129 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4130 mdd_write_unlock(env, sobj);
4135 if (!S_ISDIR(attr->la_mode))
4136 rc = mdd_iterate_linkea(env, sobj, tobj, lname, mdo2fid(tpobj),
4137 ldata, NULL, handle, mdd_update_link);
4142 static int mdd_declare_migrate_update(const struct lu_env *env,
4143 struct mdd_object *spobj,
4144 struct mdd_object *tpobj,
4145 struct mdd_object *sobj,
4146 struct mdd_object *tobj,
4147 const struct lu_name *lname,
4148 struct lu_attr *attr,
4149 struct lu_attr *spattr,
4150 struct lu_attr *tpattr,
4151 struct linkea_data *ldata,
4155 struct thandle *handle)
4157 struct mdd_thread_info *info = mdd_env_info(env);
4158 const struct lu_fid *fid = mdo2fid(do_create ? tobj : sobj);
4159 struct lu_attr *la = &info->mti_la_for_fix;
4162 rc = mdo_declare_index_delete(env, spobj, lname->ln_name, handle);
4166 if (S_ISDIR(attr->la_mode)) {
4167 rc = mdo_declare_ref_del(env, spobj, handle);
4172 rc = mdo_declare_index_insert(env, tpobj, fid, mdd_object_type(sobj),
4173 lname->ln_name, handle);
4177 rc = mdd_declare_links_add(env, do_create ? tobj : sobj, handle, ldata);
4181 if (S_ISDIR(attr->la_mode)) {
4182 rc = mdo_declare_ref_add(env, tpobj, handle);
4187 la->la_valid = LA_CTIME | LA_MTIME;
4188 rc = mdo_declare_attr_set(env, spobj, la, handle);
4192 if (tpobj != spobj) {
4193 rc = mdo_declare_attr_set(env, tpobj, la, handle);
4198 if (do_create && do_destroy) {
4199 rc = mdo_declare_ref_del(env, sobj, handle);
4203 rc = mdo_declare_destroy(env, sobj, handle);
4212 * migrate dirent from \a spobj to \a tpobj, and destroy \a sobj
4214 static int mdd_migrate_update(const struct lu_env *env,
4215 struct mdd_object *spobj,
4216 struct mdd_object *tpobj,
4217 struct mdd_object *sobj,
4218 struct mdd_object *tobj,
4219 const struct lu_name *lname,
4220 struct lu_attr *attr,
4221 struct lu_attr *spattr,
4222 struct lu_attr *tpattr,
4223 struct linkea_data *ldata,
4227 struct thandle *handle)
4229 struct mdd_thread_info *info = mdd_env_info(env);
4230 const struct lu_fid *fid = mdo2fid(do_create ? tobj : sobj);
4231 struct lu_attr *la = &info->mti_la_for_fix;
4236 CDEBUG(D_INFO, "update %s "DFID"/"DFID" to "DFID"/"DFID"\n",
4237 lname->ln_name, PFID(mdo2fid(spobj)),
4238 PFID(mdo2fid(sobj)), PFID(mdo2fid(tpobj)),
4241 rc = __mdd_index_delete(env, spobj, lname->ln_name,
4242 S_ISDIR(attr->la_mode), handle);
4246 rc = __mdd_index_insert(env, tpobj, fid, mdd_object_type(sobj),
4247 lname->ln_name, handle);
4251 rc = mdd_links_write(env, do_create ? tobj : sobj, ldata, handle);
4255 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
4256 la->la_valid = LA_CTIME | LA_MTIME;
4257 mdd_write_lock(env, spobj, DT_SRC_PARENT);
4258 rc = mdd_update_time(env, spobj, spattr, la, handle);
4259 mdd_write_unlock(env, spobj);
4263 if (tpobj != spobj) {
4264 la->la_valid = LA_CTIME | LA_MTIME;
4265 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
4266 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
4267 mdd_write_unlock(env, tpobj);
4273 * there are three situations we shouldn't destroy source:
4274 * 1. if source is not dir, and it happens to be located on the same MDT
4276 * 2. if source is not dir, and has link on the same MDT where source is
4278 * 3. if source is dir, and it's a normal, non-empty dir.
4280 * the first two situations equals to !do_create, and the 3rd equals to
4281 * !do_destroy, so the below condition is actually
4282 * !(!do_create || !do_destroy).
4284 * NB, if user has opened source dir before migration, he will get
4285 * -ENOENT error when close it later, because source is likely to be
4286 * remote, which can't be moved to orphan list, but except this error
4287 * message, this won't cause any inconsistency or trouble.
4289 if (do_create && do_destroy) {
4290 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4291 mdo_ref_del(env, sobj, handle);
4292 rc = mdo_destroy(env, sobj, handle);
4293 mdd_write_unlock(env, sobj);
4300 * Migrate directory or file.
4302 * migrate source to target in following steps:
4303 * 1. create target, append source stripes after target's if it's directory,
4304 * migrate xattrs and update fid of source links.
4305 * 2. update namespace: migrate dirent from source parent to target parent,
4306 * update file linkea, and destroy source if it's not needed any more.
4308 * \param[in] env execution environment
4309 * \param[in] md_pobj parent master object
4310 * \param[in] md_sobj source object
4311 * \param[in] lname file name
4312 * \param[in] md_tobj target object
4313 * \param[in] spec target creation spec
4314 * \param[in] ma used to update \a pobj mtime and ctime
4316 * \retval 0 on success
4317 * \retval -errno on failure
4319 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4320 struct md_object *md_sobj, const struct lu_name *lname,
4321 struct md_object *md_tobj, struct md_op_spec *spec,
4324 struct mdd_device *mdd = mdo2mdd(md_pobj);
4325 struct mdd_thread_info *info = mdd_env_info(env);
4326 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4327 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4328 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4329 struct mdd_object *spobj = NULL;
4330 struct mdd_object *tpobj = NULL;
4331 struct lu_attr *spattr = &info->mti_pattr;
4332 struct lu_attr *tpattr = &info->mti_tpattr;
4333 struct lu_attr *attr = &info->mti_cattr;
4334 struct linkea_data *ldata = &info->mti_link_data;
4335 struct dt_allocation_hint *hint = &info->mti_hint;
4336 struct lu_fid *fid = &info->mti_fid2;
4337 struct lu_buf pbuf = { NULL };
4338 struct lu_buf sbuf = { NULL };
4339 struct lmv_mds_md_v1 *plmv;
4340 struct thandle *handle;
4341 bool do_create = true;
4342 bool do_destroy = true;
4346 rc = mdd_la_get(env, sobj, attr);
4350 /* locate source and target stripe on pobj, which are the real parent */
4351 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4352 if (rc < 0 && rc != -ENODATA)
4357 __u32 hash_type = le32_to_cpu(plmv->lmv_hash_type);
4358 __u32 count = le32_to_cpu(plmv->lmv_stripe_count);
4361 /* locate target parent stripe */
4362 if (hash_type & LMV_HASH_FLAG_MIGRATION) {
4364 * fail check here to make sure top dir migration
4367 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4368 GOTO(out, rc = -EIO);
4369 hash_type &= ~LMV_HASH_FLAG_MIGRATION;
4370 count = le32_to_cpu(plmv->lmv_migrate_offset);
4372 index = lmv_name_to_stripe_index(hash_type, count,
4376 GOTO(out, rc = index);
4378 fid_le_to_cpu(fid, &plmv->lmv_stripe_fids[index]);
4379 tpobj = mdd_object_find(env, mdd, fid);
4381 GOTO(out, rc = PTR_ERR(tpobj));
4383 /* locate source parent stripe */
4384 if (le32_to_cpu(plmv->lmv_hash_type) &
4385 LMV_HASH_FLAG_MIGRATION) {
4386 hash_type = le32_to_cpu(plmv->lmv_migrate_hash);
4387 count = le32_to_cpu(plmv->lmv_stripe_count) -
4388 le32_to_cpu(plmv->lmv_migrate_offset);
4390 index = lmv_name_to_stripe_index(hash_type, count,
4394 mdd_object_put(env, tpobj);
4395 GOTO(out, rc = index);
4398 index += le32_to_cpu(plmv->lmv_migrate_offset);
4399 fid_le_to_cpu(fid, &plmv->lmv_stripe_fids[index]);
4400 spobj = mdd_object_find(env, mdd, fid);
4401 if (IS_ERR(spobj)) {
4402 mdd_object_put(env, tpobj);
4403 GOTO(out, rc = PTR_ERR(spobj));
4407 mdd_object_get(spobj);
4412 mdd_object_get(tpobj);
4413 mdd_object_get(spobj);
4416 rc = mdd_la_get(env, spobj, spattr);
4420 rc = mdd_la_get(env, tpobj, tpattr);
4424 if (S_ISDIR(attr->la_mode)) {
4425 struct lmv_user_md_v1 *lmu = spec->u.sp_ea.eadata;
4430 * if user use default value '0' for stripe_count, we need to
4431 * adjust it to '1' to create a 1-stripe directory.
4433 if (lmu->lum_stripe_count == 0) {
4434 /* eadata is from request, don't alter it */
4435 info->mti_lmu = *lmu;
4436 info->mti_lmu.lum_stripe_count = cpu_to_le32(1);
4437 spec->u.sp_ea.eadata = &info->mti_lmu;
4438 lmu = spec->u.sp_ea.eadata;
4441 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4442 if (rc == -ENODATA) {
4443 if (mdd_dir_is_empty(env, sobj) == 0) {
4445 * if sobj is empty, and target is not striped,
4446 * create target as a normal directory.
4448 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4449 info->mti_lmu = *lmu;
4450 info->mti_lmu.lum_stripe_count = 0;
4451 spec->u.sp_ea.eadata = &info->mti_lmu;
4452 lmu = spec->u.sp_ea.eadata;
4456 * sobj is not striped dir, if it's not empty,
4457 * it will be migrated to be a stripe of target,
4458 * don't destroy it after migration.
4465 struct lmv_mds_md_v1 *lmv = sbuf.lb_buf;
4467 if (le32_to_cpu(lmv->lmv_hash_type) &
4468 LMV_HASH_FLAG_MIGRATION) {
4469 __u32 lum_stripe_count = lmu->lum_stripe_count;
4470 __u32 lmv_hash_type = lmv->lmv_hash_type &
4471 cpu_to_le32(LMV_HASH_TYPE_MASK);
4473 if (!lum_stripe_count)
4474 lum_stripe_count = cpu_to_le32(1);
4476 /* TODO: check specific MDTs */
4477 if (lmv->lmv_migrate_offset !=
4479 lmv->lmv_master_mdt_index !=
4480 lmu->lum_stripe_offset ||
4481 (lmv_hash_type != 0 &&
4482 lmv_hash_type != lmu->lum_hash_type)) {
4483 CERROR("%s: \'"DNAME"\' migration was "
4484 "interrupted, run \'lfs migrate "
4485 "-m %d -c %d -H %d "DNAME"\' to "
4486 "finish migration.\n",
4487 mdd2obd_dev(mdd)->obd_name,
4490 lmv->lmv_master_mdt_index),
4492 lmv->lmv_migrate_offset),
4493 le32_to_cpu(lmv_hash_type),
4495 GOTO(out, rc = -EPERM);
4497 GOTO(out, rc = -EALREADY);
4500 } else if (!mdd_object_remote(tpobj)) {
4502 * if source is already on MDT where target parent is located,
4503 * no need to create, just update namespace.
4506 } else if (S_ISLNK(attr->la_mode)) {
4507 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4509 GOTO(out, rc = -ENOMEM);
4510 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4513 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4514 mdd2obd_dev(mdd)->obd_name,
4515 PFID(mdo2fid(sobj)), rc);
4518 spec->u.sp_symname = sbuf.lb_buf;
4519 } else if (S_ISREG(attr->la_mode)) {
4520 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
4521 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
4525 * if sobj has link on the same MDT, no need to create, just update
4526 * namespace, and it will be a remote file on target parent, which is
4527 * similar to rename.
4529 rc = migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, lname, attr,
4536 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4537 spattr, tpattr, attr);
4541 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
4543 handle = mdd_trans_create(env, mdd);
4545 GOTO(out, rc = PTR_ERR(handle));
4548 rc = mdd_declare_migrate_create(env, tpobj, sobj, tobj, lname,
4549 attr, &sbuf, ldata, spec, hint,
4552 GOTO(stop_trans, rc);
4555 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, tobj, lname,
4556 attr, spattr, tpattr, ldata, do_create,
4557 do_destroy, ma, handle);
4559 GOTO(stop_trans, rc);
4561 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4564 GOTO(stop_trans, rc);
4566 rc = mdd_trans_start(env, mdd, handle);
4568 GOTO(stop_trans, rc);
4571 rc = mdd_migrate_create(env, tpobj, sobj, tobj, lname, attr,
4572 &sbuf, ldata, spec, hint, handle);
4574 GOTO(stop_trans, rc);
4577 rc = mdd_migrate_update(env, spobj, tpobj, sobj, tobj, lname, attr,
4578 spattr, tpattr, ldata, do_create, do_destroy,
4581 GOTO(stop_trans, rc);
4583 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
4584 mdo2fid(spobj), mdo2fid(sobj),
4585 mdo2fid(tpobj), lname, lname, handle);
4587 GOTO(stop_trans, rc);
4591 rc = mdd_trans_stop(env, mdd, rc, handle);
4593 if (spobj && !IS_ERR(spobj))
4594 mdd_object_put(env, spobj);
4595 if (tpobj && !IS_ERR(tpobj))
4596 mdd_object_put(env, tpobj);
4602 static int __mdd_dir_declare_layout_shrink(const struct lu_env *env,
4603 struct mdd_object *pobj,
4604 struct mdd_object *obj,
4605 struct mdd_object *stripe,
4606 struct lu_attr *attr,
4607 struct lu_buf *lmv_buf,
4608 const struct lu_buf *lmu_buf,
4609 struct lu_name *lname,
4610 struct thandle *handle)
4612 struct mdd_thread_info *info = mdd_env_info(env);
4613 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
4614 struct lmv_user_md *lmu = (typeof(lmu))info->mti_key;
4615 struct lu_buf shrink_buf = { .lb_buf = lmu,
4616 .lb_len = sizeof(*lmu) };
4621 memcpy(lmu, lmu_buf->lb_buf, sizeof(*lmu));
4623 if (le32_to_cpu(lmu->lum_stripe_count) < 2)
4624 lmu->lum_stripe_count = 0;
4626 rc = mdd_dir_declare_layout_delete(env, obj, lmv_buf, &shrink_buf,
4631 if (lmu->lum_stripe_count == 0) {
4632 lmu->lum_stripe_count = cpu_to_le32(1);
4634 rc = mdo_declare_xattr_del(env, obj, XATTR_NAME_LMV, handle);
4639 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
4640 mdd_dir_declare_destroy_stripe);
4644 if (le32_to_cpu(lmu->lum_stripe_count) > 1)
4645 return mdo_declare_xattr_set(env, obj, lmv_buf,
4646 XATTR_NAME_LMV".set", 0, handle);
4648 rc = mdo_declare_index_insert(env, stripe, mdo2fid(pobj), S_IFDIR,
4653 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle,
4654 mdo_declare_xattr_set);
4658 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4662 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4666 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4670 rc = mdo_declare_index_insert(env, pobj, mdo2fid(stripe), attr->la_mode,
4671 lname->ln_name, handle);
4675 rc = mdo_declare_ref_del(env, obj, handle);
4679 rc = mdo_declare_ref_del(env, obj, handle);
4683 rc = mdo_declare_destroy(env, obj, handle);
4692 * after files under \a obj were migrated, shrink old stripes from \a obj,
4693 * furthermore, if it becomes a 1-stripe directory, convert it to a normal one.
4695 static int __mdd_dir_layout_shrink(const struct lu_env *env,
4696 struct mdd_object *pobj,
4697 struct mdd_object *obj,
4698 struct mdd_object *stripe,
4699 struct lu_attr *attr,
4700 struct lu_buf *lmv_buf,
4701 const struct lu_buf *lmu_buf,
4702 struct lu_name *lname,
4703 struct thandle *handle)
4705 struct mdd_thread_info *info = mdd_env_info(env);
4706 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
4707 struct lmv_user_md *lmu = (typeof(lmu))info->mti_key;
4708 struct lu_buf shrink_buf = { .lb_buf = lmu,
4709 .lb_len = sizeof(*lmu) };
4710 int len = lmv_buf->lb_len;
4711 __u32 version = le32_to_cpu(lmv->lmv_layout_version);
4716 /* lmu needs to be altered, but lmu_buf is const */
4717 memcpy(lmu, lmu_buf->lb_buf, sizeof(*lmu));
4720 * if dir will be shrunk to 1-stripe, delete all stripes, because it
4721 * will be converted to normal dir.
4723 if (le32_to_cpu(lmu->lum_stripe_count) == 1)
4724 lmu->lum_stripe_count = 0;
4726 /* delete stripes after lmu_stripe_count */
4727 rc = mdd_dir_layout_delete(env, obj, lmv_buf, &shrink_buf, handle);
4731 if (lmu->lum_stripe_count == 0) {
4732 lmu->lum_stripe_count = cpu_to_le32(1);
4734 /* delete LMV to avoid deleting stripes again upon destroy */
4735 mdd_write_lock(env, obj, DT_SRC_CHILD);
4736 rc = mdo_xattr_del(env, obj, XATTR_NAME_LMV, handle);
4737 mdd_write_unlock(env, obj);
4742 /* destroy stripes after lmu_stripe_count */
4743 mdd_write_lock(env, obj, DT_SRC_PARENT);
4744 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
4745 mdd_dir_destroy_stripe);
4746 mdd_write_unlock(env, obj);
4748 if (le32_to_cpu(lmu->lum_stripe_count) > 1) {
4749 /* update dir LMV, that's all if it's still striped. */
4750 lmv->lmv_stripe_count = lmu->lum_stripe_count;
4751 lmv->lmv_hash_type &= ~cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4752 lmv->lmv_migrate_offset = 0;
4753 lmv->lmv_migrate_hash = 0;
4754 lmv->lmv_layout_version = cpu_to_le32(++version);
4756 lmv_buf->lb_len = sizeof(*lmv);
4757 rc = mdo_xattr_set(env, obj, lmv_buf, XATTR_NAME_LMV".set", 0,
4759 lmv_buf->lb_len = len;
4763 /* replace directory with its remaining stripe */
4767 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4768 mdd_write_lock(env, obj, DT_SRC_CHILD);
4770 /* insert dotdot to stripe which points to parent */
4771 rc = __mdd_index_insert_only(env, stripe, mdo2fid(pobj), S_IFDIR,
4776 /* copy xattrs including linkea */
4777 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle, mdo_xattr_set);
4782 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4786 /* don't set nlink from parent */
4787 attr->la_valid &= ~LA_NLINK;
4789 rc = mdo_attr_set(env, stripe, attr, handle);
4793 /* delete dir name from parent */
4794 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4798 /* insert stripe to parent with dir name */
4799 rc = __mdd_index_insert_only(env, pobj, mdo2fid(stripe), attr->la_mode,
4800 lname->ln_name, handle);
4804 /* destroy dir obj */
4805 rc = mdo_ref_del(env, obj, handle);
4809 rc = mdo_ref_del(env, obj, handle);
4813 rc = mdo_destroy(env, obj, handle);
4819 mdd_write_unlock(env, obj);
4820 mdd_write_unlock(env, pobj);
4826 * shrink directory stripes to lum_stripe_count specified by lum_mds_md.
4828 int mdd_dir_layout_shrink(const struct lu_env *env,
4829 struct md_object *md_obj,
4830 const struct lu_buf *lmu_buf)
4832 struct mdd_device *mdd = mdo2mdd(md_obj);
4833 struct mdd_thread_info *info = mdd_env_info(env);
4834 struct mdd_object *obj = md2mdd_obj(md_obj);
4835 struct mdd_object *pobj = NULL;
4836 struct mdd_object *stripe = NULL;
4837 struct lu_attr *attr = &info->mti_pattr;
4838 struct lu_fid *fid = &info->mti_fid2;
4839 struct lu_name lname = { NULL };
4840 struct lu_buf lmv_buf = { NULL };
4841 struct lmv_mds_md_v1 *lmv;
4842 struct lmv_user_md *lmu;
4843 struct thandle *handle;
4848 rc = mdd_la_get(env, obj, attr);
4852 if (!S_ISDIR(attr->la_mode))
4855 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4859 lmv = lmv_buf.lb_buf;
4860 lmu = lmu_buf->lb_buf;
4862 /* this was checked in MDT */
4863 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4864 le32_to_cpu(lmv->lmv_stripe_count));
4866 rc = mdd_dir_iterate_stripes(env, obj, &lmv_buf, lmu_buf, NULL,
4867 mdd_shrink_stripe_is_empty);
4871 GOTO(out, rc = -ENOTEMPTY);
4874 * if obj stripe count will be shrunk to 1, we need to convert it to a
4875 * normal dir, which will change its fid and update parent namespace,
4876 * get obj name and parent fid from linkea.
4878 if (le32_to_cpu(lmu->lum_stripe_count) < 2) {
4879 struct linkea_data *ldata = &info->mti_link_data;
4880 char *filename = info->mti_name;
4882 rc = mdd_links_read(env, obj, ldata);
4886 if (ldata->ld_leh->leh_reccount > 1)
4887 GOTO(out, rc = -EINVAL);
4889 linkea_first_entry(ldata);
4891 GOTO(out, rc = -ENODATA);
4893 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4896 /* Note: lname might miss \0 at the end */
4897 snprintf(filename, sizeof(info->mti_name), "%.*s",
4898 lname.ln_namelen, lname.ln_name);
4899 lname.ln_name = filename;
4901 pobj = mdd_object_find(env, mdd, fid);
4908 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4910 stripe = mdd_object_find(env, mdd, fid);
4911 if (IS_ERR(stripe)) {
4912 mdd_object_put(env, pobj);
4914 GOTO(out, rc = PTR_ERR(stripe));
4918 handle = mdd_trans_create(env, mdd);
4920 GOTO(out, rc = PTR_ERR(handle));
4922 rc = __mdd_dir_declare_layout_shrink(env, pobj, obj, stripe, attr,
4923 &lmv_buf, lmu_buf, &lname, handle);
4925 GOTO(stop_trans, rc);
4927 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4930 GOTO(stop_trans, rc);
4932 rc = mdd_trans_start(env, mdd, handle);
4934 GOTO(stop_trans, rc);
4936 rc = __mdd_dir_layout_shrink(env, pobj, obj, stripe, attr, &lmv_buf,
4937 lmu_buf, &lname, handle);
4939 GOTO(stop_trans, rc);
4941 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4942 XATTR_NAME_LMV, handle);
4943 GOTO(stop_trans, rc);
4946 rc = mdd_trans_stop(env, mdd, rc, handle);
4949 mdd_object_put(env, stripe);
4950 mdd_object_put(env, pobj);
4952 lu_buf_free(&lmv_buf);
4956 const struct md_dir_operations mdd_dir_ops = {
4957 .mdo_is_subdir = mdd_is_subdir,
4958 .mdo_lookup = mdd_lookup,
4959 .mdo_create = mdd_create,
4960 .mdo_rename = mdd_rename,
4961 .mdo_link = mdd_link,
4962 .mdo_unlink = mdd_unlink,
4963 .mdo_create_data = mdd_create_data,
4964 .mdo_migrate = mdd_migrate,