4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_dir.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@intel.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <obd_class.h>
42 #include <obd_support.h>
43 #include <lustre_mds.h>
44 #include <lustre_fid.h>
45 #include <lustre_lmv.h>
46 #include <lustre_idmap.h>
48 #include "mdd_internal.h"
50 static const char dot[] = ".";
51 static const char dotdot[] = "..";
53 static struct lu_name lname_dotdot = {
54 .ln_name = (char *) dotdot,
55 .ln_namelen = sizeof(dotdot) - 1,
59 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
61 if (!lu_name_is_valid(ln))
63 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
69 /* Get FID from name and parent */
71 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
72 const struct lu_attr *pattr, const struct lu_name *lname,
73 struct lu_fid* fid, int mask)
75 const char *name = lname->ln_name;
76 const struct dt_key *key = (const struct dt_key *)name;
77 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
78 struct mdd_device *m = mdo2mdd(pobj);
79 struct dt_object *dir = mdd_object_child(mdd_obj);
83 if (unlikely(mdd_is_dead_obj(mdd_obj)))
86 if (!mdd_object_exists(mdd_obj))
89 if (mdd_object_remote(mdd_obj)) {
90 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
91 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
94 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
99 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
100 dt_try_as_dir(env, dir)))
101 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
108 int mdd_lookup(const struct lu_env *env,
109 struct md_object *pobj, const struct lu_name *lname,
110 struct lu_fid *fid, struct md_op_spec *spec)
112 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
116 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
120 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
121 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
125 /** Read the link EA into a temp buffer.
126 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
127 * A pointer to the buffer is stored in \a ldata::ld_buf.
131 static int __mdd_links_read(const struct lu_env *env,
132 struct mdd_object *mdd_obj,
133 struct linkea_data *ldata)
137 if (!mdd_object_exists(mdd_obj))
140 /* First try a small buf */
141 LASSERT(env != NULL);
142 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
144 if (ldata->ld_buf->lb_buf == NULL)
147 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
149 /* Buf was too small, figure out what we need. */
150 lu_buf_free(ldata->ld_buf);
151 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
155 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
156 if (ldata->ld_buf->lb_buf == NULL)
158 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
162 lu_buf_free(ldata->ld_buf);
163 ldata->ld_buf = NULL;
167 return linkea_init(ldata);
170 static int mdd_links_read(const struct lu_env *env,
171 struct mdd_object *mdd_obj,
172 struct linkea_data *ldata)
176 rc = __mdd_links_read(env, mdd_obj, ldata);
178 rc = linkea_init(ldata);
183 static int mdd_links_read_with_rec(const struct lu_env *env,
184 struct mdd_object *mdd_obj,
185 struct linkea_data *ldata)
189 rc = __mdd_links_read(env, mdd_obj, ldata);
191 rc = linkea_init_with_rec(ldata);
197 * Get parent FID of the directory
199 * Read parent FID from linkEA, if that fails, then do lookup
200 * dotdot to get the parent FID.
202 * \param[in] env execution environment
203 * \param[in] obj object from which to find the parent FID
204 * \param[in] attr attribute of the object
205 * \param[out] fid fid to get the parent FID
207 * \retval 0 if getting the parent FID succeeds.
208 * \retval negative errno if getting the parent FID fails.
210 static inline int mdd_parent_fid(const struct lu_env *env,
211 struct mdd_object *obj,
212 const struct lu_attr *attr,
215 struct mdd_thread_info *info = mdd_env_info(env);
216 struct linkea_data ldata = { NULL };
217 struct lu_buf *buf = &info->mti_link_buf;
218 struct lu_name lname;
223 LASSERTF(S_ISDIR(mdd_object_type(obj)),
224 "%s: FID "DFID" is not a directory type = %o\n",
225 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)),
226 mdd_object_type(obj));
228 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
229 if (buf->lb_buf == NULL)
230 GOTO(lookup, rc = 0);
233 rc = mdd_links_read_with_rec(env, obj, &ldata);
237 LASSERT(ldata.ld_leh != NULL);
238 /* Directory should only have 1 parent */
239 if (ldata.ld_leh->leh_reccount > 1)
242 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
244 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
245 if (likely(fid_is_sane(fid)))
248 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
253 * For root fid use special function, which does not compare version component
254 * of fid. Version component is different for root fids on all MDTs.
256 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
258 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
259 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
263 * return 1: if \a tfid is the fid of the ancestor of \a mo;
265 * otherwise: values < 0, errors.
267 static int mdd_is_parent(const struct lu_env *env,
268 struct mdd_device *mdd,
269 struct mdd_object *mo,
270 const struct lu_attr *attr,
271 const struct lu_fid *tfid)
273 struct mdd_object *mp;
277 LASSERT(!lu_fid_eq(mdo2fid(mo), tfid));
278 pfid = &mdd_env_info(env)->mti_fid;
280 if (mdd_is_root(mdd, mdo2fid(mo)))
283 if (mdd_is_root(mdd, tfid))
286 rc = mdd_parent_fid(env, mo, attr, pfid);
291 if (lu_fid_eq(pfid, tfid))
294 if (mdd_is_root(mdd, pfid))
297 mp = mdd_object_find(env, mdd, pfid);
301 if (!mdd_object_exists(mp)) {
302 mdd_object_put(env, mp);
306 rc = mdd_parent_fid(env, mp, attr, pfid);
307 mdd_object_put(env, mp);
316 * No permission check is needed.
318 * returns 1: if fid is ancestor of @mo;
319 * returns 0: if fid is not an ancestor of @mo;
320 * returns < 0: if error
322 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
323 const struct lu_fid *fid)
325 struct mdd_device *mdd = mdo2mdd(mo);
326 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
330 if (!mdd_object_exists(md2mdd_obj(mo)))
333 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
336 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
340 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid);
345 * Check that @dir contains no entries except (possibly) dot and dotdot.
350 * -ENOTDIR not a directory object
351 * -ENOTEMPTY not empty
355 static int mdd_dir_is_empty(const struct lu_env *env,
356 struct mdd_object *dir)
359 struct dt_object *obj;
360 const struct dt_it_ops *iops;
364 obj = mdd_object_child(dir);
365 if (!dt_try_as_dir(env, obj))
368 iops = &obj->do_index_ops->dio_it;
369 it = iops->init(env, obj, LUDA_64BITHASH);
371 result = iops->get(env, it, (const struct dt_key *)"");
374 for (result = 0, i = 0; result == 0 && i < 3; ++i)
375 result = iops->next(env, it);
378 else if (result == 1)
380 } else if (result == 0)
382 * Huh? Index contains no zero key?
389 result = PTR_ERR(it);
390 /* -ENODEV means no valid stripe */
391 if (result == -ENODEV)
398 * Determine if the target object can be hard linked, and right now it only
399 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
400 * directory nlink count might exceed the maximum link count(see
401 * osd_object_ref_add), so it only check nlink for non-directories.
403 * \param[in] env thread environment
404 * \param[in] obj object being linked to
405 * \param[in] la attributes of \a obj
407 * \retval 0 if \a obj can be hard linked
408 * \retval negative error if \a obj is a directory or has too
411 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
412 const struct lu_attr *la)
414 struct mdd_device *m = mdd_obj2mdd_dev(obj);
419 /* Subdir count limitation can be broken through
420 * (see osd_object_ref_add), so only check non-directory here. */
421 if (!S_ISDIR(la->la_mode) &&
422 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
429 * Check whether it may create the cobj under the pobj.
431 * \param[in] env execution environment
432 * \param[in] pobj the parent directory
433 * \param[in] pattr the attribute of the parent directory
434 * \param[in] cobj the child to be created
435 * \param[in] check_perm if check WRITE|EXEC permission for parent
437 * \retval = 0 create the child under this dir is allowed
438 * \retval negative errno create the child under this dir is
441 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
442 const struct lu_attr *pattr, struct mdd_object *cobj,
448 if (cobj && mdd_object_exists(cobj))
451 if (mdd_is_dead_obj(pobj))
455 rc = mdd_permission_internal_locked(env, pobj, pattr,
456 MAY_WRITE | MAY_EXEC,
462 * Check whether can unlink from the pobj in the case of "cobj == NULL".
464 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
465 const struct lu_attr *pattr, const struct lu_attr *attr)
470 if (mdd_is_dead_obj(pobj))
473 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
476 rc = mdd_permission_internal_locked(env, pobj, pattr,
477 MAY_WRITE | MAY_EXEC,
482 if (pattr->la_flags & LUSTRE_APPEND_FL)
489 * pobj == NULL is remote ops case, under such case, pobj's
490 * VTX feature has been checked already, no need check again.
492 static inline int mdd_is_sticky(const struct lu_env *env,
493 struct mdd_object *pobj,
494 const struct lu_attr *pattr,
495 struct mdd_object *cobj,
496 const struct lu_attr *cattr)
498 struct lu_ucred *uc = lu_ucred_assert(env);
501 LASSERT(pattr != NULL);
502 if (!(pattr->la_mode & S_ISVTX) ||
503 (pattr->la_uid == uc->uc_fsuid))
507 LASSERT(cattr != NULL);
508 if (cattr->la_uid == uc->uc_fsuid)
511 return !md_capable(uc, CFS_CAP_FOWNER);
514 static int mdd_may_delete_entry(const struct lu_env *env,
515 struct mdd_object *pobj,
516 const struct lu_attr *pattr,
521 LASSERT(pobj != NULL);
522 if (!mdd_object_exists(pobj))
525 if (mdd_is_dead_obj(pobj))
530 rc = mdd_permission_internal_locked(env, pobj, pattr,
531 MAY_WRITE | MAY_EXEC,
537 if (pattr->la_flags & LUSTRE_APPEND_FL)
544 * Check whether it may delete the cobj from the pobj.
547 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
548 const struct lu_attr *tpattr, struct mdd_object *tobj,
549 const struct lu_attr *tattr, const struct lu_attr *cattr,
550 int check_perm, int check_empty)
556 LASSERT(tpattr != NULL);
557 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
565 if (!mdd_object_exists(tobj))
568 if (mdd_is_dead_obj(tobj))
571 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
574 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
577 /* additional check the rename case */
579 if (S_ISDIR(cattr->la_mode)) {
580 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
582 if (!S_ISDIR(tattr->la_mode))
585 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
587 } else if (S_ISDIR(tattr->la_mode))
591 if (S_ISDIR(tattr->la_mode) && check_empty)
592 rc = mdd_dir_is_empty(env, tobj);
598 * Check whether it can create the link file(linked to @src_obj) under
599 * the target directory(@tgt_obj), and src_obj has been locked by
602 * \param[in] env execution environment
603 * \param[in] tgt_obj the target directory
604 * \param[in] tattr attributes of target directory
605 * \param[in] lname the link name
606 * \param[in] src_obj source object for link
607 * \param[in] cattr attributes for source object
609 * \retval = 0 it is allowed to create the link file under tgt_obj
610 * \retval negative error not allowed to create the link file
612 static int mdd_link_sanity_check(const struct lu_env *env,
613 struct mdd_object *tgt_obj,
614 const struct lu_attr *tattr,
615 const struct lu_name *lname,
616 struct mdd_object *src_obj,
617 const struct lu_attr *cattr)
619 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
623 if (!mdd_object_exists(src_obj))
626 if (mdd_is_dead_obj(src_obj))
629 /* Local ops, no lookup before link, check filename length here. */
630 rc = mdd_name_check(m, lname);
634 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
637 if (S_ISDIR(mdd_object_type(src_obj)))
640 LASSERT(src_obj != tgt_obj);
641 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
645 rc = __mdd_may_link(env, src_obj, cattr);
650 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
651 const char *name, struct thandle *handle)
653 struct dt_object *next = mdd_object_child(pobj);
657 if (dt_try_as_dir(env, next))
658 rc = dt_delete(env, next, (struct dt_key *)name, handle);
665 static int __mdd_index_insert_only(const struct lu_env *env,
666 struct mdd_object *pobj,
667 const struct lu_fid *lf, __u32 type,
669 struct thandle *handle)
671 struct dt_object *next = mdd_object_child(pobj);
675 if (dt_try_as_dir(env, next)) {
676 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
679 rec->rec_type = type;
680 rc = dt_insert(env, next, (const struct dt_rec *)rec,
681 (const struct dt_key *)name, handle);
688 /* insert named index, add reference if isdir */
689 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
690 const struct lu_fid *lf, __u32 type,
691 const char *name, struct thandle *handle)
696 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
697 if (rc == 0 && S_ISDIR(type)) {
698 mdd_write_lock(env, pobj, DT_TGT_PARENT);
699 mdo_ref_add(env, pobj, handle);
700 mdd_write_unlock(env, pobj);
706 /* delete named index, drop reference if isdir */
707 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
708 const char *name, int is_dir,
709 struct thandle *handle)
714 rc = __mdd_index_delete_only(env, pobj, name, handle);
715 if (rc == 0 && is_dir) {
716 mdd_write_lock(env, pobj, DT_TGT_PARENT);
717 mdo_ref_del(env, pobj, handle);
718 mdd_write_unlock(env, pobj);
724 static int mdd_llog_record_calc_size(const struct lu_env *env,
725 const struct lu_name *tname,
726 const struct lu_name *sname)
728 const struct lu_ucred *uc = lu_ucred(env);
729 enum changelog_rec_flags clf_flags = CLF_EXTRA_FLAGS;
730 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
733 clf_flags |= CLF_RENAME;
735 if (uc != NULL && uc->uc_jobid[0] != '\0')
736 clf_flags |= CLF_JOBID;
738 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
739 changelog_rec_offset(clf_flags, crfe) +
740 (tname != NULL ? tname->ln_namelen : 0) +
741 (sname != NULL ? 1 + sname->ln_namelen : 0));
744 int mdd_declare_changelog_store(const struct lu_env *env,
745 struct mdd_device *mdd,
746 enum changelog_rec_type type,
747 const struct lu_name *tname,
748 const struct lu_name *sname,
749 struct thandle *handle)
751 struct obd_device *obd = mdd2obd_dev(mdd);
752 struct llog_ctxt *ctxt;
753 struct llog_changelog_rec *rec;
755 struct thandle *llog_th;
759 if (!mdd_changelog_enabled(env, mdd, type))
762 reclen = mdd_llog_record_calc_size(env, tname, sname);
763 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
764 if (buf->lb_buf == NULL)
768 rec->cr_hdr.lrh_len = reclen;
769 rec->cr_hdr.lrh_type = CHANGELOG_REC;
771 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
775 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
777 GOTO(out_put, rc = PTR_ERR(llog_th));
779 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
787 int mdd_changelog_write_rec(const struct lu_env *env,
788 struct llog_handle *loghandle,
789 struct llog_rec_hdr *r,
790 struct llog_cookie *cookie,
791 int idx, struct thandle *th)
795 if (r->lrh_type == CHANGELOG_REC) {
796 struct mdd_device *mdd;
797 struct llog_changelog_rec *rec;
799 mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
800 rec = container_of0(r, struct llog_changelog_rec, cr_hdr);
802 spin_lock(&mdd->mdd_cl.mc_lock);
803 rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
804 spin_unlock(&mdd->mdd_cl.mc_lock);
806 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
810 * if current llog is full, we will generate a new
811 * llog, and since it's actually not an error, let's
812 * avoid increasing index so that userspace apps
813 * should not see a gap in the changelog sequence
815 if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
816 spin_lock(&mdd->mdd_cl.mc_lock);
817 ++mdd->mdd_cl.mc_index;
818 spin_unlock(&mdd->mdd_cl.mc_lock);
821 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
828 /** Add a changelog entry \a rec to the changelog llog
831 * \param handle - currently ignored since llogs start their own transaction;
832 * this will hopefully be fixed in llog rewrite
835 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
836 struct llog_changelog_rec *rec, struct thandle *th)
838 struct obd_device *obd = mdd2obd_dev(mdd);
839 struct llog_ctxt *ctxt;
840 struct thandle *llog_th;
843 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
844 changelog_rec_varsize(&rec->cr));
846 /* llog_lvfs_write_rec sets the llog tail len */
847 rec->cr_hdr.lrh_type = CHANGELOG_REC;
848 rec->cr.cr_time = cl_time();
850 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
854 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
856 GOTO(out_put, rc = PTR_ERR(llog_th));
858 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
859 /* nested journal transaction */
860 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
862 /* time to recover some space ?? */
863 if (likely(!mdd->mdd_changelog_gc ||
864 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
865 mdd->mdd_changelog_min_gc_interval >=
866 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
867 /* save a spin_lock trip */
869 spin_lock(&mdd->mdd_cl.mc_lock);
870 if (likely(mdd->mdd_changelog_gc &&
871 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
872 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
873 mdd->mdd_changelog_min_gc_interval)) {
874 if (unlikely(llog_cat_free_space(ctxt->loc_handle) <=
875 mdd->mdd_changelog_min_free_cat_entries ||
876 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD))) {
877 CWARN("%s:%s low on changelog_catalog free entries, "
878 "starting ChangeLog garbage collection thread\n",
880 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
883 /* indicate further kthread run will occur outside
884 * right after current journal transaction filling has
887 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
889 /* next check in mdd_changelog_min_gc_interval anyway
891 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
893 spin_unlock(&mdd->mdd_cl.mc_lock);
901 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
902 const struct lu_fid *sfid,
903 const struct lu_fid *spfid,
904 const struct lu_name *sname)
906 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
909 LASSERT(sfid != NULL);
910 LASSERT(spfid != NULL);
911 LASSERT(sname != NULL);
913 extsize = sname->ln_namelen + 1;
915 rnm->cr_sfid = *sfid;
916 rnm->cr_spfid = *spfid;
918 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
919 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
920 rec->cr_namelen += extsize;
923 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
925 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
927 if (jobid == NULL || jobid[0] == '\0')
930 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
933 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
935 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
937 ef->cr_extra_flags = eflags;
940 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
941 __u64 uid, __u64 gid)
943 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
945 uidgid->cr_uid = uid;
946 uidgid->cr_gid = gid;
949 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
952 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
957 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
959 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
961 omd->cr_openflags = flags;
964 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
965 const char *xattr_name)
967 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
969 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
972 /** Store a namespace change changelog record
973 * If this fails, we must fail the whole transaction; we don't
974 * want the change to commit without the log entry.
975 * \param target - mdd_object of change
976 * \param tpfid - target parent dir/object fid
977 * \param sfid - source object fid
978 * \param spfid - source parent fid
979 * \param tname - target name string
980 * \param sname - source name string
981 * \param handle - transaction handle
983 int mdd_changelog_ns_store(const struct lu_env *env,
984 struct mdd_device *mdd,
985 enum changelog_rec_type type,
986 enum changelog_rec_flags clf_flags,
987 struct mdd_object *target,
988 const struct lu_fid *tpfid,
989 const struct lu_fid *sfid,
990 const struct lu_fid *spfid,
991 const struct lu_name *tname,
992 const struct lu_name *sname,
993 struct thandle *handle)
995 const struct lu_ucred *uc = lu_ucred(env);
996 struct llog_changelog_rec *rec;
999 __u64 xflags = CLFE_INVALID;
1003 if (!mdd_changelog_enabled(env, mdd, type))
1006 LASSERT(tpfid != NULL);
1007 LASSERT(tname != NULL);
1008 LASSERT(handle != NULL);
1010 reclen = mdd_llog_record_calc_size(env, tname, sname);
1011 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1012 if (buf->lb_buf == NULL)
1016 clf_flags &= CLF_FLAGMASK;
1017 clf_flags |= CLF_EXTRA_FLAGS;
1020 if (uc->uc_jobid[0] != '\0')
1021 clf_flags |= CLF_JOBID;
1022 xflags |= CLFE_UIDGID;
1027 clf_flags |= CLF_RENAME;
1029 clf_flags |= CLF_VERSION;
1031 rec->cr.cr_flags = clf_flags;
1033 if (clf_flags & CLF_EXTRA_FLAGS) {
1034 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1035 if (xflags & CLFE_UIDGID)
1036 mdd_changelog_rec_extra_uidgid(&rec->cr,
1037 uc->uc_uid, uc->uc_gid);
1038 if (xflags & CLFE_NID)
1039 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1042 rec->cr.cr_type = (__u32)type;
1043 rec->cr.cr_pfid = *tpfid;
1044 rec->cr.cr_namelen = tname->ln_namelen;
1045 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1047 if (clf_flags & CLF_RENAME)
1048 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1050 if (clf_flags & CLF_JOBID)
1051 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1053 if (likely(target != NULL)) {
1054 rec->cr.cr_tfid = *mdo2fid(target);
1055 target->mod_cltime = ktime_get();
1057 fid_zero(&rec->cr.cr_tfid);
1060 rc = mdd_changelog_store(env, mdd, rec, handle);
1062 CERROR("%s: cannot store changelog record: type = %d, "
1063 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1064 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1065 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1072 static int __mdd_links_add(const struct lu_env *env,
1073 struct mdd_object *mdd_obj,
1074 struct linkea_data *ldata,
1075 const struct lu_name *lname,
1076 const struct lu_fid *pfid,
1077 int first, int check)
1081 if (ldata->ld_leh == NULL) {
1082 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1086 rc = linkea_data_new(ldata,
1087 &mdd_env_info(env)->mti_link_buf);
1094 rc = linkea_links_find(ldata, lname, pfid);
1095 if (rc && rc != -ENOENT)
1101 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1102 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1106 linkea_add_buf(ldata, lname, tfid);
1109 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1110 linkea_add_buf(ldata, lname, pfid);
1112 return linkea_add_buf(ldata, lname, pfid);
1115 static int __mdd_links_del(const struct lu_env *env,
1116 struct mdd_object *mdd_obj,
1117 struct linkea_data *ldata,
1118 const struct lu_name *lname,
1119 const struct lu_fid *pfid)
1123 if (ldata->ld_leh == NULL) {
1124 rc = mdd_links_read(env, mdd_obj, ldata);
1129 rc = linkea_links_find(ldata, lname, pfid);
1133 linkea_del_buf(ldata, lname);
1137 static int mdd_linkea_prepare(const struct lu_env *env,
1138 struct mdd_object *mdd_obj,
1139 const struct lu_fid *oldpfid,
1140 const struct lu_name *oldlname,
1141 const struct lu_fid *newpfid,
1142 const struct lu_name *newlname,
1143 int first, int check,
1144 struct linkea_data *ldata)
1149 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1152 LASSERT(oldpfid != NULL || newpfid != NULL);
1154 if (mdd_obj->mod_flags & DEAD_OBJ)
1155 /* Unnecessary to update linkEA for dead object. */
1158 if (oldpfid != NULL) {
1159 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1161 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1164 /* No changes done. */
1169 /* If renaming, add the new record */
1170 if (newpfid != NULL)
1171 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1177 int mdd_links_rename(const struct lu_env *env,
1178 struct mdd_object *mdd_obj,
1179 const struct lu_fid *oldpfid,
1180 const struct lu_name *oldlname,
1181 const struct lu_fid *newpfid,
1182 const struct lu_name *newlname,
1183 struct thandle *handle,
1184 struct linkea_data *ldata,
1185 int first, int check)
1190 if (ldata == NULL) {
1191 ldata = &mdd_env_info(env)->mti_link_data;
1192 memset(ldata, 0, sizeof(*ldata));
1193 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1194 newpfid, newlname, first, check, ldata);
1199 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1200 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1206 if (newlname == NULL)
1207 CERROR("link_ea add failed %d "DFID"\n",
1208 rc, PFID(mdd_object_fid(mdd_obj)));
1209 else if (oldpfid == NULL)
1210 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1211 newlname->ln_namelen, newlname->ln_name, rc,
1212 PFID(mdd_object_fid(mdd_obj)));
1213 else if (newpfid == NULL)
1214 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1215 oldlname->ln_namelen, oldlname->ln_name, rc,
1216 PFID(mdd_object_fid(mdd_obj)));
1218 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1219 "\n", oldlname->ln_namelen, oldlname->ln_name,
1220 newlname->ln_namelen, newlname->ln_name, rc,
1221 PFID(mdd_object_fid(mdd_obj)));
1224 if (is_vmalloc_addr(ldata->ld_buf))
1225 /* if we vmalloced a large buffer drop it */
1226 lu_buf_free(ldata->ld_buf);
1231 static inline int mdd_links_add(const struct lu_env *env,
1232 struct mdd_object *mdd_obj,
1233 const struct lu_fid *pfid,
1234 const struct lu_name *lname,
1235 struct thandle *handle,
1236 struct linkea_data *ldata, int first)
1238 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1239 pfid, lname, handle, ldata, first, 0);
1242 static inline int mdd_links_del(const struct lu_env *env,
1243 struct mdd_object *mdd_obj,
1244 const struct lu_fid *pfid,
1245 const struct lu_name *lname,
1246 struct thandle *handle)
1248 return mdd_links_rename(env, mdd_obj, pfid, lname,
1249 NULL, NULL, handle, NULL, 0, 0);
1252 /** Read the link EA into a temp buffer.
1253 * Uses the name_buf since it is generally large.
1254 * \retval IS_ERR err
1255 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1257 struct lu_buf *mdd_links_get(const struct lu_env *env,
1258 struct mdd_object *mdd_obj)
1260 struct linkea_data ldata = { NULL };
1263 rc = mdd_links_read(env, mdd_obj, &ldata);
1264 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1267 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1268 struct linkea_data *ldata, struct thandle *handle)
1270 const struct lu_buf *buf;
1273 if (ldata == NULL || ldata->ld_buf == NULL ||
1274 ldata->ld_leh == NULL)
1277 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1281 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1282 ldata->ld_leh->leh_len);
1283 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1284 if (unlikely(rc == -ENOSPC)) {
1285 rc = linkea_overflow_shrink(ldata);
1293 static int mdd_declare_links_add(const struct lu_env *env,
1294 struct mdd_object *mdd_obj,
1295 struct thandle *handle,
1296 struct linkea_data *ldata)
1302 if (ldata != NULL && ldata->ld_leh != NULL) {
1303 ea_len = ldata->ld_leh->leh_len;
1304 linkea = ldata->ld_buf->lb_buf;
1306 ea_len = MAX_LINKEA_SIZE;
1310 rc = mdo_declare_xattr_set(env, mdd_obj,
1311 mdd_buf_get_const(env, linkea, ea_len),
1312 XATTR_NAME_LINK, 0, handle);
1317 static inline int mdd_declare_links_del(const struct lu_env *env,
1318 struct mdd_object *c,
1319 struct thandle *handle)
1323 /* For directory, the linkEA will be removed together
1324 * with the object. */
1325 if (!S_ISDIR(mdd_object_type(c)))
1326 rc = mdd_declare_links_add(env, c, handle, NULL);
1331 static int mdd_declare_link(const struct lu_env *env,
1332 struct mdd_device *mdd,
1333 struct mdd_object *p,
1334 struct mdd_object *c,
1335 const struct lu_name *name,
1336 struct thandle *handle,
1338 struct linkea_data *data)
1340 struct lu_fid tfid = *mdo2fid(c);
1343 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1344 tfid.f_oid = cfs_fail_val;
1346 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1347 name->ln_name, handle);
1351 rc = mdo_declare_ref_add(env, c, handle);
1355 la->la_valid = LA_CTIME | LA_MTIME;
1356 rc = mdo_declare_attr_set(env, p, la, handle);
1360 la->la_valid = LA_CTIME;
1361 rc = mdo_declare_attr_set(env, c, la, handle);
1365 rc = mdd_declare_links_add(env, c, handle, data);
1369 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1375 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1376 struct md_object *src_obj, const struct lu_name *lname,
1379 const char *name = lname->ln_name;
1380 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1381 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1382 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1383 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1384 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1385 struct mdd_device *mdd = mdo2mdd(src_obj);
1386 struct thandle *handle;
1387 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1388 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1392 rc = mdd_la_get(env, mdd_sobj, cattr);
1396 rc = mdd_la_get(env, mdd_tobj, tattr);
1401 * If we are using project inheritance, we only allow hard link
1402 * creation in our tree when the project IDs are the same;
1403 * otherwise the tree quota mechanism could be circumvented.
1405 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1406 (tattr->la_projid != cattr->la_projid))
1409 handle = mdd_trans_create(env, mdd);
1411 GOTO(out_pending, rc = PTR_ERR(handle));
1413 memset(ldata, 0, sizeof(*ldata));
1415 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1416 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1418 /* Note: even this function will change ldata, but it comes from
1419 * thread_info, which is completely temporary and only seen in
1420 * this function, so we do not need reset ldata once it fails.*/
1421 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1422 lname, 0, 0, ldata);
1426 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1431 rc = mdd_trans_start(env, mdd, handle);
1435 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1436 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1439 GOTO(out_unlock, rc);
1441 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1442 rc = mdo_ref_add(env, mdd_sobj, handle);
1444 GOTO(out_unlock, rc);
1447 *tfid = *mdo2fid(mdd_sobj);
1448 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1449 tfid->f_oid = cfs_fail_val;
1451 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1452 mdd_object_type(mdd_sobj), name, handle);
1454 mdo_ref_del(env, mdd_sobj, handle);
1455 GOTO(out_unlock, rc);
1458 la->la_valid = LA_CTIME | LA_MTIME;
1459 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1461 GOTO(out_unlock, rc);
1463 la->la_valid = LA_CTIME;
1464 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1466 /* Note: The failure of links_add should not cause the
1467 * link failure, so do not check return value. */
1468 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1469 lname, handle, ldata, 0);
1473 mdd_write_unlock(env, mdd_sobj);
1475 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1476 mdo2fid(mdd_tobj), NULL, NULL,
1477 lname, NULL, handle);
1479 rc = mdd_trans_stop(env, mdd, rc, handle);
1480 if (is_vmalloc_addr(ldata->ld_buf))
1481 /* if we vmalloced a large buffer drop it */
1482 lu_buf_free(ldata->ld_buf);
1487 static int mdd_mark_orphan_object(const struct lu_env *env,
1488 struct mdd_object *obj, struct thandle *handle,
1491 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1494 attr->la_valid = LA_FLAGS;
1495 attr->la_flags = LUSTRE_ORPHAN_FL;
1498 rc = mdo_declare_attr_set(env, obj, attr, handle);
1500 rc = mdo_attr_set(env, obj, attr, handle);
1505 static int mdd_declare_finish_unlink(const struct lu_env *env,
1506 struct mdd_object *obj,
1507 struct thandle *handle)
1511 /* Sigh, we do not know if the unlink object will become orphan in
1512 * declare phase, but fortunately the flags here does not matter
1513 * in current declare implementation */
1514 rc = mdd_mark_orphan_object(env, obj, handle, true);
1518 rc = mdo_declare_destroy(env, obj, handle);
1522 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1526 return mdd_declare_links_del(env, obj, handle);
1529 /* caller should take a lock before calling */
1530 int mdd_finish_unlink(const struct lu_env *env,
1531 struct mdd_object *obj, struct md_attr *ma,
1532 const struct mdd_object *pobj,
1533 const struct lu_name *lname,
1537 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1540 LASSERT(mdd_write_locked(env, obj) != 0);
1542 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1543 /* add new orphan and the object
1544 * will be deleted during mdd_close() */
1545 obj->mod_flags |= DEAD_OBJ;
1546 if (obj->mod_count) {
1547 rc = mdd_orphan_insert(env, obj, th);
1549 CDEBUG(D_HA, "Object "DFID" is inserted into "
1550 "orphan list, open count = %d\n",
1551 PFID(mdd_object_fid(obj)),
1554 CERROR("Object "DFID" fail to be an orphan, "
1555 "open count = %d, maybe cause failed "
1557 PFID(mdd_object_fid(obj)),
1560 /* mark object as an orphan here, not
1561 * before mdd_orphan_insert() as racing
1562 * mdd_la_get() may propagate ORPHAN_OBJ
1563 * causing the asserition */
1564 rc = mdd_mark_orphan_object(env, obj, th, false);
1566 rc = mdo_destroy(env, obj, th);
1568 } else if (!is_dir) {
1569 /* old files may not have link ea; ignore errors */
1570 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1578 * has mdd_write_lock on cobj already, but not on pobj yet
1580 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1581 const struct lu_attr *pattr,
1582 struct mdd_object *cobj,
1583 const struct lu_attr *cattr)
1588 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1593 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1594 struct mdd_object *p, struct mdd_object *c,
1595 const struct lu_name *name, struct md_attr *ma,
1596 struct thandle *handle, int no_name, int is_dir)
1598 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1601 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1602 if (likely(no_name == 0)) {
1603 rc = mdo_declare_index_delete(env, p, name->ln_name,
1610 rc = mdo_declare_ref_del(env, p, handle);
1616 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1617 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1618 la->la_valid = LA_CTIME | LA_MTIME;
1619 rc = mdo_declare_attr_set(env, p, la, handle);
1624 rc = mdo_declare_ref_del(env, c, handle);
1628 rc = mdo_declare_ref_del(env, c, handle);
1632 la->la_valid = LA_CTIME;
1633 rc = mdo_declare_attr_set(env, c, la, handle);
1637 rc = mdd_declare_finish_unlink(env, c, handle);
1641 /* FIXME: need changelog for remove entry */
1642 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1650 * test if a file has an HSM archive
1651 * if HSM attributes are not found in ma update them from
1654 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1655 struct mdd_object *obj,
1660 if (!(ma->ma_valid & MA_HSM)) {
1661 /* no HSM MD provided, read xattr */
1662 struct lu_buf *hsm_buf;
1663 const size_t buflen = sizeof(struct hsm_attrs);
1666 hsm_buf = mdd_buf_get(env, NULL, 0);
1667 lu_buf_alloc(hsm_buf, buflen);
1668 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1669 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1670 lu_buf_free(hsm_buf);
1674 ma->ma_valid |= MA_HSM;
1676 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1682 * Delete name entry and the object.
1683 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1684 * does not exist for this object, and it could only happen during resending
1685 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1686 * is also needed in this case(needed by changelog), so we have to add another
1687 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1688 * the ENOENT failure should be able to be fixed by redo mechanism.
1690 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1691 struct md_object *cobj, const struct lu_name *lname,
1692 struct md_attr *ma, int no_name)
1694 const char *name = lname->ln_name;
1695 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1696 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1697 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1698 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1699 struct mdd_object *mdd_cobj = NULL;
1700 struct mdd_device *mdd = mdo2mdd(pobj);
1701 struct thandle *handle;
1702 int rc, is_dir = 0, cl_flags = 0;
1705 /* let shutdown to start */
1706 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1708 /* cobj == NULL means only delete name entry */
1709 if (likely(cobj != NULL)) {
1710 mdd_cobj = md2mdd_obj(cobj);
1711 if (mdd_object_exists(mdd_cobj) == 0)
1715 rc = mdd_la_get(env, mdd_pobj, pattr);
1719 if (likely(mdd_cobj != NULL)) {
1721 rc = mdd_la_get(env, mdd_cobj, cattr);
1725 is_dir = S_ISDIR(cattr->la_mode);
1726 /* search for an existing archive.
1727 * we should check ahead as the object
1728 * can be destroyed in this transaction */
1729 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1730 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1733 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1737 handle = mdd_trans_create(env, mdd);
1739 RETURN(PTR_ERR(handle));
1741 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1742 lname, ma, handle, no_name, is_dir);
1746 rc = mdd_trans_start(env, mdd, handle);
1750 if (likely(mdd_cobj != NULL))
1751 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1753 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1754 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1759 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1760 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1761 GOTO(cleanup, rc = 0);
1763 if (likely(mdd_cobj != NULL)) {
1764 rc = mdo_ref_del(env, mdd_cobj, handle);
1766 __mdd_index_insert_only(env, mdd_pobj,
1768 mdd_object_type(mdd_cobj),
1775 mdo_ref_del(env, mdd_cobj, handle);
1777 /* fetch updated nlink */
1778 rc = mdd_la_get(env, mdd_cobj, cattr);
1783 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1784 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1786 la->la_valid = LA_CTIME | LA_MTIME;
1787 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1791 /* Enough for only unlink the entry */
1792 if (unlikely(mdd_cobj == NULL))
1795 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1796 /* update ctime of an unlinked file only if it is still
1797 * opened or a link still exists */
1798 la->la_valid = LA_CTIME;
1799 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1804 /* XXX: this transfer to ma will be removed with LOD/OSP */
1805 ma->ma_attr = *cattr;
1806 ma->ma_valid |= MA_INODE;
1807 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1811 /* fetch updated nlink */
1812 rc = mdd_la_get(env, mdd_cobj, cattr);
1813 /* if object is removed then we can't get its attrs,
1815 if (rc == -ENOENT) {
1816 cattr->la_nlink = 0;
1820 if (cattr->la_nlink == 0) {
1821 ma->ma_attr = *cattr;
1822 ma->ma_valid |= MA_INODE;
1827 if (likely(mdd_cobj != NULL))
1828 mdd_write_unlock(env, mdd_cobj);
1831 if (cattr->la_nlink == 0)
1832 cl_flags |= CLF_UNLINK_LAST;
1834 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1836 rc = mdd_changelog_ns_store(env, mdd,
1837 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1838 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1843 rc = mdd_trans_stop(env, mdd, rc, handle);
1849 * The permission has been checked when obj created, no need check again.
1851 static int mdd_cd_sanity_check(const struct lu_env *env,
1852 struct mdd_object *obj)
1857 if (!obj || mdd_is_dead_obj(obj))
1863 static int mdd_create_data(const struct lu_env *env,
1864 struct md_object *pobj,
1865 struct md_object *cobj,
1866 const struct md_op_spec *spec,
1869 struct mdd_device *mdd = mdo2mdd(cobj);
1870 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1871 struct mdd_object *son = md2mdd_obj(cobj);
1872 struct thandle *handle;
1873 const struct lu_buf *buf;
1874 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1875 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1879 rc = mdd_cd_sanity_check(env, son);
1883 if (!md_should_create(spec->sp_cr_flags))
1887 * there are following use cases for this function:
1888 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1889 * striping can be specified or not
1892 rc = mdd_la_get(env, son, attr);
1896 /* calling ->ah_make_hint() is used to transfer information from parent */
1897 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1899 handle = mdd_trans_create(env, mdd);
1901 GOTO(out_free, rc = PTR_ERR(handle));
1904 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1905 * Should this be fixed?
1907 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1908 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1909 spec->sp_cr_flags, spec->no_create);
1911 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1912 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1913 spec->u.sp_ea.eadatalen);
1918 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1919 XATTR_NAME_LOV, 0, handle);
1923 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1928 rc = mdd_trans_start(env, mdd, handle);
1932 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1938 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1941 rc = mdd_trans_stop(env, mdd, rc, handle);
1947 static int mdd_declare_object_initialize(const struct lu_env *env,
1948 struct mdd_object *parent,
1949 struct mdd_object *child,
1950 const struct lu_attr *attr,
1951 struct thandle *handle)
1956 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1957 if (!S_ISDIR(attr->la_mode))
1960 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1965 rc = mdo_declare_ref_add(env, child, handle);
1969 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1975 static int mdd_object_initialize(const struct lu_env *env,
1976 const struct lu_fid *pfid,
1977 struct mdd_object *child,
1978 struct lu_attr *attr,
1979 struct thandle *handle)
1984 if (S_ISDIR(attr->la_mode)) {
1985 /* Add "." and ".." for newly created dir */
1986 mdo_ref_add(env, child, handle);
1987 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1988 S_IFDIR, dot, handle);
1990 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1993 mdo_ref_del(env, child, handle);
2000 * This function checks whether it can create a file/dir under the
2001 * directory(@pobj). The directory(@pobj) is not being locked by
2004 * \param[in] env execution environment
2005 * \param[in] pobj the directory to create files
2006 * \param[in] pattr the attributes of the directory
2007 * \param[in] lname the name of the created file/dir
2008 * \param[in] cattr the attributes of the file/dir
2009 * \param[in] spec create specification
2011 * \retval = 0 it is allowed to create file/dir under
2013 * \retval negative error not allowed to create file/dir
2014 * under the directory
2016 static int mdd_create_sanity_check(const struct lu_env *env,
2017 struct md_object *pobj,
2018 const struct lu_attr *pattr,
2019 const struct lu_name *lname,
2020 struct lu_attr *cattr,
2021 struct md_op_spec *spec)
2023 struct mdd_thread_info *info = mdd_env_info(env);
2024 struct lu_fid *fid = &info->mti_fid;
2025 struct mdd_object *obj = md2mdd_obj(pobj);
2026 struct mdd_device *m = mdo2mdd(pobj);
2027 bool check_perm = true;
2032 if (mdd_is_dead_obj(obj))
2036 * In some cases this lookup is not needed - we know before if name
2037 * exists or not because MDT performs lookup for it.
2038 * name length check is done in lookup.
2040 if (spec->sp_cr_lookup) {
2042 * Check if the name already exist, though it will be checked in
2043 * _index_insert also, for avoiding rolling back if exists
2046 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2047 MAY_WRITE | MAY_EXEC);
2049 RETURN(rc ? : -EEXIST);
2051 /* Permission is already being checked in mdd_lookup */
2055 if (S_ISDIR(cattr->la_mode) &&
2056 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2057 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2058 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2060 if (!lmv_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2061 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2063 CERROR("%s: invalid lmv_user_md: magic = %x, "
2064 "stripe_offset = %d, stripe_count = %u: "
2065 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2066 le32_to_cpu(lum->lum_magic),
2067 (int)le32_to_cpu(lum->lum_stripe_offset),
2068 le32_to_cpu(lum->lum_stripe_count), rc);
2073 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2078 if (pattr->la_mode & S_ISGID) {
2079 struct lu_ucred *uc = lu_ucred(env);
2081 cattr->la_gid = pattr->la_gid;
2083 /* Directories are special, and always inherit S_ISGID */
2084 if (S_ISDIR(cattr->la_mode)) {
2085 cattr->la_mode |= S_ISGID;
2086 cattr->la_valid |= LA_MODE;
2087 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2088 == (S_ISGID | S_IXGRP) &&
2089 !lustre_in_group_p(uc,
2090 (cattr->la_valid & LA_GID) ?
2091 cattr->la_gid : pattr->la_gid) &&
2092 !md_capable(uc, CFS_CAP_FSETID)) {
2093 cattr->la_mode &= ~S_ISGID;
2094 cattr->la_valid |= LA_MODE;
2098 /* Inherit project ID from parent directory */
2099 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2100 cattr->la_projid = pattr->la_projid;
2101 if (S_ISDIR(cattr->la_mode)) {
2102 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2103 cattr->la_valid |= LA_FLAGS;
2105 cattr->la_valid |= LA_PROJID;
2108 rc = mdd_name_check(m, lname);
2112 switch (cattr->la_mode & S_IFMT) {
2114 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2116 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2117 RETURN(-ENAMETOOLONG);
2136 static int mdd_declare_create_object(const struct lu_env *env,
2137 struct mdd_device *mdd,
2138 struct mdd_object *p, struct mdd_object *c,
2139 struct lu_attr *attr,
2140 struct thandle *handle,
2141 const struct md_op_spec *spec,
2142 struct lu_buf *def_acl_buf,
2143 struct lu_buf *acl_buf,
2144 struct lu_buf *hsm_buf,
2145 struct dt_allocation_hint *hint)
2147 const struct lu_buf *buf;
2150 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2155 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2156 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2157 /* if dir, then can inherit default ACl */
2158 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2159 XATTR_NAME_ACL_DEFAULT,
2165 if (acl_buf && acl_buf->lb_len > 0) {
2166 rc = mdo_declare_attr_set(env, c, attr, handle);
2170 rc = mdo_declare_xattr_set(env, c, acl_buf,
2171 XATTR_NAME_ACL_ACCESS, 0, handle);
2176 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2180 /* replay case, create LOV EA from client data */
2181 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2182 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2183 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2184 spec->u.sp_ea.eadatalen);
2185 rc = mdo_declare_xattr_set(env, c, buf,
2186 S_ISDIR(attr->la_mode) ?
2187 XATTR_NAME_LMV : XATTR_NAME_LOV,
2192 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2193 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2201 if (S_ISLNK(attr->la_mode)) {
2202 const char *target_name = spec->u.sp_symname;
2203 int sym_len = strlen(target_name);
2204 const struct lu_buf *buf;
2206 buf = mdd_buf_get_const(env, target_name, sym_len);
2207 rc = dt_declare_record_write(env, mdd_object_child(c),
2213 if (spec->sp_cr_file_secctx_name != NULL) {
2214 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2215 spec->sp_cr_file_secctx_size);
2216 rc = mdo_declare_xattr_set(env, c, buf,
2217 spec->sp_cr_file_secctx_name, 0,
2226 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2227 struct mdd_object *p, struct mdd_object *c,
2228 const struct lu_name *name,
2229 struct lu_attr *attr,
2230 struct thandle *handle,
2231 const struct md_op_spec *spec,
2232 struct linkea_data *ldata,
2233 struct lu_buf *def_acl_buf,
2234 struct lu_buf *acl_buf,
2235 struct lu_buf *hsm_buf,
2236 struct dt_allocation_hint *hint)
2240 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2241 def_acl_buf, acl_buf, hsm_buf, hint);
2245 if (S_ISDIR(attr->la_mode)) {
2246 rc = mdo_declare_ref_add(env, p, handle);
2251 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2252 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2256 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2257 enum changelog_rec_type type;
2259 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2260 name->ln_name, handle);
2264 rc = mdd_declare_links_add(env, c, handle, ldata);
2269 la->la_valid = LA_CTIME | LA_MTIME;
2270 rc = mdo_declare_attr_set(env, p, la, handle);
2274 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2275 S_ISREG(attr->la_mode) ? CL_CREATE :
2276 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2278 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2287 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2288 struct lu_attr *la, struct lu_buf *def_acl_buf,
2289 struct lu_buf *acl_buf)
2294 if (S_ISLNK(la->la_mode)) {
2295 acl_buf->lb_len = 0;
2296 def_acl_buf->lb_len = 0;
2300 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2301 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2302 XATTR_NAME_ACL_DEFAULT);
2303 mdd_read_unlock(env, pobj);
2305 /* If there are default ACL, fix mode/ACL by default ACL */
2306 def_acl_buf->lb_len = rc;
2307 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2308 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2309 acl_buf->lb_len = rc;
2310 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2313 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2314 /* If there are no default ACL, fix mode by mask */
2315 struct lu_ucred *uc = lu_ucred(env);
2317 /* The create triggered by MDT internal events, such as
2318 * LFSCK reset, will not contain valid "uc". */
2319 if (unlikely(uc != NULL))
2320 la->la_mode &= ~uc->uc_umask;
2322 acl_buf->lb_len = 0;
2323 def_acl_buf->lb_len = 0;
2330 * Create a metadata object and initialize it, set acl, xattr.
2332 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2333 struct mdd_object *son, struct lu_attr *attr,
2334 struct md_op_spec *spec, struct lu_buf *acl_buf,
2335 struct lu_buf *def_acl_buf,
2336 struct lu_buf *hsm_buf,
2337 struct dt_allocation_hint *hint,
2338 struct thandle *handle)
2340 const struct lu_buf *buf;
2343 mdd_write_lock(env, son, DT_TGT_CHILD);
2344 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2349 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2350 * created in declare phase, they also needs to be added to master
2351 * object as sub-directory entry. So it has to initialize the master
2352 * object, then set dir striped EA.(in mdo_xattr_set) */
2353 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle);
2355 GOTO(err_destroy, rc);
2358 * in case of replay we just set LOVEA provided by the client
2359 * XXX: I think it would be interesting to try "old" way where
2360 * MDT calls this xattr_set(LOV) in a different transaction.
2361 * probably this way we code can be made better.
2364 /* During creation, there are only a few cases we need do xattr_set to
2366 * 1. regular file: see comments above.
2367 * 2. dir: inherit default striping or pool settings from parent.
2368 * 3. create striped directory with provided stripeEA.
2369 * 4. create striped directory because inherit default layout from the
2372 if (spec->no_create ||
2373 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2374 S_ISDIR(attr->la_mode)) {
2375 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2376 spec->u.sp_ea.eadatalen);
2377 rc = mdo_xattr_set(env, son, buf,
2378 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2382 GOTO(err_destroy, rc);
2385 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2388 memset(&mh, 0, sizeof(mh));
2389 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2390 mh.mh_arch_id = spec->sp_archive_id;
2391 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2392 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2395 GOTO(err_destroy, rc);
2398 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2399 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2400 S_ISDIR(attr->la_mode)) {
2401 /* set default acl */
2402 rc = mdo_xattr_set(env, son, def_acl_buf,
2403 XATTR_NAME_ACL_DEFAULT, 0,
2406 GOTO(err_destroy, rc);
2408 /* set its own acl */
2409 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2410 rc = mdo_xattr_set(env, son, acl_buf,
2411 XATTR_NAME_ACL_ACCESS,
2414 GOTO(err_destroy, rc);
2418 if (S_ISLNK(attr->la_mode)) {
2419 struct dt_object *dt = mdd_object_child(son);
2420 const char *target_name = spec->u.sp_symname;
2421 int sym_len = strlen(target_name);
2424 buf = mdd_buf_get_const(env, target_name, sym_len);
2425 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2429 GOTO(err_initlized, rc = -EFAULT);
2432 if (spec->sp_cr_file_secctx_name != NULL) {
2433 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2434 spec->sp_cr_file_secctx_size);
2435 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2438 GOTO(err_initlized, rc);
2442 if (unlikely(rc != 0)) {
2444 if (S_ISDIR(attr->la_mode)) {
2445 /* Drop the reference, no need to delete "."/"..",
2446 * because the object to be destroied directly. */
2447 rc2 = mdo_ref_del(env, son, handle);
2451 rc2 = mdo_ref_del(env, son, handle);
2455 mdo_destroy(env, son, handle);
2458 mdd_write_unlock(env, son);
2462 static int mdd_index_delete(const struct lu_env *env,
2463 struct mdd_object *mdd_pobj,
2464 struct lu_attr *cattr,
2465 const struct lu_name *lname)
2467 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2468 struct thandle *handle;
2472 handle = mdd_trans_create(env, mdd);
2474 RETURN(PTR_ERR(handle));
2476 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2481 if (S_ISDIR(cattr->la_mode)) {
2482 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2487 /* Since this will only be used in the error handler path,
2488 * Let's set the thandle to be local and not mess the transno */
2489 handle->th_local = 1;
2490 rc = mdd_trans_start(env, mdd, handle);
2494 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2495 S_ISDIR(cattr->la_mode), handle);
2499 rc = mdd_trans_stop(env, mdd, rc, handle);
2505 * Create object and insert it into namespace.
2507 * Two operations have to be performed:
2509 * - an allocation of a new object (->do_create()), and
2510 * - an insertion into a parent index (->dio_insert()).
2512 * Due to locking, operation order is not important, when both are
2513 * successful, *but* error handling cases are quite different:
2515 * - if insertion is done first, and following object creation fails,
2516 * insertion has to be rolled back, but this operation might fail
2517 * also leaving us with dangling index entry.
2519 * - if creation is done first, is has to be undone if insertion fails,
2520 * leaving us with leaked space, which is not good but not fatal.
2522 * It seems that creation-first is simplest solution, but it is sub-optimal
2528 * case, because second mkdir is bound to create object, only to
2529 * destroy it immediately.
2531 * To avoid this follow local file systems that do double lookup:
2533 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2534 * 1. create (mdd_create_object_internal())
2535 * 2. insert (__mdd_index_insert(), lookup again)
2537 * \param[in] pobj parent object
2538 * \param[in] lname name of child being created
2539 * \param[in,out] child child object being created
2540 * \param[in] spec additional create parameters
2541 * \param[in] ma attributes for new child object
2543 * \retval 0 on success
2544 * \retval negative errno on failure
2546 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2547 const struct lu_name *lname, struct md_object *child,
2548 struct md_op_spec *spec, struct md_attr *ma)
2550 struct mdd_thread_info *info = mdd_env_info(env);
2551 struct lu_attr *la = &info->mti_la_for_fix;
2552 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2553 struct mdd_object *son = md2mdd_obj(child);
2554 struct mdd_device *mdd = mdo2mdd(pobj);
2555 struct lu_attr *attr = &ma->ma_attr;
2556 struct thandle *handle;
2557 struct lu_attr *pattr = &info->mti_pattr;
2558 struct lu_buf acl_buf;
2559 struct lu_buf def_acl_buf;
2560 struct lu_buf hsm_buf;
2561 struct linkea_data *ldata = &info->mti_link_data;
2562 const char *name = lname->ln_name;
2563 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2568 rc = mdd_la_get(env, mdd_pobj, pattr);
2572 /* Sanity checks before big job. */
2573 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2577 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2578 GOTO(out_free, rc = -EINPROGRESS);
2580 handle = mdd_trans_create(env, mdd);
2582 GOTO(out_free, rc = PTR_ERR(handle));
2584 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2585 MIN(mdd->mdd_dt_conf.ddp_max_ea_size, XATTR_SIZE_MAX));
2586 acl_buf = info->mti_xattr_buf;
2587 def_acl_buf.lb_buf = info->mti_key;
2588 def_acl_buf.lb_len = sizeof(info->mti_key);
2589 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2593 if (S_ISDIR(attr->la_mode)) {
2594 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2597 * migrate may create 1-stripe directory, so lod_ah_init()
2598 * doesn't adjust stripe count from lmu.
2600 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1)) {
2601 info->mti_lmu = *lmu;
2602 info->mti_lmu.lum_stripe_count = 0;
2603 spec->u.sp_ea.eadata = &info->mti_lmu;
2607 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2609 memset(ldata, 0, sizeof(*ldata));
2610 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2611 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2614 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2615 &tfid, lname, 1, 0, ldata);
2617 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2618 mdd_object_fid(mdd_pobj),
2619 lname, 1, 0, ldata);
2622 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2623 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2625 memset(&hsm_buf, 0, sizeof(hsm_buf));
2626 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2627 if (hsm_buf.lb_buf == NULL)
2628 GOTO(out_stop, rc = -ENOMEM);
2631 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2632 handle, spec, ldata, &def_acl_buf, &acl_buf,
2637 rc = mdd_trans_start(env, mdd, handle);
2641 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2642 &def_acl_buf, &hsm_buf, hint, handle);
2646 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2647 mdd_write_lock(env, son, DT_TGT_CHILD);
2648 son->mod_flags |= VOLATILE_OBJ;
2649 rc = mdd_orphan_insert(env, son, handle);
2650 GOTO(out_volatile, rc);
2652 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2653 attr->la_mode, name, handle);
2655 GOTO(err_created, rc);
2657 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2660 /* update parent directory mtime/ctime */
2662 la->la_valid = LA_CTIME | LA_MTIME;
2663 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2665 GOTO(err_insert, rc);
2671 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2672 rc2 = mdd_orphan_delete(env, son, handle);
2674 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2675 S_ISDIR(attr->la_mode),
2681 mdd_write_lock(env, son, DT_TGT_CHILD);
2682 if (S_ISDIR(attr->la_mode)) {
2683 /* Drop the reference, no need to delete "."/"..",
2684 * because the object is to be destroyed directly. */
2685 rc2 = mdo_ref_del(env, son, handle);
2687 mdd_write_unlock(env, son);
2692 /* For volatile files drop one link immediately, since there is
2693 * no filename in the namespace, and save any error returned. */
2694 rc2 = mdo_ref_del(env, son, handle);
2696 mdd_write_unlock(env, son);
2697 if (unlikely(rc == 0))
2702 /* Don't destroy the volatile object on success */
2703 if (likely(rc != 0))
2704 mdo_destroy(env, son, handle);
2705 mdd_write_unlock(env, son);
2708 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2709 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2710 rc = mdd_changelog_ns_store(env, mdd,
2711 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2712 S_ISREG(attr->la_mode) ? CL_CREATE :
2713 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2714 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2717 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2719 /* If creation fails, it is most likely due to the remote update
2720 * failure, because local transaction will mostly succeed at
2721 * this stage. There is no easy way to rollback all of previous
2722 * updates, so let's remove the object from namespace, and
2723 * LFSCK should handle the orphan object. */
2724 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2725 mdd_index_delete(env, mdd_pobj, attr, lname);
2729 if (is_vmalloc_addr(ldata->ld_buf))
2730 /* if we vmalloced a large buffer drop it */
2731 lu_buf_free(ldata->ld_buf);
2733 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2734 lu_buf_free(&hsm_buf);
2736 /* The child object shouldn't be cached anymore */
2738 set_bit(LU_OBJECT_HEARD_BANSHEE,
2739 &child->mo_lu.lo_header->loh_flags);
2744 * Get locks on parents in proper order
2745 * RETURN: < 0 - error, rename_order if successful
2753 static int mdd_rename_order(const struct lu_env *env,
2754 struct mdd_device *mdd,
2755 struct mdd_object *src_pobj,
2756 const struct lu_attr *pattr,
2757 struct mdd_object *tgt_pobj)
2759 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2763 if (src_pobj == tgt_pobj)
2764 RETURN(MDD_RN_SAME);
2766 /* compared the parent child relationship of src_p&tgt_p */
2767 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2769 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2772 rc = mdd_is_parent(env, mdd, src_pobj, pattr,
2786 /* has not mdd_write{read}_lock on any obj yet. */
2787 static int mdd_rename_sanity_check(const struct lu_env *env,
2788 struct mdd_object *src_pobj,
2789 const struct lu_attr *pattr,
2790 struct mdd_object *tgt_pobj,
2791 const struct lu_attr *tpattr,
2792 struct mdd_object *sobj,
2793 const struct lu_attr *cattr,
2794 struct mdd_object *tobj,
2795 const struct lu_attr *tattr)
2800 /* XXX: when get here, sobj must NOT be NULL,
2801 * the other case has been processed in cld_rename
2802 * before mdd_rename and enable MDS_PERM_BYPASS. */
2806 * If we are using project inheritance, we only allow renames
2807 * into our tree when the project IDs are the same; otherwise
2808 * tree quota mechanism would be circumvented.
2810 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2811 tpattr->la_projid != cattr->la_projid) ||
2812 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2813 (pattr->la_projid != tpattr->la_projid)))
2816 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2820 /* XXX: when get here, "tobj == NULL" means tobj must
2821 * NOT exist (neither on remote MDS, such case has been
2822 * processed in cld_rename before mdd_rename and enable
2824 * So check may_create, but not check may_unlink. */
2826 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2827 (src_pobj != tgt_pobj));
2829 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2830 (src_pobj != tgt_pobj), 1);
2832 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2833 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2838 static int mdd_declare_rename(const struct lu_env *env,
2839 struct mdd_device *mdd,
2840 struct mdd_object *mdd_spobj,
2841 struct mdd_object *mdd_tpobj,
2842 struct mdd_object *mdd_sobj,
2843 struct mdd_object *mdd_tobj,
2844 const struct lu_name *sname,
2845 const struct lu_name *tname,
2847 struct linkea_data *ldata,
2848 struct thandle *handle)
2850 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2853 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2854 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2860 /* name from source dir */
2861 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2865 /* .. from source child */
2866 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2867 /* source child can be directory,
2868 * counted by source dir's nlink */
2869 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2872 if (mdd_spobj != mdd_tpobj) {
2873 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2878 rc = mdo_declare_index_insert(env, mdd_sobj,
2880 S_IFDIR, dotdot, handle);
2885 /* new target child can be directory,
2886 * counted by target dir's nlink */
2887 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2892 la->la_valid = LA_CTIME | LA_MTIME;
2893 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2897 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2901 la->la_valid = LA_CTIME;
2902 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2906 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2911 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2912 mdd_object_type(mdd_sobj),
2913 tname->ln_name, handle);
2917 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2918 /* delete target child in target parent directory */
2919 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2924 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2928 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2929 /* target child can be directory,
2930 * delete "." reference in target child directory */
2931 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2935 /* delete ".." reference in target parent directory */
2936 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2941 la->la_valid = LA_CTIME;
2942 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2946 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2951 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2959 /* src object can be remote that is why we use only fid and type of object */
2960 static int mdd_rename(const struct lu_env *env,
2961 struct md_object *src_pobj, struct md_object *tgt_pobj,
2962 const struct lu_fid *lf, const struct lu_name *lsname,
2963 struct md_object *tobj, const struct lu_name *ltname,
2966 const char *sname = lsname->ln_name;
2967 const char *tname = ltname->ln_name;
2968 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2969 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2970 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2971 struct mdd_device *mdd = mdo2mdd(src_pobj);
2972 struct mdd_object *mdd_sobj = NULL; /* source object */
2973 struct mdd_object *mdd_tobj = NULL;
2974 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2975 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2976 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2977 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2978 struct thandle *handle;
2979 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2980 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2981 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2984 bool tobj_locked = 0;
2985 unsigned cl_flags = 0;
2989 /* let unlink to complete and commit */
2990 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
2993 mdd_tobj = md2mdd_obj(tobj);
2995 mdd_sobj = mdd_object_find(env, mdd, lf);
2996 if (IS_ERR(mdd_sobj))
2997 RETURN(PTR_ERR(mdd_sobj));
2999 rc = mdd_la_get(env, mdd_sobj, cattr);
3001 GOTO(out_pending, rc);
3003 rc = mdd_la_get(env, mdd_spobj, pattr);
3005 GOTO(out_pending, rc);
3008 rc = mdd_la_get(env, mdd_tobj, tattr);
3010 GOTO(out_pending, rc);
3011 /* search for an existing archive.
3012 * we should check ahead as the object
3013 * can be destroyed in this transaction */
3014 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3015 cl_flags |= CLF_RENAME_LAST_EXISTS;
3018 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3020 GOTO(out_pending, rc);
3022 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3023 mdd_sobj, cattr, mdd_tobj, tattr);
3025 GOTO(out_pending, rc);
3027 rc = mdd_name_check(mdd, ltname);
3029 GOTO(out_pending, rc);
3031 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3032 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3034 GOTO(out_pending, rc);
3036 handle = mdd_trans_create(env, mdd);
3038 GOTO(out_pending, rc = PTR_ERR(handle));
3040 memset(ldata, 0, sizeof(*ldata));
3041 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3042 lsname, mdd_object_fid(mdd_tpobj), ltname,
3047 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3048 mdd_tobj, lsname, ltname, ma, ldata, handle);
3052 rc = mdd_trans_start(env, mdd, handle);
3056 is_dir = S_ISDIR(cattr->la_mode);
3058 /* Remove source name from source directory */
3059 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3063 /* "mv dir1 dir2" needs "dir1/.." link update */
3064 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3065 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3067 GOTO(fixup_spobj2, rc);
3069 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3072 GOTO(fixup_spobj, rc);
3075 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3076 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3078 /* tname might been renamed to something else */
3079 GOTO(fixup_spobj, rc);
3082 /* Insert new fid with target name into target dir */
3083 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3086 GOTO(fixup_tpobj, rc);
3088 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3089 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3091 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3092 la->la_valid = LA_CTIME;
3093 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3095 GOTO(fixup_tpobj, rc);
3097 /* Update the linkEA for the source object */
3098 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3099 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3100 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3103 /* Old files might not have EA entry */
3104 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3105 lsname, handle, NULL, 0);
3106 mdd_write_unlock(env, mdd_sobj);
3107 /* We don't fail the transaction if the link ea can't be
3108 updated -- fid2path will use alternate lookup method. */
3111 /* Remove old target object
3112 * For tobj is remote case cmm layer has processed
3113 * and set tobj to NULL then. So when tobj is NOT NULL,
3114 * it must be local one.
3116 if (tobj && mdd_object_exists(mdd_tobj)) {
3117 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3119 if (mdd_is_dead_obj(mdd_tobj)) {
3120 /* shld not be dead, something is wrong */
3121 CERROR("tobj is dead, something is wrong\n");
3125 mdo_ref_del(env, mdd_tobj, handle);
3127 /* Remove dot reference. */
3128 if (S_ISDIR(tattr->la_mode))
3129 mdo_ref_del(env, mdd_tobj, handle);
3132 /* fetch updated nlink */
3133 rc = mdd_la_get(env, mdd_tobj, tattr);
3135 CERROR("%s: Failed to get nlink for tobj "
3137 mdd2obd_dev(mdd)->obd_name,
3138 PFID(tpobj_fid), rc);
3139 GOTO(fixup_tpobj, rc);
3142 la->la_valid = LA_CTIME;
3143 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3145 CERROR("%s: Failed to set ctime for tobj "
3147 mdd2obd_dev(mdd)->obd_name,
3148 PFID(tpobj_fid), rc);
3149 GOTO(fixup_tpobj, rc);
3152 /* XXX: this transfer to ma will be removed with LOD/OSP */
3153 ma->ma_attr = *tattr;
3154 ma->ma_valid |= MA_INODE;
3155 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3158 CERROR("%s: Failed to unlink tobj "
3160 mdd2obd_dev(mdd)->obd_name,
3161 PFID(tpobj_fid), rc);
3162 GOTO(fixup_tpobj, rc);
3165 /* fetch updated nlink */
3166 rc = mdd_la_get(env, mdd_tobj, tattr);
3167 if (rc == -ENOENT) {
3168 /* the object got removed, let's
3169 * return the latest known attributes */
3170 tattr->la_nlink = 0;
3172 } else if (rc != 0) {
3173 CERROR("%s: Failed to get nlink for tobj "
3175 mdd2obd_dev(mdd)->obd_name,
3176 PFID(tpobj_fid), rc);
3177 GOTO(fixup_tpobj, rc);
3179 /* XXX: this transfer to ma will be removed with LOD/OSP */
3180 ma->ma_attr = *tattr;
3181 ma->ma_valid |= MA_INODE;
3183 if (tattr->la_nlink == 0)
3184 cl_flags |= CLF_RENAME_LAST;
3186 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3189 la->la_valid = LA_CTIME | LA_MTIME;
3190 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3192 GOTO(fixup_tpobj, rc);
3194 if (mdd_spobj != mdd_tpobj) {
3195 la->la_valid = LA_CTIME | LA_MTIME;
3196 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3198 GOTO(fixup_tpobj, rc);
3205 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3207 CWARN("tp obj fix error %d\n",rc2);
3209 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3210 !mdd_is_dead_obj(mdd_tobj)) {
3212 mdo_ref_add(env, mdd_tobj, handle);
3214 mdo_ref_add(env, mdd_tobj, handle);
3217 rc2 = __mdd_index_insert(env, mdd_tpobj,
3219 mdd_object_type(mdd_tobj),
3222 CWARN("tp obj fix error: rc = %d\n", rc2);
3227 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3228 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3230 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3231 mdd2obd_dev(mdd)->obd_name, rc2);
3234 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3237 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3238 mdd2obd_dev(mdd)->obd_name, rc2);
3243 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3244 mdd_object_type(mdd_sobj), sname,
3247 CWARN("sp obj fix error: rc = %d\n", rc2);
3252 mdd_write_unlock(env, mdd_tobj);
3255 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3256 mdd_tobj, tpobj_fid, lf, spobj_fid,
3257 ltname, lsname, handle);
3260 rc = mdd_trans_stop(env, mdd, rc, handle);
3263 mdd_object_put(env, mdd_sobj);
3268 * Check whether we should migrate the file/dir
3270 * < 0 permission check failed or other error.
3271 * = 0 the file can be migrated.
3273 static int mdd_migrate_sanity_check(const struct lu_env *env,
3274 struct mdd_device *mdd,
3275 struct mdd_object *spobj,
3276 struct mdd_object *tpobj,
3277 struct mdd_object *sobj,
3278 struct mdd_object *tobj,
3279 const struct lu_attr *spattr,
3280 const struct lu_attr *tpattr,
3281 const struct lu_attr *attr)
3287 if (!mdd_object_remote(sobj)) {
3288 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3289 if (sobj->mod_count > 0) {
3290 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3291 mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(sobj)),
3293 mdd_read_unlock(env, sobj);
3296 mdd_read_unlock(env, sobj);
3299 if (mdd_object_exists(tobj))
3302 rc = mdd_rename_sanity_check(env, spobj, spattr, tpobj, tpattr, sobj,
3307 typedef int (*mdd_dir_stripe_cb)(const struct lu_env *env,
3308 struct mdd_object *obj,
3309 struct mdd_object *stripe,
3310 const struct lu_buf *lmv_buf,
3311 const struct lu_buf *lmu_buf,
3313 struct thandle *handle);
3315 static int mdd_dir_declare_delete_stripe(const struct lu_env *env,
3316 struct mdd_object *obj,
3317 struct mdd_object *stripe,
3318 const struct lu_buf *lmv_buf,
3319 const struct lu_buf *lmu_buf,
3321 struct thandle *handle)
3323 struct mdd_thread_info *info = mdd_env_info(env);
3324 char *stripe_name = info->mti_name;
3325 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3328 if (index < le32_to_cpu(lmu->lum_stripe_count))
3331 rc = mdo_declare_index_delete(env, stripe, dotdot, handle);
3335 snprintf(stripe_name, sizeof(info->mti_name), DFID":%d",
3336 PFID(mdd_object_fid(stripe)), index);
3338 rc = mdo_declare_index_delete(env, obj, stripe_name, handle);
3342 rc = mdo_declare_ref_del(env, obj, handle);
3347 /* delete stripe from its master object namespace */
3348 static int mdd_dir_delete_stripe(const struct lu_env *env,
3349 struct mdd_object *obj,
3350 struct mdd_object *stripe,
3351 const struct lu_buf *lmv_buf,
3352 const struct lu_buf *lmu_buf,
3354 struct thandle *handle)
3356 struct mdd_thread_info *info = mdd_env_info(env);
3357 char *stripe_name = info->mti_name;
3358 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3359 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3360 __u32 del_offset = le32_to_cpu(lmu->lum_stripe_count);
3365 /* local dir will delete via LOD */
3366 LASSERT(mdd_object_remote(obj));
3367 LASSERT(del_offset < le32_to_cpu(lmv->lmv_stripe_count));
3369 if (index < del_offset)
3372 mdd_write_lock(env, stripe, DT_SRC_CHILD);
3373 rc = __mdd_index_delete_only(env, stripe, dotdot, handle);
3377 snprintf(stripe_name, sizeof(info->mti_name), DFID":%d",
3378 PFID(mdd_object_fid(stripe)), index);
3380 rc = __mdd_index_delete_only(env, obj, stripe_name, handle);
3384 rc = mdo_ref_del(env, obj, handle);
3387 mdd_write_unlock(env, stripe);
3392 static int mdd_dir_declare_destroy_stripe(const struct lu_env *env,
3393 struct mdd_object *obj,
3394 struct mdd_object *stripe,
3395 const struct lu_buf *lmv_buf,
3396 const struct lu_buf *lmu_buf,
3398 struct thandle *handle)
3400 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3401 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3404 if (index < shrink_offset) {
3405 if (shrink_offset < 2)
3407 return mdo_declare_xattr_set(env, stripe, lmv_buf,
3408 XATTR_NAME_LMV".set", 0, handle);
3411 rc = mdo_declare_ref_del(env, stripe, handle);
3415 rc = mdo_declare_destroy(env, stripe, handle);
3420 static int mdd_dir_destroy_stripe(const struct lu_env *env,
3421 struct mdd_object *obj,
3422 struct mdd_object *stripe,
3423 const struct lu_buf *lmv_buf,
3424 const struct lu_buf *lmu_buf,
3426 struct thandle *handle)
3428 struct mdd_thread_info *info = mdd_env_info(env);
3429 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3430 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3431 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3436 /* update remaining stripes' LMV */
3437 if (index < shrink_offset) {
3438 struct lmv_mds_md_v1 *slave_lmv;
3439 struct lu_buf slave_buf = {
3440 .lb_buf = &info->mti_lmv.lmv_md_v1,
3441 .lb_len = sizeof(*slave_lmv)
3443 __u32 version = le32_to_cpu(lmv->lmv_layout_version);
3445 /* if dir will be shrunk to 1-stripe, don't update */
3446 if (shrink_offset < 2)
3449 slave_lmv = slave_buf.lb_buf;
3450 memset(slave_lmv, 0, sizeof(*slave_lmv));
3451 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
3452 slave_lmv->lmv_stripe_count = lmu->lum_stripe_count;
3453 slave_lmv->lmv_master_mdt_index = cpu_to_le32(index);
3454 slave_lmv->lmv_hash_type = lmv->lmv_hash_type &
3455 cpu_to_le32(LMV_HASH_TYPE_MASK);
3456 slave_lmv->lmv_layout_version = cpu_to_le32(++version);
3458 rc = mdo_xattr_set(env, stripe, &slave_buf,
3459 XATTR_NAME_LMV".set", 0, handle);
3463 mdd_write_lock(env, stripe, DT_SRC_CHILD);
3464 rc = mdo_ref_del(env, stripe, handle);
3466 rc = mdo_destroy(env, stripe, handle);
3467 mdd_write_unlock(env, stripe);
3472 static int mdd_shrink_stripe_is_empty(const struct lu_env *env,
3473 struct mdd_object *obj,
3474 struct mdd_object *stripe,
3475 const struct lu_buf *lmv_buf,
3476 const struct lu_buf *lmu_buf,
3478 struct thandle *handle)
3480 struct lmv_user_md *lmu = lmu_buf->lb_buf;
3481 __u32 shrink_offset = le32_to_cpu(lmu->lum_stripe_count);
3483 /* the default value is 0, but it means 1 */
3487 if (index < shrink_offset)
3490 return mdd_dir_is_empty(env, stripe);
3494 * iterate stripes of striped directory on remote MDT, local striped directory
3495 * is accessed via LOD.
3497 static int mdd_dir_iterate_stripes(const struct lu_env *env,
3498 struct mdd_object *obj,
3499 const struct lu_buf *lmv_buf,
3500 const struct lu_buf *lmu_buf,
3501 struct thandle *handle,
3502 mdd_dir_stripe_cb cb)
3504 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
3505 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3506 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
3507 struct mdd_object *stripe;
3515 for (i = 0; i < le32_to_cpu(lmv->lmv_stripe_count); i++) {
3516 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[i]);
3517 if (!fid_is_sane(fid))
3520 stripe = mdd_object_find(env, mdd, fid);
3522 RETURN(PTR_ERR(stripe));
3524 rc = cb(env, obj, stripe, lmv_buf, lmu_buf, i, handle);
3525 mdd_object_put(env, stripe);
3533 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3534 struct mdd_object *obj,
3535 const struct lu_buf *buf,
3537 int fl, struct thandle *handle);
3539 /* iterate xattrs, but ignore LMA, LMV, and LINKEA if 'skip_linkea' is set. */
3540 static int mdd_iterate_xattrs(const struct lu_env *env,
3541 struct mdd_object *sobj,
3542 struct mdd_object *tobj,
3544 struct thandle *handle,
3547 struct mdd_thread_info *info = mdd_env_info(env);
3549 struct lu_buf list_xbuf;
3550 struct lu_buf cbxbuf;
3551 struct lu_buf xbuf = { NULL };
3560 /* retrieve xattr list from the old object */
3561 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3562 if (list_xsize == -ENODATA)
3568 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3569 if (info->mti_big_buf.lb_buf == NULL)
3572 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3573 list_xbuf.lb_len = list_xsize;
3574 rc = mdo_xattr_list(env, sobj, &list_xbuf);
3580 xname = list_xbuf.lb_buf;
3582 xlen = strnlen(xname, rem - 1) + 1;
3583 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3584 strcmp(XATTR_NAME_LMV, xname) == 0)
3588 strcmp(XATTR_NAME_LINK, xname) == 0)
3591 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3592 if (xsize == -ENODATA)
3595 GOTO(out, rc = xsize);
3597 lu_buf_check_and_alloc(&xbuf, xsize);
3598 if (xbuf.lb_buf == NULL)
3599 GOTO(out, rc = -ENOMEM);
3601 rc = mdo_xattr_get(env, sobj, &xbuf, xname);
3609 rc = cb(env, tobj, &cbxbuf, xname, 0, handle);
3610 if (unlikely(rc == -ENOSPC &&
3611 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3612 rc = linkea_overflow_shrink(
3613 (struct linkea_data *)(cbxbuf.lb_buf));
3614 if (likely(rc > 0)) {
3632 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3633 struct mdd_object *sobj,
3634 struct mdd_object *tobj,
3635 const struct lu_name *sname,
3636 const struct lu_fid *sfid,
3637 const struct lu_name *lname,
3638 const struct lu_fid *fid,
3640 struct thandle *handle);
3642 static int mdd_declare_update_link(const struct lu_env *env,
3643 struct mdd_object *sobj,
3644 struct mdd_object *tobj,
3645 const struct lu_name *tname,
3646 const struct lu_fid *tpfid,
3647 const struct lu_name *lname,
3648 const struct lu_fid *fid,
3650 struct thandle *handle)
3652 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3653 struct mdd_object *pobj;
3657 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3658 !strcmp(tname->ln_name, lname->ln_name))
3661 pobj = mdd_object_find(env, mdd, fid);
3663 return PTR_ERR(pobj);
3666 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3668 rc = mdo_declare_index_insert(env, pobj, mdo2fid(tobj),
3669 mdd_object_type(sobj),
3670 lname->ln_name, handle);
3671 mdd_object_put(env, pobj);
3675 rc = mdo_declare_ref_add(env, tobj, handle);
3679 rc = mdo_declare_ref_del(env, sobj, handle);
3683 static int mdd_update_link(const struct lu_env *env,
3684 struct mdd_object *sobj,
3685 struct mdd_object *tobj,
3686 const struct lu_name *tname,
3687 const struct lu_fid *tpfid,
3688 const struct lu_name *lname,
3689 const struct lu_fid *fid,
3691 struct thandle *handle)
3693 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3694 struct mdd_object *pobj;
3699 LASSERT(lu_name_is_valid(lname));
3702 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3703 !strncmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3706 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3707 PFID(fid), PNAME(lname), PFID(mdo2fid(tobj)));
3709 pobj = mdd_object_find(env, mdd, fid);
3711 CWARN("%s: cannot find obj "DFID": %ld\n",
3712 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3713 RETURN(PTR_ERR(pobj));
3716 if (!mdd_object_exists(pobj)) {
3717 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3718 mdd_object_put(env, pobj);
3722 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3723 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3725 rc = __mdd_index_insert_only(env, pobj, mdo2fid(tobj),
3726 mdd_object_type(sobj),
3727 lname->ln_name, handle);
3728 mdd_write_unlock(env, pobj);
3729 mdd_object_put(env, pobj);
3733 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3734 rc = mdo_ref_add(env, tobj, handle);
3735 mdd_write_unlock(env, tobj);
3739 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3740 rc = mdo_ref_del(env, sobj, handle);
3741 mdd_write_unlock(env, sobj);
3746 static inline int mdd_fld_lookup(const struct lu_env *env,
3747 struct mdd_device *mdd,
3748 const struct lu_fid *fid,
3751 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
3752 struct seq_server_site *ss;
3755 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3757 range->lsr_flags = LU_SEQ_RANGE_MDT;
3758 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3762 *mdt_index = range->lsr_index;
3767 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3768 struct mdd_object *sobj,
3769 struct mdd_object *tobj,
3770 const struct lu_name *tname,
3771 const struct lu_fid *tpfid,
3772 const struct lu_name *lname,
3773 const struct lu_fid *fid,
3775 struct thandle *handle)
3777 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3778 __u32 source_mdt_index = *(__u32 *)opaque;
3779 __u32 link_mdt_index;
3785 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3786 !strcmp(tname->ln_name, lname->ln_name))
3789 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3793 RETURN(link_mdt_index == source_mdt_index);
3796 static int mdd_iterate_linkea(const struct lu_env *env,
3797 struct mdd_object *sobj,
3798 struct mdd_object *tobj,
3799 const struct lu_name *tname,
3800 const struct lu_fid *tpfid,
3801 struct linkea_data *ldata,
3803 struct thandle *handle,
3806 struct mdd_thread_info *info = mdd_env_info(env);
3807 char *filename = info->mti_name;
3808 struct lu_name lname;
3815 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3816 linkea_next_entry(ldata)) {
3817 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3820 /* Note: lname might miss \0 at the end */
3821 snprintf(filename, sizeof(info->mti_name), "%.*s",
3822 lname.ln_namelen, lname.ln_name);
3823 lname.ln_name = filename;
3825 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3827 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3835 * Prepare linkea, and check whether file needs migrate: if source still has
3836 * link on source MDT, no need to migrate, just update namespace on source and
3839 * \retval 0 do migrate
3840 * \retval 1 don't migrate
3841 * \retval -errno on failure
3843 static int migrate_linkea_prepare(const struct lu_env *env,
3844 struct mdd_device *mdd,
3845 struct mdd_object *spobj,
3846 struct mdd_object *tpobj,
3847 struct mdd_object *sobj,
3848 const struct lu_name *lname,
3849 const struct lu_attr *attr,
3850 struct linkea_data *ldata)
3852 __u32 source_mdt_index;
3857 memset(ldata, 0, sizeof(*ldata));
3858 rc = mdd_linkea_prepare(env, sobj, mdo2fid(spobj), lname,
3859 mdo2fid(tpobj), lname, 1, 0, ldata);
3864 * Then it will check if the file should be migrated. If the file has
3865 * mulitple links, we only need migrate the file if all of its entries
3866 * has been migrated to the remote MDT.
3868 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3871 /* If there are still links locally, don't migrate this file */
3872 LASSERT(ldata->ld_leh != NULL);
3875 * If linkEA is overflow, it means there are some unknown name entries
3876 * under unknown parents, which will prevent the migration.
3878 if (unlikely(ldata->ld_leh->leh_overflow_time))
3881 rc = mdd_fld_lookup(env, mdd, mdo2fid(sobj), &source_mdt_index);
3885 rc = mdd_iterate_linkea(env, sobj, NULL, lname, mdo2fid(tpobj), ldata,
3886 &source_mdt_index, NULL,
3887 mdd_is_link_on_source_mdt);
3891 static int mdd_dir_declare_layout_delete(const struct lu_env *env,
3892 struct mdd_object *obj,
3893 const struct lu_buf *lmv_buf,
3894 const struct lu_buf *lmu_buf,
3895 struct thandle *handle)
3899 if (!lmv_buf->lb_buf)
3900 rc = mdo_declare_index_delete(env, obj, dotdot, handle);
3901 else if (mdd_object_remote(obj))
3902 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, lmu_buf, handle,
3903 mdd_dir_declare_delete_stripe);
3905 rc = mdo_declare_xattr_set(env, obj, lmu_buf,
3906 XATTR_NAME_LMV".del", 0, handle);
3911 static int mdd_dir_layout_delete(const struct lu_env *env,
3912 struct mdd_object *obj,
3913 const struct lu_buf *lmv_buf,
3914 const struct lu_buf *lmu_buf,
3915 struct thandle *handle)
3921 mdd_write_lock(env, obj, DT_SRC_PARENT);
3922 if (!lmv_buf->lb_buf)
3924 rc = __mdd_index_delete_only(env, obj, dotdot, handle);
3925 else if (mdd_object_remote(obj))
3926 /* striped, but remote */
3927 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, lmu_buf, handle,
3928 mdd_dir_delete_stripe);
3930 rc = mdo_xattr_set(env, obj, lmu_buf, XATTR_NAME_LMV".del", 0,
3932 mdd_write_unlock(env, obj);
3937 static int mdd_declare_migrate_create(const struct lu_env *env,
3938 struct mdd_object *tpobj,
3939 struct mdd_object *sobj,
3940 struct mdd_object *tobj,
3941 const struct lu_name *lname,
3942 struct lu_attr *attr,
3943 struct lu_buf *sbuf,
3944 struct linkea_data *ldata,
3945 struct md_op_spec *spec,
3946 struct dt_allocation_hint *hint,
3947 struct thandle *handle)
3949 struct mdd_thread_info *info = mdd_env_info(env);
3950 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3953 if (S_ISDIR(attr->la_mode)) {
3954 struct lu_buf lmu_buf = { NULL };
3957 struct lmv_user_md *lmu = &info->mti_lmv.lmv_user_md;
3959 lmu->lum_stripe_count = 0;
3960 lmu_buf.lb_buf = lmu;
3961 lmu_buf.lb_len = sizeof(*lmu);
3964 rc = mdd_dir_declare_layout_delete(env, sobj, sbuf, &lmu_buf,
3970 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LMV,
3977 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3978 lname, attr, handle, spec, ldata, NULL, NULL,
3983 if (S_ISDIR(attr->la_mode) && mdd_dir_is_empty(env, sobj) != 0) {
3986 * if sobj is not striped, fake a 1-stripe LMV, which
3987 * will be used to generate a compound LMV for tobj.
3989 LASSERT(sizeof(info->mti_key) >
3990 lmv_mds_md_size(1, LMV_MAGIC_V1));
3991 lmv = (typeof(lmv))info->mti_key;
3992 memset(lmv, 0, sizeof(*lmv));
3993 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3994 lmv->lmv_stripe_count = cpu_to_le32(1);
3995 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdo2fid(sobj));
3997 sbuf->lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3999 rc = mdo_declare_xattr_set(env, tobj, sbuf,
4000 XATTR_NAME_LMV".add", 0,
4002 sbuf->lb_buf = NULL;
4005 rc = mdo_declare_xattr_set(env, tobj, sbuf,
4006 XATTR_NAME_LMV".add", 0,
4014 * tobj mode will be used in lod_declare_xattr_set(), but it's not
4015 * createb yet, copy from sobj.
4017 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
4018 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
4019 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
4021 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle,
4022 mdo_declare_xattr_set);
4026 if (S_ISREG(attr->la_mode)) {
4027 struct lu_buf fid_buf;
4029 handle->th_complex = 1;
4031 /* target may be remote, update PFID via sobj. */
4032 fid_buf.lb_buf = (void *)mdo2fid(tobj);
4033 fid_buf.lb_len = sizeof(struct lu_fid);
4034 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
4039 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4044 if (!S_ISDIR(attr->la_mode)) {
4045 rc = mdd_iterate_linkea(env, sobj, tobj, lname, mdo2fid(tpobj),
4046 ldata, NULL, handle,
4047 mdd_declare_update_link);
4052 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LMV,
4063 * Create target, migrate xattrs and update links.
4065 * Create target according to \a spec, and then migrate xattrs, if it's
4066 * directory, migrate source stripes to target, else update fid to target
4069 * \param[in] env execution environment
4070 * \param[in] tpobj target parent object
4071 * \param[in] sobj source object
4072 * \param[in] tobj target object
4073 * \param[in] lname file name
4074 * \param[in] attr source attributes
4075 * \param[in] sbuf source LMV buf
4076 * \param[in] ldata source linkea
4077 * \param[in] spec migrate create spec
4078 * \param[in] hint target creation hint
4079 * \param[in] handle tranasction handle
4081 * \retval 0 on success
4082 * \retval -errno on failure
4084 static int mdd_migrate_create(const struct lu_env *env,
4085 struct mdd_object *tpobj,
4086 struct mdd_object *sobj,
4087 struct mdd_object *tobj,
4088 const struct lu_name *lname,
4089 struct lu_attr *attr,
4090 const struct lu_buf *sbuf,
4091 struct linkea_data *ldata,
4092 struct md_op_spec *spec,
4093 struct dt_allocation_hint *hint,
4094 struct thandle *handle)
4101 * directory will migrate sobj stripes to tobj:
4102 * 1. delete stripes from sobj.
4103 * 2. add stripes to tobj, see lod_dir_declare_layout_add().
4104 * 3. create/attach stripes for tobj, see lod_xattr_set_lmv().
4106 if (S_ISDIR(attr->la_mode)) {
4107 struct lu_buf lmu_buf = { NULL };
4110 struct mdd_thread_info *info = mdd_env_info(env);
4111 struct lmv_user_md *lmu = &info->mti_lmv.lmv_user_md;
4113 lmu->lum_stripe_count = 0;
4114 lmu_buf.lb_buf = lmu;
4115 lmu_buf.lb_len = sizeof(*lmu);
4118 rc = mdd_dir_layout_delete(env, sobj, sbuf, &lmu_buf, handle);
4123 * delete LMV so that later when destroying sobj it won't delete
4127 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4128 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LMV, handle);
4129 mdd_write_unlock(env, sobj);
4135 /* don't set nlink from sobj */
4136 attr->la_valid &= ~LA_NLINK;
4138 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4143 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4144 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
4145 mdd_write_unlock(env, tobj);
4149 /* for regular file, update OST objects XATTR_NAME_FID */
4150 if (S_ISREG(attr->la_mode)) {
4151 struct lu_buf fid_buf;
4153 /* target may be remote, update PFID via sobj. */
4154 fid_buf.lb_buf = (void *)mdo2fid(tobj);
4155 fid_buf.lb_len = sizeof(struct lu_fid);
4156 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4161 /* delete LOV to avoid deleting OST objs when destroying sobj */
4162 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4163 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4164 mdd_write_unlock(env, sobj);
4169 if (!S_ISDIR(attr->la_mode))
4170 rc = mdd_iterate_linkea(env, sobj, tobj, lname, mdo2fid(tpobj),
4171 ldata, NULL, handle, mdd_update_link);
4176 static int mdd_declare_migrate_update(const struct lu_env *env,
4177 struct mdd_object *spobj,
4178 struct mdd_object *tpobj,
4179 struct mdd_object *sobj,
4180 struct mdd_object *tobj,
4181 const struct lu_name *lname,
4182 struct lu_attr *attr,
4183 struct lu_attr *spattr,
4184 struct lu_attr *tpattr,
4185 struct linkea_data *ldata,
4189 struct thandle *handle)
4191 struct mdd_thread_info *info = mdd_env_info(env);
4192 const struct lu_fid *fid = mdo2fid(do_create ? tobj : sobj);
4193 struct lu_attr *la = &info->mti_la_for_fix;
4196 rc = mdo_declare_index_delete(env, spobj, lname->ln_name, handle);
4200 if (S_ISDIR(attr->la_mode)) {
4201 rc = mdo_declare_ref_del(env, spobj, handle);
4206 rc = mdo_declare_index_insert(env, tpobj, fid, mdd_object_type(sobj),
4207 lname->ln_name, handle);
4211 rc = mdd_declare_links_add(env, do_create ? tobj : sobj, handle, ldata);
4215 if (S_ISDIR(attr->la_mode)) {
4216 rc = mdo_declare_ref_add(env, tpobj, handle);
4221 la->la_valid = LA_CTIME | LA_MTIME;
4222 rc = mdo_declare_attr_set(env, spobj, la, handle);
4226 if (tpobj != spobj) {
4227 rc = mdo_declare_attr_set(env, tpobj, la, handle);
4232 if (do_create && do_destroy) {
4233 rc = mdo_declare_ref_del(env, sobj, handle);
4237 rc = mdo_declare_destroy(env, sobj, handle);
4246 * migrate dirent from \a spobj to \a tpobj, and destroy \a sobj
4248 static int mdd_migrate_update(const struct lu_env *env,
4249 struct mdd_object *spobj,
4250 struct mdd_object *tpobj,
4251 struct mdd_object *sobj,
4252 struct mdd_object *tobj,
4253 const struct lu_name *lname,
4254 struct lu_attr *attr,
4255 struct lu_attr *spattr,
4256 struct lu_attr *tpattr,
4257 struct linkea_data *ldata,
4261 struct thandle *handle)
4263 struct mdd_thread_info *info = mdd_env_info(env);
4264 const struct lu_fid *fid = mdo2fid(do_create ? tobj : sobj);
4265 struct lu_attr *la = &info->mti_la_for_fix;
4270 CDEBUG(D_INFO, "update %s "DFID"/"DFID" to "DFID"/"DFID"\n",
4271 lname->ln_name, PFID(mdo2fid(spobj)),
4272 PFID(mdo2fid(sobj)), PFID(mdo2fid(tpobj)),
4275 rc = __mdd_index_delete(env, spobj, lname->ln_name,
4276 S_ISDIR(attr->la_mode), handle);
4280 rc = __mdd_index_insert(env, tpobj, fid, mdd_object_type(sobj),
4281 lname->ln_name, handle);
4285 rc = mdd_links_write(env, do_create ? tobj : sobj, ldata, handle);
4289 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
4290 la->la_valid = LA_CTIME | LA_MTIME;
4291 mdd_write_lock(env, spobj, DT_SRC_PARENT);
4292 rc = mdd_update_time(env, spobj, spattr, la, handle);
4293 mdd_write_unlock(env, spobj);
4297 if (tpobj != spobj) {
4298 la->la_valid = LA_CTIME | LA_MTIME;
4299 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
4300 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
4301 mdd_write_unlock(env, tpobj);
4307 * there are three situations we shouldn't destroy source:
4308 * 1. if source is not dir, and it happens to be located on the same MDT
4310 * 2. if source is not dir, and has link on the same MDT where source is
4312 * 3. if source is dir, and it's a normal, non-empty dir.
4314 * the first two situations equals to !do_create, and the 3rd equals to
4315 * !do_destroy, so the below condition is actually
4316 * !(!do_create || !do_destroy).
4318 * NB, if user has opened source dir before migration, he will get
4319 * -ENOENT error when close it later, because source is likely to be
4320 * remote, which can't be moved to orphan list, but except this error
4321 * message, this won't cause any inconsistency or trouble.
4323 if (do_create && do_destroy) {
4324 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4325 mdo_ref_del(env, sobj, handle);
4326 rc = mdo_destroy(env, sobj, handle);
4327 mdd_write_unlock(env, sobj);
4334 * Migrate directory or file.
4336 * migrate source to target in following steps:
4337 * 1. create target, append source stripes after target's if it's directory,
4338 * migrate xattrs and update fid of source links.
4339 * 2. update namespace: migrate dirent from source parent to target parent,
4340 * update file linkea, and destroy source if it's not needed any more.
4342 * \param[in] env execution environment
4343 * \param[in] md_pobj parent master object
4344 * \param[in] md_sobj source object
4345 * \param[in] lname file name
4346 * \param[in] md_tobj target object
4347 * \param[in] spec target creation spec
4348 * \param[in] ma used to update \a pobj mtime and ctime
4350 * \retval 0 on success
4351 * \retval -errno on failure
4353 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4354 struct md_object *md_sobj, const struct lu_name *lname,
4355 struct md_object *md_tobj, struct md_op_spec *spec,
4358 struct mdd_device *mdd = mdo2mdd(md_pobj);
4359 struct mdd_thread_info *info = mdd_env_info(env);
4360 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4361 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4362 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4363 struct mdd_object *spobj = NULL;
4364 struct mdd_object *tpobj = NULL;
4365 struct lu_attr *spattr = &info->mti_pattr;
4366 struct lu_attr *tpattr = &info->mti_tpattr;
4367 struct lu_attr *attr = &info->mti_cattr;
4368 struct linkea_data *ldata = &info->mti_link_data;
4369 struct dt_allocation_hint *hint = &info->mti_hint;
4370 struct lu_fid *fid = &info->mti_fid2;
4371 struct lu_buf pbuf = { NULL };
4372 struct lu_buf sbuf = { NULL };
4373 struct lmv_mds_md_v1 *plmv;
4374 struct thandle *handle;
4375 bool do_create = true;
4376 bool do_destroy = true;
4380 rc = mdd_la_get(env, sobj, attr);
4384 /* locate source and target stripe on pobj, which are the real parent */
4385 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4386 if (rc < 0 && rc != -ENODATA)
4391 __u32 hash_type = le32_to_cpu(plmv->lmv_hash_type);
4392 __u32 count = le32_to_cpu(plmv->lmv_stripe_count);
4395 /* locate target parent stripe */
4396 if (hash_type & LMV_HASH_FLAG_MIGRATION) {
4398 * fail check here to make sure top dir migration
4401 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4402 GOTO(out, rc = -EIO);
4403 hash_type &= ~LMV_HASH_FLAG_MIGRATION;
4404 count = le32_to_cpu(plmv->lmv_migrate_offset);
4406 index = lmv_name_to_stripe_index(hash_type, count,
4410 GOTO(out, rc = index);
4412 fid_le_to_cpu(fid, &plmv->lmv_stripe_fids[index]);
4413 tpobj = mdd_object_find(env, mdd, fid);
4415 GOTO(out, rc = PTR_ERR(tpobj));
4417 /* locate source parent stripe */
4418 if (le32_to_cpu(plmv->lmv_hash_type) &
4419 LMV_HASH_FLAG_MIGRATION) {
4420 hash_type = le32_to_cpu(plmv->lmv_migrate_hash);
4421 count = le32_to_cpu(plmv->lmv_stripe_count) -
4422 le32_to_cpu(plmv->lmv_migrate_offset);
4424 index = lmv_name_to_stripe_index(hash_type, count,
4428 mdd_object_put(env, tpobj);
4429 GOTO(out, rc = index);
4432 index += le32_to_cpu(plmv->lmv_migrate_offset);
4433 fid_le_to_cpu(fid, &plmv->lmv_stripe_fids[index]);
4434 spobj = mdd_object_find(env, mdd, fid);
4435 if (IS_ERR(spobj)) {
4436 mdd_object_put(env, tpobj);
4437 GOTO(out, rc = PTR_ERR(spobj));
4441 mdd_object_get(spobj);
4446 mdd_object_get(tpobj);
4447 mdd_object_get(spobj);
4450 rc = mdd_la_get(env, spobj, spattr);
4454 rc = mdd_la_get(env, tpobj, tpattr);
4458 if (S_ISDIR(attr->la_mode)) {
4459 struct lmv_user_md_v1 *lmu = spec->u.sp_ea.eadata;
4464 * if user use default value '0' for stripe_count, we need to
4465 * adjust it to '1' to create a 1-stripe directory.
4467 if (lmu->lum_stripe_count == 0) {
4468 /* eadata is from request, don't alter it */
4469 info->mti_lmu = *lmu;
4470 info->mti_lmu.lum_stripe_count = cpu_to_le32(1);
4471 spec->u.sp_ea.eadata = &info->mti_lmu;
4472 lmu = spec->u.sp_ea.eadata;
4475 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4476 if (rc == -ENODATA) {
4477 if (mdd_dir_is_empty(env, sobj) == 0) {
4479 * if sobj is empty, and target is not striped,
4480 * create target as a normal directory.
4482 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4483 info->mti_lmu = *lmu;
4484 info->mti_lmu.lum_stripe_count = 0;
4485 spec->u.sp_ea.eadata = &info->mti_lmu;
4486 lmu = spec->u.sp_ea.eadata;
4490 * sobj is not striped dir, if it's not empty,
4491 * it will be migrated to be a stripe of target,
4492 * don't destroy it after migration.
4499 struct lmv_mds_md_v1 *lmv = sbuf.lb_buf;
4501 if (le32_to_cpu(lmv->lmv_hash_type) &
4502 LMV_HASH_FLAG_MIGRATION) {
4503 __u32 lum_stripe_count = lmu->lum_stripe_count;
4504 __u32 lmv_hash_type = lmv->lmv_hash_type &
4505 cpu_to_le32(LMV_HASH_TYPE_MASK);
4507 if (!lum_stripe_count)
4508 lum_stripe_count = cpu_to_le32(1);
4510 /* TODO: check specific MDTs */
4511 if (lmv->lmv_migrate_offset !=
4513 lmv->lmv_master_mdt_index !=
4514 lmu->lum_stripe_offset ||
4515 (lmv_hash_type != 0 &&
4516 lmv_hash_type != lmu->lum_hash_type)) {
4517 CERROR("%s: \'"DNAME"\' migration was "
4518 "interrupted, run \'lfs migrate "
4519 "-m %d -c %d -H %d "DNAME"\' to "
4520 "finish migration.\n",
4521 mdd2obd_dev(mdd)->obd_name,
4524 lmv->lmv_master_mdt_index),
4526 lmv->lmv_migrate_offset),
4527 le32_to_cpu(lmv_hash_type),
4529 GOTO(out, rc = -EPERM);
4531 GOTO(out, rc = -EALREADY);
4534 } else if (!mdd_object_remote(tpobj)) {
4536 * if source is already on MDT where target parent is located,
4537 * no need to create, just update namespace.
4540 } else if (S_ISLNK(attr->la_mode)) {
4541 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4543 GOTO(out, rc = -ENOMEM);
4544 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4547 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4548 mdd2obd_dev(mdd)->obd_name,
4549 PFID(mdo2fid(sobj)), rc);
4552 spec->u.sp_symname = sbuf.lb_buf;
4553 } else if (S_ISREG(attr->la_mode)) {
4554 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
4555 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
4559 * if sobj has link on the same MDT, no need to create, just update
4560 * namespace, and it will be a remote file on target parent, which is
4561 * similar to rename.
4563 rc = migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, lname, attr,
4570 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4571 spattr, tpattr, attr);
4575 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
4577 handle = mdd_trans_create(env, mdd);
4579 GOTO(out, rc = PTR_ERR(handle));
4582 rc = mdd_declare_migrate_create(env, tpobj, sobj, tobj, lname,
4583 attr, &sbuf, ldata, spec, hint,
4586 GOTO(stop_trans, rc);
4589 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, tobj, lname,
4590 attr, spattr, tpattr, ldata, do_create,
4591 do_destroy, ma, handle);
4593 GOTO(stop_trans, rc);
4595 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4598 GOTO(stop_trans, rc);
4600 rc = mdd_trans_start(env, mdd, handle);
4602 GOTO(stop_trans, rc);
4605 rc = mdd_migrate_create(env, tpobj, sobj, tobj, lname, attr,
4606 &sbuf, ldata, spec, hint, handle);
4608 GOTO(stop_trans, rc);
4611 rc = mdd_migrate_update(env, spobj, tpobj, sobj, tobj, lname, attr,
4612 spattr, tpattr, ldata, do_create, do_destroy,
4615 GOTO(stop_trans, rc);
4617 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
4618 mdo2fid(spobj), mdo2fid(sobj),
4619 mdo2fid(tpobj), lname, lname, handle);
4621 GOTO(stop_trans, rc);
4625 rc = mdd_trans_stop(env, mdd, rc, handle);
4627 if (spobj && !IS_ERR(spobj))
4628 mdd_object_put(env, spobj);
4629 if (tpobj && !IS_ERR(tpobj))
4630 mdd_object_put(env, tpobj);
4636 static int __mdd_dir_declare_layout_shrink(const struct lu_env *env,
4637 struct mdd_object *pobj,
4638 struct mdd_object *obj,
4639 struct mdd_object *stripe,
4640 struct lu_attr *attr,
4641 struct lu_buf *lmv_buf,
4642 const struct lu_buf *lmu_buf,
4643 struct lu_name *lname,
4644 struct thandle *handle)
4646 struct mdd_thread_info *info = mdd_env_info(env);
4647 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
4648 struct lmv_user_md *lmu = (typeof(lmu))info->mti_key;
4649 struct lu_buf shrink_buf = { .lb_buf = lmu,
4650 .lb_len = sizeof(*lmu) };
4655 memcpy(lmu, lmu_buf->lb_buf, sizeof(*lmu));
4657 if (le32_to_cpu(lmu->lum_stripe_count) < 2)
4658 lmu->lum_stripe_count = 0;
4660 rc = mdd_dir_declare_layout_delete(env, obj, lmv_buf, &shrink_buf,
4665 if (lmu->lum_stripe_count == 0) {
4666 lmu->lum_stripe_count = cpu_to_le32(1);
4668 rc = mdo_declare_xattr_del(env, obj, XATTR_NAME_LMV, handle);
4673 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
4674 mdd_dir_declare_destroy_stripe);
4678 if (le32_to_cpu(lmu->lum_stripe_count) > 1)
4679 return mdo_declare_xattr_set(env, obj, lmv_buf,
4680 XATTR_NAME_LMV".set", 0, handle);
4682 rc = mdo_declare_index_insert(env, stripe, mdo2fid(pobj), S_IFDIR,
4687 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle,
4688 mdo_declare_xattr_set);
4692 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4696 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4700 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4704 rc = mdo_declare_index_insert(env, pobj, mdo2fid(stripe), attr->la_mode,
4705 lname->ln_name, handle);
4709 rc = mdo_declare_ref_del(env, obj, handle);
4713 rc = mdo_declare_ref_del(env, obj, handle);
4717 rc = mdo_declare_destroy(env, obj, handle);
4726 * after files under \a obj were migrated, shrink old stripes from \a obj,
4727 * furthermore, if it becomes a 1-stripe directory, convert it to a normal one.
4729 static int __mdd_dir_layout_shrink(const struct lu_env *env,
4730 struct mdd_object *pobj,
4731 struct mdd_object *obj,
4732 struct mdd_object *stripe,
4733 struct lu_attr *attr,
4734 struct lu_buf *lmv_buf,
4735 const struct lu_buf *lmu_buf,
4736 struct lu_name *lname,
4737 struct thandle *handle)
4739 struct mdd_thread_info *info = mdd_env_info(env);
4740 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
4741 struct lmv_user_md *lmu = (typeof(lmu))info->mti_key;
4742 struct lu_buf shrink_buf = { .lb_buf = lmu,
4743 .lb_len = sizeof(*lmu) };
4744 int len = lmv_buf->lb_len;
4745 __u32 version = le32_to_cpu(lmv->lmv_layout_version);
4750 /* lmu needs to be altered, but lmu_buf is const */
4751 memcpy(lmu, lmu_buf->lb_buf, sizeof(*lmu));
4754 * if dir will be shrunk to 1-stripe, delete all stripes, because it
4755 * will be converted to normal dir.
4757 if (le32_to_cpu(lmu->lum_stripe_count) == 1)
4758 lmu->lum_stripe_count = 0;
4760 /* delete stripes after lmu_stripe_count */
4761 rc = mdd_dir_layout_delete(env, obj, lmv_buf, &shrink_buf, handle);
4765 if (lmu->lum_stripe_count == 0) {
4766 lmu->lum_stripe_count = cpu_to_le32(1);
4768 /* delete LMV to avoid deleting stripes again upon destroy */
4769 mdd_write_lock(env, obj, DT_SRC_CHILD);
4770 rc = mdo_xattr_del(env, obj, XATTR_NAME_LMV, handle);
4771 mdd_write_unlock(env, obj);
4776 /* destroy stripes after lmu_stripe_count */
4777 mdd_write_lock(env, obj, DT_SRC_PARENT);
4778 rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
4779 mdd_dir_destroy_stripe);
4780 mdd_write_unlock(env, obj);
4782 if (le32_to_cpu(lmu->lum_stripe_count) > 1) {
4783 /* update dir LMV, that's all if it's still striped. */
4784 lmv->lmv_stripe_count = lmu->lum_stripe_count;
4785 lmv->lmv_hash_type &= ~cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4786 lmv->lmv_migrate_offset = 0;
4787 lmv->lmv_migrate_hash = 0;
4788 lmv->lmv_layout_version = cpu_to_le32(++version);
4790 lmv_buf->lb_len = sizeof(*lmv);
4791 rc = mdo_xattr_set(env, obj, lmv_buf, XATTR_NAME_LMV".set", 0,
4793 lmv_buf->lb_len = len;
4797 /* replace directory with its remaining stripe */
4801 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4802 mdd_write_lock(env, obj, DT_SRC_CHILD);
4804 /* insert dotdot to stripe which points to parent */
4805 rc = __mdd_index_insert_only(env, stripe, mdo2fid(pobj), S_IFDIR,
4810 /* copy xattrs including linkea */
4811 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle, mdo_xattr_set);
4816 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4820 /* don't set nlink from parent */
4821 attr->la_valid &= ~LA_NLINK;
4823 rc = mdo_attr_set(env, stripe, attr, handle);
4827 /* delete dir name from parent */
4828 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4832 /* insert stripe to parent with dir name */
4833 rc = __mdd_index_insert_only(env, pobj, mdo2fid(stripe), attr->la_mode,
4834 lname->ln_name, handle);
4838 /* destroy dir obj */
4839 rc = mdo_ref_del(env, obj, handle);
4843 rc = mdo_ref_del(env, obj, handle);
4847 rc = mdo_destroy(env, obj, handle);
4853 mdd_write_unlock(env, obj);
4854 mdd_write_unlock(env, pobj);
4860 * shrink directory stripes to lum_stripe_count specified by lum_mds_md.
4862 int mdd_dir_layout_shrink(const struct lu_env *env,
4863 struct md_object *md_obj,
4864 const struct lu_buf *lmu_buf)
4866 struct mdd_device *mdd = mdo2mdd(md_obj);
4867 struct mdd_thread_info *info = mdd_env_info(env);
4868 struct mdd_object *obj = md2mdd_obj(md_obj);
4869 struct mdd_object *pobj = NULL;
4870 struct mdd_object *stripe = NULL;
4871 struct lu_attr *attr = &info->mti_pattr;
4872 struct lu_fid *fid = &info->mti_fid2;
4873 struct lu_name lname = { NULL };
4874 struct lu_buf lmv_buf = { NULL };
4875 struct lmv_mds_md_v1 *lmv;
4876 struct lmv_user_md *lmu;
4877 struct thandle *handle;
4882 rc = mdd_la_get(env, obj, attr);
4886 if (!S_ISDIR(attr->la_mode))
4889 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4893 lmv = lmv_buf.lb_buf;
4894 lmu = lmu_buf->lb_buf;
4896 /* this was checked in MDT */
4897 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4898 le32_to_cpu(lmv->lmv_stripe_count));
4900 rc = mdd_dir_iterate_stripes(env, obj, &lmv_buf, lmu_buf, NULL,
4901 mdd_shrink_stripe_is_empty);
4905 GOTO(out, rc = -ENOTEMPTY);
4908 * if obj stripe count will be shrunk to 1, we need to convert it to a
4909 * normal dir, which will change its fid and update parent namespace,
4910 * get obj name and parent fid from linkea.
4912 if (le32_to_cpu(lmu->lum_stripe_count) < 2) {
4913 struct linkea_data *ldata = &info->mti_link_data;
4914 char *filename = info->mti_name;
4916 rc = mdd_links_read(env, obj, ldata);
4920 if (ldata->ld_leh->leh_reccount > 1)
4921 GOTO(out, rc = -EINVAL);
4923 linkea_first_entry(ldata);
4925 GOTO(out, rc = -ENODATA);
4927 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4930 /* Note: lname might miss \0 at the end */
4931 snprintf(filename, sizeof(info->mti_name), "%.*s",
4932 lname.ln_namelen, lname.ln_name);
4933 lname.ln_name = filename;
4935 pobj = mdd_object_find(env, mdd, fid);
4942 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4944 stripe = mdd_object_find(env, mdd, fid);
4945 if (IS_ERR(stripe)) {
4946 mdd_object_put(env, pobj);
4948 GOTO(out, rc = PTR_ERR(stripe));
4952 handle = mdd_trans_create(env, mdd);
4954 GOTO(out, rc = PTR_ERR(handle));
4956 rc = __mdd_dir_declare_layout_shrink(env, pobj, obj, stripe, attr,
4957 &lmv_buf, lmu_buf, &lname, handle);
4959 GOTO(stop_trans, rc);
4961 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4964 GOTO(stop_trans, rc);
4966 rc = mdd_trans_start(env, mdd, handle);
4968 GOTO(stop_trans, rc);
4970 rc = __mdd_dir_layout_shrink(env, pobj, obj, stripe, attr, &lmv_buf,
4971 lmu_buf, &lname, handle);
4973 GOTO(stop_trans, rc);
4975 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4976 XATTR_NAME_LMV, handle);
4977 GOTO(stop_trans, rc);
4980 rc = mdd_trans_stop(env, mdd, rc, handle);
4983 mdd_object_put(env, stripe);
4984 mdd_object_put(env, pobj);
4986 lu_buf_free(&lmv_buf);
4990 const struct md_dir_operations mdd_dir_ops = {
4991 .mdo_is_subdir = mdd_is_subdir,
4992 .mdo_lookup = mdd_lookup,
4993 .mdo_create = mdd_create,
4994 .mdo_rename = mdd_rename,
4995 .mdo_link = mdd_link,
4996 .mdo_unlink = mdd_unlink,
4997 .mdo_create_data = mdd_create_data,
4998 .mdo_migrate = mdd_migrate,