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 = {
52 .ln_name = (char *) dotdot,
53 .ln_namelen = sizeof(dotdot) - 1,
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);
123 /** Read the link EA into a temp buffer.
124 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
125 * A pointer to the buffer is stored in \a ldata::ld_buf.
129 static int __mdd_links_read(const struct lu_env *env,
130 struct mdd_object *mdd_obj,
131 struct linkea_data *ldata)
135 if (!mdd_object_exists(mdd_obj))
138 /* First try a small buf */
139 LASSERT(env != NULL);
140 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
142 if (ldata->ld_buf->lb_buf == NULL)
145 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
147 /* Buf was too small, figure out what we need. */
148 lu_buf_free(ldata->ld_buf);
149 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
153 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
154 if (ldata->ld_buf->lb_buf == NULL)
156 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
160 lu_buf_free(ldata->ld_buf);
161 ldata->ld_buf = NULL;
165 return linkea_init(ldata);
168 static int mdd_links_read(const struct lu_env *env,
169 struct mdd_object *mdd_obj,
170 struct linkea_data *ldata)
174 rc = __mdd_links_read(env, mdd_obj, ldata);
176 rc = linkea_init(ldata);
181 static int mdd_links_read_with_rec(const struct lu_env *env,
182 struct mdd_object *mdd_obj,
183 struct linkea_data *ldata)
187 rc = __mdd_links_read(env, mdd_obj, ldata);
189 rc = linkea_init_with_rec(ldata);
195 * Get parent FID of the directory
197 * Read parent FID from linkEA, if that fails, then do lookup
198 * dotdot to get the parent FID.
200 * \param[in] env execution environment
201 * \param[in] obj object from which to find the parent FID
202 * \param[in] attr attribute of the object
203 * \param[out] fid fid to get the parent FID
205 * \retval 0 if getting the parent FID succeeds.
206 * \retval negative errno if getting the parent FID fails.
208 static inline int mdd_parent_fid(const struct lu_env *env,
209 struct mdd_object *obj,
210 const struct lu_attr *attr,
213 struct mdd_thread_info *info = mdd_env_info(env);
214 struct linkea_data ldata = { NULL };
215 struct lu_buf *buf = &info->mti_link_buf;
216 struct lu_name lname;
221 LASSERT(S_ISDIR(mdd_object_type(obj)));
223 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
224 if (buf->lb_buf == NULL)
225 GOTO(lookup, rc = 0);
228 rc = mdd_links_read_with_rec(env, obj, &ldata);
232 LASSERT(ldata.ld_leh != NULL);
233 /* Directory should only have 1 parent */
234 if (ldata.ld_leh->leh_reccount > 1)
237 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
239 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
240 if (likely(fid_is_sane(fid)))
243 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
248 * For root fid use special function, which does not compare version component
249 * of fid. Version component is different for root fids on all MDTs.
251 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
253 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
254 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
258 * return 1: if lf is the fid of the ancestor of p1;
261 * return -EREMOTE: if remote object is found, in this
262 * case fid of remote object is saved to @pf;
264 * otherwise: values < 0, errors.
266 static int mdd_is_parent(const struct lu_env *env,
267 struct mdd_device *mdd,
268 struct mdd_object *p1,
269 const struct lu_attr *attr,
270 const struct lu_fid *lf,
273 struct mdd_object *parent = NULL;
278 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
279 pfid = &mdd_env_info(env)->mti_fid;
281 /* Check for root first. */
282 if (mdd_is_root(mdd, mdo2fid(p1)))
286 /* this is done recursively */
287 rc = mdd_parent_fid(env, p1, attr, pfid);
290 if (mdd_is_root(mdd, pfid))
292 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
294 if (lu_fid_eq(pfid, lf))
297 mdd_object_put(env, parent);
299 parent = mdd_object_find(env, mdd, pfid);
301 GOTO(out, rc = PTR_ERR(parent));
303 if (!mdd_object_exists(parent))
304 GOTO(out, rc = -EINVAL);
310 if (parent && !IS_ERR(parent))
311 mdd_object_put(env, parent);
316 * No permission check is needed.
318 * returns 1: if fid is ancestor of @mo;
319 * returns 0: if fid is not an ancestor of @mo;
321 * returns EREMOTE if remote object is found, fid of remote object is saved to
324 * returns < 0: if error
326 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
327 const struct lu_fid *fid, struct lu_fid *sfid)
329 struct mdd_device *mdd = mdo2mdd(mo);
330 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
334 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
337 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
341 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
345 } else if (rc == 1) {
346 /* found @fid is parent */
354 * Check that @dir contains no entries except (possibly) dot and dotdot.
359 * -ENOTDIR not a directory object
360 * -ENOTEMPTY not empty
364 static int mdd_dir_is_empty(const struct lu_env *env,
365 struct mdd_object *dir)
368 struct dt_object *obj;
369 const struct dt_it_ops *iops;
373 obj = mdd_object_child(dir);
374 if (!dt_try_as_dir(env, obj))
377 iops = &obj->do_index_ops->dio_it;
378 it = iops->init(env, obj, LUDA_64BITHASH);
380 result = iops->get(env, it, (const struct dt_key *)"");
383 for (result = 0, i = 0; result == 0 && i < 3; ++i)
384 result = iops->next(env, it);
387 else if (result == 1)
389 } else if (result == 0)
391 * Huh? Index contains no zero key?
398 result = PTR_ERR(it);
403 * Determine if the target object can be hard linked, and right now it only
404 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
405 * directory nlink count might exceed the maximum link count(see
406 * osd_object_ref_add), so it only check nlink for non-directories.
408 * \param[in] env thread environment
409 * \param[in] obj object being linked to
410 * \param[in] la attributes of \a obj
412 * \retval 0 if \a obj can be hard linked
413 * \retval negative error if \a obj is a directory or has too
416 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
417 const struct lu_attr *la)
419 struct mdd_device *m = mdd_obj2mdd_dev(obj);
424 /* Subdir count limitation can be broken through
425 * (see osd_object_ref_add), so only check non-directory here. */
426 if (!S_ISDIR(la->la_mode) &&
427 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
434 * Check whether it may create the cobj under the pobj.
436 * \param[in] env execution environment
437 * \param[in] pobj the parent directory
438 * \param[in] pattr the attribute of the parent directory
439 * \param[in] cobj the child to be created
440 * \param[in] check_perm if check WRITE|EXEC permission for parent
442 * \retval = 0 create the child under this dir is allowed
443 * \retval negative errno create the child under this dir is
446 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
447 const struct lu_attr *pattr, struct mdd_object *cobj,
453 if (cobj && mdd_object_exists(cobj))
456 if (mdd_is_dead_obj(pobj))
460 rc = mdd_permission_internal_locked(env, pobj, pattr,
461 MAY_WRITE | MAY_EXEC,
467 * Check whether can unlink from the pobj in the case of "cobj == NULL".
469 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
470 const struct lu_attr *pattr, const struct lu_attr *attr)
475 if (mdd_is_dead_obj(pobj))
478 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
481 rc = mdd_permission_internal_locked(env, pobj, pattr,
482 MAY_WRITE | MAY_EXEC,
487 if (pattr->la_flags & LUSTRE_APPEND_FL)
494 * pobj == NULL is remote ops case, under such case, pobj's
495 * VTX feature has been checked already, no need check again.
497 static inline int mdd_is_sticky(const struct lu_env *env,
498 struct mdd_object *pobj,
499 const struct lu_attr *pattr,
500 struct mdd_object *cobj,
501 const struct lu_attr *cattr)
503 struct lu_ucred *uc = lu_ucred_assert(env);
506 LASSERT(pattr != NULL);
507 if (!(pattr->la_mode & S_ISVTX) ||
508 (pattr->la_uid == uc->uc_fsuid))
512 LASSERT(cattr != NULL);
513 if (cattr->la_uid == uc->uc_fsuid)
516 return !md_capable(uc, CFS_CAP_FOWNER);
519 static int mdd_may_delete_entry(const struct lu_env *env,
520 struct mdd_object *pobj,
521 const struct lu_attr *pattr,
526 LASSERT(pobj != NULL);
527 if (!mdd_object_exists(pobj))
530 if (mdd_is_dead_obj(pobj))
535 rc = mdd_permission_internal_locked(env, pobj, pattr,
536 MAY_WRITE | MAY_EXEC,
542 if (pattr->la_flags & LUSTRE_APPEND_FL)
549 * Check whether it may delete the cobj from the pobj.
552 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
553 const struct lu_attr *tpattr, struct mdd_object *tobj,
554 const struct lu_attr *tattr, const struct lu_attr *cattr,
555 int check_perm, int check_empty)
561 LASSERT(tpattr != NULL);
562 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
570 if (!mdd_object_exists(tobj))
573 if (mdd_is_dead_obj(tobj))
576 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
579 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
582 /* additional check the rename case */
584 if (S_ISDIR(cattr->la_mode)) {
585 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
587 if (!S_ISDIR(tattr->la_mode))
590 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
592 } else if (S_ISDIR(tattr->la_mode))
596 if (S_ISDIR(tattr->la_mode) && check_empty)
597 rc = mdd_dir_is_empty(env, tobj);
603 * Check whether it can create the link file(linked to @src_obj) under
604 * the target directory(@tgt_obj), and src_obj has been locked by
607 * \param[in] env execution environment
608 * \param[in] tgt_obj the target directory
609 * \param[in] tattr attributes of target directory
610 * \param[in] lname the link name
611 * \param[in] src_obj source object for link
612 * \param[in] cattr attributes for source object
614 * \retval = 0 it is allowed to create the link file under tgt_obj
615 * \retval negative error not allowed to create the link file
617 static int mdd_link_sanity_check(const struct lu_env *env,
618 struct mdd_object *tgt_obj,
619 const struct lu_attr *tattr,
620 const struct lu_name *lname,
621 struct mdd_object *src_obj,
622 const struct lu_attr *cattr)
624 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
628 if (!mdd_object_exists(src_obj))
631 if (mdd_is_dead_obj(src_obj))
634 /* Local ops, no lookup before link, check filename length here. */
635 rc = mdd_name_check(m, lname);
639 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
642 if (S_ISDIR(mdd_object_type(src_obj)))
645 LASSERT(src_obj != tgt_obj);
646 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
650 rc = __mdd_may_link(env, src_obj, cattr);
655 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
656 const char *name, struct thandle *handle)
658 struct dt_object *next = mdd_object_child(pobj);
662 if (dt_try_as_dir(env, next))
663 rc = dt_delete(env, next, (struct dt_key *)name, handle);
670 static int __mdd_index_insert_only(const struct lu_env *env,
671 struct mdd_object *pobj,
672 const struct lu_fid *lf, __u32 type,
674 struct thandle *handle)
676 struct dt_object *next = mdd_object_child(pobj);
680 if (dt_try_as_dir(env, next)) {
681 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
682 struct lu_ucred *uc = lu_ucred_check(env);
686 rec->rec_type = type;
687 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
688 rc = dt_insert(env, next, (const struct dt_rec *)rec,
689 (const struct dt_key *)name, handle,
697 /* insert named index, add reference if isdir */
698 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
699 const struct lu_fid *lf, __u32 type,
700 const char *name, struct thandle *handle)
705 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
706 if (rc == 0 && S_ISDIR(type)) {
707 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
708 mdo_ref_add(env, pobj, handle);
709 mdd_write_unlock(env, pobj);
715 /* delete named index, drop reference if isdir */
716 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
717 const char *name, int is_dir,
718 struct thandle *handle)
723 rc = __mdd_index_delete_only(env, pobj, name, handle);
724 if (rc == 0 && is_dir) {
725 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
726 mdo_ref_del(env, pobj, handle);
727 mdd_write_unlock(env, pobj);
733 static int mdd_llog_record_calc_size(const struct lu_env *env,
734 const struct lu_name *tname,
735 const struct lu_name *sname)
737 const struct lu_ucred *uc = lu_ucred(env);
738 enum changelog_rec_flags crf = CLF_EXTRA_FLAGS;
739 enum changelog_rec_extra_flags crfe = CLFE_UIDGID;
744 if (uc != NULL && uc->uc_jobid[0] != '\0')
747 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
748 changelog_rec_offset(crf, crfe) +
749 (tname != NULL ? tname->ln_namelen : 0) +
750 (sname != NULL ? 1 + sname->ln_namelen : 0));
753 int mdd_declare_changelog_store(const struct lu_env *env,
754 struct mdd_device *mdd,
755 const struct lu_name *tname,
756 const struct lu_name *sname,
757 struct thandle *handle)
759 struct obd_device *obd = mdd2obd_dev(mdd);
760 struct llog_ctxt *ctxt;
761 struct llog_changelog_rec *rec;
763 struct thandle *llog_th;
768 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
771 reclen = mdd_llog_record_calc_size(env, tname, sname);
772 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
773 if (buf->lb_buf == NULL)
777 rec->cr_hdr.lrh_len = reclen;
778 rec->cr_hdr.lrh_type = CHANGELOG_REC;
780 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
784 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
786 GOTO(out_put, rc = PTR_ERR(llog_th));
788 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
796 /** Add a changelog entry \a rec to the changelog llog
799 * \param handle - currently ignored since llogs start their own transaction;
800 * this will hopefully be fixed in llog rewrite
803 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
804 struct llog_changelog_rec *rec, struct thandle *th)
806 struct obd_device *obd = mdd2obd_dev(mdd);
807 struct llog_ctxt *ctxt;
808 struct thandle *llog_th;
811 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
812 changelog_rec_varsize(&rec->cr));
814 /* llog_lvfs_write_rec sets the llog tail len */
815 rec->cr_hdr.lrh_type = CHANGELOG_REC;
816 rec->cr.cr_time = cl_time();
818 spin_lock(&mdd->mdd_cl.mc_lock);
819 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
820 * but as long as the MDD transactions are ordered correctly for e.g.
821 * rename conflicts, I don't think this should matter. */
822 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
823 spin_unlock(&mdd->mdd_cl.mc_lock);
825 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
829 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
831 GOTO(out_put, rc = PTR_ERR(llog_th));
833 /* nested journal transaction */
834 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
843 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
844 const struct lu_fid *sfid,
845 const struct lu_fid *spfid,
846 const struct lu_name *sname)
848 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
849 size_t extsize = sname->ln_namelen + 1;
851 LASSERT(sfid != NULL);
852 LASSERT(spfid != NULL);
853 LASSERT(sname != NULL);
855 rnm->cr_sfid = *sfid;
856 rnm->cr_spfid = *spfid;
858 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
859 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
860 rec->cr_namelen += extsize;
863 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
865 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
867 if (jobid == NULL || jobid[0] == '\0')
870 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
873 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
875 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
877 ef->cr_extra_flags = eflags;
880 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
881 __u64 uid, __u64 gid)
883 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
885 uidgid->cr_uid = uid;
886 uidgid->cr_gid = gid;
889 /** Store a namespace change changelog record
890 * If this fails, we must fail the whole transaction; we don't
891 * want the change to commit without the log entry.
892 * \param target - mdd_object of change
893 * \param tpfid - target parent dir/object fid
894 * \param sfid - source object fid
895 * \param spfid - source parent fid
896 * \param tname - target name string
897 * \param sname - source name string
898 * \param handle - transaction handle
900 int mdd_changelog_ns_store(const struct lu_env *env,
901 struct mdd_device *mdd,
902 enum changelog_rec_type type,
903 enum changelog_rec_flags crf,
904 struct mdd_object *target,
905 const struct lu_fid *tpfid,
906 const struct lu_fid *sfid,
907 const struct lu_fid *spfid,
908 const struct lu_name *tname,
909 const struct lu_name *sname,
910 struct thandle *handle)
912 const struct lu_ucred *uc = lu_ucred(env);
913 struct llog_changelog_rec *rec;
916 __u64 xflags = CLFE_INVALID;
921 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
924 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
927 LASSERT(tpfid != NULL);
928 LASSERT(tname != NULL);
929 LASSERT(handle != NULL);
931 reclen = mdd_llog_record_calc_size(env, tname, sname);
932 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
933 if (buf->lb_buf == NULL)
938 crf |= CLF_EXTRA_FLAGS;
940 if (uc != NULL && uc->uc_jobid[0] != '\0')
948 xflags |= CLFE_UIDGID;
950 rec->cr.cr_flags = crf;
952 if (crf & CLF_EXTRA_FLAGS) {
953 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
954 if (xflags & CLFE_UIDGID)
955 mdd_changelog_rec_extra_uidgid(&rec->cr,
956 uc->uc_uid, uc->uc_gid);
959 rec->cr.cr_type = (__u32)type;
960 rec->cr.cr_pfid = *tpfid;
961 rec->cr.cr_namelen = tname->ln_namelen;
962 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
964 if (crf & CLF_RENAME)
965 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
968 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
970 if (likely(target != NULL)) {
971 rec->cr.cr_tfid = *mdo2fid(target);
972 target->mod_cltime = ktime_get();
974 fid_zero(&rec->cr.cr_tfid);
977 rc = mdd_changelog_store(env, mdd, rec, handle);
979 CERROR("%s: cannot store changelog record: type = %d, "
980 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
981 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
982 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
989 static int __mdd_links_add(const struct lu_env *env,
990 struct mdd_object *mdd_obj,
991 struct linkea_data *ldata,
992 const struct lu_name *lname,
993 const struct lu_fid *pfid,
994 int first, int check)
998 if (ldata->ld_leh == NULL) {
999 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1003 rc = linkea_data_new(ldata,
1004 &mdd_env_info(env)->mti_link_buf);
1011 rc = linkea_links_find(ldata, lname, pfid);
1012 if (rc && rc != -ENOENT)
1018 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1019 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1023 linkea_add_buf(ldata, lname, tfid);
1026 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1027 linkea_add_buf(ldata, lname, pfid);
1029 return linkea_add_buf(ldata, lname, pfid);
1032 static int __mdd_links_del(const struct lu_env *env,
1033 struct mdd_object *mdd_obj,
1034 struct linkea_data *ldata,
1035 const struct lu_name *lname,
1036 const struct lu_fid *pfid)
1040 if (ldata->ld_leh == NULL) {
1041 rc = mdd_links_read(env, mdd_obj, ldata);
1046 rc = linkea_links_find(ldata, lname, pfid);
1050 linkea_del_buf(ldata, lname);
1054 static int mdd_linkea_prepare(const struct lu_env *env,
1055 struct mdd_object *mdd_obj,
1056 const struct lu_fid *oldpfid,
1057 const struct lu_name *oldlname,
1058 const struct lu_fid *newpfid,
1059 const struct lu_name *newlname,
1060 int first, int check,
1061 struct linkea_data *ldata)
1066 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1069 LASSERT(oldpfid != NULL || newpfid != NULL);
1071 if (mdd_obj->mod_flags & DEAD_OBJ)
1072 /* Unnecessary to update linkEA for dead object. */
1075 if (oldpfid != NULL) {
1076 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1078 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1081 /* No changes done. */
1086 /* If renaming, add the new record */
1087 if (newpfid != NULL)
1088 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1094 int mdd_links_rename(const struct lu_env *env,
1095 struct mdd_object *mdd_obj,
1096 const struct lu_fid *oldpfid,
1097 const struct lu_name *oldlname,
1098 const struct lu_fid *newpfid,
1099 const struct lu_name *newlname,
1100 struct thandle *handle,
1101 struct linkea_data *ldata,
1102 int first, int check)
1107 if (ldata == NULL) {
1108 ldata = &mdd_env_info(env)->mti_link_data;
1109 memset(ldata, 0, sizeof(*ldata));
1110 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1111 newpfid, newlname, first, check, ldata);
1116 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1117 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1123 if (newlname == NULL)
1124 CERROR("link_ea add failed %d "DFID"\n",
1125 rc, PFID(mdd_object_fid(mdd_obj)));
1126 else if (oldpfid == NULL)
1127 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1128 newlname->ln_namelen, newlname->ln_name, rc,
1129 PFID(mdd_object_fid(mdd_obj)));
1130 else if (newpfid == NULL)
1131 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1132 oldlname->ln_namelen, oldlname->ln_name, rc,
1133 PFID(mdd_object_fid(mdd_obj)));
1135 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1136 "\n", oldlname->ln_namelen, oldlname->ln_name,
1137 newlname->ln_namelen, newlname->ln_name, rc,
1138 PFID(mdd_object_fid(mdd_obj)));
1141 if (is_vmalloc_addr(ldata->ld_buf))
1142 /* if we vmalloced a large buffer drop it */
1143 lu_buf_free(ldata->ld_buf);
1148 static inline int mdd_links_add(const struct lu_env *env,
1149 struct mdd_object *mdd_obj,
1150 const struct lu_fid *pfid,
1151 const struct lu_name *lname,
1152 struct thandle *handle,
1153 struct linkea_data *ldata, int first)
1155 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1156 pfid, lname, handle, ldata, first, 0);
1159 static inline int mdd_links_del(const struct lu_env *env,
1160 struct mdd_object *mdd_obj,
1161 const struct lu_fid *pfid,
1162 const struct lu_name *lname,
1163 struct thandle *handle)
1165 return mdd_links_rename(env, mdd_obj, pfid, lname,
1166 NULL, NULL, handle, NULL, 0, 0);
1169 /** Read the link EA into a temp buffer.
1170 * Uses the name_buf since it is generally large.
1171 * \retval IS_ERR err
1172 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1174 struct lu_buf *mdd_links_get(const struct lu_env *env,
1175 struct mdd_object *mdd_obj)
1177 struct linkea_data ldata = { NULL };
1180 rc = mdd_links_read(env, mdd_obj, &ldata);
1181 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1184 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1185 struct linkea_data *ldata, struct thandle *handle)
1187 const struct lu_buf *buf;
1190 if (ldata == NULL || ldata->ld_buf == NULL ||
1191 ldata->ld_leh == NULL)
1194 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1198 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1199 ldata->ld_leh->leh_len);
1200 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1201 if (unlikely(rc == -ENOSPC)) {
1202 rc = linkea_overflow_shrink(ldata);
1210 static int mdd_declare_links_add(const struct lu_env *env,
1211 struct mdd_object *mdd_obj,
1212 struct thandle *handle,
1213 struct linkea_data *ldata)
1219 if (ldata != NULL && ldata->ld_leh != NULL) {
1220 ea_len = ldata->ld_leh->leh_len;
1221 linkea = ldata->ld_buf->lb_buf;
1223 ea_len = MAX_LINKEA_SIZE;
1227 rc = mdo_declare_xattr_set(env, mdd_obj,
1228 mdd_buf_get_const(env, linkea, ea_len),
1229 XATTR_NAME_LINK, 0, handle);
1234 static inline int mdd_declare_links_del(const struct lu_env *env,
1235 struct mdd_object *c,
1236 struct thandle *handle)
1240 /* For directory, the linkEA will be removed together
1241 * with the object. */
1242 if (!S_ISDIR(mdd_object_type(c)))
1243 rc = mdd_declare_links_add(env, c, handle, NULL);
1248 static int mdd_declare_link(const struct lu_env *env,
1249 struct mdd_device *mdd,
1250 struct mdd_object *p,
1251 struct mdd_object *c,
1252 const struct lu_name *name,
1253 struct thandle *handle,
1255 struct linkea_data *data)
1257 struct lu_fid tfid = *mdo2fid(c);
1260 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1261 tfid.f_oid = cfs_fail_val;
1263 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1264 name->ln_name, handle);
1268 rc = mdo_declare_ref_add(env, c, handle);
1272 la->la_valid = LA_CTIME | LA_MTIME;
1273 rc = mdo_declare_attr_set(env, p, la, handle);
1277 la->la_valid = LA_CTIME;
1278 rc = mdo_declare_attr_set(env, c, la, handle);
1282 rc = mdd_declare_links_add(env, c, handle, data);
1286 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1291 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1292 struct md_object *src_obj, const struct lu_name *lname,
1295 const char *name = lname->ln_name;
1296 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1297 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1298 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1299 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1300 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1301 struct mdd_device *mdd = mdo2mdd(src_obj);
1302 struct thandle *handle;
1303 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1304 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1308 rc = mdd_la_get(env, mdd_sobj, cattr);
1312 rc = mdd_la_get(env, mdd_tobj, tattr);
1317 * If we are using project inheritance, we only allow hard link
1318 * creation in our tree when the project IDs are the same;
1319 * otherwise the tree quota mechanism could be circumvented.
1321 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1322 (tattr->la_projid != cattr->la_projid))
1325 handle = mdd_trans_create(env, mdd);
1327 GOTO(out_pending, rc = PTR_ERR(handle));
1329 memset(ldata, 0, sizeof(*ldata));
1331 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1332 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1334 /* Note: even this function will change ldata, but it comes from
1335 * thread_info, which is completely temporary and only seen in
1336 * this function, so we do not need reset ldata once it fails.*/
1337 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1338 lname, 0, 0, ldata);
1342 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1347 rc = mdd_trans_start(env, mdd, handle);
1351 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1352 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1355 GOTO(out_unlock, rc);
1357 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_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 if (S_ISDIR(cattr->la_mode) &&
1970 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
1971 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
1972 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
1974 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC) &&
1975 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
1977 CERROR("%s: invalid lmv_user_md: magic = %x, "
1978 "stripe_offset = %d, stripe_count = %u: "
1979 "rc = %d\n", mdd2obd_dev(m)->obd_name,
1980 le32_to_cpu(lum->lum_magic),
1981 (int)le32_to_cpu(lum->lum_stripe_offset),
1982 le32_to_cpu(lum->lum_stripe_count), rc);
1987 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1992 if (pattr->la_mode & S_ISGID) {
1993 cattr->la_gid = pattr->la_gid;
1994 if (S_ISDIR(cattr->la_mode)) {
1995 cattr->la_mode |= S_ISGID;
1996 cattr->la_valid |= LA_MODE;
2000 /* Inherit project ID from parent directory */
2001 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2002 cattr->la_projid = pattr->la_projid;
2003 if (S_ISDIR(cattr->la_mode)) {
2004 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2005 cattr->la_valid |= LA_FLAGS;
2007 cattr->la_valid |= LA_PROJID;
2010 rc = mdd_name_check(m, lname);
2014 switch (cattr->la_mode & S_IFMT) {
2016 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2018 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2019 RETURN(-ENAMETOOLONG);
2038 static int mdd_declare_create_object(const struct lu_env *env,
2039 struct mdd_device *mdd,
2040 struct mdd_object *p, struct mdd_object *c,
2041 struct lu_attr *attr,
2042 struct thandle *handle,
2043 const struct md_op_spec *spec,
2044 struct lu_buf *def_acl_buf,
2045 struct lu_buf *acl_buf,
2046 struct dt_allocation_hint *hint)
2048 const struct lu_buf *buf;
2051 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2056 #ifdef CONFIG_FS_POSIX_ACL
2057 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2058 /* if dir, then can inherit default ACl */
2059 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2060 XATTR_NAME_ACL_DEFAULT,
2066 if (acl_buf->lb_len > 0) {
2067 rc = mdo_declare_attr_set(env, c, attr, handle);
2071 rc = mdo_declare_xattr_set(env, c, acl_buf,
2072 XATTR_NAME_ACL_ACCESS, 0, handle);
2077 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2081 /* replay case, create LOV EA from client data */
2082 if (spec->no_create ||
2083 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2084 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2085 spec->u.sp_ea.eadatalen);
2086 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2092 if (S_ISLNK(attr->la_mode)) {
2093 const char *target_name = spec->u.sp_symname;
2094 int sym_len = strlen(target_name);
2095 const struct lu_buf *buf;
2097 buf = mdd_buf_get_const(env, target_name, sym_len);
2098 rc = dt_declare_record_write(env, mdd_object_child(c),
2104 if (spec->sp_cr_file_secctx_name != NULL) {
2105 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2106 spec->sp_cr_file_secctx_size);
2107 rc = mdo_declare_xattr_set(env, c, buf,
2108 spec->sp_cr_file_secctx_name, 0,
2117 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2118 struct mdd_object *p, struct mdd_object *c,
2119 const struct lu_name *name,
2120 struct lu_attr *attr,
2121 struct thandle *handle,
2122 const struct md_op_spec *spec,
2123 struct linkea_data *ldata,
2124 struct lu_buf *def_acl_buf,
2125 struct lu_buf *acl_buf,
2126 struct dt_allocation_hint *hint)
2130 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2131 def_acl_buf, acl_buf, hint);
2135 if (S_ISDIR(attr->la_mode)) {
2136 rc = mdo_declare_ref_add(env, p, handle);
2141 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2142 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2146 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2148 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2149 name->ln_name, handle);
2153 rc = mdd_declare_links_add(env, c, handle, ldata);
2158 la->la_valid = LA_CTIME | LA_MTIME;
2159 rc = mdo_declare_attr_set(env, p, la, handle);
2163 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2171 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2172 struct lu_attr *la, struct lu_buf *def_acl_buf,
2173 struct lu_buf *acl_buf)
2178 if (S_ISLNK(la->la_mode)) {
2179 acl_buf->lb_len = 0;
2180 def_acl_buf->lb_len = 0;
2184 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2185 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2186 XATTR_NAME_ACL_DEFAULT);
2187 mdd_read_unlock(env, pobj);
2189 /* If there are default ACL, fix mode/ACL by default ACL */
2190 def_acl_buf->lb_len = rc;
2191 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2192 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2193 acl_buf->lb_len = rc;
2194 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2197 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2198 /* If there are no default ACL, fix mode by mask */
2199 struct lu_ucred *uc = lu_ucred(env);
2201 /* The create triggered by MDT internal events, such as
2202 * LFSCK reset, will not contain valid "uc". */
2203 if (unlikely(uc != NULL))
2204 la->la_mode &= ~uc->uc_umask;
2206 acl_buf->lb_len = 0;
2207 def_acl_buf->lb_len = 0;
2214 * Create a metadata object and initialize it, set acl, xattr.
2216 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2217 struct mdd_object *son, struct lu_attr *attr,
2218 struct md_op_spec *spec, struct lu_buf *acl_buf,
2219 struct lu_buf *def_acl_buf,
2220 struct dt_allocation_hint *hint,
2221 struct thandle *handle)
2223 const struct lu_buf *buf;
2226 mdd_write_lock(env, son, MOR_TGT_CHILD);
2227 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2232 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2233 * created in declare phase, they also needs to be added to master
2234 * object as sub-directory entry. So it has to initialize the master
2235 * object, then set dir striped EA.(in mdo_xattr_set) */
2236 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2239 GOTO(err_destroy, rc);
2242 * in case of replay we just set LOVEA provided by the client
2243 * XXX: I think it would be interesting to try "old" way where
2244 * MDT calls this xattr_set(LOV) in a different transaction.
2245 * probably this way we code can be made better.
2248 /* During creation, there are only a few cases we need do xattr_set to
2250 * 1. regular file: see comments above.
2251 * 2. dir: inherit default striping or pool settings from parent.
2252 * 3. create striped directory with provided stripeEA.
2253 * 4. create striped directory because inherit default layout from the
2256 if (spec->no_create ||
2257 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2258 S_ISDIR(attr->la_mode)) {
2259 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2260 spec->u.sp_ea.eadatalen);
2261 rc = mdo_xattr_set(env, son, buf,
2262 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2266 GOTO(err_destroy, rc);
2269 #ifdef CONFIG_FS_POSIX_ACL
2270 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2271 S_ISDIR(attr->la_mode)) {
2272 /* set default acl */
2273 rc = mdo_xattr_set(env, son, def_acl_buf,
2274 XATTR_NAME_ACL_DEFAULT, 0,
2277 GOTO(err_destroy, rc);
2279 /* set its own acl */
2280 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2281 rc = mdo_xattr_set(env, son, acl_buf,
2282 XATTR_NAME_ACL_ACCESS,
2285 GOTO(err_destroy, rc);
2289 if (S_ISLNK(attr->la_mode)) {
2290 struct lu_ucred *uc = lu_ucred_assert(env);
2291 struct dt_object *dt = mdd_object_child(son);
2292 const char *target_name = spec->u.sp_symname;
2293 int sym_len = strlen(target_name);
2296 buf = mdd_buf_get_const(env, target_name, sym_len);
2297 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2299 CFS_CAP_SYS_RESOURCE_MASK);
2304 GOTO(err_initlized, rc = -EFAULT);
2307 if (spec->sp_cr_file_secctx_name != NULL) {
2308 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2309 spec->sp_cr_file_secctx_size);
2310 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2313 GOTO(err_initlized, rc);
2317 if (unlikely(rc != 0)) {
2319 if (S_ISDIR(attr->la_mode)) {
2320 /* Drop the reference, no need to delete "."/"..",
2321 * because the object to be destroied directly. */
2322 rc2 = mdo_ref_del(env, son, handle);
2326 rc2 = mdo_ref_del(env, son, handle);
2330 mdo_destroy(env, son, handle);
2333 mdd_write_unlock(env, son);
2337 static int mdd_index_delete(const struct lu_env *env,
2338 struct mdd_object *mdd_pobj,
2339 struct lu_attr *cattr,
2340 const struct lu_name *lname)
2342 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2343 struct thandle *handle;
2347 handle = mdd_trans_create(env, mdd);
2349 RETURN(PTR_ERR(handle));
2351 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2356 if (S_ISDIR(cattr->la_mode)) {
2357 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2362 /* Since this will only be used in the error handler path,
2363 * Let's set the thandle to be local and not mess the transno */
2364 handle->th_local = 1;
2365 rc = mdd_trans_start(env, mdd, handle);
2369 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2370 S_ISDIR(cattr->la_mode), handle);
2374 rc = mdd_trans_stop(env, mdd, rc, handle);
2380 * Create object and insert it into namespace.
2382 * Two operations have to be performed:
2384 * - an allocation of a new object (->do_create()), and
2385 * - an insertion into a parent index (->dio_insert()).
2387 * Due to locking, operation order is not important, when both are
2388 * successful, *but* error handling cases are quite different:
2390 * - if insertion is done first, and following object creation fails,
2391 * insertion has to be rolled back, but this operation might fail
2392 * also leaving us with dangling index entry.
2394 * - if creation is done first, is has to be undone if insertion fails,
2395 * leaving us with leaked space, which is not good but not fatal.
2397 * It seems that creation-first is simplest solution, but it is sub-optimal
2403 * case, because second mkdir is bound to create object, only to
2404 * destroy it immediately.
2406 * To avoid this follow local file systems that do double lookup:
2408 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2409 * 1. create (mdd_create_object_internal())
2410 * 2. insert (__mdd_index_insert(), lookup again)
2412 * \param[in] pobj parent object
2413 * \param[in] lname name of child being created
2414 * \param[in,out] child child object being created
2415 * \param[in] spec additional create parameters
2416 * \param[in] ma attributes for new child object
2418 * \retval 0 on success
2419 * \retval negative errno on failure
2421 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2422 const struct lu_name *lname, struct md_object *child,
2423 struct md_op_spec *spec, struct md_attr *ma)
2425 struct mdd_thread_info *info = mdd_env_info(env);
2426 struct lu_attr *la = &info->mti_la_for_fix;
2427 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2428 struct mdd_object *son = md2mdd_obj(child);
2429 struct mdd_device *mdd = mdo2mdd(pobj);
2430 struct lu_attr *attr = &ma->ma_attr;
2431 struct thandle *handle;
2432 struct lu_attr *pattr = &info->mti_pattr;
2433 struct lu_buf acl_buf;
2434 struct lu_buf def_acl_buf;
2435 struct linkea_data *ldata = &info->mti_link_data;
2436 const char *name = lname->ln_name;
2437 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2442 rc = mdd_la_get(env, mdd_pobj, pattr);
2446 /* Sanity checks before big job. */
2447 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2451 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2452 GOTO(out_free, rc = -EINPROGRESS);
2454 handle = mdd_trans_create(env, mdd);
2456 GOTO(out_free, rc = PTR_ERR(handle));
2458 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2459 mdd->mdd_dt_conf.ddp_max_ea_size);
2460 acl_buf = info->mti_xattr_buf;
2461 def_acl_buf.lb_buf = info->mti_key;
2462 def_acl_buf.lb_len = sizeof(info->mti_key);
2463 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2467 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2469 memset(ldata, 0, sizeof(*ldata));
2470 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2471 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2474 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2475 &tfid, lname, 1, 0, ldata);
2477 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2478 mdd_object_fid(mdd_pobj),
2479 lname, 1, 0, ldata);
2482 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2483 handle, spec, ldata, &def_acl_buf, &acl_buf,
2488 rc = mdd_trans_start(env, mdd, handle);
2492 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2493 &def_acl_buf, hint, handle);
2497 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2498 mdd_write_lock(env, son, MOR_TGT_CHILD);
2499 son->mod_flags |= VOLATILE_OBJ;
2500 rc = __mdd_orphan_add(env, son, handle);
2501 GOTO(out_volatile, rc);
2503 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2504 attr->la_mode, name, handle);
2506 GOTO(err_created, rc);
2508 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2511 /* update parent directory mtime/ctime */
2513 la->la_valid = LA_CTIME | LA_MTIME;
2514 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2516 GOTO(err_insert, rc);
2522 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2523 rc2 = __mdd_orphan_del(env, son, handle);
2525 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2526 S_ISDIR(attr->la_mode),
2532 mdd_write_lock(env, son, MOR_TGT_CHILD);
2533 if (S_ISDIR(attr->la_mode)) {
2534 /* Drop the reference, no need to delete "."/"..",
2535 * because the object is to be destroyed directly. */
2536 rc2 = mdo_ref_del(env, son, handle);
2538 mdd_write_unlock(env, son);
2543 /* For volatile files drop one link immediately, since there is
2544 * no filename in the namespace, and save any error returned. */
2545 rc2 = mdo_ref_del(env, son, handle);
2547 mdd_write_unlock(env, son);
2548 if (unlikely(rc == 0))
2553 /* Don't destroy the volatile object on success */
2554 if (likely(rc != 0))
2555 mdo_destroy(env, son, handle);
2556 mdd_write_unlock(env, son);
2559 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2560 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2561 rc = mdd_changelog_ns_store(env, mdd,
2562 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2563 S_ISREG(attr->la_mode) ? CL_CREATE :
2564 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2565 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2568 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2570 /* If creation fails, it is most likely due to the remote update
2571 * failure, because local transaction will mostly succeed at
2572 * this stage. There is no easy way to rollback all of previous
2573 * updates, so let's remove the object from namespace, and
2574 * LFSCK should handle the orphan object. */
2575 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2576 mdd_index_delete(env, mdd_pobj, attr, lname);
2580 if (is_vmalloc_addr(ldata->ld_buf))
2581 /* if we vmalloced a large buffer drop it */
2582 lu_buf_free(ldata->ld_buf);
2584 /* The child object shouldn't be cached anymore */
2586 set_bit(LU_OBJECT_HEARD_BANSHEE,
2587 &child->mo_lu.lo_header->loh_flags);
2592 * Get locks on parents in proper order
2593 * RETURN: < 0 - error, rename_order if successful
2601 static int mdd_rename_order(const struct lu_env *env,
2602 struct mdd_device *mdd,
2603 struct mdd_object *src_pobj,
2604 const struct lu_attr *pattr,
2605 struct mdd_object *tgt_pobj)
2607 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2611 if (src_pobj == tgt_pobj)
2612 RETURN(MDD_RN_SAME);
2614 /* compared the parent child relationship of src_p&tgt_p */
2615 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2617 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2620 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2634 /* has not mdd_write{read}_lock on any obj yet. */
2635 static int mdd_rename_sanity_check(const struct lu_env *env,
2636 struct mdd_object *src_pobj,
2637 const struct lu_attr *pattr,
2638 struct mdd_object *tgt_pobj,
2639 const struct lu_attr *tpattr,
2640 struct mdd_object *sobj,
2641 const struct lu_attr *cattr,
2642 struct mdd_object *tobj,
2643 const struct lu_attr *tattr)
2648 /* XXX: when get here, sobj must NOT be NULL,
2649 * the other case has been processed in cld_rename
2650 * before mdd_rename and enable MDS_PERM_BYPASS. */
2654 * If we are using project inheritance, we only allow renames
2655 * into our tree when the project IDs are the same; otherwise
2656 * tree quota mechanism would be circumvented.
2658 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2659 tpattr->la_projid != cattr->la_projid) ||
2660 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2661 (pattr->la_projid != tpattr->la_projid)))
2664 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2668 /* XXX: when get here, "tobj == NULL" means tobj must
2669 * NOT exist (neither on remote MDS, such case has been
2670 * processed in cld_rename before mdd_rename and enable
2672 * So check may_create, but not check may_unlink. */
2674 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2675 (src_pobj != tgt_pobj));
2677 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2678 (src_pobj != tgt_pobj), 1);
2680 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2681 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2686 static int mdd_declare_rename(const struct lu_env *env,
2687 struct mdd_device *mdd,
2688 struct mdd_object *mdd_spobj,
2689 struct mdd_object *mdd_tpobj,
2690 struct mdd_object *mdd_sobj,
2691 struct mdd_object *mdd_tobj,
2692 const struct lu_name *sname,
2693 const struct lu_name *tname,
2695 struct linkea_data *ldata,
2696 struct thandle *handle)
2698 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2701 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2702 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2708 /* name from source dir */
2709 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2713 /* .. from source child */
2714 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2715 /* source child can be directory,
2716 * counted by source dir's nlink */
2717 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2720 if (mdd_spobj != mdd_tpobj) {
2721 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2726 rc = mdo_declare_index_insert(env, mdd_sobj,
2728 S_IFDIR, dotdot, handle);
2733 /* new target child can be directory,
2734 * counted by target dir's nlink */
2735 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2740 la->la_valid = LA_CTIME | LA_MTIME;
2741 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2745 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2749 la->la_valid = LA_CTIME;
2750 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2754 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2759 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2760 mdd_object_type(mdd_sobj),
2761 tname->ln_name, handle);
2765 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2766 /* delete target child in target parent directory */
2767 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2772 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2776 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2777 /* target child can be directory,
2778 * delete "." reference in target child directory */
2779 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2783 /* delete ".." reference in target parent directory */
2784 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2789 la->la_valid = LA_CTIME;
2790 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2794 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2799 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2806 /* src object can be remote that is why we use only fid and type of object */
2807 static int mdd_rename(const struct lu_env *env,
2808 struct md_object *src_pobj, struct md_object *tgt_pobj,
2809 const struct lu_fid *lf, const struct lu_name *lsname,
2810 struct md_object *tobj, const struct lu_name *ltname,
2813 const char *sname = lsname->ln_name;
2814 const char *tname = ltname->ln_name;
2815 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2816 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2817 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2818 struct mdd_device *mdd = mdo2mdd(src_pobj);
2819 struct mdd_object *mdd_sobj = NULL; /* source object */
2820 struct mdd_object *mdd_tobj = NULL;
2821 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2822 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2823 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2824 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2825 struct thandle *handle;
2826 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2827 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2828 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2831 bool tobj_locked = 0;
2832 unsigned cl_flags = 0;
2837 mdd_tobj = md2mdd_obj(tobj);
2839 mdd_sobj = mdd_object_find(env, mdd, lf);
2840 if (IS_ERR(mdd_sobj))
2841 RETURN(PTR_ERR(mdd_sobj));
2843 rc = mdd_la_get(env, mdd_sobj, cattr);
2845 GOTO(out_pending, rc);
2847 rc = mdd_la_get(env, mdd_spobj, pattr);
2849 GOTO(out_pending, rc);
2852 rc = mdd_la_get(env, mdd_tobj, tattr);
2854 GOTO(out_pending, rc);
2855 /* search for an existing archive.
2856 * we should check ahead as the object
2857 * can be destroyed in this transaction */
2858 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2859 cl_flags |= CLF_RENAME_LAST_EXISTS;
2862 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2864 GOTO(out_pending, rc);
2866 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2867 mdd_sobj, cattr, mdd_tobj, tattr);
2869 GOTO(out_pending, rc);
2871 rc = mdd_name_check(mdd, ltname);
2873 GOTO(out_pending, rc);
2875 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2876 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2878 GOTO(out_pending, rc);
2880 handle = mdd_trans_create(env, mdd);
2882 GOTO(out_pending, rc = PTR_ERR(handle));
2884 memset(ldata, 0, sizeof(*ldata));
2885 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
2886 lsname, mdd_object_fid(mdd_tpobj), ltname,
2891 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2892 mdd_tobj, lsname, ltname, ma, ldata, handle);
2896 rc = mdd_trans_start(env, mdd, handle);
2900 is_dir = S_ISDIR(cattr->la_mode);
2902 /* Remove source name from source directory */
2903 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2907 /* "mv dir1 dir2" needs "dir1/.." link update */
2908 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2909 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2911 GOTO(fixup_spobj2, rc);
2913 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2916 GOTO(fixup_spobj, rc);
2919 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2920 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2922 /* tname might been renamed to something else */
2923 GOTO(fixup_spobj, rc);
2926 /* Insert new fid with target name into target dir */
2927 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2930 GOTO(fixup_tpobj, rc);
2932 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2933 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2935 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2936 la->la_valid = LA_CTIME;
2937 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2939 GOTO(fixup_tpobj, rc);
2941 /* Update the linkEA for the source object */
2942 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2943 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2944 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2947 /* Old files might not have EA entry */
2948 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2949 lsname, handle, NULL, 0);
2950 mdd_write_unlock(env, mdd_sobj);
2951 /* We don't fail the transaction if the link ea can't be
2952 updated -- fid2path will use alternate lookup method. */
2955 /* Remove old target object
2956 * For tobj is remote case cmm layer has processed
2957 * and set tobj to NULL then. So when tobj is NOT NULL,
2958 * it must be local one.
2960 if (tobj && mdd_object_exists(mdd_tobj)) {
2961 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2963 if (mdd_is_dead_obj(mdd_tobj)) {
2964 /* shld not be dead, something is wrong */
2965 CERROR("tobj is dead, something is wrong\n");
2969 mdo_ref_del(env, mdd_tobj, handle);
2971 /* Remove dot reference. */
2972 if (S_ISDIR(tattr->la_mode))
2973 mdo_ref_del(env, mdd_tobj, handle);
2976 /* fetch updated nlink */
2977 rc = mdd_la_get(env, mdd_tobj, tattr);
2979 CERROR("%s: Failed to get nlink for tobj "
2981 mdd2obd_dev(mdd)->obd_name,
2982 PFID(tpobj_fid), rc);
2983 GOTO(fixup_tpobj, rc);
2986 la->la_valid = LA_CTIME;
2987 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2989 CERROR("%s: Failed to set ctime for tobj "
2991 mdd2obd_dev(mdd)->obd_name,
2992 PFID(tpobj_fid), rc);
2993 GOTO(fixup_tpobj, rc);
2996 /* XXX: this transfer to ma will be removed with LOD/OSP */
2997 ma->ma_attr = *tattr;
2998 ma->ma_valid |= MA_INODE;
2999 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3002 CERROR("%s: Failed to unlink tobj "
3004 mdd2obd_dev(mdd)->obd_name,
3005 PFID(tpobj_fid), rc);
3006 GOTO(fixup_tpobj, rc);
3009 /* fetch updated nlink */
3010 rc = mdd_la_get(env, mdd_tobj, tattr);
3011 if (rc == -ENOENT) {
3012 /* the object got removed, let's
3013 * return the latest known attributes */
3014 tattr->la_nlink = 0;
3016 } else if (rc != 0) {
3017 CERROR("%s: Failed to get nlink for tobj "
3019 mdd2obd_dev(mdd)->obd_name,
3020 PFID(tpobj_fid), rc);
3021 GOTO(fixup_tpobj, rc);
3023 /* XXX: this transfer to ma will be removed with LOD/OSP */
3024 ma->ma_attr = *tattr;
3025 ma->ma_valid |= MA_INODE;
3027 if (tattr->la_nlink == 0)
3028 cl_flags |= CLF_RENAME_LAST;
3030 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3033 la->la_valid = LA_CTIME | LA_MTIME;
3034 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3036 GOTO(fixup_tpobj, rc);
3038 if (mdd_spobj != mdd_tpobj) {
3039 la->la_valid = LA_CTIME | LA_MTIME;
3040 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3042 GOTO(fixup_tpobj, rc);
3049 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3051 CWARN("tp obj fix error %d\n",rc2);
3053 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3054 !mdd_is_dead_obj(mdd_tobj)) {
3056 mdo_ref_add(env, mdd_tobj, handle);
3058 mdo_ref_add(env, mdd_tobj, handle);
3061 rc2 = __mdd_index_insert(env, mdd_tpobj,
3063 mdd_object_type(mdd_tobj),
3066 CWARN("tp obj fix error: rc = %d\n", rc2);
3071 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3072 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3074 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3075 mdd2obd_dev(mdd)->obd_name, rc2);
3078 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3081 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3082 mdd2obd_dev(mdd)->obd_name, rc2);
3087 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3088 mdd_object_type(mdd_sobj), sname,
3091 CWARN("sp obj fix error: rc = %d\n", rc2);
3096 mdd_write_unlock(env, mdd_tobj);
3099 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3100 mdd_tobj, tpobj_fid, lf, spobj_fid,
3101 ltname, lsname, handle);
3104 rc = mdd_trans_stop(env, mdd, rc, handle);
3107 mdd_object_put(env, mdd_sobj);
3112 * During migration once the parent FID has been changed,
3113 * we need update the parent FID in linkea.
3115 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3116 struct mdd_object *parent,
3117 struct mdd_object *newparent,
3118 struct mdd_object *child,
3119 const char *name, int namelen,
3120 struct thandle *handle,
3123 struct mdd_thread_info *info = mdd_env_info(env);
3124 struct linkea_data ldata = { NULL };
3125 struct lu_buf *buf = &info->mti_link_buf;
3131 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3132 if (buf->lb_buf == NULL)
3136 rc = mdd_links_read(env, child, &ldata);
3138 if (rc == -ENOENT || rc == -ENODATA)
3143 LASSERT(ldata.ld_leh != NULL);
3144 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3145 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3146 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3147 struct lu_name lname;
3150 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3153 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3154 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3155 ldata.ld_lee = (struct link_ea_entry *)
3156 ((char *)ldata.ld_lee +
3161 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3162 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3163 lname.ln_namelen, lname.ln_name,
3164 PFID(mdd_object_fid(newparent)));
3165 /* update to the new parent fid */
3166 linkea_entry_pack(ldata.ld_lee, &lname,
3167 mdd_object_fid(newparent));
3169 rc = mdd_declare_links_add(env, child, handle, &ldata);
3171 rc = mdd_links_write(env, child, &ldata, handle);
3177 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3178 struct mdd_object *parent,
3179 struct mdd_object *newparent,
3180 struct mdd_object *child,
3181 const char *name, int namelen,
3182 struct thandle *handle)
3184 return mdd_linkea_update_child_internal(env, parent, newparent,
3186 namelen, handle, true);
3189 static int mdd_linkea_update_child(const struct lu_env *env,
3190 struct mdd_object *parent,
3191 struct mdd_object *newparent,
3192 struct mdd_object *child,
3193 const char *name, int namelen,
3194 struct thandle *handle)
3196 return mdd_linkea_update_child_internal(env, parent, newparent,
3198 namelen, handle, false);
3201 static int mdd_update_linkea_internal(const struct lu_env *env,
3202 struct mdd_object *mdd_pobj,
3203 struct mdd_object *mdd_sobj,
3204 struct mdd_object *mdd_tobj,
3205 const struct lu_name *child_name,
3206 struct linkea_data *ldata,
3207 struct thandle *handle,
3210 struct mdd_thread_info *info = mdd_env_info(env);
3215 LASSERT(ldata->ld_buf != NULL);
3216 LASSERT(ldata->ld_leh != NULL);
3218 /* If it is mulitple links file, we need update the name entry for
3220 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3221 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3222 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3223 struct mdd_object *pobj;
3224 struct lu_name lname;
3227 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3229 pobj = mdd_object_find(env, mdd, &fid);
3231 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3232 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3237 if (!mdd_object_exists(pobj)) {
3238 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3239 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3243 if (pobj == mdd_pobj &&
3244 lname.ln_namelen == child_name->ln_namelen &&
3245 strncmp(lname.ln_name, child_name->ln_name,
3246 lname.ln_namelen) == 0) {
3247 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3248 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3253 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3254 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3255 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3258 /* Remove source name from source directory */
3259 /* Insert new fid with target name into target dir */
3260 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3265 rc = mdo_declare_index_insert(env, pobj,
3266 mdd_object_fid(mdd_tobj),
3267 mdd_object_type(mdd_tobj),
3268 lname.ln_name, handle);
3272 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3276 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3280 char *tmp_name = info->mti_key;
3282 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3283 /* lnamelen is too big(> NAME_MAX + 16),
3284 * something wrong about this linkea, let's
3286 CWARN("%s: the name %.*s is too long under "
3287 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3288 lname.ln_namelen, lname.ln_name,
3293 /* Note: lname might be without \0 at the end, see
3294 * linkea_entry_unpack(), let's add extra \0 by
3296 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3297 lname.ln_namelen, lname.ln_name);
3298 lname.ln_name = tmp_name;
3300 /* Let's check if this linkEA still valid, before
3301 * it might be packed into the RPC buffer. */
3302 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3303 &info->mti_fid, NULL);
3304 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3305 mdd_object_fid(mdd_sobj)))
3306 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3308 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3312 rc = __mdd_index_insert(env, pobj,
3313 mdd_object_fid(mdd_tobj),
3314 mdd_object_type(mdd_tobj),
3319 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3320 rc = mdo_ref_add(env, mdd_tobj, handle);
3321 mdd_write_unlock(env, mdd_tobj);
3325 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3326 mdo_ref_del(env, mdd_sobj, handle);
3327 mdd_write_unlock(env, mdd_sobj);
3330 mdd_object_put(env, pobj);
3334 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3341 static int mdd_migrate_xattrs(const struct lu_env *env,
3342 struct mdd_object *mdd_sobj,
3343 struct mdd_object *mdd_tobj)
3345 struct mdd_thread_info *info = mdd_env_info(env);
3346 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3348 struct thandle *handle;
3354 struct lu_buf list_xbuf;
3357 /* retrieve xattr list from the old object */
3358 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3359 if (list_xsize == -ENODATA)
3365 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3366 if (info->mti_big_buf.lb_buf == NULL)
3369 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3370 list_xbuf.lb_len = list_xsize;
3371 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3376 xname = list_xbuf.lb_buf;
3378 xlen = strnlen(xname, rem - 1) + 1;
3379 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3380 strcmp(XATTR_NAME_LMV, xname) == 0)
3383 /* For directory, if there are default layout, migrate here */
3384 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3385 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3388 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3389 if (xsize == -ENODATA)
3394 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3395 if (info->mti_link_buf.lb_buf == NULL)
3396 GOTO(out, rc = -ENOMEM);
3398 xbuf.lb_len = xsize;
3399 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3400 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3406 handle = mdd_trans_create(env, mdd);
3408 GOTO(out, rc = PTR_ERR(handle));
3410 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3413 GOTO(stop_trans, rc);
3414 /* Note: this transaction is part of migration, and it is not
3415 * the last step of migration, so we set th_local = 1 to avoid
3416 * update last rcvd for this transaction */
3417 handle->th_local = 1;
3418 rc = mdd_trans_start(env, mdd, handle);
3420 GOTO(stop_trans, rc);
3423 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3425 GOTO(stop_trans, rc = 0);
3427 if (unlikely(rc == -ENOSPC &&
3428 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3429 rc = linkea_overflow_shrink(
3430 (struct linkea_data *)(xbuf.lb_buf));
3431 if (likely(rc > 0)) {
3438 GOTO(stop_trans, rc);
3440 rc = mdd_trans_stop(env, mdd, rc, handle);
3445 memmove(xname, xname + xlen, rem);
3451 static int mdd_declare_migrate_create(const struct lu_env *env,
3452 struct mdd_object *mdd_pobj,
3453 struct mdd_object *mdd_sobj,
3454 struct mdd_object *mdd_tobj,
3455 struct md_op_spec *spec,
3457 union lmv_mds_md *mgr_ea,
3458 struct linkea_data *ldata,
3459 struct thandle *handle)
3461 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3462 const struct lu_buf *buf;
3466 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3467 handle, spec, NULL);
3471 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3476 if (S_ISLNK(la->la_mode)) {
3477 const char *target_name = spec->u.sp_symname;
3478 int sym_len = strlen(target_name);
3479 const struct lu_buf *buf;
3481 buf = mdd_buf_get_const(env, target_name, sym_len);
3482 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3486 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3487 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3492 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3493 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3494 spec->u.sp_ea.eadatalen);
3495 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3501 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3502 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3503 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3508 la_flag->la_valid = LA_FLAGS;
3509 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3510 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3515 static int mdd_migrate_create(const struct lu_env *env,
3516 struct mdd_object *mdd_pobj,
3517 struct mdd_object *mdd_sobj,
3518 struct mdd_object *mdd_tobj,
3519 const struct lu_name *lname,
3522 struct mdd_thread_info *info = mdd_env_info(env);
3523 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3524 struct md_op_spec *spec = &info->mti_spec;
3525 struct lu_buf lmm_buf = { NULL };
3526 struct lu_buf link_buf = { NULL };
3527 struct lu_buf mgr_buf;
3528 struct thandle *handle;
3529 struct lmv_mds_md_v1 *mgr_ea;
3530 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3531 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3533 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3537 /* prepare spec for create */
3538 memset(spec, 0, sizeof(*spec));
3539 spec->sp_cr_lookup = 0;
3540 spec->sp_feat = &dt_directory_features;
3541 if (S_ISLNK(la->la_mode)) {
3542 const struct lu_buf *buf;
3544 buf = lu_buf_check_and_alloc(
3545 &mdd_env_info(env)->mti_big_buf,
3548 link_buf.lb_len = la->la_size + 1;
3549 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3550 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3552 rc = rc != 0 ? rc : -EFAULT;
3553 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3554 mdd2obd_dev(mdd)->obd_name,
3555 PFID(mdd_object_fid(mdd_sobj)), rc);
3558 spec->u.sp_symname = link_buf.lb_buf;
3559 } else if (S_ISREG(la->la_mode)) {
3560 /* retrieve lov of the old object */
3561 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3562 if (rc != 0 && rc != -ENODATA)
3564 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3565 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3566 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3567 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3569 } else if (S_ISDIR(la->la_mode)) {
3570 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3571 if (rc == -ENODATA) {
3572 /* ignore the non-linkEA error */
3580 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3581 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3582 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3583 mgr_buf.lb_len = mgr_easize;
3584 mgr_ea = mgr_buf.lb_buf;
3585 memset(mgr_ea, 0, sizeof(*mgr_ea));
3586 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3587 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3588 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3589 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3590 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3591 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3593 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3595 handle = mdd_trans_create(env, mdd);
3597 GOTO(out_free, rc = PTR_ERR(handle));
3599 /* Note: this transaction is part of migration, and it is not
3600 * the last step of migration, so we set th_local = 1 to avoid
3601 * update last rcvd for this transaction */
3602 handle->th_local = 1;
3603 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3604 la, mgr_buf.lb_buf, ldata, handle);
3606 GOTO(stop_trans, rc);
3608 rc = mdd_trans_start(env, mdd, handle);
3610 GOTO(stop_trans, rc);
3612 /* don't set nlink from the original object */
3613 la->la_valid &= ~LA_NLINK;
3615 /* create the target object */
3616 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3619 GOTO(stop_trans, rc);
3621 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3622 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3624 GOTO(stop_trans, rc);
3627 /* Set MIGRATE EA on the source inode, so once the migration needs
3628 * to be re-done during failover, the re-do process can locate the
3629 * target object which is already being created. */
3630 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3632 GOTO(stop_trans, rc);
3634 /* Set immutable flag, so any modification is disabled until
3635 * the migration is done. Once the migration is interrupted,
3636 * if the resume process find the migrating object has both
3637 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3638 * flag and approve the migration */
3639 la_flag->la_valid = LA_FLAGS;
3640 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3641 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3644 rc = mdd_trans_stop(env, mdd, rc, handle);
3646 if (lmm_buf.lb_buf != NULL)
3647 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3651 static int mdd_migrate_entries(const struct lu_env *env,
3652 struct mdd_object *mdd_sobj,
3653 struct mdd_object *mdd_tobj)
3655 struct dt_object *next = mdd_object_child(mdd_sobj);
3656 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3657 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3658 struct thandle *handle;
3660 const struct dt_it_ops *iops;
3662 struct lu_dirent *ent;
3666 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3670 if (!dt_try_as_dir(env, next))
3671 GOTO(out_ent, rc = -ENOTDIR);
3673 * iterate directories
3675 iops = &next->do_index_ops->dio_it;
3676 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3678 GOTO(out_ent, rc = PTR_ERR(it));
3680 rc = iops->load(env, it, 0);
3682 rc = iops->next(env, it);
3686 * At this point and across for-loop:
3688 * rc == 0 -> ok, proceed.
3689 * rc > 0 -> end of directory.
3693 struct mdd_object *child;
3694 char *name = mdd_env_info(env)->mti_key;
3697 bool target_exist = false;
3699 len = iops->key_size(env, it);
3703 result = iops->rec(env, it, (struct dt_rec *)ent,
3704 LUDA_FID | LUDA_TYPE);
3705 if (result == -ESTALE)
3712 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3714 /* Insert new fid with target name into target dir */
3715 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3716 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3717 ent->lde_name[1] == '.'))
3720 child = mdd_object_find(env, mdd, &ent->lde_fid);
3722 GOTO(out, rc = PTR_ERR(child));
3724 mdd_write_lock(env, child, MOR_SRC_CHILD);
3725 is_dir = S_ISDIR(mdd_object_type(child));
3727 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3729 /* Check whether the name has been inserted to the target */
3730 if (dt_try_as_dir(env, dt_tobj)) {
3731 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3733 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3734 (struct dt_key *)name);
3735 if (unlikely(rc == 0))
3736 target_exist = true;
3739 handle = mdd_trans_create(env, mdd);
3741 GOTO(out_put, rc = PTR_ERR(handle));
3743 /* Note: this transaction is part of migration, and it is not
3744 * the last step of migration, so we set th_local = 1 to avoid
3745 * updating last rcvd for this transaction */
3746 handle->th_local = 1;
3747 if (likely(!target_exist)) {
3748 rc = mdo_declare_index_insert(env, mdd_tobj,
3750 mdd_object_type(child),
3756 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3762 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3767 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3771 /* Update .. for child */
3772 rc = mdo_declare_index_delete(env, child, dotdot,
3777 rc = mdo_declare_index_insert(env, child,
3778 mdd_object_fid(mdd_tobj),
3779 S_IFDIR, dotdot, handle);
3784 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3791 rc = mdd_trans_start(env, mdd, handle);
3793 CERROR("%s: transaction start failed: rc = %d\n",
3794 mdd2obd_dev(mdd)->obd_name, rc);
3798 if (likely(!target_exist)) {
3799 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3800 mdd_object_type(child),
3806 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3811 rc = __mdd_index_delete_only(env, child, dotdot,
3816 rc = __mdd_index_insert_only(env, child,
3817 mdd_object_fid(mdd_tobj), S_IFDIR,
3823 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3825 strlen(name), handle);
3828 mdd_write_unlock(env, child);
3829 mdd_object_put(env, child);
3830 rc = mdd_trans_stop(env, mdd, rc, handle);
3834 result = iops->next(env, it);
3835 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3836 GOTO(out, rc = -EINTR);
3838 if (result == -ESTALE)
3840 } while (result == 0);
3843 iops->fini(env, it);
3845 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3849 static int mdd_declare_update_linkea(const struct lu_env *env,
3850 struct mdd_object *mdd_pobj,
3851 struct mdd_object *mdd_sobj,
3852 struct mdd_object *mdd_tobj,
3853 const struct lu_name *child_name,
3854 struct linkea_data *ldata,
3855 struct thandle *handle)
3857 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3858 child_name, ldata, handle, 1);
3861 static int mdd_update_linkea(const struct lu_env *env,
3862 struct mdd_object *mdd_pobj,
3863 struct mdd_object *mdd_sobj,
3864 struct mdd_object *mdd_tobj,
3865 const struct lu_name *child_name,
3866 struct linkea_data *ldata,
3867 struct thandle *handle)
3869 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3870 child_name, ldata, handle, 0);
3873 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3874 struct mdd_object *mdd_pobj,
3875 struct mdd_object *mdd_sobj,
3876 struct mdd_object *mdd_tobj,
3877 const struct lu_name *lname,
3879 struct lu_attr *parent_la,
3880 struct linkea_data *ldata,
3881 struct thandle *handle)
3883 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3884 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3887 /* Revert IMMUTABLE flag */
3888 la_flag->la_valid = LA_FLAGS;
3889 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3890 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3894 /* delete entry from source dir */
3895 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3899 if (ldata->ld_buf != NULL) {
3900 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3901 mdd_tobj, lname, ldata, handle);
3906 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3907 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3912 handle->th_complex = 1;
3913 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3915 LU_XATTR_REPLACE, handle);
3920 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3921 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3927 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3928 mdd_object_type(mdd_tobj),
3929 lname->ln_name, handle);
3933 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
3937 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3938 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3943 /* delete old object */
3944 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3948 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3949 /* delete old object */
3950 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3953 /* set nlink to 0 */
3954 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3959 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3963 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3967 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3972 static int mdd_migrate_update_name(const struct lu_env *env,
3973 struct mdd_object *mdd_pobj,
3974 struct mdd_object *mdd_sobj,
3975 struct mdd_object *mdd_tobj,
3976 const struct lu_name *lname,
3979 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3980 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3981 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3982 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3983 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3984 struct thandle *handle;
3985 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3986 const char *name = lname->ln_name;
3990 /* update time for parent */
3991 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3992 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3993 p_la->la_valid = LA_CTIME;
3995 rc = mdd_la_get(env, mdd_sobj, so_attr);
3999 ldata->ld_buf = NULL;
4000 rc = mdd_links_read(env, mdd_sobj, ldata);
4001 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4004 handle = mdd_trans_create(env, mdd);
4006 RETURN(PTR_ERR(handle));
4008 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4009 lname, so_attr, p_la, ldata,
4012 /* If the migration can not be fit in one transaction, just
4013 * leave it in the original MDT */
4015 GOTO(stop_trans, rc = 0);
4017 GOTO(stop_trans, rc);
4020 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4021 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4022 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4023 PFID(mdd_object_fid(mdd_tobj)));
4025 rc = mdd_trans_start(env, mdd, handle);
4027 GOTO(stop_trans, rc);
4029 /* Revert IMMUTABLE flag */
4030 la_flag->la_valid = LA_FLAGS;
4031 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4032 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4034 GOTO(stop_trans, rc);
4036 /* Remove source name from source directory */
4037 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4039 GOTO(stop_trans, rc);
4041 if (ldata->ld_buf != NULL) {
4042 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4043 lname, ldata, handle);
4045 GOTO(stop_trans, rc);
4047 /* linkea update might decrease the source object
4048 * nlink, let's get the attr again after ref_del */
4049 rc = mdd_la_get(env, mdd_sobj, so_attr);
4051 GOTO(stop_trans, rc);
4054 if (S_ISREG(so_attr->la_mode)) {
4055 if (so_attr->la_nlink == 1) {
4056 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4058 if (rc != 0 && rc != -ENODATA)
4059 GOTO(stop_trans, rc);
4061 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4063 LU_XATTR_REPLACE, handle);
4065 GOTO(stop_trans, rc);
4069 /* Insert new fid with target name into target dir */
4070 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4071 mdd_object_type(mdd_tobj), name, handle);
4073 GOTO(stop_trans, rc);
4075 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4077 mdd_sobj->mod_flags |= DEAD_OBJ;
4078 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4080 GOTO(out_unlock, rc);
4082 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4084 GOTO(out_unlock, rc);
4086 mdo_ref_del(env, mdd_sobj, handle);
4088 mdo_ref_del(env, mdd_sobj, handle);
4090 /* Get the attr again after ref_del */
4091 rc = mdd_la_get(env, mdd_sobj, so_attr);
4093 GOTO(out_unlock, rc);
4095 ma->ma_attr = *so_attr;
4096 ma->ma_valid |= MA_INODE;
4098 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4100 GOTO(out_unlock, rc);
4102 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4103 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4104 mdo2fid(mdd_pobj), lname, lname, handle);
4106 CWARN("%s: changelog for migrate %s "DFID
4107 "under "DFID" failed: rc = %d\n",
4108 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4109 PFID(mdd_object_fid(mdd_sobj)),
4110 PFID(mdd_object_fid(mdd_pobj)), rc);
4111 /* Sigh, there are no easy way to migrate back the object, so
4112 * let's reset the result to 0 for now XXX */
4116 mdd_write_unlock(env, mdd_sobj);
4119 rc = mdd_trans_stop(env, mdd, rc, handle);
4124 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4125 const struct lu_fid *fid, __u32 *mdt_index)
4127 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4128 struct seq_server_site *ss;
4131 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4133 range->lsr_flags = LU_SEQ_RANGE_MDT;
4134 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4138 *mdt_index = range->lsr_index;
4143 * Check whether we should migrate the file/dir
4145 * < 0 permission check failed or other error.
4146 * = 0 the file can be migrated.
4147 * > 0 the file does not need to be migrated, mostly
4148 * for multiple link file
4150 static int mdd_migrate_sanity_check(const struct lu_env *env,
4151 struct mdd_object *pobj,
4152 const struct lu_attr *pattr,
4153 struct mdd_object *sobj,
4154 struct lu_attr *sattr)
4156 struct mdd_thread_info *info = mdd_env_info(env);
4157 struct linkea_data *ldata = &info->mti_link_data;
4158 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4160 struct lu_buf *mgr_buf;
4166 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4167 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4168 if (mgr_buf->lb_buf == NULL)
4171 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4173 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4175 /* If the object has migrateEA, it means IMMUTE flag
4176 * is being set by previous migration process, so it
4177 * needs to override the IMMUTE flag, otherwise the
4178 * following sanity check will fail */
4179 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4180 LMV_HASH_FLAG_MIGRATION) {
4181 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4183 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4184 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4185 mdd2obd_dev(mdd)->obd_name,
4186 PFID(mdd_object_fid(sobj)));
4190 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4191 sobj, sattr, NULL, NULL);
4195 /* Then it will check if the file should be migrated. If the file
4196 * has mulitple links, we only need migrate the file if all of its
4197 * entries has been migrated to the remote MDT */
4198 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4201 rc = mdd_links_read(env, sobj, ldata);
4203 /* For multiple links files, if there are no linkEA data at all,
4204 * means the file might be created before linkEA is enabled, and
4205 * all of its links should not be migrated yet, otherwise it
4206 * should have some linkEA there */
4207 if (rc == -ENOENT || rc == -ENODATA)
4212 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4213 /* If there are still links locally, then the file will not be
4215 LASSERT(ldata->ld_leh != NULL);
4217 /* If the linkEA is overflow, then means there are some unknown name
4218 * entries under unknown parents, that will prevent the migration. */
4219 if (unlikely(ldata->ld_leh->leh_overflow_time))
4222 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4223 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4224 struct lu_name lname;
4226 __u32 parent_mdt_index;
4228 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4230 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4233 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4237 /* Migrate the object only if none of its parents are on the
4239 if (parent_mdt_index != mdt_index)
4242 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4243 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4244 lname.ln_name, PFID(&fid));
4252 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4253 struct md_object *sobj, const struct lu_name *lname,
4254 struct md_object *tobj, struct md_attr *ma)
4256 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4257 struct mdd_device *mdd = mdo2mdd(pobj);
4258 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4259 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4260 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4261 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4262 bool created = false;
4266 /* If the file will being migrated, it will check whether
4267 * the file is being opened by someone else right now */
4268 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4269 if (mdd_sobj->mod_count > 0) {
4271 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4272 mdd2obd_dev(mdd)->obd_name,
4273 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4274 mdd_sobj->mod_count, -EBUSY);
4275 mdd_read_unlock(env, mdd_sobj);
4276 GOTO(put, rc = -EBUSY);
4278 mdd_read_unlock(env, mdd_sobj);
4280 rc = mdd_la_get(env, mdd_sobj, so_attr);
4284 rc = mdd_la_get(env, mdd_pobj, pattr);
4288 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4295 /* Sigh, it is impossible to finish all of migration in a single
4296 * transaction, for example migrating big directory entries to the
4297 * new MDT, it needs insert all of name entries of children in the
4300 * So migration will be done in multiple steps and transactions.
4302 * 1. create an orphan object on the remote MDT in one transaction.
4303 * 2. migrate extend attributes to the new target file/directory.
4304 * 3. For directory, migrate the entries to the new MDT and update
4305 * linkEA of each children. Because we can not migrate all entries
4306 * in a single transaction, so the migrating directory will become
4307 * a striped directory during migration, so once the process is
4308 * interrupted, the directory is still accessible. (During lookup,
4309 * client will locate the name by searching both original and target
4311 * 4. Finally, update the name/FID to point to the new file/directory
4312 * in a separate transaction.
4315 /* step 1: Check whether the orphan object has been created, and create
4316 * orphan object on the remote MDT if needed */
4317 if (!mdd_object_exists(mdd_tobj)) {
4318 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4325 LASSERT(mdd_object_exists(mdd_tobj));
4326 /* step 2: migrate xattr */
4327 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4331 /* step 3: migrate name entries to the orphan object */
4332 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4333 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4336 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4337 OBD_FAIL_MDS_REINT_NET_REP)))
4340 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4343 LASSERT(mdd_object_exists(mdd_tobj));
4344 /* step 4: update name entry to the new object */
4345 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4350 /* newly created target was not locked, don't cache its attributes */
4352 mdd_invalidate(env, tobj);
4357 const struct md_dir_operations mdd_dir_ops = {
4358 .mdo_is_subdir = mdd_is_subdir,
4359 .mdo_lookup = mdd_lookup,
4360 .mdo_create = mdd_create,
4361 .mdo_rename = mdd_rename,
4362 .mdo_link = mdd_link,
4363 .mdo_unlink = mdd_unlink,
4364 .mdo_create_data = mdd_create_data,
4365 .mdo_migrate = mdd_migrate,