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/
31 * lustre/mdd/mdd_dir.c
33 * Lustre Metadata Server (mdd) routines
35 * Author: Wang Di <wangdi@intel.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <obd_class.h>
41 #include <obd_support.h>
42 #include <lustre_mds.h>
43 #include <lustre_fid.h>
44 #include <lustre_lmv.h>
45 #include <lustre_idmap.h>
47 #include "mdd_internal.h"
49 static const char dot[] = ".";
50 static const char dotdot[] = "..";
52 static struct lu_name lname_dotdot = {
53 .ln_name = (char *) dotdot,
54 .ln_namelen = sizeof(dotdot) - 1,
58 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
60 if (!lu_name_is_valid(ln))
62 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
68 /* Get FID from name and parent */
70 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
71 const struct lu_attr *pattr, const struct lu_name *lname,
72 struct lu_fid *fid, unsigned int may_mask)
74 const char *name = lname->ln_name;
75 const struct dt_key *key = (const struct dt_key *)name;
76 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
77 struct dt_object *dir = mdd_object_child(mdd_obj);
82 if (unlikely(mdd_is_dead_obj(mdd_obj)))
85 if (!mdd_object_exists(mdd_obj))
88 if (mdd_object_remote(mdd_obj)) {
89 CDEBUG(D_INFO, "%s: Object "DFID" located on remote server\n",
90 mdd_obj_dev_name(mdd_obj),
91 PFID(mdd_object_fid(mdd_obj)));
94 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, may_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_link_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 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 /* the obj is not locked, don't cache attributes */
238 mdd_invalidate(env, &obj->mod_obj);
240 LASSERT(ldata.ld_leh != NULL);
241 /* Directory should only have 1 parent */
242 if (ldata.ld_leh->leh_reccount > 1)
245 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
247 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
248 if (likely(fid_is_sane(fid)))
251 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
256 * For root fid use special function, which does not compare version component
257 * of fid. Version component is different for root fids on all MDTs.
259 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
261 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
262 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
266 * return 1: if \a tfid is the fid of the ancestor of \a mo;
268 * otherwise: values < 0, errors.
270 static int mdd_is_parent(const struct lu_env *env,
271 struct mdd_device *mdd,
272 struct mdd_object *mo,
273 const struct lu_attr *attr,
274 const struct lu_fid *tfid)
276 struct mdd_object *mp;
280 LASSERT(!lu_fid_eq(mdd_object_fid(mo), tfid));
281 pfid = &mdd_env_info(env)->mti_fid;
283 if (mdd_is_root(mdd, mdd_object_fid(mo)))
286 if (mdd_is_root(mdd, tfid))
289 rc = mdd_parent_fid(env, mo, attr, pfid);
294 if (lu_fid_eq(pfid, tfid))
297 if (mdd_is_root(mdd, pfid))
300 mp = mdd_object_find(env, mdd, pfid);
304 if (!mdd_object_exists(mp)) {
305 mdd_object_put(env, mp);
309 rc = mdd_parent_fid(env, mp, attr, pfid);
310 mdd_object_put(env, mp);
319 * No permission check is needed.
321 * returns 1: if fid is ancestor of @mo;
322 * returns 0: if fid is not an ancestor of @mo;
323 * returns < 0: if error
325 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
326 const struct lu_fid *fid)
328 struct mdd_device *mdd = mdo2mdd(mo);
329 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
333 if (!mdd_object_exists(md2mdd_obj(mo)))
336 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
339 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
343 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid);
348 * Check that @dir contains no entries except (possibly) dot and dotdot.
353 * -ENOTDIR not a directory object
354 * -ENOTEMPTY not empty
358 int mdd_dir_is_empty(const struct lu_env *env, struct mdd_object *dir)
361 struct dt_object *obj;
362 const struct dt_it_ops *iops;
366 obj = mdd_object_child(dir);
367 if (!dt_try_as_dir(env, obj))
370 iops = &obj->do_index_ops->dio_it;
371 it = iops->init(env, obj, LUDA_64BITHASH);
373 result = iops->get(env, it, (const struct dt_key *)"");
376 for (result = 0, i = 0; result == 0 && i < 3; ++i)
377 result = iops->next(env, it);
380 else if (result == 1)
382 } else if (result == 0)
384 * Huh? Index contains no zero key?
391 result = PTR_ERR(it);
392 /* -ENODEV means no valid stripe */
393 if (result == -ENODEV)
400 * Determine if the target object can be hard linked, and right now it only
401 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
402 * directory nlink count might exceed the maximum link count(see
403 * osd_object_ref_add), so it only check nlink for non-directories.
405 * \param[in] env thread environment
406 * \param[in] obj object being linked to
407 * \param[in] la attributes of \a obj
409 * \retval 0 if \a obj can be hard linked
410 * \retval negative error if \a obj is a directory or has too
413 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
414 const struct lu_attr *la)
416 struct mdd_device *m = mdd_obj2mdd_dev(obj);
421 /* Subdir count limitation can be broken through
422 * (see osd_object_ref_add), so only check non-directory here. */
423 if (!S_ISDIR(la->la_mode) &&
424 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
431 * Check whether it may create the cobj under the pobj.
433 * \param[in] env execution environment
434 * \param[in] pobj the parent directory
435 * \param[in] pattr the attribute of the parent directory
436 * \param[in] cobj the child to be created
437 * \param[in] check_perm if check WRITE|EXEC permission for parent
439 * \retval = 0 create the child under this dir is allowed
440 * \retval negative errno create the child under this dir is
443 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
444 const struct lu_attr *pattr, struct mdd_object *cobj,
450 if (cobj && mdd_object_exists(cobj))
453 if (mdd_is_dead_obj(pobj))
457 rc = mdd_permission_internal_locked(env, pobj, pattr,
458 MAY_WRITE | MAY_EXEC,
464 * Check whether can unlink from the pobj in the case of "cobj == NULL".
466 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
467 const struct lu_attr *pattr, const struct lu_attr *attr)
472 if (mdd_is_dead_obj(pobj))
475 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
478 rc = mdd_permission_internal_locked(env, pobj, pattr,
479 MAY_WRITE | MAY_EXEC,
484 if (pattr->la_flags & LUSTRE_APPEND_FL)
491 * pobj == NULL is remote ops case, under such case, pobj's
492 * VTX feature has been checked already, no need check again.
494 static inline int mdd_is_sticky(const struct lu_env *env,
495 struct mdd_object *pobj,
496 const struct lu_attr *pattr,
497 struct mdd_object *cobj,
498 const struct lu_attr *cattr)
500 struct lu_ucred *uc = lu_ucred_assert(env);
503 LASSERT(pattr != NULL);
504 if (!(pattr->la_mode & S_ISVTX) ||
505 (pattr->la_uid == uc->uc_fsuid))
509 LASSERT(cattr != NULL);
510 if (cattr->la_uid == uc->uc_fsuid)
513 return !md_capable(uc, CAP_FOWNER);
516 static int mdd_may_delete_entry(const struct lu_env *env,
517 struct mdd_object *pobj,
518 const struct lu_attr *pattr,
523 LASSERT(pobj != NULL);
524 if (!mdd_object_exists(pobj))
527 if (mdd_is_dead_obj(pobj))
532 rc = mdd_permission_internal_locked(env, pobj, pattr,
533 MAY_WRITE | MAY_EXEC,
539 if (pattr->la_flags & LUSTRE_APPEND_FL)
546 * Check whether it may delete the cobj from the pobj.
549 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
550 const struct lu_attr *tpattr, struct mdd_object *tobj,
551 const struct lu_attr *tattr, const struct lu_attr *cattr,
552 int check_perm, int check_empty)
558 LASSERT(tpattr != NULL);
559 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
567 if (!mdd_object_exists(tobj))
570 if (mdd_is_dead_obj(tobj))
573 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
576 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
579 /* additional check the rename case */
581 if (S_ISDIR(cattr->la_mode)) {
582 if (!S_ISDIR(tattr->la_mode))
585 if (mdd_is_root(mdo2mdd(&tobj->mod_obj),
586 mdd_object_fid(tobj)))
588 } else if (S_ISDIR(tattr->la_mode))
592 if (S_ISDIR(tattr->la_mode) && check_empty)
593 rc = mdd_dir_is_empty(env, tobj);
599 * Check whether it can create the link file(linked to @src_obj) under
600 * the target directory(@tgt_obj), and src_obj has been locked by
603 * \param[in] env execution environment
604 * \param[in] tgt_obj the target directory
605 * \param[in] tattr attributes of target directory
606 * \param[in] lname the link name
607 * \param[in] src_obj source object for link
608 * \param[in] cattr attributes for source object
610 * \retval = 0 it is allowed to create the link file under tgt_obj
611 * \retval negative error not allowed to create the link file
613 static int mdd_link_sanity_check(const struct lu_env *env,
614 struct mdd_object *tgt_obj,
615 const struct lu_attr *tattr,
616 const struct lu_name *lname,
617 struct mdd_object *src_obj,
618 const struct lu_attr *cattr)
620 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
624 if (!mdd_object_exists(src_obj))
627 if (mdd_is_dead_obj(src_obj))
630 /* Local ops, no lookup before link, check filename length here. */
631 rc = mdd_name_check(m, lname);
635 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
638 if (S_ISDIR(mdd_object_type(src_obj)))
641 LASSERT(src_obj != tgt_obj);
642 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
646 rc = __mdd_may_link(env, src_obj, cattr);
651 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
652 const char *name, struct thandle *handle)
654 struct dt_object *next = mdd_object_child(pobj);
658 if (dt_try_as_dir(env, next))
659 rc = dt_delete(env, next, (struct dt_key *)name, handle);
666 static int __mdd_index_insert_only(const struct lu_env *env,
667 struct mdd_object *pobj,
668 const struct lu_fid *lf, __u32 type,
670 struct thandle *handle)
672 struct dt_object *next = mdd_object_child(pobj);
676 if (dt_try_as_dir(env, next)) {
677 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
680 rec->rec_type = type;
681 rc = dt_insert(env, next, (const struct dt_rec *)rec,
682 (const struct dt_key *)name, handle);
689 /* insert named index, add reference if isdir */
690 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
691 const struct lu_fid *lf, __u32 type,
692 const char *name, struct thandle *handle)
697 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
698 if (rc == 0 && S_ISDIR(type)) {
699 mdd_write_lock(env, pobj, DT_TGT_PARENT);
700 mdo_ref_add(env, pobj, handle);
701 mdd_write_unlock(env, pobj);
707 /* delete named index, drop reference if isdir */
708 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
709 const char *name, int is_dir,
710 struct thandle *handle)
715 rc = __mdd_index_delete_only(env, pobj, name, handle);
716 if (rc == 0 && is_dir) {
717 mdd_write_lock(env, pobj, DT_TGT_PARENT);
718 mdo_ref_del(env, pobj, handle);
719 mdd_write_unlock(env, pobj);
725 static int mdd_llog_record_calc_size(const struct lu_env *env,
726 const struct lu_name *tname,
727 const struct lu_name *sname)
729 const struct lu_ucred *uc = lu_ucred(env);
730 enum changelog_rec_flags clf_flags = CLF_EXTRA_FLAGS;
731 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
734 clf_flags |= CLF_RENAME;
736 if (uc != NULL && uc->uc_jobid[0] != '\0')
737 clf_flags |= CLF_JOBID;
739 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
740 changelog_rec_offset(clf_flags, crfe) +
741 (tname != NULL ? tname->ln_namelen : 0) +
742 (sname != NULL ? 1 + sname->ln_namelen : 0));
745 int mdd_declare_changelog_store(const struct lu_env *env,
746 struct mdd_device *mdd,
747 enum changelog_rec_type type,
748 const struct lu_name *tname,
749 const struct lu_name *sname,
750 struct thandle *handle)
752 struct obd_device *obd = mdd2obd_dev(mdd);
753 struct llog_ctxt *ctxt;
754 struct llog_changelog_rec *rec;
756 struct thandle *llog_th;
760 if (!mdd_changelog_enabled(env, mdd, type))
763 reclen = mdd_llog_record_calc_size(env, tname, sname);
764 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_chlg_buf, reclen);
765 if (buf->lb_buf == NULL)
769 rec->cr_hdr.lrh_len = reclen;
770 rec->cr_hdr.lrh_type = CHANGELOG_REC;
772 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
776 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
778 GOTO(out_put, rc = PTR_ERR(llog_th));
780 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
788 int mdd_changelog_write_rec(const struct lu_env *env,
789 struct llog_handle *loghandle,
790 struct llog_rec_hdr *r,
791 struct llog_cookie *cookie,
792 int idx, struct thandle *th)
796 if (r->lrh_type == CHANGELOG_REC) {
797 struct mdd_device *mdd;
798 struct llog_changelog_rec *rec;
800 mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
801 rec = container_of(r, struct llog_changelog_rec, cr_hdr);
803 spin_lock(&mdd->mdd_cl.mc_lock);
804 rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
805 spin_unlock(&mdd->mdd_cl.mc_lock);
807 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
811 * if current llog is full, we will generate a new
812 * llog, and since it's actually not an error, let's
813 * avoid increasing index so that userspace apps
814 * should not see a gap in the changelog sequence
816 if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
817 spin_lock(&mdd->mdd_cl.mc_lock);
818 ++mdd->mdd_cl.mc_index;
819 spin_unlock(&mdd->mdd_cl.mc_lock);
822 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
829 /** Add a changelog entry \a rec to the changelog llog
832 * \param handle - currently ignored since llogs start their own transaction;
833 * this will hopefully be fixed in llog rewrite
836 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
837 struct llog_changelog_rec *rec, struct thandle *th)
839 struct obd_device *obd = mdd2obd_dev(mdd);
840 struct llog_ctxt *ctxt;
841 struct thandle *llog_th;
844 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
845 changelog_rec_varsize(&rec->cr));
847 /* llog_lvfs_write_rec sets the llog tail len */
848 rec->cr_hdr.lrh_type = CHANGELOG_REC;
849 rec->cr.cr_time = cl_time();
851 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
855 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
857 GOTO(out_put, rc = PTR_ERR(llog_th));
859 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
860 /* nested journal transaction */
861 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
863 /* time to recover some space ?? */
864 if (likely(!mdd->mdd_changelog_gc ||
865 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
866 mdd->mdd_changelog_min_gc_interval >=
867 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
868 /* save a spin_lock trip */
870 spin_lock(&mdd->mdd_cl.mc_lock);
871 if (likely(mdd->mdd_changelog_gc &&
872 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
873 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
874 mdd->mdd_changelog_min_gc_interval)) {
875 if (unlikely(llog_cat_free_space(ctxt->loc_handle) <=
876 mdd->mdd_changelog_min_free_cat_entries ||
877 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD))) {
878 CWARN("%s:%s low on changelog_catalog free entries, "
879 "starting ChangeLog garbage collection thread\n",
881 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
884 /* indicate further kthread run will occur outside
885 * right after current journal transaction filling has
888 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
890 /* next check in mdd_changelog_min_gc_interval anyway
892 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
894 spin_unlock(&mdd->mdd_cl.mc_lock);
902 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
903 const struct lu_fid *sfid,
904 const struct lu_fid *spfid,
905 const struct lu_name *sname)
907 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
910 LASSERT(sfid != NULL);
911 LASSERT(spfid != NULL);
912 LASSERT(sname != NULL);
914 extsize = sname->ln_namelen + 1;
916 rnm->cr_sfid = *sfid;
917 rnm->cr_spfid = *spfid;
919 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
920 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
921 rec->cr_namelen += extsize;
924 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
926 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
928 if (jobid == NULL || jobid[0] == '\0')
931 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
934 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
936 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
938 ef->cr_extra_flags = eflags;
941 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
942 __u64 uid, __u64 gid)
944 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
946 uidgid->cr_uid = uid;
947 uidgid->cr_gid = gid;
950 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
953 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
958 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
960 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
962 omd->cr_openflags = flags;
965 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
966 const char *xattr_name)
968 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
970 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
973 /** Store a namespace change changelog record
974 * If this fails, we must fail the whole transaction; we don't
975 * want the change to commit without the log entry.
976 * \param target - mdd_object of change
977 * \param tpfid - target parent dir/object fid
978 * \param sfid - source object fid
979 * \param spfid - source parent fid
980 * \param tname - target name string
981 * \param sname - source name string
982 * \param handle - transaction handle
984 int mdd_changelog_ns_store(const struct lu_env *env,
985 struct mdd_device *mdd,
986 enum changelog_rec_type type,
987 enum changelog_rec_flags clf_flags,
988 struct mdd_object *target,
989 const struct lu_fid *tpfid,
990 const struct lu_fid *sfid,
991 const struct lu_fid *spfid,
992 const struct lu_name *tname,
993 const struct lu_name *sname,
994 struct thandle *handle)
996 const struct lu_ucred *uc = lu_ucred(env);
997 struct llog_changelog_rec *rec;
1000 __u64 xflags = CLFE_INVALID;
1004 if (!mdd_changelog_enabled(env, mdd, type))
1007 LASSERT(tpfid != NULL);
1008 LASSERT(tname != NULL);
1009 LASSERT(handle != NULL);
1011 reclen = mdd_llog_record_calc_size(env, tname, sname);
1012 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_chlg_buf, reclen);
1013 if (buf->lb_buf == NULL)
1017 clf_flags &= CLF_FLAGMASK;
1018 clf_flags |= CLF_EXTRA_FLAGS;
1021 if (uc->uc_jobid[0] != '\0')
1022 clf_flags |= CLF_JOBID;
1023 xflags |= CLFE_UIDGID;
1028 clf_flags |= CLF_RENAME;
1030 clf_flags |= CLF_VERSION;
1032 rec->cr.cr_flags = clf_flags;
1034 if (clf_flags & CLF_EXTRA_FLAGS) {
1035 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1036 if (xflags & CLFE_UIDGID)
1037 mdd_changelog_rec_extra_uidgid(&rec->cr,
1038 uc->uc_uid, uc->uc_gid);
1039 if (xflags & CLFE_NID)
1040 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1043 rec->cr.cr_type = (__u32)type;
1044 rec->cr.cr_pfid = *tpfid;
1045 rec->cr.cr_namelen = tname->ln_namelen;
1046 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1048 if (clf_flags & CLF_RENAME)
1049 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1051 if (clf_flags & CLF_JOBID)
1052 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1054 if (likely(target != NULL)) {
1055 rec->cr.cr_tfid = *mdd_object_fid(target);
1056 target->mod_cltime = ktime_get();
1058 fid_zero(&rec->cr.cr_tfid);
1061 rc = mdd_changelog_store(env, mdd, rec, handle);
1063 CERROR("%s: cannot store changelog record: type = %d, "
1064 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1065 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1066 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1073 static int __mdd_links_add(const struct lu_env *env,
1074 struct mdd_object *mdd_obj,
1075 struct linkea_data *ldata,
1076 const struct lu_name *lname,
1077 const struct lu_fid *pfid,
1078 int first, int check)
1082 if (ldata->ld_leh == NULL) {
1083 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1087 rc = linkea_data_new(ldata,
1088 &mdd_env_info(env)->mti_link_buf);
1095 rc = linkea_links_find(ldata, lname, pfid);
1096 if (rc && rc != -ENOENT)
1102 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1103 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1107 linkea_add_buf(ldata, lname, tfid);
1110 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1111 linkea_add_buf(ldata, lname, pfid);
1113 return linkea_add_buf(ldata, lname, pfid);
1116 static int __mdd_links_del(const struct lu_env *env,
1117 struct mdd_object *mdd_obj,
1118 struct linkea_data *ldata,
1119 const struct lu_name *lname,
1120 const struct lu_fid *pfid)
1124 if (ldata->ld_leh == NULL) {
1125 rc = mdd_links_read(env, mdd_obj, ldata);
1130 rc = linkea_links_find(ldata, lname, pfid);
1134 linkea_del_buf(ldata, lname);
1138 static int mdd_linkea_prepare(const struct lu_env *env,
1139 struct mdd_object *mdd_obj,
1140 const struct lu_fid *oldpfid,
1141 const struct lu_name *oldlname,
1142 const struct lu_fid *newpfid,
1143 const struct lu_name *newlname,
1144 int first, int check,
1145 struct linkea_data *ldata)
1150 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1153 LASSERT(oldpfid != NULL || newpfid != NULL);
1155 if (mdd_obj->mod_flags & DEAD_OBJ)
1156 /* Unnecessary to update linkEA for dead object. */
1159 if (oldpfid != NULL) {
1160 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1162 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1165 /* No changes done. */
1170 /* If renaming, add the new record */
1171 if (newpfid != NULL)
1172 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1178 int mdd_links_rename(const struct lu_env *env,
1179 struct mdd_object *mdd_obj,
1180 const struct lu_fid *oldpfid,
1181 const struct lu_name *oldlname,
1182 const struct lu_fid *newpfid,
1183 const struct lu_name *newlname,
1184 struct thandle *handle,
1185 struct linkea_data *ldata,
1186 int first, int check)
1191 if (ldata == NULL) {
1192 ldata = &mdd_env_info(env)->mti_link_data;
1193 memset(ldata, 0, sizeof(*ldata));
1194 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1195 newpfid, newlname, first, check, ldata);
1200 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1201 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1207 if (newlname == NULL)
1208 CERROR("link_ea add failed %d "DFID"\n",
1209 rc, PFID(mdd_object_fid(mdd_obj)));
1210 else if (oldpfid == NULL)
1211 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1212 newlname->ln_namelen, newlname->ln_name, rc,
1213 PFID(mdd_object_fid(mdd_obj)));
1214 else if (newpfid == NULL)
1215 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1216 oldlname->ln_namelen, oldlname->ln_name, rc,
1217 PFID(mdd_object_fid(mdd_obj)));
1219 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1220 "\n", oldlname->ln_namelen, oldlname->ln_name,
1221 newlname->ln_namelen, newlname->ln_name, rc,
1222 PFID(mdd_object_fid(mdd_obj)));
1225 if (is_vmalloc_addr(ldata->ld_buf))
1226 /* if we vmalloced a large buffer drop it */
1227 lu_buf_free(ldata->ld_buf);
1232 static inline int mdd_links_add(const struct lu_env *env,
1233 struct mdd_object *mdd_obj,
1234 const struct lu_fid *pfid,
1235 const struct lu_name *lname,
1236 struct thandle *handle,
1237 struct linkea_data *ldata, int first)
1239 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1240 pfid, lname, handle, ldata, first, 0);
1243 static inline int mdd_links_del(const struct lu_env *env,
1244 struct mdd_object *mdd_obj,
1245 const struct lu_fid *pfid,
1246 const struct lu_name *lname,
1247 struct thandle *handle)
1249 return mdd_links_rename(env, mdd_obj, pfid, lname,
1250 NULL, NULL, handle, NULL, 0, 0);
1253 /** Read the link EA into a temp buffer.
1254 * Uses the name_buf since it is generally large.
1255 * \retval IS_ERR err
1256 * \retval ptr to \a lu_buf (always \a mti_link_buf)
1258 struct lu_buf *mdd_links_get(const struct lu_env *env,
1259 struct mdd_object *mdd_obj)
1261 struct linkea_data ldata = { NULL };
1264 rc = mdd_links_read(env, mdd_obj, &ldata);
1265 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1268 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1269 struct linkea_data *ldata, struct thandle *handle)
1271 const struct lu_buf *buf;
1274 if (ldata == NULL || ldata->ld_buf == NULL ||
1275 ldata->ld_leh == NULL)
1278 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1282 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1283 ldata->ld_leh->leh_len);
1284 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1285 if (unlikely(rc == -ENOSPC)) {
1286 rc = linkea_overflow_shrink(ldata);
1294 static int mdd_declare_links_add(const struct lu_env *env,
1295 struct mdd_object *mdd_obj,
1296 struct thandle *handle,
1297 struct linkea_data *ldata)
1303 if (ldata != NULL && ldata->ld_leh != NULL) {
1304 ea_len = ldata->ld_leh->leh_len;
1305 linkea = ldata->ld_buf->lb_buf;
1307 ea_len = MAX_LINKEA_SIZE;
1311 rc = mdo_declare_xattr_set(env, mdd_obj,
1312 mdd_buf_get_const(env, linkea, ea_len),
1313 XATTR_NAME_LINK, 0, handle);
1318 static inline int mdd_declare_links_del(const struct lu_env *env,
1319 struct mdd_object *c,
1320 struct thandle *handle)
1324 /* For directory, the linkEA will be removed together
1325 * with the object. */
1326 if (!S_ISDIR(mdd_object_type(c)))
1327 rc = mdd_declare_links_add(env, c, handle, NULL);
1332 static int mdd_declare_link(const struct lu_env *env,
1333 struct mdd_device *mdd,
1334 struct mdd_object *p,
1335 struct mdd_object *c,
1336 const struct lu_name *name,
1337 struct thandle *handle,
1339 struct linkea_data *data)
1341 struct lu_fid tfid = *mdd_object_fid(c);
1344 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1345 tfid.f_oid = cfs_fail_val;
1347 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1348 name->ln_name, handle);
1352 rc = mdo_declare_ref_add(env, c, handle);
1356 la->la_valid = LA_CTIME | LA_MTIME;
1357 rc = mdo_declare_attr_set(env, p, la, handle);
1361 la->la_valid = LA_CTIME;
1362 rc = mdo_declare_attr_set(env, c, la, handle);
1366 rc = mdd_declare_links_add(env, c, handle, data);
1370 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1376 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1377 struct md_object *src_obj, const struct lu_name *lname,
1380 const char *name = lname->ln_name;
1381 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1382 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1383 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1384 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1385 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1386 struct mdd_device *mdd = mdo2mdd(src_obj);
1387 struct thandle *handle;
1388 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1389 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1393 rc = mdd_la_get(env, mdd_sobj, cattr);
1397 rc = mdd_la_get(env, mdd_tobj, tattr);
1402 * If we are using project inheritance, we only allow hard link
1403 * creation in our tree when the project IDs are the same;
1404 * otherwise the tree quota mechanism could be circumvented.
1406 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1407 (tattr->la_projid != cattr->la_projid))
1410 handle = mdd_trans_create(env, mdd);
1412 GOTO(out_pending, rc = PTR_ERR(handle));
1414 memset(ldata, 0, sizeof(*ldata));
1416 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1417 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1419 /* Note: even this function will change ldata, but it comes from
1420 * thread_info, which is completely temporary and only seen in
1421 * this function, so we do not need reset ldata once it fails.*/
1422 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1423 mdd_object_fid(mdd_tobj), lname, 0, 0, ldata);
1427 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1432 rc = mdd_trans_start(env, mdd, handle);
1436 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1437 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1440 GOTO(out_unlock, rc);
1442 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1443 rc = mdo_ref_add(env, mdd_sobj, handle);
1445 GOTO(out_unlock, rc);
1448 *tfid = *mdd_object_fid(mdd_sobj);
1449 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1450 tfid->f_oid = cfs_fail_val;
1452 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1453 mdd_object_type(mdd_sobj), name, handle);
1455 mdo_ref_del(env, mdd_sobj, handle);
1456 GOTO(out_unlock, rc);
1459 la->la_valid = LA_CTIME | LA_MTIME;
1460 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1462 GOTO(out_unlock, rc);
1464 la->la_valid = LA_CTIME;
1465 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1467 /* Note: The failure of links_add should not cause the
1468 * link failure, so do not check return value. */
1469 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_tobj),
1470 lname, handle, ldata, 0);
1474 mdd_write_unlock(env, mdd_sobj);
1476 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1477 mdd_object_fid(mdd_tobj), NULL,
1478 NULL, lname, NULL, handle);
1480 rc = mdd_trans_stop(env, mdd, rc, handle);
1481 if (is_vmalloc_addr(ldata->ld_buf))
1482 /* if we vmalloced a large buffer drop it */
1483 lu_buf_free(ldata->ld_buf);
1488 static int mdd_mark_orphan_object(const struct lu_env *env,
1489 struct mdd_object *obj, struct thandle *handle,
1492 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1495 attr->la_valid = LA_FLAGS;
1496 attr->la_flags = LUSTRE_ORPHAN_FL;
1499 rc = mdo_declare_attr_set(env, obj, attr, handle);
1501 rc = mdo_attr_set(env, obj, attr, handle);
1506 static int mdd_declare_finish_unlink(const struct lu_env *env,
1507 struct mdd_object *obj,
1508 struct thandle *handle)
1512 /* Sigh, we do not know if the unlink object will become orphan in
1513 * declare phase, but fortunately the flags here does not matter
1514 * in current declare implementation */
1515 rc = mdd_mark_orphan_object(env, obj, handle, true);
1519 rc = mdo_declare_destroy(env, obj, handle);
1523 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1527 return mdd_declare_links_del(env, obj, handle);
1530 /* caller should take a lock before calling */
1531 int mdd_finish_unlink(const struct lu_env *env,
1532 struct mdd_object *obj, struct md_attr *ma,
1533 struct mdd_object *pobj,
1534 const struct lu_name *lname,
1538 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1541 LASSERT(mdd_write_locked(env, obj) != 0);
1543 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1544 /* add new orphan and the object
1545 * will be deleted during mdd_close() */
1546 obj->mod_flags |= DEAD_OBJ;
1547 if (obj->mod_count) {
1548 rc = mdd_orphan_insert(env, obj, th);
1550 CDEBUG(D_HA, "Object "DFID" is inserted into "
1551 "orphan list, open count = %d\n",
1552 PFID(mdd_object_fid(obj)),
1555 CERROR("Object "DFID" fail to be an orphan, "
1556 "open count = %d, maybe cause failed "
1558 PFID(mdd_object_fid(obj)),
1561 /* mark object as an orphan here, not
1562 * before mdd_orphan_insert() as racing
1563 * mdd_la_get() may propagate ORPHAN_OBJ
1564 * causing the asserition */
1565 rc = mdd_mark_orphan_object(env, obj, th, false);
1567 rc = mdo_destroy(env, obj, th);
1569 } else if (!is_dir) {
1570 /* old files may not have link ea; ignore errors */
1571 mdd_links_del(env, obj, mdd_object_fid(pobj), lname, th);
1579 * has mdd_write_lock on cobj already, but not on pobj yet
1581 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1582 const struct lu_attr *pattr,
1583 struct mdd_object *cobj,
1584 const struct lu_attr *cattr)
1589 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1594 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1595 struct mdd_object *p, struct mdd_object *c,
1596 const struct lu_name *name, struct md_attr *ma,
1597 struct thandle *handle, int no_name, int is_dir)
1599 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1602 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1603 if (likely(no_name == 0)) {
1604 rc = mdo_declare_index_delete(env, p, name->ln_name,
1611 rc = mdo_declare_ref_del(env, p, handle);
1617 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1618 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1619 la->la_valid = LA_CTIME | LA_MTIME;
1620 rc = mdo_declare_attr_set(env, p, la, handle);
1625 rc = mdo_declare_ref_del(env, c, handle);
1629 rc = mdo_declare_ref_del(env, c, handle);
1633 la->la_valid = LA_CTIME;
1634 rc = mdo_declare_attr_set(env, c, la, handle);
1638 rc = mdd_declare_finish_unlink(env, c, handle);
1642 /* FIXME: need changelog for remove entry */
1643 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1651 * test if a file has an HSM archive
1652 * if HSM attributes are not found in ma update them from
1655 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1656 struct mdd_object *obj,
1661 if (!(ma->ma_valid & MA_HSM)) {
1662 /* no HSM MD provided, read xattr */
1663 struct lu_buf *hsm_buf;
1664 const size_t buflen = sizeof(struct hsm_attrs);
1667 hsm_buf = mdd_buf_get(env, NULL, 0);
1668 lu_buf_alloc(hsm_buf, buflen);
1669 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1670 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1671 lu_buf_free(hsm_buf);
1675 ma->ma_valid |= MA_HSM;
1677 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1683 * Delete name entry and the object.
1684 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1685 * does not exist for this object, and it could only happen during resending
1686 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1687 * is also needed in this case(needed by changelog), so we have to add another
1688 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1689 * the ENOENT failure should be able to be fixed by redo mechanism.
1691 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1692 struct md_object *cobj, const struct lu_name *lname,
1693 struct md_attr *ma, int no_name)
1695 const char *name = lname->ln_name;
1696 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1697 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1698 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1699 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1700 struct mdd_object *mdd_cobj = NULL;
1701 struct mdd_device *mdd = mdo2mdd(pobj);
1702 struct thandle *handle;
1703 int rc, is_dir = 0, cl_flags = 0;
1706 /* let shutdown to start */
1707 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1709 /* cobj == NULL means only delete name entry */
1710 if (likely(cobj != NULL)) {
1711 mdd_cobj = md2mdd_obj(cobj);
1712 if (mdd_object_exists(mdd_cobj) == 0)
1716 rc = mdd_la_get(env, mdd_pobj, pattr);
1720 if (likely(mdd_cobj != NULL)) {
1722 rc = mdd_la_get(env, mdd_cobj, cattr);
1726 is_dir = S_ISDIR(cattr->la_mode);
1727 /* search for an existing archive.
1728 * we should check ahead as the object
1729 * can be destroyed in this transaction */
1730 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1731 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1734 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1738 handle = mdd_trans_create(env, mdd);
1740 RETURN(PTR_ERR(handle));
1742 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1743 lname, ma, handle, no_name, is_dir);
1747 rc = mdd_trans_start(env, mdd, handle);
1751 if (likely(mdd_cobj != NULL))
1752 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1754 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1755 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1760 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1761 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1762 GOTO(cleanup, rc = 0);
1764 if (likely(mdd_cobj != NULL)) {
1765 rc = mdo_ref_del(env, mdd_cobj, handle);
1767 __mdd_index_insert_only(env, mdd_pobj,
1768 mdd_object_fid(mdd_cobj),
1769 mdd_object_type(mdd_cobj),
1776 mdo_ref_del(env, mdd_cobj, handle);
1778 /* fetch updated nlink */
1779 rc = mdd_la_get(env, mdd_cobj, cattr);
1784 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1785 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1787 la->la_valid = LA_CTIME | LA_MTIME;
1788 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1792 /* Enough for only unlink the entry */
1793 if (unlikely(mdd_cobj == NULL))
1796 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1797 /* update ctime of an unlinked file only if it is still
1798 * opened or a link still exists */
1799 la->la_valid = LA_CTIME;
1800 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1805 /* XXX: this transfer to ma will be removed with LOD/OSP */
1806 ma->ma_attr = *cattr;
1807 ma->ma_valid |= MA_INODE;
1808 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1812 /* fetch updated nlink */
1813 rc = mdd_la_get(env, mdd_cobj, cattr);
1814 /* if object is removed then we can't get its attrs,
1816 if (rc == -ENOENT) {
1817 cattr->la_nlink = 0;
1821 if (cattr->la_nlink == 0) {
1822 ma->ma_attr = *cattr;
1823 ma->ma_valid |= MA_INODE;
1828 if (likely(mdd_cobj != NULL))
1829 mdd_write_unlock(env, mdd_cobj);
1832 if (cattr->la_nlink == 0)
1833 cl_flags |= CLF_UNLINK_LAST;
1835 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1837 rc = mdd_changelog_ns_store(env, mdd,
1838 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1839 mdd_cobj, mdd_object_fid(mdd_pobj), NULL, NULL,
1840 lname, NULL, handle);
1844 rc = mdd_trans_stop(env, mdd, rc, handle);
1850 * The permission has been checked when obj created, no need check again.
1852 static int mdd_cd_sanity_check(const struct lu_env *env,
1853 struct mdd_object *obj)
1858 if (!obj || mdd_is_dead_obj(obj))
1864 static int mdd_create_data(const struct lu_env *env,
1865 struct md_object *pobj,
1866 struct md_object *cobj,
1867 const struct md_op_spec *spec,
1870 struct mdd_device *mdd = mdo2mdd(cobj);
1871 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1872 struct mdd_object *son = md2mdd_obj(cobj);
1873 struct thandle *handle;
1874 const struct lu_buf *buf;
1875 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1876 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1880 rc = mdd_cd_sanity_check(env, son);
1884 if (!md_should_create(spec->sp_cr_flags))
1888 * there are following use cases for this function:
1889 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1890 * striping can be specified or not
1893 rc = mdd_la_get(env, son, attr);
1897 /* calling ->ah_make_hint() is used to transfer information from parent */
1898 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1900 handle = mdd_trans_create(env, mdd);
1902 GOTO(out_free, rc = PTR_ERR(handle));
1905 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1906 * Should this be fixed?
1908 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1909 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1910 spec->sp_cr_flags, spec->no_create);
1912 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1913 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1914 spec->u.sp_ea.eadatalen);
1919 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1920 XATTR_NAME_LOV, 0, handle);
1924 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1929 rc = mdd_trans_start(env, mdd, handle);
1933 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1939 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle,
1943 rc = mdd_trans_stop(env, mdd, rc, handle);
1949 static int mdd_declare_object_initialize(const struct lu_env *env,
1950 struct mdd_object *parent,
1951 struct mdd_object *child,
1952 const struct lu_attr *attr,
1953 struct thandle *handle)
1958 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1959 if (!S_ISDIR(attr->la_mode))
1962 rc = mdo_declare_index_insert(env, child, mdd_object_fid(child),
1963 S_IFDIR, dot, handle);
1967 rc = mdo_declare_ref_add(env, child, handle);
1971 rc = mdo_declare_index_insert(env, child, mdd_object_fid(parent),
1972 S_IFDIR, dotdot, handle);
1977 static int mdd_object_initialize(const struct lu_env *env,
1978 const struct lu_fid *pfid,
1979 struct mdd_object *child,
1980 struct lu_attr *attr,
1981 struct thandle *handle)
1986 if (S_ISDIR(attr->la_mode)) {
1987 /* Add "." and ".." for newly created dir */
1988 mdo_ref_add(env, child, handle);
1989 rc = __mdd_index_insert_only(env, child, mdd_object_fid(child),
1990 S_IFDIR, dot, handle);
1992 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1995 mdo_ref_del(env, child, handle);
2002 * This function checks whether it can create a file/dir under the
2003 * directory(@pobj). The directory(@pobj) is not being locked by
2006 * \param[in] env execution environment
2007 * \param[in] pobj the directory to create files
2008 * \param[in] pattr the attributes of the directory
2009 * \param[in] lname the name of the created file/dir
2010 * \param[in] cattr the attributes of the file/dir
2011 * \param[in] spec create specification
2013 * \retval = 0 it is allowed to create file/dir under
2015 * \retval negative error not allowed to create file/dir
2016 * under the directory
2018 static int mdd_create_sanity_check(const struct lu_env *env,
2019 struct md_object *pobj,
2020 const struct lu_attr *pattr,
2021 const struct lu_name *lname,
2022 struct lu_attr *cattr,
2023 struct md_op_spec *spec)
2025 struct mdd_thread_info *info = mdd_env_info(env);
2026 struct lu_fid *fid = &info->mti_fid;
2027 struct mdd_object *obj = md2mdd_obj(pobj);
2028 struct mdd_device *m = mdo2mdd(pobj);
2029 bool check_perm = true;
2034 if (mdd_is_dead_obj(obj))
2038 * In some cases this lookup is not needed - we know before if name
2039 * exists or not because MDT performs lookup for it.
2040 * name length check is done in lookup.
2042 if (spec->sp_cr_lookup) {
2044 * Check if the name already exist, though it will be checked in
2045 * _index_insert also, for avoiding rolling back if exists
2048 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2049 MAY_WRITE | MAY_EXEC);
2051 RETURN(rc ? : -EEXIST);
2053 /* Permission is already being checked in mdd_lookup */
2057 if (S_ISDIR(cattr->la_mode) &&
2058 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2059 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2060 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2062 if (!lmv_user_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2063 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2065 CERROR("%s: invalid lmv_user_md: magic = %x, "
2066 "stripe_offset = %d, stripe_count = %u: "
2067 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2068 le32_to_cpu(lum->lum_magic),
2069 (int)le32_to_cpu(lum->lum_stripe_offset),
2070 le32_to_cpu(lum->lum_stripe_count), rc);
2075 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2080 if (pattr->la_mode & S_ISGID) {
2081 struct lu_ucred *uc = lu_ucred(env);
2083 cattr->la_gid = pattr->la_gid;
2085 /* Directories are special, and always inherit S_ISGID */
2086 if (S_ISDIR(cattr->la_mode)) {
2087 cattr->la_mode |= S_ISGID;
2088 cattr->la_valid |= LA_MODE;
2089 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2090 == (S_ISGID | S_IXGRP) &&
2091 !lustre_in_group_p(uc,
2092 (cattr->la_valid & LA_GID) ?
2093 cattr->la_gid : pattr->la_gid) &&
2094 !md_capable(uc, CAP_FSETID)) {
2095 cattr->la_mode &= ~S_ISGID;
2096 cattr->la_valid |= LA_MODE;
2100 /* Inherit project ID from parent directory */
2101 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2102 cattr->la_projid = pattr->la_projid;
2103 if (S_ISDIR(cattr->la_mode)) {
2104 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2105 cattr->la_valid |= LA_FLAGS;
2107 cattr->la_valid |= LA_PROJID;
2110 rc = mdd_name_check(m, lname);
2114 switch (cattr->la_mode & S_IFMT) {
2116 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2118 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2119 RETURN(-ENAMETOOLONG);
2138 static int mdd_declare_create_object(const struct lu_env *env,
2139 struct mdd_device *mdd,
2140 struct mdd_object *p, struct mdd_object *c,
2141 struct lu_attr *attr,
2142 struct thandle *handle,
2143 const struct md_op_spec *spec,
2144 struct lu_buf *def_acl_buf,
2145 struct lu_buf *acl_buf,
2146 struct lu_buf *hsm_buf,
2147 struct dt_allocation_hint *hint)
2149 const struct lu_buf *buf;
2152 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2153 /* ldiskfs OSD needs this information for credit allocation */
2155 hint->dah_acl_len = def_acl_buf->lb_len;
2157 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2162 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2163 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2164 /* if dir, then can inherit default ACl */
2165 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2166 XATTR_NAME_ACL_DEFAULT,
2172 if (acl_buf && acl_buf->lb_len > 0) {
2173 rc = mdo_declare_attr_set(env, c, attr, handle);
2177 rc = mdo_declare_xattr_set(env, c, acl_buf,
2178 XATTR_NAME_ACL_ACCESS, 0, handle);
2183 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2187 /* replay case, create LOV EA from client data */
2188 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2189 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2190 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2191 spec->u.sp_ea.eadatalen);
2192 rc = mdo_declare_xattr_set(env, c, buf,
2193 S_ISDIR(attr->la_mode) ?
2194 XATTR_NAME_LMV : XATTR_NAME_LOV,
2195 LU_XATTR_CREATE, handle);
2199 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2200 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2208 if (S_ISLNK(attr->la_mode)) {
2209 const char *target_name = spec->u.sp_symname;
2210 int sym_len = strlen(target_name);
2211 const struct lu_buf *buf;
2213 buf = mdd_buf_get_const(env, target_name, sym_len);
2214 rc = dt_declare_record_write(env, mdd_object_child(c),
2220 if (spec->sp_cr_file_secctx_name != NULL) {
2221 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2222 spec->sp_cr_file_secctx_size);
2223 rc = mdo_declare_xattr_set(env, c, buf,
2224 spec->sp_cr_file_secctx_name, 0,
2230 if (spec->sp_cr_file_encctx != NULL) {
2231 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2232 spec->sp_cr_file_encctx_size);
2233 rc = mdo_declare_xattr_set(env, c, buf,
2234 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2243 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2244 struct mdd_object *p, struct mdd_object *c,
2245 const struct lu_name *name,
2246 struct lu_attr *attr,
2247 struct thandle *handle,
2248 const struct md_op_spec *spec,
2249 struct linkea_data *ldata,
2250 struct lu_buf *def_acl_buf,
2251 struct lu_buf *acl_buf,
2252 struct lu_buf *hsm_buf,
2253 struct dt_allocation_hint *hint)
2257 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2258 def_acl_buf, acl_buf, hsm_buf, hint);
2262 if (S_ISDIR(attr->la_mode)) {
2263 rc = mdo_declare_ref_add(env, p, handle);
2268 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2269 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2273 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2274 enum changelog_rec_type type;
2276 rc = mdo_declare_index_insert(env, p, mdd_object_fid(c),
2277 attr->la_mode, name->ln_name,
2282 rc = mdd_declare_links_add(env, c, handle, ldata);
2287 la->la_valid = LA_CTIME | LA_MTIME;
2288 rc = mdo_declare_attr_set(env, p, la, handle);
2292 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2293 S_ISREG(attr->la_mode) ? CL_CREATE :
2294 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2296 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2305 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2306 struct lu_attr *la, struct lu_buf *def_acl_buf,
2307 struct lu_buf *acl_buf)
2313 if (S_ISLNK(la->la_mode)) {
2314 acl_buf->lb_len = 0;
2315 def_acl_buf->lb_len = 0;
2319 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2320 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2321 XATTR_NAME_ACL_DEFAULT);
2322 mdd_read_unlock(env, pobj);
2324 /* ACL buffer size is not enough, need realloc */
2325 if (rc > acl_buf->lb_len)
2328 /* If there are default ACL, fix mode/ACL by default ACL */
2329 def_acl_buf->lb_len = rc;
2330 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2331 acl_buf->lb_len = rc;
2332 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2335 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2336 /* If there are no default ACL, fix mode by mask */
2337 struct lu_ucred *uc = lu_ucred(env);
2339 /* The create triggered by MDT internal events, such as
2340 * LFSCK reset, will not contain valid "uc". */
2341 if (unlikely(uc != NULL))
2342 la->la_mode &= ~uc->uc_umask;
2344 acl_buf->lb_len = 0;
2345 def_acl_buf->lb_len = 0;
2352 * Create a metadata object and initialize it, set acl, xattr.
2354 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2355 struct mdd_object *son, struct lu_attr *attr,
2356 struct md_op_spec *spec, struct lu_buf *acl_buf,
2357 struct lu_buf *def_acl_buf,
2358 struct lu_buf *hsm_buf,
2359 struct dt_allocation_hint *hint,
2360 struct thandle *handle, bool initsecctx)
2362 const struct lu_buf *buf;
2365 mdd_write_lock(env, son, DT_TGT_CHILD);
2366 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2371 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2372 * created in declare phase, they also needs to be added to master
2373 * object as sub-directory entry. So it has to initialize the master
2374 * object, then set dir striped EA.(in mdo_xattr_set) */
2375 rc = mdd_object_initialize(env, mdd_object_fid(pobj), son, attr,
2378 GOTO(err_destroy, rc);
2381 * in case of replay we just set LOVEA provided by the client
2382 * XXX: I think it would be interesting to try "old" way where
2383 * MDT calls this xattr_set(LOV) in a different transaction.
2384 * probably this way we code can be made better.
2387 /* During creation, there are only a few cases we need do xattr_set to
2389 * 1. regular file: see comments above.
2390 * 2. dir: inherit default striping or pool settings from parent.
2391 * 3. create striped directory with provided stripeEA.
2392 * 4. create striped directory because inherit default layout from the
2395 if (spec->no_create ||
2396 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2397 S_ISDIR(attr->la_mode)) {
2398 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2399 spec->u.sp_ea.eadatalen);
2400 rc = mdo_xattr_set(env, son, buf,
2401 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2403 LU_XATTR_CREATE, handle);
2405 GOTO(err_destroy, rc);
2408 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2411 memset(&mh, 0, sizeof(mh));
2412 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2413 mh.mh_arch_id = spec->sp_archive_id;
2414 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2415 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2418 GOTO(err_destroy, rc);
2421 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2422 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2423 S_ISDIR(attr->la_mode)) {
2424 /* set default acl */
2425 rc = mdo_xattr_set(env, son, def_acl_buf,
2426 XATTR_NAME_ACL_DEFAULT, 0,
2429 GOTO(err_destroy, rc);
2431 /* set its own acl */
2432 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2433 rc = mdo_xattr_set(env, son, acl_buf,
2434 XATTR_NAME_ACL_ACCESS,
2437 GOTO(err_destroy, rc);
2441 if (S_ISLNK(attr->la_mode)) {
2442 struct dt_object *dt = mdd_object_child(son);
2443 const char *target_name = spec->u.sp_symname;
2444 int sym_len = strlen(target_name);
2447 buf = mdd_buf_get_const(env, target_name, sym_len);
2448 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2452 GOTO(err_initlized, rc = -EFAULT);
2455 if (initsecctx && spec->sp_cr_file_secctx_name != NULL) {
2456 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2457 spec->sp_cr_file_secctx_size);
2458 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2461 GOTO(err_initlized, rc);
2464 if (spec->sp_cr_file_encctx != NULL) {
2465 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2466 spec->sp_cr_file_encctx_size);
2467 rc = mdo_xattr_set(env, son, buf,
2468 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2471 GOTO(err_initlized, rc);
2475 if (unlikely(rc != 0)) {
2477 if (S_ISDIR(attr->la_mode)) {
2478 /* Drop the reference, no need to delete "."/"..",
2479 * because the object to be destroied directly. */
2480 rc2 = mdo_ref_del(env, son, handle);
2484 rc2 = mdo_ref_del(env, son, handle);
2488 mdo_destroy(env, son, handle);
2491 mdd_write_unlock(env, son);
2495 static int mdd_index_delete(const struct lu_env *env,
2496 struct mdd_object *mdd_pobj,
2497 struct lu_attr *cattr,
2498 const struct lu_name *lname)
2500 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2501 struct thandle *handle;
2505 handle = mdd_trans_create(env, mdd);
2507 RETURN(PTR_ERR(handle));
2509 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2514 if (S_ISDIR(cattr->la_mode)) {
2515 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2520 /* Since this will only be used in the error handler path,
2521 * Let's set the thandle to be local and not mess the transno */
2522 handle->th_local = 1;
2523 rc = mdd_trans_start(env, mdd, handle);
2527 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2528 S_ISDIR(cattr->la_mode), handle);
2532 rc = mdd_trans_stop(env, mdd, rc, handle);
2538 * Create object and insert it into namespace.
2540 * Two operations have to be performed:
2542 * - an allocation of a new object (->do_create()), and
2543 * - an insertion into a parent index (->dio_insert()).
2545 * Due to locking, operation order is not important, when both are
2546 * successful, *but* error handling cases are quite different:
2548 * - if insertion is done first, and following object creation fails,
2549 * insertion has to be rolled back, but this operation might fail
2550 * also leaving us with dangling index entry.
2552 * - if creation is done first, is has to be undone if insertion fails,
2553 * leaving us with leaked space, which is not good but not fatal.
2555 * It seems that creation-first is simplest solution, but it is sub-optimal
2561 * case, because second mkdir is bound to create object, only to
2562 * destroy it immediately.
2564 * To avoid this follow local file systems that do double lookup:
2566 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2567 * 1. create (mdd_create_object_internal())
2568 * 2. insert (__mdd_index_insert(), lookup again)
2570 * \param[in] pobj parent object
2571 * \param[in] lname name of child being created
2572 * \param[in,out] child child object being created
2573 * \param[in] spec additional create parameters
2574 * \param[in] ma attributes for new child object
2576 * \retval 0 on success
2577 * \retval negative errno on failure
2579 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2580 const struct lu_name *lname, struct md_object *child,
2581 struct md_op_spec *spec, struct md_attr *ma)
2583 struct mdd_thread_info *info = mdd_env_info(env);
2584 struct lu_attr *la = &info->mti_la_for_fix;
2585 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2586 struct mdd_object *son = md2mdd_obj(child);
2587 struct mdd_device *mdd = mdo2mdd(pobj);
2588 struct lu_attr *attr = &ma->ma_attr;
2589 struct thandle *handle;
2590 struct lu_attr *pattr = &info->mti_pattr;
2591 struct lu_buf acl_buf;
2592 struct lu_buf def_acl_buf;
2593 struct lu_buf hsm_buf;
2594 struct linkea_data *ldata = &info->mti_link_data;
2595 const char *name = lname->ln_name;
2596 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2597 int acl_size = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
2602 rc = mdd_la_get(env, mdd_pobj, pattr);
2606 /* Sanity checks before big job. */
2607 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2611 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2612 GOTO(out_free, rc = -EINPROGRESS);
2614 handle = mdd_trans_create(env, mdd);
2616 GOTO(out_free, rc = PTR_ERR(handle));
2619 acl_buf = *lu_buf_check_and_alloc(&info->mti_xattr_buf, acl_size);
2620 if (!acl_buf.lb_buf)
2621 GOTO(out_stop, rc = -ENOMEM);
2623 def_acl_buf = *lu_buf_check_and_alloc(&info->mti_big_buf, acl_size);
2624 if (!def_acl_buf.lb_buf)
2625 GOTO(out_stop, rc = -ENOMEM);
2627 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2628 if (unlikely(rc == -ERANGE &&
2629 acl_size == LUSTRE_POSIX_ACL_MAX_SIZE_OLD)) {
2630 /* use maximum-sized xattr buffer for too-big default ACL */
2631 acl_size = min_t(unsigned int, mdd->mdd_dt_conf.ddp_max_ea_size,
2633 goto use_bigger_buffer;
2638 if (S_ISDIR(attr->la_mode)) {
2639 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2642 * migrate may create 1-stripe directory, so lod_ah_init()
2643 * doesn't adjust stripe count from lmu.
2645 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1))
2646 lmu->lum_stripe_count = 0;
2649 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2651 memset(ldata, 0, sizeof(*ldata));
2652 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2653 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2656 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2657 &tfid, lname, 1, 0, ldata);
2659 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2660 mdd_object_fid(mdd_pobj),
2661 lname, 1, 0, ldata);
2664 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2665 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2667 memset(&hsm_buf, 0, sizeof(hsm_buf));
2668 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2669 if (hsm_buf.lb_buf == NULL)
2670 GOTO(out_stop, rc = -ENOMEM);
2673 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2674 handle, spec, ldata, &def_acl_buf, &acl_buf,
2679 rc = mdd_trans_start(env, mdd, handle);
2683 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2684 &def_acl_buf, &hsm_buf, hint, handle, true);
2688 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2689 mdd_write_lock(env, son, DT_TGT_CHILD);
2690 son->mod_flags |= VOLATILE_OBJ;
2691 rc = mdd_orphan_insert(env, son, handle);
2692 GOTO(out_volatile, rc);
2694 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(son),
2695 attr->la_mode, name, handle);
2697 GOTO(err_created, rc);
2699 mdd_links_add(env, son, mdd_object_fid(mdd_pobj), lname,
2702 /* update parent directory mtime/ctime */
2704 la->la_valid = LA_CTIME | LA_MTIME;
2705 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2707 GOTO(err_insert, rc);
2713 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2714 rc2 = mdd_orphan_delete(env, son, handle);
2716 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2717 S_ISDIR(attr->la_mode),
2723 mdd_write_lock(env, son, DT_TGT_CHILD);
2724 if (S_ISDIR(attr->la_mode)) {
2725 /* Drop the reference, no need to delete "."/"..",
2726 * because the object is to be destroyed directly. */
2727 rc2 = mdo_ref_del(env, son, handle);
2729 mdd_write_unlock(env, son);
2734 /* For volatile files drop one link immediately, since there is
2735 * no filename in the namespace, and save any error returned. */
2736 rc2 = mdo_ref_del(env, son, handle);
2738 mdd_write_unlock(env, son);
2739 if (unlikely(rc == 0))
2744 /* Don't destroy the volatile object on success */
2745 if (likely(rc != 0))
2746 mdo_destroy(env, son, handle);
2747 mdd_write_unlock(env, son);
2750 if (rc == 0 && fid_is_namespace_visible(mdd_object_fid(son)) &&
2751 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2752 rc = mdd_changelog_ns_store(env, mdd,
2753 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2754 S_ISREG(attr->la_mode) ? CL_CREATE :
2755 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2756 0, son, mdd_object_fid(mdd_pobj), NULL, NULL,
2757 lname, NULL, handle);
2759 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2761 /* If creation fails, it is most likely due to the remote update
2762 * failure, because local transaction will mostly succeed at
2763 * this stage. There is no easy way to rollback all of previous
2764 * updates, so let's remove the object from namespace, and
2765 * LFSCK should handle the orphan object. */
2766 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2767 mdd_index_delete(env, mdd_pobj, attr, lname);
2771 if (is_vmalloc_addr(ldata->ld_buf))
2772 /* if we vmalloced a large buffer drop it */
2773 lu_buf_free(ldata->ld_buf);
2775 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2776 lu_buf_free(&hsm_buf);
2778 /* The child object shouldn't be cached anymore */
2780 set_bit(LU_OBJECT_HEARD_BANSHEE,
2781 &child->mo_lu.lo_header->loh_flags);
2785 /* has not mdd_write{read}_lock on any obj yet. */
2786 static int mdd_rename_sanity_check(const struct lu_env *env,
2787 struct mdd_object *src_pobj,
2788 const struct lu_attr *pattr,
2789 struct mdd_object *tgt_pobj,
2790 const struct lu_attr *tpattr,
2791 struct mdd_object *sobj,
2792 const struct lu_attr *cattr,
2793 struct mdd_object *tobj,
2794 const struct lu_attr *tattr)
2799 /* XXX: when get here, sobj must NOT be NULL,
2800 * the other case has been processed in cld_rename
2801 * before mdd_rename and enable MDS_PERM_BYPASS. */
2805 * If we are using project inheritance, we only allow renames
2806 * into our tree when the project IDs are the same; otherwise
2807 * tree quota mechanism would be circumvented.
2809 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2810 tpattr->la_projid != cattr->la_projid) ||
2811 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2812 (pattr->la_projid != tpattr->la_projid)))
2815 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2819 /* XXX: when get here, "tobj == NULL" means tobj must
2820 * NOT exist (neither on remote MDS, such case has been
2821 * processed in cld_rename before mdd_rename and enable
2823 * So check may_create, but not check may_unlink. */
2825 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2826 (src_pobj != tgt_pobj));
2828 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2829 (src_pobj != tgt_pobj), 1);
2831 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2832 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2837 static int mdd_declare_rename(const struct lu_env *env,
2838 struct mdd_device *mdd,
2839 struct mdd_object *mdd_spobj,
2840 struct mdd_object *mdd_tpobj,
2841 struct mdd_object *mdd_sobj,
2842 struct mdd_object *mdd_tobj,
2843 const struct lu_name *sname,
2844 const struct lu_name *tname,
2846 struct linkea_data *ldata,
2847 struct thandle *handle)
2849 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2852 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2853 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2859 /* name from source dir */
2860 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2864 /* .. from source child */
2865 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2866 /* source child can be directory, count by source dir's nlink */
2867 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2870 if (mdd_spobj != mdd_tpobj) {
2871 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2876 rc = mdo_declare_index_insert(env, mdd_sobj,
2877 mdd_object_fid(mdd_tpobj),
2878 S_IFDIR, dotdot, handle);
2883 /* new target child can be directory,
2884 * counted by target dir's nlink */
2885 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2890 la->la_valid = LA_CTIME | LA_MTIME;
2891 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2895 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2899 la->la_valid = LA_CTIME;
2900 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2904 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2909 rc = mdo_declare_index_insert(env, mdd_tpobj, mdd_object_fid(mdd_sobj),
2910 mdd_object_type(mdd_sobj),
2911 tname->ln_name, handle);
2915 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2916 /* delete target child in target parent directory */
2917 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2922 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2926 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2927 /* target child can be directory,
2928 * delete "." reference in target child directory */
2929 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2933 /* delete ".." reference in target parent directory */
2934 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2939 la->la_valid = LA_CTIME;
2940 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2944 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2949 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2957 static int mdd_migrate_object(const struct lu_env *env,
2958 struct mdd_object *spobj,
2959 struct mdd_object *tpobj,
2960 struct mdd_object *sobj,
2961 struct mdd_object *tobj,
2962 const struct lu_name *sname,
2963 const struct lu_name *tname,
2964 struct md_op_spec *spec,
2965 struct md_attr *ma);
2967 /* src object can be remote that is why we use only fid and type of object */
2968 static int mdd_rename(const struct lu_env *env,
2969 struct md_object *src_pobj, struct md_object *tgt_pobj,
2970 const struct lu_fid *lf, const struct lu_name *lsname,
2971 struct md_object *tobj, const struct lu_name *ltname,
2974 const char *sname = lsname->ln_name;
2975 const char *tname = ltname->ln_name;
2976 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2977 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2978 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2979 struct mdd_device *mdd = mdo2mdd(src_pobj);
2980 struct mdd_object *mdd_sobj = NULL; /* source object */
2981 struct mdd_object *mdd_tobj = NULL;
2982 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2983 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2984 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2985 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2986 struct thandle *handle;
2987 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2988 const struct lu_fid *tpobj_fid = mdd_object_fid(mdd_tpobj);
2989 const struct lu_fid *spobj_fid = mdd_object_fid(mdd_spobj);
2992 bool tobj_locked = 0;
2993 unsigned cl_flags = 0;
2997 /* let unlink to complete and commit */
2998 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
3001 mdd_tobj = md2mdd_obj(tobj);
3003 mdd_sobj = mdd_object_find(env, mdd, lf);
3004 if (IS_ERR(mdd_sobj))
3005 RETURN(PTR_ERR(mdd_sobj));
3007 rc = mdd_la_get(env, mdd_sobj, cattr);
3009 GOTO(out_pending, rc);
3011 /* if rename is cross MDTs, migrate symlink if it doesn't have other
3012 * hard links, and target doesn't exist.
3014 if (mdd_object_remote(mdd_sobj) && S_ISLNK(cattr->la_mode) &&
3015 cattr->la_nlink == 1 && !tobj) {
3016 struct md_op_spec *spec = &mdd_env_info(env)->mti_spec;
3017 struct lu_device *ld = &mdd->mdd_md_dev.md_lu_dev;
3020 rc = ld->ld_ops->ldo_fid_alloc(env, ld, &tfid, &tgt_pobj->mo_lu,
3023 GOTO(out_pending, rc);
3025 mdd_tobj = mdd_object_find(env, mdd, &tfid);
3026 if (IS_ERR(mdd_tobj))
3027 GOTO(out_pending, rc = PTR_ERR(mdd_tobj));
3029 memset(spec, 0, sizeof(*spec));
3030 rc = mdd_migrate_object(env, mdd_spobj, mdd_tpobj, mdd_sobj,
3031 mdd_tobj, lsname, ltname, spec, ma);
3032 mdd_object_put(env, mdd_tobj);
3033 GOTO(out_pending, rc);
3036 rc = mdd_la_get(env, mdd_spobj, pattr);
3038 GOTO(out_pending, rc);
3041 rc = mdd_la_get(env, mdd_tobj, tattr);
3043 GOTO(out_pending, rc);
3044 /* search for an existing archive.
3045 * we should check ahead as the object
3046 * can be destroyed in this transaction */
3047 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3048 cl_flags |= CLF_RENAME_LAST_EXISTS;
3051 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3053 GOTO(out_pending, rc);
3055 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3056 mdd_sobj, cattr, mdd_tobj, tattr);
3058 GOTO(out_pending, rc);
3060 rc = mdd_name_check(mdd, ltname);
3062 GOTO(out_pending, rc);
3064 handle = mdd_trans_create(env, mdd);
3066 GOTO(out_pending, rc = PTR_ERR(handle));
3068 memset(ldata, 0, sizeof(*ldata));
3069 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3070 lsname, mdd_object_fid(mdd_tpobj), ltname,
3075 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3076 mdd_tobj, lsname, ltname, ma, ldata, handle);
3080 rc = mdd_trans_start(env, mdd, handle);
3084 is_dir = S_ISDIR(cattr->la_mode);
3086 /* Remove source name from source directory */
3087 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3091 /* "mv dir1 dir2" needs "dir1/.." link update */
3092 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3093 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3095 GOTO(fixup_spobj2, rc);
3097 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3100 GOTO(fixup_spobj, rc);
3103 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3104 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3106 /* tname might been renamed to something else */
3107 GOTO(fixup_spobj, rc);
3110 /* Insert new fid with target name into target dir */
3111 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3114 GOTO(fixup_tpobj, rc);
3116 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3117 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3119 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3120 la->la_valid = LA_CTIME;
3121 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3123 GOTO(fixup_tpobj, rc);
3125 /* Update the linkEA for the source object */
3126 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3127 rc = mdd_links_rename(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3128 lsname, mdd_object_fid(mdd_tpobj), ltname,
3129 handle, ldata, 0, 0);
3131 /* Old files might not have EA entry */
3132 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3133 lsname, handle, NULL, 0);
3134 mdd_write_unlock(env, mdd_sobj);
3135 /* We don't fail the transaction if the link ea can't be
3136 updated -- fid2path will use alternate lookup method. */
3139 /* Remove old target object
3140 * For tobj is remote case cmm layer has processed
3141 * and set tobj to NULL then. So when tobj is NOT NULL,
3142 * it must be local one.
3144 if (tobj && mdd_object_exists(mdd_tobj)) {
3145 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3147 if (mdd_is_dead_obj(mdd_tobj)) {
3148 /* shld not be dead, something is wrong */
3149 CERROR("tobj is dead, something is wrong\n");
3153 mdo_ref_del(env, mdd_tobj, handle);
3155 /* Remove dot reference. */
3156 if (S_ISDIR(tattr->la_mode))
3157 mdo_ref_del(env, mdd_tobj, handle);
3160 /* fetch updated nlink */
3161 rc = mdd_la_get(env, mdd_tobj, tattr);
3163 CERROR("%s: Failed to get nlink for tobj "
3165 mdd2obd_dev(mdd)->obd_name,
3166 PFID(tpobj_fid), rc);
3167 GOTO(fixup_tpobj, rc);
3170 la->la_valid = LA_CTIME;
3171 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3173 CERROR("%s: Failed to set ctime for tobj "
3175 mdd2obd_dev(mdd)->obd_name,
3176 PFID(tpobj_fid), rc);
3177 GOTO(fixup_tpobj, rc);
3180 /* XXX: this transfer to ma will be removed with LOD/OSP */
3181 ma->ma_attr = *tattr;
3182 ma->ma_valid |= MA_INODE;
3183 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3186 CERROR("%s: Failed to unlink tobj "
3188 mdd2obd_dev(mdd)->obd_name,
3189 PFID(tpobj_fid), rc);
3190 GOTO(fixup_tpobj, rc);
3193 /* fetch updated nlink */
3194 rc = mdd_la_get(env, mdd_tobj, tattr);
3195 if (rc == -ENOENT) {
3196 /* the object got removed, let's
3197 * return the latest known attributes */
3198 tattr->la_nlink = 0;
3200 } else if (rc != 0) {
3201 CERROR("%s: Failed to get nlink for tobj "
3203 mdd2obd_dev(mdd)->obd_name,
3204 PFID(tpobj_fid), rc);
3205 GOTO(fixup_tpobj, rc);
3207 /* XXX: this transfer to ma will be removed with LOD/OSP */
3208 ma->ma_attr = *tattr;
3209 ma->ma_valid |= MA_INODE;
3211 if (tattr->la_nlink == 0)
3212 cl_flags |= CLF_RENAME_LAST;
3214 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3217 la->la_valid = LA_CTIME | LA_MTIME;
3218 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3220 GOTO(fixup_tpobj, rc);
3222 if (mdd_spobj != mdd_tpobj) {
3223 la->la_valid = LA_CTIME | LA_MTIME;
3224 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3226 GOTO(fixup_tpobj, rc);
3233 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3235 CWARN("tp obj fix error %d\n",rc2);
3237 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3238 !mdd_is_dead_obj(mdd_tobj)) {
3240 mdo_ref_add(env, mdd_tobj, handle);
3242 mdo_ref_add(env, mdd_tobj, handle);
3245 rc2 = __mdd_index_insert(env, mdd_tpobj,
3246 mdd_object_fid(mdd_tobj),
3247 mdd_object_type(mdd_tobj),
3250 CWARN("tp obj fix error: rc = %d\n", rc2);
3255 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3256 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3258 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3259 mdd2obd_dev(mdd)->obd_name, rc2);
3262 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3265 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3266 mdd2obd_dev(mdd)->obd_name, rc2);
3271 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3272 mdd_object_type(mdd_sobj), sname,
3275 CWARN("sp obj fix error: rc = %d\n", rc2);
3280 mdd_write_unlock(env, mdd_tobj);
3283 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3284 mdd_tobj, tpobj_fid, lf, spobj_fid,
3285 ltname, lsname, handle);
3288 rc = mdd_trans_stop(env, mdd, rc, handle);
3291 mdd_object_put(env, mdd_sobj);
3296 * Check whether we should migrate the file/dir
3298 * < 0 permission check failed or other error.
3299 * = 0 the file can be migrated.
3301 static int mdd_migrate_sanity_check(const struct lu_env *env,
3302 struct mdd_device *mdd,
3303 struct mdd_object *spobj,
3304 struct mdd_object *tpobj,
3305 struct mdd_object *sobj,
3306 struct mdd_object *tobj,
3307 const struct lu_attr *spattr,
3308 const struct lu_attr *tpattr,
3309 const struct lu_attr *attr)
3315 if (!mdd_object_remote(sobj)) {
3316 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3317 if (sobj->mod_count > 0) {
3318 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3319 mdd_obj_dev_name(sobj),
3320 PFID(mdd_object_fid(sobj)),
3322 mdd_read_unlock(env, sobj);
3325 mdd_read_unlock(env, sobj);
3328 if (mdd_object_exists(tobj))
3331 rc = mdd_may_delete(env, spobj, spattr, sobj, attr, NULL, 1, 0);
3335 rc = mdd_may_create(env, tpobj, tpattr, NULL, true);
3340 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3341 struct mdd_object *obj,
3342 const struct lu_buf *buf,
3344 int fl, struct thandle *handle);
3346 /* iterate xattrs, but ignore LMA, LMV, and LINKEA if 'skip_linkea' is set. */
3347 static int mdd_iterate_xattrs(const struct lu_env *env,
3348 struct mdd_object *sobj,
3349 struct mdd_object *tobj,
3351 struct thandle *handle,
3354 struct mdd_thread_info *info = mdd_env_info(env);
3356 struct lu_buf list_xbuf;
3357 struct lu_buf cbxbuf;
3358 struct lu_buf xbuf = { NULL };
3367 /* retrieve xattr list from the old object */
3368 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3369 if (list_xsize == -ENODATA)
3375 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3376 if (info->mti_big_buf.lb_buf == NULL)
3379 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3380 list_xbuf.lb_len = list_xsize;
3381 rc = mdo_xattr_list(env, sobj, &list_xbuf);
3387 xname = list_xbuf.lb_buf;
3389 xlen = strnlen(xname, rem - 1) + 1;
3390 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3391 strcmp(XATTR_NAME_LMV, xname) == 0)
3395 strcmp(XATTR_NAME_LINK, xname) == 0)
3398 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3399 if (xsize == -ENODATA)
3402 GOTO(out, rc = xsize);
3404 lu_buf_check_and_alloc(&xbuf, xsize);
3405 if (xbuf.lb_buf == NULL)
3406 GOTO(out, rc = -ENOMEM);
3408 rc = mdo_xattr_get(env, sobj, &xbuf, xname);
3415 cbxbuf.lb_len = xsize;
3417 rc = cb(env, tobj, &cbxbuf, xname, 0, handle);
3418 if (unlikely(rc == -ENOSPC &&
3419 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3420 rc = linkea_overflow_shrink(
3421 (struct linkea_data *)(cbxbuf.lb_buf));
3422 if (likely(rc > 0)) {
3440 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3441 struct mdd_object *sobj,
3442 struct mdd_object *tobj,
3443 const struct lu_name *sname,
3444 const struct lu_fid *sfid,
3445 const struct lu_name *lname,
3446 const struct lu_fid *fid,
3448 struct thandle *handle);
3450 static int mdd_declare_update_link(const struct lu_env *env,
3451 struct mdd_object *sobj,
3452 struct mdd_object *tobj,
3453 const struct lu_name *tname,
3454 const struct lu_fid *tpfid,
3455 const struct lu_name *lname,
3456 const struct lu_fid *fid,
3458 struct thandle *handle)
3460 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3461 struct mdd_object *pobj;
3465 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3466 !strcmp(tname->ln_name, lname->ln_name))
3469 pobj = mdd_object_find(env, mdd, fid);
3471 return PTR_ERR(pobj);
3474 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3476 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
3477 mdd_object_type(sobj),
3478 lname->ln_name, handle);
3479 mdd_object_put(env, pobj);
3483 rc = mdo_declare_ref_add(env, tobj, handle);
3487 rc = mdo_declare_ref_del(env, sobj, handle);
3491 static int mdd_update_link(const struct lu_env *env,
3492 struct mdd_object *sobj,
3493 struct mdd_object *tobj,
3494 const struct lu_name *tname,
3495 const struct lu_fid *tpfid,
3496 const struct lu_name *lname,
3497 const struct lu_fid *fid,
3499 struct thandle *handle)
3501 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3502 struct mdd_object *pobj;
3507 LASSERT(lu_name_is_valid(lname));
3510 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3511 !strncmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3514 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3515 PFID(fid), PNAME(lname), PFID(mdd_object_fid(tobj)));
3517 pobj = mdd_object_find(env, mdd, fid);
3519 CWARN("%s: cannot find obj "DFID": %ld\n",
3520 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3521 RETURN(PTR_ERR(pobj));
3524 if (!mdd_object_exists(pobj)) {
3525 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3526 mdd_object_put(env, pobj);
3530 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3531 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3533 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(tobj),
3534 mdd_object_type(sobj),
3535 lname->ln_name, handle);
3536 mdd_write_unlock(env, pobj);
3537 mdd_object_put(env, pobj);
3541 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3542 rc = mdo_ref_add(env, tobj, handle);
3543 mdd_write_unlock(env, tobj);
3547 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3548 rc = mdo_ref_del(env, sobj, handle);
3549 mdd_write_unlock(env, sobj);
3554 static inline int mdd_fld_lookup(const struct lu_env *env,
3555 struct mdd_device *mdd,
3556 const struct lu_fid *fid,
3559 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
3560 struct seq_server_site *ss;
3563 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3565 range->lsr_flags = LU_SEQ_RANGE_MDT;
3566 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3570 *mdt_index = range->lsr_index;
3575 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3576 struct mdd_object *sobj,
3577 struct mdd_object *tobj,
3578 const struct lu_name *tname,
3579 const struct lu_fid *tpfid,
3580 const struct lu_name *lname,
3581 const struct lu_fid *fid,
3583 struct thandle *handle)
3585 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3586 __u32 source_mdt_index = *(__u32 *)opaque;
3587 __u32 link_mdt_index;
3593 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3594 !strcmp(tname->ln_name, lname->ln_name))
3597 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3601 RETURN(link_mdt_index == source_mdt_index);
3604 static int mdd_iterate_linkea(const struct lu_env *env,
3605 struct mdd_object *sobj,
3606 struct mdd_object *tobj,
3607 const struct lu_name *tname,
3608 const struct lu_fid *tpfid,
3609 struct linkea_data *ldata,
3611 struct thandle *handle,
3614 struct mdd_thread_info *info = mdd_env_info(env);
3615 char *filename = info->mti_name;
3616 struct lu_name lname;
3623 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3624 linkea_next_entry(ldata)) {
3625 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3628 /* Note: lname might miss \0 at the end */
3629 snprintf(filename, sizeof(info->mti_name), "%.*s",
3630 lname.ln_namelen, lname.ln_name);
3631 lname.ln_name = filename;
3633 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3635 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3643 * Prepare linkea, and check whether file needs migrate: if source still has
3644 * link on source MDT, no need to migrate, just update namespace on source and
3647 * \retval 0 do migrate
3648 * \retval 1 don't migrate
3649 * \retval -errno on failure
3651 static int mdd_migrate_linkea_prepare(const struct lu_env *env,
3652 struct mdd_device *mdd,
3653 struct mdd_object *spobj,
3654 struct mdd_object *tpobj,
3655 struct mdd_object *sobj,
3656 const struct lu_name *sname,
3657 const struct lu_name *tname,
3658 const struct lu_attr *attr,
3659 struct linkea_data *ldata)
3661 __u32 source_mdt_index;
3666 memset(ldata, 0, sizeof(*ldata));
3667 rc = mdd_linkea_prepare(env, sobj, mdd_object_fid(spobj), sname,
3668 mdd_object_fid(tpobj), tname, 1, 0, ldata);
3673 * Then it will check if the file should be migrated. If the file has
3674 * mulitple links, we only need migrate the file if all of its entries
3675 * has been migrated to the remote MDT.
3677 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3680 /* If there are still links locally, don't migrate this file */
3681 LASSERT(ldata->ld_leh != NULL);
3684 * If linkEA is overflow, it means there are some unknown name entries
3685 * under unknown parents, which will prevent the migration.
3687 if (unlikely(ldata->ld_leh->leh_overflow_time))
3690 rc = mdd_fld_lookup(env, mdd, mdd_object_fid(sobj), &source_mdt_index);
3694 rc = mdd_iterate_linkea(env, sobj, NULL, tname, mdd_object_fid(tpobj),
3695 ldata, &source_mdt_index, NULL,
3696 mdd_is_link_on_source_mdt);
3700 static int mdd_declare_migrate_update(const struct lu_env *env,
3701 struct mdd_object *spobj,
3702 struct mdd_object *tpobj,
3703 struct mdd_object *obj,
3704 const struct lu_name *sname,
3705 const struct lu_name *tname,
3706 struct lu_attr *attr,
3707 struct lu_attr *spattr,
3708 struct lu_attr *tpattr,
3709 struct linkea_data *ldata,
3711 struct thandle *handle)
3713 struct mdd_thread_info *info = mdd_env_info(env);
3714 struct lu_attr *la = &info->mti_la_for_fix;
3717 rc = mdo_declare_index_delete(env, spobj, sname->ln_name, handle);
3721 if (S_ISDIR(attr->la_mode)) {
3722 rc = mdo_declare_ref_del(env, spobj, handle);
3727 rc = mdo_declare_index_insert(env, tpobj, mdd_object_fid(obj),
3728 attr->la_mode & S_IFMT,
3729 tname->ln_name, handle);
3733 rc = mdd_declare_links_add(env, obj, handle, ldata);
3737 if (S_ISDIR(attr->la_mode)) {
3738 rc = mdo_declare_ref_add(env, tpobj, handle);
3743 la->la_valid = LA_CTIME | LA_MTIME;
3744 rc = mdo_declare_attr_set(env, spobj, la, handle);
3748 if (tpobj != spobj) {
3749 rc = mdo_declare_attr_set(env, tpobj, la, handle);
3757 static int mdd_declare_migrate_create(const struct lu_env *env,
3758 struct mdd_object *spobj,
3759 struct mdd_object *tpobj,
3760 struct mdd_object *sobj,
3761 struct mdd_object *tobj,
3762 const struct lu_name *sname,
3763 const struct lu_name *tname,
3764 struct lu_attr *spattr,
3765 struct lu_attr *tpattr,
3766 struct lu_attr *attr,
3767 struct lu_buf *sbuf,
3768 struct linkea_data *ldata,
3770 struct md_op_spec *spec,
3771 struct dt_allocation_hint *hint,
3772 struct thandle *handle)
3774 struct mdd_thread_info *info = mdd_env_info(env);
3775 struct md_layout_change *mlc = &info->mti_mlc;
3776 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3781 if (S_ISDIR(attr->la_mode)) {
3782 struct lmv_user_md *lum = spec->u.sp_ea.eadata;
3784 mlc->mlc_opc = MD_LAYOUT_DETACH;
3785 rc = mdo_declare_layout_change(env, sobj, mlc, handle);
3789 lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3790 } else if (S_ISLNK(attr->la_mode)) {
3791 spec->u.sp_symname = sbuf->lb_buf;
3792 } else if (S_ISREG(attr->la_mode)) {
3793 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
3794 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
3797 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
3799 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3800 tname, attr, handle, spec, ldata, NULL, NULL,
3806 * tobj mode will be used in mdo_declare_layout_change(), but it's not
3807 * createb yet, copy from sobj.
3809 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
3810 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
3811 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
3813 if (S_ISDIR(attr->la_mode)) {
3815 /* if sobj is not striped, fake a 1-stripe LMV */
3816 LASSERT(sizeof(info->mti_key) >
3817 lmv_mds_md_size(1, LMV_MAGIC_V1));
3818 lmv = (typeof(lmv))info->mti_key;
3819 memset(lmv, 0, sizeof(*lmv));
3820 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3821 lmv->lmv_stripe_count = cpu_to_le32(1);
3822 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
3823 fid_le_to_cpu(&lmv->lmv_stripe_fids[0],
3824 mdd_object_fid(sobj));
3825 mlc->mlc_buf.lb_buf = lmv;
3826 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3828 mlc->mlc_buf = *sbuf;
3830 mlc->mlc_opc = MD_LAYOUT_ATTACH;
3831 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
3836 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle,
3837 mdo_declare_xattr_set);
3841 if (S_ISREG(attr->la_mode)) {
3842 struct lu_buf fid_buf;
3844 handle->th_complex = 1;
3846 /* target may be remote, update PFID via sobj. */
3847 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
3848 fid_buf.lb_len = sizeof(struct lu_fid);
3849 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
3854 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
3859 if (!S_ISDIR(attr->la_mode)) {
3860 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
3861 mdd_object_fid(tpobj), ldata, NULL,
3862 handle, mdd_declare_update_link);
3867 if (!S_ISDIR(attr->la_mode) || lmv) {
3868 rc = mdo_declare_ref_del(env, sobj, handle);
3872 if (S_ISDIR(attr->la_mode)) {
3873 rc = mdo_declare_ref_del(env, sobj, handle);
3878 rc = mdo_declare_destroy(env, sobj, handle);
3883 rc = mdd_declare_migrate_update(env, spobj, tpobj, tobj, sname, tname,
3884 attr, spattr, tpattr, ldata, ma,
3890 * migrate dirent from \a spobj to \a tpobj.
3892 static int mdd_migrate_update(const struct lu_env *env,
3893 struct mdd_object *spobj,
3894 struct mdd_object *tpobj,
3895 struct mdd_object *obj,
3896 const struct lu_name *sname,
3897 const struct lu_name *tname,
3898 struct lu_attr *attr,
3899 struct lu_attr *spattr,
3900 struct lu_attr *tpattr,
3901 struct linkea_data *ldata,
3903 struct thandle *handle)
3905 struct mdd_thread_info *info = mdd_env_info(env);
3906 struct lu_attr *la = &info->mti_la_for_fix;
3911 CDEBUG(D_INFO, "update "DFID" from "DFID"/%s to "DFID"/%s\n",
3912 PFID(mdd_object_fid(obj)), PFID(mdd_object_fid(spobj)),
3913 sname->ln_name, PFID(mdd_object_fid(tpobj)), tname->ln_name);
3915 rc = __mdd_index_delete(env, spobj, sname->ln_name,
3916 S_ISDIR(attr->la_mode), handle);
3920 rc = __mdd_index_insert(env, tpobj, mdd_object_fid(obj),
3921 attr->la_mode & S_IFMT,
3922 tname->ln_name, handle);
3926 rc = mdd_links_write(env, obj, ldata, handle);
3930 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3931 la->la_valid = LA_CTIME | LA_MTIME;
3932 mdd_write_lock(env, spobj, DT_SRC_PARENT);
3933 rc = mdd_update_time(env, spobj, spattr, la, handle);
3934 mdd_write_unlock(env, spobj);
3938 if (tpobj != spobj) {
3939 la->la_valid = LA_CTIME | LA_MTIME;
3940 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
3941 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
3942 mdd_write_unlock(env, tpobj);
3951 * Migrate file/dir to target MDT.
3953 * Create target according to \a spec, and then migrate xattrs, if it's
3954 * directory, migrate source stripes to target.
3956 * \param[in] env execution environment
3957 * \param[in] spobj source parent object
3958 * \param[in] tpobj target parent object
3959 * \param[in] sobj source object
3960 * \param[in] tobj target object
3961 * \param[in] lname file name
3962 * \param[in] spattr source parent attributes
3963 * \param[in] tpattr target parent attributes
3964 * \param[in] attr source attributes
3965 * \param[in] sbuf source LMV buf
3966 * \param[in] spec migrate create spec
3967 * \param[in] hint target creation hint
3968 * \param[in] handle tranasction handle
3970 * \retval 0 on success
3971 * \retval -errno on failure
3973 static int mdd_migrate_create(const struct lu_env *env,
3974 struct mdd_object *spobj,
3975 struct mdd_object *tpobj,
3976 struct mdd_object *sobj,
3977 struct mdd_object *tobj,
3978 const struct lu_name *sname,
3979 const struct lu_name *tname,
3980 struct lu_attr *spattr,
3981 struct lu_attr *tpattr,
3982 struct lu_attr *attr,
3983 const struct lu_buf *sbuf,
3984 struct linkea_data *ldata,
3986 struct md_op_spec *spec,
3987 struct dt_allocation_hint *hint,
3988 struct thandle *handle)
3995 * migrate sobj stripes to tobj if it's directory:
3996 * 1. detach stripes from sobj.
3997 * 2. attach stripes to tobj, see mdd_declare_migrate_mdt().
3998 * 3. create stripes for tobj, see lod_xattr_set_lmv().
4000 if (S_ISDIR(attr->la_mode)) {
4001 struct mdd_thread_info *info = mdd_env_info(env);
4002 struct md_layout_change *mlc = &info->mti_mlc;
4004 mlc->mlc_opc = MD_LAYOUT_DETACH;
4006 mdd_write_lock(env, sobj, DT_SRC_PARENT);
4007 rc = mdo_layout_change(env, sobj, mlc, handle);
4008 mdd_write_unlock(env, sobj);
4013 /* don't set nlink from sobj */
4014 attr->la_valid &= ~LA_NLINK;
4016 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4017 hint, handle, false);
4021 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4022 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
4023 mdd_write_unlock(env, tobj);
4027 /* for regular file, update OST objects XATTR_NAME_FID */
4028 if (S_ISREG(attr->la_mode)) {
4029 struct lu_buf fid_buf;
4031 /* target may be remote, update PFID via sobj. */
4032 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
4033 fid_buf.lb_len = sizeof(struct lu_fid);
4034 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4039 /* delete LOV to avoid deleting OST objs when destroying sobj */
4040 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4041 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4042 mdd_write_unlock(env, sobj);
4043 /* O_DELAY_CREATE file may not have LOV, ignore -ENODATA */
4044 if (rc && rc != -ENODATA)
4049 /* update links FID */
4050 if (!S_ISDIR(attr->la_mode)) {
4051 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
4052 mdd_object_fid(tpobj), ldata,
4053 NULL, handle, mdd_update_link);
4058 /* don't destroy sobj if it's plain directory */
4059 if (!S_ISDIR(attr->la_mode) || sbuf->lb_buf) {
4060 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4061 rc = mdo_ref_del(env, sobj, handle);
4063 if (S_ISDIR(attr->la_mode))
4064 rc = mdo_ref_del(env, sobj, handle);
4066 rc = mdo_destroy(env, sobj, handle);
4068 mdd_write_unlock(env, sobj);
4073 rc = mdd_migrate_update(env, spobj, tpobj, tobj, sname, tname, attr,
4074 spattr, tpattr, ldata, ma, handle);
4079 /* NB: if user issued different migrate command, we can't adjust it silently
4080 * here, because this command will decide target MDT in subdir migration in
4083 static int mdd_migrate_cmd_check(struct mdd_device *mdd,
4084 const struct lmv_mds_md_v1 *lmv,
4085 const struct lmv_user_md_v1 *lum,
4086 const struct lu_name *lname)
4088 __u32 lum_stripe_count = lum->lum_stripe_count;
4089 __u32 lmv_hash_type = lmv->lmv_hash_type;
4091 if (!lmv_is_sane(lmv))
4094 /* if stripe_count unspecified, set to 1 */
4095 if (!lum_stripe_count)
4096 lum_stripe_count = cpu_to_le32(1);
4098 lmv_hash_type &= ~cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4100 /* TODO: check specific MDTs */
4101 if (lum_stripe_count != lmv->lmv_migrate_offset ||
4102 lum->lum_stripe_offset != lmv->lmv_master_mdt_index ||
4103 (lum->lum_hash_type && lum->lum_hash_type != lmv_hash_type)) {
4104 CERROR("%s: '"DNAME"' migration was interrupted, run 'lfs migrate -m %d -c %d -H %s "DNAME"' to finish migration.\n",
4105 mdd2obd_dev(mdd)->obd_name, PNAME(lname),
4106 le32_to_cpu(lmv->lmv_master_mdt_index),
4107 le32_to_cpu(lmv->lmv_migrate_offset),
4108 mdt_hash_name[le32_to_cpu(lmv_hash_type)],
4117 * Internal function to migrate directory or file between MDTs.
4119 * migrate source to target in following steps:
4120 * 1. create target, append source stripes after target's if it's directory,
4121 * migrate xattrs and update fid of source links.
4122 * 2. update namespace: migrate dirent from source parent to target parent,
4123 * update file linkea, and destroy source if it's not needed any more.
4125 * \param[in] env execution environment
4126 * \param[in] spobj source parent object
4127 * \param[in] tpobj target parent object
4128 * \param[in] sobj source object
4129 * \param[in] tobj target object
4130 * \param[in] sname source file name
4131 * \param[in] tname target file name
4132 * \param[in] spec target creation spec
4133 * \param[in] ma used to update \a pobj mtime and ctime
4135 * \retval 0 on success
4136 * \retval -errno on failure
4138 static int mdd_migrate_object(const struct lu_env *env,
4139 struct mdd_object *spobj,
4140 struct mdd_object *tpobj,
4141 struct mdd_object *sobj,
4142 struct mdd_object *tobj,
4143 const struct lu_name *sname,
4144 const struct lu_name *tname,
4145 struct md_op_spec *spec,
4148 struct mdd_thread_info *info = mdd_env_info(env);
4149 struct mdd_device *mdd = mdo2mdd(&spobj->mod_obj);
4150 struct lu_attr *spattr = &info->mti_pattr;
4151 struct lu_attr *tpattr = &info->mti_tpattr;
4152 struct lu_attr *attr = &info->mti_cattr;
4153 struct linkea_data *ldata = &info->mti_link_data;
4154 struct dt_allocation_hint *hint = &info->mti_hint;
4155 struct lu_buf sbuf = { NULL };
4156 struct lmv_mds_md_v1 *lmv;
4157 struct thandle *handle;
4162 rc = mdd_la_get(env, sobj, attr);
4166 rc = mdd_la_get(env, spobj, spattr);
4170 rc = mdd_la_get(env, tpobj, tpattr);
4174 if (S_ISDIR(attr->la_mode) && !spec->sp_migrate_nsonly) {
4175 struct lmv_user_md_v1 *lum = spec->u.sp_ea.eadata;
4179 /* if user use default value '0' for stripe_count, we need to
4180 * adjust it to '1' to create a 1-stripe directory.
4182 if (lum->lum_stripe_count == 0)
4183 lum->lum_stripe_count = cpu_to_le32(1);
4185 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4186 if (rc && rc != -ENODATA)
4191 if (!lmv_is_sane(lmv))
4192 GOTO(out, rc = -EBADF);
4193 if (lmv_is_migrating(lmv)) {
4194 rc = mdd_migrate_cmd_check(mdd, lmv, lum,
4199 } else if (!S_ISDIR(attr->la_mode)) {
4201 GOTO(out, rc = -EALREADY);
4203 /* update namespace only if @sobj is on MDT where @tpobj is. */
4204 if (!mdd_object_remote(tpobj) && !mdd_object_remote(sobj))
4205 spec->sp_migrate_nsonly = true;
4207 if (S_ISLNK(attr->la_mode)) {
4208 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4210 GOTO(out, rc = -ENOMEM);
4212 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4215 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4216 mdd2obd_dev(mdd)->obd_name,
4217 PFID(mdd_object_fid(sobj)), rc);
4223 /* linkea needs update upon FID or parent stripe change */
4224 rc = mdd_migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, sname,
4225 tname, attr, ldata);
4227 /* update namespace only if @sobj has link on its MDT. */
4228 spec->sp_migrate_nsonly = true;
4232 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4233 spattr, tpattr, attr);
4237 handle = mdd_trans_create(env, mdd);
4239 GOTO(out, rc = PTR_ERR(handle));
4241 if (spec->sp_migrate_nsonly)
4242 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, sname,
4243 tname, attr, spattr, tpattr,
4246 rc = mdd_declare_migrate_create(env, spobj, tpobj, sobj, tobj,
4247 sname, tname, spattr, tpattr,
4248 attr, &sbuf, ldata, ma, spec,
4253 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, tname, sname,
4258 rc = mdd_trans_start(env, mdd, handle);
4262 if (spec->sp_migrate_nsonly)
4263 rc = mdd_migrate_update(env, spobj, tpobj, sobj, sname, tname,
4264 attr, spattr, tpattr, ldata, ma,
4267 rc = mdd_migrate_create(env, spobj, tpobj, sobj, tobj, sname,
4268 tname, spattr, tpattr, attr, &sbuf,
4269 ldata, ma, spec, hint, handle);
4273 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0,
4274 spec->sp_migrate_nsonly ? sobj : tobj,
4275 mdd_object_fid(spobj), mdd_object_fid(sobj),
4276 mdd_object_fid(tpobj), tname, sname,
4283 rc = mdd_trans_stop(env, mdd, rc, handle);
4291 * Migrate directory or file between MDTs.
4293 * \param[in] env execution environment
4294 * \param[in] md_pobj parent master object
4295 * \param[in] md_sobj source object
4296 * \param[in] lname file name
4297 * \param[in] md_tobj target object
4298 * \param[in] spec target creation spec
4299 * \param[in] ma used to update \a pobj mtime and ctime
4301 * \retval 0 on success
4302 * \retval -errno on failure
4304 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4305 struct md_object *md_sobj, const struct lu_name *lname,
4306 struct md_object *md_tobj, struct md_op_spec *spec,
4309 struct mdd_thread_info *info = mdd_env_info(env);
4310 struct mdd_device *mdd = mdo2mdd(md_pobj);
4311 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4312 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4313 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4314 struct mdd_object *spobj = NULL;
4315 struct mdd_object *tpobj = NULL;
4316 struct lu_buf pbuf = { NULL };
4317 struct lu_fid *fid = &info->mti_fid2;
4318 struct lmv_mds_md_v1 *lmv;
4323 /* locate source and target stripe on pobj, which are the real parent */
4324 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4325 if (rc < 0 && rc != -ENODATA)
4332 if (!lmv_is_sane(lmv))
4333 GOTO(out, rc = -EBADF);
4335 /* locate target parent stripe */
4336 /* fail check here to make sure top dir migration succeed. */
4337 if (lmv_is_migrating(lmv) &&
4338 OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4339 GOTO(out, rc = -EIO);
4341 index = lmv_name_to_stripe_index(lmv, lname->ln_name,
4344 GOTO(out, rc = index);
4346 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4347 tpobj = mdd_object_find(env, mdd, fid);
4349 GOTO(out, rc = PTR_ERR(tpobj));
4351 /* locate source parent stripe */
4352 if (lmv_is_layout_changing(lmv)) {
4353 index = lmv_name_to_stripe_index_old(lmv,
4357 GOTO(out, rc = index);
4359 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4360 spobj = mdd_object_find(env, mdd, fid);
4362 GOTO(out, rc = PTR_ERR(spobj));
4364 /* parent stripe unchanged */
4365 if (spobj == tpobj) {
4366 if (!lmv_is_restriping(lmv))
4367 GOTO(out, rc = -EINVAL);
4368 GOTO(out, rc = -EALREADY);
4372 mdd_object_get(spobj);
4377 mdd_object_get(tpobj);
4378 mdd_object_get(spobj);
4381 rc = mdd_migrate_object(env, spobj, tpobj, sobj, tobj, lname, lname,
4386 if (!IS_ERR_OR_NULL(spobj))
4387 mdd_object_put(env, spobj);
4388 if (!IS_ERR_OR_NULL(tpobj))
4389 mdd_object_put(env, tpobj);
4395 static int mdd_declare_1sd_collapse(const struct lu_env *env,
4396 struct mdd_object *pobj,
4397 struct mdd_object *obj,
4398 struct mdd_object *stripe,
4399 struct lu_attr *attr,
4400 struct md_layout_change *mlc,
4401 struct lu_name *lname,
4402 struct thandle *handle)
4406 mlc->mlc_opc = MD_LAYOUT_DETACH;
4407 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4411 rc = mdo_declare_index_insert(env, stripe, mdd_object_fid(pobj),
4412 S_IFDIR, dotdot, handle);
4416 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle,
4417 mdo_declare_xattr_set);
4421 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4425 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4429 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4433 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(stripe),
4434 attr->la_mode, lname->ln_name, handle);
4438 rc = mdo_declare_ref_del(env, obj, handle);
4442 rc = mdo_declare_ref_del(env, obj, handle);
4446 rc = mdo_declare_destroy(env, obj, handle);
4453 /* transform one-stripe directory to a plain directory */
4454 static int mdd_1sd_collapse(const struct lu_env *env,
4455 struct mdd_object *pobj,
4456 struct mdd_object *obj,
4457 struct mdd_object *stripe,
4458 struct lu_attr *attr,
4459 struct md_layout_change *mlc,
4460 struct lu_name *lname,
4461 struct thandle *handle)
4467 /* replace 1-stripe directory with its stripe */
4468 mlc->mlc_opc = MD_LAYOUT_DETACH;
4470 mdd_write_lock(env, obj, DT_SRC_PARENT);
4471 rc = mdo_layout_change(env, obj, mlc, handle);
4472 mdd_write_unlock(env, obj);
4476 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4477 mdd_write_lock(env, obj, DT_SRC_CHILD);
4479 /* insert dotdot to stripe which points to parent */
4480 rc = __mdd_index_insert_only(env, stripe, mdd_object_fid(pobj),
4481 S_IFDIR, dotdot, handle);
4485 /* copy xattrs including linkea */
4486 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle, mdo_xattr_set);
4491 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4495 /* don't set nlink from parent */
4496 attr->la_valid &= ~LA_NLINK;
4498 rc = mdo_attr_set(env, stripe, attr, handle);
4502 /* delete dir name from parent */
4503 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4507 /* insert stripe to parent with dir name */
4508 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(stripe),
4509 attr->la_mode, lname->ln_name, handle);
4513 /* destroy dir obj */
4514 rc = mdo_ref_del(env, obj, handle);
4518 rc = mdo_ref_del(env, obj, handle);
4522 rc = mdo_destroy(env, obj, handle);
4528 mdd_write_unlock(env, obj);
4529 mdd_write_unlock(env, pobj);
4535 * shrink directory stripes after migration/merge
4537 int mdd_dir_layout_shrink(const struct lu_env *env,
4538 struct md_object *md_obj,
4539 struct md_layout_change *mlc)
4541 struct mdd_device *mdd = mdo2mdd(md_obj);
4542 struct mdd_thread_info *info = mdd_env_info(env);
4543 struct mdd_object *obj = md2mdd_obj(md_obj);
4544 struct mdd_object *pobj = NULL;
4545 struct mdd_object *stripe = NULL;
4546 struct lu_attr *attr = &info->mti_pattr;
4547 struct lu_fid *fid = &info->mti_fid2;
4548 struct lu_name lname = { NULL };
4549 struct lu_buf lmv_buf = { NULL };
4550 struct lmv_mds_md_v1 *lmv;
4551 struct lmv_user_md *lmu;
4552 struct thandle *handle;
4557 rc = mdd_la_get(env, obj, attr);
4561 if (!S_ISDIR(attr->la_mode))
4564 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4568 lmv = lmv_buf.lb_buf;
4569 if (!lmv_is_sane(lmv))
4572 lmu = mlc->mlc_buf.lb_buf;
4574 /* adjust the default value '0' to '1' */
4575 if (lmu->lum_stripe_count == 0)
4576 lmu->lum_stripe_count = cpu_to_le32(1);
4578 /* these were checked in MDT */
4579 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4580 le32_to_cpu(lmv->lmv_stripe_count));
4581 LASSERT(!lmv_is_splitting(lmv));
4582 LASSERT(lmv_is_migrating(lmv) || lmv_is_merging(lmv));
4584 /* if dir stripe count will be shrunk to 1, it needs to be transformed
4585 * to a plain dir, which will cause FID change and namespace update.
4587 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4588 struct linkea_data *ldata = &info->mti_link_data;
4589 char *filename = info->mti_name;
4591 rc = mdd_links_read(env, obj, ldata);
4595 if (ldata->ld_leh->leh_reccount > 1)
4596 GOTO(out, rc = -EINVAL);
4598 linkea_first_entry(ldata);
4600 GOTO(out, rc = -ENODATA);
4602 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4605 /* Note: lname might miss \0 at the end */
4606 snprintf(filename, sizeof(info->mti_name), "%.*s",
4607 lname.ln_namelen, lname.ln_name);
4608 lname.ln_name = filename;
4610 pobj = mdd_object_find(env, mdd, fid);
4617 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4619 stripe = mdd_object_find(env, mdd, fid);
4620 if (IS_ERR(stripe)) {
4621 mdd_object_put(env, pobj);
4623 GOTO(out, rc = PTR_ERR(stripe));
4627 handle = mdd_trans_create(env, mdd);
4629 GOTO(out, rc = PTR_ERR(handle));
4631 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4632 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4634 GOTO(stop_trans, rc);
4636 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4637 rc = mdd_declare_1sd_collapse(env, pobj, obj, stripe, attr, mlc,
4640 GOTO(stop_trans, rc);
4643 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4646 GOTO(stop_trans, rc);
4648 rc = mdd_trans_start(env, mdd, handle);
4650 GOTO(stop_trans, rc);
4652 mdd_write_lock(env, obj, DT_SRC_PARENT);
4653 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4654 rc = mdo_layout_change(env, obj, mlc, handle);
4655 mdd_write_unlock(env, obj);
4657 GOTO(stop_trans, rc);
4659 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4660 rc = mdd_1sd_collapse(env, pobj, obj, stripe, attr, mlc, &lname,
4663 GOTO(stop_trans, rc);
4666 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4667 XATTR_NAME_LMV, handle);
4668 GOTO(stop_trans, rc);
4671 rc = mdd_trans_stop(env, mdd, rc, handle);
4674 mdd_object_put(env, stripe);
4675 mdd_object_put(env, pobj);
4677 lu_buf_free(&lmv_buf);
4681 static int mdd_dir_declare_split_plain(const struct lu_env *env,
4682 struct mdd_device *mdd,
4683 struct mdd_object *pobj,
4684 struct mdd_object *obj,
4685 struct mdd_object *tobj,
4686 struct md_layout_change *mlc,
4687 struct dt_allocation_hint *hint,
4688 struct thandle *handle)
4690 struct mdd_thread_info *info = mdd_env_info(env);
4691 const struct lu_name *lname = mlc->mlc_name;
4692 struct lu_attr *la = &info->mti_la_for_fix;
4693 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
4694 struct linkea_data *ldata = &info->mti_link_data;
4695 struct lmv_mds_md_v1 *lmv;
4699 mlc->mlc_opc = MD_LAYOUT_DETACH;
4700 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4704 memset(ldata, 0, sizeof(*ldata));
4705 rc = mdd_linkea_prepare(env, obj, NULL, NULL, mdd_object_fid(pobj),
4706 lname, 1, 0, ldata);
4710 count = lum->lum_stripe_count;
4711 lum->lum_stripe_count = 0;
4712 mdd_object_make_hint(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4714 rc = mdd_declare_create(env, mdo2mdd(&pobj->mod_obj), pobj, tobj,
4715 lname, mlc->mlc_attr, handle, mlc->mlc_spec,
4716 ldata, NULL, NULL, NULL, hint);
4720 /* tobj mode will be used in lod_declare_xattr_set(), but it's not
4723 tobj->mod_obj.mo_lu.lo_header->loh_attr |= S_IFDIR;
4725 lmv = (typeof(lmv))info->mti_key;
4726 memset(lmv, 0, sizeof(*lmv));
4727 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
4728 lmv->lmv_stripe_count = cpu_to_le32(1);
4729 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
4730 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdd_object_fid(obj));
4732 mlc->mlc_opc = MD_LAYOUT_ATTACH;
4733 mlc->mlc_buf.lb_buf = lmv;
4734 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
4735 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4739 rc = mdd_iterate_xattrs(env, obj, tobj, true, handle,
4740 mdo_declare_xattr_set);
4744 lum->lum_stripe_count = count;
4745 mlc->mlc_opc = MD_LAYOUT_SPLIT;
4746 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4750 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4754 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
4755 S_IFDIR, lname->ln_name, handle);
4759 la->la_valid = LA_CTIME | LA_MTIME;
4760 rc = mdo_declare_attr_set(env, obj, la, handle);
4764 rc = mdo_declare_attr_set(env, pobj, la, handle);
4768 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4774 * plain directory split:
4775 * 1. create \a tobj as plain directory.
4776 * 2. append \a obj as first stripe of \a tobj.
4777 * 3. migrate xattrs from \a obj to \a tobj.
4778 * 4. split \a tobj to specific stripe count.
4780 static int mdd_dir_split_plain(const struct lu_env *env,
4781 struct mdd_device *mdd,
4782 struct mdd_object *pobj,
4783 struct mdd_object *obj,
4784 struct mdd_object *tobj,
4785 struct md_layout_change *mlc,
4786 struct dt_allocation_hint *hint,
4787 struct thandle *handle)
4789 struct mdd_thread_info *info = mdd_env_info(env);
4790 struct lu_attr *pattr = &info->mti_pattr;
4791 struct lu_attr *la = &info->mti_la_for_fix;
4792 const struct lu_name *lname = mlc->mlc_name;
4793 struct linkea_data *ldata = &info->mti_link_data;
4798 /* copy linkea out and set on target later */
4799 rc = mdd_links_read(env, obj, ldata);
4803 mlc->mlc_opc = MD_LAYOUT_DETACH;
4804 rc = mdo_layout_change(env, obj, mlc, handle);
4808 /* don't set nlink from obj */
4809 mlc->mlc_attr->la_valid &= ~LA_NLINK;
4811 rc = mdd_create_object(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4812 NULL, NULL, NULL, hint, handle, false);
4816 rc = mdd_iterate_xattrs(env, obj, tobj, true, handle, mdo_xattr_set);
4820 rc = mdd_links_write(env, tobj, ldata, handle);
4824 rc = __mdd_index_delete(env, pobj, lname->ln_name, true, handle);
4828 rc = __mdd_index_insert(env, pobj, mdd_object_fid(tobj), S_IFDIR,
4829 lname->ln_name, handle);
4833 la->la_ctime = la->la_mtime = mlc->mlc_attr->la_mtime;
4834 la->la_valid = LA_CTIME | LA_MTIME;
4836 mdd_write_lock(env, obj, DT_SRC_CHILD);
4837 rc = mdd_update_time(env, tobj, mlc->mlc_attr, la, handle);
4838 mdd_write_unlock(env, obj);
4842 rc = mdd_la_get(env, pobj, pattr);
4846 la->la_valid = LA_CTIME | LA_MTIME;
4848 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4849 rc = mdd_update_time(env, pobj, pattr, la, handle);
4850 mdd_write_unlock(env, pobj);
4854 /* FID changes, record it as CL_MIGRATE */
4855 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
4856 mdd_object_fid(pobj), mdd_object_fid(obj),
4857 mdd_object_fid(pobj), lname, lname, handle);
4861 int mdd_dir_layout_split(const struct lu_env *env, struct md_object *o,
4862 struct md_layout_change *mlc)
4864 struct mdd_thread_info *info = mdd_env_info(env);
4865 struct mdd_device *mdd = mdo2mdd(o);
4866 struct mdd_object *obj = md2mdd_obj(o);
4867 struct mdd_object *pobj = md2mdd_obj(mlc->mlc_parent);
4868 struct mdd_object *tobj = md2mdd_obj(mlc->mlc_target);
4869 struct dt_allocation_hint *hint = &info->mti_hint;
4870 bool is_plain = false;
4871 struct thandle *handle;
4876 LASSERT(S_ISDIR(mdd_object_type(obj)));
4878 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LMV);
4884 handle = mdd_trans_create(env, mdd);
4886 RETURN(PTR_ERR(handle));
4889 rc = mdd_dir_declare_split_plain(env, mdd, pobj, obj, tobj, mlc,
4892 mlc->mlc_opc = MD_LAYOUT_SPLIT;
4893 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4895 GOTO(stop_trans, rc);
4897 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL,
4901 GOTO(stop_trans, rc);
4903 rc = mdd_trans_start(env, mdd, handle);
4905 GOTO(stop_trans, rc);
4908 rc = mdd_dir_split_plain(env, mdd, pobj, obj, tobj, mlc, hint,
4911 mdd_write_lock(env, obj, DT_TGT_CHILD);
4912 rc = mdo_xattr_set(env, obj, NULL, XATTR_NAME_LMV,
4913 LU_XATTR_CREATE, handle);
4914 mdd_write_unlock(env, obj);
4916 GOTO(stop_trans, rc);
4918 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4919 XATTR_NAME_LMV, handle);
4922 GOTO(stop_trans, rc);
4927 rc = mdd_trans_stop(env, mdd, rc, handle);
4932 const struct md_dir_operations mdd_dir_ops = {
4933 .mdo_is_subdir = mdd_is_subdir,
4934 .mdo_lookup = mdd_lookup,
4935 .mdo_create = mdd_create,
4936 .mdo_rename = mdd_rename,
4937 .mdo_link = mdd_link,
4938 .mdo_unlink = mdd_unlink,
4939 .mdo_create_data = mdd_create_data,
4940 .mdo_migrate = mdd_migrate,