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, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_dir.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@intel.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <obd_class.h>
42 #include <obd_support.h>
43 #include <lustre_mds.h>
44 #include <lustre_fid.h>
46 #include "mdd_internal.h"
48 static const char dot[] = ".";
49 static const char dotdot[] = "..";
51 static struct lu_name lname_dotdot = {
57 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
59 if (!lu_name_is_valid(ln))
61 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
67 /* Get FID from name and parent */
69 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
70 const struct lu_attr *pattr, const struct lu_name *lname,
71 struct lu_fid* fid, int mask)
73 const char *name = lname->ln_name;
74 const struct dt_key *key = (const struct dt_key *)name;
75 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
76 struct mdd_device *m = mdo2mdd(pobj);
77 struct dt_object *dir = mdd_object_child(mdd_obj);
81 if (unlikely(mdd_is_dead_obj(mdd_obj)))
84 if (!mdd_object_exists(mdd_obj))
87 if (mdd_object_remote(mdd_obj)) {
88 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
89 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
92 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
97 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
98 dt_try_as_dir(env, dir)))
99 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
106 int mdd_lookup(const struct lu_env *env,
107 struct md_object *pobj, const struct lu_name *lname,
108 struct lu_fid *fid, struct md_op_spec *spec)
110 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
114 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
118 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
119 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
124 * Get parent FID of the directory
126 * Read parent FID from linkEA, if that fails, then do lookup
127 * dotdot to get the parent FID.
129 * \param[in] env execution environment
130 * \param[in] obj object from which to find the parent FID
131 * \param[in] attr attribute of the object
132 * \param[out] fid fid to get the parent FID
134 * \retval 0 if getting the parent FID succeeds.
135 * \retval negative errno if getting the parent FID fails.
137 static inline int mdd_parent_fid(const struct lu_env *env,
138 struct mdd_object *obj,
139 const struct lu_attr *attr,
142 struct mdd_thread_info *info = mdd_env_info(env);
143 struct linkea_data ldata = { NULL };
144 struct lu_buf *buf = &info->mti_link_buf;
145 struct lu_name lname;
150 LASSERT(S_ISDIR(mdd_object_type(obj)));
152 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
153 if (buf->lb_buf == NULL)
154 GOTO(lookup, rc = 0);
157 rc = mdd_links_read(env, obj, &ldata);
161 LASSERT(ldata.ld_leh != NULL);
162 /* Directory should only have 1 parent */
163 if (ldata.ld_leh->leh_reccount > 1)
166 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
168 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
169 if (likely(fid_is_sane(fid)))
172 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
177 * For root fid use special function, which does not compare version component
178 * of fid. Version component is different for root fids on all MDTs.
180 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
182 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
183 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
187 * return 1: if lf is the fid of the ancestor of p1;
190 * return -EREMOTE: if remote object is found, in this
191 * case fid of remote object is saved to @pf;
193 * otherwise: values < 0, errors.
195 static int mdd_is_parent(const struct lu_env *env,
196 struct mdd_device *mdd,
197 struct mdd_object *p1,
198 const struct lu_attr *attr,
199 const struct lu_fid *lf,
202 struct mdd_object *parent = NULL;
207 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
208 pfid = &mdd_env_info(env)->mti_fid;
210 /* Check for root first. */
211 if (mdd_is_root(mdd, mdo2fid(p1)))
215 /* this is done recursively */
216 rc = mdd_parent_fid(env, p1, attr, pfid);
219 if (mdd_is_root(mdd, pfid))
221 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
223 if (lu_fid_eq(pfid, lf))
226 mdd_object_put(env, parent);
228 parent = mdd_object_find(env, mdd, pfid);
230 GOTO(out, rc = PTR_ERR(parent));
232 if (!mdd_object_exists(parent))
233 GOTO(out, rc = -EINVAL);
239 if (parent && !IS_ERR(parent))
240 mdd_object_put(env, parent);
245 * No permission check is needed.
247 * returns 1: if fid is ancestor of @mo;
248 * returns 0: if fid is not an ancestor of @mo;
250 * returns EREMOTE if remote object is found, fid of remote object is saved to
253 * returns < 0: if error
255 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
256 const struct lu_fid *fid, struct lu_fid *sfid)
258 struct mdd_device *mdd = mdo2mdd(mo);
259 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
263 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
266 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
270 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
274 } else if (rc == 1) {
275 /* found @fid is parent */
283 * Check that @dir contains no entries except (possibly) dot and dotdot.
288 * -ENOTDIR not a directory object
289 * -ENOTEMPTY not empty
293 static int mdd_dir_is_empty(const struct lu_env *env,
294 struct mdd_object *dir)
297 struct dt_object *obj;
298 const struct dt_it_ops *iops;
302 obj = mdd_object_child(dir);
303 if (!dt_try_as_dir(env, obj))
306 iops = &obj->do_index_ops->dio_it;
307 it = iops->init(env, obj, LUDA_64BITHASH);
309 result = iops->get(env, it, (const struct dt_key *)"");
312 for (result = 0, i = 0; result == 0 && i < 3; ++i)
313 result = iops->next(env, it);
316 else if (result == 1)
318 } else if (result == 0)
320 * Huh? Index contains no zero key?
327 result = PTR_ERR(it);
332 * Determine if the target object can be hard linked, and right now it only
333 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
334 * directory nlink count might exceed the maximum link count(see
335 * osd_object_ref_add), so it only check nlink for non-directories.
337 * \param[in] env thread environment
338 * \param[in] obj object being linked to
339 * \param[in] la attributes of \a obj
341 * \retval 0 if \a obj can be hard linked
342 * \retval negative error if \a obj is a directory or has too
345 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
346 const struct lu_attr *la)
348 struct mdd_device *m = mdd_obj2mdd_dev(obj);
353 /* Subdir count limitation can be broken through
354 * (see osd_object_ref_add), so only check non-directory here. */
355 if (!S_ISDIR(la->la_mode) &&
356 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
363 * Check whether it may create the cobj under the pobj.
365 * \param[in] env execution environment
366 * \param[in] pobj the parent directory
367 * \param[in] pattr the attribute of the parent directory
368 * \param[in] cobj the child to be created
369 * \param[in] check_perm if check WRITE|EXEC permission for parent
371 * \retval = 0 create the child under this dir is allowed
372 * \retval negative errno create the child under this dir is
375 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
376 const struct lu_attr *pattr, struct mdd_object *cobj,
382 if (cobj && mdd_object_exists(cobj))
385 if (mdd_is_dead_obj(pobj))
389 rc = mdd_permission_internal_locked(env, pobj, pattr,
390 MAY_WRITE | MAY_EXEC,
396 * Check whether can unlink from the pobj in the case of "cobj == NULL".
398 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
399 const struct lu_attr *pattr, const struct lu_attr *attr)
404 if (mdd_is_dead_obj(pobj))
407 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
410 rc = mdd_permission_internal_locked(env, pobj, pattr,
411 MAY_WRITE | MAY_EXEC,
416 if (pattr->la_flags & LUSTRE_APPEND_FL)
423 * pobj == NULL is remote ops case, under such case, pobj's
424 * VTX feature has been checked already, no need check again.
426 static inline int mdd_is_sticky(const struct lu_env *env,
427 struct mdd_object *pobj,
428 const struct lu_attr *pattr,
429 struct mdd_object *cobj,
430 const struct lu_attr *cattr)
432 struct lu_ucred *uc = lu_ucred_assert(env);
435 LASSERT(pattr != NULL);
436 if (!(pattr->la_mode & S_ISVTX) ||
437 (pattr->la_uid == uc->uc_fsuid))
441 LASSERT(cattr != NULL);
442 if (cattr->la_uid == uc->uc_fsuid)
445 return !md_capable(uc, CFS_CAP_FOWNER);
448 static int mdd_may_delete_entry(const struct lu_env *env,
449 struct mdd_object *pobj,
450 const struct lu_attr *pattr,
455 LASSERT(pobj != NULL);
456 if (!mdd_object_exists(pobj))
459 if (mdd_is_dead_obj(pobj))
464 rc = mdd_permission_internal_locked(env, pobj, pattr,
465 MAY_WRITE | MAY_EXEC,
471 if (pattr->la_flags & LUSTRE_APPEND_FL)
478 * Check whether it may delete the cobj from the pobj.
481 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
482 const struct lu_attr *tpattr, struct mdd_object *tobj,
483 const struct lu_attr *tattr, const struct lu_attr *cattr,
484 int check_perm, int check_empty)
490 LASSERT(tpattr != NULL);
491 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
499 if (!mdd_object_exists(tobj))
502 if (mdd_is_dead_obj(tobj))
505 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
508 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
511 /* additional check the rename case */
513 if (S_ISDIR(cattr->la_mode)) {
514 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
516 if (!S_ISDIR(tattr->la_mode))
519 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
521 } else if (S_ISDIR(tattr->la_mode))
525 if (S_ISDIR(tattr->la_mode) && check_empty)
526 rc = mdd_dir_is_empty(env, tobj);
532 * Check whether it can create the link file(linked to @src_obj) under
533 * the target directory(@tgt_obj), and src_obj has been locked by
536 * \param[in] env execution environment
537 * \param[in] tgt_obj the target directory
538 * \param[in] tattr attributes of target directory
539 * \param[in] lname the link name
540 * \param[in] src_obj source object for link
541 * \param[in] cattr attributes for source object
543 * \retval = 0 it is allowed to create the link file under tgt_obj
544 * \retval negative error not allowed to create the link file
546 static int mdd_link_sanity_check(const struct lu_env *env,
547 struct mdd_object *tgt_obj,
548 const struct lu_attr *tattr,
549 const struct lu_name *lname,
550 struct mdd_object *src_obj,
551 const struct lu_attr *cattr)
553 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
557 if (!mdd_object_exists(src_obj))
560 if (mdd_is_dead_obj(src_obj))
563 /* Local ops, no lookup before link, check filename length here. */
564 rc = mdd_name_check(m, lname);
568 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
571 if (S_ISDIR(mdd_object_type(src_obj)))
574 LASSERT(src_obj != tgt_obj);
575 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
579 rc = __mdd_may_link(env, src_obj, cattr);
584 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
585 const char *name, struct thandle *handle)
587 struct dt_object *next = mdd_object_child(pobj);
591 if (dt_try_as_dir(env, next))
592 rc = dt_delete(env, next, (struct dt_key *)name, handle);
599 static int __mdd_index_insert_only(const struct lu_env *env,
600 struct mdd_object *pobj,
601 const struct lu_fid *lf, __u32 type,
603 struct thandle *handle)
605 struct dt_object *next = mdd_object_child(pobj);
609 if (dt_try_as_dir(env, next)) {
610 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
611 struct lu_ucred *uc = lu_ucred_check(env);
615 rec->rec_type = type;
616 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
617 rc = dt_insert(env, next, (const struct dt_rec *)rec,
618 (const struct dt_key *)name, handle,
626 /* insert named index, add reference if isdir */
627 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
628 const struct lu_fid *lf, __u32 type,
629 const char *name, struct thandle *handle)
634 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
635 if (rc == 0 && S_ISDIR(type)) {
636 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
637 mdo_ref_add(env, pobj, handle);
638 mdd_write_unlock(env, pobj);
644 /* delete named index, drop reference if isdir */
645 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
646 const char *name, int is_dir,
647 struct thandle *handle)
652 rc = __mdd_index_delete_only(env, pobj, name, handle);
653 if (rc == 0 && is_dir) {
654 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
655 mdo_ref_del(env, pobj, handle);
656 mdd_write_unlock(env, pobj);
662 static int mdd_llog_record_calc_size(const struct lu_env *env,
663 const struct lu_name *tname,
664 const struct lu_name *sname)
666 const struct lu_ucred *uc = lu_ucred(env);
667 enum changelog_rec_flags crf = 0;
668 size_t hdr_size = sizeof(struct llog_changelog_rec) -
669 sizeof(struct changelog_rec);
674 if (uc != NULL && uc->uc_jobid[0] != '\0')
677 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
678 (tname != NULL ? tname->ln_namelen : 0) +
679 (sname != NULL ? 1 + sname->ln_namelen : 0));
682 int mdd_declare_changelog_store(const struct lu_env *env,
683 struct mdd_device *mdd,
684 const struct lu_name *tname,
685 const struct lu_name *sname,
686 struct thandle *handle)
688 struct obd_device *obd = mdd2obd_dev(mdd);
689 struct llog_ctxt *ctxt;
690 struct llog_changelog_rec *rec;
692 struct thandle *llog_th;
697 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
700 reclen = mdd_llog_record_calc_size(env, tname, sname);
701 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
702 if (buf->lb_buf == NULL)
706 rec->cr_hdr.lrh_len = reclen;
707 rec->cr_hdr.lrh_type = CHANGELOG_REC;
709 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
713 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
715 GOTO(out_put, rc = PTR_ERR(llog_th));
717 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
725 /** Add a changelog entry \a rec to the changelog llog
728 * \param handle - currently ignored since llogs start their own transaction;
729 * this will hopefully be fixed in llog rewrite
732 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
733 struct llog_changelog_rec *rec, struct thandle *th)
735 struct obd_device *obd = mdd2obd_dev(mdd);
736 struct llog_ctxt *ctxt;
737 struct thandle *llog_th;
740 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
741 changelog_rec_varsize(&rec->cr));
743 /* llog_lvfs_write_rec sets the llog tail len */
744 rec->cr_hdr.lrh_type = CHANGELOG_REC;
745 rec->cr.cr_time = cl_time();
747 spin_lock(&mdd->mdd_cl.mc_lock);
748 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
749 * but as long as the MDD transactions are ordered correctly for e.g.
750 * rename conflicts, I don't think this should matter. */
751 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
752 spin_unlock(&mdd->mdd_cl.mc_lock);
754 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
758 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
760 GOTO(out_put, rc = PTR_ERR(llog_th));
762 /* nested journal transaction */
763 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
772 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
773 const struct lu_fid *sfid,
774 const struct lu_fid *spfid,
775 const struct lu_name *sname)
777 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
778 size_t extsize = sname->ln_namelen + 1;
780 LASSERT(sfid != NULL);
781 LASSERT(spfid != NULL);
782 LASSERT(sname != NULL);
784 rnm->cr_sfid = *sfid;
785 rnm->cr_spfid = *spfid;
787 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
788 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
789 rec->cr_namelen += extsize;
792 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
794 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
796 if (jobid == NULL || jobid[0] == '\0')
799 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
802 /** Store a namespace change changelog record
803 * If this fails, we must fail the whole transaction; we don't
804 * want the change to commit without the log entry.
805 * \param target - mdd_object of change
806 * \param tpfid - target parent dir/object fid
807 * \param sfid - source object fid
808 * \param spfid - source parent fid
809 * \param tname - target name string
810 * \param sname - source name string
811 * \param handle - transaction handle
813 int mdd_changelog_ns_store(const struct lu_env *env,
814 struct mdd_device *mdd,
815 enum changelog_rec_type type,
816 enum changelog_rec_flags crf,
817 struct mdd_object *target,
818 const struct lu_fid *tpfid,
819 const struct lu_fid *sfid,
820 const struct lu_fid *spfid,
821 const struct lu_name *tname,
822 const struct lu_name *sname,
823 struct thandle *handle)
825 const struct lu_ucred *uc = lu_ucred(env);
826 struct llog_changelog_rec *rec;
833 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
836 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
839 LASSERT(tpfid != NULL);
840 LASSERT(tname != NULL);
841 LASSERT(handle != NULL);
843 reclen = mdd_llog_record_calc_size(env, tname, sname);
844 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
845 if (buf->lb_buf == NULL)
851 if (uc != NULL && uc->uc_jobid[0] != '\0')
859 rec->cr.cr_flags = crf;
860 rec->cr.cr_type = (__u32)type;
861 rec->cr.cr_pfid = *tpfid;
862 rec->cr.cr_namelen = tname->ln_namelen;
863 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
865 if (crf & CLF_RENAME)
866 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
869 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
871 if (likely(target != NULL)) {
872 rec->cr.cr_tfid = *mdo2fid(target);
873 target->mod_cltime = cfs_time_current_64();
875 fid_zero(&rec->cr.cr_tfid);
878 rc = mdd_changelog_store(env, mdd, rec, handle);
880 CERROR("%s: cannot store changelog record: type = %d, "
881 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
882 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
883 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
890 static int __mdd_links_add(const struct lu_env *env,
891 struct mdd_object *mdd_obj,
892 struct linkea_data *ldata,
893 const struct lu_name *lname,
894 const struct lu_fid *pfid,
895 int first, int check)
899 if (ldata->ld_leh == NULL) {
900 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
904 rc = linkea_data_new(ldata,
905 &mdd_env_info(env)->mti_link_buf);
912 rc = linkea_links_find(ldata, lname, pfid);
913 if (rc && rc != -ENOENT)
919 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
920 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
924 linkea_add_buf(ldata, lname, tfid);
927 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
928 linkea_add_buf(ldata, lname, pfid);
930 return linkea_add_buf(ldata, lname, pfid);
933 static int __mdd_links_del(const struct lu_env *env,
934 struct mdd_object *mdd_obj,
935 struct linkea_data *ldata,
936 const struct lu_name *lname,
937 const struct lu_fid *pfid)
941 if (ldata->ld_leh == NULL) {
942 rc = mdd_links_read(env, mdd_obj, ldata);
947 rc = linkea_links_find(ldata, lname, pfid);
951 linkea_del_buf(ldata, lname);
955 static int mdd_linkea_prepare(const struct lu_env *env,
956 struct mdd_object *mdd_obj,
957 const struct lu_fid *oldpfid,
958 const struct lu_name *oldlname,
959 const struct lu_fid *newpfid,
960 const struct lu_name *newlname,
961 int first, int check,
962 struct linkea_data *ldata)
968 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
971 LASSERT(oldpfid != NULL || newpfid != NULL);
973 if (mdd_obj->mod_flags & DEAD_OBJ) {
974 /* Prevent linkea to be updated which is NOT necessary. */
975 ldata->ld_reclen = 0;
976 /* No more links, don't bother */
980 if (oldpfid != NULL) {
981 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
984 (rc != -ENODATA && rc != -ENOENT))
986 /* No changes done. */
991 /* If renaming, add the new record */
992 if (newpfid != NULL) {
993 /* even if the add fails, we still delete the out-of-date
995 rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
999 rc = rc != 0 ? rc : rc2;
1004 int mdd_links_rename(const struct lu_env *env,
1005 struct mdd_object *mdd_obj,
1006 const struct lu_fid *oldpfid,
1007 const struct lu_name *oldlname,
1008 const struct lu_fid *newpfid,
1009 const struct lu_name *newlname,
1010 struct thandle *handle,
1011 struct linkea_data *ldata,
1012 int first, int check)
1017 if (ldata == NULL) {
1018 ldata = &mdd_env_info(env)->mti_link_data;
1019 memset(ldata, 0, sizeof(*ldata));
1020 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1021 newpfid, newlname, first, check,
1027 if (ldata->ld_reclen != 0)
1028 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1033 if (rc == -EOVERFLOW || rc == -ENOSPC)
1035 if (newlname == NULL)
1036 CDEBUG(error ? D_ERROR : D_OTHER,
1037 "link_ea add failed %d "DFID"\n",
1038 rc, PFID(mdd_object_fid(mdd_obj)));
1039 else if (oldpfid == NULL)
1040 CDEBUG(error ? D_ERROR : D_OTHER,
1041 "link_ea add '%.*s' failed %d "DFID"\n",
1042 newlname->ln_namelen, newlname->ln_name,
1043 rc, PFID(mdd_object_fid(mdd_obj)));
1044 else if (newpfid == NULL)
1045 CDEBUG(error ? D_ERROR : D_OTHER,
1046 "link_ea del '%.*s' failed %d "DFID"\n",
1047 oldlname->ln_namelen, oldlname->ln_name,
1048 rc, PFID(mdd_object_fid(mdd_obj)));
1050 CDEBUG(error ? D_ERROR : D_OTHER,
1051 "link_ea rename '%.*s'->'%.*s' failed %d "
1053 oldlname->ln_namelen, oldlname->ln_name,
1054 newlname->ln_namelen, newlname->ln_name,
1055 rc, PFID(mdd_object_fid(mdd_obj)));
1058 if (is_vmalloc_addr(ldata->ld_buf))
1059 /* if we vmalloced a large buffer drop it */
1060 lu_buf_free(ldata->ld_buf);
1065 static inline int mdd_links_add(const struct lu_env *env,
1066 struct mdd_object *mdd_obj,
1067 const struct lu_fid *pfid,
1068 const struct lu_name *lname,
1069 struct thandle *handle,
1070 struct linkea_data *ldata, int first)
1072 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1073 pfid, lname, handle, ldata, first, 0);
1076 static inline int mdd_links_del(const struct lu_env *env,
1077 struct mdd_object *mdd_obj,
1078 const struct lu_fid *pfid,
1079 const struct lu_name *lname,
1080 struct thandle *handle)
1082 return mdd_links_rename(env, mdd_obj, pfid, lname,
1083 NULL, NULL, handle, NULL, 0, 0);
1086 /** Read the link EA into a temp buffer.
1087 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
1088 * A pointer to the buffer is stored in \a ldata::ld_buf.
1090 * \retval 0 or error
1092 int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
1093 struct linkea_data *ldata)
1097 if (!mdd_object_exists(mdd_obj))
1100 /* First try a small buf */
1101 LASSERT(env != NULL);
1102 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
1104 if (ldata->ld_buf->lb_buf == NULL)
1107 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
1108 if (rc == -ERANGE) {
1109 /* Buf was too small, figure out what we need. */
1110 lu_buf_free(ldata->ld_buf);
1111 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1115 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
1116 if (ldata->ld_buf->lb_buf == NULL)
1118 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
1122 lu_buf_free(ldata->ld_buf);
1123 ldata->ld_buf = NULL;
1127 return linkea_init(ldata);
1130 /** Read the link EA into a temp buffer.
1131 * Uses the name_buf since it is generally large.
1132 * \retval IS_ERR err
1133 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1135 struct lu_buf *mdd_links_get(const struct lu_env *env,
1136 struct mdd_object *mdd_obj)
1138 struct linkea_data ldata = { NULL };
1141 rc = mdd_links_read(env, mdd_obj, &ldata);
1142 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1145 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1146 struct linkea_data *ldata, struct thandle *handle)
1148 const struct lu_buf *buf;
1151 if (ldata == NULL || ldata->ld_buf == NULL ||
1152 ldata->ld_leh == NULL)
1155 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1156 ldata->ld_leh->leh_len);
1157 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1160 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1161 if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) &&
1162 mdd_object_remote(mdd_obj) == 0) {
1163 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1164 struct thandle *sub_th;
1166 /* XXX: If the linkEA is overflow, then we need to notify the
1167 * namespace LFSCK to skip "nlink" attribute verification
1168 * on this object to avoid the "nlink" to be shrinked by
1169 * wrong. It may be not good an interaction with LFSCK
1170 * like this. We will consider to replace it with other
1171 * mechanism in future. LU-5802. */
1172 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK,
1173 LFSCK_TYPE_NAMESPACE);
1175 sub_th = thandle_get_sub_by_dt(env, handle,
1176 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom);
1177 lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1184 int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
1185 struct thandle *handle, struct linkea_data *ldata,
1186 enum mdd_links_add_overflow overflow)
1192 if (ldata != NULL && ldata->ld_leh != NULL) {
1193 ea_len = ldata->ld_leh->leh_len;
1194 linkea = ldata->ld_buf->lb_buf;
1196 ea_len = DEFAULT_LINKEA_SIZE;
1200 /* XXX: max size? */
1201 rc = mdo_declare_xattr_set(env, mdd_obj,
1202 mdd_buf_get_const(env, linkea, ea_len),
1203 XATTR_NAME_LINK, 0, handle);
1207 if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) {
1208 struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
1209 struct thandle *sub_th;
1211 /* XXX: If the linkEA is overflow, then we need to notify the
1212 * namespace LFSCK to skip "nlink" attribute verification
1213 * on this object to avoid the "nlink" to be shrinked by
1214 * wrong. It may be not good an interaction with LFSCK
1215 * like this. We will consider to replace it with other
1216 * mechanism in future. LU-5802. */
1217 lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE,
1218 LFSCK_TYPE_NAMESPACE);
1220 sub_th = thandle_get_sub_by_dt(env, handle,
1221 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom);
1222 rc = lfsck_in_notify(env,
1223 mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
1230 static inline int mdd_declare_links_del(const struct lu_env *env,
1231 struct mdd_object *c,
1232 struct thandle *handle)
1236 /* For directory, the linkEA will be removed together
1237 * with the object. */
1238 if (!S_ISDIR(mdd_object_type(c)))
1239 rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE);
1244 static int mdd_declare_link(const struct lu_env *env,
1245 struct mdd_device *mdd,
1246 struct mdd_object *p,
1247 struct mdd_object *c,
1248 const struct lu_name *name,
1249 struct thandle *handle,
1251 struct linkea_data *data)
1253 struct lu_fid tfid = *mdo2fid(c);
1256 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1257 tfid.f_oid = cfs_fail_val;
1259 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1260 name->ln_name, handle);
1264 rc = mdo_declare_ref_add(env, c, handle);
1268 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1269 rc = mdo_declare_ref_add(env, c, handle);
1274 la->la_valid = LA_CTIME | LA_MTIME;
1275 rc = mdo_declare_attr_set(env, p, la, handle);
1279 la->la_valid = LA_CTIME;
1280 rc = mdo_declare_attr_set(env, c, la, handle);
1284 rc = mdd_declare_links_add(env, c, handle, data,
1285 S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE);
1289 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1294 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1295 struct md_object *src_obj, const struct lu_name *lname,
1298 const char *name = lname->ln_name;
1299 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1300 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1301 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1302 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1303 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1304 struct mdd_device *mdd = mdo2mdd(src_obj);
1305 struct thandle *handle;
1306 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1307 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1311 rc = mdd_la_get(env, mdd_sobj, cattr);
1315 rc = mdd_la_get(env, mdd_tobj, tattr);
1319 handle = mdd_trans_create(env, mdd);
1321 GOTO(out_pending, rc = PTR_ERR(handle));
1323 memset(ldata, 0, sizeof(*ldata));
1325 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1326 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1328 /* Note: even this function will change ldata, but it comes from
1329 * thread_info, which is completely temporary and only seen in
1330 * this function, so we do not need reset ldata once it fails.*/
1331 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1332 lname, 0, 0, ldata);
1336 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1341 rc = mdd_trans_start(env, mdd, handle);
1345 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1346 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1349 GOTO(out_unlock, rc);
1351 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1352 rc = mdo_ref_add(env, mdd_sobj, handle);
1354 GOTO(out_unlock, rc);
1357 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
1358 rc = mdo_ref_add(env, mdd_sobj, handle);
1360 GOTO(out_unlock, rc);
1363 *tfid = *mdo2fid(mdd_sobj);
1364 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1365 tfid->f_oid = cfs_fail_val;
1367 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1368 mdd_object_type(mdd_sobj), name, handle);
1370 mdo_ref_del(env, mdd_sobj, handle);
1371 GOTO(out_unlock, rc);
1374 la->la_valid = LA_CTIME | LA_MTIME;
1375 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1377 GOTO(out_unlock, rc);
1379 la->la_valid = LA_CTIME;
1380 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1382 /* Note: The failure of links_add should not cause the
1383 * link failure, so do not check return value. */
1384 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1385 lname, handle, ldata, 0);
1389 mdd_write_unlock(env, mdd_sobj);
1391 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1392 mdo2fid(mdd_tobj), NULL, NULL,
1393 lname, NULL, handle);
1395 rc = mdd_trans_stop(env, mdd, rc, handle);
1396 if (is_vmalloc_addr(ldata->ld_buf))
1397 /* if we vmalloced a large buffer drop it */
1398 lu_buf_free(ldata->ld_buf);
1403 static int mdd_mark_orphan_object(const struct lu_env *env,
1404 struct mdd_object *obj, struct thandle *handle,
1407 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1410 if (!S_ISDIR(mdd_object_type(obj)))
1413 attr->la_valid = LA_FLAGS;
1414 attr->la_flags = LUSTRE_ORPHAN_FL;
1417 rc = mdo_declare_attr_set(env, obj, attr, handle);
1419 rc = mdo_attr_set(env, obj, attr, handle);
1424 static int mdd_declare_finish_unlink(const struct lu_env *env,
1425 struct mdd_object *obj,
1426 struct thandle *handle)
1430 /* Sigh, we do not know if the unlink object will become orphan in
1431 * declare phase, but fortunately the flags here does not matter
1432 * in current declare implementation */
1433 rc = mdd_mark_orphan_object(env, obj, handle, true);
1437 rc = mdo_declare_destroy(env, obj, handle);
1441 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1445 return mdd_declare_links_del(env, obj, handle);
1448 /* caller should take a lock before calling */
1449 int mdd_finish_unlink(const struct lu_env *env,
1450 struct mdd_object *obj, struct md_attr *ma,
1451 const struct mdd_object *pobj,
1452 const struct lu_name *lname,
1456 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1459 LASSERT(mdd_write_locked(env, obj) != 0);
1461 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1462 /* add new orphan and the object
1463 * will be deleted during mdd_close() */
1464 obj->mod_flags |= DEAD_OBJ;
1465 if (obj->mod_count) {
1466 rc = __mdd_orphan_add(env, obj, th);
1468 CDEBUG(D_HA, "Object "DFID" is inserted into "
1469 "orphan list, open count = %d\n",
1470 PFID(mdd_object_fid(obj)),
1473 CERROR("Object "DFID" fail to be an orphan, "
1474 "open count = %d, maybe cause failed "
1476 PFID(mdd_object_fid(obj)),
1479 /* mark object as an orphan here, not
1480 * before __mdd_orphan_add() as racing
1481 * mdd_la_get() may propagate ORPHAN_OBJ
1482 * causing the asserition */
1483 rc = mdd_mark_orphan_object(env, obj, th, false);
1485 rc = mdo_destroy(env, obj, th);
1487 } else if (!is_dir) {
1488 /* old files may not have link ea; ignore errors */
1489 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1497 * has mdd_write_lock on cobj already, but not on pobj yet
1499 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1500 const struct lu_attr *pattr,
1501 struct mdd_object *cobj,
1502 const struct lu_attr *cattr)
1507 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1512 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1513 struct mdd_object *p, struct mdd_object *c,
1514 const struct lu_name *name, struct md_attr *ma,
1515 struct thandle *handle, int no_name, int is_dir)
1517 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1520 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1521 if (likely(no_name == 0)) {
1522 rc = mdo_declare_index_delete(env, p, name->ln_name,
1529 rc = mdo_declare_ref_del(env, p, handle);
1535 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1536 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1537 la->la_valid = LA_CTIME | LA_MTIME;
1538 rc = mdo_declare_attr_set(env, p, la, handle);
1543 rc = mdo_declare_ref_del(env, c, handle);
1547 rc = mdo_declare_ref_del(env, c, handle);
1551 la->la_valid = LA_CTIME;
1552 rc = mdo_declare_attr_set(env, c, la, handle);
1556 rc = mdd_declare_finish_unlink(env, c, handle);
1560 /* FIXME: need changelog for remove entry */
1561 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1568 * test if a file has an HSM archive
1569 * if HSM attributes are not found in ma update them from
1572 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1573 struct mdd_object *obj,
1578 if (!(ma->ma_valid & MA_HSM)) {
1579 /* no HSM MD provided, read xattr */
1580 struct lu_buf *hsm_buf;
1581 const size_t buflen = sizeof(struct hsm_attrs);
1584 hsm_buf = mdd_buf_get(env, NULL, 0);
1585 lu_buf_alloc(hsm_buf, buflen);
1586 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1587 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1588 lu_buf_free(hsm_buf);
1592 ma->ma_valid |= MA_HSM;
1594 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1600 * Delete name entry and the object.
1601 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1602 * does not exist for this object, and it could only happen during resending
1603 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1604 * is also needed in this case(needed by changelog), so we have to add another
1605 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1606 * the ENOENT failure should be able to be fixed by redo mechanism.
1608 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1609 struct md_object *cobj, const struct lu_name *lname,
1610 struct md_attr *ma, int no_name)
1612 const char *name = lname->ln_name;
1613 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1614 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1615 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1616 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1617 struct mdd_object *mdd_cobj = NULL;
1618 struct mdd_device *mdd = mdo2mdd(pobj);
1619 struct thandle *handle;
1620 int rc, is_dir = 0, cl_flags = 0;
1623 /* cobj == NULL means only delete name entry */
1624 if (likely(cobj != NULL)) {
1625 mdd_cobj = md2mdd_obj(cobj);
1626 if (mdd_object_exists(mdd_cobj) == 0)
1630 rc = mdd_la_get(env, mdd_pobj, pattr);
1634 if (likely(mdd_cobj != NULL)) {
1636 rc = mdd_la_get(env, mdd_cobj, cattr);
1640 is_dir = S_ISDIR(cattr->la_mode);
1641 /* search for an existing archive.
1642 * we should check ahead as the object
1643 * can be destroyed in this transaction */
1644 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1645 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1648 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1652 handle = mdd_trans_create(env, mdd);
1654 RETURN(PTR_ERR(handle));
1656 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1657 lname, ma, handle, no_name, is_dir);
1661 rc = mdd_trans_start(env, mdd, handle);
1665 if (likely(mdd_cobj != NULL))
1666 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1668 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1669 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1674 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1675 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1676 GOTO(cleanup, rc = 0);
1678 if (likely(mdd_cobj != NULL)) {
1679 rc = mdo_ref_del(env, mdd_cobj, handle);
1681 __mdd_index_insert_only(env, mdd_pobj,
1683 mdd_object_type(mdd_cobj),
1690 mdo_ref_del(env, mdd_cobj, handle);
1692 /* fetch updated nlink */
1693 rc = mdd_la_get(env, mdd_cobj, cattr);
1698 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1699 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1701 la->la_valid = LA_CTIME | LA_MTIME;
1702 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1706 /* Enough for only unlink the entry */
1707 if (unlikely(mdd_cobj == NULL))
1710 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1711 /* update ctime of an unlinked file only if it is still
1712 * opened or a link still exists */
1713 la->la_valid = LA_CTIME;
1714 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1719 /* XXX: this transfer to ma will be removed with LOD/OSP */
1720 ma->ma_attr = *cattr;
1721 ma->ma_valid |= MA_INODE;
1722 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1726 /* fetch updated nlink */
1727 rc = mdd_la_get(env, mdd_cobj, cattr);
1728 /* if object is removed then we can't get its attrs,
1730 if (rc == -ENOENT) {
1731 cattr->la_nlink = 0;
1735 if (cattr->la_nlink == 0) {
1736 ma->ma_attr = *cattr;
1737 ma->ma_valid |= MA_INODE;
1742 if (likely(mdd_cobj != NULL))
1743 mdd_write_unlock(env, mdd_cobj);
1746 if (cattr->la_nlink == 0)
1747 cl_flags |= CLF_UNLINK_LAST;
1749 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1751 rc = mdd_changelog_ns_store(env, mdd,
1752 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1753 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1758 rc = mdd_trans_stop(env, mdd, rc, handle);
1764 * The permission has been checked when obj created, no need check again.
1766 static int mdd_cd_sanity_check(const struct lu_env *env,
1767 struct mdd_object *obj)
1772 if (!obj || mdd_is_dead_obj(obj))
1778 static int mdd_create_data(const struct lu_env *env,
1779 struct md_object *pobj,
1780 struct md_object *cobj,
1781 const struct md_op_spec *spec,
1784 struct mdd_device *mdd = mdo2mdd(cobj);
1785 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1786 struct mdd_object *son = md2mdd_obj(cobj);
1787 struct thandle *handle;
1788 const struct lu_buf *buf;
1789 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1790 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1794 rc = mdd_cd_sanity_check(env, son);
1798 if (!md_should_create(spec->sp_cr_flags))
1802 * there are following use cases for this function:
1803 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1804 * striping can be specified or not
1807 rc = mdd_la_get(env, son, attr);
1811 /* calling ->ah_make_hint() is used to transfer information from parent */
1812 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1814 handle = mdd_trans_create(env, mdd);
1816 GOTO(out_free, rc = PTR_ERR(handle));
1819 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1820 * Should this be fixed?
1822 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1823 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1824 spec->sp_cr_flags, spec->no_create);
1826 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1827 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1828 spec->u.sp_ea.eadatalen);
1833 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1834 XATTR_NAME_LOV, 0, handle);
1838 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1842 rc = mdd_trans_start(env, mdd, handle);
1846 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1852 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1855 rc = mdd_trans_stop(env, mdd, rc, handle);
1861 static int mdd_declare_object_initialize(const struct lu_env *env,
1862 struct mdd_object *parent,
1863 struct mdd_object *child,
1864 const struct lu_attr *attr,
1865 struct thandle *handle)
1870 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1871 if (!S_ISDIR(attr->la_mode))
1874 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1879 rc = mdo_declare_ref_add(env, child, handle);
1883 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1889 static int mdd_object_initialize(const struct lu_env *env,
1890 const struct lu_fid *pfid,
1891 struct mdd_object *child,
1892 struct lu_attr *attr, struct thandle *handle,
1893 const struct md_op_spec *spec)
1898 if (S_ISDIR(attr->la_mode)) {
1899 /* Add "." and ".." for newly created dir */
1900 mdo_ref_add(env, child, handle);
1901 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1902 S_IFDIR, dot, handle);
1904 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1907 mdo_ref_del(env, child, handle);
1914 * This function checks whether it can create a file/dir under the
1915 * directory(@pobj). The directory(@pobj) is not being locked by
1918 * \param[in] env execution environment
1919 * \param[in] pobj the directory to create files
1920 * \param[in] pattr the attributes of the directory
1921 * \param[in] lname the name of the created file/dir
1922 * \param[in] cattr the attributes of the file/dir
1923 * \param[in] spec create specification
1925 * \retval = 0 it is allowed to create file/dir under
1927 * \retval negative error not allowed to create file/dir
1928 * under the directory
1930 static int mdd_create_sanity_check(const struct lu_env *env,
1931 struct md_object *pobj,
1932 const struct lu_attr *pattr,
1933 const struct lu_name *lname,
1934 struct lu_attr *cattr,
1935 struct md_op_spec *spec)
1937 struct mdd_thread_info *info = mdd_env_info(env);
1938 struct lu_fid *fid = &info->mti_fid;
1939 struct mdd_object *obj = md2mdd_obj(pobj);
1940 struct mdd_device *m = mdo2mdd(pobj);
1941 bool check_perm = true;
1946 if (mdd_is_dead_obj(obj))
1950 * In some cases this lookup is not needed - we know before if name
1951 * exists or not because MDT performs lookup for it.
1952 * name length check is done in lookup.
1954 if (spec->sp_cr_lookup) {
1956 * Check if the name already exist, though it will be checked in
1957 * _index_insert also, for avoiding rolling back if exists
1960 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1961 MAY_WRITE | MAY_EXEC);
1963 RETURN(rc ? : -EEXIST);
1965 /* Permission is already being checked in mdd_lookup */
1969 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1974 if (pattr->la_mode & S_ISGID) {
1975 cattr->la_gid = pattr->la_gid;
1976 if (S_ISDIR(cattr->la_mode)) {
1977 cattr->la_mode |= S_ISGID;
1978 cattr->la_valid |= LA_MODE;
1982 rc = mdd_name_check(m, lname);
1986 switch (cattr->la_mode & S_IFMT) {
1988 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1990 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
1991 RETURN(-ENAMETOOLONG);
2010 static int mdd_declare_object_create(const struct lu_env *env,
2011 struct mdd_device *mdd,
2012 struct mdd_object *p, struct mdd_object *c,
2013 struct lu_attr *attr,
2014 struct thandle *handle,
2015 const struct md_op_spec *spec,
2016 struct lu_buf *def_acl_buf,
2017 struct lu_buf *acl_buf,
2018 struct dt_allocation_hint *hint)
2020 const struct lu_buf *buf;
2023 rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
2028 #ifdef CONFIG_FS_POSIX_ACL
2029 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2030 /* if dir, then can inherit default ACl */
2031 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2032 XATTR_NAME_ACL_DEFAULT,
2038 if (acl_buf->lb_len > 0) {
2039 rc = mdo_declare_attr_set(env, c, attr, handle);
2043 rc = mdo_declare_xattr_set(env, c, acl_buf,
2044 XATTR_NAME_ACL_ACCESS, 0, handle);
2049 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2053 /* replay case, create LOV EA from client data */
2054 if (spec->no_create ||
2055 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2056 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2057 spec->u.sp_ea.eadatalen);
2058 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2064 if (S_ISLNK(attr->la_mode)) {
2065 const char *target_name = spec->u.sp_symname;
2066 int sym_len = strlen(target_name);
2067 const struct lu_buf *buf;
2069 buf = mdd_buf_get_const(env, target_name, sym_len);
2070 rc = dt_declare_record_write(env, mdd_object_child(c),
2076 if (spec->sp_cr_file_secctx_name != NULL) {
2077 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2078 spec->sp_cr_file_secctx_size);
2079 rc = mdo_declare_xattr_set(env, c, buf,
2080 spec->sp_cr_file_secctx_name, 0,
2089 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2090 struct mdd_object *p, struct mdd_object *c,
2091 const struct lu_name *name,
2092 struct lu_attr *attr,
2093 struct thandle *handle,
2094 const struct md_op_spec *spec,
2095 struct linkea_data *ldata,
2096 struct lu_buf *def_acl_buf,
2097 struct lu_buf *acl_buf,
2098 struct dt_allocation_hint *hint)
2102 rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
2103 def_acl_buf, acl_buf, hint);
2107 if (S_ISDIR(attr->la_mode)) {
2108 rc = mdo_declare_ref_add(env, p, handle);
2113 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2114 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2118 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2120 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2121 name->ln_name, handle);
2125 rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
2130 la->la_valid = LA_CTIME | LA_MTIME;
2131 rc = mdo_declare_attr_set(env, p, la, handle);
2135 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2143 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2144 struct lu_attr *la, struct lu_buf *def_acl_buf,
2145 struct lu_buf *acl_buf)
2150 if (S_ISLNK(la->la_mode)) {
2151 acl_buf->lb_len = 0;
2152 def_acl_buf->lb_len = 0;
2156 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2157 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2158 XATTR_NAME_ACL_DEFAULT);
2159 mdd_read_unlock(env, pobj);
2161 /* If there are default ACL, fix mode/ACL by default ACL */
2162 def_acl_buf->lb_len = rc;
2163 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2164 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2165 acl_buf->lb_len = rc;
2166 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2169 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2170 /* If there are no default ACL, fix mode by mask */
2171 struct lu_ucred *uc = lu_ucred(env);
2173 /* The create triggered by MDT internal events, such as
2174 * LFSCK reset, will not contain valid "uc". */
2175 if (unlikely(uc != NULL))
2176 la->la_mode &= ~uc->uc_umask;
2178 acl_buf->lb_len = 0;
2179 def_acl_buf->lb_len = 0;
2186 * Create a metadata object and initialize it, set acl, xattr.
2188 static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
2189 struct mdd_object *son, struct lu_attr *attr,
2190 struct md_op_spec *spec, struct lu_buf *acl_buf,
2191 struct lu_buf *def_acl_buf,
2192 struct dt_allocation_hint *hint,
2193 struct thandle *handle)
2195 const struct lu_buf *buf;
2198 mdd_write_lock(env, son, MOR_TGT_CHILD);
2199 rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
2204 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2205 * created in declare phase, they also needs to be added to master
2206 * object as sub-directory entry. So it has to initialize the master
2207 * object, then set dir striped EA.(in mdo_xattr_set) */
2208 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2211 GOTO(err_destroy, rc);
2214 * in case of replay we just set LOVEA provided by the client
2215 * XXX: I think it would be interesting to try "old" way where
2216 * MDT calls this xattr_set(LOV) in a different transaction.
2217 * probably this way we code can be made better.
2220 /* During creation, there are only a few cases we need do xattr_set to
2222 * 1. regular file: see comments above.
2223 * 2. dir: inherit default striping or pool settings from parent.
2224 * 3. create striped directory with provided stripeEA.
2225 * 4. create striped directory because inherit default layout from the
2228 if (spec->no_create ||
2229 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2230 S_ISDIR(attr->la_mode)) {
2231 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2232 spec->u.sp_ea.eadatalen);
2233 rc = mdo_xattr_set(env, son, buf,
2234 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2238 GOTO(err_destroy, rc);
2241 #ifdef CONFIG_FS_POSIX_ACL
2242 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2243 S_ISDIR(attr->la_mode)) {
2244 /* set default acl */
2245 rc = mdo_xattr_set(env, son, def_acl_buf,
2246 XATTR_NAME_ACL_DEFAULT, 0,
2249 GOTO(err_destroy, rc);
2251 /* set its own acl */
2252 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2253 rc = mdo_xattr_set(env, son, acl_buf,
2254 XATTR_NAME_ACL_ACCESS,
2257 GOTO(err_destroy, rc);
2261 if (S_ISLNK(attr->la_mode)) {
2262 struct lu_ucred *uc = lu_ucred_assert(env);
2263 struct dt_object *dt = mdd_object_child(son);
2264 const char *target_name = spec->u.sp_symname;
2265 int sym_len = strlen(target_name);
2266 const struct lu_buf *buf;
2269 buf = mdd_buf_get_const(env, target_name, sym_len);
2270 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2272 CFS_CAP_SYS_RESOURCE_MASK);
2277 GOTO(err_initlized, rc = -EFAULT);
2280 if (spec->sp_cr_file_secctx_name != NULL) {
2281 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2282 spec->sp_cr_file_secctx_size);
2283 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2286 GOTO(err_initlized, rc);
2290 if (unlikely(rc != 0)) {
2292 if (S_ISDIR(attr->la_mode)) {
2293 /* Drop the reference, no need to delete "."/"..",
2294 * because the object to be destroied directly. */
2295 rc2 = mdo_ref_del(env, son, handle);
2299 rc2 = mdo_ref_del(env, son, handle);
2303 mdo_destroy(env, son, handle);
2306 mdd_write_unlock(env, son);
2310 static int mdd_index_delete(const struct lu_env *env,
2311 struct mdd_object *mdd_pobj,
2312 struct lu_attr *cattr,
2313 const struct lu_name *lname)
2315 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2316 struct thandle *handle;
2320 handle = mdd_trans_create(env, mdd);
2322 RETURN(PTR_ERR(handle));
2324 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2329 if (S_ISDIR(cattr->la_mode)) {
2330 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2335 /* Since this will only be used in the error handler path,
2336 * Let's set the thandle to be local and not mess the transno */
2337 handle->th_local = 1;
2338 rc = mdd_trans_start(env, mdd, handle);
2342 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2343 S_ISDIR(cattr->la_mode), handle);
2347 rc = mdd_trans_stop(env, mdd, rc, handle);
2353 * Create object and insert it into namespace.
2355 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2356 const struct lu_name *lname, struct md_object *child,
2357 struct md_op_spec *spec, struct md_attr* ma)
2359 struct mdd_thread_info *info = mdd_env_info(env);
2360 struct lu_attr *la = &info->mti_la_for_fix;
2361 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2362 struct mdd_object *son = md2mdd_obj(child);
2363 struct mdd_device *mdd = mdo2mdd(pobj);
2364 struct lu_attr *attr = &ma->ma_attr;
2365 struct thandle *handle;
2366 struct lu_attr *pattr = &info->mti_pattr;
2367 struct lu_buf acl_buf;
2368 struct lu_buf def_acl_buf;
2369 struct linkea_data *ldata = &info->mti_link_data;
2370 const char *name = lname->ln_name;
2371 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2377 * Two operations have to be performed:
2379 * - an allocation of a new object (->do_create()), and
2381 * - an insertion into a parent index (->dio_insert()).
2383 * Due to locking, operation order is not important, when both are
2384 * successful, *but* error handling cases are quite different:
2386 * - if insertion is done first, and following object creation fails,
2387 * insertion has to be rolled back, but this operation might fail
2388 * also leaving us with dangling index entry.
2390 * - if creation is done first, is has to be undone if insertion
2391 * fails, leaving us with leaked space, which is neither good, nor
2394 * It seems that creation-first is simplest solution, but it is
2395 * sub-optimal in the frequent
2400 * case, because second mkdir is bound to create object, only to
2401 * destroy it immediately.
2403 * To avoid this follow local file systems that do double lookup:
2405 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2407 * 1. create (mdd_object_create_internal())
2409 * 2. insert (__mdd_index_insert(), lookup again)
2412 rc = mdd_la_get(env, mdd_pobj, pattr);
2416 /* Sanity checks before big job. */
2417 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2421 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2422 GOTO(out_free, rc = -EINPROGRESS);
2424 handle = mdd_trans_create(env, mdd);
2426 GOTO(out_free, rc = PTR_ERR(handle));
2428 acl_buf.lb_buf = info->mti_xattr_buf;
2429 acl_buf.lb_len = sizeof(info->mti_xattr_buf);
2430 def_acl_buf.lb_buf = info->mti_key;
2431 def_acl_buf.lb_len = sizeof(info->mti_key);
2432 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2436 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2438 memset(ldata, 0, sizeof(*ldata));
2439 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2440 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2443 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2444 &tfid, lname, 1, 0, ldata);
2446 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2447 mdd_object_fid(mdd_pobj),
2448 lname, 1, 0, ldata);
2451 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2452 handle, spec, ldata, &def_acl_buf, &acl_buf,
2457 rc = mdd_trans_start(env, mdd, handle);
2461 rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
2462 &def_acl_buf, hint, handle);
2466 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2467 mdd_write_lock(env, son, MOR_TGT_CHILD);
2468 rc = __mdd_orphan_add(env, son, handle);
2469 GOTO(out_volatile, rc);
2471 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2472 attr->la_mode, name, handle);
2474 GOTO(err_created, rc);
2476 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2479 /* update parent directory mtime/ctime */
2481 la->la_valid = LA_CTIME | LA_MTIME;
2482 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2484 GOTO(err_insert, rc);
2490 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2491 rc2 = __mdd_orphan_del(env, son, handle);
2493 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2494 S_ISDIR(attr->la_mode),
2500 mdd_write_lock(env, son, MOR_TGT_CHILD);
2501 if (S_ISDIR(attr->la_mode)) {
2502 /* Drop the reference, no need to delete "."/"..",
2503 * because the object is to be destroyed directly. */
2504 rc2 = mdo_ref_del(env, son, handle);
2506 mdd_write_unlock(env, son);
2511 /* For volatile files drop one link immediately, since there is
2512 * no filename in the namespace, and save any error returned. */
2513 rc2 = mdo_ref_del(env, son, handle);
2515 mdd_write_unlock(env, son);
2516 if (unlikely(rc == 0))
2521 /* Don't destroy the volatile object on success */
2522 if (likely(rc != 0))
2523 mdo_destroy(env, son, handle);
2524 mdd_write_unlock(env, son);
2527 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2528 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2529 rc = mdd_changelog_ns_store(env, mdd,
2530 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2531 S_ISREG(attr->la_mode) ? CL_CREATE :
2532 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2533 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2536 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2538 /* If creation fails, it is most likely due to the remote update
2539 * failure, because local transaction will mostly succeed at
2540 * this stage. There is no easy way to rollback all of previous
2541 * updates, so let's remove the object from namespace, and
2542 * LFSCK should handle the orphan object. */
2543 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2544 mdd_index_delete(env, mdd_pobj, attr, lname);
2548 if (is_vmalloc_addr(ldata->ld_buf))
2549 /* if we vmalloced a large buffer drop it */
2550 lu_buf_free(ldata->ld_buf);
2552 /* The child object shouldn't be cached anymore */
2554 set_bit(LU_OBJECT_HEARD_BANSHEE,
2555 &child->mo_lu.lo_header->loh_flags);
2560 * Get locks on parents in proper order
2561 * RETURN: < 0 - error, rename_order if successful
2569 static int mdd_rename_order(const struct lu_env *env,
2570 struct mdd_device *mdd,
2571 struct mdd_object *src_pobj,
2572 const struct lu_attr *pattr,
2573 struct mdd_object *tgt_pobj)
2575 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2579 if (src_pobj == tgt_pobj)
2580 RETURN(MDD_RN_SAME);
2582 /* compared the parent child relationship of src_p&tgt_p */
2583 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2585 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2588 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2602 /* has not mdd_write{read}_lock on any obj yet. */
2603 static int mdd_rename_sanity_check(const struct lu_env *env,
2604 struct mdd_object *src_pobj,
2605 const struct lu_attr *pattr,
2606 struct mdd_object *tgt_pobj,
2607 const struct lu_attr *tpattr,
2608 struct mdd_object *sobj,
2609 const struct lu_attr *cattr,
2610 struct mdd_object *tobj,
2611 const struct lu_attr *tattr)
2616 /* XXX: when get here, sobj must NOT be NULL,
2617 * the other case has been processed in cld_rename
2618 * before mdd_rename and enable MDS_PERM_BYPASS. */
2621 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2625 /* XXX: when get here, "tobj == NULL" means tobj must
2626 * NOT exist (neither on remote MDS, such case has been
2627 * processed in cld_rename before mdd_rename and enable
2629 * So check may_create, but not check may_unlink. */
2631 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2632 (src_pobj != tgt_pobj));
2634 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2635 (src_pobj != tgt_pobj), 1);
2637 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2638 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2643 static int mdd_declare_rename(const struct lu_env *env,
2644 struct mdd_device *mdd,
2645 struct mdd_object *mdd_spobj,
2646 struct mdd_object *mdd_tpobj,
2647 struct mdd_object *mdd_sobj,
2648 struct mdd_object *mdd_tobj,
2649 const struct lu_name *sname,
2650 const struct lu_name *tname,
2652 struct linkea_data *ldata,
2653 struct thandle *handle)
2655 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2658 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2659 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2665 /* name from source dir */
2666 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2670 /* .. from source child */
2671 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2672 /* source child can be directory,
2673 * counted by source dir's nlink */
2674 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2677 if (mdd_spobj != mdd_tpobj) {
2678 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2683 rc = mdo_declare_index_insert(env, mdd_sobj,
2685 S_IFDIR, dotdot, handle);
2690 /* new target child can be directory,
2691 * counted by target dir's nlink */
2692 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2697 la->la_valid = LA_CTIME | LA_MTIME;
2698 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2702 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2706 la->la_valid = LA_CTIME;
2707 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2711 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
2712 S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
2717 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2718 mdd_object_type(mdd_sobj),
2719 tname->ln_name, handle);
2723 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2724 /* delete target child in target parent directory */
2725 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2730 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2734 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2735 /* target child can be directory,
2736 * delete "." reference in target child directory */
2737 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2741 /* delete ".." reference in target parent directory */
2742 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2747 la->la_valid = LA_CTIME;
2748 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2752 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2757 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2764 /* src object can be remote that is why we use only fid and type of object */
2765 static int mdd_rename(const struct lu_env *env,
2766 struct md_object *src_pobj, struct md_object *tgt_pobj,
2767 const struct lu_fid *lf, const struct lu_name *lsname,
2768 struct md_object *tobj, const struct lu_name *ltname,
2771 const char *sname = lsname->ln_name;
2772 const char *tname = ltname->ln_name;
2773 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2774 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2775 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2776 struct mdd_device *mdd = mdo2mdd(src_pobj);
2777 struct mdd_object *mdd_sobj = NULL; /* source object */
2778 struct mdd_object *mdd_tobj = NULL;
2779 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2780 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2781 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2782 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2783 struct thandle *handle;
2784 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2785 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2786 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2789 bool tobj_locked = 0;
2790 unsigned cl_flags = 0;
2795 mdd_tobj = md2mdd_obj(tobj);
2797 mdd_sobj = mdd_object_find(env, mdd, lf);
2798 if (IS_ERR(mdd_sobj))
2799 RETURN(PTR_ERR(mdd_sobj));
2801 rc = mdd_la_get(env, mdd_sobj, cattr);
2803 GOTO(out_pending, rc);
2805 rc = mdd_la_get(env, mdd_spobj, pattr);
2807 GOTO(out_pending, rc);
2810 rc = mdd_la_get(env, mdd_tobj, tattr);
2812 GOTO(out_pending, rc);
2813 /* search for an existing archive.
2814 * we should check ahead as the object
2815 * can be destroyed in this transaction */
2816 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2817 cl_flags |= CLF_RENAME_LAST_EXISTS;
2820 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2822 GOTO(out_pending, rc);
2824 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2825 mdd_sobj, cattr, mdd_tobj, tattr);
2827 GOTO(out_pending, rc);
2829 rc = mdd_name_check(mdd, ltname);
2831 GOTO(out_pending, rc);
2833 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2834 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2836 GOTO(out_pending, rc);
2838 handle = mdd_trans_create(env, mdd);
2840 GOTO(out_pending, rc = PTR_ERR(handle));
2842 memset(ldata, 0, sizeof(*ldata));
2843 mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname,
2844 mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata);
2845 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2846 mdd_tobj, lsname, ltname, ma, ldata, handle);
2850 rc = mdd_trans_start(env, mdd, handle);
2854 is_dir = S_ISDIR(cattr->la_mode);
2856 /* Remove source name from source directory */
2857 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2861 /* "mv dir1 dir2" needs "dir1/.." link update */
2862 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2863 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2865 GOTO(fixup_spobj2, rc);
2867 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2870 GOTO(fixup_spobj, rc);
2873 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2874 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2876 /* tname might been renamed to something else */
2877 GOTO(fixup_spobj, rc);
2880 /* Insert new fid with target name into target dir */
2881 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2884 GOTO(fixup_tpobj, rc);
2886 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2887 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2889 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2890 la->la_valid = LA_CTIME;
2891 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2893 GOTO(fixup_tpobj, rc);
2895 /* Update the linkEA for the source object */
2896 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2897 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2898 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2901 /* Old files might not have EA entry */
2902 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2903 lsname, handle, NULL, 0);
2904 mdd_write_unlock(env, mdd_sobj);
2905 /* We don't fail the transaction if the link ea can't be
2906 updated -- fid2path will use alternate lookup method. */
2909 /* Remove old target object
2910 * For tobj is remote case cmm layer has processed
2911 * and set tobj to NULL then. So when tobj is NOT NULL,
2912 * it must be local one.
2914 if (tobj && mdd_object_exists(mdd_tobj)) {
2915 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2917 if (mdd_is_dead_obj(mdd_tobj)) {
2918 /* shld not be dead, something is wrong */
2919 CERROR("tobj is dead, something is wrong\n");
2923 mdo_ref_del(env, mdd_tobj, handle);
2925 /* Remove dot reference. */
2926 if (S_ISDIR(tattr->la_mode))
2927 mdo_ref_del(env, mdd_tobj, handle);
2930 /* fetch updated nlink */
2931 rc = mdd_la_get(env, mdd_tobj, tattr);
2933 CERROR("%s: Failed to get nlink for tobj "
2935 mdd2obd_dev(mdd)->obd_name,
2936 PFID(tpobj_fid), rc);
2937 GOTO(fixup_tpobj, rc);
2940 la->la_valid = LA_CTIME;
2941 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2943 CERROR("%s: Failed to set ctime for tobj "
2945 mdd2obd_dev(mdd)->obd_name,
2946 PFID(tpobj_fid), rc);
2947 GOTO(fixup_tpobj, rc);
2950 /* XXX: this transfer to ma will be removed with LOD/OSP */
2951 ma->ma_attr = *tattr;
2952 ma->ma_valid |= MA_INODE;
2953 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2956 CERROR("%s: Failed to unlink tobj "
2958 mdd2obd_dev(mdd)->obd_name,
2959 PFID(tpobj_fid), rc);
2960 GOTO(fixup_tpobj, rc);
2963 /* fetch updated nlink */
2964 rc = mdd_la_get(env, mdd_tobj, tattr);
2965 if (rc == -ENOENT) {
2966 /* the object got removed, let's
2967 * return the latest known attributes */
2968 tattr->la_nlink = 0;
2970 } else if (rc != 0) {
2971 CERROR("%s: Failed to get nlink for tobj "
2973 mdd2obd_dev(mdd)->obd_name,
2974 PFID(tpobj_fid), rc);
2975 GOTO(fixup_tpobj, rc);
2977 /* XXX: this transfer to ma will be removed with LOD/OSP */
2978 ma->ma_attr = *tattr;
2979 ma->ma_valid |= MA_INODE;
2981 if (tattr->la_nlink == 0)
2982 cl_flags |= CLF_RENAME_LAST;
2984 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
2987 la->la_valid = LA_CTIME | LA_MTIME;
2988 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2990 GOTO(fixup_tpobj, rc);
2992 if (mdd_spobj != mdd_tpobj) {
2993 la->la_valid = LA_CTIME | LA_MTIME;
2994 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2996 GOTO(fixup_tpobj, rc);
3003 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3005 CWARN("tp obj fix error %d\n",rc2);
3007 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3008 !mdd_is_dead_obj(mdd_tobj)) {
3010 mdo_ref_add(env, mdd_tobj, handle);
3012 mdo_ref_add(env, mdd_tobj, handle);
3015 rc2 = __mdd_index_insert(env, mdd_tpobj,
3017 mdd_object_type(mdd_tobj),
3020 CWARN("tp obj fix error: rc = %d\n", rc2);
3025 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3026 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3028 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3029 mdd2obd_dev(mdd)->obd_name, rc2);
3032 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3035 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3036 mdd2obd_dev(mdd)->obd_name, rc2);
3041 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3042 mdd_object_type(mdd_sobj), sname,
3045 CWARN("sp obj fix error: rc = %d\n", rc2);
3050 mdd_write_unlock(env, mdd_tobj);
3053 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3054 mdd_tobj, tpobj_fid, lf, spobj_fid,
3055 ltname, lsname, handle);
3058 rc = mdd_trans_stop(env, mdd, rc, handle);
3061 mdd_object_put(env, mdd_sobj);
3066 * During migration once the parent FID has been changed,
3067 * we need update the parent FID in linkea.
3069 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3070 struct mdd_object *parent,
3071 struct mdd_object *newparent,
3072 struct mdd_object *child,
3073 const char *name, int namelen,
3074 struct thandle *handle,
3077 struct mdd_thread_info *info = mdd_env_info(env);
3078 struct linkea_data ldata = { NULL };
3079 struct lu_buf *buf = &info->mti_link_buf;
3085 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3086 if (buf->lb_buf == NULL)
3090 rc = mdd_links_read(env, child, &ldata);
3092 if (rc == -ENOENT || rc == -ENODATA)
3097 LASSERT(ldata.ld_leh != NULL);
3098 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3099 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3100 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3101 struct lu_name lname;
3104 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3107 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3108 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3109 ldata.ld_lee = (struct link_ea_entry *)
3110 ((char *)ldata.ld_lee +
3115 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3116 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3117 lname.ln_namelen, lname.ln_name,
3118 PFID(mdd_object_fid(newparent)));
3119 /* update to the new parent fid */
3120 linkea_entry_pack(ldata.ld_lee, &lname,
3121 mdd_object_fid(newparent));
3123 rc = mdd_declare_links_add(env, child, handle, &ldata,
3126 rc = mdd_links_write(env, child, &ldata, handle);
3132 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3133 struct mdd_object *parent,
3134 struct mdd_object *newparent,
3135 struct mdd_object *child,
3136 const char *name, int namelen,
3137 struct thandle *handle)
3139 return mdd_linkea_update_child_internal(env, parent, newparent,
3141 namelen, handle, true);
3144 static int mdd_linkea_update_child(const struct lu_env *env,
3145 struct mdd_object *parent,
3146 struct mdd_object *newparent,
3147 struct mdd_object *child,
3148 const char *name, int namelen,
3149 struct thandle *handle)
3151 return mdd_linkea_update_child_internal(env, parent, newparent,
3153 namelen, handle, false);
3156 static int mdd_update_linkea_internal(const struct lu_env *env,
3157 struct mdd_object *mdd_pobj,
3158 struct mdd_object *mdd_sobj,
3159 struct mdd_object *mdd_tobj,
3160 const struct lu_name *child_name,
3161 struct linkea_data *ldata,
3162 struct thandle *handle,
3165 struct mdd_thread_info *info = mdd_env_info(env);
3170 LASSERT(ldata->ld_buf != NULL);
3173 /* If it is mulitple links file, we need update the name entry for
3175 LASSERT(ldata->ld_leh != NULL);
3176 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3177 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3178 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3179 struct mdd_object *pobj;
3180 struct lu_name lname;
3183 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3185 pobj = mdd_object_find(env, mdd, &fid);
3187 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3188 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3190 linkea_del_buf(ldata, &lname);
3194 if (!mdd_object_exists(pobj)) {
3195 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3196 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3197 linkea_del_buf(ldata, &lname);
3198 mdd_object_put(env, pobj);
3202 if (pobj == mdd_pobj &&
3203 lname.ln_namelen == child_name->ln_namelen &&
3204 strncmp(lname.ln_name, child_name->ln_name,
3205 lname.ln_namelen) == 0) {
3206 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3207 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3209 linkea_del_buf(ldata, &lname);
3210 mdd_object_put(env, pobj);
3214 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3215 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3216 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3219 /* Remove source name from source directory */
3220 /* Insert new fid with target name into target dir */
3221 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3226 rc = mdo_declare_index_insert(env, pobj,
3227 mdd_object_fid(mdd_tobj),
3228 mdd_object_type(mdd_tobj),
3229 lname.ln_name, handle);
3233 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3237 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3241 char *tmp_name = info->mti_key;
3243 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3244 /* lnamelen is too big(> NAME_MAX + 16),
3245 * something wrong about this linkea, let's
3247 linkea_del_buf(ldata, &lname);
3248 mdd_object_put(env, pobj);
3252 /* Note: lname might be without \0 at the end, see
3253 * linkea_entry_unpack(), let's add extra \0 by
3255 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3256 lname.ln_namelen, lname.ln_name);
3257 lname.ln_name = tmp_name;
3259 /* Let's check if this linkEA still valid, before
3260 * it might be packed into the RPC buffer. */
3261 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3262 &info->mti_fid, NULL);
3264 !lu_fid_eq(&info->mti_fid,
3265 mdd_object_fid(mdd_sobj))) {
3266 /* skip invalid linkea entry */
3267 linkea_del_buf(ldata, &lname);
3268 mdd_object_put(env, pobj);
3272 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3276 rc = __mdd_index_insert(env, pobj,
3277 mdd_object_fid(mdd_tobj),
3278 mdd_object_type(mdd_tobj),
3283 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3284 rc = mdo_ref_add(env, mdd_tobj, handle);
3285 mdd_write_unlock(env, mdd_tobj);
3289 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3290 mdo_ref_del(env, mdd_sobj, handle);
3291 mdd_write_unlock(env, mdd_sobj);
3294 mdd_object_put(env, pobj);
3298 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3305 static int mdd_migrate_xattrs(const struct lu_env *env,
3306 struct mdd_object *mdd_sobj,
3307 struct mdd_object *mdd_tobj)
3309 struct mdd_thread_info *info = mdd_env_info(env);
3310 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3312 struct thandle *handle;
3318 struct lu_buf list_xbuf;
3321 /* retrieve xattr list from the old object */
3322 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3323 if (list_xsize == -ENODATA)
3329 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3330 if (info->mti_big_buf.lb_buf == NULL)
3333 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3334 list_xbuf.lb_len = list_xsize;
3335 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3340 xname = list_xbuf.lb_buf;
3342 xlen = strnlen(xname, rem - 1) + 1;
3343 if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
3344 strcmp(XATTR_NAME_LMA, xname) == 0 ||
3345 strcmp(XATTR_NAME_LMV, xname) == 0)
3348 /* For directory, if there are default layout, migrate here */
3349 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3350 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3353 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3354 if (xsize == -ENODATA)
3359 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3360 if (info->mti_link_buf.lb_buf == NULL)
3361 GOTO(out, rc = -ENOMEM);
3363 xbuf.lb_len = xsize;
3364 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3365 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3371 handle = mdd_trans_create(env, mdd);
3373 GOTO(out, rc = PTR_ERR(handle));
3375 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3378 GOTO(stop_trans, rc);
3379 /* Note: this transaction is part of migration, and it is not
3380 * the last step of migration, so we set th_local = 1 to avoid
3381 * update last rcvd for this transaction */
3382 handle->th_local = 1;
3383 rc = mdd_trans_start(env, mdd, handle);
3385 GOTO(stop_trans, rc);
3387 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3389 GOTO(stop_trans, rc = 0);
3392 GOTO(stop_trans, rc);
3394 rc = mdd_trans_stop(env, mdd, rc, handle);
3399 memmove(xname, xname + xlen, rem);
3405 static int mdd_declare_migrate_create(const struct lu_env *env,
3406 struct mdd_object *mdd_pobj,
3407 struct mdd_object *mdd_sobj,
3408 struct mdd_object *mdd_tobj,
3409 struct md_op_spec *spec,
3411 union lmv_mds_md *mgr_ea,
3412 struct linkea_data *ldata,
3413 struct thandle *handle)
3415 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3416 const struct lu_buf *buf;
3420 rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
3421 handle, spec, NULL);
3425 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3430 if (S_ISLNK(la->la_mode)) {
3431 const char *target_name = spec->u.sp_symname;
3432 int sym_len = strlen(target_name);
3433 const struct lu_buf *buf;
3435 buf = mdd_buf_get_const(env, target_name, sym_len);
3436 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3440 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3441 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
3447 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3448 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3449 spec->u.sp_ea.eadatalen);
3450 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3456 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3457 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3458 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3463 la_flag->la_valid = LA_FLAGS;
3464 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3465 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3470 static int mdd_migrate_create(const struct lu_env *env,
3471 struct mdd_object *mdd_pobj,
3472 struct mdd_object *mdd_sobj,
3473 struct mdd_object *mdd_tobj,
3474 const struct lu_name *lname,
3477 struct mdd_thread_info *info = mdd_env_info(env);
3478 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3479 struct md_op_spec *spec = &info->mti_spec;
3480 struct lu_buf lmm_buf = { NULL };
3481 struct lu_buf link_buf = { NULL };
3482 const struct lu_buf *buf;
3483 struct thandle *handle;
3484 struct lmv_mds_md_v1 *mgr_ea;
3485 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3486 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3488 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3492 /* prepare spec for create */
3493 memset(spec, 0, sizeof(*spec));
3494 spec->sp_cr_lookup = 0;
3495 spec->sp_feat = &dt_directory_features;
3496 if (S_ISLNK(la->la_mode)) {
3497 buf = lu_buf_check_and_alloc(
3498 &mdd_env_info(env)->mti_big_buf,
3501 link_buf.lb_len = la->la_size + 1;
3502 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3503 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3505 rc = rc != 0 ? rc : -EFAULT;
3506 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3507 mdd2obd_dev(mdd)->obd_name,
3508 PFID(mdd_object_fid(mdd_sobj)), rc);
3511 spec->u.sp_symname = link_buf.lb_buf;
3512 } else if (S_ISREG(la->la_mode)) {
3513 /* retrieve lov of the old object */
3514 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3515 if (rc != 0 && rc != -ENODATA)
3517 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3518 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3519 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3520 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3522 } else if (S_ISDIR(la->la_mode)) {
3523 rc = mdd_links_read(env, mdd_sobj, ldata);
3524 if (rc == -ENODATA) {
3525 /* ignore the non-linkEA error */
3533 mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
3534 memset(mgr_ea, 0, sizeof(*mgr_ea));
3535 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3536 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3537 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3538 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3539 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3540 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3542 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3544 handle = mdd_trans_create(env, mdd);
3546 GOTO(out_free, rc = PTR_ERR(handle));
3548 /* Note: this transaction is part of migration, and it is not
3549 * the last step of migration, so we set th_local = 1 to avoid
3550 * update last rcvd for this transaction */
3551 handle->th_local = 1;
3552 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
3554 (union lmv_mds_md *)info->mti_xattr_buf,
3557 GOTO(stop_trans, rc);
3559 rc = mdd_trans_start(env, mdd, handle);
3561 GOTO(stop_trans, rc);
3563 /* don't set nlink from the original object */
3564 la->la_valid &= ~LA_NLINK;
3566 /* create the target object */
3567 rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3570 GOTO(stop_trans, rc);
3572 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3573 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3575 GOTO(stop_trans, rc);
3578 /* Set MIGRATE EA on the source inode, so once the migration needs
3579 * to be re-done during failover, the re-do process can locate the
3580 * target object which is already being created. */
3581 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3582 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3583 rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
3585 GOTO(stop_trans, rc);
3587 /* Set immutable flag, so any modification is disabled until
3588 * the migration is done. Once the migration is interrupted,
3589 * if the resume process find the migrating object has both
3590 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3591 * flag and approve the migration */
3592 la_flag->la_valid = LA_FLAGS;
3593 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3594 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3597 rc = mdd_trans_stop(env, mdd, rc, handle);
3599 if (lmm_buf.lb_buf != NULL)
3600 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3604 static int mdd_migrate_entries(const struct lu_env *env,
3605 struct mdd_object *mdd_sobj,
3606 struct mdd_object *mdd_tobj)
3608 struct dt_object *next = mdd_object_child(mdd_sobj);
3609 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3610 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3611 struct thandle *handle;
3613 const struct dt_it_ops *iops;
3615 struct lu_dirent *ent;
3619 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3623 if (!dt_try_as_dir(env, next))
3624 GOTO(out_ent, rc = -ENOTDIR);
3626 * iterate directories
3628 iops = &next->do_index_ops->dio_it;
3629 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3631 GOTO(out_ent, rc = PTR_ERR(it));
3633 rc = iops->load(env, it, 0);
3635 rc = iops->next(env, it);
3639 * At this point and across for-loop:
3641 * rc == 0 -> ok, proceed.
3642 * rc > 0 -> end of directory.
3646 struct mdd_object *child;
3647 char *name = mdd_env_info(env)->mti_key;
3650 bool target_exist = false;
3652 len = iops->key_size(env, it);
3656 result = iops->rec(env, it, (struct dt_rec *)ent,
3657 LUDA_FID | LUDA_TYPE);
3658 if (result == -ESTALE)
3665 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3667 /* Insert new fid with target name into target dir */
3668 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3669 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3670 ent->lde_name[1] == '.'))
3673 child = mdd_object_find(env, mdd, &ent->lde_fid);
3675 GOTO(out, rc = PTR_ERR(child));
3677 mdd_write_lock(env, child, MOR_SRC_CHILD);
3678 is_dir = S_ISDIR(mdd_object_type(child));
3680 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3682 /* Check whether the name has been inserted to the target */
3683 if (dt_try_as_dir(env, dt_tobj)) {
3684 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3686 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3687 (struct dt_key *)name);
3688 if (unlikely(rc == 0))
3689 target_exist = true;
3692 handle = mdd_trans_create(env, mdd);
3694 GOTO(out_put, rc = PTR_ERR(handle));
3696 /* Note: this transaction is part of migration, and it is not
3697 * the last step of migration, so we set th_local = 1 to avoid
3698 * updating last rcvd for this transaction */
3699 handle->th_local = 1;
3700 if (likely(!target_exist)) {
3701 rc = mdo_declare_index_insert(env, mdd_tobj,
3703 mdd_object_type(child),
3709 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3715 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3720 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3724 /* Update .. for child */
3725 rc = mdo_declare_index_delete(env, child, dotdot,
3730 rc = mdo_declare_index_insert(env, child,
3731 mdd_object_fid(mdd_tobj),
3732 S_IFDIR, dotdot, handle);
3737 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3744 rc = mdd_trans_start(env, mdd, handle);
3746 CERROR("%s: transaction start failed: rc = %d\n",
3747 mdd2obd_dev(mdd)->obd_name, rc);
3751 if (likely(!target_exist)) {
3752 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3753 mdd_object_type(child),
3759 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3764 rc = __mdd_index_delete_only(env, child, dotdot,
3769 rc = __mdd_index_insert_only(env, child,
3770 mdd_object_fid(mdd_tobj), S_IFDIR,
3776 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3778 strlen(name), handle);
3781 mdd_write_unlock(env, child);
3782 mdd_object_put(env, child);
3783 rc = mdd_trans_stop(env, mdd, rc, handle);
3787 result = iops->next(env, it);
3788 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3789 GOTO(out, rc = -EINTR);
3791 if (result == -ESTALE)
3793 } while (result == 0);
3796 iops->fini(env, it);
3798 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3802 static int mdd_declare_update_linkea(const struct lu_env *env,
3803 struct mdd_object *mdd_pobj,
3804 struct mdd_object *mdd_sobj,
3805 struct mdd_object *mdd_tobj,
3806 const struct lu_name *child_name,
3807 struct linkea_data *ldata,
3808 struct thandle *handle)
3810 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3811 child_name, ldata, handle, 1);
3814 static int mdd_update_linkea(const struct lu_env *env,
3815 struct mdd_object *mdd_pobj,
3816 struct mdd_object *mdd_sobj,
3817 struct mdd_object *mdd_tobj,
3818 const struct lu_name *child_name,
3819 struct linkea_data *ldata,
3820 struct thandle *handle)
3822 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3823 child_name, ldata, handle, 0);
3826 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3827 struct mdd_object *mdd_pobj,
3828 struct mdd_object *mdd_sobj,
3829 struct mdd_object *mdd_tobj,
3830 const struct lu_name *lname,
3832 struct lu_attr *parent_la,
3833 struct linkea_data *ldata,
3834 struct thandle *handle)
3836 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3837 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3840 /* Revert IMMUTABLE flag */
3841 la_flag->la_valid = LA_FLAGS;
3842 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3843 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3847 /* delete entry from source dir */
3848 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3852 if (ldata->ld_buf != NULL) {
3853 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3854 mdd_tobj, lname, ldata, handle);
3859 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3860 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3865 handle->th_complex = 1;
3866 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3868 LU_XATTR_REPLACE, handle);
3873 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3874 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3880 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3881 mdd_object_type(mdd_tobj),
3882 lname->ln_name, handle);
3886 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE);
3890 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3891 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3896 /* delete old object */
3897 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3901 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3902 /* delete old object */
3903 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3906 /* set nlink to 0 */
3907 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3912 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3916 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3920 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3925 static int mdd_migrate_update_name(const struct lu_env *env,
3926 struct mdd_object *mdd_pobj,
3927 struct mdd_object *mdd_sobj,
3928 struct mdd_object *mdd_tobj,
3929 const struct lu_name *lname,
3932 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3933 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3934 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3935 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3936 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3937 struct thandle *handle;
3938 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3939 const char *name = lname->ln_name;
3943 /* update time for parent */
3944 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3945 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3946 p_la->la_valid = LA_CTIME;
3948 rc = mdd_la_get(env, mdd_sobj, so_attr);
3952 ldata->ld_buf = NULL;
3953 rc = mdd_links_read(env, mdd_sobj, ldata);
3954 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
3957 handle = mdd_trans_create(env, mdd);
3959 RETURN(PTR_ERR(handle));
3961 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3962 lname, so_attr, p_la, ldata,
3965 /* If the migration can not be fit in one transaction, just
3966 * leave it in the original MDT */
3968 GOTO(stop_trans, rc = 0);
3970 GOTO(stop_trans, rc);
3973 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3974 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3975 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3976 PFID(mdd_object_fid(mdd_tobj)));
3978 rc = mdd_trans_start(env, mdd, handle);
3980 GOTO(stop_trans, rc);
3982 /* Revert IMMUTABLE flag */
3983 la_flag->la_valid = LA_FLAGS;
3984 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3985 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3987 GOTO(stop_trans, rc);
3989 /* Remove source name from source directory */
3990 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
3992 GOTO(stop_trans, rc);
3994 if (ldata->ld_buf != NULL) {
3995 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3996 lname, ldata, handle);
3998 GOTO(stop_trans, rc);
4000 /* linkea update might decrease the source object
4001 * nlink, let's get the attr again after ref_del */
4002 rc = mdd_la_get(env, mdd_sobj, so_attr);
4004 GOTO(stop_trans, rc);
4007 if (S_ISREG(so_attr->la_mode)) {
4008 if (so_attr->la_nlink == 1) {
4009 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4011 if (rc != 0 && rc != -ENODATA)
4012 GOTO(stop_trans, rc);
4014 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4016 LU_XATTR_REPLACE, handle);
4018 GOTO(stop_trans, rc);
4022 /* Insert new fid with target name into target dir */
4023 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4024 mdd_object_type(mdd_tobj), name, handle);
4026 GOTO(stop_trans, rc);
4028 linkea_add_buf(ldata, lname, mdd_object_fid(mdd_pobj));
4029 rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
4032 GOTO(stop_trans, rc);
4034 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4036 mdd_sobj->mod_flags |= DEAD_OBJ;
4037 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4039 GOTO(out_unlock, rc);
4041 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4043 GOTO(out_unlock, rc);
4045 mdo_ref_del(env, mdd_sobj, handle);
4047 mdo_ref_del(env, mdd_sobj, handle);
4049 /* Get the attr again after ref_del */
4050 rc = mdd_la_get(env, mdd_sobj, so_attr);
4052 GOTO(out_unlock, rc);
4054 ma->ma_attr = *so_attr;
4055 ma->ma_valid |= MA_INODE;
4057 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4059 GOTO(out_unlock, rc);
4061 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4062 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4063 mdo2fid(mdd_pobj), lname, lname, handle);
4065 CWARN("%s: changelog for migrate %s "DFID
4066 "under "DFID" failed: rc = %d\n",
4067 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4068 PFID(mdd_object_fid(mdd_sobj)),
4069 PFID(mdd_object_fid(mdd_pobj)), rc);
4070 /* Sigh, there are no easy way to migrate back the object, so
4071 * let's reset the result to 0 for now XXX */
4075 mdd_write_unlock(env, mdd_sobj);
4078 rc = mdd_trans_stop(env, mdd, rc, handle);
4083 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4084 const struct lu_fid *fid, __u32 *mdt_index)
4086 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4087 struct seq_server_site *ss;
4090 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4092 range->lsr_flags = LU_SEQ_RANGE_MDT;
4093 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4097 *mdt_index = range->lsr_index;
4102 * Check whether we should migrate the file/dir
4104 * < 0 permission check failed or other error.
4105 * = 0 the file can be migrated.
4106 * > 0 the file does not need to be migrated, mostly
4107 * for multiple link file
4109 static int mdd_migrate_sanity_check(const struct lu_env *env,
4110 struct mdd_object *pobj,
4111 const struct lu_attr *pattr,
4112 struct mdd_object *sobj,
4113 struct lu_attr *sattr)
4115 struct mdd_thread_info *info = mdd_env_info(env);
4116 struct linkea_data *ldata = &info->mti_link_data;
4117 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4119 struct lu_buf *mgr_buf;
4125 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4126 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4127 if (mgr_buf->lb_buf == NULL)
4130 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4132 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4134 /* If the object has migrateEA, it means IMMUTE flag
4135 * is being set by previous migration process, so it
4136 * needs to override the IMMUTE flag, otherwise the
4137 * following sanity check will fail */
4138 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4139 LMV_HASH_FLAG_MIGRATION) {
4140 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4142 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4143 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4144 mdd2obd_dev(mdd)->obd_name,
4145 PFID(mdd_object_fid(sobj)));
4149 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4150 sobj, sattr, NULL, NULL);
4154 /* Then it will check if the file should be migrated. If the file
4155 * has mulitple links, we only need migrate the file if all of its
4156 * entries has been migrated to the remote MDT */
4157 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4160 rc = mdd_links_read(env, sobj, ldata);
4162 /* For multiple links files, if there are no linkEA data at all,
4163 * means the file might be created before linkEA is enabled, and
4164 * all of its links should not be migrated yet, otherwise it
4165 * should have some linkEA there */
4166 if (rc == -ENOENT || rc == -ENODATA)
4171 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4172 /* If there are still links locally, then the file will not be
4174 LASSERT(ldata->ld_leh != NULL);
4175 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4176 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4177 struct lu_name lname;
4179 __u32 parent_mdt_index;
4181 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4183 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4186 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4190 /* Migrate the object only if none of its parents are on the
4192 if (parent_mdt_index != mdt_index)
4195 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4196 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4197 lname.ln_name, PFID(&fid));
4205 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4206 struct md_object *sobj, const struct lu_name *lname,
4207 struct md_object *tobj, struct md_attr *ma)
4209 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4210 struct mdd_device *mdd = mdo2mdd(pobj);
4211 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4212 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4213 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4214 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4215 bool created = false;
4219 /* If the file will being migrated, it will check whether
4220 * the file is being opened by someone else right now */
4221 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4222 if (mdd_sobj->mod_count > 0) {
4224 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4225 mdd2obd_dev(mdd)->obd_name,
4226 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4227 mdd_sobj->mod_count, -EBUSY);
4228 mdd_read_unlock(env, mdd_sobj);
4229 GOTO(put, rc = -EBUSY);
4231 mdd_read_unlock(env, mdd_sobj);
4233 rc = mdd_la_get(env, mdd_sobj, so_attr);
4237 rc = mdd_la_get(env, mdd_pobj, pattr);
4241 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4248 /* Sigh, it is impossible to finish all of migration in a single
4249 * transaction, for example migrating big directory entries to the
4250 * new MDT, it needs insert all of name entries of children in the
4253 * So migration will be done in multiple steps and transactions.
4255 * 1. create an orphan object on the remote MDT in one transaction.
4256 * 2. migrate extend attributes to the new target file/directory.
4257 * 3. For directory, migrate the entries to the new MDT and update
4258 * linkEA of each children. Because we can not migrate all entries
4259 * in a single transaction, so the migrating directory will become
4260 * a striped directory during migration, so once the process is
4261 * interrupted, the directory is still accessible. (During lookup,
4262 * client will locate the name by searching both original and target
4264 * 4. Finally, update the name/FID to point to the new file/directory
4265 * in a separate transaction.
4268 /* step 1: Check whether the orphan object has been created, and create
4269 * orphan object on the remote MDT if needed */
4270 if (!mdd_object_exists(mdd_tobj)) {
4271 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4278 LASSERT(mdd_object_exists(mdd_tobj));
4279 /* step 2: migrate xattr */
4280 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4284 /* step 3: migrate name entries to the orphan object */
4285 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4286 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4289 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4290 OBD_FAIL_MDS_REINT_NET_REP)))
4293 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4296 LASSERT(mdd_object_exists(mdd_tobj));
4297 /* step 4: update name entry to the new object */
4298 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4303 /* newly created target was not locked, don't cache its attributes */
4305 mdd_invalidate(env, tobj);
4310 const struct md_dir_operations mdd_dir_ops = {
4311 .mdo_is_subdir = mdd_is_subdir,
4312 .mdo_lookup = mdd_lookup,
4313 .mdo_create = mdd_create,
4314 .mdo_rename = mdd_rename,
4315 .mdo_link = mdd_link,
4316 .mdo_unlink = mdd_unlink,
4317 .mdo_create_data = mdd_create_data,
4318 .mdo_migrate = mdd_migrate,