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 attr->la_valid = LA_FLAGS;
1460 attr->la_flags = LUSTRE_ORPHAN_FL;
1463 rc = mdo_declare_attr_set(env, obj, attr, handle);
1465 rc = mdo_attr_set(env, obj, attr, handle);
1470 static int mdd_declare_finish_unlink(const struct lu_env *env,
1471 struct mdd_object *obj,
1472 struct thandle *handle)
1476 /* Sigh, we do not know if the unlink object will become orphan in
1477 * declare phase, but fortunately the flags here does not matter
1478 * in current declare implementation */
1479 rc = mdd_mark_orphan_object(env, obj, handle, true);
1483 rc = mdo_declare_destroy(env, obj, handle);
1487 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1491 return mdd_declare_links_del(env, obj, handle);
1494 /* caller should take a lock before calling */
1495 int mdd_finish_unlink(const struct lu_env *env,
1496 struct mdd_object *obj, struct md_attr *ma,
1497 const struct mdd_object *pobj,
1498 const struct lu_name *lname,
1502 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1505 LASSERT(mdd_write_locked(env, obj) != 0);
1507 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1508 /* add new orphan and the object
1509 * will be deleted during mdd_close() */
1510 obj->mod_flags |= DEAD_OBJ;
1511 if (obj->mod_count) {
1512 rc = mdd_orphan_insert(env, obj, th);
1514 CDEBUG(D_HA, "Object "DFID" is inserted into "
1515 "orphan list, open count = %d\n",
1516 PFID(mdd_object_fid(obj)),
1519 CERROR("Object "DFID" fail to be an orphan, "
1520 "open count = %d, maybe cause failed "
1522 PFID(mdd_object_fid(obj)),
1525 /* mark object as an orphan here, not
1526 * before mdd_orphan_insert() as racing
1527 * mdd_la_get() may propagate ORPHAN_OBJ
1528 * causing the asserition */
1529 rc = mdd_mark_orphan_object(env, obj, th, false);
1531 rc = mdo_destroy(env, obj, th);
1533 } else if (!is_dir) {
1534 /* old files may not have link ea; ignore errors */
1535 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1543 * has mdd_write_lock on cobj already, but not on pobj yet
1545 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1546 const struct lu_attr *pattr,
1547 struct mdd_object *cobj,
1548 const struct lu_attr *cattr)
1553 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1558 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1559 struct mdd_object *p, struct mdd_object *c,
1560 const struct lu_name *name, struct md_attr *ma,
1561 struct thandle *handle, int no_name, int is_dir)
1563 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1566 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1567 if (likely(no_name == 0)) {
1568 rc = mdo_declare_index_delete(env, p, name->ln_name,
1575 rc = mdo_declare_ref_del(env, p, handle);
1581 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1582 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1583 la->la_valid = LA_CTIME | LA_MTIME;
1584 rc = mdo_declare_attr_set(env, p, la, handle);
1589 rc = mdo_declare_ref_del(env, c, handle);
1593 rc = mdo_declare_ref_del(env, c, handle);
1597 la->la_valid = LA_CTIME;
1598 rc = mdo_declare_attr_set(env, c, la, handle);
1602 rc = mdd_declare_finish_unlink(env, c, handle);
1606 /* FIXME: need changelog for remove entry */
1607 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1615 * test if a file has an HSM archive
1616 * if HSM attributes are not found in ma update them from
1619 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1620 struct mdd_object *obj,
1625 if (!(ma->ma_valid & MA_HSM)) {
1626 /* no HSM MD provided, read xattr */
1627 struct lu_buf *hsm_buf;
1628 const size_t buflen = sizeof(struct hsm_attrs);
1631 hsm_buf = mdd_buf_get(env, NULL, 0);
1632 lu_buf_alloc(hsm_buf, buflen);
1633 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1634 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1635 lu_buf_free(hsm_buf);
1639 ma->ma_valid |= MA_HSM;
1641 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1647 * Delete name entry and the object.
1648 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1649 * does not exist for this object, and it could only happen during resending
1650 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1651 * is also needed in this case(needed by changelog), so we have to add another
1652 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1653 * the ENOENT failure should be able to be fixed by redo mechanism.
1655 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1656 struct md_object *cobj, const struct lu_name *lname,
1657 struct md_attr *ma, int no_name)
1659 const char *name = lname->ln_name;
1660 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1661 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1662 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1663 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1664 struct mdd_object *mdd_cobj = NULL;
1665 struct mdd_device *mdd = mdo2mdd(pobj);
1666 struct thandle *handle;
1667 int rc, is_dir = 0, cl_flags = 0;
1670 /* let shutdown to start */
1671 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1673 /* cobj == NULL means only delete name entry */
1674 if (likely(cobj != NULL)) {
1675 mdd_cobj = md2mdd_obj(cobj);
1676 if (mdd_object_exists(mdd_cobj) == 0)
1680 rc = mdd_la_get(env, mdd_pobj, pattr);
1684 if (likely(mdd_cobj != NULL)) {
1686 rc = mdd_la_get(env, mdd_cobj, cattr);
1690 is_dir = S_ISDIR(cattr->la_mode);
1691 /* search for an existing archive.
1692 * we should check ahead as the object
1693 * can be destroyed in this transaction */
1694 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1695 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1698 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1702 handle = mdd_trans_create(env, mdd);
1704 RETURN(PTR_ERR(handle));
1706 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1707 lname, ma, handle, no_name, is_dir);
1711 rc = mdd_trans_start(env, mdd, handle);
1715 if (likely(mdd_cobj != NULL))
1716 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1718 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1719 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1724 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1725 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1726 GOTO(cleanup, rc = 0);
1728 if (likely(mdd_cobj != NULL)) {
1729 rc = mdo_ref_del(env, mdd_cobj, handle);
1731 __mdd_index_insert_only(env, mdd_pobj,
1733 mdd_object_type(mdd_cobj),
1740 mdo_ref_del(env, mdd_cobj, handle);
1742 /* fetch updated nlink */
1743 rc = mdd_la_get(env, mdd_cobj, cattr);
1748 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1749 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1751 la->la_valid = LA_CTIME | LA_MTIME;
1752 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1756 /* Enough for only unlink the entry */
1757 if (unlikely(mdd_cobj == NULL))
1760 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1761 /* update ctime of an unlinked file only if it is still
1762 * opened or a link still exists */
1763 la->la_valid = LA_CTIME;
1764 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1769 /* XXX: this transfer to ma will be removed with LOD/OSP */
1770 ma->ma_attr = *cattr;
1771 ma->ma_valid |= MA_INODE;
1772 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1776 /* fetch updated nlink */
1777 rc = mdd_la_get(env, mdd_cobj, cattr);
1778 /* if object is removed then we can't get its attrs,
1780 if (rc == -ENOENT) {
1781 cattr->la_nlink = 0;
1785 if (cattr->la_nlink == 0) {
1786 ma->ma_attr = *cattr;
1787 ma->ma_valid |= MA_INODE;
1792 if (likely(mdd_cobj != NULL))
1793 mdd_write_unlock(env, mdd_cobj);
1796 if (cattr->la_nlink == 0)
1797 cl_flags |= CLF_UNLINK_LAST;
1799 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1801 rc = mdd_changelog_ns_store(env, mdd,
1802 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1803 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1808 rc = mdd_trans_stop(env, mdd, rc, handle);
1814 * The permission has been checked when obj created, no need check again.
1816 static int mdd_cd_sanity_check(const struct lu_env *env,
1817 struct mdd_object *obj)
1822 if (!obj || mdd_is_dead_obj(obj))
1828 static int mdd_create_data(const struct lu_env *env,
1829 struct md_object *pobj,
1830 struct md_object *cobj,
1831 const struct md_op_spec *spec,
1834 struct mdd_device *mdd = mdo2mdd(cobj);
1835 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1836 struct mdd_object *son = md2mdd_obj(cobj);
1837 struct thandle *handle;
1838 const struct lu_buf *buf;
1839 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1840 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1844 rc = mdd_cd_sanity_check(env, son);
1848 if (!md_should_create(spec->sp_cr_flags))
1852 * there are following use cases for this function:
1853 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1854 * striping can be specified or not
1857 rc = mdd_la_get(env, son, attr);
1861 /* calling ->ah_make_hint() is used to transfer information from parent */
1862 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1864 handle = mdd_trans_create(env, mdd);
1866 GOTO(out_free, rc = PTR_ERR(handle));
1869 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1870 * Should this be fixed?
1872 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1873 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1874 spec->sp_cr_flags, spec->no_create);
1876 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1877 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1878 spec->u.sp_ea.eadatalen);
1883 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1884 XATTR_NAME_LOV, 0, handle);
1888 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1893 rc = mdd_trans_start(env, mdd, handle);
1897 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1903 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1906 rc = mdd_trans_stop(env, mdd, rc, handle);
1912 static int mdd_declare_object_initialize(const struct lu_env *env,
1913 struct mdd_object *parent,
1914 struct mdd_object *child,
1915 const struct lu_attr *attr,
1916 struct thandle *handle)
1921 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1922 if (!S_ISDIR(attr->la_mode))
1925 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1930 rc = mdo_declare_ref_add(env, child, handle);
1934 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1940 static int mdd_object_initialize(const struct lu_env *env,
1941 const struct lu_fid *pfid,
1942 struct mdd_object *child,
1943 struct lu_attr *attr,
1944 struct thandle *handle)
1949 if (S_ISDIR(attr->la_mode)) {
1950 /* Add "." and ".." for newly created dir */
1951 mdo_ref_add(env, child, handle);
1952 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1953 S_IFDIR, dot, handle);
1955 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1958 mdo_ref_del(env, child, handle);
1965 * This function checks whether it can create a file/dir under the
1966 * directory(@pobj). The directory(@pobj) is not being locked by
1969 * \param[in] env execution environment
1970 * \param[in] pobj the directory to create files
1971 * \param[in] pattr the attributes of the directory
1972 * \param[in] lname the name of the created file/dir
1973 * \param[in] cattr the attributes of the file/dir
1974 * \param[in] spec create specification
1976 * \retval = 0 it is allowed to create file/dir under
1978 * \retval negative error not allowed to create file/dir
1979 * under the directory
1981 static int mdd_create_sanity_check(const struct lu_env *env,
1982 struct md_object *pobj,
1983 const struct lu_attr *pattr,
1984 const struct lu_name *lname,
1985 struct lu_attr *cattr,
1986 struct md_op_spec *spec)
1988 struct mdd_thread_info *info = mdd_env_info(env);
1989 struct lu_fid *fid = &info->mti_fid;
1990 struct mdd_object *obj = md2mdd_obj(pobj);
1991 struct mdd_device *m = mdo2mdd(pobj);
1992 bool check_perm = true;
1997 if (mdd_is_dead_obj(obj))
2001 * In some cases this lookup is not needed - we know before if name
2002 * exists or not because MDT performs lookup for it.
2003 * name length check is done in lookup.
2005 if (spec->sp_cr_lookup) {
2007 * Check if the name already exist, though it will be checked in
2008 * _index_insert also, for avoiding rolling back if exists
2011 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2012 MAY_WRITE | MAY_EXEC);
2014 RETURN(rc ? : -EEXIST);
2016 /* Permission is already being checked in mdd_lookup */
2020 if (S_ISDIR(cattr->la_mode) &&
2021 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2022 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2023 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2025 if (!lmv_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2026 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2028 CERROR("%s: invalid lmv_user_md: magic = %x, "
2029 "stripe_offset = %d, stripe_count = %u: "
2030 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2031 le32_to_cpu(lum->lum_magic),
2032 (int)le32_to_cpu(lum->lum_stripe_offset),
2033 le32_to_cpu(lum->lum_stripe_count), rc);
2038 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2043 if (pattr->la_mode & S_ISGID) {
2044 struct lu_ucred *uc = lu_ucred(env);
2046 cattr->la_gid = pattr->la_gid;
2048 /* Directories are special, and always inherit S_ISGID */
2049 if (S_ISDIR(cattr->la_mode)) {
2050 cattr->la_mode |= S_ISGID;
2051 cattr->la_valid |= LA_MODE;
2052 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2053 == (S_ISGID | S_IXGRP) &&
2054 !lustre_in_group_p(uc,
2055 (cattr->la_valid & LA_GID) ?
2056 cattr->la_gid : pattr->la_gid) &&
2057 !md_capable(uc, CFS_CAP_FSETID)) {
2058 cattr->la_mode &= ~S_ISGID;
2059 cattr->la_valid |= LA_MODE;
2063 /* Inherit project ID from parent directory */
2064 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2065 cattr->la_projid = pattr->la_projid;
2066 if (S_ISDIR(cattr->la_mode)) {
2067 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2068 cattr->la_valid |= LA_FLAGS;
2070 cattr->la_valid |= LA_PROJID;
2073 rc = mdd_name_check(m, lname);
2077 switch (cattr->la_mode & S_IFMT) {
2079 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2081 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2082 RETURN(-ENAMETOOLONG);
2101 static int mdd_declare_create_object(const struct lu_env *env,
2102 struct mdd_device *mdd,
2103 struct mdd_object *p, struct mdd_object *c,
2104 struct lu_attr *attr,
2105 struct thandle *handle,
2106 const struct md_op_spec *spec,
2107 struct lu_buf *def_acl_buf,
2108 struct lu_buf *acl_buf,
2109 struct lu_buf *hsm_buf,
2110 struct dt_allocation_hint *hint)
2112 const struct lu_buf *buf;
2115 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2120 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2121 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2122 /* if dir, then can inherit default ACl */
2123 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2124 XATTR_NAME_ACL_DEFAULT,
2130 if (acl_buf && acl_buf->lb_len > 0) {
2131 rc = mdo_declare_attr_set(env, c, attr, handle);
2135 rc = mdo_declare_xattr_set(env, c, acl_buf,
2136 XATTR_NAME_ACL_ACCESS, 0, handle);
2141 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2145 /* replay case, create LOV EA from client data */
2146 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2147 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2148 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2149 spec->u.sp_ea.eadatalen);
2150 rc = mdo_declare_xattr_set(env, c, buf,
2151 S_ISDIR(attr->la_mode) ?
2152 XATTR_NAME_LMV : XATTR_NAME_LOV,
2157 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2158 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2166 if (S_ISLNK(attr->la_mode)) {
2167 const char *target_name = spec->u.sp_symname;
2168 int sym_len = strlen(target_name);
2169 const struct lu_buf *buf;
2171 buf = mdd_buf_get_const(env, target_name, sym_len);
2172 rc = dt_declare_record_write(env, mdd_object_child(c),
2178 if (spec->sp_cr_file_secctx_name != NULL) {
2179 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2180 spec->sp_cr_file_secctx_size);
2181 rc = mdo_declare_xattr_set(env, c, buf,
2182 spec->sp_cr_file_secctx_name, 0,
2191 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2192 struct mdd_object *p, struct mdd_object *c,
2193 const struct lu_name *name,
2194 struct lu_attr *attr,
2195 struct thandle *handle,
2196 const struct md_op_spec *spec,
2197 struct linkea_data *ldata,
2198 struct lu_buf *def_acl_buf,
2199 struct lu_buf *acl_buf,
2200 struct lu_buf *hsm_buf,
2201 struct dt_allocation_hint *hint)
2205 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2206 def_acl_buf, acl_buf, hsm_buf, hint);
2210 if (S_ISDIR(attr->la_mode)) {
2211 rc = mdo_declare_ref_add(env, p, handle);
2216 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2217 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2221 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2222 enum changelog_rec_type type;
2224 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2225 name->ln_name, handle);
2229 rc = mdd_declare_links_add(env, c, handle, ldata);
2234 la->la_valid = LA_CTIME | LA_MTIME;
2235 rc = mdo_declare_attr_set(env, p, la, handle);
2239 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2240 S_ISREG(attr->la_mode) ? CL_CREATE :
2241 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2243 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2252 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2253 struct lu_attr *la, struct lu_buf *def_acl_buf,
2254 struct lu_buf *acl_buf)
2259 if (S_ISLNK(la->la_mode)) {
2260 acl_buf->lb_len = 0;
2261 def_acl_buf->lb_len = 0;
2265 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2266 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2267 XATTR_NAME_ACL_DEFAULT);
2268 mdd_read_unlock(env, pobj);
2270 /* If there are default ACL, fix mode/ACL by default ACL */
2271 def_acl_buf->lb_len = rc;
2272 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2273 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2274 acl_buf->lb_len = rc;
2275 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2278 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2279 /* If there are no default ACL, fix mode by mask */
2280 struct lu_ucred *uc = lu_ucred(env);
2282 /* The create triggered by MDT internal events, such as
2283 * LFSCK reset, will not contain valid "uc". */
2284 if (unlikely(uc != NULL))
2285 la->la_mode &= ~uc->uc_umask;
2287 acl_buf->lb_len = 0;
2288 def_acl_buf->lb_len = 0;
2295 * Create a metadata object and initialize it, set acl, xattr.
2297 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2298 struct mdd_object *son, struct lu_attr *attr,
2299 struct md_op_spec *spec, struct lu_buf *acl_buf,
2300 struct lu_buf *def_acl_buf,
2301 struct lu_buf *hsm_buf,
2302 struct dt_allocation_hint *hint,
2303 struct thandle *handle)
2305 const struct lu_buf *buf;
2308 mdd_write_lock(env, son, DT_TGT_CHILD);
2309 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2314 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2315 * created in declare phase, they also needs to be added to master
2316 * object as sub-directory entry. So it has to initialize the master
2317 * object, then set dir striped EA.(in mdo_xattr_set) */
2318 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle);
2320 GOTO(err_destroy, rc);
2323 * in case of replay we just set LOVEA provided by the client
2324 * XXX: I think it would be interesting to try "old" way where
2325 * MDT calls this xattr_set(LOV) in a different transaction.
2326 * probably this way we code can be made better.
2329 /* During creation, there are only a few cases we need do xattr_set to
2331 * 1. regular file: see comments above.
2332 * 2. dir: inherit default striping or pool settings from parent.
2333 * 3. create striped directory with provided stripeEA.
2334 * 4. create striped directory because inherit default layout from the
2337 if (spec->no_create ||
2338 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2339 S_ISDIR(attr->la_mode)) {
2340 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2341 spec->u.sp_ea.eadatalen);
2342 rc = mdo_xattr_set(env, son, buf,
2343 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2347 GOTO(err_destroy, rc);
2350 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2353 memset(&mh, 0, sizeof(mh));
2354 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2355 mh.mh_arch_id = spec->sp_archive_id;
2356 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2357 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2360 GOTO(err_destroy, rc);
2363 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2364 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2365 S_ISDIR(attr->la_mode)) {
2366 /* set default acl */
2367 rc = mdo_xattr_set(env, son, def_acl_buf,
2368 XATTR_NAME_ACL_DEFAULT, 0,
2371 GOTO(err_destroy, rc);
2373 /* set its own acl */
2374 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2375 rc = mdo_xattr_set(env, son, acl_buf,
2376 XATTR_NAME_ACL_ACCESS,
2379 GOTO(err_destroy, rc);
2383 if (S_ISLNK(attr->la_mode)) {
2384 struct dt_object *dt = mdd_object_child(son);
2385 const char *target_name = spec->u.sp_symname;
2386 int sym_len = strlen(target_name);
2389 buf = mdd_buf_get_const(env, target_name, sym_len);
2390 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2394 GOTO(err_initlized, rc = -EFAULT);
2397 if (spec->sp_cr_file_secctx_name != NULL) {
2398 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2399 spec->sp_cr_file_secctx_size);
2400 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2403 GOTO(err_initlized, rc);
2407 if (unlikely(rc != 0)) {
2409 if (S_ISDIR(attr->la_mode)) {
2410 /* Drop the reference, no need to delete "."/"..",
2411 * because the object to be destroied directly. */
2412 rc2 = mdo_ref_del(env, son, handle);
2416 rc2 = mdo_ref_del(env, son, handle);
2420 mdo_destroy(env, son, handle);
2423 mdd_write_unlock(env, son);
2427 static int mdd_index_delete(const struct lu_env *env,
2428 struct mdd_object *mdd_pobj,
2429 struct lu_attr *cattr,
2430 const struct lu_name *lname)
2432 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2433 struct thandle *handle;
2437 handle = mdd_trans_create(env, mdd);
2439 RETURN(PTR_ERR(handle));
2441 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2446 if (S_ISDIR(cattr->la_mode)) {
2447 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2452 /* Since this will only be used in the error handler path,
2453 * Let's set the thandle to be local and not mess the transno */
2454 handle->th_local = 1;
2455 rc = mdd_trans_start(env, mdd, handle);
2459 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2460 S_ISDIR(cattr->la_mode), handle);
2464 rc = mdd_trans_stop(env, mdd, rc, handle);
2470 * Create object and insert it into namespace.
2472 * Two operations have to be performed:
2474 * - an allocation of a new object (->do_create()), and
2475 * - an insertion into a parent index (->dio_insert()).
2477 * Due to locking, operation order is not important, when both are
2478 * successful, *but* error handling cases are quite different:
2480 * - if insertion is done first, and following object creation fails,
2481 * insertion has to be rolled back, but this operation might fail
2482 * also leaving us with dangling index entry.
2484 * - if creation is done first, is has to be undone if insertion fails,
2485 * leaving us with leaked space, which is not good but not fatal.
2487 * It seems that creation-first is simplest solution, but it is sub-optimal
2493 * case, because second mkdir is bound to create object, only to
2494 * destroy it immediately.
2496 * To avoid this follow local file systems that do double lookup:
2498 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2499 * 1. create (mdd_create_object_internal())
2500 * 2. insert (__mdd_index_insert(), lookup again)
2502 * \param[in] pobj parent object
2503 * \param[in] lname name of child being created
2504 * \param[in,out] child child object being created
2505 * \param[in] spec additional create parameters
2506 * \param[in] ma attributes for new child object
2508 * \retval 0 on success
2509 * \retval negative errno on failure
2511 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2512 const struct lu_name *lname, struct md_object *child,
2513 struct md_op_spec *spec, struct md_attr *ma)
2515 struct mdd_thread_info *info = mdd_env_info(env);
2516 struct lu_attr *la = &info->mti_la_for_fix;
2517 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2518 struct mdd_object *son = md2mdd_obj(child);
2519 struct mdd_device *mdd = mdo2mdd(pobj);
2520 struct lu_attr *attr = &ma->ma_attr;
2521 struct thandle *handle;
2522 struct lu_attr *pattr = &info->mti_pattr;
2523 struct lu_buf acl_buf;
2524 struct lu_buf def_acl_buf;
2525 struct lu_buf hsm_buf;
2526 struct linkea_data *ldata = &info->mti_link_data;
2527 const char *name = lname->ln_name;
2528 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2533 rc = mdd_la_get(env, mdd_pobj, pattr);
2537 /* Sanity checks before big job. */
2538 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2542 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2543 GOTO(out_free, rc = -EINPROGRESS);
2545 handle = mdd_trans_create(env, mdd);
2547 GOTO(out_free, rc = PTR_ERR(handle));
2549 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2550 MIN(mdd->mdd_dt_conf.ddp_max_ea_size, XATTR_SIZE_MAX));
2551 acl_buf = info->mti_xattr_buf;
2552 def_acl_buf.lb_buf = info->mti_key;
2553 def_acl_buf.lb_len = sizeof(info->mti_key);
2554 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2558 if (S_ISDIR(attr->la_mode)) {
2559 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2562 * migrate may create 1-stripe directory, so lod_ah_init()
2563 * doesn't adjust stripe count from lmu.
2565 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1)) {
2566 info->mti_lmu = *lmu;
2567 info->mti_lmu.lum_stripe_count = 0;
2568 spec->u.sp_ea.eadata = &info->mti_lmu;
2572 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2574 memset(ldata, 0, sizeof(*ldata));
2575 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2576 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2579 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2580 &tfid, lname, 1, 0, ldata);
2582 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2583 mdd_object_fid(mdd_pobj),
2584 lname, 1, 0, ldata);
2587 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2588 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2590 memset(&hsm_buf, 0, sizeof(hsm_buf));
2591 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2592 if (hsm_buf.lb_buf == NULL)
2593 GOTO(out_stop, rc = -ENOMEM);
2596 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2597 handle, spec, ldata, &def_acl_buf, &acl_buf,
2602 rc = mdd_trans_start(env, mdd, handle);
2606 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2607 &def_acl_buf, &hsm_buf, hint, handle);
2611 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2612 mdd_write_lock(env, son, DT_TGT_CHILD);
2613 son->mod_flags |= VOLATILE_OBJ;
2614 rc = mdd_orphan_insert(env, son, handle);
2615 GOTO(out_volatile, rc);
2617 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2618 attr->la_mode, name, handle);
2620 GOTO(err_created, rc);
2622 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2625 /* update parent directory mtime/ctime */
2627 la->la_valid = LA_CTIME | LA_MTIME;
2628 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2630 GOTO(err_insert, rc);
2636 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2637 rc2 = mdd_orphan_delete(env, son, handle);
2639 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2640 S_ISDIR(attr->la_mode),
2646 mdd_write_lock(env, son, DT_TGT_CHILD);
2647 if (S_ISDIR(attr->la_mode)) {
2648 /* Drop the reference, no need to delete "."/"..",
2649 * because the object is to be destroyed directly. */
2650 rc2 = mdo_ref_del(env, son, handle);
2652 mdd_write_unlock(env, son);
2657 /* For volatile files drop one link immediately, since there is
2658 * no filename in the namespace, and save any error returned. */
2659 rc2 = mdo_ref_del(env, son, handle);
2661 mdd_write_unlock(env, son);
2662 if (unlikely(rc == 0))
2667 /* Don't destroy the volatile object on success */
2668 if (likely(rc != 0))
2669 mdo_destroy(env, son, handle);
2670 mdd_write_unlock(env, son);
2673 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2674 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2675 rc = mdd_changelog_ns_store(env, mdd,
2676 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2677 S_ISREG(attr->la_mode) ? CL_CREATE :
2678 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2679 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2682 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2684 /* If creation fails, it is most likely due to the remote update
2685 * failure, because local transaction will mostly succeed at
2686 * this stage. There is no easy way to rollback all of previous
2687 * updates, so let's remove the object from namespace, and
2688 * LFSCK should handle the orphan object. */
2689 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2690 mdd_index_delete(env, mdd_pobj, attr, lname);
2694 if (is_vmalloc_addr(ldata->ld_buf))
2695 /* if we vmalloced a large buffer drop it */
2696 lu_buf_free(ldata->ld_buf);
2698 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2699 lu_buf_free(&hsm_buf);
2701 /* The child object shouldn't be cached anymore */
2703 set_bit(LU_OBJECT_HEARD_BANSHEE,
2704 &child->mo_lu.lo_header->loh_flags);
2709 * Get locks on parents in proper order
2710 * RETURN: < 0 - error, rename_order if successful
2718 static int mdd_rename_order(const struct lu_env *env,
2719 struct mdd_device *mdd,
2720 struct mdd_object *src_pobj,
2721 const struct lu_attr *pattr,
2722 struct mdd_object *tgt_pobj)
2724 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2728 if (src_pobj == tgt_pobj)
2729 RETURN(MDD_RN_SAME);
2731 /* compared the parent child relationship of src_p&tgt_p */
2732 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2734 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2737 rc = mdd_is_parent(env, mdd, src_pobj, pattr,
2751 /* has not mdd_write{read}_lock on any obj yet. */
2752 static int mdd_rename_sanity_check(const struct lu_env *env,
2753 struct mdd_object *src_pobj,
2754 const struct lu_attr *pattr,
2755 struct mdd_object *tgt_pobj,
2756 const struct lu_attr *tpattr,
2757 struct mdd_object *sobj,
2758 const struct lu_attr *cattr,
2759 struct mdd_object *tobj,
2760 const struct lu_attr *tattr)
2765 /* XXX: when get here, sobj must NOT be NULL,
2766 * the other case has been processed in cld_rename
2767 * before mdd_rename and enable MDS_PERM_BYPASS. */
2771 * If we are using project inheritance, we only allow renames
2772 * into our tree when the project IDs are the same; otherwise
2773 * tree quota mechanism would be circumvented.
2775 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2776 tpattr->la_projid != cattr->la_projid) ||
2777 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2778 (pattr->la_projid != tpattr->la_projid)))
2781 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2785 /* XXX: when get here, "tobj == NULL" means tobj must
2786 * NOT exist (neither on remote MDS, such case has been
2787 * processed in cld_rename before mdd_rename and enable
2789 * So check may_create, but not check may_unlink. */
2791 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2792 (src_pobj != tgt_pobj));
2794 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2795 (src_pobj != tgt_pobj), 1);
2797 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2798 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2803 static int mdd_declare_rename(const struct lu_env *env,
2804 struct mdd_device *mdd,
2805 struct mdd_object *mdd_spobj,
2806 struct mdd_object *mdd_tpobj,
2807 struct mdd_object *mdd_sobj,
2808 struct mdd_object *mdd_tobj,
2809 const struct lu_name *sname,
2810 const struct lu_name *tname,
2812 struct linkea_data *ldata,
2813 struct thandle *handle)
2815 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2818 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2819 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2825 /* name from source dir */
2826 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2830 /* .. from source child */
2831 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2832 /* source child can be directory,
2833 * counted by source dir's nlink */
2834 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2837 if (mdd_spobj != mdd_tpobj) {
2838 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2843 rc = mdo_declare_index_insert(env, mdd_sobj,
2845 S_IFDIR, dotdot, handle);
2850 /* new target child can be directory,
2851 * counted by target dir's nlink */
2852 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2857 la->la_valid = LA_CTIME | LA_MTIME;
2858 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2862 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2866 la->la_valid = LA_CTIME;
2867 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2871 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2876 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2877 mdd_object_type(mdd_sobj),
2878 tname->ln_name, handle);
2882 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2883 /* delete target child in target parent directory */
2884 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2889 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2893 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2894 /* target child can be directory,
2895 * delete "." reference in target child directory */
2896 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2900 /* delete ".." reference in target parent directory */
2901 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2906 la->la_valid = LA_CTIME;
2907 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2911 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2916 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2924 /* src object can be remote that is why we use only fid and type of object */
2925 static int mdd_rename(const struct lu_env *env,
2926 struct md_object *src_pobj, struct md_object *tgt_pobj,
2927 const struct lu_fid *lf, const struct lu_name *lsname,
2928 struct md_object *tobj, const struct lu_name *ltname,
2931 const char *sname = lsname->ln_name;
2932 const char *tname = ltname->ln_name;
2933 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2934 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2935 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2936 struct mdd_device *mdd = mdo2mdd(src_pobj);
2937 struct mdd_object *mdd_sobj = NULL; /* source object */
2938 struct mdd_object *mdd_tobj = NULL;
2939 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2940 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2941 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2942 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2943 struct thandle *handle;
2944 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2945 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2946 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2949 bool tobj_locked = 0;
2950 unsigned cl_flags = 0;
2954 /* let unlink to complete and commit */
2955 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
2958 mdd_tobj = md2mdd_obj(tobj);
2960 mdd_sobj = mdd_object_find(env, mdd, lf);
2961 if (IS_ERR(mdd_sobj))
2962 RETURN(PTR_ERR(mdd_sobj));
2964 rc = mdd_la_get(env, mdd_sobj, cattr);
2966 GOTO(out_pending, rc);
2968 rc = mdd_la_get(env, mdd_spobj, pattr);
2970 GOTO(out_pending, rc);
2973 rc = mdd_la_get(env, mdd_tobj, tattr);
2975 GOTO(out_pending, rc);
2976 /* search for an existing archive.
2977 * we should check ahead as the object
2978 * can be destroyed in this transaction */
2979 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2980 cl_flags |= CLF_RENAME_LAST_EXISTS;
2983 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2985 GOTO(out_pending, rc);
2987 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2988 mdd_sobj, cattr, mdd_tobj, tattr);
2990 GOTO(out_pending, rc);
2992 rc = mdd_name_check(mdd, ltname);
2994 GOTO(out_pending, rc);
2996 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2997 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2999 GOTO(out_pending, rc);
3001 handle = mdd_trans_create(env, mdd);
3003 GOTO(out_pending, rc = PTR_ERR(handle));
3005 memset(ldata, 0, sizeof(*ldata));
3006 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3007 lsname, mdd_object_fid(mdd_tpobj), ltname,
3012 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3013 mdd_tobj, lsname, ltname, ma, ldata, handle);
3017 rc = mdd_trans_start(env, mdd, handle);
3021 is_dir = S_ISDIR(cattr->la_mode);
3023 /* Remove source name from source directory */
3024 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3028 /* "mv dir1 dir2" needs "dir1/.." link update */
3029 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3030 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3032 GOTO(fixup_spobj2, rc);
3034 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3037 GOTO(fixup_spobj, rc);
3040 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3041 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3043 /* tname might been renamed to something else */
3044 GOTO(fixup_spobj, rc);
3047 /* Insert new fid with target name into target dir */
3048 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3051 GOTO(fixup_tpobj, rc);
3053 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3054 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3056 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3057 la->la_valid = LA_CTIME;
3058 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3060 GOTO(fixup_tpobj, rc);
3062 /* Update the linkEA for the source object */
3063 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3064 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3065 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3068 /* Old files might not have EA entry */
3069 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3070 lsname, handle, NULL, 0);
3071 mdd_write_unlock(env, mdd_sobj);
3072 /* We don't fail the transaction if the link ea can't be
3073 updated -- fid2path will use alternate lookup method. */
3076 /* Remove old target object
3077 * For tobj is remote case cmm layer has processed
3078 * and set tobj to NULL then. So when tobj is NOT NULL,
3079 * it must be local one.
3081 if (tobj && mdd_object_exists(mdd_tobj)) {
3082 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3084 if (mdd_is_dead_obj(mdd_tobj)) {
3085 /* shld not be dead, something is wrong */
3086 CERROR("tobj is dead, something is wrong\n");
3090 mdo_ref_del(env, mdd_tobj, handle);
3092 /* Remove dot reference. */
3093 if (S_ISDIR(tattr->la_mode))
3094 mdo_ref_del(env, mdd_tobj, handle);
3097 /* fetch updated nlink */
3098 rc = mdd_la_get(env, mdd_tobj, tattr);
3100 CERROR("%s: Failed to get nlink for tobj "
3102 mdd2obd_dev(mdd)->obd_name,
3103 PFID(tpobj_fid), rc);
3104 GOTO(fixup_tpobj, rc);
3107 la->la_valid = LA_CTIME;
3108 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3110 CERROR("%s: Failed to set ctime for tobj "
3112 mdd2obd_dev(mdd)->obd_name,
3113 PFID(tpobj_fid), rc);
3114 GOTO(fixup_tpobj, rc);
3117 /* XXX: this transfer to ma will be removed with LOD/OSP */
3118 ma->ma_attr = *tattr;
3119 ma->ma_valid |= MA_INODE;
3120 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3123 CERROR("%s: Failed to unlink tobj "
3125 mdd2obd_dev(mdd)->obd_name,
3126 PFID(tpobj_fid), rc);
3127 GOTO(fixup_tpobj, rc);
3130 /* fetch updated nlink */
3131 rc = mdd_la_get(env, mdd_tobj, tattr);
3132 if (rc == -ENOENT) {
3133 /* the object got removed, let's
3134 * return the latest known attributes */
3135 tattr->la_nlink = 0;
3137 } else if (rc != 0) {
3138 CERROR("%s: Failed to get nlink for tobj "
3140 mdd2obd_dev(mdd)->obd_name,
3141 PFID(tpobj_fid), rc);
3142 GOTO(fixup_tpobj, rc);
3144 /* XXX: this transfer to ma will be removed with LOD/OSP */
3145 ma->ma_attr = *tattr;
3146 ma->ma_valid |= MA_INODE;
3148 if (tattr->la_nlink == 0)
3149 cl_flags |= CLF_RENAME_LAST;
3151 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3154 la->la_valid = LA_CTIME | LA_MTIME;
3155 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3157 GOTO(fixup_tpobj, rc);
3159 if (mdd_spobj != mdd_tpobj) {
3160 la->la_valid = LA_CTIME | LA_MTIME;
3161 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3163 GOTO(fixup_tpobj, rc);
3170 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3172 CWARN("tp obj fix error %d\n",rc2);
3174 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3175 !mdd_is_dead_obj(mdd_tobj)) {
3177 mdo_ref_add(env, mdd_tobj, handle);
3179 mdo_ref_add(env, mdd_tobj, handle);
3182 rc2 = __mdd_index_insert(env, mdd_tpobj,
3184 mdd_object_type(mdd_tobj),
3187 CWARN("tp obj fix error: rc = %d\n", rc2);
3192 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3193 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3195 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3196 mdd2obd_dev(mdd)->obd_name, rc2);
3199 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3202 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3203 mdd2obd_dev(mdd)->obd_name, rc2);
3208 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3209 mdd_object_type(mdd_sobj), sname,
3212 CWARN("sp obj fix error: rc = %d\n", rc2);
3217 mdd_write_unlock(env, mdd_tobj);
3220 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3221 mdd_tobj, tpobj_fid, lf, spobj_fid,
3222 ltname, lsname, handle);
3225 rc = mdd_trans_stop(env, mdd, rc, handle);
3228 mdd_object_put(env, mdd_sobj);
3233 * Check whether we should migrate the file/dir
3235 * < 0 permission check failed or other error.
3236 * = 0 the file can be migrated.
3238 static int mdd_migrate_sanity_check(const struct lu_env *env,
3239 struct mdd_device *mdd,
3240 struct mdd_object *spobj,
3241 struct mdd_object *tpobj,
3242 struct mdd_object *sobj,
3243 struct mdd_object *tobj,
3244 const struct lu_attr *spattr,
3245 const struct lu_attr *tpattr,
3246 const struct lu_attr *attr)
3252 if (!mdd_object_remote(sobj)) {
3253 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3254 if (sobj->mod_count > 0) {
3255 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3256 mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(sobj)),
3258 mdd_read_unlock(env, sobj);
3261 mdd_read_unlock(env, sobj);
3264 if (mdd_object_exists(tobj))
3267 rc = mdd_rename_sanity_check(env, spobj, spattr, tpobj, tpattr, sobj,
3272 typedef int (*mdd_dir_stripe_cb)(const struct lu_env *env,
3273 struct mdd_object *obj,
3274 struct mdd_object *stripe,
3275 const struct lu_buf *lmv_buf,
3276 const struct lu_buf *lmu_buf,
3278 struct thandle *handle);
3280 static int mdd_dir_declare_delete_stripe(const struct lu_env *env,
3281 struct mdd_object *obj,
3282 struct mdd_object *stripe,
3283 const struct lu_buf *lmv_buf,
3284 const struct lu_buf *lmu_buf,
3286 struct thandle *handle)
3288 struct mdd_thread_info *info = mdd_env_info(env);
3289 char *stripe_name = info->mti_name;
3290 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3293 if (index < le32_to_cpu(lmu->lum_stripe_count))
3296 rc = mdo_declare_index_delete(env, stripe, dotdot, handle);
3300 snprintf(stripe_name, sizeof(info->mti_name), DFID":%d",
3301 PFID(mdd_object_fid(stripe)), index);
3303 rc = mdo_declare_index_delete(env, obj, stripe_name, handle);
3307 rc = mdo_declare_ref_del(env, obj, handle);
3312 /* delete stripe from its master object namespace */
3313 static int mdd_dir_delete_stripe(const struct lu_env *env,
3314 struct mdd_object *obj,
3315 struct mdd_object *stripe,
3316 const struct lu_buf *lmv_buf,
3317 const struct lu_buf *lmu_buf,
3319 struct thandle *handle)
3321 struct mdd_thread_info *info = mdd_env_info(env);
3322 char *stripe_name = info->mti_name;
3323 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3324 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3325 __u32 del_offset = le32_to_cpu(lmu->lum_stripe_count);
3330 /* local dir will delete via LOD */
3331 LASSERT(mdd_object_remote(obj));
3332 LASSERT(del_offset < le32_to_cpu(lmv->lmv_stripe_count));
3334 if (index < del_offset)
3337 mdd_write_lock(env, stripe, DT_SRC_CHILD);
3338 rc = __mdd_index_delete_only(env, stripe, dotdot, handle);
3342 snprintf(stripe_name, sizeof(info->mti_name), DFID":%d",
3343 PFID(mdd_object_fid(stripe)), index);
3345 rc = __mdd_index_delete_only(env, obj, stripe_name, handle);
3349 rc = mdo_ref_del(env, obj, handle);
3352 mdd_write_unlock(env, stripe);
3357 static int mdd_dir_declare_destroy_stripe(const struct lu_env *env,
3358 struct mdd_object *obj,
3359 struct mdd_object *stripe,
3360 const struct lu_buf *lmv_buf,
3361 const struct lu_buf *lmu_buf,
3363 struct thandle *handle)
3365 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3366 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3369 if (index < shrink_offset) {
3370 if (shrink_offset < 2)
3372 return mdo_declare_xattr_set(env, stripe, lmv_buf,
3373 XATTR_NAME_LMV".set", 0, handle);
3376 rc = mdo_declare_ref_del(env, stripe, handle);
3380 rc = mdo_declare_destroy(env, stripe, handle);
3385 static int mdd_dir_destroy_stripe(const struct lu_env *env,
3386 struct mdd_object *obj,
3387 struct mdd_object *stripe,
3388 const struct lu_buf *lmv_buf,
3389 const struct lu_buf *lmu_buf,
3391 struct thandle *handle)
3393 struct mdd_thread_info *info = mdd_env_info(env);
3394 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3395 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3396 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3401 /* update remaining stripes' LMV */
3402 if (index < shrink_offset) {
3403 struct lmv_mds_md_v1 *slave_lmv;
3404 struct lu_buf slave_buf = {
3405 .lb_buf = &info->mti_lmv.lmv_md_v1,
3406 .lb_len = sizeof(*slave_lmv)
3408 __u32 version = le32_to_cpu(lmv->lmv_layout_version);
3410 /* if dir will be shrunk to 1-stripe, don't update */
3411 if (shrink_offset < 2)
3414 slave_lmv = slave_buf.lb_buf;
3415 memset(slave_lmv, 0, sizeof(*slave_lmv));
3416 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
3417 slave_lmv->lmv_stripe_count = lmu->lum_stripe_count;
3418 slave_lmv->lmv_master_mdt_index = cpu_to_le32(index);
3419 slave_lmv->lmv_hash_type = lmv->lmv_hash_type &
3420 cpu_to_le32(LMV_HASH_TYPE_MASK);
3421 slave_lmv->lmv_layout_version = cpu_to_le32(++version);
3423 rc = mdo_xattr_set(env, stripe, &slave_buf,
3424 XATTR_NAME_LMV".set", 0, handle);
3428 mdd_write_lock(env, stripe, DT_SRC_CHILD);
3429 rc = mdo_ref_del(env, stripe, handle);
3431 rc = mdo_destroy(env, stripe, handle);
3432 mdd_write_unlock(env, stripe);
3437 static int mdd_shrink_stripe_is_empty(const struct lu_env *env,
3438 struct mdd_object *obj,
3439 struct mdd_object *stripe,
3440 const struct lu_buf *lmv_buf,
3441 const struct lu_buf *lmu_buf,
3443 struct thandle *handle)
3445 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3446 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3448 /* the default value is 0, but it means 1 */
3452 if (index < shrink_offset)
3455 return mdd_dir_is_empty(env, stripe);
3459 * iterate stripes of striped directory on remote MDT, local striped directory
3460 * is accessed via LOD.
3462 static int mdd_dir_iterate_stripes(const struct lu_env *env,
3463 struct mdd_object *obj,
3464 const struct lu_buf *lmv_buf,
3465 const struct lu_buf *lmu_buf,
3466 struct thandle *handle,
3467 mdd_dir_stripe_cb cb)
3469 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
3470 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3471 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3472 struct mdd_object *stripe;
3480 for (i = 0; i < le32_to_cpu(lmv->lmv_stripe_count); i++) {
3481 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[i]);
3482 if (!fid_is_sane(fid))
3485 stripe = mdd_object_find(env, mdd, fid);
3487 RETURN(PTR_ERR(stripe));
3489 rc = cb(env, obj, stripe, lmv_buf, lmu_buf, i, handle);
3490 mdd_object_put(env, stripe);
3498 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3499 struct mdd_object *obj,
3500 const struct lu_buf *buf,
3502 int fl, struct thandle *handle);
3504 /* iterate xattrs, but ignore LMA, LMV, and LINKEA if 'skip_linkea' is set. */
3505 static int mdd_iterate_xattrs(const struct lu_env *env,
3506 struct mdd_object *sobj,
3507 struct mdd_object *tobj,
3509 struct thandle *handle,
3512 struct mdd_thread_info *info = mdd_env_info(env);
3514 struct lu_buf list_xbuf;
3515 struct lu_buf cbxbuf;
3516 struct lu_buf xbuf = { NULL };
3525 /* retrieve xattr list from the old object */
3526 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3527 if (list_xsize == -ENODATA)
3533 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3534 if (info->mti_big_buf.lb_buf == NULL)
3537 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3538 list_xbuf.lb_len = list_xsize;
3539 rc = mdo_xattr_list(env, sobj, &list_xbuf);
3545 xname = list_xbuf.lb_buf;
3547 xlen = strnlen(xname, rem - 1) + 1;
3548 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3549 strcmp(XATTR_NAME_LMV, xname) == 0)
3553 strcmp(XATTR_NAME_LINK, xname) == 0)
3556 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3557 if (xsize == -ENODATA)
3560 GOTO(out, rc = xsize);
3562 lu_buf_check_and_alloc(&xbuf, xsize);
3563 if (xbuf.lb_buf == NULL)
3564 GOTO(out, rc = -ENOMEM);
3566 rc = mdo_xattr_get(env, sobj, &xbuf, xname);
3574 rc = cb(env, tobj, &cbxbuf, xname, 0, handle);
3575 if (unlikely(rc == -ENOSPC &&
3576 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3577 rc = linkea_overflow_shrink(
3578 (struct linkea_data *)(cbxbuf.lb_buf));
3579 if (likely(rc > 0)) {
3597 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3598 struct mdd_object *sobj,
3599 struct mdd_object *tobj,
3600 const struct lu_name *sname,
3601 const struct lu_fid *sfid,
3602 const struct lu_name *lname,
3603 const struct lu_fid *fid,
3605 struct thandle *handle);
3607 static int mdd_declare_update_link(const struct lu_env *env,
3608 struct mdd_object *sobj,
3609 struct mdd_object *tobj,
3610 const struct lu_name *tname,
3611 const struct lu_fid *tpfid,
3612 const struct lu_name *lname,
3613 const struct lu_fid *fid,
3615 struct thandle *handle)
3617 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3618 struct mdd_object *pobj;
3622 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3623 !strcmp(tname->ln_name, lname->ln_name))
3626 pobj = mdd_object_find(env, mdd, fid);
3628 return PTR_ERR(pobj);
3631 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3633 rc = mdo_declare_index_insert(env, pobj, mdo2fid(tobj),
3634 mdd_object_type(sobj),
3635 lname->ln_name, handle);
3636 mdd_object_put(env, pobj);
3640 rc = mdo_declare_ref_add(env, tobj, handle);
3644 rc = mdo_declare_ref_del(env, sobj, handle);
3648 static int mdd_update_link(const struct lu_env *env,
3649 struct mdd_object *sobj,
3650 struct mdd_object *tobj,
3651 const struct lu_name *tname,
3652 const struct lu_fid *tpfid,
3653 const struct lu_name *lname,
3654 const struct lu_fid *fid,
3656 struct thandle *handle)
3658 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3659 struct mdd_object *pobj;
3664 LASSERT(lu_name_is_valid(lname));
3667 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3668 !strncmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3671 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3672 PFID(fid), PNAME(lname), PFID(mdo2fid(tobj)));
3674 pobj = mdd_object_find(env, mdd, fid);
3676 CWARN("%s: cannot find obj "DFID": %ld\n",
3677 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3678 RETURN(PTR_ERR(pobj));
3681 if (!mdd_object_exists(pobj)) {
3682 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3683 mdd_object_put(env, pobj);
3687 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3688 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3690 rc = __mdd_index_insert_only(env, pobj, mdo2fid(tobj),
3691 mdd_object_type(sobj),
3692 lname->ln_name, handle);
3693 mdd_write_unlock(env, pobj);
3694 mdd_object_put(env, pobj);
3698 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3699 rc = mdo_ref_add(env, tobj, handle);
3700 mdd_write_unlock(env, tobj);
3704 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3705 rc = mdo_ref_del(env, sobj, handle);
3706 mdd_write_unlock(env, sobj);
3711 static inline int mdd_fld_lookup(const struct lu_env *env,
3712 struct mdd_device *mdd,
3713 const struct lu_fid *fid,
3716 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
3717 struct seq_server_site *ss;
3720 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3722 range->lsr_flags = LU_SEQ_RANGE_MDT;
3723 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3727 *mdt_index = range->lsr_index;
3732 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3733 struct mdd_object *sobj,
3734 struct mdd_object *tobj,
3735 const struct lu_name *tname,
3736 const struct lu_fid *tpfid,
3737 const struct lu_name *lname,
3738 const struct lu_fid *fid,
3740 struct thandle *handle)
3742 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3743 __u32 source_mdt_index = *(__u32 *)opaque;
3744 __u32 link_mdt_index;
3750 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3751 !strcmp(tname->ln_name, lname->ln_name))
3754 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3758 RETURN(link_mdt_index == source_mdt_index);
3761 static int mdd_iterate_linkea(const struct lu_env *env,
3762 struct mdd_object *sobj,
3763 struct mdd_object *tobj,
3764 const struct lu_name *tname,
3765 const struct lu_fid *tpfid,
3766 struct linkea_data *ldata,
3768 struct thandle *handle,
3771 struct mdd_thread_info *info = mdd_env_info(env);
3772 char *filename = info->mti_name;
3773 struct lu_name lname;
3780 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3781 linkea_next_entry(ldata)) {
3782 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3785 /* Note: lname might miss \0 at the end */
3786 snprintf(filename, sizeof(info->mti_name), "%.*s",
3787 lname.ln_namelen, lname.ln_name);
3788 lname.ln_name = filename;
3790 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3792 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3800 * Prepare linkea, and check whether file needs migrate: if source still has
3801 * link on source MDT, no need to migrate, just update namespace on source and
3804 * \retval 0 do migrate
3805 * \retval 1 don't migrate
3806 * \retval -errno on failure
3808 static int migrate_linkea_prepare(const struct lu_env *env,
3809 struct mdd_device *mdd,
3810 struct mdd_object *spobj,
3811 struct mdd_object *tpobj,
3812 struct mdd_object *sobj,
3813 const struct lu_name *lname,
3814 const struct lu_attr *attr,
3815 struct linkea_data *ldata)
3817 __u32 source_mdt_index;
3822 memset(ldata, 0, sizeof(*ldata));
3823 rc = mdd_linkea_prepare(env, sobj, mdo2fid(spobj), lname,
3824 mdo2fid(tpobj), lname, 1, 0, ldata);
3829 * Then it will check if the file should be migrated. If the file has
3830 * mulitple links, we only need migrate the file if all of its entries
3831 * has been migrated to the remote MDT.
3833 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3836 /* If there are still links locally, don't migrate this file */
3837 LASSERT(ldata->ld_leh != NULL);
3840 * If linkEA is overflow, it means there are some unknown name entries
3841 * under unknown parents, which will prevent the migration.
3843 if (unlikely(ldata->ld_leh->leh_overflow_time))
3846 rc = mdd_fld_lookup(env, mdd, mdo2fid(sobj), &source_mdt_index);
3850 rc = mdd_iterate_linkea(env, sobj, NULL, lname, mdo2fid(tpobj), ldata,
3851 &source_mdt_index, NULL,
3852 mdd_is_link_on_source_mdt);
3856 static int mdd_dir_declare_layout_delete(const struct lu_env *env,
3857 struct mdd_object *obj,
3858 const struct lu_buf *lmv_buf,
3859 const struct lu_buf *lmu_buf,
3860 struct thandle *handle)
3864 if (!lmv_buf->lb_buf)
3865 rc = mdo_declare_index_delete(env, obj, dotdot, handle);
3866 else if (mdd_object_remote(obj))
3867 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, lmu_buf, handle,
3868 mdd_dir_declare_delete_stripe);
3870 rc = mdo_declare_xattr_set(env, obj, lmu_buf,
3871 XATTR_NAME_LMV".del", 0, handle);
3876 static int mdd_dir_layout_delete(const struct lu_env *env,
3877 struct mdd_object *obj,
3878 const struct lu_buf *lmv_buf,
3879 const struct lu_buf *lmu_buf,
3880 struct thandle *handle)
3886 mdd_write_lock(env, obj, DT_SRC_PARENT);
3887 if (!lmv_buf->lb_buf)
3889 rc = __mdd_index_delete_only(env, obj, dotdot, handle);
3890 else if (mdd_object_remote(obj))
3891 /* striped, but remote */
3892 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, lmu_buf, handle,
3893 mdd_dir_delete_stripe);
3895 rc = mdo_xattr_set(env, obj, lmu_buf, XATTR_NAME_LMV".del", 0,
3897 mdd_write_unlock(env, obj);
3902 static int mdd_declare_migrate_create(const struct lu_env *env,
3903 struct mdd_object *tpobj,
3904 struct mdd_object *sobj,
3905 struct mdd_object *tobj,
3906 const struct lu_name *lname,
3907 struct lu_attr *attr,
3908 struct lu_buf *sbuf,
3909 struct linkea_data *ldata,
3910 struct md_op_spec *spec,
3911 struct dt_allocation_hint *hint,
3912 struct thandle *handle)
3914 struct mdd_thread_info *info = mdd_env_info(env);
3915 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3918 if (S_ISDIR(attr->la_mode)) {
3919 struct lu_buf lmu_buf = { NULL };
3922 struct lmv_user_md *lmu = &info->mti_lmv.lmv_user_md;
3924 lmu->lum_stripe_count = 0;
3925 lmu_buf.lb_buf = lmu;
3926 lmu_buf.lb_len = sizeof(*lmu);
3929 rc = mdd_dir_declare_layout_delete(env, sobj, sbuf, &lmu_buf,
3935 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LMV,
3942 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3943 lname, attr, handle, spec, ldata, NULL, NULL,
3948 if (S_ISDIR(attr->la_mode) && mdd_dir_is_empty(env, sobj) != 0) {
3951 * if sobj is not striped, fake a 1-stripe LMV, which
3952 * will be used to generate a compound LMV for tobj.
3954 LASSERT(sizeof(info->mti_key) >
3955 lmv_mds_md_size(1, LMV_MAGIC_V1));
3956 lmv = (typeof(lmv))info->mti_key;
3957 memset(lmv, 0, sizeof(*lmv));
3958 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3959 lmv->lmv_stripe_count = cpu_to_le32(1);
3960 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdo2fid(sobj));
3962 sbuf->lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3964 rc = mdo_declare_xattr_set(env, tobj, sbuf,
3965 XATTR_NAME_LMV".add", 0,
3967 sbuf->lb_buf = NULL;
3970 rc = mdo_declare_xattr_set(env, tobj, sbuf,
3971 XATTR_NAME_LMV".add", 0,
3979 * tobj mode will be used in lod_declare_xattr_set(), but it's not
3980 * createb yet, copy from sobj.
3982 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
3983 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
3984 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
3986 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle,
3987 mdo_declare_xattr_set);
3991 if (S_ISREG(attr->la_mode)) {
3992 struct lu_buf fid_buf;
3994 handle->th_complex = 1;
3996 /* target may be remote, update PFID via sobj. */
3997 fid_buf.lb_buf = (void *)mdo2fid(tobj);
3998 fid_buf.lb_len = sizeof(struct lu_fid);
3999 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
4004 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4009 if (!S_ISDIR(attr->la_mode)) {
4010 rc = mdd_iterate_linkea(env, sobj, tobj, lname, mdo2fid(tpobj),
4011 ldata, NULL, handle,
4012 mdd_declare_update_link);
4017 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LMV,
4028 * Create target, migrate xattrs and update links.
4030 * Create target according to \a spec, and then migrate xattrs, if it's
4031 * directory, migrate source stripes to target, else update fid to target
4034 * \param[in] env execution environment
4035 * \param[in] tpobj target parent object
4036 * \param[in] sobj source object
4037 * \param[in] tobj target object
4038 * \param[in] lname file name
4039 * \param[in] attr source attributes
4040 * \param[in] sbuf source LMV buf
4041 * \param[in] ldata source linkea
4042 * \param[in] spec migrate create spec
4043 * \param[in] hint target creation hint
4044 * \param[in] handle tranasction handle
4046 * \retval 0 on success
4047 * \retval -errno on failure
4049 static int mdd_migrate_create(const struct lu_env *env,
4050 struct mdd_object *tpobj,
4051 struct mdd_object *sobj,
4052 struct mdd_object *tobj,
4053 const struct lu_name *lname,
4054 struct lu_attr *attr,
4055 const struct lu_buf *sbuf,
4056 struct linkea_data *ldata,
4057 struct md_op_spec *spec,
4058 struct dt_allocation_hint *hint,
4059 struct thandle *handle)
4066 * directory will migrate sobj stripes to tobj:
4067 * 1. delete stripes from sobj.
4068 * 2. add stripes to tobj, see lod_dir_declare_layout_add().
4069 * 3. create/attach stripes for tobj, see lod_xattr_set_lmv().
4071 if (S_ISDIR(attr->la_mode)) {
4072 struct lu_buf lmu_buf = { NULL };
4075 struct mdd_thread_info *info = mdd_env_info(env);
4076 struct lmv_user_md *lmu = &info->mti_lmv.lmv_user_md;
4078 lmu->lum_stripe_count = 0;
4079 lmu_buf.lb_buf = lmu;
4080 lmu_buf.lb_len = sizeof(*lmu);
4083 rc = mdd_dir_layout_delete(env, sobj, sbuf, &lmu_buf, handle);
4088 * delete LMV so that later when destroying sobj it won't delete
4092 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4093 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LMV, handle);
4094 mdd_write_unlock(env, sobj);
4100 /* don't set nlink from sobj */
4101 attr->la_valid &= ~LA_NLINK;
4103 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4108 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4109 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
4110 mdd_write_unlock(env, tobj);
4114 /* for regular file, update OST objects XATTR_NAME_FID */
4115 if (S_ISREG(attr->la_mode)) {
4116 struct lu_buf fid_buf;
4118 /* target may be remote, update PFID via sobj. */
4119 fid_buf.lb_buf = (void *)mdo2fid(tobj);
4120 fid_buf.lb_len = sizeof(struct lu_fid);
4121 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4126 /* delete LOV to avoid deleting OST objs when destroying sobj */
4127 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4128 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4129 mdd_write_unlock(env, sobj);
4134 if (!S_ISDIR(attr->la_mode))
4135 rc = mdd_iterate_linkea(env, sobj, tobj, lname, mdo2fid(tpobj),
4136 ldata, NULL, handle, mdd_update_link);
4141 static int mdd_declare_migrate_update(const struct lu_env *env,
4142 struct mdd_object *spobj,
4143 struct mdd_object *tpobj,
4144 struct mdd_object *sobj,
4145 struct mdd_object *tobj,
4146 const struct lu_name *lname,
4147 struct lu_attr *attr,
4148 struct lu_attr *spattr,
4149 struct lu_attr *tpattr,
4150 struct linkea_data *ldata,
4154 struct thandle *handle)
4156 struct mdd_thread_info *info = mdd_env_info(env);
4157 const struct lu_fid *fid = mdo2fid(do_create ? tobj : sobj);
4158 struct lu_attr *la = &info->mti_la_for_fix;
4161 rc = mdo_declare_index_delete(env, spobj, lname->ln_name, handle);
4165 if (S_ISDIR(attr->la_mode)) {
4166 rc = mdo_declare_ref_del(env, spobj, handle);
4171 rc = mdo_declare_index_insert(env, tpobj, fid, mdd_object_type(sobj),
4172 lname->ln_name, handle);
4176 rc = mdd_declare_links_add(env, do_create ? tobj : sobj, handle, ldata);
4180 if (S_ISDIR(attr->la_mode)) {
4181 rc = mdo_declare_ref_add(env, tpobj, handle);
4186 la->la_valid = LA_CTIME | LA_MTIME;
4187 rc = mdo_declare_attr_set(env, spobj, la, handle);
4191 if (tpobj != spobj) {
4192 rc = mdo_declare_attr_set(env, tpobj, la, handle);
4197 if (do_create && do_destroy) {
4198 rc = mdo_declare_ref_del(env, sobj, handle);
4202 rc = mdo_declare_destroy(env, sobj, handle);
4211 * migrate dirent from \a spobj to \a tpobj, and destroy \a sobj
4213 static int mdd_migrate_update(const struct lu_env *env,
4214 struct mdd_object *spobj,
4215 struct mdd_object *tpobj,
4216 struct mdd_object *sobj,
4217 struct mdd_object *tobj,
4218 const struct lu_name *lname,
4219 struct lu_attr *attr,
4220 struct lu_attr *spattr,
4221 struct lu_attr *tpattr,
4222 struct linkea_data *ldata,
4226 struct thandle *handle)
4228 struct mdd_thread_info *info = mdd_env_info(env);
4229 const struct lu_fid *fid = mdo2fid(do_create ? tobj : sobj);
4230 struct lu_attr *la = &info->mti_la_for_fix;
4235 CDEBUG(D_INFO, "update %s "DFID"/"DFID" to "DFID"/"DFID"\n",
4236 lname->ln_name, PFID(mdo2fid(spobj)),
4237 PFID(mdo2fid(sobj)), PFID(mdo2fid(tpobj)),
4240 rc = __mdd_index_delete(env, spobj, lname->ln_name,
4241 S_ISDIR(attr->la_mode), handle);
4245 rc = __mdd_index_insert(env, tpobj, fid, mdd_object_type(sobj),
4246 lname->ln_name, handle);
4250 rc = mdd_links_write(env, do_create ? tobj : sobj, ldata, handle);
4254 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
4255 la->la_valid = LA_CTIME | LA_MTIME;
4256 mdd_write_lock(env, spobj, DT_SRC_PARENT);
4257 rc = mdd_update_time(env, spobj, spattr, la, handle);
4258 mdd_write_unlock(env, spobj);
4262 if (tpobj != spobj) {
4263 la->la_valid = LA_CTIME | LA_MTIME;
4264 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
4265 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
4266 mdd_write_unlock(env, tpobj);
4272 * there are three situations we shouldn't destroy source:
4273 * 1. if source is not dir, and it happens to be located on the same MDT
4275 * 2. if source is not dir, and has link on the same MDT where source is
4277 * 3. if source is dir, and it's a normal, non-empty dir.
4279 * the first two situations equals to !do_create, and the 3rd equals to
4280 * !do_destroy, so the below condition is actually
4281 * !(!do_create || !do_destroy).
4283 * NB, if user has opened source dir before migration, he will get
4284 * -ENOENT error when close it later, because source is likely to be
4285 * remote, which can't be moved to orphan list, but except this error
4286 * message, this won't cause any inconsistency or trouble.
4288 if (do_create && do_destroy) {
4289 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4290 mdo_ref_del(env, sobj, handle);
4291 rc = mdo_destroy(env, sobj, handle);
4292 mdd_write_unlock(env, sobj);
4299 * Migrate directory or file.
4301 * migrate source to target in following steps:
4302 * 1. create target, append source stripes after target's if it's directory,
4303 * migrate xattrs and update fid of source links.
4304 * 2. update namespace: migrate dirent from source parent to target parent,
4305 * update file linkea, and destroy source if it's not needed any more.
4307 * \param[in] env execution environment
4308 * \param[in] md_pobj parent master object
4309 * \param[in] md_sobj source object
4310 * \param[in] lname file name
4311 * \param[in] md_tobj target object
4312 * \param[in] spec target creation spec
4313 * \param[in] ma used to update \a pobj mtime and ctime
4315 * \retval 0 on success
4316 * \retval -errno on failure
4318 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4319 struct md_object *md_sobj, const struct lu_name *lname,
4320 struct md_object *md_tobj, struct md_op_spec *spec,
4323 struct mdd_device *mdd = mdo2mdd(md_pobj);
4324 struct mdd_thread_info *info = mdd_env_info(env);
4325 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4326 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4327 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4328 struct mdd_object *spobj = NULL;
4329 struct mdd_object *tpobj = NULL;
4330 struct lu_attr *spattr = &info->mti_pattr;
4331 struct lu_attr *tpattr = &info->mti_tpattr;
4332 struct lu_attr *attr = &info->mti_cattr;
4333 struct linkea_data *ldata = &info->mti_link_data;
4334 struct dt_allocation_hint *hint = &info->mti_hint;
4335 struct lu_fid *fid = &info->mti_fid2;
4336 struct lu_buf pbuf = { NULL };
4337 struct lu_buf sbuf = { NULL };
4338 struct lmv_mds_md_v1 *plmv;
4339 struct thandle *handle;
4340 bool do_create = true;
4341 bool do_destroy = true;
4345 rc = mdd_la_get(env, sobj, attr);
4349 /* locate source and target stripe on pobj, which are the real parent */
4350 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4351 if (rc < 0 && rc != -ENODATA)
4356 __u32 hash_type = le32_to_cpu(plmv->lmv_hash_type);
4357 __u32 count = le32_to_cpu(plmv->lmv_stripe_count);
4360 /* locate target parent stripe */
4361 if (hash_type & LMV_HASH_FLAG_MIGRATION) {
4363 * fail check here to make sure top dir migration
4366 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4367 GOTO(out, rc = -EIO);
4368 hash_type &= ~LMV_HASH_FLAG_MIGRATION;
4369 count = le32_to_cpu(plmv->lmv_migrate_offset);
4371 index = lmv_name_to_stripe_index(hash_type, count,
4375 GOTO(out, rc = index);
4377 fid_le_to_cpu(fid, &plmv->lmv_stripe_fids[index]);
4378 tpobj = mdd_object_find(env, mdd, fid);
4380 GOTO(out, rc = PTR_ERR(tpobj));
4382 /* locate source parent stripe */
4383 if (le32_to_cpu(plmv->lmv_hash_type) &
4384 LMV_HASH_FLAG_MIGRATION) {
4385 hash_type = le32_to_cpu(plmv->lmv_migrate_hash);
4386 count = le32_to_cpu(plmv->lmv_stripe_count) -
4387 le32_to_cpu(plmv->lmv_migrate_offset);
4389 index = lmv_name_to_stripe_index(hash_type, count,
4393 mdd_object_put(env, tpobj);
4394 GOTO(out, rc = index);
4397 index += le32_to_cpu(plmv->lmv_migrate_offset);
4398 fid_le_to_cpu(fid, &plmv->lmv_stripe_fids[index]);
4399 spobj = mdd_object_find(env, mdd, fid);
4400 if (IS_ERR(spobj)) {
4401 mdd_object_put(env, tpobj);
4402 GOTO(out, rc = PTR_ERR(spobj));
4406 mdd_object_get(spobj);
4411 mdd_object_get(tpobj);
4412 mdd_object_get(spobj);
4415 rc = mdd_la_get(env, spobj, spattr);
4419 rc = mdd_la_get(env, tpobj, tpattr);
4423 if (S_ISDIR(attr->la_mode)) {
4424 struct lmv_user_md_v1 *lmu = spec->u.sp_ea.eadata;
4429 * if user use default value '0' for stripe_count, we need to
4430 * adjust it to '1' to create a 1-stripe directory.
4432 if (lmu->lum_stripe_count == 0) {
4433 /* eadata is from request, don't alter it */
4434 info->mti_lmu = *lmu;
4435 info->mti_lmu.lum_stripe_count = cpu_to_le32(1);
4436 spec->u.sp_ea.eadata = &info->mti_lmu;
4437 lmu = spec->u.sp_ea.eadata;
4440 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4441 if (rc == -ENODATA) {
4442 if (mdd_dir_is_empty(env, sobj) == 0) {
4444 * if sobj is empty, and target is not striped,
4445 * create target as a normal directory.
4447 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4448 info->mti_lmu = *lmu;
4449 info->mti_lmu.lum_stripe_count = 0;
4450 spec->u.sp_ea.eadata = &info->mti_lmu;
4451 lmu = spec->u.sp_ea.eadata;
4455 * sobj is not striped dir, if it's not empty,
4456 * it will be migrated to be a stripe of target,
4457 * don't destroy it after migration.
4464 struct lmv_mds_md_v1 *lmv = sbuf.lb_buf;
4466 if (le32_to_cpu(lmv->lmv_hash_type) &
4467 LMV_HASH_FLAG_MIGRATION) {
4468 __u32 lum_stripe_count = lmu->lum_stripe_count;
4469 __u32 lmv_hash_type = lmv->lmv_hash_type &
4470 cpu_to_le32(LMV_HASH_TYPE_MASK);
4472 if (!lum_stripe_count)
4473 lum_stripe_count = cpu_to_le32(1);
4475 /* TODO: check specific MDTs */
4476 if (lmv->lmv_migrate_offset !=
4478 lmv->lmv_master_mdt_index !=
4479 lmu->lum_stripe_offset ||
4480 (lmv_hash_type != 0 &&
4481 lmv_hash_type != lmu->lum_hash_type)) {
4482 CERROR("%s: \'"DNAME"\' migration was "
4483 "interrupted, run \'lfs migrate "
4484 "-m %d -c %d -H %d "DNAME"\' to "
4485 "finish migration.\n",
4486 mdd2obd_dev(mdd)->obd_name,
4489 lmv->lmv_master_mdt_index),
4491 lmv->lmv_migrate_offset),
4492 le32_to_cpu(lmv_hash_type),
4494 GOTO(out, rc = -EPERM);
4496 GOTO(out, rc = -EALREADY);
4499 } else if (!mdd_object_remote(tpobj)) {
4501 * if source is already on MDT where target parent is located,
4502 * no need to create, just update namespace.
4505 } else if (S_ISLNK(attr->la_mode)) {
4506 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4508 GOTO(out, rc = -ENOMEM);
4509 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4512 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4513 mdd2obd_dev(mdd)->obd_name,
4514 PFID(mdo2fid(sobj)), rc);
4517 spec->u.sp_symname = sbuf.lb_buf;
4518 } else if (S_ISREG(attr->la_mode)) {
4519 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
4520 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
4524 * if sobj has link on the same MDT, no need to create, just update
4525 * namespace, and it will be a remote file on target parent, which is
4526 * similar to rename.
4528 rc = migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, lname, attr,
4535 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4536 spattr, tpattr, attr);
4540 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
4542 handle = mdd_trans_create(env, mdd);
4544 GOTO(out, rc = PTR_ERR(handle));
4547 rc = mdd_declare_migrate_create(env, tpobj, sobj, tobj, lname,
4548 attr, &sbuf, ldata, spec, hint,
4551 GOTO(stop_trans, rc);
4554 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, tobj, lname,
4555 attr, spattr, tpattr, ldata, do_create,
4556 do_destroy, ma, handle);
4558 GOTO(stop_trans, rc);
4560 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4563 GOTO(stop_trans, rc);
4565 rc = mdd_trans_start(env, mdd, handle);
4567 GOTO(stop_trans, rc);
4570 rc = mdd_migrate_create(env, tpobj, sobj, tobj, lname, attr,
4571 &sbuf, ldata, spec, hint, handle);
4573 GOTO(stop_trans, rc);
4576 rc = mdd_migrate_update(env, spobj, tpobj, sobj, tobj, lname, attr,
4577 spattr, tpattr, ldata, do_create, do_destroy,
4580 GOTO(stop_trans, rc);
4582 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
4583 mdo2fid(spobj), mdo2fid(sobj),
4584 mdo2fid(tpobj), lname, lname, handle);
4586 GOTO(stop_trans, rc);
4590 rc = mdd_trans_stop(env, mdd, rc, handle);
4592 if (spobj && !IS_ERR(spobj))
4593 mdd_object_put(env, spobj);
4594 if (tpobj && !IS_ERR(tpobj))
4595 mdd_object_put(env, tpobj);
4601 static int __mdd_dir_declare_layout_shrink(const struct lu_env *env,
4602 struct mdd_object *pobj,
4603 struct mdd_object *obj,
4604 struct mdd_object *stripe,
4605 struct lu_attr *attr,
4606 struct lu_buf *lmv_buf,
4607 const struct lu_buf *lmu_buf,
4608 struct lu_name *lname,
4609 struct thandle *handle)
4611 struct mdd_thread_info *info = mdd_env_info(env);
4612 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
4613 struct lmv_user_md *lmu = (typeof(lmu))info->mti_key;
4614 struct lu_buf shrink_buf = { .lb_buf = lmu,
4615 .lb_len = sizeof(*lmu) };
4620 memcpy(lmu, lmu_buf->lb_buf, sizeof(*lmu));
4622 if (le32_to_cpu(lmu->lum_stripe_count) < 2)
4623 lmu->lum_stripe_count = 0;
4625 rc = mdd_dir_declare_layout_delete(env, obj, lmv_buf, &shrink_buf,
4630 if (lmu->lum_stripe_count == 0) {
4631 lmu->lum_stripe_count = cpu_to_le32(1);
4633 rc = mdo_declare_xattr_del(env, obj, XATTR_NAME_LMV, handle);
4638 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
4639 mdd_dir_declare_destroy_stripe);
4643 if (le32_to_cpu(lmu->lum_stripe_count) > 1)
4644 return mdo_declare_xattr_set(env, obj, lmv_buf,
4645 XATTR_NAME_LMV".set", 0, handle);
4647 rc = mdo_declare_index_insert(env, stripe, mdo2fid(pobj), S_IFDIR,
4652 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle,
4653 mdo_declare_xattr_set);
4657 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4661 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4665 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4669 rc = mdo_declare_index_insert(env, pobj, mdo2fid(stripe), attr->la_mode,
4670 lname->ln_name, handle);
4674 rc = mdo_declare_ref_del(env, obj, handle);
4678 rc = mdo_declare_ref_del(env, obj, handle);
4682 rc = mdo_declare_destroy(env, obj, handle);
4691 * after files under \a obj were migrated, shrink old stripes from \a obj,
4692 * furthermore, if it becomes a 1-stripe directory, convert it to a normal one.
4694 static int __mdd_dir_layout_shrink(const struct lu_env *env,
4695 struct mdd_object *pobj,
4696 struct mdd_object *obj,
4697 struct mdd_object *stripe,
4698 struct lu_attr *attr,
4699 struct lu_buf *lmv_buf,
4700 const struct lu_buf *lmu_buf,
4701 struct lu_name *lname,
4702 struct thandle *handle)
4704 struct mdd_thread_info *info = mdd_env_info(env);
4705 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
4706 struct lmv_user_md *lmu = (typeof(lmu))info->mti_key;
4707 struct lu_buf shrink_buf = { .lb_buf = lmu,
4708 .lb_len = sizeof(*lmu) };
4709 int len = lmv_buf->lb_len;
4710 __u32 version = le32_to_cpu(lmv->lmv_layout_version);
4715 /* lmu needs to be altered, but lmu_buf is const */
4716 memcpy(lmu, lmu_buf->lb_buf, sizeof(*lmu));
4719 * if dir will be shrunk to 1-stripe, delete all stripes, because it
4720 * will be converted to normal dir.
4722 if (le32_to_cpu(lmu->lum_stripe_count) == 1)
4723 lmu->lum_stripe_count = 0;
4725 /* delete stripes after lmu_stripe_count */
4726 rc = mdd_dir_layout_delete(env, obj, lmv_buf, &shrink_buf, handle);
4730 if (lmu->lum_stripe_count == 0) {
4731 lmu->lum_stripe_count = cpu_to_le32(1);
4733 /* delete LMV to avoid deleting stripes again upon destroy */
4734 mdd_write_lock(env, obj, DT_SRC_CHILD);
4735 rc = mdo_xattr_del(env, obj, XATTR_NAME_LMV, handle);
4736 mdd_write_unlock(env, obj);
4741 /* destroy stripes after lmu_stripe_count */
4742 mdd_write_lock(env, obj, DT_SRC_PARENT);
4743 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
4744 mdd_dir_destroy_stripe);
4745 mdd_write_unlock(env, obj);
4747 if (le32_to_cpu(lmu->lum_stripe_count) > 1) {
4748 /* update dir LMV, that's all if it's still striped. */
4749 lmv->lmv_stripe_count = lmu->lum_stripe_count;
4750 lmv->lmv_hash_type &= ~cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4751 lmv->lmv_migrate_offset = 0;
4752 lmv->lmv_migrate_hash = 0;
4753 lmv->lmv_layout_version = cpu_to_le32(++version);
4755 lmv_buf->lb_len = sizeof(*lmv);
4756 rc = mdo_xattr_set(env, obj, lmv_buf, XATTR_NAME_LMV".set", 0,
4758 lmv_buf->lb_len = len;
4762 /* replace directory with its remaining stripe */
4766 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4767 mdd_write_lock(env, obj, DT_SRC_CHILD);
4769 /* insert dotdot to stripe which points to parent */
4770 rc = __mdd_index_insert_only(env, stripe, mdo2fid(pobj), S_IFDIR,
4775 /* copy xattrs including linkea */
4776 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle, mdo_xattr_set);
4781 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4785 /* don't set nlink from parent */
4786 attr->la_valid &= ~LA_NLINK;
4788 rc = mdo_attr_set(env, stripe, attr, handle);
4792 /* delete dir name from parent */
4793 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4797 /* insert stripe to parent with dir name */
4798 rc = __mdd_index_insert_only(env, pobj, mdo2fid(stripe), attr->la_mode,
4799 lname->ln_name, handle);
4803 /* destroy dir obj */
4804 rc = mdo_ref_del(env, obj, handle);
4808 rc = mdo_ref_del(env, obj, handle);
4812 rc = mdo_destroy(env, obj, handle);
4818 mdd_write_unlock(env, obj);
4819 mdd_write_unlock(env, pobj);
4825 * shrink directory stripes to lum_stripe_count specified by lum_mds_md.
4827 int mdd_dir_layout_shrink(const struct lu_env *env,
4828 struct md_object *md_obj,
4829 const struct lu_buf *lmu_buf)
4831 struct mdd_device *mdd = mdo2mdd(md_obj);
4832 struct mdd_thread_info *info = mdd_env_info(env);
4833 struct mdd_object *obj = md2mdd_obj(md_obj);
4834 struct mdd_object *pobj = NULL;
4835 struct mdd_object *stripe = NULL;
4836 struct lu_attr *attr = &info->mti_pattr;
4837 struct lu_fid *fid = &info->mti_fid2;
4838 struct lu_name lname = { NULL };
4839 struct lu_buf lmv_buf = { NULL };
4840 struct lmv_mds_md_v1 *lmv;
4841 struct lmv_user_md *lmu;
4842 struct thandle *handle;
4847 rc = mdd_la_get(env, obj, attr);
4851 if (!S_ISDIR(attr->la_mode))
4854 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4858 lmv = lmv_buf.lb_buf;
4859 lmu = lmu_buf->lb_buf;
4861 /* this was checked in MDT */
4862 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4863 le32_to_cpu(lmv->lmv_stripe_count));
4865 rc = mdd_dir_iterate_stripes(env, obj, &lmv_buf, lmu_buf, NULL,
4866 mdd_shrink_stripe_is_empty);
4870 GOTO(out, rc = -ENOTEMPTY);
4873 * if obj stripe count will be shrunk to 1, we need to convert it to a
4874 * normal dir, which will change its fid and update parent namespace,
4875 * get obj name and parent fid from linkea.
4877 if (le32_to_cpu(lmu->lum_stripe_count) < 2) {
4878 struct linkea_data *ldata = &info->mti_link_data;
4879 char *filename = info->mti_name;
4881 rc = mdd_links_read(env, obj, ldata);
4885 if (ldata->ld_leh->leh_reccount > 1)
4886 GOTO(out, rc = -EINVAL);
4888 linkea_first_entry(ldata);
4890 GOTO(out, rc = -ENODATA);
4892 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4895 /* Note: lname might miss \0 at the end */
4896 snprintf(filename, sizeof(info->mti_name), "%.*s",
4897 lname.ln_namelen, lname.ln_name);
4898 lname.ln_name = filename;
4900 pobj = mdd_object_find(env, mdd, fid);
4907 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4909 stripe = mdd_object_find(env, mdd, fid);
4910 if (IS_ERR(stripe)) {
4911 mdd_object_put(env, pobj);
4913 GOTO(out, rc = PTR_ERR(stripe));
4917 handle = mdd_trans_create(env, mdd);
4919 GOTO(out, rc = PTR_ERR(handle));
4921 rc = __mdd_dir_declare_layout_shrink(env, pobj, obj, stripe, attr,
4922 &lmv_buf, lmu_buf, &lname, handle);
4924 GOTO(stop_trans, rc);
4926 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4929 GOTO(stop_trans, rc);
4931 rc = mdd_trans_start(env, mdd, handle);
4933 GOTO(stop_trans, rc);
4935 rc = __mdd_dir_layout_shrink(env, pobj, obj, stripe, attr, &lmv_buf,
4936 lmu_buf, &lname, handle);
4938 GOTO(stop_trans, rc);
4940 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4941 XATTR_NAME_LMV, handle);
4942 GOTO(stop_trans, rc);
4945 rc = mdd_trans_stop(env, mdd, rc, handle);
4948 mdd_object_put(env, stripe);
4949 mdd_object_put(env, pobj);
4951 lu_buf_free(&lmv_buf);
4955 const struct md_dir_operations mdd_dir_ops = {
4956 .mdo_is_subdir = mdd_is_subdir,
4957 .mdo_lookup = mdd_lookup,
4958 .mdo_create = mdd_create,
4959 .mdo_rename = mdd_rename,
4960 .mdo_link = mdd_link,
4961 .mdo_unlink = mdd_unlink,
4962 .mdo_create_data = mdd_create_data,
4963 .mdo_migrate = mdd_migrate,