4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/mdd/mdd_dir.c
33 * Lustre Metadata Server (mdd) routines
35 * Author: Wang Di <wangdi@intel.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <obd_class.h>
41 #include <obd_support.h>
42 #include <lustre_mds.h>
43 #include <lustre_fid.h>
44 #include <lustre_lmv.h>
45 #include <lustre_idmap.h>
47 #include "mdd_internal.h"
49 static const char dot[] = ".";
50 static const char dotdot[] = "..";
52 static struct lu_name lname_dotdot = {
53 .ln_name = (char *) dotdot,
54 .ln_namelen = sizeof(dotdot) - 1,
58 mdd_name_check(const struct lu_env *env, struct mdd_device *m,
59 const struct lu_name *ln)
61 struct mdd_thread_info *info = mdd_env_info(env);
62 bool enc = info->mdi_pattr.la_valid & LA_FLAGS &&
63 info->mdi_pattr.la_flags & LUSTRE_ENCRYPT_FL;
65 if (!lu_name_is_valid(ln))
67 else if (!enc && ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
73 /* Get FID from name and parent */
75 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
76 const struct lu_attr *pattr, const struct lu_name *lname,
77 struct lu_fid *fid, unsigned int may_mask)
79 const char *name = lname->ln_name;
80 const struct dt_key *key = (const struct dt_key *)name;
81 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
82 struct dt_object *dir = mdd_object_child(mdd_obj);
87 if (unlikely(mdd_is_dead_obj(mdd_obj)))
90 if (!mdd_object_exists(mdd_obj))
93 if (mdd_object_remote(mdd_obj)) {
94 CDEBUG(D_INFO, "%s: Object "DFID" located on remote server\n",
95 mdd_obj_dev_name(mdd_obj),
96 PFID(mdd_object_fid(mdd_obj)));
99 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, may_mask,
104 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
105 dt_try_as_dir(env, dir)))
106 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
113 int mdd_lookup(const struct lu_env *env,
114 struct md_object *pobj, const struct lu_name *lname,
115 struct lu_fid *fid, struct md_op_spec *spec)
117 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
121 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
125 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
126 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
130 /** Read the link EA into a temp buffer.
131 * Uses the mdd_thread_info::mdi_link_buf since it is generally large.
132 * A pointer to the buffer is stored in \a ldata::ld_buf.
136 static int __mdd_links_read(const struct lu_env *env,
137 struct mdd_object *mdd_obj,
138 struct linkea_data *ldata)
142 if (!mdd_object_exists(mdd_obj))
145 /* First try a small buf */
146 LASSERT(env != NULL);
147 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_link_buf,
149 if (ldata->ld_buf->lb_buf == NULL)
152 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
154 /* Buf was too small, figure out what we need. */
155 lu_buf_free(ldata->ld_buf);
156 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
160 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
161 if (ldata->ld_buf->lb_buf == NULL)
163 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
167 lu_buf_free(ldata->ld_buf);
168 ldata->ld_buf = NULL;
172 return linkea_init(ldata);
175 int mdd_links_read(const struct lu_env *env,
176 struct mdd_object *mdd_obj,
177 struct linkea_data *ldata)
181 rc = __mdd_links_read(env, mdd_obj, ldata);
183 rc = linkea_init(ldata);
188 static int mdd_links_read_with_rec(const struct lu_env *env,
189 struct mdd_object *mdd_obj,
190 struct linkea_data *ldata)
194 rc = __mdd_links_read(env, mdd_obj, ldata);
196 rc = linkea_init_with_rec(ldata);
202 * Get parent FID of the directory
204 * Read parent FID from linkEA, if that fails, then do lookup
205 * dotdot to get the parent FID.
207 * \param[in] env execution environment
208 * \param[in] obj object from which to find the parent FID
209 * \param[in] attr attribute of the object
210 * \param[out] fid fid to get the parent FID
212 * \retval 0 if getting the parent FID succeeds.
213 * \retval negative errno if getting the parent FID fails.
215 static inline int mdd_parent_fid(const struct lu_env *env,
216 struct mdd_object *obj,
217 const struct lu_attr *attr,
220 struct mdd_thread_info *info = mdd_env_info(env);
221 struct linkea_data ldata = { NULL };
222 struct lu_buf *buf = &info->mdi_link_buf;
223 struct lu_name lname;
228 LASSERTF(S_ISDIR(mdd_object_type(obj)),
229 "%s: FID "DFID" is not a directory type = %o\n",
230 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)),
231 mdd_object_type(obj));
233 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
234 if (buf->lb_buf == NULL)
235 GOTO(lookup, rc = 0);
238 rc = mdd_links_read_with_rec(env, obj, &ldata);
242 /* the obj is not locked, don't cache attributes */
243 mdd_invalidate(env, &obj->mod_obj);
245 LASSERT(ldata.ld_leh != NULL);
246 /* Directory should only have 1 parent */
247 if (ldata.ld_leh->leh_reccount > 1)
250 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
252 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
253 if (likely(fid_is_sane(fid)))
256 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
261 * For root fid use special function, which does not compare version component
262 * of fid. Version component is different for root fids on all MDTs.
264 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
266 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
267 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
271 * return 1: if \a tfid is the fid of the ancestor of \a mo;
273 * otherwise: values < 0, errors.
275 static int mdd_is_parent(const struct lu_env *env,
276 struct mdd_device *mdd,
277 struct mdd_object *mo,
278 const struct lu_attr *attr,
279 const struct lu_fid *tfid)
281 struct mdd_object *mp;
285 LASSERT(!lu_fid_eq(mdd_object_fid(mo), tfid));
286 pfid = &mdd_env_info(env)->mdi_fid;
288 if (mdd_is_root(mdd, mdd_object_fid(mo)))
291 if (mdd_is_root(mdd, tfid))
294 rc = mdd_parent_fid(env, mo, attr, pfid);
299 if (lu_fid_eq(pfid, tfid))
302 if (mdd_is_root(mdd, pfid))
305 mp = mdd_object_find(env, mdd, pfid);
309 if (!mdd_object_exists(mp)) {
310 mdd_object_put(env, mp);
314 rc = mdd_parent_fid(env, mp, attr, pfid);
315 mdd_object_put(env, mp);
324 * No permission check is needed.
326 * returns 1: if fid is ancestor of @mo;
327 * returns 0: if fid is not an ancestor of @mo;
328 * returns < 0: if error
330 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
331 const struct lu_fid *fid)
333 struct mdd_device *mdd = mdo2mdd(mo);
334 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
338 if (!mdd_object_exists(md2mdd_obj(mo)))
341 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
344 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
348 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid);
353 * Check that @dir contains no entries except (possibly) dot and dotdot.
358 * -ENOTDIR not a directory object
359 * -ENOTEMPTY not empty
363 int mdd_dir_is_empty(const struct lu_env *env, struct mdd_object *dir)
366 struct dt_object *obj;
367 const struct dt_it_ops *iops;
371 obj = mdd_object_child(dir);
372 if (!dt_try_as_dir(env, obj))
375 iops = &obj->do_index_ops->dio_it;
376 it = iops->init(env, obj, LUDA_64BITHASH);
378 result = iops->get(env, it, (const struct dt_key *)"");
381 for (result = 0, i = 0; result == 0 && i < 3; ++i)
382 result = iops->next(env, it);
385 else if (result == 1)
387 } else if (result == 0)
389 * Huh? Index contains no zero key?
396 result = PTR_ERR(it);
397 /* -ENODEV means no valid stripe */
398 if (result == -ENODEV)
405 * Determine if the target object can be hard linked, and right now it only
406 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
407 * directory nlink count might exceed the maximum link count(see
408 * osd_object_ref_add), so it only check nlink for non-directories.
410 * \param[in] env thread environment
411 * \param[in] obj object being linked to
412 * \param[in] la attributes of \a obj
414 * \retval 0 if \a obj can be hard linked
415 * \retval negative error if \a obj is a directory or has too
418 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
419 const struct lu_attr *la)
421 struct mdd_device *m = mdd_obj2mdd_dev(obj);
426 /* Subdir count limitation can be broken through
427 * (see osd_object_ref_add), so only check non-directory here. */
428 if (!S_ISDIR(la->la_mode) &&
429 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
436 * Check whether it may create the cobj under the pobj.
438 * \param[in] env execution environment
439 * \param[in] pobj the parent directory
440 * \param[in] pattr the attribute of the parent directory
441 * \param[in] cobj the child to be created
442 * \param[in] check_perm if check WRITE|EXEC permission for parent
444 * \retval = 0 create the child under this dir is allowed
445 * \retval negative errno create the child under this dir is
448 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
449 const struct lu_attr *pattr, struct mdd_object *cobj,
455 if (cobj && mdd_object_exists(cobj))
458 if (mdd_is_dead_obj(pobj))
462 rc = mdd_permission_internal_locked(env, pobj, pattr,
463 MAY_WRITE | MAY_EXEC,
469 * Check whether can unlink from the pobj in the case of "cobj == NULL".
471 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
472 const struct lu_attr *pattr, const struct lu_attr *attr)
477 if (mdd_is_dead_obj(pobj))
480 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
483 rc = mdd_permission_internal_locked(env, pobj, pattr,
484 MAY_WRITE | MAY_EXEC,
489 if (pattr->la_flags & LUSTRE_APPEND_FL)
496 * pobj == NULL is remote ops case, under such case, pobj's
497 * VTX feature has been checked already, no need check again.
499 static inline int mdd_is_sticky(const struct lu_env *env,
500 struct mdd_object *pobj,
501 const struct lu_attr *pattr,
502 struct mdd_object *cobj,
503 const struct lu_attr *cattr)
505 struct lu_ucred *uc = lu_ucred_assert(env);
508 LASSERT(pattr != NULL);
509 if (!(pattr->la_mode & S_ISVTX) ||
510 (pattr->la_uid == uc->uc_fsuid))
514 LASSERT(cattr != NULL);
515 if (cattr->la_uid == uc->uc_fsuid)
518 return !cap_raised(uc->uc_cap, CAP_FOWNER);
521 static int mdd_may_delete_entry(const struct lu_env *env,
522 struct mdd_object *pobj,
523 const struct lu_attr *pattr,
528 LASSERT(pobj != NULL);
529 if (!mdd_object_exists(pobj))
532 if (mdd_is_dead_obj(pobj))
537 rc = mdd_permission_internal_locked(env, pobj, pattr,
538 MAY_WRITE | MAY_EXEC,
544 if (pattr->la_flags & LUSTRE_APPEND_FL)
551 * Check whether it may delete the cobj from the pobj.
554 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
555 const struct lu_attr *tpattr, struct mdd_object *tobj,
556 const struct lu_attr *tattr, const struct lu_attr *cattr,
557 int check_perm, int check_empty)
563 LASSERT(tpattr != NULL);
564 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
572 if (!mdd_object_exists(tobj))
575 if (mdd_is_dead_obj(tobj))
578 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
581 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
584 /* additional check the rename case */
586 if (S_ISDIR(cattr->la_mode)) {
587 if (!S_ISDIR(tattr->la_mode))
590 if (mdd_is_root(mdo2mdd(&tobj->mod_obj),
591 mdd_object_fid(tobj)))
593 } else if (S_ISDIR(tattr->la_mode))
597 if (S_ISDIR(tattr->la_mode) && check_empty)
598 rc = mdd_dir_is_empty(env, tobj);
604 * Check whether it can create the link file(linked to @src_obj) under
605 * the target directory(@tgt_obj), and src_obj has been locked by
608 * \param[in] env execution environment
609 * \param[in] tgt_obj the target directory
610 * \param[in] tattr attributes of target directory
611 * \param[in] lname the link name
612 * \param[in] src_obj source object for link
613 * \param[in] cattr attributes for source object
615 * \retval = 0 it is allowed to create the link file under tgt_obj
616 * \retval negative error not allowed to create the link file
618 static int mdd_link_sanity_check(const struct lu_env *env,
619 struct mdd_object *tgt_obj,
620 const struct lu_attr *tattr,
621 const struct lu_name *lname,
622 struct mdd_object *src_obj,
623 const struct lu_attr *cattr)
625 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
629 if (!mdd_object_exists(src_obj))
632 if (mdd_is_dead_obj(src_obj))
635 /* Local ops, no lookup before link, check filename length here. */
636 rc = mdd_name_check(env, m, lname);
640 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
643 if (S_ISDIR(mdd_object_type(src_obj)))
646 LASSERT(src_obj != tgt_obj);
647 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
651 rc = __mdd_may_link(env, src_obj, cattr);
656 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
657 const char *name, struct thandle *handle)
659 struct dt_object *next = mdd_object_child(pobj);
663 if (dt_try_as_dir(env, next))
664 rc = dt_delete(env, next, (struct dt_key *)name, handle);
671 static int __mdd_index_insert_only(const struct lu_env *env,
672 struct mdd_object *pobj,
673 const struct lu_fid *lf, __u32 type,
674 const char *name, struct thandle *handle)
676 struct dt_object *next = mdd_object_child(pobj);
680 if (dt_try_as_dir(env, next)) {
681 struct dt_insert_rec *rec = &mdd_env_info(env)->mdi_dt_rec;
684 rec->rec_type = type;
685 rc = dt_insert(env, next, (const struct dt_rec *)rec,
686 (const struct dt_key *)name, handle);
693 /* insert named index, add reference if isdir */
694 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
695 const struct lu_fid *lf, __u32 type,
696 const char *name, struct thandle *handle)
701 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
702 if (rc == 0 && S_ISDIR(type)) {
703 mdd_write_lock(env, pobj, DT_TGT_PARENT);
704 mdo_ref_add(env, pobj, handle);
705 mdd_write_unlock(env, pobj);
711 /* delete named index, drop reference if isdir */
712 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
713 const char *name, int is_dir,
714 struct thandle *handle)
719 rc = __mdd_index_delete_only(env, pobj, name, handle);
720 if (rc == 0 && is_dir) {
721 mdd_write_lock(env, pobj, DT_TGT_PARENT);
722 mdo_ref_del(env, pobj, handle);
723 mdd_write_unlock(env, pobj);
729 static int mdd_llog_record_calc_size(const struct lu_env *env,
730 const struct lu_name *tname,
731 const struct lu_name *sname)
733 const struct lu_ucred *uc = lu_ucred(env);
734 enum changelog_rec_flags clf_flags = CLF_EXTRA_FLAGS;
735 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
738 clf_flags |= CLF_RENAME;
740 if (uc != NULL && uc->uc_jobid[0] != '\0')
741 clf_flags |= CLF_JOBID;
743 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
744 changelog_rec_offset(clf_flags, crfe) +
745 (tname != NULL ? tname->ln_namelen : 0) +
746 (sname != NULL ? 1 + sname->ln_namelen : 0));
749 int mdd_declare_changelog_store(const struct lu_env *env,
750 struct mdd_device *mdd,
751 enum changelog_rec_type type,
752 const struct lu_name *tname,
753 const struct lu_name *sname,
754 struct thandle *handle)
756 struct obd_device *obd = mdd2obd_dev(mdd);
757 struct llog_ctxt *ctxt;
758 struct llog_rec_hdr rec_hdr;
759 struct thandle *llog_th;
762 if (!mdd_changelog_enabled(env, mdd, type))
765 rec_hdr.lrh_len = mdd_llog_record_calc_size(env, tname, sname);
766 rec_hdr.lrh_type = CHANGELOG_REC;
768 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
772 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
774 GOTO(out_put, rc = PTR_ERR(llog_th));
776 rc = llog_declare_add(env, ctxt->loc_handle, &rec_hdr, llog_th);
784 int mdd_changelog_write_rec(const struct lu_env *env,
785 struct llog_handle *loghandle,
786 struct llog_rec_hdr *r,
787 struct llog_cookie *cookie,
788 int idx, struct thandle *th)
792 if (r->lrh_type == CHANGELOG_REC) {
793 struct mdd_device *mdd;
794 struct llog_changelog_rec *rec;
796 mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
797 rec = container_of(r, struct llog_changelog_rec, cr_hdr);
799 spin_lock(&mdd->mdd_cl.mc_lock);
800 rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
801 spin_unlock(&mdd->mdd_cl.mc_lock);
803 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
807 * if current llog is full, we will generate a new
808 * llog, and since it's actually not an error, let's
809 * avoid increasing index so that userspace apps
810 * should not see a gap in the changelog sequence
812 if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
813 spin_lock(&mdd->mdd_cl.mc_lock);
814 ++mdd->mdd_cl.mc_index;
815 spin_unlock(&mdd->mdd_cl.mc_lock);
818 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
825 bool mdd_changelog_need_gc(const struct lu_env *env, struct mdd_device *mdd,
826 struct llog_handle *lgh)
828 unsigned long free_cat_entries = llog_cat_free_space(lgh);
829 struct mdd_changelog *mc = &mdd->mdd_cl;
831 return free_cat_entries <= mdd->mdd_changelog_min_free_cat_entries ||
832 mdd_changelog_is_too_idle(mdd, mc->mc_minrec, mc->mc_mintime) ||
833 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD);
836 /** Add a changelog entry \a rec to the changelog llog
839 * \param handle - currently ignored since llogs start their own transaction;
840 * this will hopefully be fixed in llog rewrite
843 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
844 struct llog_changelog_rec *rec, struct thandle *th)
846 struct obd_device *obd = mdd2obd_dev(mdd);
847 struct llog_ctxt *ctxt;
848 struct thandle *llog_th;
852 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
853 changelog_rec_varsize(&rec->cr));
855 /* llog_lvfs_write_rec sets the llog tail len */
856 rec->cr_hdr.lrh_type = CHANGELOG_REC;
857 rec->cr.cr_time = cl_time();
859 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
863 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
865 GOTO(out_put, rc = PTR_ERR(llog_th));
867 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
868 /* nested journal transaction */
869 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
871 /* time to recover some space ?? */
872 if (likely(!mdd->mdd_changelog_gc ||
873 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
874 mdd->mdd_changelog_min_gc_interval >=
875 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
876 /* save a spin_lock trip */
879 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_IDX_PUMP)) {
880 spin_lock(&mdd->mdd_cl.mc_lock);
881 mdd->mdd_cl.mc_index += cfs_fail_val;
882 spin_unlock(&mdd->mdd_cl.mc_lock);
885 need_gc = mdd_changelog_need_gc(env, mdd, ctxt->loc_handle);
886 spin_lock(&mdd->mdd_cl.mc_lock);
887 if (likely(mdd->mdd_changelog_gc &&
888 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
889 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
890 mdd->mdd_changelog_min_gc_interval)) {
891 if (unlikely(need_gc)) {
892 CWARN("%s:%s starting changelog garbage collection\n",
894 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
896 /* indicate further kthread run will occur outside
897 * right after current journal transaction filling has
900 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
902 /* next check in mdd_changelog_min_gc_interval anyway
904 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
906 spin_unlock(&mdd->mdd_cl.mc_lock);
914 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
915 const struct lu_fid *sfid,
916 const struct lu_fid *spfid,
917 const struct lu_name *sname)
919 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
922 LASSERT(sfid != NULL);
923 LASSERT(spfid != NULL);
924 LASSERT(sname != NULL);
926 extsize = sname->ln_namelen + 1;
928 rnm->cr_sfid = *sfid;
929 rnm->cr_spfid = *spfid;
931 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
932 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
933 rec->cr_namelen += extsize;
936 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
938 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
940 if (jobid == NULL || jobid[0] == '\0')
943 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
946 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
948 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
950 ef->cr_extra_flags = eflags;
953 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
954 __u64 uid, __u64 gid)
956 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
958 uidgid->cr_uid = uid;
959 uidgid->cr_gid = gid;
962 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
965 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
970 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
972 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
974 omd->cr_openflags = flags;
977 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
978 const char *xattr_name)
980 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
982 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
985 /** Store a namespace change changelog record
986 * If this fails, we must fail the whole transaction; we don't
987 * want the change to commit without the log entry.
988 * \param target - mdd_object of change
989 * \param tpfid - target parent dir/object fid
990 * \param sfid - source object fid
991 * \param spfid - source parent fid
992 * \param tname - target name string
993 * \param sname - source name string
994 * \param handle - transaction handle
996 int mdd_changelog_ns_store(const struct lu_env *env,
997 struct mdd_device *mdd,
998 enum changelog_rec_type type,
999 enum changelog_rec_flags clf_flags,
1000 struct mdd_object *target,
1001 const struct lu_fid *tpfid,
1002 const struct lu_fid *sfid,
1003 const struct lu_fid *spfid,
1004 const struct lu_name *tname,
1005 const struct lu_name *sname,
1006 struct thandle *handle)
1008 const struct lu_ucred *uc = lu_ucred(env);
1009 struct llog_changelog_rec *rec;
1012 __u64 xflags = CLFE_INVALID;
1016 if (!mdd_changelog_enabled(env, mdd, type))
1019 LASSERT(tpfid != NULL);
1020 LASSERT(tname != NULL);
1021 LASSERT(handle != NULL);
1023 reclen = mdd_llog_record_calc_size(env, tname, sname);
1024 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
1025 if (buf->lb_buf == NULL)
1029 clf_flags &= CLF_FLAGMASK;
1030 clf_flags |= CLF_EXTRA_FLAGS;
1033 if (uc->uc_jobid[0] != '\0')
1034 clf_flags |= CLF_JOBID;
1035 xflags |= CLFE_UIDGID;
1040 clf_flags |= CLF_RENAME;
1042 clf_flags |= CLF_VERSION;
1044 rec->cr.cr_flags = clf_flags;
1046 if (clf_flags & CLF_EXTRA_FLAGS) {
1047 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1048 if (xflags & CLFE_UIDGID)
1049 mdd_changelog_rec_extra_uidgid(&rec->cr,
1050 uc->uc_uid, uc->uc_gid);
1051 if (xflags & CLFE_NID)
1052 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1055 rec->cr.cr_type = (__u32)type;
1056 rec->cr.cr_pfid = *tpfid;
1057 rec->cr.cr_namelen = tname->ln_namelen;
1058 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1060 if (clf_flags & CLF_RENAME)
1061 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1063 if (clf_flags & CLF_JOBID)
1064 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1066 if (likely(target != NULL)) {
1067 rec->cr.cr_tfid = *mdd_object_fid(target);
1068 target->mod_cltime = ktime_get();
1070 fid_zero(&rec->cr.cr_tfid);
1073 rc = mdd_changelog_store(env, mdd, rec, handle);
1075 CERROR("%s: cannot store changelog record: type = %d, "
1076 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1077 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1078 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1085 static int __mdd_links_add(const struct lu_env *env,
1086 struct mdd_object *mdd_obj,
1087 struct linkea_data *ldata,
1088 const struct lu_name *lname,
1089 const struct lu_fid *pfid,
1090 int first, int check)
1092 /* cattr is set in mdd_link */
1093 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1096 if (ldata->ld_leh == NULL) {
1097 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1101 rc = linkea_data_new(ldata,
1102 &mdd_env_info(env)->mdi_link_buf);
1109 rc = linkea_links_find(ldata, lname, pfid);
1110 if (rc && rc != -ENOENT)
1116 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1117 struct lu_fid *tfid = &mdd_env_info(env)->mdi_fid2;
1121 linkea_add_buf(ldata, lname, tfid, false);
1124 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1125 linkea_add_buf(ldata, lname, pfid, false);
1127 /* For encrypted file, we want to limit number of hard links to what
1128 * linkEA can contain. So ask to return error in case of overflow.
1129 * Currently linkEA stores 4KiB of links, that is 14 NAME_MAX links,
1130 * or 119 16-byte names.
1132 return linkea_add_buf(ldata, lname, pfid,
1133 cattr->la_valid & LA_FLAGS &&
1134 cattr->la_flags & LUSTRE_ENCRYPT_FL);
1137 static int __mdd_links_del(const struct lu_env *env,
1138 struct mdd_object *mdd_obj,
1139 struct linkea_data *ldata,
1140 const struct lu_name *lname,
1141 const struct lu_fid *pfid)
1143 /* cattr is set in mdd_link */
1144 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1147 if (ldata->ld_leh == NULL) {
1148 rc = mdd_links_read(env, mdd_obj, ldata);
1153 rc = linkea_links_find(ldata, lname, pfid);
1157 linkea_del_buf(ldata, lname,
1158 cattr->la_valid & LA_FLAGS &&
1159 cattr->la_flags & LUSTRE_ENCRYPT_FL);
1163 static int mdd_linkea_prepare(const struct lu_env *env,
1164 struct mdd_object *mdd_obj,
1165 const struct lu_fid *oldpfid,
1166 const struct lu_name *oldlname,
1167 const struct lu_fid *newpfid,
1168 const struct lu_name *newlname,
1169 int first, int check,
1170 struct linkea_data *ldata)
1175 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1178 LASSERT(oldpfid != NULL || newpfid != NULL);
1180 if (mdd_obj->mod_flags & DEAD_OBJ)
1181 /* Unnecessary to update linkEA for dead object. */
1184 if (oldpfid != NULL) {
1185 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1187 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1190 /* No changes done. */
1195 /* If renaming, add the new record */
1196 if (newpfid != NULL)
1197 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1203 int mdd_links_rename(const struct lu_env *env,
1204 struct mdd_object *mdd_obj,
1205 const struct lu_fid *oldpfid,
1206 const struct lu_name *oldlname,
1207 const struct lu_fid *newpfid,
1208 const struct lu_name *newlname,
1209 struct thandle *handle,
1210 struct linkea_data *ldata,
1211 int first, int check)
1216 if (ldata == NULL) {
1217 ldata = &mdd_env_info(env)->mdi_link_data;
1218 memset(ldata, 0, sizeof(*ldata));
1219 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1220 newpfid, newlname, first, check, ldata);
1225 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1226 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1232 if (newlname == NULL)
1233 CERROR("link_ea add failed %d "DFID"\n",
1234 rc, PFID(mdd_object_fid(mdd_obj)));
1235 else if (oldpfid == NULL)
1236 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1237 newlname->ln_namelen, newlname->ln_name, rc,
1238 PFID(mdd_object_fid(mdd_obj)));
1239 else if (newpfid == NULL)
1240 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1241 oldlname->ln_namelen, oldlname->ln_name, rc,
1242 PFID(mdd_object_fid(mdd_obj)));
1244 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1245 "\n", oldlname->ln_namelen, oldlname->ln_name,
1246 newlname->ln_namelen, newlname->ln_name, rc,
1247 PFID(mdd_object_fid(mdd_obj)));
1250 if (is_vmalloc_addr(ldata->ld_buf))
1251 /* if we vmalloced a large buffer drop it */
1252 lu_buf_free(ldata->ld_buf);
1257 static inline int mdd_links_add(const struct lu_env *env,
1258 struct mdd_object *mdd_obj,
1259 const struct lu_fid *pfid,
1260 const struct lu_name *lname,
1261 struct thandle *handle,
1262 struct linkea_data *ldata, int first)
1264 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1265 pfid, lname, handle, ldata, first, 0);
1268 static inline int mdd_links_del(const struct lu_env *env,
1269 struct mdd_object *mdd_obj,
1270 const struct lu_fid *pfid,
1271 const struct lu_name *lname,
1272 struct thandle *handle)
1274 return mdd_links_rename(env, mdd_obj, pfid, lname,
1275 NULL, NULL, handle, NULL, 0, 0);
1278 /** Read the link EA into a temp buffer.
1279 * Uses the name_buf since it is generally large.
1280 * \retval IS_ERR err
1281 * \retval ptr to \a lu_buf (always \a mdi_link_buf)
1283 struct lu_buf *mdd_links_get(const struct lu_env *env,
1284 struct mdd_object *mdd_obj)
1286 struct linkea_data ldata = { NULL };
1289 rc = mdd_links_read(env, mdd_obj, &ldata);
1290 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1293 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1294 struct linkea_data *ldata, struct thandle *handle)
1296 const struct lu_buf *buf;
1299 if (ldata == NULL || ldata->ld_buf == NULL ||
1300 ldata->ld_leh == NULL)
1303 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1307 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1308 ldata->ld_leh->leh_len);
1309 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1310 if (unlikely(rc == -ENOSPC)) {
1311 rc = linkea_overflow_shrink(ldata);
1319 static int mdd_declare_links_add(const struct lu_env *env,
1320 struct mdd_object *mdd_obj,
1321 struct thandle *handle,
1322 struct linkea_data *ldata)
1328 if (ldata != NULL && ldata->ld_leh != NULL) {
1329 ea_len = ldata->ld_leh->leh_len;
1330 linkea = ldata->ld_buf->lb_buf;
1332 ea_len = MAX_LINKEA_SIZE;
1336 rc = mdo_declare_xattr_set(env, mdd_obj,
1337 mdd_buf_get_const(env, linkea, ea_len),
1338 XATTR_NAME_LINK, 0, handle);
1343 static inline int mdd_declare_links_del(const struct lu_env *env,
1344 struct mdd_object *c,
1345 struct thandle *handle)
1349 /* For directory, the linkEA will be removed together
1350 * with the object. */
1351 if (!S_ISDIR(mdd_object_type(c)))
1352 rc = mdd_declare_links_add(env, c, handle, NULL);
1357 static int mdd_declare_link(const struct lu_env *env,
1358 struct mdd_device *mdd,
1359 struct mdd_object *p,
1360 struct mdd_object *c,
1361 const struct lu_name *name,
1362 struct thandle *handle,
1364 struct linkea_data *data)
1366 struct lu_fid tfid = *mdd_object_fid(c);
1369 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1370 tfid.f_oid = cfs_fail_val;
1372 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1373 name->ln_name, handle);
1377 rc = mdo_declare_ref_add(env, c, handle);
1381 la->la_valid = LA_CTIME | LA_MTIME;
1382 rc = mdo_declare_attr_set(env, p, la, handle);
1386 la->la_valid = LA_CTIME;
1387 rc = mdo_declare_attr_set(env, c, la, handle);
1391 rc = mdd_declare_links_add(env, c, handle, data);
1395 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1401 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1402 struct md_object *src_obj, const struct lu_name *lname,
1405 const char *name = lname->ln_name;
1406 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
1407 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1408 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1409 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1410 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1411 struct mdd_device *mdd = mdo2mdd(src_obj);
1412 struct thandle *handle;
1413 struct lu_fid *tfid = &mdd_env_info(env)->mdi_fid2;
1414 struct linkea_data *ldata = &mdd_env_info(env)->mdi_link_data;
1418 rc = mdd_la_get(env, mdd_sobj, cattr);
1422 rc = mdd_la_get(env, mdd_tobj, tattr);
1427 * If we are using project inheritance, we only allow hard link
1428 * creation in our tree when the project IDs are the same;
1429 * otherwise the tree quota mechanism could be circumvented.
1431 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1432 (tattr->la_projid != cattr->la_projid))
1435 handle = mdd_trans_create(env, mdd);
1437 GOTO(out_pending, rc = PTR_ERR(handle));
1439 memset(ldata, 0, sizeof(*ldata));
1441 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1442 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1444 /* Note: even this function will change ldata, but it comes from
1445 * thread_info, which is completely temporary and only seen in
1446 * this function, so we do not need reset ldata once it fails.*/
1447 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1448 mdd_object_fid(mdd_tobj), lname, 0, 0, ldata);
1452 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1457 rc = mdd_trans_start(env, mdd, handle);
1461 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1462 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1465 GOTO(out_unlock, rc);
1467 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1468 rc = mdo_ref_add(env, mdd_sobj, handle);
1470 GOTO(out_unlock, rc);
1473 *tfid = *mdd_object_fid(mdd_sobj);
1474 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1475 tfid->f_oid = cfs_fail_val;
1477 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1478 mdd_object_type(mdd_sobj), name, handle);
1480 mdo_ref_del(env, mdd_sobj, handle);
1481 GOTO(out_unlock, rc);
1484 la->la_valid = LA_CTIME | LA_MTIME;
1485 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1487 GOTO(out_unlock, rc);
1489 la->la_valid = LA_CTIME;
1490 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1492 /* Note: The failure of links_add should not cause the
1493 * link failure, so do not check return value. */
1494 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_tobj),
1495 lname, handle, ldata, 0);
1499 mdd_write_unlock(env, mdd_sobj);
1501 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1502 mdd_object_fid(mdd_tobj), NULL,
1503 NULL, lname, NULL, handle);
1505 rc = mdd_trans_stop(env, mdd, rc, handle);
1506 if (is_vmalloc_addr(ldata->ld_buf))
1507 /* if we vmalloced a large buffer drop it */
1508 lu_buf_free(ldata->ld_buf);
1513 static int mdd_mark_orphan_object(const struct lu_env *env,
1514 struct mdd_object *obj, struct thandle *handle,
1517 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1520 attr->la_valid = LA_FLAGS;
1521 attr->la_flags = LUSTRE_ORPHAN_FL;
1524 rc = mdo_declare_attr_set(env, obj, attr, handle);
1526 rc = mdo_attr_set(env, obj, attr, handle);
1531 static int mdd_declare_finish_unlink(const struct lu_env *env,
1532 struct mdd_object *obj,
1533 struct thandle *handle)
1537 /* Sigh, we do not know if the unlink object will become orphan in
1538 * declare phase, but fortunately the flags here does not matter
1539 * in current declare implementation */
1540 rc = mdd_mark_orphan_object(env, obj, handle, true);
1544 rc = mdo_declare_destroy(env, obj, handle);
1548 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1552 return mdd_declare_links_del(env, obj, handle);
1555 /* caller should take a lock before calling */
1556 int mdd_finish_unlink(const struct lu_env *env,
1557 struct mdd_object *obj, struct md_attr *ma,
1558 struct mdd_object *pobj,
1559 const struct lu_name *lname,
1563 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1566 LASSERT(mdd_write_locked(env, obj) != 0);
1568 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1569 /* add new orphan and the object
1570 * will be deleted during mdd_close() */
1571 obj->mod_flags |= DEAD_OBJ;
1572 if (obj->mod_count) {
1573 rc = mdd_orphan_insert(env, obj, th);
1575 CDEBUG(D_HA, "Object "DFID" is inserted into "
1576 "orphan list, open count = %d\n",
1577 PFID(mdd_object_fid(obj)),
1580 CERROR("Object "DFID" fail to be an orphan, "
1581 "open count = %d, maybe cause failed "
1583 PFID(mdd_object_fid(obj)),
1586 /* mark object as an orphan here, not
1587 * before mdd_orphan_insert() as racing
1588 * mdd_la_get() may propagate ORPHAN_OBJ
1589 * causing the asserition */
1590 rc = mdd_mark_orphan_object(env, obj, th, false);
1592 rc = mdo_destroy(env, obj, th);
1594 } else if (!is_dir) {
1595 /* old files may not have link ea; ignore errors */
1596 mdd_links_del(env, obj, mdd_object_fid(pobj), lname, th);
1604 * has mdd_write_lock on cobj already, but not on pobj yet
1606 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1607 const struct lu_attr *pattr,
1608 struct mdd_object *cobj,
1609 const struct lu_attr *cattr)
1614 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1619 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1620 struct mdd_object *p, struct mdd_object *c,
1621 const struct lu_name *name, struct md_attr *ma,
1622 struct thandle *handle, int no_name, int is_dir)
1624 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
1627 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1628 if (likely(no_name == 0)) {
1629 rc = mdo_declare_index_delete(env, p, name->ln_name,
1636 rc = mdo_declare_ref_del(env, p, handle);
1642 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1643 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1644 la->la_valid = LA_CTIME | LA_MTIME;
1645 rc = mdo_declare_attr_set(env, p, la, handle);
1650 rc = mdo_declare_ref_del(env, c, handle);
1654 rc = mdo_declare_ref_del(env, c, handle);
1658 la->la_valid = LA_CTIME;
1659 rc = mdo_declare_attr_set(env, c, la, handle);
1663 rc = mdd_declare_finish_unlink(env, c, handle);
1667 /* FIXME: need changelog for remove entry */
1668 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1676 * test if a file has an HSM archive
1677 * if HSM attributes are not found in ma update them from
1680 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1681 struct mdd_object *obj,
1686 if (!(ma->ma_valid & MA_HSM)) {
1687 /* no HSM MD provided, read xattr */
1688 struct lu_buf *hsm_buf;
1689 const size_t buflen = sizeof(struct hsm_attrs);
1692 hsm_buf = mdd_buf_get(env, NULL, 0);
1693 lu_buf_alloc(hsm_buf, buflen);
1694 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1695 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1696 lu_buf_free(hsm_buf);
1700 ma->ma_valid |= MA_HSM;
1702 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1708 * Delete name entry and the object.
1709 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1710 * does not exist for this object, and it could only happen during resending
1711 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1712 * is also needed in this case(needed by changelog), so we have to add another
1713 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1714 * the ENOENT failure should be able to be fixed by redo mechanism.
1716 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1717 struct md_object *cobj, const struct lu_name *lname,
1718 struct md_attr *ma, int no_name)
1720 char *name = (char *)lname->ln_name;
1721 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1722 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1723 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
1724 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1725 struct mdd_object *mdd_cobj = NULL;
1726 struct mdd_device *mdd = mdo2mdd(pobj);
1727 struct thandle *handle;
1728 int rc, is_dir = 0, cl_flags = 0;
1731 /* let shutdown to start */
1732 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1734 /* cobj == NULL means only delete name entry */
1735 if (likely(cobj != NULL)) {
1736 mdd_cobj = md2mdd_obj(cobj);
1737 if (mdd_object_exists(mdd_cobj) == 0)
1741 rc = mdd_la_get(env, mdd_pobj, pattr);
1745 if (likely(mdd_cobj != NULL)) {
1747 rc = mdd_la_get(env, mdd_cobj, cattr);
1751 is_dir = S_ISDIR(cattr->la_mode);
1752 /* search for an existing archive.
1753 * we should check ahead as the object
1754 * can be destroyed in this transaction */
1755 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1756 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1759 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1763 handle = mdd_trans_create(env, mdd);
1765 RETURN(PTR_ERR(handle));
1767 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1768 lname, ma, handle, no_name, is_dir);
1772 rc = mdd_trans_start(env, mdd, handle);
1776 if (likely(mdd_cobj != NULL))
1777 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1779 if (lname->ln_name[lname->ln_namelen] != '\0') {
1780 /* lname->ln_name is not necessarily NUL terminated */
1781 name = kmalloc(lname->ln_namelen + 1, GFP_NOFS);
1783 GOTO(cleanup, rc = -ENOMEM);
1785 memcpy(name, lname->ln_name, lname->ln_namelen);
1786 name[lname->ln_namelen] = '\0';
1789 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1790 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1795 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1796 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1797 GOTO(cleanup, rc = 0);
1799 if (likely(mdd_cobj != NULL)) {
1800 rc = mdo_ref_del(env, mdd_cobj, handle);
1802 __mdd_index_insert_only(env, mdd_pobj,
1803 mdd_object_fid(mdd_cobj),
1804 mdd_object_type(mdd_cobj),
1811 mdo_ref_del(env, mdd_cobj, handle);
1813 /* fetch updated nlink */
1814 rc = mdd_la_get(env, mdd_cobj, cattr);
1819 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1820 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1822 la->la_valid = LA_CTIME | LA_MTIME;
1823 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1827 /* Enough for only unlink the entry */
1828 if (unlikely(mdd_cobj == NULL))
1831 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1832 /* update ctime of an unlinked file only if it is still
1833 * opened or a link still exists */
1834 la->la_valid = LA_CTIME;
1835 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1840 /* XXX: this transfer to ma will be removed with LOD/OSP */
1841 ma->ma_attr = *cattr;
1842 ma->ma_valid |= MA_INODE;
1843 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1847 /* fetch updated nlink */
1848 rc = mdd_la_get(env, mdd_cobj, cattr);
1849 /* if object is removed then we can't get its attrs,
1851 if (rc == -ENOENT) {
1852 cattr->la_nlink = 0;
1856 if (cattr->la_nlink == 0) {
1857 ma->ma_attr = *cattr;
1858 ma->ma_valid |= MA_INODE;
1863 if (name != lname->ln_name)
1866 if (likely(mdd_cobj != NULL))
1867 mdd_write_unlock(env, mdd_cobj);
1870 if (cattr->la_nlink == 0)
1871 cl_flags |= CLF_UNLINK_LAST;
1873 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1875 rc = mdd_changelog_ns_store(env, mdd,
1876 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1877 mdd_cobj, mdd_object_fid(mdd_pobj), NULL, NULL,
1878 lname, NULL, handle);
1882 rc = mdd_trans_stop(env, mdd, rc, handle);
1888 * The permission has been checked when obj created, no need check again.
1890 static int mdd_cd_sanity_check(const struct lu_env *env,
1891 struct mdd_object *obj)
1896 if (!obj || mdd_is_dead_obj(obj))
1902 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1903 struct md_object *cobj,
1904 const struct md_op_spec *spec, struct md_attr *ma)
1906 struct mdd_device *mdd = mdo2mdd(cobj);
1907 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1908 struct mdd_object *son = md2mdd_obj(cobj);
1909 struct thandle *handle;
1910 const struct lu_buf *buf;
1911 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1912 struct dt_allocation_hint *hint = &mdd_env_info(env)->mdi_hint;
1916 rc = mdd_cd_sanity_check(env, son);
1920 if (!md_should_create(spec->sp_cr_flags))
1924 * there are following use cases for this function:
1925 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1926 * striping can be specified or not
1929 rc = mdd_la_get(env, son, attr);
1933 /* calling ->ah_make_hint() is used to transfer information from parent */
1934 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1936 handle = mdd_trans_create(env, mdd);
1938 GOTO(out_free, rc = PTR_ERR(handle));
1941 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1942 * Should this be fixed?
1944 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1945 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1946 spec->sp_cr_flags, spec->no_create);
1948 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1949 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1950 spec->u.sp_ea.eadatalen);
1955 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1956 XATTR_NAME_LOV, 0, handle);
1960 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1965 rc = mdd_trans_start(env, mdd, handle);
1969 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1975 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle,
1979 rc = mdd_trans_stop(env, mdd, rc, handle);
1985 static int mdd_declare_object_initialize(const struct lu_env *env,
1986 struct mdd_object *parent,
1987 struct mdd_object *child,
1988 const struct lu_attr *attr,
1989 struct thandle *handle)
1994 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1995 if (!S_ISDIR(attr->la_mode))
1998 rc = mdo_declare_index_insert(env, child, mdd_object_fid(child),
1999 S_IFDIR, dot, handle);
2003 rc = mdo_declare_ref_add(env, child, handle);
2007 rc = mdo_declare_index_insert(env, child, mdd_object_fid(parent),
2008 S_IFDIR, dotdot, handle);
2013 static int mdd_object_initialize(const struct lu_env *env,
2014 const struct lu_fid *pfid,
2015 struct mdd_object *child,
2016 struct lu_attr *attr,
2017 struct thandle *handle)
2022 if (S_ISDIR(attr->la_mode)) {
2023 /* Add "." and ".." for newly created dir */
2024 mdo_ref_add(env, child, handle);
2025 rc = __mdd_index_insert_only(env, child, mdd_object_fid(child),
2026 S_IFDIR, dot, handle);
2028 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2031 mdo_ref_del(env, child, handle);
2038 * This function checks whether it can create a file/dir under the
2039 * directory(@pobj). The directory(@pobj) is not being locked by
2042 * \param[in] env execution environment
2043 * \param[in] pobj the directory to create files
2044 * \param[in] pattr the attributes of the directory
2045 * \param[in] lname the name of the created file/dir
2046 * \param[in] cattr the attributes of the file/dir
2047 * \param[in] spec create specification
2049 * \retval = 0 it is allowed to create file/dir under
2051 * \retval negative error not allowed to create file/dir
2052 * under the directory
2054 static int mdd_create_sanity_check(const struct lu_env *env,
2055 struct md_object *pobj,
2056 const struct lu_attr *pattr,
2057 const struct lu_name *lname,
2058 struct lu_attr *cattr,
2059 struct md_op_spec *spec)
2061 struct mdd_thread_info *info = mdd_env_info(env);
2062 struct lu_fid *fid = &info->mdi_fid;
2063 struct mdd_object *obj = md2mdd_obj(pobj);
2064 struct mdd_device *m = mdo2mdd(pobj);
2065 bool check_perm = true;
2070 if (mdd_is_dead_obj(obj))
2074 * In some cases this lookup is not needed - we know before if name
2075 * exists or not because MDT performs lookup for it.
2076 * name length check is done in lookup.
2078 if (spec->sp_cr_lookup) {
2080 * Check if the name already exist, though it will be checked in
2081 * _index_insert also, for avoiding rolling back if exists
2084 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2085 MAY_WRITE | MAY_EXEC);
2087 RETURN(rc ? : -EEXIST);
2089 /* Permission is already being checked in mdd_lookup */
2093 if (S_ISDIR(cattr->la_mode) &&
2094 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2095 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2096 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2098 if (!lmv_user_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2099 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2101 CERROR("%s: invalid lmv_user_md: magic=%x hash=%x stripe_offset=%d stripe_count=%u: rc = %d\n",
2102 mdd2obd_dev(m)->obd_name,
2103 le32_to_cpu(lum->lum_magic),
2104 le32_to_cpu(lum->lum_hash_type),
2105 (int)le32_to_cpu(lum->lum_stripe_offset),
2106 le32_to_cpu(lum->lum_stripe_count), rc);
2112 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2117 if (pattr->la_mode & S_ISGID) {
2118 struct lu_ucred *uc = lu_ucred(env);
2120 cattr->la_gid = pattr->la_gid;
2122 /* Directories are special, and always inherit S_ISGID */
2123 if (S_ISDIR(cattr->la_mode)) {
2124 cattr->la_mode |= S_ISGID;
2125 cattr->la_valid |= LA_MODE;
2126 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2127 == (S_ISGID | S_IXGRP) &&
2128 !lustre_in_group_p(uc,
2129 (cattr->la_valid & LA_GID) ?
2130 cattr->la_gid : pattr->la_gid) &&
2131 !cap_raised(uc->uc_cap, CAP_FSETID)) {
2132 cattr->la_mode &= ~S_ISGID;
2133 cattr->la_valid |= LA_MODE;
2137 /* Inherit project ID from parent directory */
2138 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2139 cattr->la_projid = pattr->la_projid;
2140 if (S_ISDIR(cattr->la_mode)) {
2141 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2142 cattr->la_valid |= LA_FLAGS;
2144 cattr->la_valid |= LA_PROJID;
2147 rc = mdd_name_check(env, m, lname);
2151 switch (cattr->la_mode & S_IFMT) {
2153 unsigned int symlen = spec->u.sp_symname.ln_namelen + 1;
2155 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2156 RETURN(-ENAMETOOLONG);
2175 static int mdd_declare_create_object(const struct lu_env *env,
2176 struct mdd_device *mdd,
2177 struct mdd_object *p, struct mdd_object *c,
2178 struct lu_attr *attr,
2179 struct thandle *handle,
2180 const struct md_op_spec *spec,
2181 struct lu_buf *def_acl_buf,
2182 struct lu_buf *acl_buf,
2183 struct lu_buf *hsm_buf,
2184 struct dt_allocation_hint *hint)
2186 const struct lu_buf *buf;
2189 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2190 /* ldiskfs OSD needs this information for credit allocation */
2192 hint->dah_acl_len = def_acl_buf->lb_len;
2194 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2199 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2200 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2201 /* if dir, then can inherit default ACl */
2202 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2203 XATTR_NAME_ACL_DEFAULT,
2209 if (acl_buf && acl_buf->lb_len > 0) {
2210 rc = mdo_declare_attr_set(env, c, attr, handle);
2214 rc = mdo_declare_xattr_set(env, c, acl_buf,
2215 XATTR_NAME_ACL_ACCESS, 0, handle);
2220 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2224 /* replay case, create LOV EA from client data */
2225 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2226 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2227 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2228 spec->u.sp_ea.eadatalen);
2229 rc = mdo_declare_xattr_set(env, c, buf,
2230 S_ISDIR(attr->la_mode) ?
2231 XATTR_NAME_LMV : XATTR_NAME_LOV,
2232 LU_XATTR_CREATE, handle);
2236 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2237 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2245 if (S_ISLNK(attr->la_mode)) {
2246 const char *target_name = spec->u.sp_symname.ln_name;
2247 int sym_len = spec->u.sp_symname.ln_namelen;
2248 const struct lu_buf *buf;
2250 buf = mdd_buf_get_const(env, target_name, sym_len);
2251 rc = dt_declare_record_write(env, mdd_object_child(c),
2257 if (spec->sp_cr_file_secctx_name != NULL) {
2258 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2259 spec->sp_cr_file_secctx_size);
2260 rc = mdo_declare_xattr_set(env, c, buf,
2261 spec->sp_cr_file_secctx_name, 0,
2267 if (spec->sp_cr_file_encctx != NULL) {
2268 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2269 spec->sp_cr_file_encctx_size);
2270 rc = mdo_declare_xattr_set(env, c, buf,
2271 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2280 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2281 struct mdd_object *p, struct mdd_object *c,
2282 const struct lu_name *name,
2283 struct lu_attr *attr,
2284 struct thandle *handle,
2285 const struct md_op_spec *spec,
2286 struct linkea_data *ldata,
2287 struct lu_buf *def_acl_buf,
2288 struct lu_buf *acl_buf,
2289 struct lu_buf *hsm_buf,
2290 struct dt_allocation_hint *hint)
2294 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2295 def_acl_buf, acl_buf, hsm_buf, hint);
2299 if (S_ISDIR(attr->la_mode)) {
2300 rc = mdo_declare_ref_add(env, p, handle);
2305 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2306 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2310 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
2311 enum changelog_rec_type type;
2313 rc = mdo_declare_index_insert(env, p, mdd_object_fid(c),
2314 attr->la_mode, name->ln_name,
2319 rc = mdd_declare_links_add(env, c, handle, ldata);
2324 la->la_valid = LA_CTIME | LA_MTIME;
2325 rc = mdo_declare_attr_set(env, p, la, handle);
2329 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2330 S_ISREG(attr->la_mode) ? CL_CREATE :
2331 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2333 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2342 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2343 struct lu_attr *la, struct lu_buf *def_acl_buf,
2344 struct lu_buf *acl_buf)
2350 if (S_ISLNK(la->la_mode)) {
2351 acl_buf->lb_len = 0;
2352 def_acl_buf->lb_len = 0;
2356 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2357 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2358 XATTR_NAME_ACL_DEFAULT);
2359 mdd_read_unlock(env, pobj);
2361 /* ACL buffer size is not enough, need realloc */
2362 if (rc > acl_buf->lb_len)
2365 /* If there are default ACL, fix mode/ACL by default ACL */
2366 def_acl_buf->lb_len = rc;
2367 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2368 acl_buf->lb_len = rc;
2369 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2372 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2373 /* If there are no default ACL, fix mode by mask */
2374 struct lu_ucred *uc = lu_ucred(env);
2376 /* The create triggered by MDT internal events, such as
2377 * LFSCK reset, will not contain valid "uc". */
2378 if (unlikely(uc != NULL))
2379 la->la_mode &= ~uc->uc_umask;
2381 acl_buf->lb_len = 0;
2382 def_acl_buf->lb_len = 0;
2389 * Create a metadata object and initialize it, set acl, xattr.
2391 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2392 struct mdd_object *son, struct lu_attr *attr,
2393 struct md_op_spec *spec, struct lu_buf *acl_buf,
2394 struct lu_buf *def_acl_buf,
2395 struct lu_buf *hsm_buf,
2396 struct dt_allocation_hint *hint,
2397 struct thandle *handle, bool initsecctx)
2399 const struct lu_buf *buf;
2402 mdd_write_lock(env, son, DT_TGT_CHILD);
2403 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2408 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2409 * created in declare phase, they also needs to be added to master
2410 * object as sub-directory entry. So it has to initialize the master
2411 * object, then set dir striped EA.(in mdo_xattr_set) */
2412 rc = mdd_object_initialize(env, mdd_object_fid(pobj), son, attr,
2415 GOTO(err_destroy, rc);
2418 * in case of replay we just set LOVEA provided by the client
2419 * XXX: I think it would be interesting to try "old" way where
2420 * MDT calls this xattr_set(LOV) in a different transaction.
2421 * probably this way we code can be made better.
2424 /* During creation, there are only a few cases we need do xattr_set to
2426 * 1. regular file: see comments above.
2427 * 2. dir: inherit default striping or pool settings from parent.
2428 * 3. create striped directory with provided stripeEA.
2429 * 4. create striped directory because inherit default layout from the
2432 if (spec->no_create ||
2433 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2434 S_ISDIR(attr->la_mode)) {
2435 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2436 spec->u.sp_ea.eadatalen);
2437 rc = mdo_xattr_set(env, son, buf,
2438 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2440 LU_XATTR_CREATE, handle);
2442 GOTO(err_destroy, rc);
2445 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2448 memset(&mh, 0, sizeof(mh));
2449 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2450 mh.mh_arch_id = spec->sp_archive_id;
2451 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2452 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2455 GOTO(err_destroy, rc);
2458 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2459 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2460 S_ISDIR(attr->la_mode)) {
2461 /* set default acl */
2462 rc = mdo_xattr_set(env, son, def_acl_buf,
2463 XATTR_NAME_ACL_DEFAULT, 0,
2466 GOTO(err_destroy, rc);
2468 /* set its own acl */
2469 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2470 rc = mdo_xattr_set(env, son, acl_buf,
2471 XATTR_NAME_ACL_ACCESS,
2474 GOTO(err_destroy, rc);
2478 if (S_ISLNK(attr->la_mode)) {
2479 struct dt_object *dt = mdd_object_child(son);
2480 const char *target_name = spec->u.sp_symname.ln_name;
2481 int sym_len = spec->u.sp_symname.ln_namelen;
2484 buf = mdd_buf_get_const(env, target_name, sym_len);
2485 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2489 GOTO(err_initlized, rc = -EFAULT);
2492 if (initsecctx && spec->sp_cr_file_secctx_name != NULL) {
2493 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2494 spec->sp_cr_file_secctx_size);
2495 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2498 GOTO(err_initlized, rc);
2501 if (spec->sp_cr_file_encctx != NULL) {
2502 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2503 spec->sp_cr_file_encctx_size);
2504 rc = mdo_xattr_set(env, son, buf,
2505 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2508 GOTO(err_initlized, rc);
2512 if (unlikely(rc != 0)) {
2514 if (S_ISDIR(attr->la_mode)) {
2515 /* Drop the reference, no need to delete "."/"..",
2516 * because the object to be destroied directly. */
2517 rc2 = mdo_ref_del(env, son, handle);
2521 rc2 = mdo_ref_del(env, son, handle);
2525 mdo_destroy(env, son, handle);
2528 mdd_write_unlock(env, son);
2532 static int mdd_index_delete(const struct lu_env *env,
2533 struct mdd_object *mdd_pobj,
2534 struct lu_attr *cattr,
2535 const struct lu_name *lname)
2537 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2538 struct thandle *handle;
2542 handle = mdd_trans_create(env, mdd);
2544 RETURN(PTR_ERR(handle));
2546 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2551 if (S_ISDIR(cattr->la_mode)) {
2552 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2557 /* Since this will only be used in the error handler path,
2558 * Let's set the thandle to be local and not mess the transno */
2559 handle->th_local = 1;
2560 rc = mdd_trans_start(env, mdd, handle);
2564 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2565 S_ISDIR(cattr->la_mode), handle);
2569 rc = mdd_trans_stop(env, mdd, rc, handle);
2575 * Create object and insert it into namespace.
2577 * Two operations have to be performed:
2579 * - an allocation of a new object (->do_create()), and
2580 * - an insertion into a parent index (->dio_insert()).
2582 * Due to locking, operation order is not important, when both are
2583 * successful, *but* error handling cases are quite different:
2585 * - if insertion is done first, and following object creation fails,
2586 * insertion has to be rolled back, but this operation might fail
2587 * also leaving us with dangling index entry.
2589 * - if creation is done first, is has to be undone if insertion fails,
2590 * leaving us with leaked space, which is not good but not fatal.
2592 * It seems that creation-first is simplest solution, but it is sub-optimal
2598 * case, because second mkdir is bound to create object, only to
2599 * destroy it immediately.
2601 * To avoid this follow local file systems that do double lookup:
2603 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2604 * 1. create (mdd_create_object_internal())
2605 * 2. insert (__mdd_index_insert(), lookup again)
2607 * \param[in] pobj parent object
2608 * \param[in] lname name of child being created
2609 * \param[in,out] child child object being created
2610 * \param[in] spec additional create parameters
2611 * \param[in] ma attributes for new child object
2613 * \retval 0 on success
2614 * \retval negative errno on failure
2616 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2617 const struct lu_name *lname, struct md_object *child,
2618 struct md_op_spec *spec, struct md_attr *ma)
2620 struct mdd_thread_info *info = mdd_env_info(env);
2621 struct lu_attr *la = &info->mdi_la_for_fix;
2622 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2623 struct mdd_object *son = md2mdd_obj(child);
2624 struct mdd_device *mdd = mdo2mdd(pobj);
2625 struct lu_attr *attr = &ma->ma_attr;
2626 struct thandle *handle;
2627 struct lu_attr *pattr = &info->mdi_pattr;
2628 struct lu_buf acl_buf;
2629 struct lu_buf def_acl_buf;
2630 struct lu_buf hsm_buf;
2631 struct linkea_data *ldata = &info->mdi_link_data;
2632 const char *name = lname->ln_name;
2633 struct dt_allocation_hint *hint = &mdd_env_info(env)->mdi_hint;
2634 int acl_size = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
2639 rc = mdd_la_get(env, mdd_pobj, pattr);
2643 /* Sanity checks before big job. */
2644 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2648 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2649 GOTO(out_free, rc = -EINPROGRESS);
2651 handle = mdd_trans_create(env, mdd);
2653 GOTO(out_free, rc = PTR_ERR(handle));
2656 acl_buf = *lu_buf_check_and_alloc(&info->mdi_xattr_buf, acl_size);
2657 if (!acl_buf.lb_buf)
2658 GOTO(out_stop, rc = -ENOMEM);
2660 def_acl_buf = *lu_buf_check_and_alloc(&info->mdi_big_buf, acl_size);
2661 if (!def_acl_buf.lb_buf)
2662 GOTO(out_stop, rc = -ENOMEM);
2664 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2665 if (unlikely(rc == -ERANGE &&
2666 acl_size == LUSTRE_POSIX_ACL_MAX_SIZE_OLD)) {
2667 /* use maximum-sized xattr buffer for too-big default ACL */
2668 acl_size = min_t(unsigned int, mdd->mdd_dt_conf.ddp_max_ea_size,
2670 goto use_bigger_buffer;
2675 if (S_ISDIR(attr->la_mode)) {
2676 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2679 * migrate may create 1-stripe directory, so lod_ah_init()
2680 * doesn't adjust stripe count from lmu.
2682 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1))
2683 lmu->lum_stripe_count = 0;
2686 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2688 memset(ldata, 0, sizeof(*ldata));
2689 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2690 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2693 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2694 &tfid, lname, 1, 0, ldata);
2696 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2697 mdd_object_fid(mdd_pobj),
2698 lname, 1, 0, ldata);
2701 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2702 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2704 memset(&hsm_buf, 0, sizeof(hsm_buf));
2705 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2706 if (hsm_buf.lb_buf == NULL)
2707 GOTO(out_stop, rc = -ENOMEM);
2710 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2711 handle, spec, ldata, &def_acl_buf, &acl_buf,
2716 rc = mdd_trans_start(env, mdd, handle);
2720 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2721 &def_acl_buf, &hsm_buf, hint, handle, true);
2725 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2726 mdd_write_lock(env, son, DT_TGT_CHILD);
2727 son->mod_flags |= VOLATILE_OBJ;
2728 rc = mdd_orphan_insert(env, son, handle);
2729 GOTO(out_volatile, rc);
2731 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(son),
2732 attr->la_mode, name, handle);
2734 GOTO(err_created, rc);
2736 mdd_links_add(env, son, mdd_object_fid(mdd_pobj), lname,
2739 /* update parent directory mtime/ctime */
2741 la->la_valid = LA_CTIME | LA_MTIME;
2742 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2744 GOTO(err_insert, rc);
2750 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2751 rc2 = mdd_orphan_delete(env, son, handle);
2753 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2754 S_ISDIR(attr->la_mode),
2760 mdd_write_lock(env, son, DT_TGT_CHILD);
2761 if (S_ISDIR(attr->la_mode)) {
2762 /* Drop the reference, no need to delete "."/"..",
2763 * because the object is to be destroyed directly. */
2764 rc2 = mdo_ref_del(env, son, handle);
2766 mdd_write_unlock(env, son);
2771 /* For volatile files drop one link immediately, since there is
2772 * no filename in the namespace, and save any error returned. */
2773 rc2 = mdo_ref_del(env, son, handle);
2775 mdd_write_unlock(env, son);
2776 if (unlikely(rc == 0))
2781 /* Don't destroy the volatile object on success */
2782 if (likely(rc != 0))
2783 mdo_destroy(env, son, handle);
2784 mdd_write_unlock(env, son);
2787 if (rc == 0 && fid_is_namespace_visible(mdd_object_fid(son)) &&
2788 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2789 rc = mdd_changelog_ns_store(env, mdd,
2790 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2791 S_ISREG(attr->la_mode) ? CL_CREATE :
2792 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2793 0, son, mdd_object_fid(mdd_pobj), NULL, NULL,
2794 lname, NULL, handle);
2796 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2798 /* If creation fails, it is most likely due to the remote update
2799 * failure, because local transaction will mostly succeed at
2800 * this stage. There is no easy way to rollback all of previous
2801 * updates, so let's remove the object from namespace, and
2802 * LFSCK should handle the orphan object. */
2803 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2804 mdd_index_delete(env, mdd_pobj, attr, lname);
2808 if (is_vmalloc_addr(ldata->ld_buf))
2809 /* if we vmalloced a large buffer drop it */
2810 lu_buf_free(ldata->ld_buf);
2812 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2813 lu_buf_free(&hsm_buf);
2815 /* The child object shouldn't be cached anymore */
2817 set_bit(LU_OBJECT_HEARD_BANSHEE,
2818 &child->mo_lu.lo_header->loh_flags);
2822 /* has not mdd_write{read}_lock on any obj yet. */
2823 static int mdd_rename_sanity_check(const struct lu_env *env,
2824 struct mdd_object *src_pobj,
2825 const struct lu_attr *pattr,
2826 struct mdd_object *tgt_pobj,
2827 const struct lu_attr *tpattr,
2828 struct mdd_object *sobj,
2829 const struct lu_attr *cattr,
2830 struct mdd_object *tobj,
2831 const struct lu_attr *tattr)
2836 /* XXX: when get here, sobj must NOT be NULL,
2837 * the other case has been processed in cld_rename
2838 * before mdd_rename and enable MDS_PERM_BYPASS. */
2842 * If we are using project inheritance, we only allow renames
2843 * into our tree when the project IDs are the same; otherwise
2844 * tree quota mechanism would be circumvented.
2846 if ((((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2847 tpattr->la_projid != cattr->la_projid) ||
2848 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2849 (pattr->la_projid != tpattr->la_projid))) &&
2850 S_ISDIR(cattr->la_mode))
2853 /* we prevent an encrypted file from being renamed
2854 * into an unencrypted dir
2856 if ((pattr->la_valid & LA_FLAGS &&
2857 pattr->la_flags & LUSTRE_ENCRYPT_FL) &&
2858 !(tpattr->la_valid & LA_FLAGS &&
2859 tpattr->la_flags & LUSTRE_ENCRYPT_FL))
2862 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2866 /* XXX: when get here, "tobj == NULL" means tobj must
2867 * NOT exist (neither on remote MDS, such case has been
2868 * processed in cld_rename before mdd_rename and enable
2870 * So check may_create, but not check may_unlink. */
2872 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2873 (src_pobj != tgt_pobj));
2875 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2876 (src_pobj != tgt_pobj), 1);
2878 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2879 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2884 static int mdd_declare_rename(const struct lu_env *env,
2885 struct mdd_device *mdd,
2886 struct mdd_object *mdd_spobj,
2887 struct mdd_object *mdd_tpobj,
2888 struct mdd_object *mdd_sobj,
2889 struct mdd_object *mdd_tobj,
2890 const struct lu_name *sname,
2891 const struct lu_name *tname,
2893 struct linkea_data *ldata, bool change_projid,
2894 struct thandle *handle)
2896 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
2899 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2900 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2906 /* name from source dir */
2907 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2911 /* .. from source child */
2912 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2913 /* source child can be directory, count by source dir's nlink */
2914 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2917 if (mdd_spobj != mdd_tpobj) {
2918 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2923 rc = mdo_declare_index_insert(env, mdd_sobj,
2924 mdd_object_fid(mdd_tpobj),
2925 S_IFDIR, dotdot, handle);
2930 /* new target child can be directory,
2931 * counted by target dir's nlink */
2932 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2937 la->la_valid = LA_CTIME | LA_MTIME;
2938 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2942 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2946 la->la_valid = LA_CTIME;
2948 la->la_valid |= LA_PROJID;
2949 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2953 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2958 rc = mdo_declare_index_insert(env, mdd_tpobj, mdd_object_fid(mdd_sobj),
2959 mdd_object_type(mdd_sobj),
2960 tname->ln_name, handle);
2964 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2965 /* delete target child in target parent directory */
2966 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2971 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2975 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2976 /* target child can be directory,
2977 * delete "." reference in target child directory */
2978 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2982 /* delete ".." reference in target parent directory */
2983 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2988 la->la_valid = LA_CTIME;
2989 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2993 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2998 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
3006 static int mdd_migrate_object(const struct lu_env *env,
3007 struct mdd_object *spobj,
3008 struct mdd_object *tpobj,
3009 struct mdd_object *sobj,
3010 struct mdd_object *tobj,
3011 const struct lu_name *sname,
3012 const struct lu_name *tname,
3013 struct md_op_spec *spec,
3014 struct md_attr *ma);
3016 /* src object can be remote that is why we use only fid and type of object */
3017 static int mdd_rename(const struct lu_env *env,
3018 struct md_object *src_pobj, struct md_object *tgt_pobj,
3019 const struct lu_fid *lf, const struct lu_name *lsname,
3020 struct md_object *tobj, const struct lu_name *ltname,
3023 const char *sname = lsname->ln_name;
3024 const char *tname = ltname->ln_name;
3025 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
3026 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
3027 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
3028 struct mdd_device *mdd = mdo2mdd(src_pobj);
3029 struct mdd_object *mdd_sobj = NULL; /* source object */
3030 struct mdd_object *mdd_tobj = NULL;
3031 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
3032 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3033 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
3034 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3035 struct thandle *handle;
3036 struct linkea_data *ldata = &mdd_env_info(env)->mdi_link_data;
3037 const struct lu_fid *tpobj_fid = mdd_object_fid(mdd_tpobj);
3038 const struct lu_fid *spobj_fid = mdd_object_fid(mdd_spobj);
3041 bool tobj_locked = 0;
3042 bool change_projid = false;
3043 unsigned cl_flags = 0;
3047 /* let unlink to complete and commit */
3048 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
3051 mdd_tobj = md2mdd_obj(tobj);
3053 mdd_sobj = mdd_object_find(env, mdd, lf);
3054 if (IS_ERR(mdd_sobj))
3055 RETURN(PTR_ERR(mdd_sobj));
3057 rc = mdd_la_get(env, mdd_sobj, cattr);
3059 GOTO(out_pending, rc);
3061 /* if rename is cross MDTs, migrate symlink if it doesn't have other
3062 * hard links, and target doesn't exist.
3064 if (mdd_object_remote(mdd_sobj) && S_ISLNK(cattr->la_mode) &&
3065 cattr->la_nlink == 1 && !tobj) {
3066 struct md_op_spec *spec = &mdd_env_info(env)->mdi_spec;
3067 struct lu_device *ld = &mdd->mdd_md_dev.md_lu_dev;
3070 rc = ld->ld_ops->ldo_fid_alloc(env, ld, &tfid, &tgt_pobj->mo_lu,
3073 GOTO(out_pending, rc);
3075 mdd_tobj = mdd_object_find(env, mdd, &tfid);
3076 if (IS_ERR(mdd_tobj))
3077 GOTO(out_pending, rc = PTR_ERR(mdd_tobj));
3079 memset(spec, 0, sizeof(*spec));
3080 rc = mdd_migrate_object(env, mdd_spobj, mdd_tpobj, mdd_sobj,
3081 mdd_tobj, lsname, ltname, spec, ma);
3082 mdd_object_put(env, mdd_tobj);
3083 GOTO(out_pending, rc);
3086 rc = mdd_la_get(env, mdd_spobj, pattr);
3088 GOTO(out_pending, rc);
3091 rc = mdd_la_get(env, mdd_tobj, tattr);
3093 GOTO(out_pending, rc);
3094 /* search for an existing archive.
3095 * we should check ahead as the object
3096 * can be destroyed in this transaction */
3097 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3098 cl_flags |= CLF_RENAME_LAST_EXISTS;
3101 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3103 GOTO(out_pending, rc);
3105 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3106 mdd_sobj, cattr, mdd_tobj, tattr);
3108 GOTO(out_pending, rc);
3110 rc = mdd_name_check(env, mdd, ltname);
3112 GOTO(out_pending, rc);
3114 handle = mdd_trans_create(env, mdd);
3116 GOTO(out_pending, rc = PTR_ERR(handle));
3118 memset(ldata, 0, sizeof(*ldata));
3119 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3120 lsname, mdd_object_fid(mdd_tpobj), ltname,
3125 if (tpattr->la_projid != cattr->la_projid &&
3126 tpattr->la_flags & LUSTRE_PROJINHERIT_FL)
3127 change_projid = true;
3129 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3130 mdd_tobj, lsname, ltname, ma, ldata,
3131 change_projid, handle);
3135 rc = mdd_trans_start(env, mdd, handle);
3139 is_dir = S_ISDIR(cattr->la_mode);
3141 /* Remove source name from source directory */
3142 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3146 /* "mv dir1 dir2" needs "dir1/.." link update */
3147 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3148 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3150 GOTO(fixup_spobj2, rc);
3152 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3155 GOTO(fixup_spobj, rc);
3158 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3159 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3161 /* tname might been renamed to something else */
3162 GOTO(fixup_spobj, rc);
3165 /* Insert new fid with target name into target dir */
3166 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3169 GOTO(fixup_tpobj, rc);
3171 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3172 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3174 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3175 la->la_valid = LA_CTIME;
3176 if (change_projid) {
3177 /* mdd_update_time honors other valid flags except TIME ones */
3178 la->la_valid |= LA_PROJID;
3179 la->la_projid = tpattr->la_projid;
3181 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3183 GOTO(fixup_tpobj, rc);
3185 /* Update the linkEA for the source object */
3186 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3187 rc = mdd_links_rename(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3188 lsname, mdd_object_fid(mdd_tpobj), ltname,
3189 handle, ldata, 0, 0);
3191 /* Old files might not have EA entry */
3192 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3193 lsname, handle, NULL, 0);
3194 mdd_write_unlock(env, mdd_sobj);
3195 /* We don't fail the transaction if the link ea can't be
3196 updated -- fid2path will use alternate lookup method. */
3199 /* Remove old target object
3200 * For tobj is remote case cmm layer has processed
3201 * and set tobj to NULL then. So when tobj is NOT NULL,
3202 * it must be local one.
3204 if (tobj && mdd_object_exists(mdd_tobj)) {
3205 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3207 if (mdd_is_dead_obj(mdd_tobj)) {
3208 /* shld not be dead, something is wrong */
3209 CERROR("tobj is dead, something is wrong\n");
3213 mdo_ref_del(env, mdd_tobj, handle);
3215 /* Remove dot reference. */
3216 if (S_ISDIR(tattr->la_mode))
3217 mdo_ref_del(env, mdd_tobj, handle);
3220 /* fetch updated nlink */
3221 rc = mdd_la_get(env, mdd_tobj, tattr);
3223 CERROR("%s: Failed to get nlink for tobj "
3225 mdd2obd_dev(mdd)->obd_name,
3226 PFID(tpobj_fid), rc);
3227 GOTO(fixup_tpobj, rc);
3230 la->la_valid = LA_CTIME;
3231 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3233 CERROR("%s: Failed to set ctime for tobj "
3235 mdd2obd_dev(mdd)->obd_name,
3236 PFID(tpobj_fid), rc);
3237 GOTO(fixup_tpobj, rc);
3240 /* XXX: this transfer to ma will be removed with LOD/OSP */
3241 ma->ma_attr = *tattr;
3242 ma->ma_valid |= MA_INODE;
3243 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3246 CERROR("%s: Failed to unlink tobj "
3248 mdd2obd_dev(mdd)->obd_name,
3249 PFID(tpobj_fid), rc);
3250 GOTO(fixup_tpobj, rc);
3253 /* fetch updated nlink */
3254 rc = mdd_la_get(env, mdd_tobj, tattr);
3255 if (rc == -ENOENT) {
3256 /* the object got removed, let's
3257 * return the latest known attributes */
3258 tattr->la_nlink = 0;
3260 } else if (rc != 0) {
3261 CERROR("%s: Failed to get nlink for tobj "
3263 mdd2obd_dev(mdd)->obd_name,
3264 PFID(tpobj_fid), rc);
3265 GOTO(fixup_tpobj, rc);
3267 /* XXX: this transfer to ma will be removed with LOD/OSP */
3268 ma->ma_attr = *tattr;
3269 ma->ma_valid |= MA_INODE;
3271 if (tattr->la_nlink == 0)
3272 cl_flags |= CLF_RENAME_LAST;
3274 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3277 la->la_valid = LA_CTIME | LA_MTIME;
3278 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3280 GOTO(fixup_tpobj, rc);
3282 if (mdd_spobj != mdd_tpobj) {
3283 la->la_valid = LA_CTIME | LA_MTIME;
3284 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3286 GOTO(fixup_tpobj, rc);
3293 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3295 CWARN("tp obj fix error %d\n",rc2);
3297 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3298 !mdd_is_dead_obj(mdd_tobj)) {
3300 mdo_ref_add(env, mdd_tobj, handle);
3302 mdo_ref_add(env, mdd_tobj, handle);
3305 rc2 = __mdd_index_insert(env, mdd_tpobj,
3306 mdd_object_fid(mdd_tobj),
3307 mdd_object_type(mdd_tobj),
3310 CWARN("tp obj fix error: rc = %d\n", rc2);
3315 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3316 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3318 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3319 mdd2obd_dev(mdd)->obd_name, rc2);
3322 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3325 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3326 mdd2obd_dev(mdd)->obd_name, rc2);
3331 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3332 mdd_object_type(mdd_sobj), sname,
3335 CWARN("sp obj fix error: rc = %d\n", rc2);
3340 mdd_write_unlock(env, mdd_tobj);
3343 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3344 mdd_tobj, tpobj_fid, lf, spobj_fid,
3345 ltname, lsname, handle);
3348 rc = mdd_trans_stop(env, mdd, rc, handle);
3351 mdd_object_put(env, mdd_sobj);
3356 * Check whether we should migrate the file/dir
3358 * < 0 permission check failed or other error.
3359 * = 0 the file can be migrated.
3361 static int mdd_migrate_sanity_check(const struct lu_env *env,
3362 struct mdd_device *mdd,
3363 struct mdd_object *spobj,
3364 struct mdd_object *tpobj,
3365 struct mdd_object *sobj,
3366 struct mdd_object *tobj,
3367 const struct lu_attr *spattr,
3368 const struct lu_attr *tpattr,
3369 const struct lu_attr *attr)
3375 if (!mdd_object_remote(sobj)) {
3376 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3377 if (sobj->mod_count > 0) {
3378 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3379 mdd_obj_dev_name(sobj),
3380 PFID(mdd_object_fid(sobj)),
3382 mdd_read_unlock(env, sobj);
3385 mdd_read_unlock(env, sobj);
3388 if (mdd_object_exists(tobj))
3391 rc = mdd_may_delete(env, spobj, spattr, sobj, attr, NULL, 1, 0);
3395 rc = mdd_may_create(env, tpobj, tpattr, NULL, true);
3400 struct mdd_xattr_entry {
3401 struct list_head mxe_linkage;
3403 struct lu_buf mxe_buf;
3407 struct lu_buf mx_namebuf;
3408 struct list_head mx_list;
3411 static inline void mdd_xattrs_init(struct mdd_xattrs *xattrs)
3413 INIT_LIST_HEAD(&xattrs->mx_list);
3414 xattrs->mx_namebuf.lb_buf = NULL;
3415 xattrs->mx_namebuf.lb_len = 0;
3418 static inline void mdd_xattrs_fini(struct mdd_xattrs *xattrs)
3420 struct mdd_xattr_entry *entry;
3421 struct mdd_xattr_entry *tmp;
3423 list_for_each_entry_safe(entry, tmp, &xattrs->mx_list, mxe_linkage) {
3424 lu_buf_free(&entry->mxe_buf);
3425 list_del(&entry->mxe_linkage);
3426 OBD_FREE_PTR(entry);
3429 lu_buf_free(&xattrs->mx_namebuf);
3432 /* read xattrs into buf, but skip LMA, LMV, LINKEA if 'skip_linkea' is
3433 * set, and DMV if 'skip_dmv" is set.
3435 static int mdd_xattrs_migrate_prep(const struct lu_env *env,
3436 struct mdd_xattrs *xattrs,
3437 struct mdd_object *sobj,
3441 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
3442 struct mdd_xattr_entry *entry;
3443 bool needencxattr = false;
3444 bool encxattrfound = false;
3454 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3455 if (list_xsize == -ENODATA)
3461 if (attr->la_valid & LA_FLAGS &&
3462 attr->la_flags & LUSTRE_ENCRYPT_FL) {
3463 needencxattr = true;
3465 strlen(LL_XATTR_NAME_ENCRYPTION_CONTEXT) + 1;
3468 lu_buf_alloc(&xattrs->mx_namebuf, list_xsize);
3469 if (xattrs->mx_namebuf.lb_buf == NULL)
3472 rc = mdo_xattr_list(env, sobj, &xattrs->mx_namebuf);
3478 xname = xattrs->mx_namebuf.lb_buf;
3480 for (; rem > 0; xname += xlen, rem -= xlen) {
3482 strcmp(xname, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0)
3483 encxattrfound = true;
3484 xlen = strnlen(xname, rem - 1) + 1;
3485 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3486 strcmp(XATTR_NAME_LMV, xname) == 0)
3490 strcmp(XATTR_NAME_LINK, xname) == 0)
3494 strcmp(XATTR_NAME_DEFAULT_LMV, xname) == 0)
3497 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3498 if (xsize == -ENODATA)
3501 GOTO(fini, rc = xsize);
3503 OBD_ALLOC_PTR(entry);
3505 GOTO(fini, rc = -ENOMEM);
3507 lu_buf_alloc(&entry->mxe_buf, xsize);
3508 if (!entry->mxe_buf.lb_buf) {
3509 OBD_FREE_PTR(entry);
3510 GOTO(fini, rc = -ENOMEM);
3513 rc = mdo_xattr_get(env, sobj, &entry->mxe_buf, xname);
3515 lu_buf_free(&entry->mxe_buf);
3516 OBD_FREE_PTR(entry);
3522 entry->mxe_name = xname;
3523 list_add_tail(&entry->mxe_linkage, &xattrs->mx_list);
3526 if (needencxattr && !encxattrfound) {
3527 xlen = strlen(LL_XATTR_NAME_ENCRYPTION_CONTEXT) + 1;
3528 strncpy(xname, LL_XATTR_NAME_ENCRYPTION_CONTEXT, xlen);
3535 mdd_xattrs_fini(xattrs);
3539 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3540 struct mdd_object *obj,
3541 const struct lu_buf *buf,
3543 int fl, struct thandle *handle);
3545 static int mdd_foreach_xattr(const struct lu_env *env,
3546 struct mdd_object *tobj,
3547 struct mdd_xattrs *xattrs,
3548 struct thandle *handle,
3551 struct mdd_xattr_entry *entry;
3554 list_for_each_entry(entry, &xattrs->mx_list, mxe_linkage) {
3555 rc = cb(env, tobj, &entry->mxe_buf, entry->mxe_name, 0, handle);
3563 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3564 struct mdd_object *sobj,
3565 struct mdd_object *tobj,
3566 const struct lu_name *sname,
3567 const struct lu_fid *sfid,
3568 const struct lu_name *lname,
3569 const struct lu_fid *fid,
3571 struct thandle *handle);
3573 static int mdd_declare_update_link(const struct lu_env *env,
3574 struct mdd_object *sobj,
3575 struct mdd_object *tobj,
3576 const struct lu_name *tname,
3577 const struct lu_fid *tpfid,
3578 const struct lu_name *lname,
3579 const struct lu_fid *fid,
3581 struct thandle *handle)
3583 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3584 struct mdd_object *pobj;
3588 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3589 !strcmp(tname->ln_name, lname->ln_name))
3592 pobj = mdd_object_find(env, mdd, fid);
3594 return PTR_ERR(pobj);
3597 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3599 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
3600 mdd_object_type(sobj),
3601 lname->ln_name, handle);
3602 mdd_object_put(env, pobj);
3606 rc = mdo_declare_ref_add(env, tobj, handle);
3610 rc = mdo_declare_ref_del(env, sobj, handle);
3614 static int mdd_update_link(const struct lu_env *env,
3615 struct mdd_object *sobj,
3616 struct mdd_object *tobj,
3617 const struct lu_name *tname,
3618 const struct lu_fid *tpfid,
3619 const struct lu_name *lname,
3620 const struct lu_fid *fid,
3622 struct thandle *handle)
3624 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3625 struct mdd_object *pobj;
3631 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3632 !memcmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3635 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3636 PFID(fid), PNAME(lname), PFID(mdd_object_fid(tobj)));
3638 pobj = mdd_object_find(env, mdd, fid);
3640 CWARN("%s: cannot find obj "DFID": %ld\n",
3641 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3642 RETURN(PTR_ERR(pobj));
3645 if (!mdd_object_exists(pobj)) {
3646 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3647 mdd_object_put(env, pobj);
3651 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3652 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3654 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(tobj),
3655 mdd_object_type(sobj),
3656 lname->ln_name, handle);
3657 mdd_write_unlock(env, pobj);
3658 mdd_object_put(env, pobj);
3662 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3663 rc = mdo_ref_add(env, tobj, handle);
3664 mdd_write_unlock(env, tobj);
3668 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3669 rc = mdo_ref_del(env, sobj, handle);
3670 mdd_write_unlock(env, sobj);
3675 static inline int mdd_fld_lookup(const struct lu_env *env,
3676 struct mdd_device *mdd,
3677 const struct lu_fid *fid,
3680 struct lu_seq_range *range = &mdd_env_info(env)->mdi_range;
3681 struct seq_server_site *ss;
3684 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3686 range->lsr_flags = LU_SEQ_RANGE_MDT;
3687 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3691 *mdt_index = range->lsr_index;
3696 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3697 struct mdd_object *sobj,
3698 struct mdd_object *tobj,
3699 const struct lu_name *tname,
3700 const struct lu_fid *tpfid,
3701 const struct lu_name *lname,
3702 const struct lu_fid *fid,
3704 struct thandle *handle)
3706 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3707 __u32 source_mdt_index = *(__u32 *)opaque;
3708 __u32 link_mdt_index;
3714 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3715 !memcmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3718 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3722 RETURN(link_mdt_index == source_mdt_index);
3725 static int mdd_iterate_linkea(const struct lu_env *env,
3726 struct mdd_object *sobj,
3727 struct mdd_object *tobj,
3728 const struct lu_name *tname,
3729 const struct lu_fid *tpfid,
3730 struct linkea_data *ldata,
3732 struct thandle *handle,
3735 struct mdd_thread_info *info = mdd_env_info(env);
3736 char *filename = info->mdi_name;
3737 struct lu_name lname;
3744 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3745 linkea_next_entry(ldata)) {
3746 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3749 /* Note: lname might miss \0 at the end */
3750 snprintf(filename, sizeof(info->mdi_name), "%.*s",
3751 lname.ln_namelen, lname.ln_name);
3752 lname.ln_name = filename;
3754 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3756 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3764 * Prepare linkea, and check whether file needs migrate: if source still has
3765 * link on source MDT, no need to migrate, just update namespace on source and
3768 * \retval 0 do migrate
3769 * \retval 1 don't migrate
3770 * \retval -errno on failure
3772 static int mdd_migrate_linkea_prepare(const struct lu_env *env,
3773 struct mdd_device *mdd,
3774 struct mdd_object *spobj,
3775 struct mdd_object *tpobj,
3776 struct mdd_object *sobj,
3777 const struct lu_name *sname,
3778 const struct lu_name *tname,
3779 const struct lu_attr *attr,
3780 struct linkea_data *ldata)
3782 __u32 source_mdt_index;
3787 memset(ldata, 0, sizeof(*ldata));
3788 rc = mdd_linkea_prepare(env, sobj, mdd_object_fid(spobj), sname,
3789 mdd_object_fid(tpobj), tname, 1, 0, ldata);
3794 * Then it will check if the file should be migrated. If the file has
3795 * mulitple links, we only need migrate the file if all of its entries
3796 * has been migrated to the remote MDT.
3798 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3801 /* If there are still links locally, don't migrate this file */
3802 LASSERT(ldata->ld_leh != NULL);
3805 * If linkEA is overflow, it means there are some unknown name entries
3806 * under unknown parents, which will prevent the migration.
3808 if (unlikely(ldata->ld_leh->leh_overflow_time))
3811 rc = mdd_fld_lookup(env, mdd, mdd_object_fid(sobj), &source_mdt_index);
3815 rc = mdd_iterate_linkea(env, sobj, NULL, tname, mdd_object_fid(tpobj),
3816 ldata, &source_mdt_index, NULL,
3817 mdd_is_link_on_source_mdt);
3821 static int mdd_declare_migrate_update(const struct lu_env *env,
3822 struct mdd_object *spobj,
3823 struct mdd_object *tpobj,
3824 struct mdd_object *obj,
3825 const struct lu_name *sname,
3826 const struct lu_name *tname,
3827 struct lu_attr *attr,
3828 struct lu_attr *spattr,
3829 struct lu_attr *tpattr,
3830 struct linkea_data *ldata,
3832 struct thandle *handle)
3834 struct mdd_thread_info *info = mdd_env_info(env);
3835 struct lu_attr *la = &info->mdi_la_for_fix;
3838 rc = mdo_declare_index_delete(env, spobj, sname->ln_name, handle);
3842 if (S_ISDIR(attr->la_mode)) {
3843 rc = mdo_declare_ref_del(env, spobj, handle);
3848 rc = mdo_declare_index_insert(env, tpobj, mdd_object_fid(obj),
3849 attr->la_mode & S_IFMT,
3850 tname->ln_name, handle);
3854 rc = mdd_declare_links_add(env, obj, handle, ldata);
3858 if (S_ISDIR(attr->la_mode)) {
3859 rc = mdo_declare_ref_add(env, tpobj, handle);
3864 la->la_valid = LA_CTIME | LA_MTIME;
3865 rc = mdo_declare_attr_set(env, spobj, la, handle);
3869 if (tpobj != spobj) {
3870 rc = mdo_declare_attr_set(env, tpobj, la, handle);
3878 static int mdd_declare_migrate_create(const struct lu_env *env,
3879 struct mdd_object *spobj,
3880 struct mdd_object *tpobj,
3881 struct mdd_object *sobj,
3882 struct mdd_object *tobj,
3883 const struct lu_name *sname,
3884 const struct lu_name *tname,
3885 struct lu_attr *spattr,
3886 struct lu_attr *tpattr,
3887 struct lu_attr *attr,
3888 struct lu_buf *sbuf,
3889 struct linkea_data *ldata,
3890 struct mdd_xattrs *xattrs,
3892 struct md_op_spec *spec,
3893 struct dt_allocation_hint *hint,
3894 struct thandle *handle)
3896 struct mdd_thread_info *info = mdd_env_info(env);
3897 struct md_layout_change *mlc = &info->mdi_mlc;
3898 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
3903 if (S_ISDIR(attr->la_mode)) {
3904 struct lmv_user_md *lum = spec->u.sp_ea.eadata;
3906 mlc->mlc_opc = MD_LAYOUT_DETACH;
3907 rc = mdo_declare_layout_change(env, sobj, mlc, handle);
3911 lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3912 } else if (S_ISLNK(attr->la_mode)) {
3913 spec->u.sp_symname.ln_name = sbuf->lb_buf;
3914 /* don't count NUL */
3915 spec->u.sp_symname.ln_namelen = sbuf->lb_len - 1;
3916 } else if (S_ISREG(attr->la_mode)) {
3917 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
3918 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
3921 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
3923 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
3924 tname, attr, handle, spec, ldata, NULL, NULL,
3930 * tobj mode will be used in mdo_declare_layout_change(), but it's not
3931 * createb yet, copy from sobj.
3933 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
3934 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
3935 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
3937 if (S_ISDIR(attr->la_mode)) {
3939 /* if sobj is not striped, fake a 1-stripe LMV */
3940 LASSERT(sizeof(info->mdi_key) >
3941 lmv_mds_md_size(1, LMV_MAGIC_V1));
3942 lmv = (typeof(lmv))info->mdi_key;
3943 memset(lmv, 0, sizeof(*lmv));
3944 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3945 lmv->lmv_stripe_count = cpu_to_le32(1);
3946 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
3947 fid_le_to_cpu(&lmv->lmv_stripe_fids[0],
3948 mdd_object_fid(sobj));
3949 mlc->mlc_buf.lb_buf = lmv;
3950 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
3952 mlc->mlc_buf = *sbuf;
3954 mlc->mlc_opc = MD_LAYOUT_ATTACH;
3955 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
3960 rc = mdd_foreach_xattr(env, tobj, xattrs, handle,
3961 mdo_declare_xattr_set);
3965 if (S_ISREG(attr->la_mode)) {
3966 struct lu_buf fid_buf;
3968 handle->th_complex = 1;
3970 /* target may be remote, update PFID via sobj. */
3971 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
3972 fid_buf.lb_len = sizeof(struct lu_fid);
3973 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
3978 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
3983 if (!S_ISDIR(attr->la_mode)) {
3984 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
3985 mdd_object_fid(tpobj), ldata, NULL,
3986 handle, mdd_declare_update_link);
3991 if (!S_ISDIR(attr->la_mode) || lmv) {
3992 rc = mdo_declare_ref_del(env, sobj, handle);
3996 if (S_ISDIR(attr->la_mode)) {
3997 rc = mdo_declare_ref_del(env, sobj, handle);
4002 rc = mdo_declare_destroy(env, sobj, handle);
4007 rc = mdd_declare_migrate_update(env, spobj, tpobj, tobj, sname, tname,
4008 attr, spattr, tpattr, ldata, ma,
4014 * migrate dirent from \a spobj to \a tpobj.
4016 static int mdd_migrate_update(const struct lu_env *env,
4017 struct mdd_object *spobj,
4018 struct mdd_object *tpobj,
4019 struct mdd_object *obj,
4020 const struct lu_name *sname,
4021 const struct lu_name *tname,
4022 struct lu_attr *attr,
4023 struct lu_attr *spattr,
4024 struct lu_attr *tpattr,
4025 struct linkea_data *ldata,
4027 struct thandle *handle)
4029 struct mdd_thread_info *info = mdd_env_info(env);
4030 struct lu_attr *la = &info->mdi_la_for_fix;
4035 CDEBUG(D_INFO, "update "DFID" from "DFID"/%s to "DFID"/%s\n",
4036 PFID(mdd_object_fid(obj)), PFID(mdd_object_fid(spobj)),
4037 sname->ln_name, PFID(mdd_object_fid(tpobj)), tname->ln_name);
4039 rc = __mdd_index_delete(env, spobj, sname->ln_name,
4040 S_ISDIR(attr->la_mode), handle);
4044 rc = __mdd_index_insert(env, tpobj, mdd_object_fid(obj),
4045 attr->la_mode & S_IFMT,
4046 tname->ln_name, handle);
4050 rc = mdd_links_write(env, obj, ldata, handle);
4054 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
4055 la->la_valid = LA_CTIME | LA_MTIME;
4056 mdd_write_lock(env, spobj, DT_SRC_PARENT);
4057 rc = mdd_update_time(env, spobj, spattr, la, handle);
4058 mdd_write_unlock(env, spobj);
4062 if (tpobj != spobj) {
4063 la->la_valid = LA_CTIME | LA_MTIME;
4064 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
4065 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
4066 mdd_write_unlock(env, tpobj);
4075 * Migrate file/dir to target MDT.
4077 * Create target according to \a spec, and then migrate xattrs, if it's
4078 * directory, migrate source stripes to target.
4080 * \param[in] env execution environment
4081 * \param[in] spobj source parent object
4082 * \param[in] tpobj target parent object
4083 * \param[in] sobj source object
4084 * \param[in] tobj target object
4085 * \param[in] lname file name
4086 * \param[in] spattr source parent attributes
4087 * \param[in] tpattr target parent attributes
4088 * \param[in] attr source attributes
4089 * \param[in] sbuf source LMV buf
4090 * \param[in] spec migrate create spec
4091 * \param[in] hint target creation hint
4092 * \param[in] handle tranasction handle
4094 * \retval 0 on success
4095 * \retval -errno on failure
4097 static int mdd_migrate_create(const struct lu_env *env,
4098 struct mdd_object *spobj,
4099 struct mdd_object *tpobj,
4100 struct mdd_object *sobj,
4101 struct mdd_object *tobj,
4102 const struct lu_name *sname,
4103 const struct lu_name *tname,
4104 struct lu_attr *spattr,
4105 struct lu_attr *tpattr,
4106 struct lu_attr *attr,
4107 const struct lu_buf *sbuf,
4108 struct linkea_data *ldata,
4109 struct mdd_xattrs *xattrs,
4111 struct md_op_spec *spec,
4112 struct dt_allocation_hint *hint,
4113 struct thandle *handle)
4120 * migrate sobj stripes to tobj if it's directory:
4121 * 1. detach stripes from sobj.
4122 * 2. attach stripes to tobj, see mdd_declare_migrate_mdt().
4123 * 3. create stripes for tobj, see lod_xattr_set_lmv().
4125 if (S_ISDIR(attr->la_mode)) {
4126 struct mdd_thread_info *info = mdd_env_info(env);
4127 struct md_layout_change *mlc = &info->mdi_mlc;
4129 mlc->mlc_opc = MD_LAYOUT_DETACH;
4131 mdd_write_lock(env, sobj, DT_SRC_PARENT);
4132 rc = mdo_layout_change(env, sobj, mlc, handle);
4133 mdd_write_unlock(env, sobj);
4138 /* don't set nlink from sobj */
4139 attr->la_valid &= ~LA_NLINK;
4141 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4142 hint, handle, false);
4146 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4147 rc = mdd_foreach_xattr(env, tobj, xattrs, handle, mdo_xattr_set);
4148 mdd_write_unlock(env, tobj);
4152 /* for regular file, update OST objects XATTR_NAME_FID */
4153 if (S_ISREG(attr->la_mode)) {
4154 struct lu_buf fid_buf;
4156 /* target may be remote, update PFID via sobj. */
4157 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
4158 fid_buf.lb_len = sizeof(struct lu_fid);
4159 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4164 /* delete LOV to avoid deleting OST objs when destroying sobj */
4165 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4166 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4167 mdd_write_unlock(env, sobj);
4168 /* O_DELAY_CREATE file may not have LOV, ignore -ENODATA */
4169 if (rc && rc != -ENODATA)
4174 /* update links FID */
4175 if (!S_ISDIR(attr->la_mode)) {
4176 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
4177 mdd_object_fid(tpobj), ldata,
4178 NULL, handle, mdd_update_link);
4183 /* don't destroy sobj if it's plain directory */
4184 if (!S_ISDIR(attr->la_mode) || sbuf->lb_buf) {
4185 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4186 rc = mdo_ref_del(env, sobj, handle);
4188 if (S_ISDIR(attr->la_mode))
4189 rc = mdo_ref_del(env, sobj, handle);
4191 rc = mdo_destroy(env, sobj, handle);
4193 mdd_write_unlock(env, sobj);
4198 rc = mdd_migrate_update(env, spobj, tpobj, tobj, sname, tname, attr,
4199 spattr, tpattr, ldata, ma, handle);
4204 /* NB: if user issued different migrate command, we can't adjust it silently
4205 * here, because this command will decide target MDT in subdir migration in
4208 static int mdd_migrate_cmd_check(struct mdd_device *mdd,
4209 const struct lmv_mds_md_v1 *lmv,
4210 const struct lmv_user_md_v1 *lum,
4211 const struct lu_name *lname)
4213 __u32 lum_stripe_count = lum->lum_stripe_count;
4214 __u32 lum_hash_type = lum->lum_hash_type &
4215 cpu_to_le32(LMV_HASH_TYPE_MASK);
4216 __u32 lmv_hash_type = lmv->lmv_hash_type &
4217 cpu_to_le32(LMV_HASH_TYPE_MASK);
4219 if (!lmv_is_sane(lmv))
4222 /* if stripe_count unspecified, set to 1 */
4223 if (!lum_stripe_count)
4224 lum_stripe_count = cpu_to_le32(1);
4226 /* TODO: check specific MDTs */
4227 if (lum_stripe_count != lmv->lmv_migrate_offset ||
4228 lum->lum_stripe_offset != lmv->lmv_master_mdt_index ||
4229 (lum_hash_type && lum_hash_type != lmv_hash_type)) {
4230 CERROR("%s: '"DNAME"' migration was interrupted, run 'lfs migrate -m %d -c %d -H %s "DNAME"' to finish migration.\n",
4231 mdd2obd_dev(mdd)->obd_name, PNAME(lname),
4232 le32_to_cpu(lmv->lmv_master_mdt_index),
4233 le32_to_cpu(lmv->lmv_migrate_offset),
4234 mdt_hash_name[le32_to_cpu(lmv_hash_type)],
4243 * Internal function to migrate directory or file between MDTs.
4245 * migrate source to target in following steps:
4246 * 1. create target, append source stripes after target's if it's directory,
4247 * migrate xattrs and update fid of source links.
4248 * 2. update namespace: migrate dirent from source parent to target parent,
4249 * update file linkea, and destroy source if it's not needed any more.
4251 * \param[in] env execution environment
4252 * \param[in] spobj source parent object
4253 * \param[in] tpobj target parent object
4254 * \param[in] sobj source object
4255 * \param[in] tobj target object
4256 * \param[in] sname source file name
4257 * \param[in] tname target file name
4258 * \param[in] spec target creation spec
4259 * \param[in] ma used to update \a pobj mtime and ctime
4261 * \retval 0 on success
4262 * \retval -errno on failure
4264 static int mdd_migrate_object(const struct lu_env *env,
4265 struct mdd_object *spobj,
4266 struct mdd_object *tpobj,
4267 struct mdd_object *sobj,
4268 struct mdd_object *tobj,
4269 const struct lu_name *sname,
4270 const struct lu_name *tname,
4271 struct md_op_spec *spec,
4274 struct mdd_thread_info *info = mdd_env_info(env);
4275 struct mdd_device *mdd = mdo2mdd(&spobj->mod_obj);
4276 struct lu_attr *spattr = &info->mdi_pattr;
4277 struct lu_attr *tpattr = &info->mdi_tpattr;
4278 struct lu_attr *attr = &info->mdi_cattr;
4279 struct linkea_data *ldata = &info->mdi_link_data;
4280 struct dt_allocation_hint *hint = &info->mdi_hint;
4281 struct lu_buf sbuf = { NULL };
4282 struct mdd_xattrs xattrs;
4283 struct lmv_mds_md_v1 *lmv;
4284 struct thandle *handle;
4289 rc = mdd_la_get(env, sobj, attr);
4293 rc = mdd_la_get(env, spobj, spattr);
4297 rc = mdd_la_get(env, tpobj, tpattr);
4301 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4302 spattr, tpattr, attr);
4306 mdd_xattrs_init(&xattrs);
4308 if (S_ISDIR(attr->la_mode) && !spec->sp_migrate_nsonly) {
4309 struct lmv_user_md_v1 *lum = spec->u.sp_ea.eadata;
4313 /* if user use default value '0' for stripe_count, we need to
4314 * adjust it to '1' to create a 1-stripe directory.
4316 if (lum->lum_stripe_count == 0)
4317 lum->lum_stripe_count = cpu_to_le32(1);
4319 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4320 if (rc && rc != -ENODATA)
4325 if (!lmv_is_sane(lmv))
4326 GOTO(out, rc = -EBADF);
4327 if (lmv_is_migrating(lmv)) {
4328 rc = mdd_migrate_cmd_check(mdd, lmv, lum,
4333 } else if (!S_ISDIR(attr->la_mode)) {
4335 GOTO(out, rc = -EALREADY);
4337 /* update namespace only if @sobj is on MDT where @tpobj is. */
4338 if (!mdd_object_remote(tpobj) && !mdd_object_remote(sobj))
4339 spec->sp_migrate_nsonly = true;
4341 if (S_ISLNK(attr->la_mode)) {
4342 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4344 GOTO(out, rc = -ENOMEM);
4346 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4349 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4350 mdd2obd_dev(mdd)->obd_name,
4351 PFID(mdd_object_fid(sobj)), rc);
4357 /* linkea needs update upon FID or parent stripe change */
4358 rc = mdd_migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, sname,
4359 tname, attr, ldata);
4361 /* update namespace only if @sobj has link on its MDT. */
4362 spec->sp_migrate_nsonly = true;
4366 /* migrate inode will migrate xattrs, prepare xattrs early to avoid
4367 * RPCs inside transaction.
4369 if (!spec->sp_migrate_nsonly) {
4370 rc = mdd_xattrs_migrate_prep(env, &xattrs, sobj, true, true);
4375 handle = mdd_trans_create(env, mdd);
4377 GOTO(out, rc = PTR_ERR(handle));
4379 if (spec->sp_migrate_nsonly)
4380 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, sname,
4381 tname, attr, spattr, tpattr,
4384 rc = mdd_declare_migrate_create(env, spobj, tpobj, sobj, tobj,
4385 sname, tname, spattr, tpattr,
4386 attr, &sbuf, ldata, &xattrs, ma,
4387 spec, hint, handle);
4391 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, tname, sname,
4396 rc = mdd_trans_start(env, mdd, handle);
4400 if (spec->sp_migrate_nsonly)
4401 rc = mdd_migrate_update(env, spobj, tpobj, sobj, sname, tname,
4402 attr, spattr, tpattr, ldata, ma,
4405 rc = mdd_migrate_create(env, spobj, tpobj, sobj, tobj, sname,
4406 tname, spattr, tpattr, attr, &sbuf,
4407 ldata, &xattrs, ma, spec, hint, handle);
4411 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0,
4412 spec->sp_migrate_nsonly ? sobj : tobj,
4413 mdd_object_fid(spobj), mdd_object_fid(sobj),
4414 mdd_object_fid(tpobj), tname, sname,
4421 rc = mdd_trans_stop(env, mdd, rc, handle);
4423 mdd_xattrs_fini(&xattrs);
4430 * Migrate directory or file between MDTs.
4432 * \param[in] env execution environment
4433 * \param[in] md_pobj parent master object
4434 * \param[in] md_sobj source object
4435 * \param[in] lname file name
4436 * \param[in] md_tobj target object
4437 * \param[in] spec target creation spec
4438 * \param[in] ma used to update \a pobj mtime and ctime
4440 * \retval 0 on success
4441 * \retval -errno on failure
4443 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4444 struct md_object *md_sobj, const struct lu_name *lname,
4445 struct md_object *md_tobj, struct md_op_spec *spec,
4448 struct mdd_thread_info *info = mdd_env_info(env);
4449 struct mdd_device *mdd = mdo2mdd(md_pobj);
4450 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4451 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4452 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4453 struct mdd_object *spobj = NULL;
4454 struct mdd_object *tpobj = NULL;
4455 struct lu_buf pbuf = { NULL };
4456 struct lu_fid *fid = &info->mdi_fid2;
4457 struct lmv_mds_md_v1 *lmv;
4462 /* locate source and target stripe on pobj, which are the real parent */
4463 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4464 if (rc < 0 && rc != -ENODATA)
4471 if (!lmv_is_sane(lmv))
4472 GOTO(out, rc = -EBADF);
4474 /* locate target parent stripe */
4475 /* fail check here to make sure top dir migration succeed. */
4476 if (lmv_is_migrating(lmv) &&
4477 OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4478 GOTO(out, rc = -EIO);
4480 index = lmv_name_to_stripe_index(lmv, lname->ln_name,
4483 GOTO(out, rc = index);
4485 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4486 tpobj = mdd_object_find(env, mdd, fid);
4488 GOTO(out, rc = PTR_ERR(tpobj));
4490 /* locate source parent stripe */
4491 if (lmv_is_layout_changing(lmv)) {
4492 index = lmv_name_to_stripe_index_old(lmv,
4496 GOTO(out, rc = index);
4498 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4499 spobj = mdd_object_find(env, mdd, fid);
4501 GOTO(out, rc = PTR_ERR(spobj));
4503 /* parent stripe unchanged */
4504 if (spobj == tpobj) {
4505 if (!lmv_is_restriping(lmv))
4506 GOTO(out, rc = -EINVAL);
4507 GOTO(out, rc = -EALREADY);
4511 mdd_object_get(spobj);
4516 mdd_object_get(tpobj);
4517 mdd_object_get(spobj);
4520 rc = mdd_migrate_object(env, spobj, tpobj, sobj, tobj, lname, lname,
4525 if (!IS_ERR_OR_NULL(spobj))
4526 mdd_object_put(env, spobj);
4527 if (!IS_ERR_OR_NULL(tpobj))
4528 mdd_object_put(env, tpobj);
4534 static int mdd_declare_1sd_collapse(const struct lu_env *env,
4535 struct mdd_object *pobj,
4536 struct mdd_object *obj,
4537 struct mdd_object *stripe,
4538 struct lu_attr *attr,
4539 struct mdd_xattrs *xattrs,
4540 struct md_layout_change *mlc,
4541 struct lu_name *lname,
4542 struct thandle *handle)
4546 mlc->mlc_opc = MD_LAYOUT_DETACH;
4547 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4551 rc = mdo_declare_index_insert(env, stripe, mdd_object_fid(pobj),
4552 S_IFDIR, dotdot, handle);
4556 rc = mdd_foreach_xattr(env, stripe, xattrs, handle,
4557 mdo_declare_xattr_set);
4561 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4565 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4569 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4573 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(stripe),
4574 attr->la_mode, lname->ln_name, handle);
4578 rc = mdo_declare_ref_del(env, obj, handle);
4582 rc = mdo_declare_ref_del(env, obj, handle);
4586 rc = mdo_declare_destroy(env, obj, handle);
4593 /* transform one-stripe directory to a plain directory */
4594 static int mdd_1sd_collapse(const struct lu_env *env,
4595 struct mdd_object *pobj,
4596 struct mdd_object *obj,
4597 struct mdd_object *stripe,
4598 struct lu_attr *attr,
4599 struct mdd_xattrs *xattrs,
4600 struct md_layout_change *mlc,
4601 struct lu_name *lname,
4602 struct thandle *handle)
4608 /* replace 1-stripe directory with its stripe */
4609 mlc->mlc_opc = MD_LAYOUT_DETACH;
4611 mdd_write_lock(env, obj, DT_SRC_PARENT);
4612 rc = mdo_layout_change(env, obj, mlc, handle);
4613 mdd_write_unlock(env, obj);
4617 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4618 mdd_write_lock(env, obj, DT_SRC_CHILD);
4620 /* insert dotdot to stripe which points to parent */
4621 rc = __mdd_index_insert_only(env, stripe, mdd_object_fid(pobj),
4622 S_IFDIR, dotdot, handle);
4626 rc = mdd_foreach_xattr(env, stripe, xattrs, handle, mdo_xattr_set);
4631 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4635 /* don't set nlink from parent */
4636 attr->la_valid &= ~LA_NLINK;
4638 rc = mdo_attr_set(env, stripe, attr, handle);
4642 /* delete dir name from parent */
4643 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4647 /* insert stripe to parent with dir name */
4648 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(stripe),
4649 attr->la_mode, lname->ln_name, handle);
4653 /* destroy dir obj */
4654 rc = mdo_ref_del(env, obj, handle);
4658 rc = mdo_ref_del(env, obj, handle);
4662 rc = mdo_destroy(env, obj, handle);
4668 mdd_write_unlock(env, obj);
4669 mdd_write_unlock(env, pobj);
4675 * shrink directory stripes after migration/merge
4677 int mdd_dir_layout_shrink(const struct lu_env *env,
4678 struct md_object *md_obj,
4679 struct md_layout_change *mlc)
4681 struct mdd_device *mdd = mdo2mdd(md_obj);
4682 struct mdd_thread_info *info = mdd_env_info(env);
4683 struct mdd_object *obj = md2mdd_obj(md_obj);
4684 struct mdd_object *pobj = NULL;
4685 struct mdd_object *stripe = NULL;
4686 struct lu_attr *attr = &info->mdi_pattr;
4687 struct lu_fid *fid = &info->mdi_fid2;
4688 struct lu_name lname = { NULL };
4689 struct lu_buf lmv_buf = { NULL };
4690 struct mdd_xattrs xattrs;
4691 struct lmv_mds_md_v1 *lmv;
4692 struct lmv_user_md *lmu;
4693 struct thandle *handle;
4698 rc = mdd_la_get(env, obj, attr);
4702 if (!S_ISDIR(attr->la_mode))
4705 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4709 lmv = lmv_buf.lb_buf;
4710 if (!lmv_is_sane(lmv))
4713 lmu = mlc->mlc_buf.lb_buf;
4715 /* adjust the default value '0' to '1' */
4716 if (lmu->lum_stripe_count == 0)
4717 lmu->lum_stripe_count = cpu_to_le32(1);
4719 /* these were checked in MDT */
4720 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4721 le32_to_cpu(lmv->lmv_stripe_count));
4722 LASSERT(!lmv_is_splitting(lmv));
4723 LASSERT(lmv_is_migrating(lmv) || lmv_is_merging(lmv));
4725 mdd_xattrs_init(&xattrs);
4727 /* if dir stripe count will be shrunk to 1, it needs to be transformed
4728 * to a plain dir, which will cause FID change and namespace update.
4730 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4731 struct linkea_data *ldata = &info->mdi_link_data;
4732 char *filename = info->mdi_name;
4734 rc = mdd_links_read(env, obj, ldata);
4738 if (ldata->ld_leh->leh_reccount > 1)
4739 GOTO(out, rc = -EINVAL);
4741 linkea_first_entry(ldata);
4743 GOTO(out, rc = -ENODATA);
4745 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4748 /* Note: lname might miss \0 at the end */
4749 snprintf(filename, sizeof(info->mdi_name), "%.*s",
4750 lname.ln_namelen, lname.ln_name);
4751 lname.ln_name = filename;
4753 pobj = mdd_object_find(env, mdd, fid);
4760 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4762 stripe = mdd_object_find(env, mdd, fid);
4763 if (IS_ERR(stripe)) {
4764 mdd_object_put(env, pobj);
4766 GOTO(out, rc = PTR_ERR(stripe));
4769 if (!lmv_is_fixed(lmv))
4770 rc = mdd_xattrs_migrate_prep(env, &xattrs, obj, false,
4774 handle = mdd_trans_create(env, mdd);
4776 GOTO(out, rc = PTR_ERR(handle));
4778 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4779 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4781 GOTO(stop_trans, rc);
4783 if (le32_to_cpu(lmu->lum_stripe_count) == 1 && !lmv_is_fixed(lmv)) {
4784 rc = mdd_declare_1sd_collapse(env, pobj, obj, stripe, attr,
4785 &xattrs, mlc, &lname, handle);
4787 GOTO(stop_trans, rc);
4790 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4793 GOTO(stop_trans, rc);
4795 rc = mdd_trans_start(env, mdd, handle);
4797 GOTO(stop_trans, rc);
4799 mdd_write_lock(env, obj, DT_SRC_PARENT);
4800 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4801 rc = mdo_layout_change(env, obj, mlc, handle);
4802 mdd_write_unlock(env, obj);
4804 GOTO(stop_trans, rc);
4806 if (le32_to_cpu(lmu->lum_stripe_count) == 1 && !lmv_is_fixed(lmv)) {
4807 rc = mdd_1sd_collapse(env, pobj, obj, stripe, attr, &xattrs,
4808 mlc, &lname, handle);
4810 GOTO(stop_trans, rc);
4813 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4814 XATTR_NAME_LMV, handle);
4815 GOTO(stop_trans, rc);
4818 rc = mdd_trans_stop(env, mdd, rc, handle);
4820 mdd_xattrs_fini(&xattrs);
4822 mdd_object_put(env, stripe);
4823 mdd_object_put(env, pobj);
4825 lu_buf_free(&lmv_buf);
4829 static int mdd_dir_declare_split_plain(const struct lu_env *env,
4830 struct mdd_device *mdd,
4831 struct mdd_object *pobj,
4832 struct mdd_object *obj,
4833 struct mdd_object *tobj,
4834 struct mdd_xattrs *xattrs,
4835 struct md_layout_change *mlc,
4836 struct dt_allocation_hint *hint,
4837 struct thandle *handle)
4839 struct mdd_thread_info *info = mdd_env_info(env);
4840 const struct lu_name *lname = mlc->mlc_name;
4841 struct lu_attr *la = &info->mdi_la_for_fix;
4842 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
4843 struct linkea_data *ldata = &info->mdi_link_data;
4844 struct lmv_mds_md_v1 *lmv;
4848 mlc->mlc_opc = MD_LAYOUT_DETACH;
4849 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4853 memset(ldata, 0, sizeof(*ldata));
4854 rc = mdd_linkea_prepare(env, obj, NULL, NULL, mdd_object_fid(pobj),
4855 lname, 1, 0, ldata);
4859 count = lum->lum_stripe_count;
4860 lum->lum_stripe_count = 0;
4861 /* don't set default LMV since it will become a striped dir */
4862 lum->lum_max_inherit = LMV_INHERIT_NONE;
4863 mdd_object_make_hint(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4865 rc = mdd_declare_create(env, mdo2mdd(&pobj->mod_obj), pobj, tobj,
4866 lname, mlc->mlc_attr, handle, mlc->mlc_spec,
4867 ldata, NULL, NULL, NULL, hint);
4871 /* tobj mode will be used in lod_declare_xattr_set(), but it's not
4874 tobj->mod_obj.mo_lu.lo_header->loh_attr |= S_IFDIR;
4876 lmv = (typeof(lmv))info->mdi_key;
4877 memset(lmv, 0, sizeof(*lmv));
4878 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
4879 lmv->lmv_stripe_count = cpu_to_le32(1);
4880 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
4881 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdd_object_fid(obj));
4883 mlc->mlc_opc = MD_LAYOUT_ATTACH;
4884 mlc->mlc_buf.lb_buf = lmv;
4885 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
4886 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4890 rc = mdd_foreach_xattr(env, tobj, xattrs, handle,
4891 mdo_declare_xattr_set);
4895 lum->lum_stripe_count = count;
4896 mlc->mlc_opc = MD_LAYOUT_SPLIT;
4897 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4901 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4905 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
4906 S_IFDIR, lname->ln_name, handle);
4910 la->la_valid = LA_CTIME | LA_MTIME;
4911 rc = mdo_declare_attr_set(env, obj, la, handle);
4915 rc = mdo_declare_attr_set(env, pobj, la, handle);
4919 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4925 * plain directory split:
4926 * 1. create \a tobj as plain directory.
4927 * 2. append \a obj as first stripe of \a tobj.
4928 * 3. migrate xattrs from \a obj to \a tobj.
4929 * 4. split \a tobj to specific stripe count.
4931 static int mdd_dir_split_plain(const struct lu_env *env,
4932 struct mdd_device *mdd,
4933 struct mdd_object *pobj,
4934 struct mdd_object *obj,
4935 struct mdd_object *tobj,
4936 struct mdd_xattrs *xattrs,
4937 struct md_layout_change *mlc,
4938 struct dt_allocation_hint *hint,
4939 struct thandle *handle)
4941 struct mdd_thread_info *info = mdd_env_info(env);
4942 struct lu_attr *pattr = &info->mdi_pattr;
4943 struct lu_attr *la = &info->mdi_la_for_fix;
4944 const struct lu_name *lname = mlc->mlc_name;
4945 struct linkea_data *ldata = &info->mdi_link_data;
4950 /* copy linkea out and set on target later */
4951 rc = mdd_links_read(env, obj, ldata);
4955 mlc->mlc_opc = MD_LAYOUT_DETACH;
4956 rc = mdo_layout_change(env, obj, mlc, handle);
4960 /* don't set nlink from obj */
4961 mlc->mlc_attr->la_valid &= ~LA_NLINK;
4963 rc = mdd_create_object(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4964 NULL, NULL, NULL, hint, handle, false);
4968 rc = mdd_foreach_xattr(env, tobj, xattrs, handle, mdo_xattr_set);
4972 rc = mdd_links_write(env, tobj, ldata, handle);
4976 rc = __mdd_index_delete(env, pobj, lname->ln_name, true, handle);
4980 rc = __mdd_index_insert(env, pobj, mdd_object_fid(tobj), S_IFDIR,
4981 lname->ln_name, handle);
4985 la->la_ctime = la->la_mtime = mlc->mlc_attr->la_mtime;
4986 la->la_valid = LA_CTIME | LA_MTIME;
4988 mdd_write_lock(env, obj, DT_SRC_CHILD);
4989 rc = mdd_update_time(env, tobj, mlc->mlc_attr, la, handle);
4990 mdd_write_unlock(env, obj);
4994 rc = mdd_la_get(env, pobj, pattr);
4998 la->la_valid = LA_CTIME | LA_MTIME;
5000 mdd_write_lock(env, pobj, DT_SRC_PARENT);
5001 rc = mdd_update_time(env, pobj, pattr, la, handle);
5002 mdd_write_unlock(env, pobj);
5006 /* FID changes, record it as CL_MIGRATE */
5007 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
5008 mdd_object_fid(pobj), mdd_object_fid(obj),
5009 mdd_object_fid(pobj), lname, lname, handle);
5013 int mdd_dir_layout_split(const struct lu_env *env, struct md_object *o,
5014 struct md_layout_change *mlc)
5016 struct mdd_thread_info *info = mdd_env_info(env);
5017 struct mdd_device *mdd = mdo2mdd(o);
5018 struct mdd_object *obj = md2mdd_obj(o);
5019 struct mdd_object *pobj = md2mdd_obj(mlc->mlc_parent);
5020 struct mdd_object *tobj = md2mdd_obj(mlc->mlc_target);
5021 struct dt_allocation_hint *hint = &info->mdi_hint;
5022 bool is_plain = false;
5023 struct mdd_xattrs xattrs;
5024 struct thandle *handle;
5029 LASSERT(S_ISDIR(mdd_object_type(obj)));
5031 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LMV);
5037 mdd_xattrs_init(&xattrs);
5039 rc = mdd_xattrs_migrate_prep(env, &xattrs, obj, true, true);
5041 handle = mdd_trans_create(env, mdd);
5043 GOTO(out, rc = PTR_ERR(handle));
5046 rc = mdd_dir_declare_split_plain(env, mdd, pobj, obj, tobj,
5047 &xattrs, mlc, hint, handle);
5049 mlc->mlc_opc = MD_LAYOUT_SPLIT;
5050 rc = mdo_declare_layout_change(env, obj, mlc, handle);
5052 GOTO(stop_trans, rc);
5054 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL,
5058 GOTO(stop_trans, rc);
5060 rc = mdd_trans_start(env, mdd, handle);
5062 GOTO(stop_trans, rc);
5065 rc = mdd_dir_split_plain(env, mdd, pobj, obj, tobj, &xattrs,
5068 mdd_write_lock(env, obj, DT_TGT_CHILD);
5069 rc = mdo_xattr_set(env, obj, NULL, XATTR_NAME_LMV,
5070 LU_XATTR_CREATE, handle);
5071 mdd_write_unlock(env, obj);
5073 GOTO(stop_trans, rc);
5075 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
5076 XATTR_NAME_LMV, handle);
5079 GOTO(stop_trans, rc);
5084 rc = mdd_trans_stop(env, mdd, rc, handle);
5086 mdd_xattrs_fini(&xattrs);
5091 const struct md_dir_operations mdd_dir_ops = {
5092 .mdo_is_subdir = mdd_is_subdir,
5093 .mdo_lookup = mdd_lookup,
5094 .mdo_create = mdd_create,
5095 .mdo_rename = mdd_rename,
5096 .mdo_link = mdd_link,
5097 .mdo_unlink = mdd_unlink,
5098 .mdo_create_data = mdd_create_data,
5099 .mdo_migrate = mdd_migrate,