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 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" located on remote server\n",
91 mdd_obj_dev_name(mdd_obj),
92 PFID(mdd_object_fid(mdd_obj)));
95 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
100 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
101 dt_try_as_dir(env, dir)))
102 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
109 int mdd_lookup(const struct lu_env *env,
110 struct md_object *pobj, const struct lu_name *lname,
111 struct lu_fid *fid, struct md_op_spec *spec)
113 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
117 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
121 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
122 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
126 /** Read the link EA into a temp buffer.
127 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
128 * A pointer to the buffer is stored in \a ldata::ld_buf.
132 static int __mdd_links_read(const struct lu_env *env,
133 struct mdd_object *mdd_obj,
134 struct linkea_data *ldata)
138 if (!mdd_object_exists(mdd_obj))
141 /* First try a small buf */
142 LASSERT(env != NULL);
143 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
145 if (ldata->ld_buf->lb_buf == NULL)
148 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
150 /* Buf was too small, figure out what we need. */
151 lu_buf_free(ldata->ld_buf);
152 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
156 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
157 if (ldata->ld_buf->lb_buf == NULL)
159 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
163 lu_buf_free(ldata->ld_buf);
164 ldata->ld_buf = NULL;
168 return linkea_init(ldata);
171 static int mdd_links_read(const struct lu_env *env,
172 struct mdd_object *mdd_obj,
173 struct linkea_data *ldata)
177 rc = __mdd_links_read(env, mdd_obj, ldata);
179 rc = linkea_init(ldata);
184 static int mdd_links_read_with_rec(const struct lu_env *env,
185 struct mdd_object *mdd_obj,
186 struct linkea_data *ldata)
190 rc = __mdd_links_read(env, mdd_obj, ldata);
192 rc = linkea_init_with_rec(ldata);
198 * Get parent FID of the directory
200 * Read parent FID from linkEA, if that fails, then do lookup
201 * dotdot to get the parent FID.
203 * \param[in] env execution environment
204 * \param[in] obj object from which to find the parent FID
205 * \param[in] attr attribute of the object
206 * \param[out] fid fid to get the parent FID
208 * \retval 0 if getting the parent FID succeeds.
209 * \retval negative errno if getting the parent FID fails.
211 static inline int mdd_parent_fid(const struct lu_env *env,
212 struct mdd_object *obj,
213 const struct lu_attr *attr,
216 struct mdd_thread_info *info = mdd_env_info(env);
217 struct linkea_data ldata = { NULL };
218 struct lu_buf *buf = &info->mti_link_buf;
219 struct lu_name lname;
224 LASSERTF(S_ISDIR(mdd_object_type(obj)),
225 "%s: FID "DFID" is not a directory type = %o\n",
226 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)),
227 mdd_object_type(obj));
229 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
230 if (buf->lb_buf == NULL)
231 GOTO(lookup, rc = 0);
234 rc = mdd_links_read_with_rec(env, obj, &ldata);
238 /* the obj is not locked, don't cache attributes */
239 mdd_invalidate(env, &obj->mod_obj);
241 LASSERT(ldata.ld_leh != NULL);
242 /* Directory should only have 1 parent */
243 if (ldata.ld_leh->leh_reccount > 1)
246 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
248 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
249 if (likely(fid_is_sane(fid)))
252 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
257 * For root fid use special function, which does not compare version component
258 * of fid. Version component is different for root fids on all MDTs.
260 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
262 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
263 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
267 * return 1: if \a tfid is the fid of the ancestor of \a mo;
269 * otherwise: values < 0, errors.
271 static int mdd_is_parent(const struct lu_env *env,
272 struct mdd_device *mdd,
273 struct mdd_object *mo,
274 const struct lu_attr *attr,
275 const struct lu_fid *tfid)
277 struct mdd_object *mp;
281 LASSERT(!lu_fid_eq(mdd_object_fid(mo), tfid));
282 pfid = &mdd_env_info(env)->mti_fid;
284 if (mdd_is_root(mdd, mdd_object_fid(mo)))
287 if (mdd_is_root(mdd, tfid))
290 rc = mdd_parent_fid(env, mo, attr, pfid);
295 if (lu_fid_eq(pfid, tfid))
298 if (mdd_is_root(mdd, pfid))
301 mp = mdd_object_find(env, mdd, pfid);
305 if (!mdd_object_exists(mp)) {
306 mdd_object_put(env, mp);
310 rc = mdd_parent_fid(env, mp, attr, pfid);
311 mdd_object_put(env, mp);
320 * No permission check is needed.
322 * returns 1: if fid is ancestor of @mo;
323 * returns 0: if fid is not an ancestor of @mo;
324 * returns < 0: if error
326 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
327 const struct lu_fid *fid)
329 struct mdd_device *mdd = mdo2mdd(mo);
330 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
334 if (!mdd_object_exists(md2mdd_obj(mo)))
337 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
340 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
344 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid);
349 * Check that @dir contains no entries except (possibly) dot and dotdot.
354 * -ENOTDIR not a directory object
355 * -ENOTEMPTY not empty
359 static int mdd_dir_is_empty(const struct lu_env *env,
360 struct mdd_object *dir)
363 struct dt_object *obj;
364 const struct dt_it_ops *iops;
368 obj = mdd_object_child(dir);
369 if (!dt_try_as_dir(env, obj))
372 iops = &obj->do_index_ops->dio_it;
373 it = iops->init(env, obj, LUDA_64BITHASH);
375 result = iops->get(env, it, (const struct dt_key *)"");
378 for (result = 0, i = 0; result == 0 && i < 3; ++i)
379 result = iops->next(env, it);
382 else if (result == 1)
384 } else if (result == 0)
386 * Huh? Index contains no zero key?
393 result = PTR_ERR(it);
394 /* -ENODEV means no valid stripe */
395 if (result == -ENODEV)
402 * Determine if the target object can be hard linked, and right now it only
403 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
404 * directory nlink count might exceed the maximum link count(see
405 * osd_object_ref_add), so it only check nlink for non-directories.
407 * \param[in] env thread environment
408 * \param[in] obj object being linked to
409 * \param[in] la attributes of \a obj
411 * \retval 0 if \a obj can be hard linked
412 * \retval negative error if \a obj is a directory or has too
415 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
416 const struct lu_attr *la)
418 struct mdd_device *m = mdd_obj2mdd_dev(obj);
423 /* Subdir count limitation can be broken through
424 * (see osd_object_ref_add), so only check non-directory here. */
425 if (!S_ISDIR(la->la_mode) &&
426 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
433 * Check whether it may create the cobj under the pobj.
435 * \param[in] env execution environment
436 * \param[in] pobj the parent directory
437 * \param[in] pattr the attribute of the parent directory
438 * \param[in] cobj the child to be created
439 * \param[in] check_perm if check WRITE|EXEC permission for parent
441 * \retval = 0 create the child under this dir is allowed
442 * \retval negative errno create the child under this dir is
445 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
446 const struct lu_attr *pattr, struct mdd_object *cobj,
452 if (cobj && mdd_object_exists(cobj))
455 if (mdd_is_dead_obj(pobj))
459 rc = mdd_permission_internal_locked(env, pobj, pattr,
460 MAY_WRITE | MAY_EXEC,
466 * Check whether can unlink from the pobj in the case of "cobj == NULL".
468 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
469 const struct lu_attr *pattr, const struct lu_attr *attr)
474 if (mdd_is_dead_obj(pobj))
477 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
480 rc = mdd_permission_internal_locked(env, pobj, pattr,
481 MAY_WRITE | MAY_EXEC,
486 if (pattr->la_flags & LUSTRE_APPEND_FL)
493 * pobj == NULL is remote ops case, under such case, pobj's
494 * VTX feature has been checked already, no need check again.
496 static inline int mdd_is_sticky(const struct lu_env *env,
497 struct mdd_object *pobj,
498 const struct lu_attr *pattr,
499 struct mdd_object *cobj,
500 const struct lu_attr *cattr)
502 struct lu_ucred *uc = lu_ucred_assert(env);
505 LASSERT(pattr != NULL);
506 if (!(pattr->la_mode & S_ISVTX) ||
507 (pattr->la_uid == uc->uc_fsuid))
511 LASSERT(cattr != NULL);
512 if (cattr->la_uid == uc->uc_fsuid)
515 return !md_capable(uc, CFS_CAP_FOWNER);
518 static int mdd_may_delete_entry(const struct lu_env *env,
519 struct mdd_object *pobj,
520 const struct lu_attr *pattr,
525 LASSERT(pobj != NULL);
526 if (!mdd_object_exists(pobj))
529 if (mdd_is_dead_obj(pobj))
534 rc = mdd_permission_internal_locked(env, pobj, pattr,
535 MAY_WRITE | MAY_EXEC,
541 if (pattr->la_flags & LUSTRE_APPEND_FL)
548 * Check whether it may delete the cobj from the pobj.
551 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
552 const struct lu_attr *tpattr, struct mdd_object *tobj,
553 const struct lu_attr *tattr, const struct lu_attr *cattr,
554 int check_perm, int check_empty)
560 LASSERT(tpattr != NULL);
561 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
569 if (!mdd_object_exists(tobj))
572 if (mdd_is_dead_obj(tobj))
575 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
578 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
581 /* additional check the rename case */
583 if (S_ISDIR(cattr->la_mode)) {
584 if (!S_ISDIR(tattr->la_mode))
587 if (mdd_is_root(mdo2mdd(&tobj->mod_obj),
588 mdd_object_fid(tobj)))
590 } else if (S_ISDIR(tattr->la_mode))
594 if (S_ISDIR(tattr->la_mode) && check_empty)
595 rc = mdd_dir_is_empty(env, tobj);
601 * Check whether it can create the link file(linked to @src_obj) under
602 * the target directory(@tgt_obj), and src_obj has been locked by
605 * \param[in] env execution environment
606 * \param[in] tgt_obj the target directory
607 * \param[in] tattr attributes of target directory
608 * \param[in] lname the link name
609 * \param[in] src_obj source object for link
610 * \param[in] cattr attributes for source object
612 * \retval = 0 it is allowed to create the link file under tgt_obj
613 * \retval negative error not allowed to create the link file
615 static int mdd_link_sanity_check(const struct lu_env *env,
616 struct mdd_object *tgt_obj,
617 const struct lu_attr *tattr,
618 const struct lu_name *lname,
619 struct mdd_object *src_obj,
620 const struct lu_attr *cattr)
622 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
626 if (!mdd_object_exists(src_obj))
629 if (mdd_is_dead_obj(src_obj))
632 /* Local ops, no lookup before link, check filename length here. */
633 rc = mdd_name_check(m, lname);
637 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
640 if (S_ISDIR(mdd_object_type(src_obj)))
643 LASSERT(src_obj != tgt_obj);
644 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
648 rc = __mdd_may_link(env, src_obj, cattr);
653 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
654 const char *name, struct thandle *handle)
656 struct dt_object *next = mdd_object_child(pobj);
660 if (dt_try_as_dir(env, next))
661 rc = dt_delete(env, next, (struct dt_key *)name, handle);
668 static int __mdd_index_insert_only(const struct lu_env *env,
669 struct mdd_object *pobj,
670 const struct lu_fid *lf, __u32 type,
672 struct thandle *handle)
674 struct dt_object *next = mdd_object_child(pobj);
678 if (dt_try_as_dir(env, next)) {
679 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
682 rec->rec_type = type;
683 rc = dt_insert(env, next, (const struct dt_rec *)rec,
684 (const struct dt_key *)name, handle);
691 /* insert named index, add reference if isdir */
692 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
693 const struct lu_fid *lf, __u32 type,
694 const char *name, struct thandle *handle)
699 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
700 if (rc == 0 && S_ISDIR(type)) {
701 mdd_write_lock(env, pobj, DT_TGT_PARENT);
702 mdo_ref_add(env, pobj, handle);
703 mdd_write_unlock(env, pobj);
709 /* delete named index, drop reference if isdir */
710 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
711 const char *name, int is_dir,
712 struct thandle *handle)
717 rc = __mdd_index_delete_only(env, pobj, name, handle);
718 if (rc == 0 && is_dir) {
719 mdd_write_lock(env, pobj, DT_TGT_PARENT);
720 mdo_ref_del(env, pobj, handle);
721 mdd_write_unlock(env, pobj);
727 static int mdd_llog_record_calc_size(const struct lu_env *env,
728 const struct lu_name *tname,
729 const struct lu_name *sname)
731 const struct lu_ucred *uc = lu_ucred(env);
732 enum changelog_rec_flags clf_flags = CLF_EXTRA_FLAGS;
733 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
736 clf_flags |= CLF_RENAME;
738 if (uc != NULL && uc->uc_jobid[0] != '\0')
739 clf_flags |= CLF_JOBID;
741 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
742 changelog_rec_offset(clf_flags, crfe) +
743 (tname != NULL ? tname->ln_namelen : 0) +
744 (sname != NULL ? 1 + sname->ln_namelen : 0));
747 int mdd_declare_changelog_store(const struct lu_env *env,
748 struct mdd_device *mdd,
749 enum changelog_rec_type type,
750 const struct lu_name *tname,
751 const struct lu_name *sname,
752 struct thandle *handle)
754 struct obd_device *obd = mdd2obd_dev(mdd);
755 struct llog_ctxt *ctxt;
756 struct llog_changelog_rec *rec;
758 struct thandle *llog_th;
762 if (!mdd_changelog_enabled(env, mdd, type))
765 reclen = mdd_llog_record_calc_size(env, tname, sname);
766 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
767 if (buf->lb_buf == NULL)
771 rec->cr_hdr.lrh_len = reclen;
772 rec->cr_hdr.lrh_type = CHANGELOG_REC;
774 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
778 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
780 GOTO(out_put, rc = PTR_ERR(llog_th));
782 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
790 int mdd_changelog_write_rec(const struct lu_env *env,
791 struct llog_handle *loghandle,
792 struct llog_rec_hdr *r,
793 struct llog_cookie *cookie,
794 int idx, struct thandle *th)
798 if (r->lrh_type == CHANGELOG_REC) {
799 struct mdd_device *mdd;
800 struct llog_changelog_rec *rec;
802 mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
803 rec = container_of0(r, struct llog_changelog_rec, cr_hdr);
805 spin_lock(&mdd->mdd_cl.mc_lock);
806 rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
807 spin_unlock(&mdd->mdd_cl.mc_lock);
809 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
813 * if current llog is full, we will generate a new
814 * llog, and since it's actually not an error, let's
815 * avoid increasing index so that userspace apps
816 * should not see a gap in the changelog sequence
818 if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
819 spin_lock(&mdd->mdd_cl.mc_lock);
820 ++mdd->mdd_cl.mc_index;
821 spin_unlock(&mdd->mdd_cl.mc_lock);
824 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
831 /** Add a changelog entry \a rec to the changelog llog
834 * \param handle - currently ignored since llogs start their own transaction;
835 * this will hopefully be fixed in llog rewrite
838 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
839 struct llog_changelog_rec *rec, struct thandle *th)
841 struct obd_device *obd = mdd2obd_dev(mdd);
842 struct llog_ctxt *ctxt;
843 struct thandle *llog_th;
846 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
847 changelog_rec_varsize(&rec->cr));
849 /* llog_lvfs_write_rec sets the llog tail len */
850 rec->cr_hdr.lrh_type = CHANGELOG_REC;
851 rec->cr.cr_time = cl_time();
853 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
857 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
859 GOTO(out_put, rc = PTR_ERR(llog_th));
861 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
862 /* nested journal transaction */
863 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
865 /* time to recover some space ?? */
866 if (likely(!mdd->mdd_changelog_gc ||
867 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
868 mdd->mdd_changelog_min_gc_interval >=
869 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
870 /* save a spin_lock trip */
872 spin_lock(&mdd->mdd_cl.mc_lock);
873 if (likely(mdd->mdd_changelog_gc &&
874 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
875 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
876 mdd->mdd_changelog_min_gc_interval)) {
877 if (unlikely(llog_cat_free_space(ctxt->loc_handle) <=
878 mdd->mdd_changelog_min_free_cat_entries ||
879 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD))) {
880 CWARN("%s:%s low on changelog_catalog free entries, "
881 "starting ChangeLog garbage collection thread\n",
883 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
886 /* indicate further kthread run will occur outside
887 * right after current journal transaction filling has
890 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
892 /* next check in mdd_changelog_min_gc_interval anyway
894 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
896 spin_unlock(&mdd->mdd_cl.mc_lock);
904 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
905 const struct lu_fid *sfid,
906 const struct lu_fid *spfid,
907 const struct lu_name *sname)
909 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
912 LASSERT(sfid != NULL);
913 LASSERT(spfid != NULL);
914 LASSERT(sname != NULL);
916 extsize = sname->ln_namelen + 1;
918 rnm->cr_sfid = *sfid;
919 rnm->cr_spfid = *spfid;
921 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
922 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
923 rec->cr_namelen += extsize;
926 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
928 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
930 if (jobid == NULL || jobid[0] == '\0')
933 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
936 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
938 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
940 ef->cr_extra_flags = eflags;
943 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
944 __u64 uid, __u64 gid)
946 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
948 uidgid->cr_uid = uid;
949 uidgid->cr_gid = gid;
952 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
955 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
960 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
962 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
964 omd->cr_openflags = flags;
967 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
968 const char *xattr_name)
970 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
972 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
975 /** Store a namespace change changelog record
976 * If this fails, we must fail the whole transaction; we don't
977 * want the change to commit without the log entry.
978 * \param target - mdd_object of change
979 * \param tpfid - target parent dir/object fid
980 * \param sfid - source object fid
981 * \param spfid - source parent fid
982 * \param tname - target name string
983 * \param sname - source name string
984 * \param handle - transaction handle
986 int mdd_changelog_ns_store(const struct lu_env *env,
987 struct mdd_device *mdd,
988 enum changelog_rec_type type,
989 enum changelog_rec_flags clf_flags,
990 struct mdd_object *target,
991 const struct lu_fid *tpfid,
992 const struct lu_fid *sfid,
993 const struct lu_fid *spfid,
994 const struct lu_name *tname,
995 const struct lu_name *sname,
996 struct thandle *handle)
998 const struct lu_ucred *uc = lu_ucred(env);
999 struct llog_changelog_rec *rec;
1002 __u64 xflags = CLFE_INVALID;
1006 if (!mdd_changelog_enabled(env, mdd, type))
1009 LASSERT(tpfid != NULL);
1010 LASSERT(tname != NULL);
1011 LASSERT(handle != NULL);
1013 reclen = mdd_llog_record_calc_size(env, tname, sname);
1014 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1015 if (buf->lb_buf == NULL)
1019 clf_flags &= CLF_FLAGMASK;
1020 clf_flags |= CLF_EXTRA_FLAGS;
1023 if (uc->uc_jobid[0] != '\0')
1024 clf_flags |= CLF_JOBID;
1025 xflags |= CLFE_UIDGID;
1030 clf_flags |= CLF_RENAME;
1032 clf_flags |= CLF_VERSION;
1034 rec->cr.cr_flags = clf_flags;
1036 if (clf_flags & CLF_EXTRA_FLAGS) {
1037 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1038 if (xflags & CLFE_UIDGID)
1039 mdd_changelog_rec_extra_uidgid(&rec->cr,
1040 uc->uc_uid, uc->uc_gid);
1041 if (xflags & CLFE_NID)
1042 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1045 rec->cr.cr_type = (__u32)type;
1046 rec->cr.cr_pfid = *tpfid;
1047 rec->cr.cr_namelen = tname->ln_namelen;
1048 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1050 if (clf_flags & CLF_RENAME)
1051 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1053 if (clf_flags & CLF_JOBID)
1054 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1056 if (likely(target != NULL)) {
1057 rec->cr.cr_tfid = *mdd_object_fid(target);
1058 target->mod_cltime = ktime_get();
1060 fid_zero(&rec->cr.cr_tfid);
1063 rc = mdd_changelog_store(env, mdd, rec, handle);
1065 CERROR("%s: cannot store changelog record: type = %d, "
1066 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1067 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1068 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1075 static int __mdd_links_add(const struct lu_env *env,
1076 struct mdd_object *mdd_obj,
1077 struct linkea_data *ldata,
1078 const struct lu_name *lname,
1079 const struct lu_fid *pfid,
1080 int first, int check)
1084 if (ldata->ld_leh == NULL) {
1085 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1089 rc = linkea_data_new(ldata,
1090 &mdd_env_info(env)->mti_link_buf);
1097 rc = linkea_links_find(ldata, lname, pfid);
1098 if (rc && rc != -ENOENT)
1104 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1105 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1109 linkea_add_buf(ldata, lname, tfid);
1112 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1113 linkea_add_buf(ldata, lname, pfid);
1115 return linkea_add_buf(ldata, lname, pfid);
1118 static int __mdd_links_del(const struct lu_env *env,
1119 struct mdd_object *mdd_obj,
1120 struct linkea_data *ldata,
1121 const struct lu_name *lname,
1122 const struct lu_fid *pfid)
1126 if (ldata->ld_leh == NULL) {
1127 rc = mdd_links_read(env, mdd_obj, ldata);
1132 rc = linkea_links_find(ldata, lname, pfid);
1136 linkea_del_buf(ldata, lname);
1140 static int mdd_linkea_prepare(const struct lu_env *env,
1141 struct mdd_object *mdd_obj,
1142 const struct lu_fid *oldpfid,
1143 const struct lu_name *oldlname,
1144 const struct lu_fid *newpfid,
1145 const struct lu_name *newlname,
1146 int first, int check,
1147 struct linkea_data *ldata)
1152 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1155 LASSERT(oldpfid != NULL || newpfid != NULL);
1157 if (mdd_obj->mod_flags & DEAD_OBJ)
1158 /* Unnecessary to update linkEA for dead object. */
1161 if (oldpfid != NULL) {
1162 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1164 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1167 /* No changes done. */
1172 /* If renaming, add the new record */
1173 if (newpfid != NULL)
1174 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1180 int mdd_links_rename(const struct lu_env *env,
1181 struct mdd_object *mdd_obj,
1182 const struct lu_fid *oldpfid,
1183 const struct lu_name *oldlname,
1184 const struct lu_fid *newpfid,
1185 const struct lu_name *newlname,
1186 struct thandle *handle,
1187 struct linkea_data *ldata,
1188 int first, int check)
1193 if (ldata == NULL) {
1194 ldata = &mdd_env_info(env)->mti_link_data;
1195 memset(ldata, 0, sizeof(*ldata));
1196 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1197 newpfid, newlname, first, check, ldata);
1202 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1203 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1209 if (newlname == NULL)
1210 CERROR("link_ea add failed %d "DFID"\n",
1211 rc, PFID(mdd_object_fid(mdd_obj)));
1212 else if (oldpfid == NULL)
1213 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1214 newlname->ln_namelen, newlname->ln_name, rc,
1215 PFID(mdd_object_fid(mdd_obj)));
1216 else if (newpfid == NULL)
1217 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1218 oldlname->ln_namelen, oldlname->ln_name, rc,
1219 PFID(mdd_object_fid(mdd_obj)));
1221 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1222 "\n", oldlname->ln_namelen, oldlname->ln_name,
1223 newlname->ln_namelen, newlname->ln_name, rc,
1224 PFID(mdd_object_fid(mdd_obj)));
1227 if (is_vmalloc_addr(ldata->ld_buf))
1228 /* if we vmalloced a large buffer drop it */
1229 lu_buf_free(ldata->ld_buf);
1234 static inline int mdd_links_add(const struct lu_env *env,
1235 struct mdd_object *mdd_obj,
1236 const struct lu_fid *pfid,
1237 const struct lu_name *lname,
1238 struct thandle *handle,
1239 struct linkea_data *ldata, int first)
1241 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1242 pfid, lname, handle, ldata, first, 0);
1245 static inline int mdd_links_del(const struct lu_env *env,
1246 struct mdd_object *mdd_obj,
1247 const struct lu_fid *pfid,
1248 const struct lu_name *lname,
1249 struct thandle *handle)
1251 return mdd_links_rename(env, mdd_obj, pfid, lname,
1252 NULL, NULL, handle, NULL, 0, 0);
1255 /** Read the link EA into a temp buffer.
1256 * Uses the name_buf since it is generally large.
1257 * \retval IS_ERR err
1258 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1260 struct lu_buf *mdd_links_get(const struct lu_env *env,
1261 struct mdd_object *mdd_obj)
1263 struct linkea_data ldata = { NULL };
1266 rc = mdd_links_read(env, mdd_obj, &ldata);
1267 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1270 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1271 struct linkea_data *ldata, struct thandle *handle)
1273 const struct lu_buf *buf;
1276 if (ldata == NULL || ldata->ld_buf == NULL ||
1277 ldata->ld_leh == NULL)
1280 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1284 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1285 ldata->ld_leh->leh_len);
1286 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1287 if (unlikely(rc == -ENOSPC)) {
1288 rc = linkea_overflow_shrink(ldata);
1296 static int mdd_declare_links_add(const struct lu_env *env,
1297 struct mdd_object *mdd_obj,
1298 struct thandle *handle,
1299 struct linkea_data *ldata)
1305 if (ldata != NULL && ldata->ld_leh != NULL) {
1306 ea_len = ldata->ld_leh->leh_len;
1307 linkea = ldata->ld_buf->lb_buf;
1309 ea_len = MAX_LINKEA_SIZE;
1313 rc = mdo_declare_xattr_set(env, mdd_obj,
1314 mdd_buf_get_const(env, linkea, ea_len),
1315 XATTR_NAME_LINK, 0, handle);
1320 static inline int mdd_declare_links_del(const struct lu_env *env,
1321 struct mdd_object *c,
1322 struct thandle *handle)
1326 /* For directory, the linkEA will be removed together
1327 * with the object. */
1328 if (!S_ISDIR(mdd_object_type(c)))
1329 rc = mdd_declare_links_add(env, c, handle, NULL);
1334 static int mdd_declare_link(const struct lu_env *env,
1335 struct mdd_device *mdd,
1336 struct mdd_object *p,
1337 struct mdd_object *c,
1338 const struct lu_name *name,
1339 struct thandle *handle,
1341 struct linkea_data *data)
1343 struct lu_fid tfid = *mdd_object_fid(c);
1346 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1347 tfid.f_oid = cfs_fail_val;
1349 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1350 name->ln_name, handle);
1354 rc = mdo_declare_ref_add(env, c, handle);
1358 la->la_valid = LA_CTIME | LA_MTIME;
1359 rc = mdo_declare_attr_set(env, p, la, handle);
1363 la->la_valid = LA_CTIME;
1364 rc = mdo_declare_attr_set(env, c, la, handle);
1368 rc = mdd_declare_links_add(env, c, handle, data);
1372 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1378 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1379 struct md_object *src_obj, const struct lu_name *lname,
1382 const char *name = lname->ln_name;
1383 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1384 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1385 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1386 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1387 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1388 struct mdd_device *mdd = mdo2mdd(src_obj);
1389 struct thandle *handle;
1390 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1391 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1395 rc = mdd_la_get(env, mdd_sobj, cattr);
1399 rc = mdd_la_get(env, mdd_tobj, tattr);
1404 * If we are using project inheritance, we only allow hard link
1405 * creation in our tree when the project IDs are the same;
1406 * otherwise the tree quota mechanism could be circumvented.
1408 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1409 (tattr->la_projid != cattr->la_projid))
1412 handle = mdd_trans_create(env, mdd);
1414 GOTO(out_pending, rc = PTR_ERR(handle));
1416 memset(ldata, 0, sizeof(*ldata));
1418 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1419 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1421 /* Note: even this function will change ldata, but it comes from
1422 * thread_info, which is completely temporary and only seen in
1423 * this function, so we do not need reset ldata once it fails.*/
1424 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1425 mdd_object_fid(mdd_tobj), lname, 0, 0, ldata);
1429 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1434 rc = mdd_trans_start(env, mdd, handle);
1438 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1439 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1442 GOTO(out_unlock, rc);
1444 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1445 rc = mdo_ref_add(env, mdd_sobj, handle);
1447 GOTO(out_unlock, rc);
1450 *tfid = *mdd_object_fid(mdd_sobj);
1451 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1452 tfid->f_oid = cfs_fail_val;
1454 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1455 mdd_object_type(mdd_sobj), name, handle);
1457 mdo_ref_del(env, mdd_sobj, handle);
1458 GOTO(out_unlock, rc);
1461 la->la_valid = LA_CTIME | LA_MTIME;
1462 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1464 GOTO(out_unlock, rc);
1466 la->la_valid = LA_CTIME;
1467 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1469 /* Note: The failure of links_add should not cause the
1470 * link failure, so do not check return value. */
1471 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_tobj),
1472 lname, handle, ldata, 0);
1476 mdd_write_unlock(env, mdd_sobj);
1478 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1479 mdd_object_fid(mdd_tobj), NULL,
1480 NULL, lname, NULL, handle);
1482 rc = mdd_trans_stop(env, mdd, rc, handle);
1483 if (is_vmalloc_addr(ldata->ld_buf))
1484 /* if we vmalloced a large buffer drop it */
1485 lu_buf_free(ldata->ld_buf);
1490 static int mdd_mark_orphan_object(const struct lu_env *env,
1491 struct mdd_object *obj, struct thandle *handle,
1494 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1497 attr->la_valid = LA_FLAGS;
1498 attr->la_flags = LUSTRE_ORPHAN_FL;
1501 rc = mdo_declare_attr_set(env, obj, attr, handle);
1503 rc = mdo_attr_set(env, obj, attr, handle);
1508 static int mdd_declare_finish_unlink(const struct lu_env *env,
1509 struct mdd_object *obj,
1510 struct thandle *handle)
1514 /* Sigh, we do not know if the unlink object will become orphan in
1515 * declare phase, but fortunately the flags here does not matter
1516 * in current declare implementation */
1517 rc = mdd_mark_orphan_object(env, obj, handle, true);
1521 rc = mdo_declare_destroy(env, obj, handle);
1525 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1529 return mdd_declare_links_del(env, obj, handle);
1532 /* caller should take a lock before calling */
1533 int mdd_finish_unlink(const struct lu_env *env,
1534 struct mdd_object *obj, struct md_attr *ma,
1535 struct mdd_object *pobj,
1536 const struct lu_name *lname,
1540 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1543 LASSERT(mdd_write_locked(env, obj) != 0);
1545 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1546 /* add new orphan and the object
1547 * will be deleted during mdd_close() */
1548 obj->mod_flags |= DEAD_OBJ;
1549 if (obj->mod_count) {
1550 rc = mdd_orphan_insert(env, obj, th);
1552 CDEBUG(D_HA, "Object "DFID" is inserted into "
1553 "orphan list, open count = %d\n",
1554 PFID(mdd_object_fid(obj)),
1557 CERROR("Object "DFID" fail to be an orphan, "
1558 "open count = %d, maybe cause failed "
1560 PFID(mdd_object_fid(obj)),
1563 /* mark object as an orphan here, not
1564 * before mdd_orphan_insert() as racing
1565 * mdd_la_get() may propagate ORPHAN_OBJ
1566 * causing the asserition */
1567 rc = mdd_mark_orphan_object(env, obj, th, false);
1569 rc = mdo_destroy(env, obj, th);
1571 } else if (!is_dir) {
1572 /* old files may not have link ea; ignore errors */
1573 mdd_links_del(env, obj, mdd_object_fid(pobj), lname, th);
1581 * has mdd_write_lock on cobj already, but not on pobj yet
1583 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1584 const struct lu_attr *pattr,
1585 struct mdd_object *cobj,
1586 const struct lu_attr *cattr)
1591 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1596 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1597 struct mdd_object *p, struct mdd_object *c,
1598 const struct lu_name *name, struct md_attr *ma,
1599 struct thandle *handle, int no_name, int is_dir)
1601 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1604 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1605 if (likely(no_name == 0)) {
1606 rc = mdo_declare_index_delete(env, p, name->ln_name,
1613 rc = mdo_declare_ref_del(env, p, handle);
1619 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1620 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1621 la->la_valid = LA_CTIME | LA_MTIME;
1622 rc = mdo_declare_attr_set(env, p, la, handle);
1627 rc = mdo_declare_ref_del(env, c, handle);
1631 rc = mdo_declare_ref_del(env, c, handle);
1635 la->la_valid = LA_CTIME;
1636 rc = mdo_declare_attr_set(env, c, la, handle);
1640 rc = mdd_declare_finish_unlink(env, c, handle);
1644 /* FIXME: need changelog for remove entry */
1645 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1653 * test if a file has an HSM archive
1654 * if HSM attributes are not found in ma update them from
1657 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1658 struct mdd_object *obj,
1663 if (!(ma->ma_valid & MA_HSM)) {
1664 /* no HSM MD provided, read xattr */
1665 struct lu_buf *hsm_buf;
1666 const size_t buflen = sizeof(struct hsm_attrs);
1669 hsm_buf = mdd_buf_get(env, NULL, 0);
1670 lu_buf_alloc(hsm_buf, buflen);
1671 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1672 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1673 lu_buf_free(hsm_buf);
1677 ma->ma_valid |= MA_HSM;
1679 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1685 * Delete name entry and the object.
1686 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1687 * does not exist for this object, and it could only happen during resending
1688 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1689 * is also needed in this case(needed by changelog), so we have to add another
1690 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1691 * the ENOENT failure should be able to be fixed by redo mechanism.
1693 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1694 struct md_object *cobj, const struct lu_name *lname,
1695 struct md_attr *ma, int no_name)
1697 const char *name = lname->ln_name;
1698 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1699 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1700 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1701 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1702 struct mdd_object *mdd_cobj = NULL;
1703 struct mdd_device *mdd = mdo2mdd(pobj);
1704 struct thandle *handle;
1705 int rc, is_dir = 0, cl_flags = 0;
1708 /* let shutdown to start */
1709 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1711 /* cobj == NULL means only delete name entry */
1712 if (likely(cobj != NULL)) {
1713 mdd_cobj = md2mdd_obj(cobj);
1714 if (mdd_object_exists(mdd_cobj) == 0)
1718 rc = mdd_la_get(env, mdd_pobj, pattr);
1722 if (likely(mdd_cobj != NULL)) {
1724 rc = mdd_la_get(env, mdd_cobj, cattr);
1728 is_dir = S_ISDIR(cattr->la_mode);
1729 /* search for an existing archive.
1730 * we should check ahead as the object
1731 * can be destroyed in this transaction */
1732 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1733 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1736 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1740 handle = mdd_trans_create(env, mdd);
1742 RETURN(PTR_ERR(handle));
1744 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1745 lname, ma, handle, no_name, is_dir);
1749 rc = mdd_trans_start(env, mdd, handle);
1753 if (likely(mdd_cobj != NULL))
1754 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1756 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1757 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1762 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1763 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1764 GOTO(cleanup, rc = 0);
1766 if (likely(mdd_cobj != NULL)) {
1767 rc = mdo_ref_del(env, mdd_cobj, handle);
1769 __mdd_index_insert_only(env, mdd_pobj,
1770 mdd_object_fid(mdd_cobj),
1771 mdd_object_type(mdd_cobj),
1778 mdo_ref_del(env, mdd_cobj, handle);
1780 /* fetch updated nlink */
1781 rc = mdd_la_get(env, mdd_cobj, cattr);
1786 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1787 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1789 la->la_valid = LA_CTIME | LA_MTIME;
1790 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1794 /* Enough for only unlink the entry */
1795 if (unlikely(mdd_cobj == NULL))
1798 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1799 /* update ctime of an unlinked file only if it is still
1800 * opened or a link still exists */
1801 la->la_valid = LA_CTIME;
1802 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1807 /* XXX: this transfer to ma will be removed with LOD/OSP */
1808 ma->ma_attr = *cattr;
1809 ma->ma_valid |= MA_INODE;
1810 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1814 /* fetch updated nlink */
1815 rc = mdd_la_get(env, mdd_cobj, cattr);
1816 /* if object is removed then we can't get its attrs,
1818 if (rc == -ENOENT) {
1819 cattr->la_nlink = 0;
1823 if (cattr->la_nlink == 0) {
1824 ma->ma_attr = *cattr;
1825 ma->ma_valid |= MA_INODE;
1830 if (likely(mdd_cobj != NULL))
1831 mdd_write_unlock(env, mdd_cobj);
1834 if (cattr->la_nlink == 0)
1835 cl_flags |= CLF_UNLINK_LAST;
1837 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1839 rc = mdd_changelog_ns_store(env, mdd,
1840 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1841 mdd_cobj, mdd_object_fid(mdd_pobj), NULL, NULL,
1842 lname, NULL, handle);
1846 rc = mdd_trans_stop(env, mdd, rc, handle);
1852 * The permission has been checked when obj created, no need check again.
1854 static int mdd_cd_sanity_check(const struct lu_env *env,
1855 struct mdd_object *obj)
1860 if (!obj || mdd_is_dead_obj(obj))
1866 static int mdd_create_data(const struct lu_env *env,
1867 struct md_object *pobj,
1868 struct md_object *cobj,
1869 const struct md_op_spec *spec,
1872 struct mdd_device *mdd = mdo2mdd(cobj);
1873 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1874 struct mdd_object *son = md2mdd_obj(cobj);
1875 struct thandle *handle;
1876 const struct lu_buf *buf;
1877 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1878 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1882 rc = mdd_cd_sanity_check(env, son);
1886 if (!md_should_create(spec->sp_cr_flags))
1890 * there are following use cases for this function:
1891 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1892 * striping can be specified or not
1895 rc = mdd_la_get(env, son, attr);
1899 /* calling ->ah_make_hint() is used to transfer information from parent */
1900 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1902 handle = mdd_trans_create(env, mdd);
1904 GOTO(out_free, rc = PTR_ERR(handle));
1907 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1908 * Should this be fixed?
1910 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1911 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1912 spec->sp_cr_flags, spec->no_create);
1914 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1915 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1916 spec->u.sp_ea.eadatalen);
1921 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1922 XATTR_NAME_LOV, 0, handle);
1926 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1931 rc = mdd_trans_start(env, mdd, handle);
1935 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1941 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1944 rc = mdd_trans_stop(env, mdd, rc, handle);
1950 static int mdd_declare_object_initialize(const struct lu_env *env,
1951 struct mdd_object *parent,
1952 struct mdd_object *child,
1953 const struct lu_attr *attr,
1954 struct thandle *handle)
1959 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1960 if (!S_ISDIR(attr->la_mode))
1963 rc = mdo_declare_index_insert(env, child, mdd_object_fid(child),
1964 S_IFDIR, dot, handle);
1968 rc = mdo_declare_ref_add(env, child, handle);
1972 rc = mdo_declare_index_insert(env, child, mdd_object_fid(parent),
1973 S_IFDIR, dotdot, handle);
1978 static int mdd_object_initialize(const struct lu_env *env,
1979 const struct lu_fid *pfid,
1980 struct mdd_object *child,
1981 struct lu_attr *attr,
1982 struct thandle *handle)
1987 if (S_ISDIR(attr->la_mode)) {
1988 /* Add "." and ".." for newly created dir */
1989 mdo_ref_add(env, child, handle);
1990 rc = __mdd_index_insert_only(env, child, mdd_object_fid(child),
1991 S_IFDIR, dot, handle);
1993 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1996 mdo_ref_del(env, child, handle);
2003 * This function checks whether it can create a file/dir under the
2004 * directory(@pobj). The directory(@pobj) is not being locked by
2007 * \param[in] env execution environment
2008 * \param[in] pobj the directory to create files
2009 * \param[in] pattr the attributes of the directory
2010 * \param[in] lname the name of the created file/dir
2011 * \param[in] cattr the attributes of the file/dir
2012 * \param[in] spec create specification
2014 * \retval = 0 it is allowed to create file/dir under
2016 * \retval negative error not allowed to create file/dir
2017 * under the directory
2019 static int mdd_create_sanity_check(const struct lu_env *env,
2020 struct md_object *pobj,
2021 const struct lu_attr *pattr,
2022 const struct lu_name *lname,
2023 struct lu_attr *cattr,
2024 struct md_op_spec *spec)
2026 struct mdd_thread_info *info = mdd_env_info(env);
2027 struct lu_fid *fid = &info->mti_fid;
2028 struct mdd_object *obj = md2mdd_obj(pobj);
2029 struct mdd_device *m = mdo2mdd(pobj);
2030 bool check_perm = true;
2035 if (mdd_is_dead_obj(obj))
2039 * In some cases this lookup is not needed - we know before if name
2040 * exists or not because MDT performs lookup for it.
2041 * name length check is done in lookup.
2043 if (spec->sp_cr_lookup) {
2045 * Check if the name already exist, though it will be checked in
2046 * _index_insert also, for avoiding rolling back if exists
2049 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2050 MAY_WRITE | MAY_EXEC);
2052 RETURN(rc ? : -EEXIST);
2054 /* Permission is already being checked in mdd_lookup */
2058 if (S_ISDIR(cattr->la_mode) &&
2059 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2060 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2061 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2063 if (!lmv_user_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2064 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2066 CERROR("%s: invalid lmv_user_md: magic = %x, "
2067 "stripe_offset = %d, stripe_count = %u: "
2068 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2069 le32_to_cpu(lum->lum_magic),
2070 (int)le32_to_cpu(lum->lum_stripe_offset),
2071 le32_to_cpu(lum->lum_stripe_count), rc);
2076 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2081 if (pattr->la_mode & S_ISGID) {
2082 struct lu_ucred *uc = lu_ucred(env);
2084 cattr->la_gid = pattr->la_gid;
2086 /* Directories are special, and always inherit S_ISGID */
2087 if (S_ISDIR(cattr->la_mode)) {
2088 cattr->la_mode |= S_ISGID;
2089 cattr->la_valid |= LA_MODE;
2090 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2091 == (S_ISGID | S_IXGRP) &&
2092 !lustre_in_group_p(uc,
2093 (cattr->la_valid & LA_GID) ?
2094 cattr->la_gid : pattr->la_gid) &&
2095 !md_capable(uc, CFS_CAP_FSETID)) {
2096 cattr->la_mode &= ~S_ISGID;
2097 cattr->la_valid |= LA_MODE;
2101 /* Inherit project ID from parent directory */
2102 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2103 cattr->la_projid = pattr->la_projid;
2104 if (S_ISDIR(cattr->la_mode)) {
2105 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2106 cattr->la_valid |= LA_FLAGS;
2108 cattr->la_valid |= LA_PROJID;
2111 rc = mdd_name_check(m, lname);
2115 switch (cattr->la_mode & S_IFMT) {
2117 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2119 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2120 RETURN(-ENAMETOOLONG);
2139 static int mdd_declare_create_object(const struct lu_env *env,
2140 struct mdd_device *mdd,
2141 struct mdd_object *p, struct mdd_object *c,
2142 struct lu_attr *attr,
2143 struct thandle *handle,
2144 const struct md_op_spec *spec,
2145 struct lu_buf *def_acl_buf,
2146 struct lu_buf *acl_buf,
2147 struct lu_buf *hsm_buf,
2148 struct dt_allocation_hint *hint)
2150 const struct lu_buf *buf;
2153 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2158 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2159 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2160 /* if dir, then can inherit default ACl */
2161 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2162 XATTR_NAME_ACL_DEFAULT,
2168 if (acl_buf && acl_buf->lb_len > 0) {
2169 rc = mdo_declare_attr_set(env, c, attr, handle);
2173 rc = mdo_declare_xattr_set(env, c, acl_buf,
2174 XATTR_NAME_ACL_ACCESS, 0, handle);
2179 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2183 /* replay case, create LOV EA from client data */
2184 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2185 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2186 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2187 spec->u.sp_ea.eadatalen);
2188 rc = mdo_declare_xattr_set(env, c, buf,
2189 S_ISDIR(attr->la_mode) ?
2190 XATTR_NAME_LMV : XATTR_NAME_LOV,
2191 LU_XATTR_CREATE, handle);
2195 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2196 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2204 if (S_ISLNK(attr->la_mode)) {
2205 const char *target_name = spec->u.sp_symname;
2206 int sym_len = strlen(target_name);
2207 const struct lu_buf *buf;
2209 buf = mdd_buf_get_const(env, target_name, sym_len);
2210 rc = dt_declare_record_write(env, mdd_object_child(c),
2216 if (spec->sp_cr_file_secctx_name != NULL) {
2217 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2218 spec->sp_cr_file_secctx_size);
2219 rc = mdo_declare_xattr_set(env, c, buf,
2220 spec->sp_cr_file_secctx_name, 0,
2229 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2230 struct mdd_object *p, struct mdd_object *c,
2231 const struct lu_name *name,
2232 struct lu_attr *attr,
2233 struct thandle *handle,
2234 const struct md_op_spec *spec,
2235 struct linkea_data *ldata,
2236 struct lu_buf *def_acl_buf,
2237 struct lu_buf *acl_buf,
2238 struct lu_buf *hsm_buf,
2239 struct dt_allocation_hint *hint)
2243 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2244 def_acl_buf, acl_buf, hsm_buf, hint);
2248 if (S_ISDIR(attr->la_mode)) {
2249 rc = mdo_declare_ref_add(env, p, handle);
2254 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2255 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2259 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2260 enum changelog_rec_type type;
2262 rc = mdo_declare_index_insert(env, p, mdd_object_fid(c),
2263 attr->la_mode, name->ln_name,
2268 rc = mdd_declare_links_add(env, c, handle, ldata);
2273 la->la_valid = LA_CTIME | LA_MTIME;
2274 rc = mdo_declare_attr_set(env, p, la, handle);
2278 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2279 S_ISREG(attr->la_mode) ? CL_CREATE :
2280 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2282 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2291 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2292 struct lu_attr *la, struct lu_buf *def_acl_buf,
2293 struct lu_buf *acl_buf)
2298 if (S_ISLNK(la->la_mode)) {
2299 acl_buf->lb_len = 0;
2300 def_acl_buf->lb_len = 0;
2304 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2305 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2306 XATTR_NAME_ACL_DEFAULT);
2307 mdd_read_unlock(env, pobj);
2309 /* If there are default ACL, fix mode/ACL by default ACL */
2310 def_acl_buf->lb_len = rc;
2311 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2312 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2313 acl_buf->lb_len = rc;
2314 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2317 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2318 /* If there are no default ACL, fix mode by mask */
2319 struct lu_ucred *uc = lu_ucred(env);
2321 /* The create triggered by MDT internal events, such as
2322 * LFSCK reset, will not contain valid "uc". */
2323 if (unlikely(uc != NULL))
2324 la->la_mode &= ~uc->uc_umask;
2326 acl_buf->lb_len = 0;
2327 def_acl_buf->lb_len = 0;
2334 * Create a metadata object and initialize it, set acl, xattr.
2336 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2337 struct mdd_object *son, struct lu_attr *attr,
2338 struct md_op_spec *spec, struct lu_buf *acl_buf,
2339 struct lu_buf *def_acl_buf,
2340 struct lu_buf *hsm_buf,
2341 struct dt_allocation_hint *hint,
2342 struct thandle *handle, bool initsecctx)
2344 const struct lu_buf *buf;
2347 mdd_write_lock(env, son, DT_TGT_CHILD);
2348 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2353 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2354 * created in declare phase, they also needs to be added to master
2355 * object as sub-directory entry. So it has to initialize the master
2356 * object, then set dir striped EA.(in mdo_xattr_set) */
2357 rc = mdd_object_initialize(env, mdd_object_fid(pobj), son, attr,
2360 GOTO(err_destroy, rc);
2363 * in case of replay we just set LOVEA provided by the client
2364 * XXX: I think it would be interesting to try "old" way where
2365 * MDT calls this xattr_set(LOV) in a different transaction.
2366 * probably this way we code can be made better.
2369 /* During creation, there are only a few cases we need do xattr_set to
2371 * 1. regular file: see comments above.
2372 * 2. dir: inherit default striping or pool settings from parent.
2373 * 3. create striped directory with provided stripeEA.
2374 * 4. create striped directory because inherit default layout from the
2377 if (spec->no_create ||
2378 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2379 S_ISDIR(attr->la_mode)) {
2380 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2381 spec->u.sp_ea.eadatalen);
2382 rc = mdo_xattr_set(env, son, buf,
2383 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2385 LU_XATTR_CREATE, handle);
2387 GOTO(err_destroy, rc);
2390 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2393 memset(&mh, 0, sizeof(mh));
2394 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2395 mh.mh_arch_id = spec->sp_archive_id;
2396 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2397 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2400 GOTO(err_destroy, rc);
2403 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2404 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2405 S_ISDIR(attr->la_mode)) {
2406 /* set default acl */
2407 rc = mdo_xattr_set(env, son, def_acl_buf,
2408 XATTR_NAME_ACL_DEFAULT, 0,
2411 GOTO(err_destroy, rc);
2413 /* set its own acl */
2414 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2415 rc = mdo_xattr_set(env, son, acl_buf,
2416 XATTR_NAME_ACL_ACCESS,
2419 GOTO(err_destroy, rc);
2423 if (S_ISLNK(attr->la_mode)) {
2424 struct dt_object *dt = mdd_object_child(son);
2425 const char *target_name = spec->u.sp_symname;
2426 int sym_len = strlen(target_name);
2429 buf = mdd_buf_get_const(env, target_name, sym_len);
2430 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2434 GOTO(err_initlized, rc = -EFAULT);
2437 if (initsecctx && spec->sp_cr_file_secctx_name != NULL) {
2438 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2439 spec->sp_cr_file_secctx_size);
2440 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2443 GOTO(err_initlized, rc);
2447 if (unlikely(rc != 0)) {
2449 if (S_ISDIR(attr->la_mode)) {
2450 /* Drop the reference, no need to delete "."/"..",
2451 * because the object to be destroied directly. */
2452 rc2 = mdo_ref_del(env, son, handle);
2456 rc2 = mdo_ref_del(env, son, handle);
2460 mdo_destroy(env, son, handle);
2463 mdd_write_unlock(env, son);
2467 static int mdd_index_delete(const struct lu_env *env,
2468 struct mdd_object *mdd_pobj,
2469 struct lu_attr *cattr,
2470 const struct lu_name *lname)
2472 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2473 struct thandle *handle;
2477 handle = mdd_trans_create(env, mdd);
2479 RETURN(PTR_ERR(handle));
2481 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2486 if (S_ISDIR(cattr->la_mode)) {
2487 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2492 /* Since this will only be used in the error handler path,
2493 * Let's set the thandle to be local and not mess the transno */
2494 handle->th_local = 1;
2495 rc = mdd_trans_start(env, mdd, handle);
2499 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2500 S_ISDIR(cattr->la_mode), handle);
2504 rc = mdd_trans_stop(env, mdd, rc, handle);
2510 * Create object and insert it into namespace.
2512 * Two operations have to be performed:
2514 * - an allocation of a new object (->do_create()), and
2515 * - an insertion into a parent index (->dio_insert()).
2517 * Due to locking, operation order is not important, when both are
2518 * successful, *but* error handling cases are quite different:
2520 * - if insertion is done first, and following object creation fails,
2521 * insertion has to be rolled back, but this operation might fail
2522 * also leaving us with dangling index entry.
2524 * - if creation is done first, is has to be undone if insertion fails,
2525 * leaving us with leaked space, which is not good but not fatal.
2527 * It seems that creation-first is simplest solution, but it is sub-optimal
2533 * case, because second mkdir is bound to create object, only to
2534 * destroy it immediately.
2536 * To avoid this follow local file systems that do double lookup:
2538 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2539 * 1. create (mdd_create_object_internal())
2540 * 2. insert (__mdd_index_insert(), lookup again)
2542 * \param[in] pobj parent object
2543 * \param[in] lname name of child being created
2544 * \param[in,out] child child object being created
2545 * \param[in] spec additional create parameters
2546 * \param[in] ma attributes for new child object
2548 * \retval 0 on success
2549 * \retval negative errno on failure
2551 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2552 const struct lu_name *lname, struct md_object *child,
2553 struct md_op_spec *spec, struct md_attr *ma)
2555 struct mdd_thread_info *info = mdd_env_info(env);
2556 struct lu_attr *la = &info->mti_la_for_fix;
2557 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2558 struct mdd_object *son = md2mdd_obj(child);
2559 struct mdd_device *mdd = mdo2mdd(pobj);
2560 struct lu_attr *attr = &ma->ma_attr;
2561 struct thandle *handle;
2562 struct lu_attr *pattr = &info->mti_pattr;
2563 struct lu_buf acl_buf;
2564 struct lu_buf def_acl_buf;
2565 struct lu_buf hsm_buf;
2566 struct linkea_data *ldata = &info->mti_link_data;
2567 const char *name = lname->ln_name;
2568 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2573 rc = mdd_la_get(env, mdd_pobj, pattr);
2577 /* Sanity checks before big job. */
2578 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2582 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2583 GOTO(out_free, rc = -EINPROGRESS);
2585 handle = mdd_trans_create(env, mdd);
2587 GOTO(out_free, rc = PTR_ERR(handle));
2589 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2590 min_t(unsigned int, mdd->mdd_dt_conf.ddp_max_ea_size,
2592 acl_buf = info->mti_xattr_buf;
2593 def_acl_buf.lb_buf = info->mti_key;
2594 def_acl_buf.lb_len = sizeof(info->mti_key);
2595 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2599 if (S_ISDIR(attr->la_mode)) {
2600 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2603 * migrate may create 1-stripe directory, so lod_ah_init()
2604 * doesn't adjust stripe count from lmu.
2606 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1))
2607 lmu->lum_stripe_count = 0;
2610 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2612 memset(ldata, 0, sizeof(*ldata));
2613 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2614 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2617 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2618 &tfid, lname, 1, 0, ldata);
2620 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2621 mdd_object_fid(mdd_pobj),
2622 lname, 1, 0, ldata);
2625 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2626 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2628 memset(&hsm_buf, 0, sizeof(hsm_buf));
2629 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2630 if (hsm_buf.lb_buf == NULL)
2631 GOTO(out_stop, rc = -ENOMEM);
2634 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2635 handle, spec, ldata, &def_acl_buf, &acl_buf,
2640 rc = mdd_trans_start(env, mdd, handle);
2644 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2645 &def_acl_buf, &hsm_buf, hint, handle, true);
2649 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2650 mdd_write_lock(env, son, DT_TGT_CHILD);
2651 son->mod_flags |= VOLATILE_OBJ;
2652 rc = mdd_orphan_insert(env, son, handle);
2653 GOTO(out_volatile, rc);
2655 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(son),
2656 attr->la_mode, name, handle);
2658 GOTO(err_created, rc);
2660 mdd_links_add(env, son, mdd_object_fid(mdd_pobj), lname,
2663 /* update parent directory mtime/ctime */
2665 la->la_valid = LA_CTIME | LA_MTIME;
2666 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2668 GOTO(err_insert, rc);
2674 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2675 rc2 = mdd_orphan_delete(env, son, handle);
2677 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2678 S_ISDIR(attr->la_mode),
2684 mdd_write_lock(env, son, DT_TGT_CHILD);
2685 if (S_ISDIR(attr->la_mode)) {
2686 /* Drop the reference, no need to delete "."/"..",
2687 * because the object is to be destroyed directly. */
2688 rc2 = mdo_ref_del(env, son, handle);
2690 mdd_write_unlock(env, son);
2695 /* For volatile files drop one link immediately, since there is
2696 * no filename in the namespace, and save any error returned. */
2697 rc2 = mdo_ref_del(env, son, handle);
2699 mdd_write_unlock(env, son);
2700 if (unlikely(rc == 0))
2705 /* Don't destroy the volatile object on success */
2706 if (likely(rc != 0))
2707 mdo_destroy(env, son, handle);
2708 mdd_write_unlock(env, son);
2711 if (rc == 0 && fid_is_namespace_visible(mdd_object_fid(son)) &&
2712 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2713 rc = mdd_changelog_ns_store(env, mdd,
2714 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2715 S_ISREG(attr->la_mode) ? CL_CREATE :
2716 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2717 0, son, mdd_object_fid(mdd_pobj), NULL, NULL,
2718 lname, NULL, handle);
2720 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2722 /* If creation fails, it is most likely due to the remote update
2723 * failure, because local transaction will mostly succeed at
2724 * this stage. There is no easy way to rollback all of previous
2725 * updates, so let's remove the object from namespace, and
2726 * LFSCK should handle the orphan object. */
2727 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2728 mdd_index_delete(env, mdd_pobj, attr, lname);
2732 if (is_vmalloc_addr(ldata->ld_buf))
2733 /* if we vmalloced a large buffer drop it */
2734 lu_buf_free(ldata->ld_buf);
2736 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2737 lu_buf_free(&hsm_buf);
2739 /* The child object shouldn't be cached anymore */
2741 set_bit(LU_OBJECT_HEARD_BANSHEE,
2742 &child->mo_lu.lo_header->loh_flags);
2747 * Get locks on parents in proper order
2748 * RETURN: < 0 - error, rename_order if successful
2756 static int mdd_rename_order(const struct lu_env *env,
2757 struct mdd_device *mdd,
2758 struct mdd_object *src_pobj,
2759 const struct lu_attr *pattr,
2760 struct mdd_object *tgt_pobj)
2762 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2766 if (src_pobj == tgt_pobj)
2767 RETURN(MDD_RN_SAME);
2769 /* compared the parent child relationship of src_p & tgt_p */
2770 if (lu_fid_eq(&mdd->mdd_root_fid, mdd_object_fid(src_pobj))) {
2772 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdd_object_fid(tgt_pobj))) {
2775 rc = mdd_is_parent(env, mdd, src_pobj, pattr,
2776 mdd_object_fid(tgt_pobj));
2789 /* has not mdd_write{read}_lock on any obj yet. */
2790 static int mdd_rename_sanity_check(const struct lu_env *env,
2791 struct mdd_object *src_pobj,
2792 const struct lu_attr *pattr,
2793 struct mdd_object *tgt_pobj,
2794 const struct lu_attr *tpattr,
2795 struct mdd_object *sobj,
2796 const struct lu_attr *cattr,
2797 struct mdd_object *tobj,
2798 const struct lu_attr *tattr)
2803 /* XXX: when get here, sobj must NOT be NULL,
2804 * the other case has been processed in cld_rename
2805 * before mdd_rename and enable MDS_PERM_BYPASS. */
2809 * If we are using project inheritance, we only allow renames
2810 * into our tree when the project IDs are the same; otherwise
2811 * tree quota mechanism would be circumvented.
2813 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2814 tpattr->la_projid != cattr->la_projid) ||
2815 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2816 (pattr->la_projid != tpattr->la_projid)))
2819 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2823 /* XXX: when get here, "tobj == NULL" means tobj must
2824 * NOT exist (neither on remote MDS, such case has been
2825 * processed in cld_rename before mdd_rename and enable
2827 * So check may_create, but not check may_unlink. */
2829 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2830 (src_pobj != tgt_pobj));
2832 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2833 (src_pobj != tgt_pobj), 1);
2835 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2836 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2841 static int mdd_declare_rename(const struct lu_env *env,
2842 struct mdd_device *mdd,
2843 struct mdd_object *mdd_spobj,
2844 struct mdd_object *mdd_tpobj,
2845 struct mdd_object *mdd_sobj,
2846 struct mdd_object *mdd_tobj,
2847 const struct lu_name *sname,
2848 const struct lu_name *tname,
2850 struct linkea_data *ldata,
2851 struct thandle *handle)
2853 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2856 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2857 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2863 /* name from source dir */
2864 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2868 /* .. from source child */
2869 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2870 /* source child can be directory, count by source dir's nlink */
2871 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2874 if (mdd_spobj != mdd_tpobj) {
2875 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2880 rc = mdo_declare_index_insert(env, mdd_sobj,
2881 mdd_object_fid(mdd_tpobj),
2882 S_IFDIR, dotdot, handle);
2887 /* new target child can be directory,
2888 * counted by target dir's nlink */
2889 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2894 la->la_valid = LA_CTIME | LA_MTIME;
2895 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2899 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2903 la->la_valid = LA_CTIME;
2904 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2908 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2913 rc = mdo_declare_index_insert(env, mdd_tpobj, mdd_object_fid(mdd_sobj),
2914 mdd_object_type(mdd_sobj),
2915 tname->ln_name, handle);
2919 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2920 /* delete target child in target parent directory */
2921 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2926 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2930 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2931 /* target child can be directory,
2932 * delete "." reference in target child directory */
2933 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2937 /* delete ".." reference in target parent directory */
2938 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2943 la->la_valid = LA_CTIME;
2944 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2948 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2953 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2961 /* src object can be remote that is why we use only fid and type of object */
2962 static int mdd_rename(const struct lu_env *env,
2963 struct md_object *src_pobj, struct md_object *tgt_pobj,
2964 const struct lu_fid *lf, const struct lu_name *lsname,
2965 struct md_object *tobj, const struct lu_name *ltname,
2968 const char *sname = lsname->ln_name;
2969 const char *tname = ltname->ln_name;
2970 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2971 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2972 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2973 struct mdd_device *mdd = mdo2mdd(src_pobj);
2974 struct mdd_object *mdd_sobj = NULL; /* source object */
2975 struct mdd_object *mdd_tobj = NULL;
2976 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2977 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2978 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2979 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2980 struct thandle *handle;
2981 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2982 const struct lu_fid *tpobj_fid = mdd_object_fid(mdd_tpobj);
2983 const struct lu_fid *spobj_fid = mdd_object_fid(mdd_spobj);
2986 bool tobj_locked = 0;
2987 unsigned cl_flags = 0;
2991 /* let unlink to complete and commit */
2992 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
2995 mdd_tobj = md2mdd_obj(tobj);
2997 mdd_sobj = mdd_object_find(env, mdd, lf);
2998 if (IS_ERR(mdd_sobj))
2999 RETURN(PTR_ERR(mdd_sobj));
3001 rc = mdd_la_get(env, mdd_sobj, cattr);
3003 GOTO(out_pending, rc);
3005 rc = mdd_la_get(env, mdd_spobj, pattr);
3007 GOTO(out_pending, rc);
3010 rc = mdd_la_get(env, mdd_tobj, tattr);
3012 GOTO(out_pending, rc);
3013 /* search for an existing archive.
3014 * we should check ahead as the object
3015 * can be destroyed in this transaction */
3016 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3017 cl_flags |= CLF_RENAME_LAST_EXISTS;
3020 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3022 GOTO(out_pending, rc);
3024 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3025 mdd_sobj, cattr, mdd_tobj, tattr);
3027 GOTO(out_pending, rc);
3029 rc = mdd_name_check(mdd, ltname);
3031 GOTO(out_pending, rc);
3033 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3034 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3036 GOTO(out_pending, rc);
3038 handle = mdd_trans_create(env, mdd);
3040 GOTO(out_pending, rc = PTR_ERR(handle));
3042 memset(ldata, 0, sizeof(*ldata));
3043 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3044 lsname, mdd_object_fid(mdd_tpobj), ltname,
3049 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3050 mdd_tobj, lsname, ltname, ma, ldata, handle);
3054 rc = mdd_trans_start(env, mdd, handle);
3058 is_dir = S_ISDIR(cattr->la_mode);
3060 /* Remove source name from source directory */
3061 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3065 /* "mv dir1 dir2" needs "dir1/.." link update */
3066 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3067 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3069 GOTO(fixup_spobj2, rc);
3071 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3074 GOTO(fixup_spobj, rc);
3077 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3078 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3080 /* tname might been renamed to something else */
3081 GOTO(fixup_spobj, rc);
3084 /* Insert new fid with target name into target dir */
3085 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3088 GOTO(fixup_tpobj, rc);
3090 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3091 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3093 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3094 la->la_valid = LA_CTIME;
3095 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3097 GOTO(fixup_tpobj, rc);
3099 /* Update the linkEA for the source object */
3100 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3101 rc = mdd_links_rename(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3102 lsname, mdd_object_fid(mdd_tpobj), ltname,
3103 handle, ldata, 0, 0);
3105 /* Old files might not have EA entry */
3106 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3107 lsname, handle, NULL, 0);
3108 mdd_write_unlock(env, mdd_sobj);
3109 /* We don't fail the transaction if the link ea can't be
3110 updated -- fid2path will use alternate lookup method. */
3113 /* Remove old target object
3114 * For tobj is remote case cmm layer has processed
3115 * and set tobj to NULL then. So when tobj is NOT NULL,
3116 * it must be local one.
3118 if (tobj && mdd_object_exists(mdd_tobj)) {
3119 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3121 if (mdd_is_dead_obj(mdd_tobj)) {
3122 /* shld not be dead, something is wrong */
3123 CERROR("tobj is dead, something is wrong\n");
3127 mdo_ref_del(env, mdd_tobj, handle);
3129 /* Remove dot reference. */
3130 if (S_ISDIR(tattr->la_mode))
3131 mdo_ref_del(env, mdd_tobj, handle);
3134 /* fetch updated nlink */
3135 rc = mdd_la_get(env, mdd_tobj, tattr);
3137 CERROR("%s: Failed to get nlink for tobj "
3139 mdd2obd_dev(mdd)->obd_name,
3140 PFID(tpobj_fid), rc);
3141 GOTO(fixup_tpobj, rc);
3144 la->la_valid = LA_CTIME;
3145 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3147 CERROR("%s: Failed to set ctime for tobj "
3149 mdd2obd_dev(mdd)->obd_name,
3150 PFID(tpobj_fid), rc);
3151 GOTO(fixup_tpobj, rc);
3154 /* XXX: this transfer to ma will be removed with LOD/OSP */
3155 ma->ma_attr = *tattr;
3156 ma->ma_valid |= MA_INODE;
3157 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3160 CERROR("%s: Failed to unlink tobj "
3162 mdd2obd_dev(mdd)->obd_name,
3163 PFID(tpobj_fid), rc);
3164 GOTO(fixup_tpobj, rc);
3167 /* fetch updated nlink */
3168 rc = mdd_la_get(env, mdd_tobj, tattr);
3169 if (rc == -ENOENT) {
3170 /* the object got removed, let's
3171 * return the latest known attributes */
3172 tattr->la_nlink = 0;
3174 } else if (rc != 0) {
3175 CERROR("%s: Failed to get nlink for tobj "
3177 mdd2obd_dev(mdd)->obd_name,
3178 PFID(tpobj_fid), rc);
3179 GOTO(fixup_tpobj, rc);
3181 /* XXX: this transfer to ma will be removed with LOD/OSP */
3182 ma->ma_attr = *tattr;
3183 ma->ma_valid |= MA_INODE;
3185 if (tattr->la_nlink == 0)
3186 cl_flags |= CLF_RENAME_LAST;
3188 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3191 la->la_valid = LA_CTIME | LA_MTIME;
3192 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3194 GOTO(fixup_tpobj, rc);
3196 if (mdd_spobj != mdd_tpobj) {
3197 la->la_valid = LA_CTIME | LA_MTIME;
3198 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3200 GOTO(fixup_tpobj, rc);
3207 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3209 CWARN("tp obj fix error %d\n",rc2);
3211 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3212 !mdd_is_dead_obj(mdd_tobj)) {
3214 mdo_ref_add(env, mdd_tobj, handle);
3216 mdo_ref_add(env, mdd_tobj, handle);
3219 rc2 = __mdd_index_insert(env, mdd_tpobj,
3220 mdd_object_fid(mdd_tobj),
3221 mdd_object_type(mdd_tobj),
3224 CWARN("tp obj fix error: rc = %d\n", rc2);
3229 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3230 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3232 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3233 mdd2obd_dev(mdd)->obd_name, rc2);
3236 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3239 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3240 mdd2obd_dev(mdd)->obd_name, rc2);
3245 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3246 mdd_object_type(mdd_sobj), sname,
3249 CWARN("sp obj fix error: rc = %d\n", rc2);
3254 mdd_write_unlock(env, mdd_tobj);
3257 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3258 mdd_tobj, tpobj_fid, lf, spobj_fid,
3259 ltname, lsname, handle);
3262 rc = mdd_trans_stop(env, mdd, rc, handle);
3265 mdd_object_put(env, mdd_sobj);
3270 * Check whether we should migrate the file/dir
3272 * < 0 permission check failed or other error.
3273 * = 0 the file can be migrated.
3275 static int mdd_migrate_sanity_check(const struct lu_env *env,
3276 struct mdd_device *mdd,
3277 struct mdd_object *spobj,
3278 struct mdd_object *tpobj,
3279 struct mdd_object *sobj,
3280 struct mdd_object *tobj,
3281 const struct lu_attr *spattr,
3282 const struct lu_attr *tpattr,
3283 const struct lu_attr *attr)
3289 if (!mdd_object_remote(sobj)) {
3290 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3291 if (sobj->mod_count > 0) {
3292 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3293 mdd_obj_dev_name(sobj),
3294 PFID(mdd_object_fid(sobj)),
3296 mdd_read_unlock(env, sobj);
3299 mdd_read_unlock(env, sobj);
3302 if (mdd_object_exists(tobj))
3305 rc = mdd_rename_sanity_check(env, spobj, spattr, tpobj, tpattr, sobj,
3310 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3311 struct mdd_object *obj,
3312 const struct lu_buf *buf,
3314 int fl, struct thandle *handle);
3316 /* iterate xattrs, but ignore LMA, LMV, and LINKEA if 'skip_linkea' is set. */
3317 static int mdd_iterate_xattrs(const struct lu_env *env,
3318 struct mdd_object *sobj,
3319 struct mdd_object *tobj,
3321 struct thandle *handle,
3324 struct mdd_thread_info *info = mdd_env_info(env);
3326 struct lu_buf list_xbuf;
3327 struct lu_buf cbxbuf;
3328 struct lu_buf xbuf = { NULL };
3337 /* retrieve xattr list from the old object */
3338 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3339 if (list_xsize == -ENODATA)
3345 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3346 if (info->mti_big_buf.lb_buf == NULL)
3349 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3350 list_xbuf.lb_len = list_xsize;
3351 rc = mdo_xattr_list(env, sobj, &list_xbuf);
3357 xname = list_xbuf.lb_buf;
3359 xlen = strnlen(xname, rem - 1) + 1;
3360 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3361 strcmp(XATTR_NAME_LMV, xname) == 0)
3365 strcmp(XATTR_NAME_LINK, xname) == 0)
3368 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3369 if (xsize == -ENODATA)
3372 GOTO(out, rc = xsize);
3374 lu_buf_check_and_alloc(&xbuf, xsize);
3375 if (xbuf.lb_buf == NULL)
3376 GOTO(out, rc = -ENOMEM);
3378 rc = mdo_xattr_get(env, sobj, &xbuf, xname);
3385 cbxbuf.lb_len = xsize;
3387 rc = cb(env, tobj, &cbxbuf, xname, 0, handle);
3388 if (unlikely(rc == -ENOSPC &&
3389 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3390 rc = linkea_overflow_shrink(
3391 (struct linkea_data *)(cbxbuf.lb_buf));
3392 if (likely(rc > 0)) {
3410 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3411 struct mdd_object *sobj,
3412 struct mdd_object *tobj,
3413 const struct lu_name *sname,
3414 const struct lu_fid *sfid,
3415 const struct lu_name *lname,
3416 const struct lu_fid *fid,
3418 struct thandle *handle);
3420 static int mdd_declare_update_link(const struct lu_env *env,
3421 struct mdd_object *sobj,
3422 struct mdd_object *tobj,
3423 const struct lu_name *tname,
3424 const struct lu_fid *tpfid,
3425 const struct lu_name *lname,
3426 const struct lu_fid *fid,
3428 struct thandle *handle)
3430 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3431 struct mdd_object *pobj;
3435 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3436 !strcmp(tname->ln_name, lname->ln_name))
3439 pobj = mdd_object_find(env, mdd, fid);
3441 return PTR_ERR(pobj);
3444 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3446 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
3447 mdd_object_type(sobj),
3448 lname->ln_name, handle);
3449 mdd_object_put(env, pobj);
3453 rc = mdo_declare_ref_add(env, tobj, handle);
3457 rc = mdo_declare_ref_del(env, sobj, handle);
3461 static int mdd_update_link(const struct lu_env *env,
3462 struct mdd_object *sobj,
3463 struct mdd_object *tobj,
3464 const struct lu_name *tname,
3465 const struct lu_fid *tpfid,
3466 const struct lu_name *lname,
3467 const struct lu_fid *fid,
3469 struct thandle *handle)
3471 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3472 struct mdd_object *pobj;
3477 LASSERT(lu_name_is_valid(lname));
3480 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3481 !strncmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3484 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3485 PFID(fid), PNAME(lname), PFID(mdd_object_fid(tobj)));
3487 pobj = mdd_object_find(env, mdd, fid);
3489 CWARN("%s: cannot find obj "DFID": %ld\n",
3490 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3491 RETURN(PTR_ERR(pobj));
3494 if (!mdd_object_exists(pobj)) {
3495 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3496 mdd_object_put(env, pobj);
3500 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3501 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3503 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(tobj),
3504 mdd_object_type(sobj),
3505 lname->ln_name, handle);
3506 mdd_write_unlock(env, pobj);
3507 mdd_object_put(env, pobj);
3511 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3512 rc = mdo_ref_add(env, tobj, handle);
3513 mdd_write_unlock(env, tobj);
3517 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3518 rc = mdo_ref_del(env, sobj, handle);
3519 mdd_write_unlock(env, sobj);
3524 static inline int mdd_fld_lookup(const struct lu_env *env,
3525 struct mdd_device *mdd,
3526 const struct lu_fid *fid,
3529 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
3530 struct seq_server_site *ss;
3533 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3535 range->lsr_flags = LU_SEQ_RANGE_MDT;
3536 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3540 *mdt_index = range->lsr_index;
3545 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3546 struct mdd_object *sobj,
3547 struct mdd_object *tobj,
3548 const struct lu_name *tname,
3549 const struct lu_fid *tpfid,
3550 const struct lu_name *lname,
3551 const struct lu_fid *fid,
3553 struct thandle *handle)
3555 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3556 __u32 source_mdt_index = *(__u32 *)opaque;
3557 __u32 link_mdt_index;
3563 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3564 !strcmp(tname->ln_name, lname->ln_name))
3567 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3571 RETURN(link_mdt_index == source_mdt_index);
3574 static int mdd_iterate_linkea(const struct lu_env *env,
3575 struct mdd_object *sobj,
3576 struct mdd_object *tobj,
3577 const struct lu_name *tname,
3578 const struct lu_fid *tpfid,
3579 struct linkea_data *ldata,
3581 struct thandle *handle,
3584 struct mdd_thread_info *info = mdd_env_info(env);
3585 char *filename = info->mti_name;
3586 struct lu_name lname;
3593 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3594 linkea_next_entry(ldata)) {
3595 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3598 /* Note: lname might miss \0 at the end */
3599 snprintf(filename, sizeof(info->mti_name), "%.*s",
3600 lname.ln_namelen, lname.ln_name);
3601 lname.ln_name = filename;
3603 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3605 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3613 * Prepare linkea, and check whether file needs migrate: if source still has
3614 * link on source MDT, no need to migrate, just update namespace on source and
3617 * \retval 0 do migrate
3618 * \retval 1 don't migrate
3619 * \retval -errno on failure
3621 static int mdd_migrate_linkea_prepare(const struct lu_env *env,
3622 struct mdd_device *mdd,
3623 struct mdd_object *spobj,
3624 struct mdd_object *tpobj,
3625 struct mdd_object *sobj,
3626 const struct lu_name *lname,
3627 const struct lu_attr *attr,
3628 struct linkea_data *ldata)
3630 __u32 source_mdt_index;
3635 memset(ldata, 0, sizeof(*ldata));
3636 rc = mdd_linkea_prepare(env, sobj, mdd_object_fid(spobj), lname,
3637 mdd_object_fid(tpobj), lname, 1, 0, ldata);
3642 * Then it will check if the file should be migrated. If the file has
3643 * mulitple links, we only need migrate the file if all of its entries
3644 * has been migrated to the remote MDT.
3646 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3649 /* If there are still links locally, don't migrate this file */
3650 LASSERT(ldata->ld_leh != NULL);
3653 * If linkEA is overflow, it means there are some unknown name entries
3654 * under unknown parents, which will prevent the migration.
3656 if (unlikely(ldata->ld_leh->leh_overflow_time))
3659 rc = mdd_fld_lookup(env, mdd, mdd_object_fid(sobj), &source_mdt_index);
3663 rc = mdd_iterate_linkea(env, sobj, NULL, lname, mdd_object_fid(tpobj),
3664 ldata, &source_mdt_index, NULL,
3665 mdd_is_link_on_source_mdt);
3669 static int mdd_declare_migrate_update(const struct lu_env *env,
3670 struct mdd_object *spobj,
3671 struct mdd_object *tpobj,
3672 struct mdd_object *obj,
3673 const struct lu_name *lname,
3674 struct lu_attr *attr,
3675 struct lu_attr *spattr,
3676 struct lu_attr *tpattr,
3677 struct linkea_data *ldata,
3679 struct thandle *handle)
3681 struct mdd_thread_info *info = mdd_env_info(env);
3682 struct lu_attr *la = &info->mti_la_for_fix;
3685 rc = mdo_declare_index_delete(env, spobj, lname->ln_name, handle);
3689 if (S_ISDIR(attr->la_mode)) {
3690 rc = mdo_declare_ref_del(env, spobj, handle);
3695 rc = mdo_declare_index_insert(env, tpobj, mdd_object_fid(obj),
3696 attr->la_mode & S_IFMT,
3697 lname->ln_name, handle);
3701 rc = mdd_declare_links_add(env, obj, handle, ldata);
3705 if (S_ISDIR(attr->la_mode)) {
3706 rc = mdo_declare_ref_add(env, tpobj, handle);
3711 la->la_valid = LA_CTIME | LA_MTIME;
3712 rc = mdo_declare_attr_set(env, spobj, la, handle);
3716 if (tpobj != spobj) {
3717 rc = mdo_declare_attr_set(env, tpobj, la, handle);
3725 static int mdd_declare_migrate_create(const struct lu_env *env,
3726 struct mdd_object *spobj,
3727 struct mdd_object *tpobj,
3728 struct mdd_object *sobj,
3729 struct mdd_object *tobj,
3730 const struct lu_name *lname,
3731 struct lu_attr *spattr,
3732 struct lu_attr *tpattr,
3733 struct lu_attr *attr,
3734 struct lu_buf *sbuf,
3735 struct linkea_data *ldata,
3737 struct md_op_spec *spec,
3738 struct dt_allocation_hint *hint,
3739 struct thandle *handle)
3741 struct mdd_thread_info *info = mdd_env_info(env);
3742 struct md_layout_change *mlc = &info->mti_mlc;
3743 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3748 if (S_ISDIR(attr->la_mode)) {
3749 struct lmv_user_md *lum = spec->u.sp_ea.eadata;
3751 mlc->mlc_opc = MD_LAYOUT_DETACH;
3752 rc = mdo_declare_layout_change(env, sobj, mlc, handle);
3756 lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3757 } else if (S_ISLNK(attr->la_mode)) {
3758 spec->u.sp_symname = sbuf->lb_buf;
3759 } else if (S_ISREG(attr->la_mode)) {
3760 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
3761 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
3764 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
3766 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3767 lname, attr, handle, spec, ldata, NULL, NULL,
3773 * tobj mode will be used in mdo_declare_layout_change(), but it's not
3774 * createb yet, copy from sobj.
3776 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
3777 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
3778 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
3780 if (S_ISDIR(attr->la_mode)) {
3782 /* if sobj is not striped, fake a 1-stripe LMV */
3783 LASSERT(sizeof(info->mti_key) >
3784 lmv_mds_md_size(1, LMV_MAGIC_V1));
3785 lmv = (typeof(lmv))info->mti_key;
3786 memset(lmv, 0, sizeof(*lmv));
3787 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3788 lmv->lmv_stripe_count = cpu_to_le32(1);
3789 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
3790 fid_le_to_cpu(&lmv->lmv_stripe_fids[0],
3791 mdd_object_fid(sobj));
3792 mlc->mlc_buf.lb_buf = lmv;
3793 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3795 mlc->mlc_buf = *sbuf;
3797 mlc->mlc_opc = MD_LAYOUT_ATTACH;
3798 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
3803 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle,
3804 mdo_declare_xattr_set);
3808 if (S_ISREG(attr->la_mode)) {
3809 struct lu_buf fid_buf;
3811 handle->th_complex = 1;
3813 /* target may be remote, update PFID via sobj. */
3814 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
3815 fid_buf.lb_len = sizeof(struct lu_fid);
3816 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
3821 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
3826 if (!S_ISDIR(attr->la_mode)) {
3827 rc = mdd_iterate_linkea(env, sobj, tobj, lname,
3828 mdd_object_fid(tpobj), ldata, NULL,
3829 handle, mdd_declare_update_link);
3834 if (!S_ISDIR(attr->la_mode) || lmv) {
3835 rc = mdo_declare_ref_del(env, sobj, handle);
3839 if (S_ISDIR(attr->la_mode)) {
3840 rc = mdo_declare_ref_del(env, sobj, handle);
3845 rc = mdo_declare_destroy(env, sobj, handle);
3850 rc = mdd_declare_migrate_update(env, spobj, tpobj, tobj, lname, attr,
3851 spattr, tpattr, ldata, ma, handle);
3856 * migrate dirent from \a spobj to \a tpobj.
3858 static int mdd_migrate_update(const struct lu_env *env,
3859 struct mdd_object *spobj,
3860 struct mdd_object *tpobj,
3861 struct mdd_object *obj,
3862 const struct lu_name *lname,
3863 struct lu_attr *attr,
3864 struct lu_attr *spattr,
3865 struct lu_attr *tpattr,
3866 struct linkea_data *ldata,
3868 struct thandle *handle)
3870 struct mdd_thread_info *info = mdd_env_info(env);
3871 struct lu_attr *la = &info->mti_la_for_fix;
3876 CDEBUG(D_INFO, "update "DFID"/%s to "DFID"/"DFID"\n",
3877 PFID(mdd_object_fid(spobj)), lname->ln_name,
3878 PFID(mdd_object_fid(tpobj)), PFID(mdd_object_fid(obj)));
3880 rc = __mdd_index_delete(env, spobj, lname->ln_name,
3881 S_ISDIR(attr->la_mode), handle);
3885 rc = __mdd_index_insert(env, tpobj, mdd_object_fid(obj),
3886 attr->la_mode & S_IFMT,
3887 lname->ln_name, handle);
3891 rc = mdd_links_write(env, obj, ldata, handle);
3895 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3896 la->la_valid = LA_CTIME | LA_MTIME;
3897 mdd_write_lock(env, spobj, DT_SRC_PARENT);
3898 rc = mdd_update_time(env, spobj, spattr, la, handle);
3899 mdd_write_unlock(env, spobj);
3903 if (tpobj != spobj) {
3904 la->la_valid = LA_CTIME | LA_MTIME;
3905 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
3906 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
3907 mdd_write_unlock(env, tpobj);
3916 * Migrate file/dir to target MDT.
3918 * Create target according to \a spec, and then migrate xattrs, if it's
3919 * directory, migrate source stripes to target.
3921 * \param[in] env execution environment
3922 * \param[in] spobj source parent object
3923 * \param[in] tpobj target parent object
3924 * \param[in] sobj source object
3925 * \param[in] tobj target object
3926 * \param[in] lname file name
3927 * \param[in] spattr source parent attributes
3928 * \param[in] tpattr target parent attributes
3929 * \param[in] attr source attributes
3930 * \param[in] sbuf source LMV buf
3931 * \param[in] spec migrate create spec
3932 * \param[in] hint target creation hint
3933 * \param[in] handle tranasction handle
3935 * \retval 0 on success
3936 * \retval -errno on failure
3938 static int mdd_migrate_create(const struct lu_env *env,
3939 struct mdd_object *spobj,
3940 struct mdd_object *tpobj,
3941 struct mdd_object *sobj,
3942 struct mdd_object *tobj,
3943 const struct lu_name *lname,
3944 struct lu_attr *spattr,
3945 struct lu_attr *tpattr,
3946 struct lu_attr *attr,
3947 const struct lu_buf *sbuf,
3948 struct linkea_data *ldata,
3950 struct md_op_spec *spec,
3951 struct dt_allocation_hint *hint,
3952 struct thandle *handle)
3959 * migrate sobj stripes to tobj if it's directory:
3960 * 1. detach stripes from sobj.
3961 * 2. attach stripes to tobj, see mdd_declare_migrate_mdt().
3962 * 3. create stripes for tobj, see lod_xattr_set_lmv().
3964 if (S_ISDIR(attr->la_mode)) {
3965 struct mdd_thread_info *info = mdd_env_info(env);
3966 struct md_layout_change *mlc = &info->mti_mlc;
3968 mlc->mlc_opc = MD_LAYOUT_DETACH;
3970 mdd_write_lock(env, sobj, DT_SRC_PARENT);
3971 rc = mdo_layout_change(env, sobj, mlc, handle);
3972 mdd_write_unlock(env, sobj);
3977 /* don't set nlink from sobj */
3978 attr->la_valid &= ~LA_NLINK;
3980 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
3981 hint, handle, false);
3985 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3986 rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
3987 mdd_write_unlock(env, tobj);
3991 /* for regular file, update OST objects XATTR_NAME_FID */
3992 if (S_ISREG(attr->la_mode)) {
3993 struct lu_buf fid_buf;
3995 /* target may be remote, update PFID via sobj. */
3996 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
3997 fid_buf.lb_len = sizeof(struct lu_fid);
3998 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4003 /* delete LOV to avoid deleting OST objs when destroying sobj */
4004 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4005 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4006 mdd_write_unlock(env, sobj);
4007 /* O_DELAY_CREATE file may not have LOV, ignore -ENODATA */
4008 if (rc && rc != -ENODATA)
4013 /* update links FID */
4014 if (!S_ISDIR(attr->la_mode)) {
4015 rc = mdd_iterate_linkea(env, sobj, tobj, lname,
4016 mdd_object_fid(tpobj), ldata,
4017 NULL, handle, mdd_update_link);
4022 /* don't destroy sobj if it's plain directory */
4023 if (!S_ISDIR(attr->la_mode) || sbuf->lb_buf) {
4024 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4025 rc = mdo_ref_del(env, sobj, handle);
4027 if (S_ISDIR(attr->la_mode))
4028 rc = mdo_ref_del(env, sobj, handle);
4030 rc = mdo_destroy(env, sobj, handle);
4032 mdd_write_unlock(env, sobj);
4037 rc = mdd_migrate_update(env, spobj, tpobj, tobj, lname, attr,
4038 spattr, tpattr, ldata, ma, handle);
4043 static int mdd_migrate_cmd_check(struct mdd_device *mdd,
4044 const struct lmv_mds_md_v1 *lmv,
4045 const struct lmv_user_md_v1 *lum,
4046 const struct lu_name *lname)
4048 __u32 lum_stripe_count = lum->lum_stripe_count;
4049 __u32 lmv_hash_type = lmv->lmv_hash_type;
4050 char *mdt_hash_name[] = { "none",
4051 LMV_HASH_NAME_ALL_CHARS,
4052 LMV_HASH_NAME_FNV_1A_64,
4053 LMV_HASH_NAME_CRUSH,
4056 if (!lmv_is_sane(lmv))
4059 /* if stripe_count unspecified, set to 1 */
4060 if (!lum_stripe_count)
4061 lum_stripe_count = cpu_to_le32(1);
4063 lmv_hash_type &= ~cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4065 /* TODO: check specific MDTs */
4066 if (lum_stripe_count != lmv->lmv_migrate_offset ||
4067 lum->lum_stripe_offset != lmv->lmv_master_mdt_index ||
4068 (lum->lum_hash_type && lum->lum_hash_type != lmv_hash_type)) {
4069 CERROR("%s: '"DNAME"' migration was interrupted, run 'lfs migrate -m %d -c %d -H %s "DNAME"' to finish migration.\n",
4070 mdd2obd_dev(mdd)->obd_name, PNAME(lname),
4071 le32_to_cpu(lmv->lmv_master_mdt_index),
4072 le32_to_cpu(lmv->lmv_migrate_offset),
4073 mdt_hash_name[le32_to_cpu(lmv_hash_type)],
4082 * Migrate directory or file.
4084 * migrate source to target in following steps:
4085 * 1. create target, append source stripes after target's if it's directory,
4086 * migrate xattrs and update fid of source links.
4087 * 2. update namespace: migrate dirent from source parent to target parent,
4088 * update file linkea, and destroy source if it's not needed any more.
4090 * \param[in] env execution environment
4091 * \param[in] md_pobj parent master object
4092 * \param[in] md_sobj source object
4093 * \param[in] lname file name
4094 * \param[in] md_tobj target object
4095 * \param[in] spec target creation spec
4096 * \param[in] ma used to update \a pobj mtime and ctime
4098 * \retval 0 on success
4099 * \retval -errno on failure
4101 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4102 struct md_object *md_sobj, const struct lu_name *lname,
4103 struct md_object *md_tobj, struct md_op_spec *spec,
4106 struct mdd_device *mdd = mdo2mdd(md_pobj);
4107 struct mdd_thread_info *info = mdd_env_info(env);
4108 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4109 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4110 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4111 struct mdd_object *spobj = NULL;
4112 struct mdd_object *tpobj = NULL;
4113 struct lu_attr *spattr = &info->mti_pattr;
4114 struct lu_attr *tpattr = &info->mti_tpattr;
4115 struct lu_attr *attr = &info->mti_cattr;
4116 struct linkea_data *ldata = &info->mti_link_data;
4117 struct dt_allocation_hint *hint = &info->mti_hint;
4118 struct lu_fid *fid = &info->mti_fid2;
4119 struct lu_buf pbuf = { NULL };
4120 struct lu_buf sbuf = { NULL };
4121 struct lmv_mds_md_v1 *lmv;
4122 struct thandle *handle;
4123 bool nsonly = false;
4128 rc = mdd_la_get(env, sobj, attr);
4132 /* locate source and target stripe on pobj, which are the real parent */
4133 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4134 if (rc < 0 && rc != -ENODATA)
4139 __u32 hash_type = le32_to_cpu(lmv->lmv_hash_type);
4142 /* locate target parent stripe */
4143 /* fail check here to make sure top dir migration succeed. */
4144 if ((hash_type & LMV_HASH_FLAG_MIGRATION) &&
4145 OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4146 GOTO(out, rc = -EIO);
4148 index = lmv_name_to_stripe_index(lmv, lname->ln_name,
4151 GOTO(out, rc = index);
4153 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4154 tpobj = mdd_object_find(env, mdd, fid);
4156 GOTO(out, rc = PTR_ERR(tpobj));
4158 /* locate source parent stripe */
4159 if (hash_type & LMV_HASH_FLAG_LAYOUT_CHANGE) {
4160 index = lmv_name_to_stripe_index_old(lmv,
4164 GOTO(out, rc = index);
4166 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4167 spobj = mdd_object_find(env, mdd, fid);
4169 GOTO(out, rc = PTR_ERR(spobj));
4172 mdd_object_get(spobj);
4177 mdd_object_get(tpobj);
4178 mdd_object_get(spobj);
4181 rc = mdd_la_get(env, spobj, spattr);
4185 rc = mdd_la_get(env, tpobj, tpattr);
4189 if (S_ISDIR(attr->la_mode)) {
4190 struct lmv_user_md_v1 *lum = spec->u.sp_ea.eadata;
4194 /* if user use default value '0' for stripe_count, we need to
4195 * adjust it to '1' to create a 1-stripe directory.
4197 if (lum->lum_stripe_count == 0)
4198 lum->lum_stripe_count = cpu_to_le32(1);
4200 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4201 if (rc && rc != -ENODATA)
4206 (le32_to_cpu(lmv->lmv_hash_type) &
4207 LMV_HASH_FLAG_MIGRATION)) {
4208 rc = mdd_migrate_cmd_check(mdd, lmv, lum, lname);
4213 GOTO(out, rc = -EALREADY);
4215 /* update namespace only if @sobj is on MDT where @tpobj is. */
4216 if (!mdd_object_remote(tpobj))
4219 if (S_ISLNK(attr->la_mode)) {
4220 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4222 GOTO(out, rc = -ENOMEM);
4224 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4227 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4228 mdd2obd_dev(mdd)->obd_name,
4229 PFID(mdd_object_fid(sobj)), rc);
4235 /* linkea needs update upon FID or parent stripe change */
4236 rc = mdd_migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, lname,
4239 /* update namespace only if @sobj has link on its MDT. */
4244 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4245 spattr, tpattr, attr);
4249 handle = mdd_trans_create(env, mdd);
4251 GOTO(out, rc = PTR_ERR(handle));
4254 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, lname,
4255 attr, spattr, tpattr, ldata, ma,
4258 rc = mdd_declare_migrate_create(env, spobj, tpobj, sobj, tobj,
4259 lname, spattr, tpattr, attr,
4260 &sbuf, ldata, ma, spec, hint,
4263 GOTO(stop_trans, rc);
4265 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4268 GOTO(stop_trans, rc);
4270 rc = mdd_trans_start(env, mdd, handle);
4272 GOTO(stop_trans, rc);
4275 rc = mdd_migrate_update(env, spobj, tpobj, sobj, lname, attr,
4276 spattr, tpattr, ldata, ma, handle);
4278 rc = mdd_migrate_create(env, spobj, tpobj, sobj, tobj, lname,
4279 spattr, tpattr, attr, &sbuf, ldata, ma,
4280 spec, hint, handle);
4282 GOTO(stop_trans, rc);
4284 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0,
4285 nsonly ? sobj : tobj, mdd_object_fid(spobj),
4286 mdd_object_fid(sobj), mdd_object_fid(tpobj),
4287 lname, lname, handle);
4289 GOTO(stop_trans, rc);
4293 rc = mdd_trans_stop(env, mdd, rc, handle);
4295 if (!IS_ERR_OR_NULL(spobj))
4296 mdd_object_put(env, spobj);
4297 if (!IS_ERR_OR_NULL(tpobj))
4298 mdd_object_put(env, tpobj);
4305 static int mdd_declare_1sd_collapse(const struct lu_env *env,
4306 struct mdd_object *pobj,
4307 struct mdd_object *obj,
4308 struct mdd_object *stripe,
4309 struct lu_attr *attr,
4310 struct md_layout_change *mlc,
4311 struct lu_name *lname,
4312 struct thandle *handle)
4316 mlc->mlc_opc = MD_LAYOUT_DETACH;
4317 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4321 rc = mdo_declare_index_insert(env, stripe, mdd_object_fid(pobj),
4322 S_IFDIR, dotdot, handle);
4326 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle,
4327 mdo_declare_xattr_set);
4331 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4335 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4339 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4343 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(stripe),
4344 attr->la_mode, lname->ln_name, handle);
4348 rc = mdo_declare_ref_del(env, obj, handle);
4352 rc = mdo_declare_ref_del(env, obj, handle);
4356 rc = mdo_declare_destroy(env, obj, handle);
4363 /* transform one-stripe directory to a plain directory */
4364 static int mdd_1sd_collapse(const struct lu_env *env,
4365 struct mdd_object *pobj,
4366 struct mdd_object *obj,
4367 struct mdd_object *stripe,
4368 struct lu_attr *attr,
4369 struct md_layout_change *mlc,
4370 struct lu_name *lname,
4371 struct thandle *handle)
4377 /* replace 1-stripe directory with its stripe */
4378 mlc->mlc_opc = MD_LAYOUT_DETACH;
4380 mdd_write_lock(env, obj, DT_SRC_PARENT);
4381 rc = mdo_layout_change(env, obj, mlc, handle);
4382 mdd_write_unlock(env, obj);
4386 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4387 mdd_write_lock(env, obj, DT_SRC_CHILD);
4389 /* insert dotdot to stripe which points to parent */
4390 rc = __mdd_index_insert_only(env, stripe, mdd_object_fid(pobj),
4391 S_IFDIR, dotdot, handle);
4395 /* copy xattrs including linkea */
4396 rc = mdd_iterate_xattrs(env, obj, stripe, false, handle, mdo_xattr_set);
4401 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4405 /* don't set nlink from parent */
4406 attr->la_valid &= ~LA_NLINK;
4408 rc = mdo_attr_set(env, stripe, attr, handle);
4412 /* delete dir name from parent */
4413 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4417 /* insert stripe to parent with dir name */
4418 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(stripe),
4419 attr->la_mode, lname->ln_name, handle);
4423 /* destroy dir obj */
4424 rc = mdo_ref_del(env, obj, handle);
4428 rc = mdo_ref_del(env, obj, handle);
4432 rc = mdo_destroy(env, obj, handle);
4438 mdd_write_unlock(env, obj);
4439 mdd_write_unlock(env, pobj);
4445 * shrink directory stripes after migration/merge
4447 int mdd_dir_layout_shrink(const struct lu_env *env,
4448 struct md_object *md_obj,
4449 struct md_layout_change *mlc)
4451 struct mdd_device *mdd = mdo2mdd(md_obj);
4452 struct mdd_thread_info *info = mdd_env_info(env);
4453 struct mdd_object *obj = md2mdd_obj(md_obj);
4454 struct mdd_object *pobj = NULL;
4455 struct mdd_object *stripe = NULL;
4456 struct lu_attr *attr = &info->mti_pattr;
4457 struct lu_fid *fid = &info->mti_fid2;
4458 struct lu_name lname = { NULL };
4459 struct lu_buf lmv_buf = { NULL };
4460 struct lmv_mds_md_v1 *lmv;
4461 struct lmv_user_md *lmu;
4462 struct thandle *handle;
4467 rc = mdd_la_get(env, obj, attr);
4471 if (!S_ISDIR(attr->la_mode))
4474 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4478 lmv = lmv_buf.lb_buf;
4479 lmu = mlc->mlc_buf.lb_buf;
4481 /* adjust the default value '0' to '1' */
4482 if (lmu->lum_stripe_count == 0)
4483 lmu->lum_stripe_count = cpu_to_le32(1);
4485 /* this was checked in MDT */
4486 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4487 le32_to_cpu(lmv->lmv_stripe_count));
4490 * if obj stripe count will be shrunk to 1, we need to convert it to a
4491 * normal dir, which will change its fid and update parent namespace,
4492 * get obj name and parent fid from linkea.
4494 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4495 struct linkea_data *ldata = &info->mti_link_data;
4496 char *filename = info->mti_name;
4498 rc = mdd_links_read(env, obj, ldata);
4502 if (ldata->ld_leh->leh_reccount > 1)
4503 GOTO(out, rc = -EINVAL);
4505 linkea_first_entry(ldata);
4507 GOTO(out, rc = -ENODATA);
4509 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4512 /* Note: lname might miss \0 at the end */
4513 snprintf(filename, sizeof(info->mti_name), "%.*s",
4514 lname.ln_namelen, lname.ln_name);
4515 lname.ln_name = filename;
4517 pobj = mdd_object_find(env, mdd, fid);
4524 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4526 stripe = mdd_object_find(env, mdd, fid);
4527 if (IS_ERR(stripe)) {
4528 mdd_object_put(env, pobj);
4530 GOTO(out, rc = PTR_ERR(stripe));
4534 handle = mdd_trans_create(env, mdd);
4536 GOTO(out, rc = PTR_ERR(handle));
4538 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4539 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4541 GOTO(stop_trans, rc);
4543 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4544 rc = mdd_declare_1sd_collapse(env, pobj, obj, stripe, attr, mlc,
4547 GOTO(stop_trans, rc);
4550 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4553 GOTO(stop_trans, rc);
4555 rc = mdd_trans_start(env, mdd, handle);
4557 GOTO(stop_trans, rc);
4559 mdd_write_lock(env, obj, DT_SRC_PARENT);
4560 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4561 rc = mdo_layout_change(env, obj, mlc, handle);
4562 mdd_write_unlock(env, obj);
4564 GOTO(stop_trans, rc);
4566 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4567 rc = mdd_1sd_collapse(env, pobj, obj, stripe, attr, mlc, &lname,
4570 GOTO(stop_trans, rc);
4573 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4574 XATTR_NAME_LMV, handle);
4575 GOTO(stop_trans, rc);
4578 rc = mdd_trans_stop(env, mdd, rc, handle);
4581 mdd_object_put(env, stripe);
4582 mdd_object_put(env, pobj);
4584 lu_buf_free(&lmv_buf);
4588 const struct md_dir_operations mdd_dir_ops = {
4589 .mdo_is_subdir = mdd_is_subdir,
4590 .mdo_lookup = mdd_lookup,
4591 .mdo_create = mdd_create,
4592 .mdo_rename = mdd_rename,
4593 .mdo_link = mdd_link,
4594 .mdo_unlink = mdd_unlink,
4595 .mdo_create_data = mdd_create_data,
4596 .mdo_migrate = mdd_migrate,