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 = 0;
739 size_t hdr_size = sizeof(struct llog_changelog_rec) -
740 sizeof(struct changelog_rec);
745 if (uc != NULL && uc->uc_jobid[0] != '\0')
748 return llog_data_len(hdr_size + changelog_rec_offset(crf) +
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 /** Store a namespace change changelog record
874 * If this fails, we must fail the whole transaction; we don't
875 * want the change to commit without the log entry.
876 * \param target - mdd_object of change
877 * \param tpfid - target parent dir/object fid
878 * \param sfid - source object fid
879 * \param spfid - source parent fid
880 * \param tname - target name string
881 * \param sname - source name string
882 * \param handle - transaction handle
884 int mdd_changelog_ns_store(const struct lu_env *env,
885 struct mdd_device *mdd,
886 enum changelog_rec_type type,
887 enum changelog_rec_flags crf,
888 struct mdd_object *target,
889 const struct lu_fid *tpfid,
890 const struct lu_fid *sfid,
891 const struct lu_fid *spfid,
892 const struct lu_name *tname,
893 const struct lu_name *sname,
894 struct thandle *handle)
896 const struct lu_ucred *uc = lu_ucred(env);
897 struct llog_changelog_rec *rec;
904 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
907 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
910 LASSERT(tpfid != NULL);
911 LASSERT(tname != NULL);
912 LASSERT(handle != NULL);
914 reclen = mdd_llog_record_calc_size(env, tname, sname);
915 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
916 if (buf->lb_buf == NULL)
922 if (uc != NULL && uc->uc_jobid[0] != '\0')
930 rec->cr.cr_flags = crf;
931 rec->cr.cr_type = (__u32)type;
932 rec->cr.cr_pfid = *tpfid;
933 rec->cr.cr_namelen = tname->ln_namelen;
934 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
936 if (crf & CLF_RENAME)
937 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
940 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
942 if (likely(target != NULL)) {
943 rec->cr.cr_tfid = *mdo2fid(target);
944 target->mod_cltime = cfs_time_current_64();
946 fid_zero(&rec->cr.cr_tfid);
949 rc = mdd_changelog_store(env, mdd, rec, handle);
951 CERROR("%s: cannot store changelog record: type = %d, "
952 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
953 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
954 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
961 static int __mdd_links_add(const struct lu_env *env,
962 struct mdd_object *mdd_obj,
963 struct linkea_data *ldata,
964 const struct lu_name *lname,
965 const struct lu_fid *pfid,
966 int first, int check)
970 if (ldata->ld_leh == NULL) {
971 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
975 rc = linkea_data_new(ldata,
976 &mdd_env_info(env)->mti_link_buf);
983 rc = linkea_links_find(ldata, lname, pfid);
984 if (rc && rc != -ENOENT)
990 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
991 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
995 linkea_add_buf(ldata, lname, tfid);
998 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
999 linkea_add_buf(ldata, lname, pfid);
1001 return linkea_add_buf(ldata, lname, pfid);
1004 static int __mdd_links_del(const struct lu_env *env,
1005 struct mdd_object *mdd_obj,
1006 struct linkea_data *ldata,
1007 const struct lu_name *lname,
1008 const struct lu_fid *pfid)
1012 if (ldata->ld_leh == NULL) {
1013 rc = mdd_links_read(env, mdd_obj, ldata);
1018 rc = linkea_links_find(ldata, lname, pfid);
1022 linkea_del_buf(ldata, lname);
1026 static int mdd_linkea_prepare(const struct lu_env *env,
1027 struct mdd_object *mdd_obj,
1028 const struct lu_fid *oldpfid,
1029 const struct lu_name *oldlname,
1030 const struct lu_fid *newpfid,
1031 const struct lu_name *newlname,
1032 int first, int check,
1033 struct linkea_data *ldata)
1038 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1041 LASSERT(oldpfid != NULL || newpfid != NULL);
1043 if (mdd_obj->mod_flags & DEAD_OBJ)
1044 /* Unnecessary to update linkEA for dead object. */
1047 if (oldpfid != NULL) {
1048 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1050 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1053 /* No changes done. */
1058 /* If renaming, add the new record */
1059 if (newpfid != NULL)
1060 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1066 int mdd_links_rename(const struct lu_env *env,
1067 struct mdd_object *mdd_obj,
1068 const struct lu_fid *oldpfid,
1069 const struct lu_name *oldlname,
1070 const struct lu_fid *newpfid,
1071 const struct lu_name *newlname,
1072 struct thandle *handle,
1073 struct linkea_data *ldata,
1074 int first, int check)
1079 if (ldata == NULL) {
1080 ldata = &mdd_env_info(env)->mti_link_data;
1081 memset(ldata, 0, sizeof(*ldata));
1082 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1083 newpfid, newlname, first, check, ldata);
1088 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1089 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1095 if (newlname == NULL)
1096 CERROR("link_ea add failed %d "DFID"\n",
1097 rc, PFID(mdd_object_fid(mdd_obj)));
1098 else if (oldpfid == NULL)
1099 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1100 newlname->ln_namelen, newlname->ln_name, rc,
1101 PFID(mdd_object_fid(mdd_obj)));
1102 else if (newpfid == NULL)
1103 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1104 oldlname->ln_namelen, oldlname->ln_name, rc,
1105 PFID(mdd_object_fid(mdd_obj)));
1107 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1108 "\n", oldlname->ln_namelen, oldlname->ln_name,
1109 newlname->ln_namelen, newlname->ln_name, rc,
1110 PFID(mdd_object_fid(mdd_obj)));
1113 if (is_vmalloc_addr(ldata->ld_buf))
1114 /* if we vmalloced a large buffer drop it */
1115 lu_buf_free(ldata->ld_buf);
1120 static inline int mdd_links_add(const struct lu_env *env,
1121 struct mdd_object *mdd_obj,
1122 const struct lu_fid *pfid,
1123 const struct lu_name *lname,
1124 struct thandle *handle,
1125 struct linkea_data *ldata, int first)
1127 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1128 pfid, lname, handle, ldata, first, 0);
1131 static inline int mdd_links_del(const struct lu_env *env,
1132 struct mdd_object *mdd_obj,
1133 const struct lu_fid *pfid,
1134 const struct lu_name *lname,
1135 struct thandle *handle)
1137 return mdd_links_rename(env, mdd_obj, pfid, lname,
1138 NULL, NULL, handle, NULL, 0, 0);
1141 /** Read the link EA into a temp buffer.
1142 * Uses the name_buf since it is generally large.
1143 * \retval IS_ERR err
1144 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1146 struct lu_buf *mdd_links_get(const struct lu_env *env,
1147 struct mdd_object *mdd_obj)
1149 struct linkea_data ldata = { NULL };
1152 rc = mdd_links_read(env, mdd_obj, &ldata);
1153 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1156 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1157 struct linkea_data *ldata, struct thandle *handle)
1159 const struct lu_buf *buf;
1162 if (ldata == NULL || ldata->ld_buf == NULL ||
1163 ldata->ld_leh == NULL)
1166 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1170 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1171 ldata->ld_leh->leh_len);
1172 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1173 if (unlikely(rc == -ENOSPC)) {
1174 rc = linkea_overflow_shrink(ldata);
1182 static int mdd_declare_links_add(const struct lu_env *env,
1183 struct mdd_object *mdd_obj,
1184 struct thandle *handle,
1185 struct linkea_data *ldata)
1191 if (ldata != NULL && ldata->ld_leh != NULL) {
1192 ea_len = ldata->ld_leh->leh_len;
1193 linkea = ldata->ld_buf->lb_buf;
1195 ea_len = MAX_LINKEA_SIZE;
1199 rc = mdo_declare_xattr_set(env, mdd_obj,
1200 mdd_buf_get_const(env, linkea, ea_len),
1201 XATTR_NAME_LINK, 0, handle);
1206 static inline int mdd_declare_links_del(const struct lu_env *env,
1207 struct mdd_object *c,
1208 struct thandle *handle)
1212 /* For directory, the linkEA will be removed together
1213 * with the object. */
1214 if (!S_ISDIR(mdd_object_type(c)))
1215 rc = mdd_declare_links_add(env, c, handle, NULL);
1220 static int mdd_declare_link(const struct lu_env *env,
1221 struct mdd_device *mdd,
1222 struct mdd_object *p,
1223 struct mdd_object *c,
1224 const struct lu_name *name,
1225 struct thandle *handle,
1227 struct linkea_data *data)
1229 struct lu_fid tfid = *mdo2fid(c);
1232 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1233 tfid.f_oid = cfs_fail_val;
1235 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1236 name->ln_name, handle);
1240 rc = mdo_declare_ref_add(env, c, handle);
1244 la->la_valid = LA_CTIME | LA_MTIME;
1245 rc = mdo_declare_attr_set(env, p, la, handle);
1249 la->la_valid = LA_CTIME;
1250 rc = mdo_declare_attr_set(env, c, la, handle);
1254 rc = mdd_declare_links_add(env, c, handle, data);
1258 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1263 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1264 struct md_object *src_obj, const struct lu_name *lname,
1267 const char *name = lname->ln_name;
1268 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1269 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1270 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1271 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1272 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1273 struct mdd_device *mdd = mdo2mdd(src_obj);
1274 struct thandle *handle;
1275 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1276 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1280 rc = mdd_la_get(env, mdd_sobj, cattr);
1284 rc = mdd_la_get(env, mdd_tobj, tattr);
1289 * If we are using project inheritance, we only allow hard link
1290 * creation in our tree when the project IDs are the same;
1291 * otherwise the tree quota mechanism could be circumvented.
1293 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1294 (tattr->la_projid != cattr->la_projid))
1297 handle = mdd_trans_create(env, mdd);
1299 GOTO(out_pending, rc = PTR_ERR(handle));
1301 memset(ldata, 0, sizeof(*ldata));
1303 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1304 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1306 /* Note: even this function will change ldata, but it comes from
1307 * thread_info, which is completely temporary and only seen in
1308 * this function, so we do not need reset ldata once it fails.*/
1309 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1310 lname, 0, 0, ldata);
1314 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1319 rc = mdd_trans_start(env, mdd, handle);
1323 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1324 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1327 GOTO(out_unlock, rc);
1329 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1330 rc = mdo_ref_add(env, mdd_sobj, handle);
1332 GOTO(out_unlock, rc);
1335 *tfid = *mdo2fid(mdd_sobj);
1336 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1337 tfid->f_oid = cfs_fail_val;
1339 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1340 mdd_object_type(mdd_sobj), name, handle);
1342 mdo_ref_del(env, mdd_sobj, handle);
1343 GOTO(out_unlock, rc);
1346 la->la_valid = LA_CTIME | LA_MTIME;
1347 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1349 GOTO(out_unlock, rc);
1351 la->la_valid = LA_CTIME;
1352 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1354 /* Note: The failure of links_add should not cause the
1355 * link failure, so do not check return value. */
1356 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1357 lname, handle, ldata, 0);
1361 mdd_write_unlock(env, mdd_sobj);
1363 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1364 mdo2fid(mdd_tobj), NULL, NULL,
1365 lname, NULL, handle);
1367 rc = mdd_trans_stop(env, mdd, rc, handle);
1368 if (is_vmalloc_addr(ldata->ld_buf))
1369 /* if we vmalloced a large buffer drop it */
1370 lu_buf_free(ldata->ld_buf);
1375 static int mdd_mark_orphan_object(const struct lu_env *env,
1376 struct mdd_object *obj, struct thandle *handle,
1379 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1382 if (!S_ISDIR(mdd_object_type(obj)))
1385 attr->la_valid = LA_FLAGS;
1386 attr->la_flags = LUSTRE_ORPHAN_FL;
1389 rc = mdo_declare_attr_set(env, obj, attr, handle);
1391 rc = mdo_attr_set(env, obj, attr, handle);
1396 static int mdd_declare_finish_unlink(const struct lu_env *env,
1397 struct mdd_object *obj,
1398 struct thandle *handle)
1402 /* Sigh, we do not know if the unlink object will become orphan in
1403 * declare phase, but fortunately the flags here does not matter
1404 * in current declare implementation */
1405 rc = mdd_mark_orphan_object(env, obj, handle, true);
1409 rc = mdo_declare_destroy(env, obj, handle);
1413 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1417 return mdd_declare_links_del(env, obj, handle);
1420 /* caller should take a lock before calling */
1421 int mdd_finish_unlink(const struct lu_env *env,
1422 struct mdd_object *obj, struct md_attr *ma,
1423 const struct mdd_object *pobj,
1424 const struct lu_name *lname,
1428 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1431 LASSERT(mdd_write_locked(env, obj) != 0);
1433 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1434 /* add new orphan and the object
1435 * will be deleted during mdd_close() */
1436 obj->mod_flags |= DEAD_OBJ;
1437 if (obj->mod_count) {
1438 rc = __mdd_orphan_add(env, obj, th);
1440 CDEBUG(D_HA, "Object "DFID" is inserted into "
1441 "orphan list, open count = %d\n",
1442 PFID(mdd_object_fid(obj)),
1445 CERROR("Object "DFID" fail to be an orphan, "
1446 "open count = %d, maybe cause failed "
1448 PFID(mdd_object_fid(obj)),
1451 /* mark object as an orphan here, not
1452 * before __mdd_orphan_add() as racing
1453 * mdd_la_get() may propagate ORPHAN_OBJ
1454 * causing the asserition */
1455 rc = mdd_mark_orphan_object(env, obj, th, false);
1457 rc = mdo_destroy(env, obj, th);
1459 } else if (!is_dir) {
1460 /* old files may not have link ea; ignore errors */
1461 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1469 * has mdd_write_lock on cobj already, but not on pobj yet
1471 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1472 const struct lu_attr *pattr,
1473 struct mdd_object *cobj,
1474 const struct lu_attr *cattr)
1479 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1484 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1485 struct mdd_object *p, struct mdd_object *c,
1486 const struct lu_name *name, struct md_attr *ma,
1487 struct thandle *handle, int no_name, int is_dir)
1489 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1492 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1493 if (likely(no_name == 0)) {
1494 rc = mdo_declare_index_delete(env, p, name->ln_name,
1501 rc = mdo_declare_ref_del(env, p, handle);
1507 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1508 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1509 la->la_valid = LA_CTIME | LA_MTIME;
1510 rc = mdo_declare_attr_set(env, p, la, handle);
1515 rc = mdo_declare_ref_del(env, c, handle);
1519 rc = mdo_declare_ref_del(env, c, handle);
1523 la->la_valid = LA_CTIME;
1524 rc = mdo_declare_attr_set(env, c, la, handle);
1528 rc = mdd_declare_finish_unlink(env, c, handle);
1532 /* FIXME: need changelog for remove entry */
1533 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1540 * test if a file has an HSM archive
1541 * if HSM attributes are not found in ma update them from
1544 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1545 struct mdd_object *obj,
1550 if (!(ma->ma_valid & MA_HSM)) {
1551 /* no HSM MD provided, read xattr */
1552 struct lu_buf *hsm_buf;
1553 const size_t buflen = sizeof(struct hsm_attrs);
1556 hsm_buf = mdd_buf_get(env, NULL, 0);
1557 lu_buf_alloc(hsm_buf, buflen);
1558 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1559 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1560 lu_buf_free(hsm_buf);
1564 ma->ma_valid |= MA_HSM;
1566 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1572 * Delete name entry and the object.
1573 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1574 * does not exist for this object, and it could only happen during resending
1575 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1576 * is also needed in this case(needed by changelog), so we have to add another
1577 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1578 * the ENOENT failure should be able to be fixed by redo mechanism.
1580 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1581 struct md_object *cobj, const struct lu_name *lname,
1582 struct md_attr *ma, int no_name)
1584 const char *name = lname->ln_name;
1585 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1586 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1587 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1588 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1589 struct mdd_object *mdd_cobj = NULL;
1590 struct mdd_device *mdd = mdo2mdd(pobj);
1591 struct thandle *handle;
1592 int rc, is_dir = 0, cl_flags = 0;
1595 /* cobj == NULL means only delete name entry */
1596 if (likely(cobj != NULL)) {
1597 mdd_cobj = md2mdd_obj(cobj);
1598 if (mdd_object_exists(mdd_cobj) == 0)
1602 rc = mdd_la_get(env, mdd_pobj, pattr);
1606 if (likely(mdd_cobj != NULL)) {
1608 rc = mdd_la_get(env, mdd_cobj, cattr);
1612 is_dir = S_ISDIR(cattr->la_mode);
1613 /* search for an existing archive.
1614 * we should check ahead as the object
1615 * can be destroyed in this transaction */
1616 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1617 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1620 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1624 handle = mdd_trans_create(env, mdd);
1626 RETURN(PTR_ERR(handle));
1628 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1629 lname, ma, handle, no_name, is_dir);
1633 rc = mdd_trans_start(env, mdd, handle);
1637 if (likely(mdd_cobj != NULL))
1638 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1640 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1641 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1646 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1647 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1648 GOTO(cleanup, rc = 0);
1650 if (likely(mdd_cobj != NULL)) {
1651 rc = mdo_ref_del(env, mdd_cobj, handle);
1653 __mdd_index_insert_only(env, mdd_pobj,
1655 mdd_object_type(mdd_cobj),
1662 mdo_ref_del(env, mdd_cobj, handle);
1664 /* fetch updated nlink */
1665 rc = mdd_la_get(env, mdd_cobj, cattr);
1670 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1671 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1673 la->la_valid = LA_CTIME | LA_MTIME;
1674 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1678 /* Enough for only unlink the entry */
1679 if (unlikely(mdd_cobj == NULL))
1682 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1683 /* update ctime of an unlinked file only if it is still
1684 * opened or a link still exists */
1685 la->la_valid = LA_CTIME;
1686 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1691 /* XXX: this transfer to ma will be removed with LOD/OSP */
1692 ma->ma_attr = *cattr;
1693 ma->ma_valid |= MA_INODE;
1694 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1698 /* fetch updated nlink */
1699 rc = mdd_la_get(env, mdd_cobj, cattr);
1700 /* if object is removed then we can't get its attrs,
1702 if (rc == -ENOENT) {
1703 cattr->la_nlink = 0;
1707 if (cattr->la_nlink == 0) {
1708 ma->ma_attr = *cattr;
1709 ma->ma_valid |= MA_INODE;
1714 if (likely(mdd_cobj != NULL))
1715 mdd_write_unlock(env, mdd_cobj);
1718 if (cattr->la_nlink == 0)
1719 cl_flags |= CLF_UNLINK_LAST;
1721 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1723 rc = mdd_changelog_ns_store(env, mdd,
1724 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1725 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1730 rc = mdd_trans_stop(env, mdd, rc, handle);
1736 * The permission has been checked when obj created, no need check again.
1738 static int mdd_cd_sanity_check(const struct lu_env *env,
1739 struct mdd_object *obj)
1744 if (!obj || mdd_is_dead_obj(obj))
1750 static int mdd_create_data(const struct lu_env *env,
1751 struct md_object *pobj,
1752 struct md_object *cobj,
1753 const struct md_op_spec *spec,
1756 struct mdd_device *mdd = mdo2mdd(cobj);
1757 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1758 struct mdd_object *son = md2mdd_obj(cobj);
1759 struct thandle *handle;
1760 const struct lu_buf *buf;
1761 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1762 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1766 rc = mdd_cd_sanity_check(env, son);
1770 if (!md_should_create(spec->sp_cr_flags))
1774 * there are following use cases for this function:
1775 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1776 * striping can be specified or not
1779 rc = mdd_la_get(env, son, attr);
1783 /* calling ->ah_make_hint() is used to transfer information from parent */
1784 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1786 handle = mdd_trans_create(env, mdd);
1788 GOTO(out_free, rc = PTR_ERR(handle));
1791 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1792 * Should this be fixed?
1794 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1795 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1796 spec->sp_cr_flags, spec->no_create);
1798 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1799 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1800 spec->u.sp_ea.eadatalen);
1805 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1806 XATTR_NAME_LOV, 0, handle);
1810 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1814 rc = mdd_trans_start(env, mdd, handle);
1818 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1824 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1827 rc = mdd_trans_stop(env, mdd, rc, handle);
1833 static int mdd_declare_object_initialize(const struct lu_env *env,
1834 struct mdd_object *parent,
1835 struct mdd_object *child,
1836 const struct lu_attr *attr,
1837 struct thandle *handle)
1842 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1843 if (!S_ISDIR(attr->la_mode))
1846 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1851 rc = mdo_declare_ref_add(env, child, handle);
1855 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1861 static int mdd_object_initialize(const struct lu_env *env,
1862 const struct lu_fid *pfid,
1863 struct mdd_object *child,
1864 struct lu_attr *attr, struct thandle *handle,
1865 const struct md_op_spec *spec)
1870 if (S_ISDIR(attr->la_mode)) {
1871 /* Add "." and ".." for newly created dir */
1872 mdo_ref_add(env, child, handle);
1873 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1874 S_IFDIR, dot, handle);
1876 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1879 mdo_ref_del(env, child, handle);
1886 * This function checks whether it can create a file/dir under the
1887 * directory(@pobj). The directory(@pobj) is not being locked by
1890 * \param[in] env execution environment
1891 * \param[in] pobj the directory to create files
1892 * \param[in] pattr the attributes of the directory
1893 * \param[in] lname the name of the created file/dir
1894 * \param[in] cattr the attributes of the file/dir
1895 * \param[in] spec create specification
1897 * \retval = 0 it is allowed to create file/dir under
1899 * \retval negative error not allowed to create file/dir
1900 * under the directory
1902 static int mdd_create_sanity_check(const struct lu_env *env,
1903 struct md_object *pobj,
1904 const struct lu_attr *pattr,
1905 const struct lu_name *lname,
1906 struct lu_attr *cattr,
1907 struct md_op_spec *spec)
1909 struct mdd_thread_info *info = mdd_env_info(env);
1910 struct lu_fid *fid = &info->mti_fid;
1911 struct mdd_object *obj = md2mdd_obj(pobj);
1912 struct mdd_device *m = mdo2mdd(pobj);
1913 bool check_perm = true;
1918 if (mdd_is_dead_obj(obj))
1922 * In some cases this lookup is not needed - we know before if name
1923 * exists or not because MDT performs lookup for it.
1924 * name length check is done in lookup.
1926 if (spec->sp_cr_lookup) {
1928 * Check if the name already exist, though it will be checked in
1929 * _index_insert also, for avoiding rolling back if exists
1932 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1933 MAY_WRITE | MAY_EXEC);
1935 RETURN(rc ? : -EEXIST);
1937 /* Permission is already being checked in mdd_lookup */
1941 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1946 if (pattr->la_mode & S_ISGID) {
1947 cattr->la_gid = pattr->la_gid;
1948 if (S_ISDIR(cattr->la_mode)) {
1949 cattr->la_mode |= S_ISGID;
1950 cattr->la_valid |= LA_MODE;
1954 /* Inherit project ID from parent directory */
1955 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
1956 cattr->la_projid = pattr->la_projid;
1957 if (S_ISDIR(cattr->la_mode)) {
1958 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
1959 cattr->la_valid |= LA_FLAGS;
1961 cattr->la_valid |= LA_PROJID;
1964 rc = mdd_name_check(m, lname);
1968 switch (cattr->la_mode & S_IFMT) {
1970 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1972 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
1973 RETURN(-ENAMETOOLONG);
1992 static int mdd_declare_create_object(const struct lu_env *env,
1993 struct mdd_device *mdd,
1994 struct mdd_object *p, struct mdd_object *c,
1995 struct lu_attr *attr,
1996 struct thandle *handle,
1997 const struct md_op_spec *spec,
1998 struct lu_buf *def_acl_buf,
1999 struct lu_buf *acl_buf,
2000 struct dt_allocation_hint *hint)
2002 const struct lu_buf *buf;
2005 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2010 #ifdef CONFIG_FS_POSIX_ACL
2011 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2012 /* if dir, then can inherit default ACl */
2013 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2014 XATTR_NAME_ACL_DEFAULT,
2020 if (acl_buf->lb_len > 0) {
2021 rc = mdo_declare_attr_set(env, c, attr, handle);
2025 rc = mdo_declare_xattr_set(env, c, acl_buf,
2026 XATTR_NAME_ACL_ACCESS, 0, handle);
2031 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2035 /* replay case, create LOV EA from client data */
2036 if (spec->no_create ||
2037 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2038 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2039 spec->u.sp_ea.eadatalen);
2040 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2046 if (S_ISLNK(attr->la_mode)) {
2047 const char *target_name = spec->u.sp_symname;
2048 int sym_len = strlen(target_name);
2049 const struct lu_buf *buf;
2051 buf = mdd_buf_get_const(env, target_name, sym_len);
2052 rc = dt_declare_record_write(env, mdd_object_child(c),
2058 if (spec->sp_cr_file_secctx_name != NULL) {
2059 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2060 spec->sp_cr_file_secctx_size);
2061 rc = mdo_declare_xattr_set(env, c, buf,
2062 spec->sp_cr_file_secctx_name, 0,
2071 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2072 struct mdd_object *p, struct mdd_object *c,
2073 const struct lu_name *name,
2074 struct lu_attr *attr,
2075 struct thandle *handle,
2076 const struct md_op_spec *spec,
2077 struct linkea_data *ldata,
2078 struct lu_buf *def_acl_buf,
2079 struct lu_buf *acl_buf,
2080 struct dt_allocation_hint *hint)
2084 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2085 def_acl_buf, acl_buf, hint);
2089 if (S_ISDIR(attr->la_mode)) {
2090 rc = mdo_declare_ref_add(env, p, handle);
2095 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2096 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2100 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2102 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2103 name->ln_name, handle);
2107 rc = mdd_declare_links_add(env, c, handle, ldata);
2112 la->la_valid = LA_CTIME | LA_MTIME;
2113 rc = mdo_declare_attr_set(env, p, la, handle);
2117 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2125 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2126 struct lu_attr *la, struct lu_buf *def_acl_buf,
2127 struct lu_buf *acl_buf)
2132 if (S_ISLNK(la->la_mode)) {
2133 acl_buf->lb_len = 0;
2134 def_acl_buf->lb_len = 0;
2138 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2139 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2140 XATTR_NAME_ACL_DEFAULT);
2141 mdd_read_unlock(env, pobj);
2143 /* If there are default ACL, fix mode/ACL by default ACL */
2144 def_acl_buf->lb_len = rc;
2145 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2146 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2147 acl_buf->lb_len = rc;
2148 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2151 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2152 /* If there are no default ACL, fix mode by mask */
2153 struct lu_ucred *uc = lu_ucred(env);
2155 /* The create triggered by MDT internal events, such as
2156 * LFSCK reset, will not contain valid "uc". */
2157 if (unlikely(uc != NULL))
2158 la->la_mode &= ~uc->uc_umask;
2160 acl_buf->lb_len = 0;
2161 def_acl_buf->lb_len = 0;
2168 * Create a metadata object and initialize it, set acl, xattr.
2170 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2171 struct mdd_object *son, struct lu_attr *attr,
2172 struct md_op_spec *spec, struct lu_buf *acl_buf,
2173 struct lu_buf *def_acl_buf,
2174 struct dt_allocation_hint *hint,
2175 struct thandle *handle)
2177 const struct lu_buf *buf;
2180 mdd_write_lock(env, son, MOR_TGT_CHILD);
2181 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2186 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2187 * created in declare phase, they also needs to be added to master
2188 * object as sub-directory entry. So it has to initialize the master
2189 * object, then set dir striped EA.(in mdo_xattr_set) */
2190 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2193 GOTO(err_destroy, rc);
2196 * in case of replay we just set LOVEA provided by the client
2197 * XXX: I think it would be interesting to try "old" way where
2198 * MDT calls this xattr_set(LOV) in a different transaction.
2199 * probably this way we code can be made better.
2202 /* During creation, there are only a few cases we need do xattr_set to
2204 * 1. regular file: see comments above.
2205 * 2. dir: inherit default striping or pool settings from parent.
2206 * 3. create striped directory with provided stripeEA.
2207 * 4. create striped directory because inherit default layout from the
2210 if (spec->no_create ||
2211 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2212 S_ISDIR(attr->la_mode)) {
2213 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2214 spec->u.sp_ea.eadatalen);
2215 rc = mdo_xattr_set(env, son, buf,
2216 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2220 GOTO(err_destroy, rc);
2223 #ifdef CONFIG_FS_POSIX_ACL
2224 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2225 S_ISDIR(attr->la_mode)) {
2226 /* set default acl */
2227 rc = mdo_xattr_set(env, son, def_acl_buf,
2228 XATTR_NAME_ACL_DEFAULT, 0,
2231 GOTO(err_destroy, rc);
2233 /* set its own acl */
2234 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2235 rc = mdo_xattr_set(env, son, acl_buf,
2236 XATTR_NAME_ACL_ACCESS,
2239 GOTO(err_destroy, rc);
2243 if (S_ISLNK(attr->la_mode)) {
2244 struct lu_ucred *uc = lu_ucred_assert(env);
2245 struct dt_object *dt = mdd_object_child(son);
2246 const char *target_name = spec->u.sp_symname;
2247 int sym_len = strlen(target_name);
2250 buf = mdd_buf_get_const(env, target_name, sym_len);
2251 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2253 CFS_CAP_SYS_RESOURCE_MASK);
2258 GOTO(err_initlized, rc = -EFAULT);
2261 if (spec->sp_cr_file_secctx_name != NULL) {
2262 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2263 spec->sp_cr_file_secctx_size);
2264 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2267 GOTO(err_initlized, rc);
2271 if (unlikely(rc != 0)) {
2273 if (S_ISDIR(attr->la_mode)) {
2274 /* Drop the reference, no need to delete "."/"..",
2275 * because the object to be destroied directly. */
2276 rc2 = mdo_ref_del(env, son, handle);
2280 rc2 = mdo_ref_del(env, son, handle);
2284 mdo_destroy(env, son, handle);
2287 mdd_write_unlock(env, son);
2291 static int mdd_index_delete(const struct lu_env *env,
2292 struct mdd_object *mdd_pobj,
2293 struct lu_attr *cattr,
2294 const struct lu_name *lname)
2296 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2297 struct thandle *handle;
2301 handle = mdd_trans_create(env, mdd);
2303 RETURN(PTR_ERR(handle));
2305 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2310 if (S_ISDIR(cattr->la_mode)) {
2311 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2316 /* Since this will only be used in the error handler path,
2317 * Let's set the thandle to be local and not mess the transno */
2318 handle->th_local = 1;
2319 rc = mdd_trans_start(env, mdd, handle);
2323 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2324 S_ISDIR(cattr->la_mode), handle);
2328 rc = mdd_trans_stop(env, mdd, rc, handle);
2334 * Create object and insert it into namespace.
2336 * Two operations have to be performed:
2338 * - an allocation of a new object (->do_create()), and
2339 * - an insertion into a parent index (->dio_insert()).
2341 * Due to locking, operation order is not important, when both are
2342 * successful, *but* error handling cases are quite different:
2344 * - if insertion is done first, and following object creation fails,
2345 * insertion has to be rolled back, but this operation might fail
2346 * also leaving us with dangling index entry.
2348 * - if creation is done first, is has to be undone if insertion fails,
2349 * leaving us with leaked space, which is not good but not fatal.
2351 * It seems that creation-first is simplest solution, but it is sub-optimal
2357 * case, because second mkdir is bound to create object, only to
2358 * destroy it immediately.
2360 * To avoid this follow local file systems that do double lookup:
2362 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2363 * 1. create (mdd_create_object_internal())
2364 * 2. insert (__mdd_index_insert(), lookup again)
2366 * \param[in] pobj parent object
2367 * \param[in] lname name of child being created
2368 * \param[in,out] child child object being created
2369 * \param[in] spec additional create parameters
2370 * \param[in] ma attributes for new child object
2372 * \retval 0 on success
2373 * \retval negative errno on failure
2375 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2376 const struct lu_name *lname, struct md_object *child,
2377 struct md_op_spec *spec, struct md_attr *ma)
2379 struct mdd_thread_info *info = mdd_env_info(env);
2380 struct lu_attr *la = &info->mti_la_for_fix;
2381 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2382 struct mdd_object *son = md2mdd_obj(child);
2383 struct mdd_device *mdd = mdo2mdd(pobj);
2384 struct lu_attr *attr = &ma->ma_attr;
2385 struct thandle *handle;
2386 struct lu_attr *pattr = &info->mti_pattr;
2387 struct lu_buf acl_buf;
2388 struct lu_buf def_acl_buf;
2389 struct linkea_data *ldata = &info->mti_link_data;
2390 const char *name = lname->ln_name;
2391 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2396 rc = mdd_la_get(env, mdd_pobj, pattr);
2400 /* Sanity checks before big job. */
2401 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2405 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2406 GOTO(out_free, rc = -EINPROGRESS);
2408 handle = mdd_trans_create(env, mdd);
2410 GOTO(out_free, rc = PTR_ERR(handle));
2412 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2413 mdd->mdd_dt_conf.ddp_max_ea_size);
2414 acl_buf = info->mti_xattr_buf;
2415 def_acl_buf.lb_buf = info->mti_key;
2416 def_acl_buf.lb_len = sizeof(info->mti_key);
2417 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2421 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2423 memset(ldata, 0, sizeof(*ldata));
2424 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2425 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2428 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2429 &tfid, lname, 1, 0, ldata);
2431 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2432 mdd_object_fid(mdd_pobj),
2433 lname, 1, 0, ldata);
2436 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2437 handle, spec, ldata, &def_acl_buf, &acl_buf,
2442 rc = mdd_trans_start(env, mdd, handle);
2446 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2447 &def_acl_buf, hint, handle);
2451 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2452 mdd_write_lock(env, son, MOR_TGT_CHILD);
2453 son->mod_flags |= VOLATILE_OBJ;
2454 rc = __mdd_orphan_add(env, son, handle);
2455 GOTO(out_volatile, rc);
2457 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2458 attr->la_mode, name, handle);
2460 GOTO(err_created, rc);
2462 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2465 /* update parent directory mtime/ctime */
2467 la->la_valid = LA_CTIME | LA_MTIME;
2468 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2470 GOTO(err_insert, rc);
2476 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2477 rc2 = __mdd_orphan_del(env, son, handle);
2479 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2480 S_ISDIR(attr->la_mode),
2486 mdd_write_lock(env, son, MOR_TGT_CHILD);
2487 if (S_ISDIR(attr->la_mode)) {
2488 /* Drop the reference, no need to delete "."/"..",
2489 * because the object is to be destroyed directly. */
2490 rc2 = mdo_ref_del(env, son, handle);
2492 mdd_write_unlock(env, son);
2497 /* For volatile files drop one link immediately, since there is
2498 * no filename in the namespace, and save any error returned. */
2499 rc2 = mdo_ref_del(env, son, handle);
2501 mdd_write_unlock(env, son);
2502 if (unlikely(rc == 0))
2507 /* Don't destroy the volatile object on success */
2508 if (likely(rc != 0))
2509 mdo_destroy(env, son, handle);
2510 mdd_write_unlock(env, son);
2513 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2514 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2515 rc = mdd_changelog_ns_store(env, mdd,
2516 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2517 S_ISREG(attr->la_mode) ? CL_CREATE :
2518 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2519 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2522 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2524 /* If creation fails, it is most likely due to the remote update
2525 * failure, because local transaction will mostly succeed at
2526 * this stage. There is no easy way to rollback all of previous
2527 * updates, so let's remove the object from namespace, and
2528 * LFSCK should handle the orphan object. */
2529 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2530 mdd_index_delete(env, mdd_pobj, attr, lname);
2534 if (is_vmalloc_addr(ldata->ld_buf))
2535 /* if we vmalloced a large buffer drop it */
2536 lu_buf_free(ldata->ld_buf);
2538 /* The child object shouldn't be cached anymore */
2540 set_bit(LU_OBJECT_HEARD_BANSHEE,
2541 &child->mo_lu.lo_header->loh_flags);
2546 * Get locks on parents in proper order
2547 * RETURN: < 0 - error, rename_order if successful
2555 static int mdd_rename_order(const struct lu_env *env,
2556 struct mdd_device *mdd,
2557 struct mdd_object *src_pobj,
2558 const struct lu_attr *pattr,
2559 struct mdd_object *tgt_pobj)
2561 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2565 if (src_pobj == tgt_pobj)
2566 RETURN(MDD_RN_SAME);
2568 /* compared the parent child relationship of src_p&tgt_p */
2569 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2571 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2574 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2588 /* has not mdd_write{read}_lock on any obj yet. */
2589 static int mdd_rename_sanity_check(const struct lu_env *env,
2590 struct mdd_object *src_pobj,
2591 const struct lu_attr *pattr,
2592 struct mdd_object *tgt_pobj,
2593 const struct lu_attr *tpattr,
2594 struct mdd_object *sobj,
2595 const struct lu_attr *cattr,
2596 struct mdd_object *tobj,
2597 const struct lu_attr *tattr)
2602 /* XXX: when get here, sobj must NOT be NULL,
2603 * the other case has been processed in cld_rename
2604 * before mdd_rename and enable MDS_PERM_BYPASS. */
2608 * If we are using project inheritance, we only allow renames
2609 * into our tree when the project IDs are the same; otherwise
2610 * tree quota mechanism would be circumvented.
2612 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2613 tpattr->la_projid != cattr->la_projid) ||
2614 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2615 (pattr->la_projid != tpattr->la_projid)))
2618 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2622 /* XXX: when get here, "tobj == NULL" means tobj must
2623 * NOT exist (neither on remote MDS, such case has been
2624 * processed in cld_rename before mdd_rename and enable
2626 * So check may_create, but not check may_unlink. */
2628 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2629 (src_pobj != tgt_pobj));
2631 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2632 (src_pobj != tgt_pobj), 1);
2634 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2635 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2640 static int mdd_declare_rename(const struct lu_env *env,
2641 struct mdd_device *mdd,
2642 struct mdd_object *mdd_spobj,
2643 struct mdd_object *mdd_tpobj,
2644 struct mdd_object *mdd_sobj,
2645 struct mdd_object *mdd_tobj,
2646 const struct lu_name *sname,
2647 const struct lu_name *tname,
2649 struct linkea_data *ldata,
2650 struct thandle *handle)
2652 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2655 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2656 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2662 /* name from source dir */
2663 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2667 /* .. from source child */
2668 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2669 /* source child can be directory,
2670 * counted by source dir's nlink */
2671 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2674 if (mdd_spobj != mdd_tpobj) {
2675 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2680 rc = mdo_declare_index_insert(env, mdd_sobj,
2682 S_IFDIR, dotdot, handle);
2687 /* new target child can be directory,
2688 * counted by target dir's nlink */
2689 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2694 la->la_valid = LA_CTIME | LA_MTIME;
2695 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2699 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2703 la->la_valid = LA_CTIME;
2704 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2708 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2713 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2714 mdd_object_type(mdd_sobj),
2715 tname->ln_name, handle);
2719 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2720 /* delete target child in target parent directory */
2721 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2726 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2730 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2731 /* target child can be directory,
2732 * delete "." reference in target child directory */
2733 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2737 /* delete ".." reference in target parent directory */
2738 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2743 la->la_valid = LA_CTIME;
2744 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2748 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2753 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2760 /* src object can be remote that is why we use only fid and type of object */
2761 static int mdd_rename(const struct lu_env *env,
2762 struct md_object *src_pobj, struct md_object *tgt_pobj,
2763 const struct lu_fid *lf, const struct lu_name *lsname,
2764 struct md_object *tobj, const struct lu_name *ltname,
2767 const char *sname = lsname->ln_name;
2768 const char *tname = ltname->ln_name;
2769 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2770 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2771 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2772 struct mdd_device *mdd = mdo2mdd(src_pobj);
2773 struct mdd_object *mdd_sobj = NULL; /* source object */
2774 struct mdd_object *mdd_tobj = NULL;
2775 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2776 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2777 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2778 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2779 struct thandle *handle;
2780 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2781 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2782 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2785 bool tobj_locked = 0;
2786 unsigned cl_flags = 0;
2791 mdd_tobj = md2mdd_obj(tobj);
2793 mdd_sobj = mdd_object_find(env, mdd, lf);
2794 if (IS_ERR(mdd_sobj))
2795 RETURN(PTR_ERR(mdd_sobj));
2797 rc = mdd_la_get(env, mdd_sobj, cattr);
2799 GOTO(out_pending, rc);
2801 rc = mdd_la_get(env, mdd_spobj, pattr);
2803 GOTO(out_pending, rc);
2806 rc = mdd_la_get(env, mdd_tobj, tattr);
2808 GOTO(out_pending, rc);
2809 /* search for an existing archive.
2810 * we should check ahead as the object
2811 * can be destroyed in this transaction */
2812 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2813 cl_flags |= CLF_RENAME_LAST_EXISTS;
2816 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2818 GOTO(out_pending, rc);
2820 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2821 mdd_sobj, cattr, mdd_tobj, tattr);
2823 GOTO(out_pending, rc);
2825 rc = mdd_name_check(mdd, ltname);
2827 GOTO(out_pending, rc);
2829 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2830 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2832 GOTO(out_pending, rc);
2834 handle = mdd_trans_create(env, mdd);
2836 GOTO(out_pending, rc = PTR_ERR(handle));
2838 memset(ldata, 0, sizeof(*ldata));
2839 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
2840 lsname, mdd_object_fid(mdd_tpobj), ltname,
2845 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2846 mdd_tobj, lsname, ltname, ma, ldata, handle);
2850 rc = mdd_trans_start(env, mdd, handle);
2854 is_dir = S_ISDIR(cattr->la_mode);
2856 /* Remove source name from source directory */
2857 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2861 /* "mv dir1 dir2" needs "dir1/.." link update */
2862 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2863 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2865 GOTO(fixup_spobj2, rc);
2867 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2870 GOTO(fixup_spobj, rc);
2873 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2874 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2876 /* tname might been renamed to something else */
2877 GOTO(fixup_spobj, rc);
2880 /* Insert new fid with target name into target dir */
2881 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2884 GOTO(fixup_tpobj, rc);
2886 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2887 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2889 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2890 la->la_valid = LA_CTIME;
2891 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2893 GOTO(fixup_tpobj, rc);
2895 /* Update the linkEA for the source object */
2896 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2897 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2898 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2901 /* Old files might not have EA entry */
2902 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2903 lsname, handle, NULL, 0);
2904 mdd_write_unlock(env, mdd_sobj);
2905 /* We don't fail the transaction if the link ea can't be
2906 updated -- fid2path will use alternate lookup method. */
2909 /* Remove old target object
2910 * For tobj is remote case cmm layer has processed
2911 * and set tobj to NULL then. So when tobj is NOT NULL,
2912 * it must be local one.
2914 if (tobj && mdd_object_exists(mdd_tobj)) {
2915 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2917 if (mdd_is_dead_obj(mdd_tobj)) {
2918 /* shld not be dead, something is wrong */
2919 CERROR("tobj is dead, something is wrong\n");
2923 mdo_ref_del(env, mdd_tobj, handle);
2925 /* Remove dot reference. */
2926 if (S_ISDIR(tattr->la_mode))
2927 mdo_ref_del(env, mdd_tobj, handle);
2930 /* fetch updated nlink */
2931 rc = mdd_la_get(env, mdd_tobj, tattr);
2933 CERROR("%s: Failed to get nlink for tobj "
2935 mdd2obd_dev(mdd)->obd_name,
2936 PFID(tpobj_fid), rc);
2937 GOTO(fixup_tpobj, rc);
2940 la->la_valid = LA_CTIME;
2941 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2943 CERROR("%s: Failed to set ctime for tobj "
2945 mdd2obd_dev(mdd)->obd_name,
2946 PFID(tpobj_fid), rc);
2947 GOTO(fixup_tpobj, rc);
2950 /* XXX: this transfer to ma will be removed with LOD/OSP */
2951 ma->ma_attr = *tattr;
2952 ma->ma_valid |= MA_INODE;
2953 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2956 CERROR("%s: Failed to unlink tobj "
2958 mdd2obd_dev(mdd)->obd_name,
2959 PFID(tpobj_fid), rc);
2960 GOTO(fixup_tpobj, rc);
2963 /* fetch updated nlink */
2964 rc = mdd_la_get(env, mdd_tobj, tattr);
2965 if (rc == -ENOENT) {
2966 /* the object got removed, let's
2967 * return the latest known attributes */
2968 tattr->la_nlink = 0;
2970 } else if (rc != 0) {
2971 CERROR("%s: Failed to get nlink for tobj "
2973 mdd2obd_dev(mdd)->obd_name,
2974 PFID(tpobj_fid), rc);
2975 GOTO(fixup_tpobj, rc);
2977 /* XXX: this transfer to ma will be removed with LOD/OSP */
2978 ma->ma_attr = *tattr;
2979 ma->ma_valid |= MA_INODE;
2981 if (tattr->la_nlink == 0)
2982 cl_flags |= CLF_RENAME_LAST;
2984 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
2987 la->la_valid = LA_CTIME | LA_MTIME;
2988 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
2990 GOTO(fixup_tpobj, rc);
2992 if (mdd_spobj != mdd_tpobj) {
2993 la->la_valid = LA_CTIME | LA_MTIME;
2994 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
2996 GOTO(fixup_tpobj, rc);
3003 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3005 CWARN("tp obj fix error %d\n",rc2);
3007 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3008 !mdd_is_dead_obj(mdd_tobj)) {
3010 mdo_ref_add(env, mdd_tobj, handle);
3012 mdo_ref_add(env, mdd_tobj, handle);
3015 rc2 = __mdd_index_insert(env, mdd_tpobj,
3017 mdd_object_type(mdd_tobj),
3020 CWARN("tp obj fix error: rc = %d\n", rc2);
3025 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3026 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3028 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3029 mdd2obd_dev(mdd)->obd_name, rc2);
3032 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3035 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3036 mdd2obd_dev(mdd)->obd_name, rc2);
3041 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3042 mdd_object_type(mdd_sobj), sname,
3045 CWARN("sp obj fix error: rc = %d\n", rc2);
3050 mdd_write_unlock(env, mdd_tobj);
3053 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3054 mdd_tobj, tpobj_fid, lf, spobj_fid,
3055 ltname, lsname, handle);
3058 rc = mdd_trans_stop(env, mdd, rc, handle);
3061 mdd_object_put(env, mdd_sobj);
3066 * During migration once the parent FID has been changed,
3067 * we need update the parent FID in linkea.
3069 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3070 struct mdd_object *parent,
3071 struct mdd_object *newparent,
3072 struct mdd_object *child,
3073 const char *name, int namelen,
3074 struct thandle *handle,
3077 struct mdd_thread_info *info = mdd_env_info(env);
3078 struct linkea_data ldata = { NULL };
3079 struct lu_buf *buf = &info->mti_link_buf;
3085 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3086 if (buf->lb_buf == NULL)
3090 rc = mdd_links_read(env, child, &ldata);
3092 if (rc == -ENOENT || rc == -ENODATA)
3097 LASSERT(ldata.ld_leh != NULL);
3098 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3099 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3100 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3101 struct lu_name lname;
3104 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3107 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3108 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3109 ldata.ld_lee = (struct link_ea_entry *)
3110 ((char *)ldata.ld_lee +
3115 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3116 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3117 lname.ln_namelen, lname.ln_name,
3118 PFID(mdd_object_fid(newparent)));
3119 /* update to the new parent fid */
3120 linkea_entry_pack(ldata.ld_lee, &lname,
3121 mdd_object_fid(newparent));
3123 rc = mdd_declare_links_add(env, child, handle, &ldata);
3125 rc = mdd_links_write(env, child, &ldata, handle);
3131 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3132 struct mdd_object *parent,
3133 struct mdd_object *newparent,
3134 struct mdd_object *child,
3135 const char *name, int namelen,
3136 struct thandle *handle)
3138 return mdd_linkea_update_child_internal(env, parent, newparent,
3140 namelen, handle, true);
3143 static int mdd_linkea_update_child(const struct lu_env *env,
3144 struct mdd_object *parent,
3145 struct mdd_object *newparent,
3146 struct mdd_object *child,
3147 const char *name, int namelen,
3148 struct thandle *handle)
3150 return mdd_linkea_update_child_internal(env, parent, newparent,
3152 namelen, handle, false);
3155 static int mdd_update_linkea_internal(const struct lu_env *env,
3156 struct mdd_object *mdd_pobj,
3157 struct mdd_object *mdd_sobj,
3158 struct mdd_object *mdd_tobj,
3159 const struct lu_name *child_name,
3160 struct linkea_data *ldata,
3161 struct thandle *handle,
3164 struct mdd_thread_info *info = mdd_env_info(env);
3169 LASSERT(ldata->ld_buf != NULL);
3170 LASSERT(ldata->ld_leh != NULL);
3172 /* If it is mulitple links file, we need update the name entry for
3174 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3175 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3176 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3177 struct mdd_object *pobj;
3178 struct lu_name lname;
3181 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3183 pobj = mdd_object_find(env, mdd, &fid);
3185 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3186 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3191 if (!mdd_object_exists(pobj)) {
3192 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3193 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3197 if (pobj == mdd_pobj &&
3198 lname.ln_namelen == child_name->ln_namelen &&
3199 strncmp(lname.ln_name, child_name->ln_name,
3200 lname.ln_namelen) == 0) {
3201 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3202 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3207 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3208 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3209 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3212 /* Remove source name from source directory */
3213 /* Insert new fid with target name into target dir */
3214 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3219 rc = mdo_declare_index_insert(env, pobj,
3220 mdd_object_fid(mdd_tobj),
3221 mdd_object_type(mdd_tobj),
3222 lname.ln_name, handle);
3226 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3230 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3234 char *tmp_name = info->mti_key;
3236 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3237 /* lnamelen is too big(> NAME_MAX + 16),
3238 * something wrong about this linkea, let's
3240 CWARN("%s: the name %.*s is too long under "
3241 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3242 lname.ln_namelen, lname.ln_name,
3247 /* Note: lname might be without \0 at the end, see
3248 * linkea_entry_unpack(), let's add extra \0 by
3250 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3251 lname.ln_namelen, lname.ln_name);
3252 lname.ln_name = tmp_name;
3254 /* Let's check if this linkEA still valid, before
3255 * it might be packed into the RPC buffer. */
3256 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3257 &info->mti_fid, NULL);
3258 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3259 mdd_object_fid(mdd_sobj)))
3260 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3262 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3266 rc = __mdd_index_insert(env, pobj,
3267 mdd_object_fid(mdd_tobj),
3268 mdd_object_type(mdd_tobj),
3273 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3274 rc = mdo_ref_add(env, mdd_tobj, handle);
3275 mdd_write_unlock(env, mdd_tobj);
3279 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3280 mdo_ref_del(env, mdd_sobj, handle);
3281 mdd_write_unlock(env, mdd_sobj);
3284 mdd_object_put(env, pobj);
3288 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3295 static int mdd_migrate_xattrs(const struct lu_env *env,
3296 struct mdd_object *mdd_sobj,
3297 struct mdd_object *mdd_tobj)
3299 struct mdd_thread_info *info = mdd_env_info(env);
3300 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3302 struct thandle *handle;
3308 struct lu_buf list_xbuf;
3311 /* retrieve xattr list from the old object */
3312 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3313 if (list_xsize == -ENODATA)
3319 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3320 if (info->mti_big_buf.lb_buf == NULL)
3323 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3324 list_xbuf.lb_len = list_xsize;
3325 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3330 xname = list_xbuf.lb_buf;
3332 xlen = strnlen(xname, rem - 1) + 1;
3333 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3334 strcmp(XATTR_NAME_LMV, xname) == 0)
3337 /* For directory, if there are default layout, migrate here */
3338 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3339 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3342 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3343 if (xsize == -ENODATA)
3348 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3349 if (info->mti_link_buf.lb_buf == NULL)
3350 GOTO(out, rc = -ENOMEM);
3352 xbuf.lb_len = xsize;
3353 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3354 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3360 handle = mdd_trans_create(env, mdd);
3362 GOTO(out, rc = PTR_ERR(handle));
3364 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3367 GOTO(stop_trans, rc);
3368 /* Note: this transaction is part of migration, and it is not
3369 * the last step of migration, so we set th_local = 1 to avoid
3370 * update last rcvd for this transaction */
3371 handle->th_local = 1;
3372 rc = mdd_trans_start(env, mdd, handle);
3374 GOTO(stop_trans, rc);
3377 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3379 GOTO(stop_trans, rc = 0);
3381 if (unlikely(rc == -ENOSPC &&
3382 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3383 rc = linkea_overflow_shrink(
3384 (struct linkea_data *)(xbuf.lb_buf));
3385 if (likely(rc > 0)) {
3392 GOTO(stop_trans, rc);
3394 rc = mdd_trans_stop(env, mdd, rc, handle);
3399 memmove(xname, xname + xlen, rem);
3405 static int mdd_declare_migrate_create(const struct lu_env *env,
3406 struct mdd_object *mdd_pobj,
3407 struct mdd_object *mdd_sobj,
3408 struct mdd_object *mdd_tobj,
3409 struct md_op_spec *spec,
3411 union lmv_mds_md *mgr_ea,
3412 struct linkea_data *ldata,
3413 struct thandle *handle)
3415 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3416 const struct lu_buf *buf;
3420 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3421 handle, spec, NULL);
3425 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3430 if (S_ISLNK(la->la_mode)) {
3431 const char *target_name = spec->u.sp_symname;
3432 int sym_len = strlen(target_name);
3433 const struct lu_buf *buf;
3435 buf = mdd_buf_get_const(env, target_name, sym_len);
3436 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3440 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3441 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3446 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3447 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3448 spec->u.sp_ea.eadatalen);
3449 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3455 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3456 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3457 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3462 la_flag->la_valid = LA_FLAGS;
3463 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3464 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3469 static int mdd_migrate_create(const struct lu_env *env,
3470 struct mdd_object *mdd_pobj,
3471 struct mdd_object *mdd_sobj,
3472 struct mdd_object *mdd_tobj,
3473 const struct lu_name *lname,
3476 struct mdd_thread_info *info = mdd_env_info(env);
3477 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3478 struct md_op_spec *spec = &info->mti_spec;
3479 struct lu_buf lmm_buf = { NULL };
3480 struct lu_buf link_buf = { NULL };
3481 struct lu_buf mgr_buf;
3482 struct thandle *handle;
3483 struct lmv_mds_md_v1 *mgr_ea;
3484 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3485 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3487 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3491 /* prepare spec for create */
3492 memset(spec, 0, sizeof(*spec));
3493 spec->sp_cr_lookup = 0;
3494 spec->sp_feat = &dt_directory_features;
3495 if (S_ISLNK(la->la_mode)) {
3496 const struct lu_buf *buf;
3498 buf = lu_buf_check_and_alloc(
3499 &mdd_env_info(env)->mti_big_buf,
3502 link_buf.lb_len = la->la_size + 1;
3503 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3504 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3506 rc = rc != 0 ? rc : -EFAULT;
3507 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3508 mdd2obd_dev(mdd)->obd_name,
3509 PFID(mdd_object_fid(mdd_sobj)), rc);
3512 spec->u.sp_symname = link_buf.lb_buf;
3513 } else if (S_ISREG(la->la_mode)) {
3514 /* retrieve lov of the old object */
3515 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3516 if (rc != 0 && rc != -ENODATA)
3518 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3519 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3520 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3521 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3523 } else if (S_ISDIR(la->la_mode)) {
3524 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3525 if (rc == -ENODATA) {
3526 /* ignore the non-linkEA error */
3534 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3535 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3536 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3537 mgr_buf.lb_len = mgr_easize;
3538 mgr_ea = mgr_buf.lb_buf;
3539 memset(mgr_ea, 0, sizeof(*mgr_ea));
3540 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3541 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3542 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3543 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3544 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3545 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3547 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3549 handle = mdd_trans_create(env, mdd);
3551 GOTO(out_free, rc = PTR_ERR(handle));
3553 /* Note: this transaction is part of migration, and it is not
3554 * the last step of migration, so we set th_local = 1 to avoid
3555 * update last rcvd for this transaction */
3556 handle->th_local = 1;
3557 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3558 la, mgr_buf.lb_buf, ldata, handle);
3560 GOTO(stop_trans, rc);
3562 rc = mdd_trans_start(env, mdd, handle);
3564 GOTO(stop_trans, rc);
3566 /* don't set nlink from the original object */
3567 la->la_valid &= ~LA_NLINK;
3569 /* create the target object */
3570 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3573 GOTO(stop_trans, rc);
3575 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3576 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3578 GOTO(stop_trans, rc);
3581 /* Set MIGRATE EA on the source inode, so once the migration needs
3582 * to be re-done during failover, the re-do process can locate the
3583 * target object which is already being created. */
3584 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3586 GOTO(stop_trans, rc);
3588 /* Set immutable flag, so any modification is disabled until
3589 * the migration is done. Once the migration is interrupted,
3590 * if the resume process find the migrating object has both
3591 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3592 * flag and approve the migration */
3593 la_flag->la_valid = LA_FLAGS;
3594 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3595 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3598 rc = mdd_trans_stop(env, mdd, rc, handle);
3600 if (lmm_buf.lb_buf != NULL)
3601 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3605 static int mdd_migrate_entries(const struct lu_env *env,
3606 struct mdd_object *mdd_sobj,
3607 struct mdd_object *mdd_tobj)
3609 struct dt_object *next = mdd_object_child(mdd_sobj);
3610 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3611 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3612 struct thandle *handle;
3614 const struct dt_it_ops *iops;
3616 struct lu_dirent *ent;
3620 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3624 if (!dt_try_as_dir(env, next))
3625 GOTO(out_ent, rc = -ENOTDIR);
3627 * iterate directories
3629 iops = &next->do_index_ops->dio_it;
3630 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3632 GOTO(out_ent, rc = PTR_ERR(it));
3634 rc = iops->load(env, it, 0);
3636 rc = iops->next(env, it);
3640 * At this point and across for-loop:
3642 * rc == 0 -> ok, proceed.
3643 * rc > 0 -> end of directory.
3647 struct mdd_object *child;
3648 char *name = mdd_env_info(env)->mti_key;
3651 bool target_exist = false;
3653 len = iops->key_size(env, it);
3657 result = iops->rec(env, it, (struct dt_rec *)ent,
3658 LUDA_FID | LUDA_TYPE);
3659 if (result == -ESTALE)
3666 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3668 /* Insert new fid with target name into target dir */
3669 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3670 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3671 ent->lde_name[1] == '.'))
3674 child = mdd_object_find(env, mdd, &ent->lde_fid);
3676 GOTO(out, rc = PTR_ERR(child));
3678 mdd_write_lock(env, child, MOR_SRC_CHILD);
3679 is_dir = S_ISDIR(mdd_object_type(child));
3681 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3683 /* Check whether the name has been inserted to the target */
3684 if (dt_try_as_dir(env, dt_tobj)) {
3685 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3687 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3688 (struct dt_key *)name);
3689 if (unlikely(rc == 0))
3690 target_exist = true;
3693 handle = mdd_trans_create(env, mdd);
3695 GOTO(out_put, rc = PTR_ERR(handle));
3697 /* Note: this transaction is part of migration, and it is not
3698 * the last step of migration, so we set th_local = 1 to avoid
3699 * updating last rcvd for this transaction */
3700 handle->th_local = 1;
3701 if (likely(!target_exist)) {
3702 rc = mdo_declare_index_insert(env, mdd_tobj,
3704 mdd_object_type(child),
3710 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3716 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3721 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3725 /* Update .. for child */
3726 rc = mdo_declare_index_delete(env, child, dotdot,
3731 rc = mdo_declare_index_insert(env, child,
3732 mdd_object_fid(mdd_tobj),
3733 S_IFDIR, dotdot, handle);
3738 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3745 rc = mdd_trans_start(env, mdd, handle);
3747 CERROR("%s: transaction start failed: rc = %d\n",
3748 mdd2obd_dev(mdd)->obd_name, rc);
3752 if (likely(!target_exist)) {
3753 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3754 mdd_object_type(child),
3760 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3765 rc = __mdd_index_delete_only(env, child, dotdot,
3770 rc = __mdd_index_insert_only(env, child,
3771 mdd_object_fid(mdd_tobj), S_IFDIR,
3777 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3779 strlen(name), handle);
3782 mdd_write_unlock(env, child);
3783 mdd_object_put(env, child);
3784 rc = mdd_trans_stop(env, mdd, rc, handle);
3788 result = iops->next(env, it);
3789 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3790 GOTO(out, rc = -EINTR);
3792 if (result == -ESTALE)
3794 } while (result == 0);
3797 iops->fini(env, it);
3799 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3803 static int mdd_declare_update_linkea(const struct lu_env *env,
3804 struct mdd_object *mdd_pobj,
3805 struct mdd_object *mdd_sobj,
3806 struct mdd_object *mdd_tobj,
3807 const struct lu_name *child_name,
3808 struct linkea_data *ldata,
3809 struct thandle *handle)
3811 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3812 child_name, ldata, handle, 1);
3815 static int mdd_update_linkea(const struct lu_env *env,
3816 struct mdd_object *mdd_pobj,
3817 struct mdd_object *mdd_sobj,
3818 struct mdd_object *mdd_tobj,
3819 const struct lu_name *child_name,
3820 struct linkea_data *ldata,
3821 struct thandle *handle)
3823 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3824 child_name, ldata, handle, 0);
3827 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3828 struct mdd_object *mdd_pobj,
3829 struct mdd_object *mdd_sobj,
3830 struct mdd_object *mdd_tobj,
3831 const struct lu_name *lname,
3833 struct lu_attr *parent_la,
3834 struct linkea_data *ldata,
3835 struct thandle *handle)
3837 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3838 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3841 /* Revert IMMUTABLE flag */
3842 la_flag->la_valid = LA_FLAGS;
3843 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3844 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3848 /* delete entry from source dir */
3849 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3853 if (ldata->ld_buf != NULL) {
3854 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3855 mdd_tobj, lname, ldata, handle);
3860 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3861 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3866 handle->th_complex = 1;
3867 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3869 LU_XATTR_REPLACE, handle);
3874 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3875 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3881 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3882 mdd_object_type(mdd_tobj),
3883 lname->ln_name, handle);
3887 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
3891 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3892 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3897 /* delete old object */
3898 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3902 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3903 /* delete old object */
3904 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3907 /* set nlink to 0 */
3908 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3913 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3917 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3921 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3926 static int mdd_migrate_update_name(const struct lu_env *env,
3927 struct mdd_object *mdd_pobj,
3928 struct mdd_object *mdd_sobj,
3929 struct mdd_object *mdd_tobj,
3930 const struct lu_name *lname,
3933 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3934 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3935 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3936 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3937 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3938 struct thandle *handle;
3939 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3940 const char *name = lname->ln_name;
3944 /* update time for parent */
3945 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3946 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3947 p_la->la_valid = LA_CTIME;
3949 rc = mdd_la_get(env, mdd_sobj, so_attr);
3953 ldata->ld_buf = NULL;
3954 rc = mdd_links_read(env, mdd_sobj, ldata);
3955 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
3958 handle = mdd_trans_create(env, mdd);
3960 RETURN(PTR_ERR(handle));
3962 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3963 lname, so_attr, p_la, ldata,
3966 /* If the migration can not be fit in one transaction, just
3967 * leave it in the original MDT */
3969 GOTO(stop_trans, rc = 0);
3971 GOTO(stop_trans, rc);
3974 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
3975 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
3976 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
3977 PFID(mdd_object_fid(mdd_tobj)));
3979 rc = mdd_trans_start(env, mdd, handle);
3981 GOTO(stop_trans, rc);
3983 /* Revert IMMUTABLE flag */
3984 la_flag->la_valid = LA_FLAGS;
3985 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
3986 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3988 GOTO(stop_trans, rc);
3990 /* Remove source name from source directory */
3991 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
3993 GOTO(stop_trans, rc);
3995 if (ldata->ld_buf != NULL) {
3996 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
3997 lname, ldata, handle);
3999 GOTO(stop_trans, rc);
4001 /* linkea update might decrease the source object
4002 * nlink, let's get the attr again after ref_del */
4003 rc = mdd_la_get(env, mdd_sobj, so_attr);
4005 GOTO(stop_trans, rc);
4008 if (S_ISREG(so_attr->la_mode)) {
4009 if (so_attr->la_nlink == 1) {
4010 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4012 if (rc != 0 && rc != -ENODATA)
4013 GOTO(stop_trans, rc);
4015 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4017 LU_XATTR_REPLACE, handle);
4019 GOTO(stop_trans, rc);
4023 /* Insert new fid with target name into target dir */
4024 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4025 mdd_object_type(mdd_tobj), name, handle);
4027 GOTO(stop_trans, rc);
4029 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4031 mdd_sobj->mod_flags |= DEAD_OBJ;
4032 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4034 GOTO(out_unlock, rc);
4036 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4038 GOTO(out_unlock, rc);
4040 mdo_ref_del(env, mdd_sobj, handle);
4042 mdo_ref_del(env, mdd_sobj, handle);
4044 /* Get the attr again after ref_del */
4045 rc = mdd_la_get(env, mdd_sobj, so_attr);
4047 GOTO(out_unlock, rc);
4049 ma->ma_attr = *so_attr;
4050 ma->ma_valid |= MA_INODE;
4052 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4054 GOTO(out_unlock, rc);
4056 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4057 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4058 mdo2fid(mdd_pobj), lname, lname, handle);
4060 CWARN("%s: changelog for migrate %s "DFID
4061 "under "DFID" failed: rc = %d\n",
4062 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4063 PFID(mdd_object_fid(mdd_sobj)),
4064 PFID(mdd_object_fid(mdd_pobj)), rc);
4065 /* Sigh, there are no easy way to migrate back the object, so
4066 * let's reset the result to 0 for now XXX */
4070 mdd_write_unlock(env, mdd_sobj);
4073 rc = mdd_trans_stop(env, mdd, rc, handle);
4078 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4079 const struct lu_fid *fid, __u32 *mdt_index)
4081 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4082 struct seq_server_site *ss;
4085 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4087 range->lsr_flags = LU_SEQ_RANGE_MDT;
4088 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4092 *mdt_index = range->lsr_index;
4097 * Check whether we should migrate the file/dir
4099 * < 0 permission check failed or other error.
4100 * = 0 the file can be migrated.
4101 * > 0 the file does not need to be migrated, mostly
4102 * for multiple link file
4104 static int mdd_migrate_sanity_check(const struct lu_env *env,
4105 struct mdd_object *pobj,
4106 const struct lu_attr *pattr,
4107 struct mdd_object *sobj,
4108 struct lu_attr *sattr)
4110 struct mdd_thread_info *info = mdd_env_info(env);
4111 struct linkea_data *ldata = &info->mti_link_data;
4112 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4114 struct lu_buf *mgr_buf;
4120 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4121 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4122 if (mgr_buf->lb_buf == NULL)
4125 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4127 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4129 /* If the object has migrateEA, it means IMMUTE flag
4130 * is being set by previous migration process, so it
4131 * needs to override the IMMUTE flag, otherwise the
4132 * following sanity check will fail */
4133 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4134 LMV_HASH_FLAG_MIGRATION) {
4135 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4137 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4138 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4139 mdd2obd_dev(mdd)->obd_name,
4140 PFID(mdd_object_fid(sobj)));
4144 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4145 sobj, sattr, NULL, NULL);
4149 /* Then it will check if the file should be migrated. If the file
4150 * has mulitple links, we only need migrate the file if all of its
4151 * entries has been migrated to the remote MDT */
4152 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4155 rc = mdd_links_read(env, sobj, ldata);
4157 /* For multiple links files, if there are no linkEA data at all,
4158 * means the file might be created before linkEA is enabled, and
4159 * all of its links should not be migrated yet, otherwise it
4160 * should have some linkEA there */
4161 if (rc == -ENOENT || rc == -ENODATA)
4166 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4167 /* If there are still links locally, then the file will not be
4169 LASSERT(ldata->ld_leh != NULL);
4171 /* If the linkEA is overflow, then means there are some unknown name
4172 * entries under unknown parents, that will prevent the migration. */
4173 if (unlikely(ldata->ld_leh->leh_overflow_time))
4176 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4177 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4178 struct lu_name lname;
4180 __u32 parent_mdt_index;
4182 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4184 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4187 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4191 /* Migrate the object only if none of its parents are on the
4193 if (parent_mdt_index != mdt_index)
4196 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4197 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4198 lname.ln_name, PFID(&fid));
4206 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4207 struct md_object *sobj, const struct lu_name *lname,
4208 struct md_object *tobj, struct md_attr *ma)
4210 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4211 struct mdd_device *mdd = mdo2mdd(pobj);
4212 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4213 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4214 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4215 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4216 bool created = false;
4220 /* If the file will being migrated, it will check whether
4221 * the file is being opened by someone else right now */
4222 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4223 if (mdd_sobj->mod_count > 0) {
4225 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4226 mdd2obd_dev(mdd)->obd_name,
4227 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4228 mdd_sobj->mod_count, -EBUSY);
4229 mdd_read_unlock(env, mdd_sobj);
4230 GOTO(put, rc = -EBUSY);
4232 mdd_read_unlock(env, mdd_sobj);
4234 rc = mdd_la_get(env, mdd_sobj, so_attr);
4238 rc = mdd_la_get(env, mdd_pobj, pattr);
4242 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4249 /* Sigh, it is impossible to finish all of migration in a single
4250 * transaction, for example migrating big directory entries to the
4251 * new MDT, it needs insert all of name entries of children in the
4254 * So migration will be done in multiple steps and transactions.
4256 * 1. create an orphan object on the remote MDT in one transaction.
4257 * 2. migrate extend attributes to the new target file/directory.
4258 * 3. For directory, migrate the entries to the new MDT and update
4259 * linkEA of each children. Because we can not migrate all entries
4260 * in a single transaction, so the migrating directory will become
4261 * a striped directory during migration, so once the process is
4262 * interrupted, the directory is still accessible. (During lookup,
4263 * client will locate the name by searching both original and target
4265 * 4. Finally, update the name/FID to point to the new file/directory
4266 * in a separate transaction.
4269 /* step 1: Check whether the orphan object has been created, and create
4270 * orphan object on the remote MDT if needed */
4271 if (!mdd_object_exists(mdd_tobj)) {
4272 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4279 LASSERT(mdd_object_exists(mdd_tobj));
4280 /* step 2: migrate xattr */
4281 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4285 /* step 3: migrate name entries to the orphan object */
4286 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4287 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4290 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4291 OBD_FAIL_MDS_REINT_NET_REP)))
4294 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4297 LASSERT(mdd_object_exists(mdd_tobj));
4298 /* step 4: update name entry to the new object */
4299 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4304 /* newly created target was not locked, don't cache its attributes */
4306 mdd_invalidate(env, tobj);
4311 const struct md_dir_operations mdd_dir_ops = {
4312 .mdo_is_subdir = mdd_is_subdir,
4313 .mdo_lookup = mdd_lookup,
4314 .mdo_create = mdd_create,
4315 .mdo_rename = mdd_rename,
4316 .mdo_link = mdd_link,
4317 .mdo_unlink = mdd_unlink,
4318 .mdo_create_data = mdd_create_data,
4319 .mdo_migrate = mdd_migrate,