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_rec_hdr rec_hdr;
755 struct thandle *llog_th;
758 if (!mdd_changelog_enabled(env, mdd, type))
761 rec_hdr.lrh_len = mdd_llog_record_calc_size(env, tname, sname);
762 rec_hdr.lrh_type = CHANGELOG_REC;
764 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
768 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
770 GOTO(out_put, rc = PTR_ERR(llog_th));
772 rc = llog_declare_add(env, ctxt->loc_handle, &rec_hdr, llog_th);
780 int mdd_changelog_write_rec(const struct lu_env *env,
781 struct llog_handle *loghandle,
782 struct llog_rec_hdr *r,
783 struct llog_cookie *cookie,
784 int idx, struct thandle *th)
788 if (r->lrh_type == CHANGELOG_REC) {
789 struct mdd_device *mdd;
790 struct llog_changelog_rec *rec;
792 mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
793 rec = container_of(r, struct llog_changelog_rec, cr_hdr);
795 spin_lock(&mdd->mdd_cl.mc_lock);
796 rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
797 spin_unlock(&mdd->mdd_cl.mc_lock);
799 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
803 * if current llog is full, we will generate a new
804 * llog, and since it's actually not an error, let's
805 * avoid increasing index so that userspace apps
806 * should not see a gap in the changelog sequence
808 if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
809 spin_lock(&mdd->mdd_cl.mc_lock);
810 ++mdd->mdd_cl.mc_index;
811 spin_unlock(&mdd->mdd_cl.mc_lock);
814 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
821 /** Add a changelog entry \a rec to the changelog llog
824 * \param handle - currently ignored since llogs start their own transaction;
825 * this will hopefully be fixed in llog rewrite
828 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
829 struct llog_changelog_rec *rec, struct thandle *th)
831 struct obd_device *obd = mdd2obd_dev(mdd);
832 struct llog_ctxt *ctxt;
833 struct thandle *llog_th;
836 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
837 changelog_rec_varsize(&rec->cr));
839 /* llog_lvfs_write_rec sets the llog tail len */
840 rec->cr_hdr.lrh_type = CHANGELOG_REC;
841 rec->cr.cr_time = cl_time();
843 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
847 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
849 GOTO(out_put, rc = PTR_ERR(llog_th));
851 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
852 /* nested journal transaction */
853 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
855 /* time to recover some space ?? */
856 if (likely(!mdd->mdd_changelog_gc ||
857 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
858 mdd->mdd_changelog_min_gc_interval >=
859 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
860 /* save a spin_lock trip */
862 spin_lock(&mdd->mdd_cl.mc_lock);
863 if (likely(mdd->mdd_changelog_gc &&
864 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
865 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
866 mdd->mdd_changelog_min_gc_interval)) {
867 if (unlikely(llog_cat_free_space(ctxt->loc_handle) <=
868 mdd->mdd_changelog_min_free_cat_entries ||
869 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD))) {
870 CWARN("%s:%s low on changelog_catalog free entries, "
871 "starting ChangeLog garbage collection thread\n",
873 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
876 /* indicate further kthread run will occur outside
877 * right after current journal transaction filling has
880 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
882 /* next check in mdd_changelog_min_gc_interval anyway
884 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
886 spin_unlock(&mdd->mdd_cl.mc_lock);
894 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
895 const struct lu_fid *sfid,
896 const struct lu_fid *spfid,
897 const struct lu_name *sname)
899 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
902 LASSERT(sfid != NULL);
903 LASSERT(spfid != NULL);
904 LASSERT(sname != NULL);
906 extsize = sname->ln_namelen + 1;
908 rnm->cr_sfid = *sfid;
909 rnm->cr_spfid = *spfid;
911 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
912 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
913 rec->cr_namelen += extsize;
916 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
918 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
920 if (jobid == NULL || jobid[0] == '\0')
923 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
926 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
928 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
930 ef->cr_extra_flags = eflags;
933 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
934 __u64 uid, __u64 gid)
936 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
938 uidgid->cr_uid = uid;
939 uidgid->cr_gid = gid;
942 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
945 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
950 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
952 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
954 omd->cr_openflags = flags;
957 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
958 const char *xattr_name)
960 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
962 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
965 /** Store a namespace change changelog record
966 * If this fails, we must fail the whole transaction; we don't
967 * want the change to commit without the log entry.
968 * \param target - mdd_object of change
969 * \param tpfid - target parent dir/object fid
970 * \param sfid - source object fid
971 * \param spfid - source parent fid
972 * \param tname - target name string
973 * \param sname - source name string
974 * \param handle - transaction handle
976 int mdd_changelog_ns_store(const struct lu_env *env,
977 struct mdd_device *mdd,
978 enum changelog_rec_type type,
979 enum changelog_rec_flags clf_flags,
980 struct mdd_object *target,
981 const struct lu_fid *tpfid,
982 const struct lu_fid *sfid,
983 const struct lu_fid *spfid,
984 const struct lu_name *tname,
985 const struct lu_name *sname,
986 struct thandle *handle)
988 const struct lu_ucred *uc = lu_ucred(env);
989 struct llog_changelog_rec *rec;
992 __u64 xflags = CLFE_INVALID;
996 if (!mdd_changelog_enabled(env, mdd, type))
999 LASSERT(tpfid != NULL);
1000 LASSERT(tname != NULL);
1001 LASSERT(handle != NULL);
1003 reclen = mdd_llog_record_calc_size(env, tname, sname);
1004 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_chlg_buf, reclen);
1005 if (buf->lb_buf == NULL)
1009 clf_flags &= CLF_FLAGMASK;
1010 clf_flags |= CLF_EXTRA_FLAGS;
1013 if (uc->uc_jobid[0] != '\0')
1014 clf_flags |= CLF_JOBID;
1015 xflags |= CLFE_UIDGID;
1020 clf_flags |= CLF_RENAME;
1022 clf_flags |= CLF_VERSION;
1024 rec->cr.cr_flags = clf_flags;
1026 if (clf_flags & CLF_EXTRA_FLAGS) {
1027 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1028 if (xflags & CLFE_UIDGID)
1029 mdd_changelog_rec_extra_uidgid(&rec->cr,
1030 uc->uc_uid, uc->uc_gid);
1031 if (xflags & CLFE_NID)
1032 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1035 rec->cr.cr_type = (__u32)type;
1036 rec->cr.cr_pfid = *tpfid;
1037 rec->cr.cr_namelen = tname->ln_namelen;
1038 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1040 if (clf_flags & CLF_RENAME)
1041 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1043 if (clf_flags & CLF_JOBID)
1044 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1046 if (likely(target != NULL)) {
1047 rec->cr.cr_tfid = *mdd_object_fid(target);
1048 target->mod_cltime = ktime_get();
1050 fid_zero(&rec->cr.cr_tfid);
1053 rc = mdd_changelog_store(env, mdd, rec, handle);
1055 CERROR("%s: cannot store changelog record: type = %d, "
1056 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1057 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1058 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1065 static int __mdd_links_add(const struct lu_env *env,
1066 struct mdd_object *mdd_obj,
1067 struct linkea_data *ldata,
1068 const struct lu_name *lname,
1069 const struct lu_fid *pfid,
1070 int first, int check)
1072 /* cattr is set in mdd_link */
1073 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1076 if (ldata->ld_leh == NULL) {
1077 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1081 rc = linkea_data_new(ldata,
1082 &mdd_env_info(env)->mti_link_buf);
1089 rc = linkea_links_find(ldata, lname, pfid);
1090 if (rc && rc != -ENOENT)
1096 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1097 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1101 linkea_add_buf(ldata, lname, tfid, false);
1104 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1105 linkea_add_buf(ldata, lname, pfid, false);
1107 /* For encrypted file, we want to limit number of hard links to what
1108 * linkEA can contain. So ask to return error in case of overflow.
1109 * Currently linkEA stores 4KiB of links, that is 14 NAME_MAX links,
1110 * or 119 16-byte names.
1112 return linkea_add_buf(ldata, lname, pfid,
1113 cattr->la_valid & LA_FLAGS &&
1114 cattr->la_flags & LUSTRE_ENCRYPT_FL);
1117 static int __mdd_links_del(const struct lu_env *env,
1118 struct mdd_object *mdd_obj,
1119 struct linkea_data *ldata,
1120 const struct lu_name *lname,
1121 const struct lu_fid *pfid)
1123 /* cattr is set in mdd_link */
1124 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1127 if (ldata->ld_leh == NULL) {
1128 rc = mdd_links_read(env, mdd_obj, ldata);
1133 rc = linkea_links_find(ldata, lname, pfid);
1137 linkea_del_buf(ldata, lname,
1138 cattr->la_valid & LA_FLAGS &&
1139 cattr->la_flags & LUSTRE_ENCRYPT_FL);
1143 static int mdd_linkea_prepare(const struct lu_env *env,
1144 struct mdd_object *mdd_obj,
1145 const struct lu_fid *oldpfid,
1146 const struct lu_name *oldlname,
1147 const struct lu_fid *newpfid,
1148 const struct lu_name *newlname,
1149 int first, int check,
1150 struct linkea_data *ldata)
1155 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1158 LASSERT(oldpfid != NULL || newpfid != NULL);
1160 if (mdd_obj->mod_flags & DEAD_OBJ)
1161 /* Unnecessary to update linkEA for dead object. */
1164 if (oldpfid != NULL) {
1165 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1167 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1170 /* No changes done. */
1175 /* If renaming, add the new record */
1176 if (newpfid != NULL)
1177 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1183 int mdd_links_rename(const struct lu_env *env,
1184 struct mdd_object *mdd_obj,
1185 const struct lu_fid *oldpfid,
1186 const struct lu_name *oldlname,
1187 const struct lu_fid *newpfid,
1188 const struct lu_name *newlname,
1189 struct thandle *handle,
1190 struct linkea_data *ldata,
1191 int first, int check)
1196 if (ldata == NULL) {
1197 ldata = &mdd_env_info(env)->mti_link_data;
1198 memset(ldata, 0, sizeof(*ldata));
1199 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1200 newpfid, newlname, first, check, ldata);
1205 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1206 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1212 if (newlname == NULL)
1213 CERROR("link_ea add failed %d "DFID"\n",
1214 rc, PFID(mdd_object_fid(mdd_obj)));
1215 else if (oldpfid == NULL)
1216 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1217 newlname->ln_namelen, newlname->ln_name, rc,
1218 PFID(mdd_object_fid(mdd_obj)));
1219 else if (newpfid == NULL)
1220 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1221 oldlname->ln_namelen, oldlname->ln_name, rc,
1222 PFID(mdd_object_fid(mdd_obj)));
1224 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1225 "\n", oldlname->ln_namelen, oldlname->ln_name,
1226 newlname->ln_namelen, newlname->ln_name, rc,
1227 PFID(mdd_object_fid(mdd_obj)));
1230 if (is_vmalloc_addr(ldata->ld_buf))
1231 /* if we vmalloced a large buffer drop it */
1232 lu_buf_free(ldata->ld_buf);
1237 static inline int mdd_links_add(const struct lu_env *env,
1238 struct mdd_object *mdd_obj,
1239 const struct lu_fid *pfid,
1240 const struct lu_name *lname,
1241 struct thandle *handle,
1242 struct linkea_data *ldata, int first)
1244 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1245 pfid, lname, handle, ldata, first, 0);
1248 static inline int mdd_links_del(const struct lu_env *env,
1249 struct mdd_object *mdd_obj,
1250 const struct lu_fid *pfid,
1251 const struct lu_name *lname,
1252 struct thandle *handle)
1254 return mdd_links_rename(env, mdd_obj, pfid, lname,
1255 NULL, NULL, handle, NULL, 0, 0);
1258 /** Read the link EA into a temp buffer.
1259 * Uses the name_buf since it is generally large.
1260 * \retval IS_ERR err
1261 * \retval ptr to \a lu_buf (always \a mti_link_buf)
1263 struct lu_buf *mdd_links_get(const struct lu_env *env,
1264 struct mdd_object *mdd_obj)
1266 struct linkea_data ldata = { NULL };
1269 rc = mdd_links_read(env, mdd_obj, &ldata);
1270 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1273 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1274 struct linkea_data *ldata, struct thandle *handle)
1276 const struct lu_buf *buf;
1279 if (ldata == NULL || ldata->ld_buf == NULL ||
1280 ldata->ld_leh == NULL)
1283 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1287 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1288 ldata->ld_leh->leh_len);
1289 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1290 if (unlikely(rc == -ENOSPC)) {
1291 rc = linkea_overflow_shrink(ldata);
1299 static int mdd_declare_links_add(const struct lu_env *env,
1300 struct mdd_object *mdd_obj,
1301 struct thandle *handle,
1302 struct linkea_data *ldata)
1308 if (ldata != NULL && ldata->ld_leh != NULL) {
1309 ea_len = ldata->ld_leh->leh_len;
1310 linkea = ldata->ld_buf->lb_buf;
1312 ea_len = MAX_LINKEA_SIZE;
1316 rc = mdo_declare_xattr_set(env, mdd_obj,
1317 mdd_buf_get_const(env, linkea, ea_len),
1318 XATTR_NAME_LINK, 0, handle);
1323 static inline int mdd_declare_links_del(const struct lu_env *env,
1324 struct mdd_object *c,
1325 struct thandle *handle)
1329 /* For directory, the linkEA will be removed together
1330 * with the object. */
1331 if (!S_ISDIR(mdd_object_type(c)))
1332 rc = mdd_declare_links_add(env, c, handle, NULL);
1337 static int mdd_declare_link(const struct lu_env *env,
1338 struct mdd_device *mdd,
1339 struct mdd_object *p,
1340 struct mdd_object *c,
1341 const struct lu_name *name,
1342 struct thandle *handle,
1344 struct linkea_data *data)
1346 struct lu_fid tfid = *mdd_object_fid(c);
1349 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1350 tfid.f_oid = cfs_fail_val;
1352 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1353 name->ln_name, handle);
1357 rc = mdo_declare_ref_add(env, c, handle);
1361 la->la_valid = LA_CTIME | LA_MTIME;
1362 rc = mdo_declare_attr_set(env, p, la, handle);
1366 la->la_valid = LA_CTIME;
1367 rc = mdo_declare_attr_set(env, c, la, handle);
1371 rc = mdd_declare_links_add(env, c, handle, data);
1375 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1381 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1382 struct md_object *src_obj, const struct lu_name *lname,
1385 const char *name = lname->ln_name;
1386 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1387 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1388 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1389 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1390 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1391 struct mdd_device *mdd = mdo2mdd(src_obj);
1392 struct thandle *handle;
1393 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1394 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1398 rc = mdd_la_get(env, mdd_sobj, cattr);
1402 rc = mdd_la_get(env, mdd_tobj, tattr);
1407 * If we are using project inheritance, we only allow hard link
1408 * creation in our tree when the project IDs are the same;
1409 * otherwise the tree quota mechanism could be circumvented.
1411 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1412 (tattr->la_projid != cattr->la_projid))
1415 handle = mdd_trans_create(env, mdd);
1417 GOTO(out_pending, rc = PTR_ERR(handle));
1419 memset(ldata, 0, sizeof(*ldata));
1421 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1422 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1424 /* Note: even this function will change ldata, but it comes from
1425 * thread_info, which is completely temporary and only seen in
1426 * this function, so we do not need reset ldata once it fails.*/
1427 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1428 mdd_object_fid(mdd_tobj), lname, 0, 0, ldata);
1432 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1437 rc = mdd_trans_start(env, mdd, handle);
1441 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1442 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1445 GOTO(out_unlock, rc);
1447 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1448 rc = mdo_ref_add(env, mdd_sobj, handle);
1450 GOTO(out_unlock, rc);
1453 *tfid = *mdd_object_fid(mdd_sobj);
1454 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1455 tfid->f_oid = cfs_fail_val;
1457 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1458 mdd_object_type(mdd_sobj), name, handle);
1460 mdo_ref_del(env, mdd_sobj, handle);
1461 GOTO(out_unlock, rc);
1464 la->la_valid = LA_CTIME | LA_MTIME;
1465 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1467 GOTO(out_unlock, rc);
1469 la->la_valid = LA_CTIME;
1470 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1472 /* Note: The failure of links_add should not cause the
1473 * link failure, so do not check return value. */
1474 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_tobj),
1475 lname, handle, ldata, 0);
1479 mdd_write_unlock(env, mdd_sobj);
1481 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1482 mdd_object_fid(mdd_tobj), NULL,
1483 NULL, lname, NULL, handle);
1485 rc = mdd_trans_stop(env, mdd, rc, handle);
1486 if (is_vmalloc_addr(ldata->ld_buf))
1487 /* if we vmalloced a large buffer drop it */
1488 lu_buf_free(ldata->ld_buf);
1493 static int mdd_mark_orphan_object(const struct lu_env *env,
1494 struct mdd_object *obj, struct thandle *handle,
1497 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1500 attr->la_valid = LA_FLAGS;
1501 attr->la_flags = LUSTRE_ORPHAN_FL;
1504 rc = mdo_declare_attr_set(env, obj, attr, handle);
1506 rc = mdo_attr_set(env, obj, attr, handle);
1511 static int mdd_declare_finish_unlink(const struct lu_env *env,
1512 struct mdd_object *obj,
1513 struct thandle *handle)
1517 /* Sigh, we do not know if the unlink object will become orphan in
1518 * declare phase, but fortunately the flags here does not matter
1519 * in current declare implementation */
1520 rc = mdd_mark_orphan_object(env, obj, handle, true);
1524 rc = mdo_declare_destroy(env, obj, handle);
1528 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1532 return mdd_declare_links_del(env, obj, handle);
1535 /* caller should take a lock before calling */
1536 int mdd_finish_unlink(const struct lu_env *env,
1537 struct mdd_object *obj, struct md_attr *ma,
1538 struct mdd_object *pobj,
1539 const struct lu_name *lname,
1543 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1546 LASSERT(mdd_write_locked(env, obj) != 0);
1548 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1549 /* add new orphan and the object
1550 * will be deleted during mdd_close() */
1551 obj->mod_flags |= DEAD_OBJ;
1552 if (obj->mod_count) {
1553 rc = mdd_orphan_insert(env, obj, th);
1555 CDEBUG(D_HA, "Object "DFID" is inserted into "
1556 "orphan list, open count = %d\n",
1557 PFID(mdd_object_fid(obj)),
1560 CERROR("Object "DFID" fail to be an orphan, "
1561 "open count = %d, maybe cause failed "
1563 PFID(mdd_object_fid(obj)),
1566 /* mark object as an orphan here, not
1567 * before mdd_orphan_insert() as racing
1568 * mdd_la_get() may propagate ORPHAN_OBJ
1569 * causing the asserition */
1570 rc = mdd_mark_orphan_object(env, obj, th, false);
1572 rc = mdo_destroy(env, obj, th);
1574 } else if (!is_dir) {
1575 /* old files may not have link ea; ignore errors */
1576 mdd_links_del(env, obj, mdd_object_fid(pobj), lname, th);
1584 * has mdd_write_lock on cobj already, but not on pobj yet
1586 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1587 const struct lu_attr *pattr,
1588 struct mdd_object *cobj,
1589 const struct lu_attr *cattr)
1594 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1599 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1600 struct mdd_object *p, struct mdd_object *c,
1601 const struct lu_name *name, struct md_attr *ma,
1602 struct thandle *handle, int no_name, int is_dir)
1604 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1607 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1608 if (likely(no_name == 0)) {
1609 rc = mdo_declare_index_delete(env, p, name->ln_name,
1616 rc = mdo_declare_ref_del(env, p, handle);
1622 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1623 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1624 la->la_valid = LA_CTIME | LA_MTIME;
1625 rc = mdo_declare_attr_set(env, p, la, handle);
1630 rc = mdo_declare_ref_del(env, c, handle);
1634 rc = mdo_declare_ref_del(env, c, handle);
1638 la->la_valid = LA_CTIME;
1639 rc = mdo_declare_attr_set(env, c, la, handle);
1643 rc = mdd_declare_finish_unlink(env, c, handle);
1647 /* FIXME: need changelog for remove entry */
1648 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1656 * test if a file has an HSM archive
1657 * if HSM attributes are not found in ma update them from
1660 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1661 struct mdd_object *obj,
1666 if (!(ma->ma_valid & MA_HSM)) {
1667 /* no HSM MD provided, read xattr */
1668 struct lu_buf *hsm_buf;
1669 const size_t buflen = sizeof(struct hsm_attrs);
1672 hsm_buf = mdd_buf_get(env, NULL, 0);
1673 lu_buf_alloc(hsm_buf, buflen);
1674 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1675 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1676 lu_buf_free(hsm_buf);
1680 ma->ma_valid |= MA_HSM;
1682 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1688 * Delete name entry and the object.
1689 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1690 * does not exist for this object, and it could only happen during resending
1691 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1692 * is also needed in this case(needed by changelog), so we have to add another
1693 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1694 * the ENOENT failure should be able to be fixed by redo mechanism.
1696 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1697 struct md_object *cobj, const struct lu_name *lname,
1698 struct md_attr *ma, int no_name)
1700 const char *name = lname->ln_name;
1701 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1702 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1703 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1704 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1705 struct mdd_object *mdd_cobj = NULL;
1706 struct mdd_device *mdd = mdo2mdd(pobj);
1707 struct thandle *handle;
1708 int rc, is_dir = 0, cl_flags = 0;
1711 /* let shutdown to start */
1712 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1714 /* cobj == NULL means only delete name entry */
1715 if (likely(cobj != NULL)) {
1716 mdd_cobj = md2mdd_obj(cobj);
1717 if (mdd_object_exists(mdd_cobj) == 0)
1721 rc = mdd_la_get(env, mdd_pobj, pattr);
1725 if (likely(mdd_cobj != NULL)) {
1727 rc = mdd_la_get(env, mdd_cobj, cattr);
1731 is_dir = S_ISDIR(cattr->la_mode);
1732 /* search for an existing archive.
1733 * we should check ahead as the object
1734 * can be destroyed in this transaction */
1735 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1736 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1739 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1743 handle = mdd_trans_create(env, mdd);
1745 RETURN(PTR_ERR(handle));
1747 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1748 lname, ma, handle, no_name, is_dir);
1752 rc = mdd_trans_start(env, mdd, handle);
1756 if (likely(mdd_cobj != NULL))
1757 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1759 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1760 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1765 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1766 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1767 GOTO(cleanup, rc = 0);
1769 if (likely(mdd_cobj != NULL)) {
1770 rc = mdo_ref_del(env, mdd_cobj, handle);
1772 __mdd_index_insert_only(env, mdd_pobj,
1773 mdd_object_fid(mdd_cobj),
1774 mdd_object_type(mdd_cobj),
1781 mdo_ref_del(env, mdd_cobj, handle);
1783 /* fetch updated nlink */
1784 rc = mdd_la_get(env, mdd_cobj, cattr);
1789 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1790 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1792 la->la_valid = LA_CTIME | LA_MTIME;
1793 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1797 /* Enough for only unlink the entry */
1798 if (unlikely(mdd_cobj == NULL))
1801 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1802 /* update ctime of an unlinked file only if it is still
1803 * opened or a link still exists */
1804 la->la_valid = LA_CTIME;
1805 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1810 /* XXX: this transfer to ma will be removed with LOD/OSP */
1811 ma->ma_attr = *cattr;
1812 ma->ma_valid |= MA_INODE;
1813 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1817 /* fetch updated nlink */
1818 rc = mdd_la_get(env, mdd_cobj, cattr);
1819 /* if object is removed then we can't get its attrs,
1821 if (rc == -ENOENT) {
1822 cattr->la_nlink = 0;
1826 if (cattr->la_nlink == 0) {
1827 ma->ma_attr = *cattr;
1828 ma->ma_valid |= MA_INODE;
1833 if (likely(mdd_cobj != NULL))
1834 mdd_write_unlock(env, mdd_cobj);
1837 if (cattr->la_nlink == 0)
1838 cl_flags |= CLF_UNLINK_LAST;
1840 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1842 rc = mdd_changelog_ns_store(env, mdd,
1843 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1844 mdd_cobj, mdd_object_fid(mdd_pobj), NULL, NULL,
1845 lname, NULL, handle);
1849 rc = mdd_trans_stop(env, mdd, rc, handle);
1855 * The permission has been checked when obj created, no need check again.
1857 static int mdd_cd_sanity_check(const struct lu_env *env,
1858 struct mdd_object *obj)
1863 if (!obj || mdd_is_dead_obj(obj))
1869 static int mdd_create_data(const struct lu_env *env,
1870 struct md_object *pobj,
1871 struct md_object *cobj,
1872 const struct md_op_spec *spec,
1875 struct mdd_device *mdd = mdo2mdd(cobj);
1876 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1877 struct mdd_object *son = md2mdd_obj(cobj);
1878 struct thandle *handle;
1879 const struct lu_buf *buf;
1880 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1881 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1885 rc = mdd_cd_sanity_check(env, son);
1889 if (!md_should_create(spec->sp_cr_flags))
1893 * there are following use cases for this function:
1894 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1895 * striping can be specified or not
1898 rc = mdd_la_get(env, son, attr);
1902 /* calling ->ah_make_hint() is used to transfer information from parent */
1903 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1905 handle = mdd_trans_create(env, mdd);
1907 GOTO(out_free, rc = PTR_ERR(handle));
1910 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1911 * Should this be fixed?
1913 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1914 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1915 spec->sp_cr_flags, spec->no_create);
1917 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1918 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1919 spec->u.sp_ea.eadatalen);
1924 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1925 XATTR_NAME_LOV, 0, handle);
1929 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1934 rc = mdd_trans_start(env, mdd, handle);
1938 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1944 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle,
1948 rc = mdd_trans_stop(env, mdd, rc, handle);
1954 static int mdd_declare_object_initialize(const struct lu_env *env,
1955 struct mdd_object *parent,
1956 struct mdd_object *child,
1957 const struct lu_attr *attr,
1958 struct thandle *handle)
1963 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1964 if (!S_ISDIR(attr->la_mode))
1967 rc = mdo_declare_index_insert(env, child, mdd_object_fid(child),
1968 S_IFDIR, dot, handle);
1972 rc = mdo_declare_ref_add(env, child, handle);
1976 rc = mdo_declare_index_insert(env, child, mdd_object_fid(parent),
1977 S_IFDIR, dotdot, handle);
1982 static int mdd_object_initialize(const struct lu_env *env,
1983 const struct lu_fid *pfid,
1984 struct mdd_object *child,
1985 struct lu_attr *attr,
1986 struct thandle *handle)
1991 if (S_ISDIR(attr->la_mode)) {
1992 /* Add "." and ".." for newly created dir */
1993 mdo_ref_add(env, child, handle);
1994 rc = __mdd_index_insert_only(env, child, mdd_object_fid(child),
1995 S_IFDIR, dot, handle);
1997 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2000 mdo_ref_del(env, child, handle);
2007 * This function checks whether it can create a file/dir under the
2008 * directory(@pobj). The directory(@pobj) is not being locked by
2011 * \param[in] env execution environment
2012 * \param[in] pobj the directory to create files
2013 * \param[in] pattr the attributes of the directory
2014 * \param[in] lname the name of the created file/dir
2015 * \param[in] cattr the attributes of the file/dir
2016 * \param[in] spec create specification
2018 * \retval = 0 it is allowed to create file/dir under
2020 * \retval negative error not allowed to create file/dir
2021 * under the directory
2023 static int mdd_create_sanity_check(const struct lu_env *env,
2024 struct md_object *pobj,
2025 const struct lu_attr *pattr,
2026 const struct lu_name *lname,
2027 struct lu_attr *cattr,
2028 struct md_op_spec *spec)
2030 struct mdd_thread_info *info = mdd_env_info(env);
2031 struct lu_fid *fid = &info->mti_fid;
2032 struct mdd_object *obj = md2mdd_obj(pobj);
2033 struct mdd_device *m = mdo2mdd(pobj);
2034 bool check_perm = true;
2039 if (mdd_is_dead_obj(obj))
2043 * In some cases this lookup is not needed - we know before if name
2044 * exists or not because MDT performs lookup for it.
2045 * name length check is done in lookup.
2047 if (spec->sp_cr_lookup) {
2049 * Check if the name already exist, though it will be checked in
2050 * _index_insert also, for avoiding rolling back if exists
2053 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2054 MAY_WRITE | MAY_EXEC);
2056 RETURN(rc ? : -EEXIST);
2058 /* Permission is already being checked in mdd_lookup */
2062 if (S_ISDIR(cattr->la_mode) &&
2063 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2064 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2065 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2067 if (!lmv_user_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2068 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2070 CERROR("%s: invalid lmv_user_md: magic = %x, "
2071 "stripe_offset = %d, stripe_count = %u: "
2072 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2073 le32_to_cpu(lum->lum_magic),
2074 (int)le32_to_cpu(lum->lum_stripe_offset),
2075 le32_to_cpu(lum->lum_stripe_count), rc);
2080 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2085 if (pattr->la_mode & S_ISGID) {
2086 struct lu_ucred *uc = lu_ucred(env);
2088 cattr->la_gid = pattr->la_gid;
2090 /* Directories are special, and always inherit S_ISGID */
2091 if (S_ISDIR(cattr->la_mode)) {
2092 cattr->la_mode |= S_ISGID;
2093 cattr->la_valid |= LA_MODE;
2094 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2095 == (S_ISGID | S_IXGRP) &&
2096 !lustre_in_group_p(uc,
2097 (cattr->la_valid & LA_GID) ?
2098 cattr->la_gid : pattr->la_gid) &&
2099 !md_capable(uc, CAP_FSETID)) {
2100 cattr->la_mode &= ~S_ISGID;
2101 cattr->la_valid |= LA_MODE;
2105 /* Inherit project ID from parent directory */
2106 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2107 cattr->la_projid = pattr->la_projid;
2108 if (S_ISDIR(cattr->la_mode)) {
2109 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2110 cattr->la_valid |= LA_FLAGS;
2112 cattr->la_valid |= LA_PROJID;
2115 rc = mdd_name_check(m, lname);
2119 switch (cattr->la_mode & S_IFMT) {
2121 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2123 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2124 RETURN(-ENAMETOOLONG);
2143 static int mdd_declare_create_object(const struct lu_env *env,
2144 struct mdd_device *mdd,
2145 struct mdd_object *p, struct mdd_object *c,
2146 struct lu_attr *attr,
2147 struct thandle *handle,
2148 const struct md_op_spec *spec,
2149 struct lu_buf *def_acl_buf,
2150 struct lu_buf *acl_buf,
2151 struct lu_buf *hsm_buf,
2152 struct dt_allocation_hint *hint)
2154 const struct lu_buf *buf;
2157 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2158 /* ldiskfs OSD needs this information for credit allocation */
2160 hint->dah_acl_len = def_acl_buf->lb_len;
2162 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2167 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2168 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2169 /* if dir, then can inherit default ACl */
2170 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2171 XATTR_NAME_ACL_DEFAULT,
2177 if (acl_buf && acl_buf->lb_len > 0) {
2178 rc = mdo_declare_attr_set(env, c, attr, handle);
2182 rc = mdo_declare_xattr_set(env, c, acl_buf,
2183 XATTR_NAME_ACL_ACCESS, 0, handle);
2188 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2192 /* replay case, create LOV EA from client data */
2193 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2194 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2195 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2196 spec->u.sp_ea.eadatalen);
2197 rc = mdo_declare_xattr_set(env, c, buf,
2198 S_ISDIR(attr->la_mode) ?
2199 XATTR_NAME_LMV : XATTR_NAME_LOV,
2200 LU_XATTR_CREATE, handle);
2204 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2205 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2213 if (S_ISLNK(attr->la_mode)) {
2214 const char *target_name = spec->u.sp_symname;
2215 int sym_len = strlen(target_name);
2216 const struct lu_buf *buf;
2218 buf = mdd_buf_get_const(env, target_name, sym_len);
2219 rc = dt_declare_record_write(env, mdd_object_child(c),
2225 if (spec->sp_cr_file_secctx_name != NULL) {
2226 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2227 spec->sp_cr_file_secctx_size);
2228 rc = mdo_declare_xattr_set(env, c, buf,
2229 spec->sp_cr_file_secctx_name, 0,
2235 if (spec->sp_cr_file_encctx != NULL) {
2236 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2237 spec->sp_cr_file_encctx_size);
2238 rc = mdo_declare_xattr_set(env, c, buf,
2239 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2248 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2249 struct mdd_object *p, struct mdd_object *c,
2250 const struct lu_name *name,
2251 struct lu_attr *attr,
2252 struct thandle *handle,
2253 const struct md_op_spec *spec,
2254 struct linkea_data *ldata,
2255 struct lu_buf *def_acl_buf,
2256 struct lu_buf *acl_buf,
2257 struct lu_buf *hsm_buf,
2258 struct dt_allocation_hint *hint)
2262 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2263 def_acl_buf, acl_buf, hsm_buf, hint);
2267 if (S_ISDIR(attr->la_mode)) {
2268 rc = mdo_declare_ref_add(env, p, handle);
2273 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2274 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2278 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2279 enum changelog_rec_type type;
2281 rc = mdo_declare_index_insert(env, p, mdd_object_fid(c),
2282 attr->la_mode, name->ln_name,
2287 rc = mdd_declare_links_add(env, c, handle, ldata);
2292 la->la_valid = LA_CTIME | LA_MTIME;
2293 rc = mdo_declare_attr_set(env, p, la, handle);
2297 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2298 S_ISREG(attr->la_mode) ? CL_CREATE :
2299 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2301 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2310 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2311 struct lu_attr *la, struct lu_buf *def_acl_buf,
2312 struct lu_buf *acl_buf)
2318 if (S_ISLNK(la->la_mode)) {
2319 acl_buf->lb_len = 0;
2320 def_acl_buf->lb_len = 0;
2324 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2325 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2326 XATTR_NAME_ACL_DEFAULT);
2327 mdd_read_unlock(env, pobj);
2329 /* ACL buffer size is not enough, need realloc */
2330 if (rc > acl_buf->lb_len)
2333 /* If there are default ACL, fix mode/ACL by default ACL */
2334 def_acl_buf->lb_len = rc;
2335 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2336 acl_buf->lb_len = rc;
2337 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2340 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2341 /* If there are no default ACL, fix mode by mask */
2342 struct lu_ucred *uc = lu_ucred(env);
2344 /* The create triggered by MDT internal events, such as
2345 * LFSCK reset, will not contain valid "uc". */
2346 if (unlikely(uc != NULL))
2347 la->la_mode &= ~uc->uc_umask;
2349 acl_buf->lb_len = 0;
2350 def_acl_buf->lb_len = 0;
2357 * Create a metadata object and initialize it, set acl, xattr.
2359 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2360 struct mdd_object *son, struct lu_attr *attr,
2361 struct md_op_spec *spec, struct lu_buf *acl_buf,
2362 struct lu_buf *def_acl_buf,
2363 struct lu_buf *hsm_buf,
2364 struct dt_allocation_hint *hint,
2365 struct thandle *handle, bool initsecctx)
2367 const struct lu_buf *buf;
2370 mdd_write_lock(env, son, DT_TGT_CHILD);
2371 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2376 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2377 * created in declare phase, they also needs to be added to master
2378 * object as sub-directory entry. So it has to initialize the master
2379 * object, then set dir striped EA.(in mdo_xattr_set) */
2380 rc = mdd_object_initialize(env, mdd_object_fid(pobj), son, attr,
2383 GOTO(err_destroy, rc);
2386 * in case of replay we just set LOVEA provided by the client
2387 * XXX: I think it would be interesting to try "old" way where
2388 * MDT calls this xattr_set(LOV) in a different transaction.
2389 * probably this way we code can be made better.
2392 /* During creation, there are only a few cases we need do xattr_set to
2394 * 1. regular file: see comments above.
2395 * 2. dir: inherit default striping or pool settings from parent.
2396 * 3. create striped directory with provided stripeEA.
2397 * 4. create striped directory because inherit default layout from the
2400 if (spec->no_create ||
2401 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2402 S_ISDIR(attr->la_mode)) {
2403 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2404 spec->u.sp_ea.eadatalen);
2405 rc = mdo_xattr_set(env, son, buf,
2406 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2408 LU_XATTR_CREATE, handle);
2410 GOTO(err_destroy, rc);
2413 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2416 memset(&mh, 0, sizeof(mh));
2417 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2418 mh.mh_arch_id = spec->sp_archive_id;
2419 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2420 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2423 GOTO(err_destroy, rc);
2426 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2427 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2428 S_ISDIR(attr->la_mode)) {
2429 /* set default acl */
2430 rc = mdo_xattr_set(env, son, def_acl_buf,
2431 XATTR_NAME_ACL_DEFAULT, 0,
2434 GOTO(err_destroy, rc);
2436 /* set its own acl */
2437 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2438 rc = mdo_xattr_set(env, son, acl_buf,
2439 XATTR_NAME_ACL_ACCESS,
2442 GOTO(err_destroy, rc);
2446 if (S_ISLNK(attr->la_mode)) {
2447 struct dt_object *dt = mdd_object_child(son);
2448 const char *target_name = spec->u.sp_symname;
2449 int sym_len = strlen(target_name);
2452 buf = mdd_buf_get_const(env, target_name, sym_len);
2453 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2457 GOTO(err_initlized, rc = -EFAULT);
2460 if (initsecctx && spec->sp_cr_file_secctx_name != NULL) {
2461 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2462 spec->sp_cr_file_secctx_size);
2463 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2466 GOTO(err_initlized, rc);
2469 if (spec->sp_cr_file_encctx != NULL) {
2470 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2471 spec->sp_cr_file_encctx_size);
2472 rc = mdo_xattr_set(env, son, buf,
2473 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2476 GOTO(err_initlized, rc);
2480 if (unlikely(rc != 0)) {
2482 if (S_ISDIR(attr->la_mode)) {
2483 /* Drop the reference, no need to delete "."/"..",
2484 * because the object to be destroied directly. */
2485 rc2 = mdo_ref_del(env, son, handle);
2489 rc2 = mdo_ref_del(env, son, handle);
2493 mdo_destroy(env, son, handle);
2496 mdd_write_unlock(env, son);
2500 static int mdd_index_delete(const struct lu_env *env,
2501 struct mdd_object *mdd_pobj,
2502 struct lu_attr *cattr,
2503 const struct lu_name *lname)
2505 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2506 struct thandle *handle;
2510 handle = mdd_trans_create(env, mdd);
2512 RETURN(PTR_ERR(handle));
2514 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2519 if (S_ISDIR(cattr->la_mode)) {
2520 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2525 /* Since this will only be used in the error handler path,
2526 * Let's set the thandle to be local and not mess the transno */
2527 handle->th_local = 1;
2528 rc = mdd_trans_start(env, mdd, handle);
2532 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2533 S_ISDIR(cattr->la_mode), handle);
2537 rc = mdd_trans_stop(env, mdd, rc, handle);
2543 * Create object and insert it into namespace.
2545 * Two operations have to be performed:
2547 * - an allocation of a new object (->do_create()), and
2548 * - an insertion into a parent index (->dio_insert()).
2550 * Due to locking, operation order is not important, when both are
2551 * successful, *but* error handling cases are quite different:
2553 * - if insertion is done first, and following object creation fails,
2554 * insertion has to be rolled back, but this operation might fail
2555 * also leaving us with dangling index entry.
2557 * - if creation is done first, is has to be undone if insertion fails,
2558 * leaving us with leaked space, which is not good but not fatal.
2560 * It seems that creation-first is simplest solution, but it is sub-optimal
2566 * case, because second mkdir is bound to create object, only to
2567 * destroy it immediately.
2569 * To avoid this follow local file systems that do double lookup:
2571 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2572 * 1. create (mdd_create_object_internal())
2573 * 2. insert (__mdd_index_insert(), lookup again)
2575 * \param[in] pobj parent object
2576 * \param[in] lname name of child being created
2577 * \param[in,out] child child object being created
2578 * \param[in] spec additional create parameters
2579 * \param[in] ma attributes for new child object
2581 * \retval 0 on success
2582 * \retval negative errno on failure
2584 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2585 const struct lu_name *lname, struct md_object *child,
2586 struct md_op_spec *spec, struct md_attr *ma)
2588 struct mdd_thread_info *info = mdd_env_info(env);
2589 struct lu_attr *la = &info->mti_la_for_fix;
2590 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2591 struct mdd_object *son = md2mdd_obj(child);
2592 struct mdd_device *mdd = mdo2mdd(pobj);
2593 struct lu_attr *attr = &ma->ma_attr;
2594 struct thandle *handle;
2595 struct lu_attr *pattr = &info->mti_pattr;
2596 struct lu_buf acl_buf;
2597 struct lu_buf def_acl_buf;
2598 struct lu_buf hsm_buf;
2599 struct linkea_data *ldata = &info->mti_link_data;
2600 const char *name = lname->ln_name;
2601 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2602 int acl_size = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
2607 rc = mdd_la_get(env, mdd_pobj, pattr);
2611 /* Sanity checks before big job. */
2612 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2616 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2617 GOTO(out_free, rc = -EINPROGRESS);
2619 handle = mdd_trans_create(env, mdd);
2621 GOTO(out_free, rc = PTR_ERR(handle));
2624 acl_buf = *lu_buf_check_and_alloc(&info->mti_xattr_buf, acl_size);
2625 if (!acl_buf.lb_buf)
2626 GOTO(out_stop, rc = -ENOMEM);
2628 def_acl_buf = *lu_buf_check_and_alloc(&info->mti_big_buf, acl_size);
2629 if (!def_acl_buf.lb_buf)
2630 GOTO(out_stop, rc = -ENOMEM);
2632 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2633 if (unlikely(rc == -ERANGE &&
2634 acl_size == LUSTRE_POSIX_ACL_MAX_SIZE_OLD)) {
2635 /* use maximum-sized xattr buffer for too-big default ACL */
2636 acl_size = min_t(unsigned int, mdd->mdd_dt_conf.ddp_max_ea_size,
2638 goto use_bigger_buffer;
2643 if (S_ISDIR(attr->la_mode)) {
2644 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2647 * migrate may create 1-stripe directory, so lod_ah_init()
2648 * doesn't adjust stripe count from lmu.
2650 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1))
2651 lmu->lum_stripe_count = 0;
2654 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2656 memset(ldata, 0, sizeof(*ldata));
2657 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2658 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2661 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2662 &tfid, lname, 1, 0, ldata);
2664 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2665 mdd_object_fid(mdd_pobj),
2666 lname, 1, 0, ldata);
2669 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2670 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2672 memset(&hsm_buf, 0, sizeof(hsm_buf));
2673 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2674 if (hsm_buf.lb_buf == NULL)
2675 GOTO(out_stop, rc = -ENOMEM);
2678 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2679 handle, spec, ldata, &def_acl_buf, &acl_buf,
2684 rc = mdd_trans_start(env, mdd, handle);
2688 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2689 &def_acl_buf, &hsm_buf, hint, handle, true);
2693 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2694 mdd_write_lock(env, son, DT_TGT_CHILD);
2695 son->mod_flags |= VOLATILE_OBJ;
2696 rc = mdd_orphan_insert(env, son, handle);
2697 GOTO(out_volatile, rc);
2699 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(son),
2700 attr->la_mode, name, handle);
2702 GOTO(err_created, rc);
2704 mdd_links_add(env, son, mdd_object_fid(mdd_pobj), lname,
2707 /* update parent directory mtime/ctime */
2709 la->la_valid = LA_CTIME | LA_MTIME;
2710 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2712 GOTO(err_insert, rc);
2718 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2719 rc2 = mdd_orphan_delete(env, son, handle);
2721 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2722 S_ISDIR(attr->la_mode),
2728 mdd_write_lock(env, son, DT_TGT_CHILD);
2729 if (S_ISDIR(attr->la_mode)) {
2730 /* Drop the reference, no need to delete "."/"..",
2731 * because the object is to be destroyed directly. */
2732 rc2 = mdo_ref_del(env, son, handle);
2734 mdd_write_unlock(env, son);
2739 /* For volatile files drop one link immediately, since there is
2740 * no filename in the namespace, and save any error returned. */
2741 rc2 = mdo_ref_del(env, son, handle);
2743 mdd_write_unlock(env, son);
2744 if (unlikely(rc == 0))
2749 /* Don't destroy the volatile object on success */
2750 if (likely(rc != 0))
2751 mdo_destroy(env, son, handle);
2752 mdd_write_unlock(env, son);
2755 if (rc == 0 && fid_is_namespace_visible(mdd_object_fid(son)) &&
2756 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2757 rc = mdd_changelog_ns_store(env, mdd,
2758 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2759 S_ISREG(attr->la_mode) ? CL_CREATE :
2760 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2761 0, son, mdd_object_fid(mdd_pobj), NULL, NULL,
2762 lname, NULL, handle);
2764 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2766 /* If creation fails, it is most likely due to the remote update
2767 * failure, because local transaction will mostly succeed at
2768 * this stage. There is no easy way to rollback all of previous
2769 * updates, so let's remove the object from namespace, and
2770 * LFSCK should handle the orphan object. */
2771 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2772 mdd_index_delete(env, mdd_pobj, attr, lname);
2776 if (is_vmalloc_addr(ldata->ld_buf))
2777 /* if we vmalloced a large buffer drop it */
2778 lu_buf_free(ldata->ld_buf);
2780 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2781 lu_buf_free(&hsm_buf);
2783 /* The child object shouldn't be cached anymore */
2785 set_bit(LU_OBJECT_HEARD_BANSHEE,
2786 &child->mo_lu.lo_header->loh_flags);
2790 /* has not mdd_write{read}_lock on any obj yet. */
2791 static int mdd_rename_sanity_check(const struct lu_env *env,
2792 struct mdd_object *src_pobj,
2793 const struct lu_attr *pattr,
2794 struct mdd_object *tgt_pobj,
2795 const struct lu_attr *tpattr,
2796 struct mdd_object *sobj,
2797 const struct lu_attr *cattr,
2798 struct mdd_object *tobj,
2799 const struct lu_attr *tattr)
2804 /* XXX: when get here, sobj must NOT be NULL,
2805 * the other case has been processed in cld_rename
2806 * before mdd_rename and enable MDS_PERM_BYPASS. */
2810 * If we are using project inheritance, we only allow renames
2811 * into our tree when the project IDs are the same; otherwise
2812 * tree quota mechanism would be circumvented.
2814 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2815 tpattr->la_projid != cattr->la_projid) ||
2816 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2817 (pattr->la_projid != tpattr->la_projid)))
2820 /* we prevent an encrypted file from being renamed
2821 * into an unencrypted dir
2823 if ((cattr->la_valid & LA_FLAGS &&
2824 cattr->la_flags & LUSTRE_ENCRYPT_FL) &&
2825 !(tpattr->la_valid & LA_FLAGS &&
2826 tpattr->la_flags & LUSTRE_ENCRYPT_FL))
2829 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2833 /* XXX: when get here, "tobj == NULL" means tobj must
2834 * NOT exist (neither on remote MDS, such case has been
2835 * processed in cld_rename before mdd_rename and enable
2837 * So check may_create, but not check may_unlink. */
2839 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2840 (src_pobj != tgt_pobj));
2842 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2843 (src_pobj != tgt_pobj), 1);
2845 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2846 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2851 static int mdd_declare_rename(const struct lu_env *env,
2852 struct mdd_device *mdd,
2853 struct mdd_object *mdd_spobj,
2854 struct mdd_object *mdd_tpobj,
2855 struct mdd_object *mdd_sobj,
2856 struct mdd_object *mdd_tobj,
2857 const struct lu_name *sname,
2858 const struct lu_name *tname,
2860 struct linkea_data *ldata,
2861 struct thandle *handle)
2863 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2866 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2867 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2873 /* name from source dir */
2874 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2878 /* .. from source child */
2879 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2880 /* source child can be directory, count by source dir's nlink */
2881 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2884 if (mdd_spobj != mdd_tpobj) {
2885 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2890 rc = mdo_declare_index_insert(env, mdd_sobj,
2891 mdd_object_fid(mdd_tpobj),
2892 S_IFDIR, dotdot, handle);
2897 /* new target child can be directory,
2898 * counted by target dir's nlink */
2899 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2904 la->la_valid = LA_CTIME | LA_MTIME;
2905 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2909 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2913 la->la_valid = LA_CTIME;
2914 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2918 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2923 rc = mdo_declare_index_insert(env, mdd_tpobj, mdd_object_fid(mdd_sobj),
2924 mdd_object_type(mdd_sobj),
2925 tname->ln_name, handle);
2929 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2930 /* delete target child in target parent directory */
2931 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2936 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2940 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2941 /* target child can be directory,
2942 * delete "." reference in target child directory */
2943 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2947 /* delete ".." reference in target parent directory */
2948 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2953 la->la_valid = LA_CTIME;
2954 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2958 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2963 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2971 static int mdd_migrate_object(const struct lu_env *env,
2972 struct mdd_object *spobj,
2973 struct mdd_object *tpobj,
2974 struct mdd_object *sobj,
2975 struct mdd_object *tobj,
2976 const struct lu_name *sname,
2977 const struct lu_name *tname,
2978 struct md_op_spec *spec,
2979 struct md_attr *ma);
2981 /* src object can be remote that is why we use only fid and type of object */
2982 static int mdd_rename(const struct lu_env *env,
2983 struct md_object *src_pobj, struct md_object *tgt_pobj,
2984 const struct lu_fid *lf, const struct lu_name *lsname,
2985 struct md_object *tobj, const struct lu_name *ltname,
2988 const char *sname = lsname->ln_name;
2989 const char *tname = ltname->ln_name;
2990 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2991 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2992 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2993 struct mdd_device *mdd = mdo2mdd(src_pobj);
2994 struct mdd_object *mdd_sobj = NULL; /* source object */
2995 struct mdd_object *mdd_tobj = NULL;
2996 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2997 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2998 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2999 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3000 struct thandle *handle;
3001 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3002 const struct lu_fid *tpobj_fid = mdd_object_fid(mdd_tpobj);
3003 const struct lu_fid *spobj_fid = mdd_object_fid(mdd_spobj);
3006 bool tobj_locked = 0;
3007 unsigned cl_flags = 0;
3011 /* let unlink to complete and commit */
3012 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
3015 mdd_tobj = md2mdd_obj(tobj);
3017 mdd_sobj = mdd_object_find(env, mdd, lf);
3018 if (IS_ERR(mdd_sobj))
3019 RETURN(PTR_ERR(mdd_sobj));
3021 rc = mdd_la_get(env, mdd_sobj, cattr);
3023 GOTO(out_pending, rc);
3025 /* if rename is cross MDTs, migrate symlink if it doesn't have other
3026 * hard links, and target doesn't exist.
3028 if (mdd_object_remote(mdd_sobj) && S_ISLNK(cattr->la_mode) &&
3029 cattr->la_nlink == 1 && !tobj) {
3030 struct md_op_spec *spec = &mdd_env_info(env)->mti_spec;
3031 struct lu_device *ld = &mdd->mdd_md_dev.md_lu_dev;
3034 rc = ld->ld_ops->ldo_fid_alloc(env, ld, &tfid, &tgt_pobj->mo_lu,
3037 GOTO(out_pending, rc);
3039 mdd_tobj = mdd_object_find(env, mdd, &tfid);
3040 if (IS_ERR(mdd_tobj))
3041 GOTO(out_pending, rc = PTR_ERR(mdd_tobj));
3043 memset(spec, 0, sizeof(*spec));
3044 rc = mdd_migrate_object(env, mdd_spobj, mdd_tpobj, mdd_sobj,
3045 mdd_tobj, lsname, ltname, spec, ma);
3046 mdd_object_put(env, mdd_tobj);
3047 GOTO(out_pending, rc);
3050 rc = mdd_la_get(env, mdd_spobj, pattr);
3052 GOTO(out_pending, rc);
3055 rc = mdd_la_get(env, mdd_tobj, tattr);
3057 GOTO(out_pending, rc);
3058 /* search for an existing archive.
3059 * we should check ahead as the object
3060 * can be destroyed in this transaction */
3061 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3062 cl_flags |= CLF_RENAME_LAST_EXISTS;
3065 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3067 GOTO(out_pending, rc);
3069 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3070 mdd_sobj, cattr, mdd_tobj, tattr);
3072 GOTO(out_pending, rc);
3074 rc = mdd_name_check(mdd, ltname);
3076 GOTO(out_pending, rc);
3078 handle = mdd_trans_create(env, mdd);
3080 GOTO(out_pending, rc = PTR_ERR(handle));
3082 memset(ldata, 0, sizeof(*ldata));
3083 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3084 lsname, mdd_object_fid(mdd_tpobj), ltname,
3089 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3090 mdd_tobj, lsname, ltname, ma, ldata, handle);
3094 rc = mdd_trans_start(env, mdd, handle);
3098 is_dir = S_ISDIR(cattr->la_mode);
3100 /* Remove source name from source directory */
3101 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3105 /* "mv dir1 dir2" needs "dir1/.." link update */
3106 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3107 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3109 GOTO(fixup_spobj2, rc);
3111 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3114 GOTO(fixup_spobj, rc);
3117 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3118 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3120 /* tname might been renamed to something else */
3121 GOTO(fixup_spobj, rc);
3124 /* Insert new fid with target name into target dir */
3125 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3128 GOTO(fixup_tpobj, rc);
3130 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3131 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3133 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3134 la->la_valid = LA_CTIME;
3135 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3137 GOTO(fixup_tpobj, rc);
3139 /* Update the linkEA for the source object */
3140 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3141 rc = mdd_links_rename(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3142 lsname, mdd_object_fid(mdd_tpobj), ltname,
3143 handle, ldata, 0, 0);
3145 /* Old files might not have EA entry */
3146 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3147 lsname, handle, NULL, 0);
3148 mdd_write_unlock(env, mdd_sobj);
3149 /* We don't fail the transaction if the link ea can't be
3150 updated -- fid2path will use alternate lookup method. */
3153 /* Remove old target object
3154 * For tobj is remote case cmm layer has processed
3155 * and set tobj to NULL then. So when tobj is NOT NULL,
3156 * it must be local one.
3158 if (tobj && mdd_object_exists(mdd_tobj)) {
3159 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3161 if (mdd_is_dead_obj(mdd_tobj)) {
3162 /* shld not be dead, something is wrong */
3163 CERROR("tobj is dead, something is wrong\n");
3167 mdo_ref_del(env, mdd_tobj, handle);
3169 /* Remove dot reference. */
3170 if (S_ISDIR(tattr->la_mode))
3171 mdo_ref_del(env, mdd_tobj, handle);
3174 /* fetch updated nlink */
3175 rc = mdd_la_get(env, mdd_tobj, tattr);
3177 CERROR("%s: Failed to get nlink for tobj "
3179 mdd2obd_dev(mdd)->obd_name,
3180 PFID(tpobj_fid), rc);
3181 GOTO(fixup_tpobj, rc);
3184 la->la_valid = LA_CTIME;
3185 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3187 CERROR("%s: Failed to set ctime for tobj "
3189 mdd2obd_dev(mdd)->obd_name,
3190 PFID(tpobj_fid), rc);
3191 GOTO(fixup_tpobj, rc);
3194 /* XXX: this transfer to ma will be removed with LOD/OSP */
3195 ma->ma_attr = *tattr;
3196 ma->ma_valid |= MA_INODE;
3197 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3200 CERROR("%s: Failed to unlink tobj "
3202 mdd2obd_dev(mdd)->obd_name,
3203 PFID(tpobj_fid), rc);
3204 GOTO(fixup_tpobj, rc);
3207 /* fetch updated nlink */
3208 rc = mdd_la_get(env, mdd_tobj, tattr);
3209 if (rc == -ENOENT) {
3210 /* the object got removed, let's
3211 * return the latest known attributes */
3212 tattr->la_nlink = 0;
3214 } else if (rc != 0) {
3215 CERROR("%s: Failed to get nlink for tobj "
3217 mdd2obd_dev(mdd)->obd_name,
3218 PFID(tpobj_fid), rc);
3219 GOTO(fixup_tpobj, rc);
3221 /* XXX: this transfer to ma will be removed with LOD/OSP */
3222 ma->ma_attr = *tattr;
3223 ma->ma_valid |= MA_INODE;
3225 if (tattr->la_nlink == 0)
3226 cl_flags |= CLF_RENAME_LAST;
3228 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3231 la->la_valid = LA_CTIME | LA_MTIME;
3232 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3234 GOTO(fixup_tpobj, rc);
3236 if (mdd_spobj != mdd_tpobj) {
3237 la->la_valid = LA_CTIME | LA_MTIME;
3238 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3240 GOTO(fixup_tpobj, rc);
3247 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3249 CWARN("tp obj fix error %d\n",rc2);
3251 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3252 !mdd_is_dead_obj(mdd_tobj)) {
3254 mdo_ref_add(env, mdd_tobj, handle);
3256 mdo_ref_add(env, mdd_tobj, handle);
3259 rc2 = __mdd_index_insert(env, mdd_tpobj,
3260 mdd_object_fid(mdd_tobj),
3261 mdd_object_type(mdd_tobj),
3264 CWARN("tp obj fix error: rc = %d\n", rc2);
3269 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3270 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3272 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3273 mdd2obd_dev(mdd)->obd_name, rc2);
3276 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3279 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3280 mdd2obd_dev(mdd)->obd_name, rc2);
3285 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3286 mdd_object_type(mdd_sobj), sname,
3289 CWARN("sp obj fix error: rc = %d\n", rc2);
3294 mdd_write_unlock(env, mdd_tobj);
3297 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3298 mdd_tobj, tpobj_fid, lf, spobj_fid,
3299 ltname, lsname, handle);
3302 rc = mdd_trans_stop(env, mdd, rc, handle);
3305 mdd_object_put(env, mdd_sobj);
3310 * Check whether we should migrate the file/dir
3312 * < 0 permission check failed or other error.
3313 * = 0 the file can be migrated.
3315 static int mdd_migrate_sanity_check(const struct lu_env *env,
3316 struct mdd_device *mdd,
3317 struct mdd_object *spobj,
3318 struct mdd_object *tpobj,
3319 struct mdd_object *sobj,
3320 struct mdd_object *tobj,
3321 const struct lu_attr *spattr,
3322 const struct lu_attr *tpattr,
3323 const struct lu_attr *attr)
3329 if (!mdd_object_remote(sobj)) {
3330 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3331 if (sobj->mod_count > 0) {
3332 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3333 mdd_obj_dev_name(sobj),
3334 PFID(mdd_object_fid(sobj)),
3336 mdd_read_unlock(env, sobj);
3339 mdd_read_unlock(env, sobj);
3342 if (mdd_object_exists(tobj))
3345 rc = mdd_may_delete(env, spobj, spattr, sobj, attr, NULL, 1, 0);
3349 rc = mdd_may_create(env, tpobj, tpattr, NULL, true);
3354 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3355 struct mdd_object *obj,
3356 const struct lu_buf *buf,
3358 int fl, struct thandle *handle);
3360 /* iterate xattrs, but ignore LMA, LMV, and LINKEA if 'skip_linkea' is set. */
3361 static int mdd_iterate_xattrs(const struct lu_env *env,
3362 struct mdd_object *sobj,
3363 struct mdd_object *tobj,
3365 struct thandle *handle,
3368 struct mdd_thread_info *info = mdd_env_info(env);
3370 struct lu_buf list_xbuf;
3371 struct lu_buf cbxbuf;
3372 struct lu_buf xbuf = { NULL };
3381 /* retrieve xattr list from the old object */
3382 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3383 if (list_xsize == -ENODATA)
3389 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3390 if (info->mti_big_buf.lb_buf == NULL)
3393 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3394 list_xbuf.lb_len = list_xsize;
3395 rc = mdo_xattr_list(env, sobj, &list_xbuf);
3401 xname = list_xbuf.lb_buf;
3403 xlen = strnlen(xname, rem - 1) + 1;
3404 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3405 strcmp(XATTR_NAME_LMV, xname) == 0)
3409 strcmp(XATTR_NAME_LINK, xname) == 0)
3412 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3413 if (xsize == -ENODATA)
3416 GOTO(out, rc = xsize);
3418 lu_buf_check_and_alloc(&xbuf, xsize);
3419 if (xbuf.lb_buf == NULL)
3420 GOTO(out, rc = -ENOMEM);
3422 rc = mdo_xattr_get(env, sobj, &xbuf, xname);
3429 cbxbuf.lb_len = xsize;
3431 rc = cb(env, tobj, &cbxbuf, xname, 0, handle);
3432 if (unlikely(rc == -ENOSPC &&
3433 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3434 rc = linkea_overflow_shrink(
3435 (struct linkea_data *)(cbxbuf.lb_buf));
3436 if (likely(rc > 0)) {
3454 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3455 struct mdd_object *sobj,
3456 struct mdd_object *tobj,
3457 const struct lu_name *sname,
3458 const struct lu_fid *sfid,
3459 const struct lu_name *lname,
3460 const struct lu_fid *fid,
3462 struct thandle *handle);
3464 static int mdd_declare_update_link(const struct lu_env *env,
3465 struct mdd_object *sobj,
3466 struct mdd_object *tobj,
3467 const struct lu_name *tname,
3468 const struct lu_fid *tpfid,
3469 const struct lu_name *lname,
3470 const struct lu_fid *fid,
3472 struct thandle *handle)
3474 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3475 struct mdd_object *pobj;
3479 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3480 !strcmp(tname->ln_name, lname->ln_name))
3483 pobj = mdd_object_find(env, mdd, fid);
3485 return PTR_ERR(pobj);
3488 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3490 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
3491 mdd_object_type(sobj),
3492 lname->ln_name, handle);
3493 mdd_object_put(env, pobj);
3497 rc = mdo_declare_ref_add(env, tobj, handle);
3501 rc = mdo_declare_ref_del(env, sobj, handle);
3505 static int mdd_update_link(const struct lu_env *env,
3506 struct mdd_object *sobj,
3507 struct mdd_object *tobj,
3508 const struct lu_name *tname,
3509 const struct lu_fid *tpfid,
3510 const struct lu_name *lname,
3511 const struct lu_fid *fid,
3513 struct thandle *handle)
3515 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3516 struct mdd_object *pobj;
3521 LASSERT(lu_name_is_valid(lname));
3524 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3525 !strncmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3528 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3529 PFID(fid), PNAME(lname), PFID(mdd_object_fid(tobj)));
3531 pobj = mdd_object_find(env, mdd, fid);
3533 CWARN("%s: cannot find obj "DFID": %ld\n",
3534 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3535 RETURN(PTR_ERR(pobj));
3538 if (!mdd_object_exists(pobj)) {
3539 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3540 mdd_object_put(env, pobj);
3544 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3545 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3547 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(tobj),
3548 mdd_object_type(sobj),
3549 lname->ln_name, handle);
3550 mdd_write_unlock(env, pobj);
3551 mdd_object_put(env, pobj);
3555 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3556 rc = mdo_ref_add(env, tobj, handle);
3557 mdd_write_unlock(env, tobj);
3561 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3562 rc = mdo_ref_del(env, sobj, handle);
3563 mdd_write_unlock(env, sobj);
3568 static inline int mdd_fld_lookup(const struct lu_env *env,
3569 struct mdd_device *mdd,
3570 const struct lu_fid *fid,
3573 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
3574 struct seq_server_site *ss;
3577 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3579 range->lsr_flags = LU_SEQ_RANGE_MDT;
3580 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3584 *mdt_index = range->lsr_index;
3589 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3590 struct mdd_object *sobj,
3591 struct mdd_object *tobj,
3592 const struct lu_name *tname,
3593 const struct lu_fid *tpfid,
3594 const struct lu_name *lname,
3595 const struct lu_fid *fid,
3597 struct thandle *handle)
3599 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3600 __u32 source_mdt_index = *(__u32 *)opaque;
3601 __u32 link_mdt_index;
3607 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3608 !strcmp(tname->ln_name, lname->ln_name))
3611 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3615 RETURN(link_mdt_index == source_mdt_index);
3618 static int mdd_iterate_linkea(const struct lu_env *env,
3619 struct mdd_object *sobj,
3620 struct mdd_object *tobj,
3621 const struct lu_name *tname,
3622 const struct lu_fid *tpfid,
3623 struct linkea_data *ldata,
3625 struct thandle *handle,
3628 struct mdd_thread_info *info = mdd_env_info(env);
3629 char *filename = info->mti_name;
3630 struct lu_name lname;
3637 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3638 linkea_next_entry(ldata)) {
3639 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3642 /* Note: lname might miss \0 at the end */
3643 snprintf(filename, sizeof(info->mti_name), "%.*s",
3644 lname.ln_namelen, lname.ln_name);
3645 lname.ln_name = filename;
3647 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3649 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3657 * Prepare linkea, and check whether file needs migrate: if source still has
3658 * link on source MDT, no need to migrate, just update namespace on source and
3661 * \retval 0 do migrate
3662 * \retval 1 don't migrate
3663 * \retval -errno on failure
3665 static int mdd_migrate_linkea_prepare(const struct lu_env *env,
3666 struct mdd_device *mdd,
3667 struct mdd_object *spobj,
3668 struct mdd_object *tpobj,
3669 struct mdd_object *sobj,
3670 const struct lu_name *sname,
3671 const struct lu_name *tname,
3672 const struct lu_attr *attr,
3673 struct linkea_data *ldata)
3675 __u32 source_mdt_index;
3680 memset(ldata, 0, sizeof(*ldata));
3681 rc = mdd_linkea_prepare(env, sobj, mdd_object_fid(spobj), sname,
3682 mdd_object_fid(tpobj), tname, 1, 0, ldata);
3687 * Then it will check if the file should be migrated. If the file has
3688 * mulitple links, we only need migrate the file if all of its entries
3689 * has been migrated to the remote MDT.
3691 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3694 /* If there are still links locally, don't migrate this file */
3695 LASSERT(ldata->ld_leh != NULL);
3698 * If linkEA is overflow, it means there are some unknown name entries
3699 * under unknown parents, which will prevent the migration.
3701 if (unlikely(ldata->ld_leh->leh_overflow_time))
3704 rc = mdd_fld_lookup(env, mdd, mdd_object_fid(sobj), &source_mdt_index);
3708 rc = mdd_iterate_linkea(env, sobj, NULL, tname, mdd_object_fid(tpobj),
3709 ldata, &source_mdt_index, NULL,
3710 mdd_is_link_on_source_mdt);
3714 static int mdd_declare_migrate_update(const struct lu_env *env,
3715 struct mdd_object *spobj,
3716 struct mdd_object *tpobj,
3717 struct mdd_object *obj,
3718 const struct lu_name *sname,
3719 const struct lu_name *tname,
3720 struct lu_attr *attr,
3721 struct lu_attr *spattr,
3722 struct lu_attr *tpattr,
3723 struct linkea_data *ldata,
3725 struct thandle *handle)
3727 struct mdd_thread_info *info = mdd_env_info(env);
3728 struct lu_attr *la = &info->mti_la_for_fix;
3731 rc = mdo_declare_index_delete(env, spobj, sname->ln_name, handle);
3735 if (S_ISDIR(attr->la_mode)) {
3736 rc = mdo_declare_ref_del(env, spobj, handle);
3741 rc = mdo_declare_index_insert(env, tpobj, mdd_object_fid(obj),
3742 attr->la_mode & S_IFMT,
3743 tname->ln_name, handle);
3747 rc = mdd_declare_links_add(env, obj, handle, ldata);
3751 if (S_ISDIR(attr->la_mode)) {
3752 rc = mdo_declare_ref_add(env, tpobj, handle);
3757 la->la_valid = LA_CTIME | LA_MTIME;
3758 rc = mdo_declare_attr_set(env, spobj, la, handle);
3762 if (tpobj != spobj) {
3763 rc = mdo_declare_attr_set(env, tpobj, la, handle);
3771 static int mdd_declare_migrate_create(const struct lu_env *env,
3772 struct mdd_object *spobj,
3773 struct mdd_object *tpobj,
3774 struct mdd_object *sobj,
3775 struct mdd_object *tobj,
3776 const struct lu_name *sname,
3777 const struct lu_name *tname,
3778 struct lu_attr *spattr,
3779 struct lu_attr *tpattr,
3780 struct lu_attr *attr,
3781 struct lu_buf *sbuf,
3782 struct linkea_data *ldata,
3784 struct md_op_spec *spec,
3785 struct dt_allocation_hint *hint,
3786 struct thandle *handle)
3788 struct mdd_thread_info *info = mdd_env_info(env);
3789 struct md_layout_change *mlc = &info->mti_mlc;
3790 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3795 if (S_ISDIR(attr->la_mode)) {
3796 struct lmv_user_md *lum = spec->u.sp_ea.eadata;
3798 mlc->mlc_opc = MD_LAYOUT_DETACH;
3799 rc = mdo_declare_layout_change(env, sobj, mlc, handle);
3803 lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3804 } else if (S_ISLNK(attr->la_mode)) {
3805 spec->u.sp_symname = sbuf->lb_buf;
3806 } else if (S_ISREG(attr->la_mode)) {
3807 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
3808 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
3811 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
3813 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3814 tname, attr, handle, spec, ldata, NULL, NULL,
3820 * tobj mode will be used in mdo_declare_layout_change(), but it's not
3821 * createb yet, copy from sobj.
3823 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
3824 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
3825 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
3827 if (S_ISDIR(attr->la_mode)) {
3829 /* if sobj is not striped, fake a 1-stripe LMV */
3830 LASSERT(sizeof(info->mti_key) >
3831 lmv_mds_md_size(1, LMV_MAGIC_V1));
3832 lmv = (typeof(lmv))info->mti_key;
3833 memset(lmv, 0, sizeof(*lmv));
3834 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3835 lmv->lmv_stripe_count = cpu_to_le32(1);
3836 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
3837 fid_le_to_cpu(&lmv->lmv_stripe_fids[0],
3838 mdd_object_fid(sobj));
3839 mlc->mlc_buf.lb_buf = lmv;
3840 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3842 mlc->mlc_buf = *sbuf;
3844 mlc->mlc_opc = MD_LAYOUT_ATTACH;
3845 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
3850 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle,
3851 mdo_declare_xattr_set);
3855 if (S_ISREG(attr->la_mode)) {
3856 struct lu_buf fid_buf;
3858 handle->th_complex = 1;
3860 /* target may be remote, update PFID via sobj. */
3861 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
3862 fid_buf.lb_len = sizeof(struct lu_fid);
3863 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
3868 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
3873 if (!S_ISDIR(attr->la_mode)) {
3874 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
3875 mdd_object_fid(tpobj), ldata, NULL,
3876 handle, mdd_declare_update_link);
3881 if (!S_ISDIR(attr->la_mode) || lmv) {
3882 rc = mdo_declare_ref_del(env, sobj, handle);
3886 if (S_ISDIR(attr->la_mode)) {
3887 rc = mdo_declare_ref_del(env, sobj, handle);
3892 rc = mdo_declare_destroy(env, sobj, handle);
3897 rc = mdd_declare_migrate_update(env, spobj, tpobj, tobj, sname, tname,
3898 attr, spattr, tpattr, ldata, ma,
3904 * migrate dirent from \a spobj to \a tpobj.
3906 static int mdd_migrate_update(const struct lu_env *env,
3907 struct mdd_object *spobj,
3908 struct mdd_object *tpobj,
3909 struct mdd_object *obj,
3910 const struct lu_name *sname,
3911 const struct lu_name *tname,
3912 struct lu_attr *attr,
3913 struct lu_attr *spattr,
3914 struct lu_attr *tpattr,
3915 struct linkea_data *ldata,
3917 struct thandle *handle)
3919 struct mdd_thread_info *info = mdd_env_info(env);
3920 struct lu_attr *la = &info->mti_la_for_fix;
3925 CDEBUG(D_INFO, "update "DFID" from "DFID"/%s to "DFID"/%s\n",
3926 PFID(mdd_object_fid(obj)), PFID(mdd_object_fid(spobj)),
3927 sname->ln_name, PFID(mdd_object_fid(tpobj)), tname->ln_name);
3929 rc = __mdd_index_delete(env, spobj, sname->ln_name,
3930 S_ISDIR(attr->la_mode), handle);
3934 rc = __mdd_index_insert(env, tpobj, mdd_object_fid(obj),
3935 attr->la_mode & S_IFMT,
3936 tname->ln_name, handle);
3940 rc = mdd_links_write(env, obj, ldata, handle);
3944 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3945 la->la_valid = LA_CTIME | LA_MTIME;
3946 mdd_write_lock(env, spobj, DT_SRC_PARENT);
3947 rc = mdd_update_time(env, spobj, spattr, la, handle);
3948 mdd_write_unlock(env, spobj);
3952 if (tpobj != spobj) {
3953 la->la_valid = LA_CTIME | LA_MTIME;
3954 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
3955 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
3956 mdd_write_unlock(env, tpobj);
3965 * Migrate file/dir to target MDT.
3967 * Create target according to \a spec, and then migrate xattrs, if it's
3968 * directory, migrate source stripes to target.
3970 * \param[in] env execution environment
3971 * \param[in] spobj source parent object
3972 * \param[in] tpobj target parent object
3973 * \param[in] sobj source object
3974 * \param[in] tobj target object
3975 * \param[in] lname file name
3976 * \param[in] spattr source parent attributes
3977 * \param[in] tpattr target parent attributes
3978 * \param[in] attr source attributes
3979 * \param[in] sbuf source LMV buf
3980 * \param[in] spec migrate create spec
3981 * \param[in] hint target creation hint
3982 * \param[in] handle tranasction handle
3984 * \retval 0 on success
3985 * \retval -errno on failure
3987 static int mdd_migrate_create(const struct lu_env *env,
3988 struct mdd_object *spobj,
3989 struct mdd_object *tpobj,
3990 struct mdd_object *sobj,
3991 struct mdd_object *tobj,
3992 const struct lu_name *sname,
3993 const struct lu_name *tname,
3994 struct lu_attr *spattr,
3995 struct lu_attr *tpattr,
3996 struct lu_attr *attr,
3997 const struct lu_buf *sbuf,
3998 struct linkea_data *ldata,
4000 struct md_op_spec *spec,
4001 struct dt_allocation_hint *hint,
4002 struct thandle *handle)
4009 * migrate sobj stripes to tobj if it's directory:
4010 * 1. detach stripes from sobj.
4011 * 2. attach stripes to tobj, see mdd_declare_migrate_mdt().
4012 * 3. create stripes for tobj, see lod_xattr_set_lmv().
4014 if (S_ISDIR(attr->la_mode)) {
4015 struct mdd_thread_info *info = mdd_env_info(env);
4016 struct md_layout_change *mlc = &info->mti_mlc;
4018 mlc->mlc_opc = MD_LAYOUT_DETACH;
4020 mdd_write_lock(env, sobj, DT_SRC_PARENT);
4021 rc = mdo_layout_change(env, sobj, mlc, handle);
4022 mdd_write_unlock(env, sobj);
4027 /* don't set nlink from sobj */
4028 attr->la_valid &= ~LA_NLINK;
4030 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4031 hint, handle, false);
4035 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4036 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
4037 mdd_write_unlock(env, tobj);
4041 /* for regular file, update OST objects XATTR_NAME_FID */
4042 if (S_ISREG(attr->la_mode)) {
4043 struct lu_buf fid_buf;
4045 /* target may be remote, update PFID via sobj. */
4046 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
4047 fid_buf.lb_len = sizeof(struct lu_fid);
4048 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4053 /* delete LOV to avoid deleting OST objs when destroying sobj */
4054 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4055 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4056 mdd_write_unlock(env, sobj);
4057 /* O_DELAY_CREATE file may not have LOV, ignore -ENODATA */
4058 if (rc && rc != -ENODATA)
4063 /* update links FID */
4064 if (!S_ISDIR(attr->la_mode)) {
4065 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
4066 mdd_object_fid(tpobj), ldata,
4067 NULL, handle, mdd_update_link);
4072 /* don't destroy sobj if it's plain directory */
4073 if (!S_ISDIR(attr->la_mode) || sbuf->lb_buf) {
4074 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4075 rc = mdo_ref_del(env, sobj, handle);
4077 if (S_ISDIR(attr->la_mode))
4078 rc = mdo_ref_del(env, sobj, handle);
4080 rc = mdo_destroy(env, sobj, handle);
4082 mdd_write_unlock(env, sobj);
4087 rc = mdd_migrate_update(env, spobj, tpobj, tobj, sname, tname, attr,
4088 spattr, tpattr, ldata, ma, handle);
4093 /* NB: if user issued different migrate command, we can't adjust it silently
4094 * here, because this command will decide target MDT in subdir migration in
4097 static int mdd_migrate_cmd_check(struct mdd_device *mdd,
4098 const struct lmv_mds_md_v1 *lmv,
4099 const struct lmv_user_md_v1 *lum,
4100 const struct lu_name *lname)
4102 __u32 lum_stripe_count = lum->lum_stripe_count;
4103 __u32 lmv_hash_type = lmv->lmv_hash_type;
4105 if (!lmv_is_sane(lmv))
4108 /* if stripe_count unspecified, set to 1 */
4109 if (!lum_stripe_count)
4110 lum_stripe_count = cpu_to_le32(1);
4112 lmv_hash_type &= ~cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4114 /* TODO: check specific MDTs */
4115 if (lum_stripe_count != lmv->lmv_migrate_offset ||
4116 lum->lum_stripe_offset != lmv->lmv_master_mdt_index ||
4117 (lum->lum_hash_type && lum->lum_hash_type != lmv_hash_type)) {
4118 CERROR("%s: '"DNAME"' migration was interrupted, run 'lfs migrate -m %d -c %d -H %s "DNAME"' to finish migration.\n",
4119 mdd2obd_dev(mdd)->obd_name, PNAME(lname),
4120 le32_to_cpu(lmv->lmv_master_mdt_index),
4121 le32_to_cpu(lmv->lmv_migrate_offset),
4122 mdt_hash_name[le32_to_cpu(lmv_hash_type)],
4131 * Internal function to migrate directory or file between MDTs.
4133 * migrate source to target in following steps:
4134 * 1. create target, append source stripes after target's if it's directory,
4135 * migrate xattrs and update fid of source links.
4136 * 2. update namespace: migrate dirent from source parent to target parent,
4137 * update file linkea, and destroy source if it's not needed any more.
4139 * \param[in] env execution environment
4140 * \param[in] spobj source parent object
4141 * \param[in] tpobj target parent object
4142 * \param[in] sobj source object
4143 * \param[in] tobj target object
4144 * \param[in] sname source file name
4145 * \param[in] tname target file name
4146 * \param[in] spec target creation spec
4147 * \param[in] ma used to update \a pobj mtime and ctime
4149 * \retval 0 on success
4150 * \retval -errno on failure
4152 static int mdd_migrate_object(const struct lu_env *env,
4153 struct mdd_object *spobj,
4154 struct mdd_object *tpobj,
4155 struct mdd_object *sobj,
4156 struct mdd_object *tobj,
4157 const struct lu_name *sname,
4158 const struct lu_name *tname,
4159 struct md_op_spec *spec,
4162 struct mdd_thread_info *info = mdd_env_info(env);
4163 struct mdd_device *mdd = mdo2mdd(&spobj->mod_obj);
4164 struct lu_attr *spattr = &info->mti_pattr;
4165 struct lu_attr *tpattr = &info->mti_tpattr;
4166 struct lu_attr *attr = &info->mti_cattr;
4167 struct linkea_data *ldata = &info->mti_link_data;
4168 struct dt_allocation_hint *hint = &info->mti_hint;
4169 struct lu_buf sbuf = { NULL };
4170 struct lmv_mds_md_v1 *lmv;
4171 struct thandle *handle;
4176 rc = mdd_la_get(env, sobj, attr);
4180 rc = mdd_la_get(env, spobj, spattr);
4184 rc = mdd_la_get(env, tpobj, tpattr);
4188 if (S_ISDIR(attr->la_mode) && !spec->sp_migrate_nsonly) {
4189 struct lmv_user_md_v1 *lum = spec->u.sp_ea.eadata;
4193 /* if user use default value '0' for stripe_count, we need to
4194 * adjust it to '1' to create a 1-stripe directory.
4196 if (lum->lum_stripe_count == 0)
4197 lum->lum_stripe_count = cpu_to_le32(1);
4199 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4200 if (rc && rc != -ENODATA)
4205 if (!lmv_is_sane(lmv))
4206 GOTO(out, rc = -EBADF);
4207 if (lmv_is_migrating(lmv)) {
4208 rc = mdd_migrate_cmd_check(mdd, lmv, lum,
4213 } else if (!S_ISDIR(attr->la_mode)) {
4215 GOTO(out, rc = -EALREADY);
4217 /* update namespace only if @sobj is on MDT where @tpobj is. */
4218 if (!mdd_object_remote(tpobj) && !mdd_object_remote(sobj))
4219 spec->sp_migrate_nsonly = true;
4221 if (S_ISLNK(attr->la_mode)) {
4222 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4224 GOTO(out, rc = -ENOMEM);
4226 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4229 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4230 mdd2obd_dev(mdd)->obd_name,
4231 PFID(mdd_object_fid(sobj)), rc);
4237 /* linkea needs update upon FID or parent stripe change */
4238 rc = mdd_migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, sname,
4239 tname, attr, ldata);
4241 /* update namespace only if @sobj has link on its MDT. */
4242 spec->sp_migrate_nsonly = true;
4246 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4247 spattr, tpattr, attr);
4251 handle = mdd_trans_create(env, mdd);
4253 GOTO(out, rc = PTR_ERR(handle));
4255 if (spec->sp_migrate_nsonly)
4256 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, sname,
4257 tname, attr, spattr, tpattr,
4260 rc = mdd_declare_migrate_create(env, spobj, tpobj, sobj, tobj,
4261 sname, tname, spattr, tpattr,
4262 attr, &sbuf, ldata, ma, spec,
4267 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, tname, sname,
4272 rc = mdd_trans_start(env, mdd, handle);
4276 if (spec->sp_migrate_nsonly)
4277 rc = mdd_migrate_update(env, spobj, tpobj, sobj, sname, tname,
4278 attr, spattr, tpattr, ldata, ma,
4281 rc = mdd_migrate_create(env, spobj, tpobj, sobj, tobj, sname,
4282 tname, spattr, tpattr, attr, &sbuf,
4283 ldata, ma, spec, hint, handle);
4287 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0,
4288 spec->sp_migrate_nsonly ? sobj : tobj,
4289 mdd_object_fid(spobj), mdd_object_fid(sobj),
4290 mdd_object_fid(tpobj), tname, sname,
4297 rc = mdd_trans_stop(env, mdd, rc, handle);
4305 * Migrate directory or file between MDTs.
4307 * \param[in] env execution environment
4308 * \param[in] md_pobj parent master object
4309 * \param[in] md_sobj source object
4310 * \param[in] lname file name
4311 * \param[in] md_tobj target object
4312 * \param[in] spec target creation spec
4313 * \param[in] ma used to update \a pobj mtime and ctime
4315 * \retval 0 on success
4316 * \retval -errno on failure
4318 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4319 struct md_object *md_sobj, const struct lu_name *lname,
4320 struct md_object *md_tobj, struct md_op_spec *spec,
4323 struct mdd_thread_info *info = mdd_env_info(env);
4324 struct mdd_device *mdd = mdo2mdd(md_pobj);
4325 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4326 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4327 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4328 struct mdd_object *spobj = NULL;
4329 struct mdd_object *tpobj = NULL;
4330 struct lu_buf pbuf = { NULL };
4331 struct lu_fid *fid = &info->mti_fid2;
4332 struct lmv_mds_md_v1 *lmv;
4337 /* locate source and target stripe on pobj, which are the real parent */
4338 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4339 if (rc < 0 && rc != -ENODATA)
4346 if (!lmv_is_sane(lmv))
4347 GOTO(out, rc = -EBADF);
4349 /* locate target parent stripe */
4350 /* fail check here to make sure top dir migration succeed. */
4351 if (lmv_is_migrating(lmv) &&
4352 OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4353 GOTO(out, rc = -EIO);
4355 index = lmv_name_to_stripe_index(lmv, lname->ln_name,
4358 GOTO(out, rc = index);
4360 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4361 tpobj = mdd_object_find(env, mdd, fid);
4363 GOTO(out, rc = PTR_ERR(tpobj));
4365 /* locate source parent stripe */
4366 if (lmv_is_layout_changing(lmv)) {
4367 index = lmv_name_to_stripe_index_old(lmv,
4371 GOTO(out, rc = index);
4373 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4374 spobj = mdd_object_find(env, mdd, fid);
4376 GOTO(out, rc = PTR_ERR(spobj));
4378 /* parent stripe unchanged */
4379 if (spobj == tpobj) {
4380 if (!lmv_is_restriping(lmv))
4381 GOTO(out, rc = -EINVAL);
4382 GOTO(out, rc = -EALREADY);
4386 mdd_object_get(spobj);
4391 mdd_object_get(tpobj);
4392 mdd_object_get(spobj);
4395 rc = mdd_migrate_object(env, spobj, tpobj, sobj, tobj, lname, lname,
4400 if (!IS_ERR_OR_NULL(spobj))
4401 mdd_object_put(env, spobj);
4402 if (!IS_ERR_OR_NULL(tpobj))
4403 mdd_object_put(env, tpobj);
4409 static int mdd_declare_1sd_collapse(const struct lu_env *env,
4410 struct mdd_object *pobj,
4411 struct mdd_object *obj,
4412 struct mdd_object *stripe,
4413 struct lu_attr *attr,
4414 struct md_layout_change *mlc,
4415 struct lu_name *lname,
4416 struct thandle *handle)
4420 mlc->mlc_opc = MD_LAYOUT_DETACH;
4421 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4425 rc = mdo_declare_index_insert(env, stripe, mdd_object_fid(pobj),
4426 S_IFDIR, dotdot, handle);
4430 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle,
4431 mdo_declare_xattr_set);
4435 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4439 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4443 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4447 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(stripe),
4448 attr->la_mode, lname->ln_name, handle);
4452 rc = mdo_declare_ref_del(env, obj, handle);
4456 rc = mdo_declare_ref_del(env, obj, handle);
4460 rc = mdo_declare_destroy(env, obj, handle);
4467 /* transform one-stripe directory to a plain directory */
4468 static int mdd_1sd_collapse(const struct lu_env *env,
4469 struct mdd_object *pobj,
4470 struct mdd_object *obj,
4471 struct mdd_object *stripe,
4472 struct lu_attr *attr,
4473 struct md_layout_change *mlc,
4474 struct lu_name *lname,
4475 struct thandle *handle)
4481 /* replace 1-stripe directory with its stripe */
4482 mlc->mlc_opc = MD_LAYOUT_DETACH;
4484 mdd_write_lock(env, obj, DT_SRC_PARENT);
4485 rc = mdo_layout_change(env, obj, mlc, handle);
4486 mdd_write_unlock(env, obj);
4490 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4491 mdd_write_lock(env, obj, DT_SRC_CHILD);
4493 /* insert dotdot to stripe which points to parent */
4494 rc = __mdd_index_insert_only(env, stripe, mdd_object_fid(pobj),
4495 S_IFDIR, dotdot, handle);
4499 /* copy xattrs including linkea */
4500 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle, mdo_xattr_set);
4505 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4509 /* don't set nlink from parent */
4510 attr->la_valid &= ~LA_NLINK;
4512 rc = mdo_attr_set(env, stripe, attr, handle);
4516 /* delete dir name from parent */
4517 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4521 /* insert stripe to parent with dir name */
4522 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(stripe),
4523 attr->la_mode, lname->ln_name, handle);
4527 /* destroy dir obj */
4528 rc = mdo_ref_del(env, obj, handle);
4532 rc = mdo_ref_del(env, obj, handle);
4536 rc = mdo_destroy(env, obj, handle);
4542 mdd_write_unlock(env, obj);
4543 mdd_write_unlock(env, pobj);
4549 * shrink directory stripes after migration/merge
4551 int mdd_dir_layout_shrink(const struct lu_env *env,
4552 struct md_object *md_obj,
4553 struct md_layout_change *mlc)
4555 struct mdd_device *mdd = mdo2mdd(md_obj);
4556 struct mdd_thread_info *info = mdd_env_info(env);
4557 struct mdd_object *obj = md2mdd_obj(md_obj);
4558 struct mdd_object *pobj = NULL;
4559 struct mdd_object *stripe = NULL;
4560 struct lu_attr *attr = &info->mti_pattr;
4561 struct lu_fid *fid = &info->mti_fid2;
4562 struct lu_name lname = { NULL };
4563 struct lu_buf lmv_buf = { NULL };
4564 struct lmv_mds_md_v1 *lmv;
4565 struct lmv_user_md *lmu;
4566 struct thandle *handle;
4571 rc = mdd_la_get(env, obj, attr);
4575 if (!S_ISDIR(attr->la_mode))
4578 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4582 lmv = lmv_buf.lb_buf;
4583 if (!lmv_is_sane(lmv))
4586 lmu = mlc->mlc_buf.lb_buf;
4588 /* adjust the default value '0' to '1' */
4589 if (lmu->lum_stripe_count == 0)
4590 lmu->lum_stripe_count = cpu_to_le32(1);
4592 /* these were checked in MDT */
4593 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4594 le32_to_cpu(lmv->lmv_stripe_count));
4595 LASSERT(!lmv_is_splitting(lmv));
4596 LASSERT(lmv_is_migrating(lmv) || lmv_is_merging(lmv));
4598 /* if dir stripe count will be shrunk to 1, it needs to be transformed
4599 * to a plain dir, which will cause FID change and namespace update.
4601 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4602 struct linkea_data *ldata = &info->mti_link_data;
4603 char *filename = info->mti_name;
4605 rc = mdd_links_read(env, obj, ldata);
4609 if (ldata->ld_leh->leh_reccount > 1)
4610 GOTO(out, rc = -EINVAL);
4612 linkea_first_entry(ldata);
4614 GOTO(out, rc = -ENODATA);
4616 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4619 /* Note: lname might miss \0 at the end */
4620 snprintf(filename, sizeof(info->mti_name), "%.*s",
4621 lname.ln_namelen, lname.ln_name);
4622 lname.ln_name = filename;
4624 pobj = mdd_object_find(env, mdd, fid);
4631 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4633 stripe = mdd_object_find(env, mdd, fid);
4634 if (IS_ERR(stripe)) {
4635 mdd_object_put(env, pobj);
4637 GOTO(out, rc = PTR_ERR(stripe));
4641 handle = mdd_trans_create(env, mdd);
4643 GOTO(out, rc = PTR_ERR(handle));
4645 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4646 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4648 GOTO(stop_trans, rc);
4650 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4651 rc = mdd_declare_1sd_collapse(env, pobj, obj, stripe, attr, mlc,
4654 GOTO(stop_trans, rc);
4657 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4660 GOTO(stop_trans, rc);
4662 rc = mdd_trans_start(env, mdd, handle);
4664 GOTO(stop_trans, rc);
4666 mdd_write_lock(env, obj, DT_SRC_PARENT);
4667 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4668 rc = mdo_layout_change(env, obj, mlc, handle);
4669 mdd_write_unlock(env, obj);
4671 GOTO(stop_trans, rc);
4673 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4674 rc = mdd_1sd_collapse(env, pobj, obj, stripe, attr, mlc, &lname,
4677 GOTO(stop_trans, rc);
4680 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4681 XATTR_NAME_LMV, handle);
4682 GOTO(stop_trans, rc);
4685 rc = mdd_trans_stop(env, mdd, rc, handle);
4688 mdd_object_put(env, stripe);
4689 mdd_object_put(env, pobj);
4691 lu_buf_free(&lmv_buf);
4695 static int mdd_dir_declare_split_plain(const struct lu_env *env,
4696 struct mdd_device *mdd,
4697 struct mdd_object *pobj,
4698 struct mdd_object *obj,
4699 struct mdd_object *tobj,
4700 struct md_layout_change *mlc,
4701 struct dt_allocation_hint *hint,
4702 struct thandle *handle)
4704 struct mdd_thread_info *info = mdd_env_info(env);
4705 const struct lu_name *lname = mlc->mlc_name;
4706 struct lu_attr *la = &info->mti_la_for_fix;
4707 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
4708 struct linkea_data *ldata = &info->mti_link_data;
4709 struct lmv_mds_md_v1 *lmv;
4713 mlc->mlc_opc = MD_LAYOUT_DETACH;
4714 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4718 memset(ldata, 0, sizeof(*ldata));
4719 rc = mdd_linkea_prepare(env, obj, NULL, NULL, mdd_object_fid(pobj),
4720 lname, 1, 0, ldata);
4724 count = lum->lum_stripe_count;
4725 lum->lum_stripe_count = 0;
4726 mdd_object_make_hint(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4728 rc = mdd_declare_create(env, mdo2mdd(&pobj->mod_obj), pobj, tobj,
4729 lname, mlc->mlc_attr, handle, mlc->mlc_spec,
4730 ldata, NULL, NULL, NULL, hint);
4734 /* tobj mode will be used in lod_declare_xattr_set(), but it's not
4737 tobj->mod_obj.mo_lu.lo_header->loh_attr |= S_IFDIR;
4739 lmv = (typeof(lmv))info->mti_key;
4740 memset(lmv, 0, sizeof(*lmv));
4741 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
4742 lmv->lmv_stripe_count = cpu_to_le32(1);
4743 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
4744 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdd_object_fid(obj));
4746 mlc->mlc_opc = MD_LAYOUT_ATTACH;
4747 mlc->mlc_buf.lb_buf = lmv;
4748 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
4749 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4753 rc = mdd_iterate_xattrs(env, obj, tobj, true, handle,
4754 mdo_declare_xattr_set);
4758 lum->lum_stripe_count = count;
4759 mlc->mlc_opc = MD_LAYOUT_SPLIT;
4760 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4764 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4768 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
4769 S_IFDIR, lname->ln_name, handle);
4773 la->la_valid = LA_CTIME | LA_MTIME;
4774 rc = mdo_declare_attr_set(env, obj, la, handle);
4778 rc = mdo_declare_attr_set(env, pobj, la, handle);
4782 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4788 * plain directory split:
4789 * 1. create \a tobj as plain directory.
4790 * 2. append \a obj as first stripe of \a tobj.
4791 * 3. migrate xattrs from \a obj to \a tobj.
4792 * 4. split \a tobj to specific stripe count.
4794 static int mdd_dir_split_plain(const struct lu_env *env,
4795 struct mdd_device *mdd,
4796 struct mdd_object *pobj,
4797 struct mdd_object *obj,
4798 struct mdd_object *tobj,
4799 struct md_layout_change *mlc,
4800 struct dt_allocation_hint *hint,
4801 struct thandle *handle)
4803 struct mdd_thread_info *info = mdd_env_info(env);
4804 struct lu_attr *pattr = &info->mti_pattr;
4805 struct lu_attr *la = &info->mti_la_for_fix;
4806 const struct lu_name *lname = mlc->mlc_name;
4807 struct linkea_data *ldata = &info->mti_link_data;
4812 /* copy linkea out and set on target later */
4813 rc = mdd_links_read(env, obj, ldata);
4817 mlc->mlc_opc = MD_LAYOUT_DETACH;
4818 rc = mdo_layout_change(env, obj, mlc, handle);
4822 /* don't set nlink from obj */
4823 mlc->mlc_attr->la_valid &= ~LA_NLINK;
4825 rc = mdd_create_object(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4826 NULL, NULL, NULL, hint, handle, false);
4830 rc = mdd_iterate_xattrs(env, obj, tobj, true, handle, mdo_xattr_set);
4834 rc = mdd_links_write(env, tobj, ldata, handle);
4838 rc = __mdd_index_delete(env, pobj, lname->ln_name, true, handle);
4842 rc = __mdd_index_insert(env, pobj, mdd_object_fid(tobj), S_IFDIR,
4843 lname->ln_name, handle);
4847 la->la_ctime = la->la_mtime = mlc->mlc_attr->la_mtime;
4848 la->la_valid = LA_CTIME | LA_MTIME;
4850 mdd_write_lock(env, obj, DT_SRC_CHILD);
4851 rc = mdd_update_time(env, tobj, mlc->mlc_attr, la, handle);
4852 mdd_write_unlock(env, obj);
4856 rc = mdd_la_get(env, pobj, pattr);
4860 la->la_valid = LA_CTIME | LA_MTIME;
4862 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4863 rc = mdd_update_time(env, pobj, pattr, la, handle);
4864 mdd_write_unlock(env, pobj);
4868 /* FID changes, record it as CL_MIGRATE */
4869 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
4870 mdd_object_fid(pobj), mdd_object_fid(obj),
4871 mdd_object_fid(pobj), lname, lname, handle);
4875 int mdd_dir_layout_split(const struct lu_env *env, struct md_object *o,
4876 struct md_layout_change *mlc)
4878 struct mdd_thread_info *info = mdd_env_info(env);
4879 struct mdd_device *mdd = mdo2mdd(o);
4880 struct mdd_object *obj = md2mdd_obj(o);
4881 struct mdd_object *pobj = md2mdd_obj(mlc->mlc_parent);
4882 struct mdd_object *tobj = md2mdd_obj(mlc->mlc_target);
4883 struct dt_allocation_hint *hint = &info->mti_hint;
4884 bool is_plain = false;
4885 struct thandle *handle;
4890 LASSERT(S_ISDIR(mdd_object_type(obj)));
4892 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LMV);
4898 handle = mdd_trans_create(env, mdd);
4900 RETURN(PTR_ERR(handle));
4903 rc = mdd_dir_declare_split_plain(env, mdd, pobj, obj, tobj, mlc,
4906 mlc->mlc_opc = MD_LAYOUT_SPLIT;
4907 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4909 GOTO(stop_trans, rc);
4911 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL,
4915 GOTO(stop_trans, rc);
4917 rc = mdd_trans_start(env, mdd, handle);
4919 GOTO(stop_trans, rc);
4922 rc = mdd_dir_split_plain(env, mdd, pobj, obj, tobj, mlc, hint,
4925 mdd_write_lock(env, obj, DT_TGT_CHILD);
4926 rc = mdo_xattr_set(env, obj, NULL, XATTR_NAME_LMV,
4927 LU_XATTR_CREATE, handle);
4928 mdd_write_unlock(env, obj);
4930 GOTO(stop_trans, rc);
4932 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4933 XATTR_NAME_LMV, handle);
4936 GOTO(stop_trans, rc);
4941 rc = mdd_trans_stop(env, mdd, rc, handle);
4946 const struct md_dir_operations mdd_dir_ops = {
4947 .mdo_is_subdir = mdd_is_subdir,
4948 .mdo_lookup = mdd_lookup,
4949 .mdo_create = mdd_create,
4950 .mdo_rename = mdd_rename,
4951 .mdo_link = mdd_link,
4952 .mdo_unlink = mdd_unlink,
4953 .mdo_create_data = mdd_create_data,
4954 .mdo_migrate = mdd_migrate,