4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_dir.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@intel.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <obd_class.h>
42 #include <obd_support.h>
43 #include <lustre_mds.h>
44 #include <lustre_fid.h>
46 #include "mdd_internal.h"
48 static const char dot[] = ".";
49 static const char dotdot[] = "..";
51 static struct lu_name lname_dotdot = {
52 .ln_name = (char *) dotdot,
53 .ln_namelen = sizeof(dotdot) - 1,
57 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
59 if (!lu_name_is_valid(ln))
61 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
67 /* Get FID from name and parent */
69 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
70 const struct lu_attr *pattr, const struct lu_name *lname,
71 struct lu_fid* fid, int mask)
73 const char *name = lname->ln_name;
74 const struct dt_key *key = (const struct dt_key *)name;
75 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
76 struct mdd_device *m = mdo2mdd(pobj);
77 struct dt_object *dir = mdd_object_child(mdd_obj);
81 if (unlikely(mdd_is_dead_obj(mdd_obj)))
84 if (!mdd_object_exists(mdd_obj))
87 if (mdd_object_remote(mdd_obj)) {
88 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
89 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
92 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
97 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
98 dt_try_as_dir(env, dir)))
99 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
106 int mdd_lookup(const struct lu_env *env,
107 struct md_object *pobj, const struct lu_name *lname,
108 struct lu_fid *fid, struct md_op_spec *spec)
110 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
114 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
118 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
119 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
123 /** Read the link EA into a temp buffer.
124 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
125 * A pointer to the buffer is stored in \a ldata::ld_buf.
129 static int __mdd_links_read(const struct lu_env *env,
130 struct mdd_object *mdd_obj,
131 struct linkea_data *ldata)
135 if (!mdd_object_exists(mdd_obj))
138 /* First try a small buf */
139 LASSERT(env != NULL);
140 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
142 if (ldata->ld_buf->lb_buf == NULL)
145 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
147 /* Buf was too small, figure out what we need. */
148 lu_buf_free(ldata->ld_buf);
149 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
153 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
154 if (ldata->ld_buf->lb_buf == NULL)
156 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
160 lu_buf_free(ldata->ld_buf);
161 ldata->ld_buf = NULL;
165 return linkea_init(ldata);
168 static int mdd_links_read(const struct lu_env *env,
169 struct mdd_object *mdd_obj,
170 struct linkea_data *ldata)
174 rc = __mdd_links_read(env, mdd_obj, ldata);
176 rc = linkea_init(ldata);
181 static int mdd_links_read_with_rec(const struct lu_env *env,
182 struct mdd_object *mdd_obj,
183 struct linkea_data *ldata)
187 rc = __mdd_links_read(env, mdd_obj, ldata);
189 rc = linkea_init_with_rec(ldata);
195 * Get parent FID of the directory
197 * Read parent FID from linkEA, if that fails, then do lookup
198 * dotdot to get the parent FID.
200 * \param[in] env execution environment
201 * \param[in] obj object from which to find the parent FID
202 * \param[in] attr attribute of the object
203 * \param[out] fid fid to get the parent FID
205 * \retval 0 if getting the parent FID succeeds.
206 * \retval negative errno if getting the parent FID fails.
208 static inline int mdd_parent_fid(const struct lu_env *env,
209 struct mdd_object *obj,
210 const struct lu_attr *attr,
213 struct mdd_thread_info *info = mdd_env_info(env);
214 struct linkea_data ldata = { NULL };
215 struct lu_buf *buf = &info->mti_link_buf;
216 struct lu_name lname;
221 LASSERT(S_ISDIR(mdd_object_type(obj)));
223 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
224 if (buf->lb_buf == NULL)
225 GOTO(lookup, rc = 0);
228 rc = mdd_links_read_with_rec(env, obj, &ldata);
232 LASSERT(ldata.ld_leh != NULL);
233 /* Directory should only have 1 parent */
234 if (ldata.ld_leh->leh_reccount > 1)
237 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
239 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
240 if (likely(fid_is_sane(fid)))
243 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
248 * For root fid use special function, which does not compare version component
249 * of fid. Version component is different for root fids on all MDTs.
251 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
253 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
254 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
258 * return 1: if lf is the fid of the ancestor of p1;
261 * return -EREMOTE: if remote object is found, in this
262 * case fid of remote object is saved to @pf;
264 * otherwise: values < 0, errors.
266 static int mdd_is_parent(const struct lu_env *env,
267 struct mdd_device *mdd,
268 struct mdd_object *p1,
269 const struct lu_attr *attr,
270 const struct lu_fid *lf,
273 struct mdd_object *parent = NULL;
278 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
279 pfid = &mdd_env_info(env)->mti_fid;
281 /* Check for root first. */
282 if (mdd_is_root(mdd, mdo2fid(p1)))
286 /* this is done recursively */
287 rc = mdd_parent_fid(env, p1, attr, pfid);
290 if (mdd_is_root(mdd, pfid))
292 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
294 if (lu_fid_eq(pfid, lf))
297 mdd_object_put(env, parent);
299 parent = mdd_object_find(env, mdd, pfid);
301 GOTO(out, rc = PTR_ERR(parent));
303 if (!mdd_object_exists(parent))
304 GOTO(out, rc = -EINVAL);
310 if (parent && !IS_ERR(parent))
311 mdd_object_put(env, parent);
316 * No permission check is needed.
318 * returns 1: if fid is ancestor of @mo;
319 * returns 0: if fid is not an ancestor of @mo;
321 * returns EREMOTE if remote object is found, fid of remote object is saved to
324 * returns < 0: if error
326 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
327 const struct lu_fid *fid, struct lu_fid *sfid)
329 struct mdd_device *mdd = mdo2mdd(mo);
330 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
334 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
337 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
341 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
345 } else if (rc == 1) {
346 /* found @fid is parent */
354 * Check that @dir contains no entries except (possibly) dot and dotdot.
359 * -ENOTDIR not a directory object
360 * -ENOTEMPTY not empty
364 static int mdd_dir_is_empty(const struct lu_env *env,
365 struct mdd_object *dir)
368 struct dt_object *obj;
369 const struct dt_it_ops *iops;
373 obj = mdd_object_child(dir);
374 if (!dt_try_as_dir(env, obj))
377 iops = &obj->do_index_ops->dio_it;
378 it = iops->init(env, obj, LUDA_64BITHASH);
380 result = iops->get(env, it, (const struct dt_key *)"");
383 for (result = 0, i = 0; result == 0 && i < 3; ++i)
384 result = iops->next(env, it);
387 else if (result == 1)
389 } else if (result == 0)
391 * Huh? Index contains no zero key?
398 result = PTR_ERR(it);
403 * Determine if the target object can be hard linked, and right now it only
404 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
405 * directory nlink count might exceed the maximum link count(see
406 * osd_object_ref_add), so it only check nlink for non-directories.
408 * \param[in] env thread environment
409 * \param[in] obj object being linked to
410 * \param[in] la attributes of \a obj
412 * \retval 0 if \a obj can be hard linked
413 * \retval negative error if \a obj is a directory or has too
416 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
417 const struct lu_attr *la)
419 struct mdd_device *m = mdd_obj2mdd_dev(obj);
424 /* Subdir count limitation can be broken through
425 * (see osd_object_ref_add), so only check non-directory here. */
426 if (!S_ISDIR(la->la_mode) &&
427 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
434 * Check whether it may create the cobj under the pobj.
436 * \param[in] env execution environment
437 * \param[in] pobj the parent directory
438 * \param[in] pattr the attribute of the parent directory
439 * \param[in] cobj the child to be created
440 * \param[in] check_perm if check WRITE|EXEC permission for parent
442 * \retval = 0 create the child under this dir is allowed
443 * \retval negative errno create the child under this dir is
446 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
447 const struct lu_attr *pattr, struct mdd_object *cobj,
453 if (cobj && mdd_object_exists(cobj))
456 if (mdd_is_dead_obj(pobj))
460 rc = mdd_permission_internal_locked(env, pobj, pattr,
461 MAY_WRITE | MAY_EXEC,
467 * Check whether can unlink from the pobj in the case of "cobj == NULL".
469 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
470 const struct lu_attr *pattr, const struct lu_attr *attr)
475 if (mdd_is_dead_obj(pobj))
478 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
481 rc = mdd_permission_internal_locked(env, pobj, pattr,
482 MAY_WRITE | MAY_EXEC,
487 if (pattr->la_flags & LUSTRE_APPEND_FL)
494 * pobj == NULL is remote ops case, under such case, pobj's
495 * VTX feature has been checked already, no need check again.
497 static inline int mdd_is_sticky(const struct lu_env *env,
498 struct mdd_object *pobj,
499 const struct lu_attr *pattr,
500 struct mdd_object *cobj,
501 const struct lu_attr *cattr)
503 struct lu_ucred *uc = lu_ucred_assert(env);
506 LASSERT(pattr != NULL);
507 if (!(pattr->la_mode & S_ISVTX) ||
508 (pattr->la_uid == uc->uc_fsuid))
512 LASSERT(cattr != NULL);
513 if (cattr->la_uid == uc->uc_fsuid)
516 return !md_capable(uc, CFS_CAP_FOWNER);
519 static int mdd_may_delete_entry(const struct lu_env *env,
520 struct mdd_object *pobj,
521 const struct lu_attr *pattr,
526 LASSERT(pobj != NULL);
527 if (!mdd_object_exists(pobj))
530 if (mdd_is_dead_obj(pobj))
535 rc = mdd_permission_internal_locked(env, pobj, pattr,
536 MAY_WRITE | MAY_EXEC,
542 if (pattr->la_flags & LUSTRE_APPEND_FL)
549 * Check whether it may delete the cobj from the pobj.
552 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
553 const struct lu_attr *tpattr, struct mdd_object *tobj,
554 const struct lu_attr *tattr, const struct lu_attr *cattr,
555 int check_perm, int check_empty)
561 LASSERT(tpattr != NULL);
562 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
570 if (!mdd_object_exists(tobj))
573 if (mdd_is_dead_obj(tobj))
576 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
579 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
582 /* additional check the rename case */
584 if (S_ISDIR(cattr->la_mode)) {
585 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
587 if (!S_ISDIR(tattr->la_mode))
590 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
592 } else if (S_ISDIR(tattr->la_mode))
596 if (S_ISDIR(tattr->la_mode) && check_empty)
597 rc = mdd_dir_is_empty(env, tobj);
603 * Check whether it can create the link file(linked to @src_obj) under
604 * the target directory(@tgt_obj), and src_obj has been locked by
607 * \param[in] env execution environment
608 * \param[in] tgt_obj the target directory
609 * \param[in] tattr attributes of target directory
610 * \param[in] lname the link name
611 * \param[in] src_obj source object for link
612 * \param[in] cattr attributes for source object
614 * \retval = 0 it is allowed to create the link file under tgt_obj
615 * \retval negative error not allowed to create the link file
617 static int mdd_link_sanity_check(const struct lu_env *env,
618 struct mdd_object *tgt_obj,
619 const struct lu_attr *tattr,
620 const struct lu_name *lname,
621 struct mdd_object *src_obj,
622 const struct lu_attr *cattr)
624 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
628 if (!mdd_object_exists(src_obj))
631 if (mdd_is_dead_obj(src_obj))
634 /* Local ops, no lookup before link, check filename length here. */
635 rc = mdd_name_check(m, lname);
639 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
642 if (S_ISDIR(mdd_object_type(src_obj)))
645 LASSERT(src_obj != tgt_obj);
646 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
650 rc = __mdd_may_link(env, src_obj, cattr);
655 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
656 const char *name, struct thandle *handle)
658 struct dt_object *next = mdd_object_child(pobj);
662 if (dt_try_as_dir(env, next))
663 rc = dt_delete(env, next, (struct dt_key *)name, handle);
670 static int __mdd_index_insert_only(const struct lu_env *env,
671 struct mdd_object *pobj,
672 const struct lu_fid *lf, __u32 type,
674 struct thandle *handle)
676 struct dt_object *next = mdd_object_child(pobj);
680 if (dt_try_as_dir(env, next)) {
681 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
682 struct lu_ucred *uc = lu_ucred_check(env);
686 rec->rec_type = type;
687 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
688 rc = dt_insert(env, next, (const struct dt_rec *)rec,
689 (const struct dt_key *)name, handle,
697 /* insert named index, add reference if isdir */
698 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
699 const struct lu_fid *lf, __u32 type,
700 const char *name, struct thandle *handle)
705 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
706 if (rc == 0 && S_ISDIR(type)) {
707 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
708 mdo_ref_add(env, pobj, handle);
709 mdd_write_unlock(env, pobj);
715 /* delete named index, drop reference if isdir */
716 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
717 const char *name, int is_dir,
718 struct thandle *handle)
723 rc = __mdd_index_delete_only(env, pobj, name, handle);
724 if (rc == 0 && is_dir) {
725 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
726 mdo_ref_del(env, pobj, handle);
727 mdd_write_unlock(env, pobj);
733 static int mdd_llog_record_calc_size(const struct lu_env *env,
734 const struct lu_name *tname,
735 const struct lu_name *sname)
737 const struct lu_ucred *uc = lu_ucred(env);
738 enum changelog_rec_flags crf = CLF_EXTRA_FLAGS;
739 enum changelog_rec_extra_flags crfe = CLFE_INVALID;
744 if (uc != NULL && uc->uc_jobid[0] != '\0')
747 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
748 changelog_rec_offset(crf, crfe) +
749 (tname != NULL ? tname->ln_namelen : 0) +
750 (sname != NULL ? 1 + sname->ln_namelen : 0));
753 int mdd_declare_changelog_store(const struct lu_env *env,
754 struct mdd_device *mdd,
755 const struct lu_name *tname,
756 const struct lu_name *sname,
757 struct thandle *handle)
759 struct obd_device *obd = mdd2obd_dev(mdd);
760 struct llog_ctxt *ctxt;
761 struct llog_changelog_rec *rec;
763 struct thandle *llog_th;
768 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
771 reclen = mdd_llog_record_calc_size(env, tname, sname);
772 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
773 if (buf->lb_buf == NULL)
777 rec->cr_hdr.lrh_len = reclen;
778 rec->cr_hdr.lrh_type = CHANGELOG_REC;
780 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
784 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
786 GOTO(out_put, rc = PTR_ERR(llog_th));
788 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
796 /** Add a changelog entry \a rec to the changelog llog
799 * \param handle - currently ignored since llogs start their own transaction;
800 * this will hopefully be fixed in llog rewrite
803 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
804 struct llog_changelog_rec *rec, struct thandle *th)
806 struct obd_device *obd = mdd2obd_dev(mdd);
807 struct llog_ctxt *ctxt;
808 struct thandle *llog_th;
811 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
812 changelog_rec_varsize(&rec->cr));
814 /* llog_lvfs_write_rec sets the llog tail len */
815 rec->cr_hdr.lrh_type = CHANGELOG_REC;
816 rec->cr.cr_time = cl_time();
818 spin_lock(&mdd->mdd_cl.mc_lock);
819 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
820 * but as long as the MDD transactions are ordered correctly for e.g.
821 * rename conflicts, I don't think this should matter. */
822 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
823 spin_unlock(&mdd->mdd_cl.mc_lock);
825 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
829 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
831 GOTO(out_put, rc = PTR_ERR(llog_th));
833 /* nested journal transaction */
834 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
843 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
844 const struct lu_fid *sfid,
845 const struct lu_fid *spfid,
846 const struct lu_name *sname)
848 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
849 size_t extsize = sname->ln_namelen + 1;
851 LASSERT(sfid != NULL);
852 LASSERT(spfid != NULL);
853 LASSERT(sname != NULL);
855 rnm->cr_sfid = *sfid;
856 rnm->cr_spfid = *spfid;
858 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
859 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
860 rec->cr_namelen += extsize;
863 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
865 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
867 if (jobid == NULL || jobid[0] == '\0')
870 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
873 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
875 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
877 ef->cr_extra_flags = eflags;
880 /** Store a namespace change changelog record
881 * If this fails, we must fail the whole transaction; we don't
882 * want the change to commit without the log entry.
883 * \param target - mdd_object of change
884 * \param tpfid - target parent dir/object fid
885 * \param sfid - source object fid
886 * \param spfid - source parent fid
887 * \param tname - target name string
888 * \param sname - source name string
889 * \param handle - transaction handle
891 int mdd_changelog_ns_store(const struct lu_env *env,
892 struct mdd_device *mdd,
893 enum changelog_rec_type type,
894 enum changelog_rec_flags crf,
895 struct mdd_object *target,
896 const struct lu_fid *tpfid,
897 const struct lu_fid *sfid,
898 const struct lu_fid *spfid,
899 const struct lu_name *tname,
900 const struct lu_name *sname,
901 struct thandle *handle)
903 const struct lu_ucred *uc = lu_ucred(env);
904 struct llog_changelog_rec *rec;
907 __u64 xflags = CLFE_INVALID;
912 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
915 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
918 LASSERT(tpfid != NULL);
919 LASSERT(tname != NULL);
920 LASSERT(handle != NULL);
922 reclen = mdd_llog_record_calc_size(env, tname, sname);
923 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
924 if (buf->lb_buf == NULL)
929 crf |= CLF_EXTRA_FLAGS;
931 if (uc != NULL && uc->uc_jobid[0] != '\0')
939 rec->cr.cr_flags = crf;
941 if (crf & CLF_EXTRA_FLAGS) {
942 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
945 rec->cr.cr_type = (__u32)type;
946 rec->cr.cr_pfid = *tpfid;
947 rec->cr.cr_namelen = tname->ln_namelen;
948 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
950 if (crf & CLF_RENAME)
951 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
954 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
956 if (likely(target != NULL)) {
957 rec->cr.cr_tfid = *mdo2fid(target);
958 target->mod_cltime = ktime_get();
960 fid_zero(&rec->cr.cr_tfid);
963 rc = mdd_changelog_store(env, mdd, rec, handle);
965 CERROR("%s: cannot store changelog record: type = %d, "
966 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
967 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
968 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
975 static int __mdd_links_add(const struct lu_env *env,
976 struct mdd_object *mdd_obj,
977 struct linkea_data *ldata,
978 const struct lu_name *lname,
979 const struct lu_fid *pfid,
980 int first, int check)
984 if (ldata->ld_leh == NULL) {
985 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
989 rc = linkea_data_new(ldata,
990 &mdd_env_info(env)->mti_link_buf);
997 rc = linkea_links_find(ldata, lname, pfid);
998 if (rc && rc != -ENOENT)
1004 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1005 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1009 linkea_add_buf(ldata, lname, tfid);
1012 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1013 linkea_add_buf(ldata, lname, pfid);
1015 return linkea_add_buf(ldata, lname, pfid);
1018 static int __mdd_links_del(const struct lu_env *env,
1019 struct mdd_object *mdd_obj,
1020 struct linkea_data *ldata,
1021 const struct lu_name *lname,
1022 const struct lu_fid *pfid)
1026 if (ldata->ld_leh == NULL) {
1027 rc = mdd_links_read(env, mdd_obj, ldata);
1032 rc = linkea_links_find(ldata, lname, pfid);
1036 linkea_del_buf(ldata, lname);
1040 static int mdd_linkea_prepare(const struct lu_env *env,
1041 struct mdd_object *mdd_obj,
1042 const struct lu_fid *oldpfid,
1043 const struct lu_name *oldlname,
1044 const struct lu_fid *newpfid,
1045 const struct lu_name *newlname,
1046 int first, int check,
1047 struct linkea_data *ldata)
1052 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1055 LASSERT(oldpfid != NULL || newpfid != NULL);
1057 if (mdd_obj->mod_flags & DEAD_OBJ)
1058 /* Unnecessary to update linkEA for dead object. */
1061 if (oldpfid != NULL) {
1062 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1064 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1067 /* No changes done. */
1072 /* If renaming, add the new record */
1073 if (newpfid != NULL)
1074 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1080 int mdd_links_rename(const struct lu_env *env,
1081 struct mdd_object *mdd_obj,
1082 const struct lu_fid *oldpfid,
1083 const struct lu_name *oldlname,
1084 const struct lu_fid *newpfid,
1085 const struct lu_name *newlname,
1086 struct thandle *handle,
1087 struct linkea_data *ldata,
1088 int first, int check)
1093 if (ldata == NULL) {
1094 ldata = &mdd_env_info(env)->mti_link_data;
1095 memset(ldata, 0, sizeof(*ldata));
1096 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1097 newpfid, newlname, first, check, ldata);
1102 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1103 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1109 if (newlname == NULL)
1110 CERROR("link_ea add failed %d "DFID"\n",
1111 rc, PFID(mdd_object_fid(mdd_obj)));
1112 else if (oldpfid == NULL)
1113 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1114 newlname->ln_namelen, newlname->ln_name, rc,
1115 PFID(mdd_object_fid(mdd_obj)));
1116 else if (newpfid == NULL)
1117 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1118 oldlname->ln_namelen, oldlname->ln_name, rc,
1119 PFID(mdd_object_fid(mdd_obj)));
1121 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1122 "\n", oldlname->ln_namelen, oldlname->ln_name,
1123 newlname->ln_namelen, newlname->ln_name, rc,
1124 PFID(mdd_object_fid(mdd_obj)));
1127 if (is_vmalloc_addr(ldata->ld_buf))
1128 /* if we vmalloced a large buffer drop it */
1129 lu_buf_free(ldata->ld_buf);
1134 static inline int mdd_links_add(const struct lu_env *env,
1135 struct mdd_object *mdd_obj,
1136 const struct lu_fid *pfid,
1137 const struct lu_name *lname,
1138 struct thandle *handle,
1139 struct linkea_data *ldata, int first)
1141 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1142 pfid, lname, handle, ldata, first, 0);
1145 static inline int mdd_links_del(const struct lu_env *env,
1146 struct mdd_object *mdd_obj,
1147 const struct lu_fid *pfid,
1148 const struct lu_name *lname,
1149 struct thandle *handle)
1151 return mdd_links_rename(env, mdd_obj, pfid, lname,
1152 NULL, NULL, handle, NULL, 0, 0);
1155 /** Read the link EA into a temp buffer.
1156 * Uses the name_buf since it is generally large.
1157 * \retval IS_ERR err
1158 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1160 struct lu_buf *mdd_links_get(const struct lu_env *env,
1161 struct mdd_object *mdd_obj)
1163 struct linkea_data ldata = { NULL };
1166 rc = mdd_links_read(env, mdd_obj, &ldata);
1167 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1170 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1171 struct linkea_data *ldata, struct thandle *handle)
1173 const struct lu_buf *buf;
1176 if (ldata == NULL || ldata->ld_buf == NULL ||
1177 ldata->ld_leh == NULL)
1180 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1184 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1185 ldata->ld_leh->leh_len);
1186 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1187 if (unlikely(rc == -ENOSPC)) {
1188 rc = linkea_overflow_shrink(ldata);
1196 static int mdd_declare_links_add(const struct lu_env *env,
1197 struct mdd_object *mdd_obj,
1198 struct thandle *handle,
1199 struct linkea_data *ldata)
1205 if (ldata != NULL && ldata->ld_leh != NULL) {
1206 ea_len = ldata->ld_leh->leh_len;
1207 linkea = ldata->ld_buf->lb_buf;
1209 ea_len = MAX_LINKEA_SIZE;
1213 rc = mdo_declare_xattr_set(env, mdd_obj,
1214 mdd_buf_get_const(env, linkea, ea_len),
1215 XATTR_NAME_LINK, 0, handle);
1220 static inline int mdd_declare_links_del(const struct lu_env *env,
1221 struct mdd_object *c,
1222 struct thandle *handle)
1226 /* For directory, the linkEA will be removed together
1227 * with the object. */
1228 if (!S_ISDIR(mdd_object_type(c)))
1229 rc = mdd_declare_links_add(env, c, handle, NULL);
1234 static int mdd_declare_link(const struct lu_env *env,
1235 struct mdd_device *mdd,
1236 struct mdd_object *p,
1237 struct mdd_object *c,
1238 const struct lu_name *name,
1239 struct thandle *handle,
1241 struct linkea_data *data)
1243 struct lu_fid tfid = *mdo2fid(c);
1246 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1247 tfid.f_oid = cfs_fail_val;
1249 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1250 name->ln_name, handle);
1254 rc = mdo_declare_ref_add(env, c, handle);
1258 la->la_valid = LA_CTIME | LA_MTIME;
1259 rc = mdo_declare_attr_set(env, p, la, handle);
1263 la->la_valid = LA_CTIME;
1264 rc = mdo_declare_attr_set(env, c, la, handle);
1268 rc = mdd_declare_links_add(env, c, handle, data);
1272 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1277 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1278 struct md_object *src_obj, const struct lu_name *lname,
1281 const char *name = lname->ln_name;
1282 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1283 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1284 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1285 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1286 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1287 struct mdd_device *mdd = mdo2mdd(src_obj);
1288 struct thandle *handle;
1289 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1290 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1294 rc = mdd_la_get(env, mdd_sobj, cattr);
1298 rc = mdd_la_get(env, mdd_tobj, tattr);
1303 * If we are using project inheritance, we only allow hard link
1304 * creation in our tree when the project IDs are the same;
1305 * otherwise the tree quota mechanism could be circumvented.
1307 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1308 (tattr->la_projid != cattr->la_projid))
1311 handle = mdd_trans_create(env, mdd);
1313 GOTO(out_pending, rc = PTR_ERR(handle));
1315 memset(ldata, 0, sizeof(*ldata));
1317 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1318 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1320 /* Note: even this function will change ldata, but it comes from
1321 * thread_info, which is completely temporary and only seen in
1322 * this function, so we do not need reset ldata once it fails.*/
1323 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1324 lname, 0, 0, ldata);
1328 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1333 rc = mdd_trans_start(env, mdd, handle);
1337 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1338 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1341 GOTO(out_unlock, rc);
1343 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1344 rc = mdo_ref_add(env, mdd_sobj, handle);
1346 GOTO(out_unlock, rc);
1349 *tfid = *mdo2fid(mdd_sobj);
1350 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1351 tfid->f_oid = cfs_fail_val;
1353 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1354 mdd_object_type(mdd_sobj), name, handle);
1356 mdo_ref_del(env, mdd_sobj, handle);
1357 GOTO(out_unlock, rc);
1360 la->la_valid = LA_CTIME | LA_MTIME;
1361 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1363 GOTO(out_unlock, rc);
1365 la->la_valid = LA_CTIME;
1366 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1368 /* Note: The failure of links_add should not cause the
1369 * link failure, so do not check return value. */
1370 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1371 lname, handle, ldata, 0);
1375 mdd_write_unlock(env, mdd_sobj);
1377 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1378 mdo2fid(mdd_tobj), NULL, NULL,
1379 lname, NULL, handle);
1381 rc = mdd_trans_stop(env, mdd, rc, handle);
1382 if (is_vmalloc_addr(ldata->ld_buf))
1383 /* if we vmalloced a large buffer drop it */
1384 lu_buf_free(ldata->ld_buf);
1389 static int mdd_mark_orphan_object(const struct lu_env *env,
1390 struct mdd_object *obj, struct thandle *handle,
1393 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1396 if (!S_ISDIR(mdd_object_type(obj)))
1399 attr->la_valid = LA_FLAGS;
1400 attr->la_flags = LUSTRE_ORPHAN_FL;
1403 rc = mdo_declare_attr_set(env, obj, attr, handle);
1405 rc = mdo_attr_set(env, obj, attr, handle);
1410 static int mdd_declare_finish_unlink(const struct lu_env *env,
1411 struct mdd_object *obj,
1412 struct thandle *handle)
1416 /* Sigh, we do not know if the unlink object will become orphan in
1417 * declare phase, but fortunately the flags here does not matter
1418 * in current declare implementation */
1419 rc = mdd_mark_orphan_object(env, obj, handle, true);
1423 rc = mdo_declare_destroy(env, obj, handle);
1427 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1431 return mdd_declare_links_del(env, obj, handle);
1434 /* caller should take a lock before calling */
1435 int mdd_finish_unlink(const struct lu_env *env,
1436 struct mdd_object *obj, struct md_attr *ma,
1437 const struct mdd_object *pobj,
1438 const struct lu_name *lname,
1442 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1445 LASSERT(mdd_write_locked(env, obj) != 0);
1447 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1448 /* add new orphan and the object
1449 * will be deleted during mdd_close() */
1450 obj->mod_flags |= DEAD_OBJ;
1451 if (obj->mod_count) {
1452 rc = __mdd_orphan_add(env, obj, th);
1454 CDEBUG(D_HA, "Object "DFID" is inserted into "
1455 "orphan list, open count = %d\n",
1456 PFID(mdd_object_fid(obj)),
1459 CERROR("Object "DFID" fail to be an orphan, "
1460 "open count = %d, maybe cause failed "
1462 PFID(mdd_object_fid(obj)),
1465 /* mark object as an orphan here, not
1466 * before __mdd_orphan_add() as racing
1467 * mdd_la_get() may propagate ORPHAN_OBJ
1468 * causing the asserition */
1469 rc = mdd_mark_orphan_object(env, obj, th, false);
1471 rc = mdo_destroy(env, obj, th);
1473 } else if (!is_dir) {
1474 /* old files may not have link ea; ignore errors */
1475 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1483 * has mdd_write_lock on cobj already, but not on pobj yet
1485 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1486 const struct lu_attr *pattr,
1487 struct mdd_object *cobj,
1488 const struct lu_attr *cattr)
1493 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1498 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1499 struct mdd_object *p, struct mdd_object *c,
1500 const struct lu_name *name, struct md_attr *ma,
1501 struct thandle *handle, int no_name, int is_dir)
1503 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1506 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1507 if (likely(no_name == 0)) {
1508 rc = mdo_declare_index_delete(env, p, name->ln_name,
1515 rc = mdo_declare_ref_del(env, p, handle);
1521 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1522 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1523 la->la_valid = LA_CTIME | LA_MTIME;
1524 rc = mdo_declare_attr_set(env, p, la, handle);
1529 rc = mdo_declare_ref_del(env, c, handle);
1533 rc = mdo_declare_ref_del(env, c, handle);
1537 la->la_valid = LA_CTIME;
1538 rc = mdo_declare_attr_set(env, c, la, handle);
1542 rc = mdd_declare_finish_unlink(env, c, handle);
1546 /* FIXME: need changelog for remove entry */
1547 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1554 * test if a file has an HSM archive
1555 * if HSM attributes are not found in ma update them from
1558 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1559 struct mdd_object *obj,
1564 if (!(ma->ma_valid & MA_HSM)) {
1565 /* no HSM MD provided, read xattr */
1566 struct lu_buf *hsm_buf;
1567 const size_t buflen = sizeof(struct hsm_attrs);
1570 hsm_buf = mdd_buf_get(env, NULL, 0);
1571 lu_buf_alloc(hsm_buf, buflen);
1572 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1573 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1574 lu_buf_free(hsm_buf);
1578 ma->ma_valid |= MA_HSM;
1580 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1586 * Delete name entry and the object.
1587 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1588 * does not exist for this object, and it could only happen during resending
1589 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1590 * is also needed in this case(needed by changelog), so we have to add another
1591 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1592 * the ENOENT failure should be able to be fixed by redo mechanism.
1594 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1595 struct md_object *cobj, const struct lu_name *lname,
1596 struct md_attr *ma, int no_name)
1598 const char *name = lname->ln_name;
1599 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1600 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1601 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1602 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1603 struct mdd_object *mdd_cobj = NULL;
1604 struct mdd_device *mdd = mdo2mdd(pobj);
1605 struct thandle *handle;
1606 int rc, is_dir = 0, cl_flags = 0;
1609 /* cobj == NULL means only delete name entry */
1610 if (likely(cobj != NULL)) {
1611 mdd_cobj = md2mdd_obj(cobj);
1612 if (mdd_object_exists(mdd_cobj) == 0)
1616 rc = mdd_la_get(env, mdd_pobj, pattr);
1620 if (likely(mdd_cobj != NULL)) {
1622 rc = mdd_la_get(env, mdd_cobj, cattr);
1626 is_dir = S_ISDIR(cattr->la_mode);
1627 /* search for an existing archive.
1628 * we should check ahead as the object
1629 * can be destroyed in this transaction */
1630 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1631 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1634 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1638 handle = mdd_trans_create(env, mdd);
1640 RETURN(PTR_ERR(handle));
1642 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1643 lname, ma, handle, no_name, is_dir);
1647 rc = mdd_trans_start(env, mdd, handle);
1651 if (likely(mdd_cobj != NULL))
1652 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1654 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1655 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1660 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1661 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1662 GOTO(cleanup, rc = 0);
1664 if (likely(mdd_cobj != NULL)) {
1665 rc = mdo_ref_del(env, mdd_cobj, handle);
1667 __mdd_index_insert_only(env, mdd_pobj,
1669 mdd_object_type(mdd_cobj),
1676 mdo_ref_del(env, mdd_cobj, handle);
1678 /* fetch updated nlink */
1679 rc = mdd_la_get(env, mdd_cobj, cattr);
1684 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1685 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1687 la->la_valid = LA_CTIME | LA_MTIME;
1688 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1692 /* Enough for only unlink the entry */
1693 if (unlikely(mdd_cobj == NULL))
1696 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1697 /* update ctime of an unlinked file only if it is still
1698 * opened or a link still exists */
1699 la->la_valid = LA_CTIME;
1700 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1705 /* XXX: this transfer to ma will be removed with LOD/OSP */
1706 ma->ma_attr = *cattr;
1707 ma->ma_valid |= MA_INODE;
1708 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1712 /* fetch updated nlink */
1713 rc = mdd_la_get(env, mdd_cobj, cattr);
1714 /* if object is removed then we can't get its attrs,
1716 if (rc == -ENOENT) {
1717 cattr->la_nlink = 0;
1721 if (cattr->la_nlink == 0) {
1722 ma->ma_attr = *cattr;
1723 ma->ma_valid |= MA_INODE;
1728 if (likely(mdd_cobj != NULL))
1729 mdd_write_unlock(env, mdd_cobj);
1732 if (cattr->la_nlink == 0)
1733 cl_flags |= CLF_UNLINK_LAST;
1735 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1737 rc = mdd_changelog_ns_store(env, mdd,
1738 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1739 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1744 rc = mdd_trans_stop(env, mdd, rc, handle);
1750 * The permission has been checked when obj created, no need check again.
1752 static int mdd_cd_sanity_check(const struct lu_env *env,
1753 struct mdd_object *obj)
1758 if (!obj || mdd_is_dead_obj(obj))
1764 static int mdd_create_data(const struct lu_env *env,
1765 struct md_object *pobj,
1766 struct md_object *cobj,
1767 const struct md_op_spec *spec,
1770 struct mdd_device *mdd = mdo2mdd(cobj);
1771 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1772 struct mdd_object *son = md2mdd_obj(cobj);
1773 struct thandle *handle;
1774 const struct lu_buf *buf;
1775 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1776 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1780 rc = mdd_cd_sanity_check(env, son);
1784 if (!md_should_create(spec->sp_cr_flags))
1788 * there are following use cases for this function:
1789 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1790 * striping can be specified or not
1793 rc = mdd_la_get(env, son, attr);
1797 /* calling ->ah_make_hint() is used to transfer information from parent */
1798 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1800 handle = mdd_trans_create(env, mdd);
1802 GOTO(out_free, rc = PTR_ERR(handle));
1805 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1806 * Should this be fixed?
1808 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1809 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1810 spec->sp_cr_flags, spec->no_create);
1812 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1813 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1814 spec->u.sp_ea.eadatalen);
1819 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1820 XATTR_NAME_LOV, 0, handle);
1824 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1828 rc = mdd_trans_start(env, mdd, handle);
1832 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1838 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1841 rc = mdd_trans_stop(env, mdd, rc, handle);
1847 static int mdd_declare_object_initialize(const struct lu_env *env,
1848 struct mdd_object *parent,
1849 struct mdd_object *child,
1850 const struct lu_attr *attr,
1851 struct thandle *handle)
1856 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1857 if (!S_ISDIR(attr->la_mode))
1860 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1865 rc = mdo_declare_ref_add(env, child, handle);
1869 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1875 static int mdd_object_initialize(const struct lu_env *env,
1876 const struct lu_fid *pfid,
1877 struct mdd_object *child,
1878 struct lu_attr *attr, struct thandle *handle,
1879 const struct md_op_spec *spec)
1884 if (S_ISDIR(attr->la_mode)) {
1885 /* Add "." and ".." for newly created dir */
1886 mdo_ref_add(env, child, handle);
1887 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1888 S_IFDIR, dot, handle);
1890 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1893 mdo_ref_del(env, child, handle);
1900 * This function checks whether it can create a file/dir under the
1901 * directory(@pobj). The directory(@pobj) is not being locked by
1904 * \param[in] env execution environment
1905 * \param[in] pobj the directory to create files
1906 * \param[in] pattr the attributes of the directory
1907 * \param[in] lname the name of the created file/dir
1908 * \param[in] cattr the attributes of the file/dir
1909 * \param[in] spec create specification
1911 * \retval = 0 it is allowed to create file/dir under
1913 * \retval negative error not allowed to create file/dir
1914 * under the directory
1916 static int mdd_create_sanity_check(const struct lu_env *env,
1917 struct md_object *pobj,
1918 const struct lu_attr *pattr,
1919 const struct lu_name *lname,
1920 struct lu_attr *cattr,
1921 struct md_op_spec *spec)
1923 struct mdd_thread_info *info = mdd_env_info(env);
1924 struct lu_fid *fid = &info->mti_fid;
1925 struct mdd_object *obj = md2mdd_obj(pobj);
1926 struct mdd_device *m = mdo2mdd(pobj);
1927 bool check_perm = true;
1932 if (mdd_is_dead_obj(obj))
1936 * In some cases this lookup is not needed - we know before if name
1937 * exists or not because MDT performs lookup for it.
1938 * name length check is done in lookup.
1940 if (spec->sp_cr_lookup) {
1942 * Check if the name already exist, though it will be checked in
1943 * _index_insert also, for avoiding rolling back if exists
1946 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
1947 MAY_WRITE | MAY_EXEC);
1949 RETURN(rc ? : -EEXIST);
1951 /* Permission is already being checked in mdd_lookup */
1955 if (S_ISDIR(cattr->la_mode) &&
1956 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
1957 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
1958 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
1960 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC) &&
1961 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
1963 CERROR("%s: invalid lmv_user_md: magic = %x, "
1964 "stripe_offset = %d, stripe_count = %u: "
1965 "rc = %d\n", mdd2obd_dev(m)->obd_name,
1966 le32_to_cpu(lum->lum_magic),
1967 (int)le32_to_cpu(lum->lum_stripe_offset),
1968 le32_to_cpu(lum->lum_stripe_count), rc);
1973 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
1978 if (pattr->la_mode & S_ISGID) {
1979 cattr->la_gid = pattr->la_gid;
1980 if (S_ISDIR(cattr->la_mode)) {
1981 cattr->la_mode |= S_ISGID;
1982 cattr->la_valid |= LA_MODE;
1986 /* Inherit project ID from parent directory */
1987 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
1988 cattr->la_projid = pattr->la_projid;
1989 if (S_ISDIR(cattr->la_mode)) {
1990 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
1991 cattr->la_valid |= LA_FLAGS;
1993 cattr->la_valid |= LA_PROJID;
1996 rc = mdd_name_check(m, lname);
2000 switch (cattr->la_mode & S_IFMT) {
2002 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2004 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2005 RETURN(-ENAMETOOLONG);
2024 static int mdd_declare_create_object(const struct lu_env *env,
2025 struct mdd_device *mdd,
2026 struct mdd_object *p, struct mdd_object *c,
2027 struct lu_attr *attr,
2028 struct thandle *handle,
2029 const struct md_op_spec *spec,
2030 struct lu_buf *def_acl_buf,
2031 struct lu_buf *acl_buf,
2032 struct dt_allocation_hint *hint)
2034 const struct lu_buf *buf;
2037 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2042 #ifdef CONFIG_FS_POSIX_ACL
2043 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2044 /* if dir, then can inherit default ACl */
2045 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2046 XATTR_NAME_ACL_DEFAULT,
2052 if (acl_buf->lb_len > 0) {
2053 rc = mdo_declare_attr_set(env, c, attr, handle);
2057 rc = mdo_declare_xattr_set(env, c, acl_buf,
2058 XATTR_NAME_ACL_ACCESS, 0, handle);
2063 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2067 /* replay case, create LOV EA from client data */
2068 if (spec->no_create ||
2069 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2070 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2071 spec->u.sp_ea.eadatalen);
2072 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2078 if (S_ISLNK(attr->la_mode)) {
2079 const char *target_name = spec->u.sp_symname;
2080 int sym_len = strlen(target_name);
2081 const struct lu_buf *buf;
2083 buf = mdd_buf_get_const(env, target_name, sym_len);
2084 rc = dt_declare_record_write(env, mdd_object_child(c),
2090 if (spec->sp_cr_file_secctx_name != NULL) {
2091 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2092 spec->sp_cr_file_secctx_size);
2093 rc = mdo_declare_xattr_set(env, c, buf,
2094 spec->sp_cr_file_secctx_name, 0,
2103 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2104 struct mdd_object *p, struct mdd_object *c,
2105 const struct lu_name *name,
2106 struct lu_attr *attr,
2107 struct thandle *handle,
2108 const struct md_op_spec *spec,
2109 struct linkea_data *ldata,
2110 struct lu_buf *def_acl_buf,
2111 struct lu_buf *acl_buf,
2112 struct dt_allocation_hint *hint)
2116 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2117 def_acl_buf, acl_buf, hint);
2121 if (S_ISDIR(attr->la_mode)) {
2122 rc = mdo_declare_ref_add(env, p, handle);
2127 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2128 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2132 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2134 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2135 name->ln_name, handle);
2139 rc = mdd_declare_links_add(env, c, handle, ldata);
2144 la->la_valid = LA_CTIME | LA_MTIME;
2145 rc = mdo_declare_attr_set(env, p, la, handle);
2149 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2157 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2158 struct lu_attr *la, struct lu_buf *def_acl_buf,
2159 struct lu_buf *acl_buf)
2164 if (S_ISLNK(la->la_mode)) {
2165 acl_buf->lb_len = 0;
2166 def_acl_buf->lb_len = 0;
2170 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2171 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2172 XATTR_NAME_ACL_DEFAULT);
2173 mdd_read_unlock(env, pobj);
2175 /* If there are default ACL, fix mode/ACL by default ACL */
2176 def_acl_buf->lb_len = rc;
2177 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2178 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2179 acl_buf->lb_len = rc;
2180 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2183 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2184 /* If there are no default ACL, fix mode by mask */
2185 struct lu_ucred *uc = lu_ucred(env);
2187 /* The create triggered by MDT internal events, such as
2188 * LFSCK reset, will not contain valid "uc". */
2189 if (unlikely(uc != NULL))
2190 la->la_mode &= ~uc->uc_umask;
2192 acl_buf->lb_len = 0;
2193 def_acl_buf->lb_len = 0;
2200 * Create a metadata object and initialize it, set acl, xattr.
2202 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2203 struct mdd_object *son, struct lu_attr *attr,
2204 struct md_op_spec *spec, struct lu_buf *acl_buf,
2205 struct lu_buf *def_acl_buf,
2206 struct dt_allocation_hint *hint,
2207 struct thandle *handle)
2209 const struct lu_buf *buf;
2212 mdd_write_lock(env, son, MOR_TGT_CHILD);
2213 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2218 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2219 * created in declare phase, they also needs to be added to master
2220 * object as sub-directory entry. So it has to initialize the master
2221 * object, then set dir striped EA.(in mdo_xattr_set) */
2222 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2225 GOTO(err_destroy, rc);
2228 * in case of replay we just set LOVEA provided by the client
2229 * XXX: I think it would be interesting to try "old" way where
2230 * MDT calls this xattr_set(LOV) in a different transaction.
2231 * probably this way we code can be made better.
2234 /* During creation, there are only a few cases we need do xattr_set to
2236 * 1. regular file: see comments above.
2237 * 2. dir: inherit default striping or pool settings from parent.
2238 * 3. create striped directory with provided stripeEA.
2239 * 4. create striped directory because inherit default layout from the
2242 if (spec->no_create ||
2243 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2244 S_ISDIR(attr->la_mode)) {
2245 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2246 spec->u.sp_ea.eadatalen);
2247 rc = mdo_xattr_set(env, son, buf,
2248 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2252 GOTO(err_destroy, rc);
2255 #ifdef CONFIG_FS_POSIX_ACL
2256 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2257 S_ISDIR(attr->la_mode)) {
2258 /* set default acl */
2259 rc = mdo_xattr_set(env, son, def_acl_buf,
2260 XATTR_NAME_ACL_DEFAULT, 0,
2263 GOTO(err_destroy, rc);
2265 /* set its own acl */
2266 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2267 rc = mdo_xattr_set(env, son, acl_buf,
2268 XATTR_NAME_ACL_ACCESS,
2271 GOTO(err_destroy, rc);
2275 if (S_ISLNK(attr->la_mode)) {
2276 struct lu_ucred *uc = lu_ucred_assert(env);
2277 struct dt_object *dt = mdd_object_child(son);
2278 const char *target_name = spec->u.sp_symname;
2279 int sym_len = strlen(target_name);
2282 buf = mdd_buf_get_const(env, target_name, sym_len);
2283 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2285 CFS_CAP_SYS_RESOURCE_MASK);
2290 GOTO(err_initlized, rc = -EFAULT);
2293 if (spec->sp_cr_file_secctx_name != NULL) {
2294 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2295 spec->sp_cr_file_secctx_size);
2296 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2299 GOTO(err_initlized, rc);
2303 if (unlikely(rc != 0)) {
2305 if (S_ISDIR(attr->la_mode)) {
2306 /* Drop the reference, no need to delete "."/"..",
2307 * because the object to be destroied directly. */
2308 rc2 = mdo_ref_del(env, son, handle);
2312 rc2 = mdo_ref_del(env, son, handle);
2316 mdo_destroy(env, son, handle);
2319 mdd_write_unlock(env, son);
2323 static int mdd_index_delete(const struct lu_env *env,
2324 struct mdd_object *mdd_pobj,
2325 struct lu_attr *cattr,
2326 const struct lu_name *lname)
2328 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2329 struct thandle *handle;
2333 handle = mdd_trans_create(env, mdd);
2335 RETURN(PTR_ERR(handle));
2337 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2342 if (S_ISDIR(cattr->la_mode)) {
2343 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2348 /* Since this will only be used in the error handler path,
2349 * Let's set the thandle to be local and not mess the transno */
2350 handle->th_local = 1;
2351 rc = mdd_trans_start(env, mdd, handle);
2355 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2356 S_ISDIR(cattr->la_mode), handle);
2360 rc = mdd_trans_stop(env, mdd, rc, handle);
2366 * Create object and insert it into namespace.
2368 * Two operations have to be performed:
2370 * - an allocation of a new object (->do_create()), and
2371 * - an insertion into a parent index (->dio_insert()).
2373 * Due to locking, operation order is not important, when both are
2374 * successful, *but* error handling cases are quite different:
2376 * - if insertion is done first, and following object creation fails,
2377 * insertion has to be rolled back, but this operation might fail
2378 * also leaving us with dangling index entry.
2380 * - if creation is done first, is has to be undone if insertion fails,
2381 * leaving us with leaked space, which is not good but not fatal.
2383 * It seems that creation-first is simplest solution, but it is sub-optimal
2389 * case, because second mkdir is bound to create object, only to
2390 * destroy it immediately.
2392 * To avoid this follow local file systems that do double lookup:
2394 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2395 * 1. create (mdd_create_object_internal())
2396 * 2. insert (__mdd_index_insert(), lookup again)
2398 * \param[in] pobj parent object
2399 * \param[in] lname name of child being created
2400 * \param[in,out] child child object being created
2401 * \param[in] spec additional create parameters
2402 * \param[in] ma attributes for new child object
2404 * \retval 0 on success
2405 * \retval negative errno on failure
2407 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2408 const struct lu_name *lname, struct md_object *child,
2409 struct md_op_spec *spec, struct md_attr *ma)
2411 struct mdd_thread_info *info = mdd_env_info(env);
2412 struct lu_attr *la = &info->mti_la_for_fix;
2413 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2414 struct mdd_object *son = md2mdd_obj(child);
2415 struct mdd_device *mdd = mdo2mdd(pobj);
2416 struct lu_attr *attr = &ma->ma_attr;
2417 struct thandle *handle;
2418 struct lu_attr *pattr = &info->mti_pattr;
2419 struct lu_buf acl_buf;
2420 struct lu_buf def_acl_buf;
2421 struct linkea_data *ldata = &info->mti_link_data;
2422 const char *name = lname->ln_name;
2423 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2428 rc = mdd_la_get(env, mdd_pobj, pattr);
2432 /* Sanity checks before big job. */
2433 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2437 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2438 GOTO(out_free, rc = -EINPROGRESS);
2440 handle = mdd_trans_create(env, mdd);
2442 GOTO(out_free, rc = PTR_ERR(handle));
2444 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2445 mdd->mdd_dt_conf.ddp_max_ea_size);
2446 acl_buf = info->mti_xattr_buf;
2447 def_acl_buf.lb_buf = info->mti_key;
2448 def_acl_buf.lb_len = sizeof(info->mti_key);
2449 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2453 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2455 memset(ldata, 0, sizeof(*ldata));
2456 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2457 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2460 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2461 &tfid, lname, 1, 0, ldata);
2463 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2464 mdd_object_fid(mdd_pobj),
2465 lname, 1, 0, ldata);
2468 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2469 handle, spec, ldata, &def_acl_buf, &acl_buf,
2474 rc = mdd_trans_start(env, mdd, handle);
2478 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2479 &def_acl_buf, hint, handle);
2483 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2484 mdd_write_lock(env, son, MOR_TGT_CHILD);
2485 son->mod_flags |= VOLATILE_OBJ;
2486 rc = __mdd_orphan_add(env, son, handle);
2487 GOTO(out_volatile, rc);
2489 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2490 attr->la_mode, name, handle);
2492 GOTO(err_created, rc);
2494 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2497 /* update parent directory mtime/ctime */
2499 la->la_valid = LA_CTIME | LA_MTIME;
2500 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2502 GOTO(err_insert, rc);
2508 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2509 rc2 = __mdd_orphan_del(env, son, handle);
2511 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2512 S_ISDIR(attr->la_mode),
2518 mdd_write_lock(env, son, MOR_TGT_CHILD);
2519 if (S_ISDIR(attr->la_mode)) {
2520 /* Drop the reference, no need to delete "."/"..",
2521 * because the object is to be destroyed directly. */
2522 rc2 = mdo_ref_del(env, son, handle);
2524 mdd_write_unlock(env, son);
2529 /* For volatile files drop one link immediately, since there is
2530 * no filename in the namespace, and save any error returned. */
2531 rc2 = mdo_ref_del(env, son, handle);
2533 mdd_write_unlock(env, son);
2534 if (unlikely(rc == 0))
2539 /* Don't destroy the volatile object on success */
2540 if (likely(rc != 0))
2541 mdo_destroy(env, son, handle);
2542 mdd_write_unlock(env, son);
2545 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2546 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2547 rc = mdd_changelog_ns_store(env, mdd,
2548 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2549 S_ISREG(attr->la_mode) ? CL_CREATE :
2550 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2551 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2554 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2556 /* If creation fails, it is most likely due to the remote update
2557 * failure, because local transaction will mostly succeed at
2558 * this stage. There is no easy way to rollback all of previous
2559 * updates, so let's remove the object from namespace, and
2560 * LFSCK should handle the orphan object. */
2561 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2562 mdd_index_delete(env, mdd_pobj, attr, lname);
2566 if (is_vmalloc_addr(ldata->ld_buf))
2567 /* if we vmalloced a large buffer drop it */
2568 lu_buf_free(ldata->ld_buf);
2570 /* The child object shouldn't be cached anymore */
2572 set_bit(LU_OBJECT_HEARD_BANSHEE,
2573 &child->mo_lu.lo_header->loh_flags);
2578 * Get locks on parents in proper order
2579 * RETURN: < 0 - error, rename_order if successful
2587 static int mdd_rename_order(const struct lu_env *env,
2588 struct mdd_device *mdd,
2589 struct mdd_object *src_pobj,
2590 const struct lu_attr *pattr,
2591 struct mdd_object *tgt_pobj)
2593 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2597 if (src_pobj == tgt_pobj)
2598 RETURN(MDD_RN_SAME);
2600 /* compared the parent child relationship of src_p&tgt_p */
2601 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2603 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2606 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2620 /* has not mdd_write{read}_lock on any obj yet. */
2621 static int mdd_rename_sanity_check(const struct lu_env *env,
2622 struct mdd_object *src_pobj,
2623 const struct lu_attr *pattr,
2624 struct mdd_object *tgt_pobj,
2625 const struct lu_attr *tpattr,
2626 struct mdd_object *sobj,
2627 const struct lu_attr *cattr,
2628 struct mdd_object *tobj,
2629 const struct lu_attr *tattr)
2634 /* XXX: when get here, sobj must NOT be NULL,
2635 * the other case has been processed in cld_rename
2636 * before mdd_rename and enable MDS_PERM_BYPASS. */
2640 * If we are using project inheritance, we only allow renames
2641 * into our tree when the project IDs are the same; otherwise
2642 * tree quota mechanism would be circumvented.
2644 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2645 tpattr->la_projid != cattr->la_projid) ||
2646 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2647 (pattr->la_projid != tpattr->la_projid)))
2650 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2654 /* XXX: when get here, "tobj == NULL" means tobj must
2655 * NOT exist (neither on remote MDS, such case has been
2656 * processed in cld_rename before mdd_rename and enable
2658 * So check may_create, but not check may_unlink. */
2660 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2661 (src_pobj != tgt_pobj));
2663 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2664 (src_pobj != tgt_pobj), 1);
2666 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2667 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2672 static int mdd_declare_rename(const struct lu_env *env,
2673 struct mdd_device *mdd,
2674 struct mdd_object *mdd_spobj,
2675 struct mdd_object *mdd_tpobj,
2676 struct mdd_object *mdd_sobj,
2677 struct mdd_object *mdd_tobj,
2678 const struct lu_name *sname,
2679 const struct lu_name *tname,
2681 struct linkea_data *ldata,
2682 struct thandle *handle)
2684 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2687 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2688 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2694 /* name from source dir */
2695 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2699 /* .. from source child */
2700 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2701 /* source child can be directory,
2702 * counted by source dir's nlink */
2703 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2706 if (mdd_spobj != mdd_tpobj) {
2707 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2712 rc = mdo_declare_index_insert(env, mdd_sobj,
2714 S_IFDIR, dotdot, handle);
2719 /* new target child can be directory,
2720 * counted by target dir's nlink */
2721 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2726 la->la_valid = LA_CTIME | LA_MTIME;
2727 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2731 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2735 la->la_valid = LA_CTIME;
2736 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2740 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2745 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2746 mdd_object_type(mdd_sobj),
2747 tname->ln_name, handle);
2751 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2752 /* delete target child in target parent directory */
2753 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2758 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2762 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2763 /* target child can be directory,
2764 * delete "." reference in target child directory */
2765 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2769 /* delete ".." reference in target parent directory */
2770 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2775 la->la_valid = LA_CTIME;
2776 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2780 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2785 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2792 /* src object can be remote that is why we use only fid and type of object */
2793 static int mdd_rename(const struct lu_env *env,
2794 struct md_object *src_pobj, struct md_object *tgt_pobj,
2795 const struct lu_fid *lf, const struct lu_name *lsname,
2796 struct md_object *tobj, const struct lu_name *ltname,
2799 const char *sname = lsname->ln_name;
2800 const char *tname = ltname->ln_name;
2801 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2802 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2803 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2804 struct mdd_device *mdd = mdo2mdd(src_pobj);
2805 struct mdd_object *mdd_sobj = NULL; /* source object */
2806 struct mdd_object *mdd_tobj = NULL;
2807 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2808 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2809 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2810 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2811 struct thandle *handle;
2812 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2813 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2814 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2817 bool tobj_locked = 0;
2818 unsigned cl_flags = 0;
2823 mdd_tobj = md2mdd_obj(tobj);
2825 mdd_sobj = mdd_object_find(env, mdd, lf);
2826 if (IS_ERR(mdd_sobj))
2827 RETURN(PTR_ERR(mdd_sobj));
2829 rc = mdd_la_get(env, mdd_sobj, cattr);
2831 GOTO(out_pending, rc);
2833 rc = mdd_la_get(env, mdd_spobj, pattr);
2835 GOTO(out_pending, rc);
2838 rc = mdd_la_get(env, mdd_tobj, tattr);
2840 GOTO(out_pending, rc);
2841 /* search for an existing archive.
2842 * we should check ahead as the object
2843 * can be destroyed in this transaction */
2844 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2845 cl_flags |= CLF_RENAME_LAST_EXISTS;
2848 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2850 GOTO(out_pending, rc);
2852 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2853 mdd_sobj, cattr, mdd_tobj, tattr);
2855 GOTO(out_pending, rc);
2857 rc = mdd_name_check(mdd, ltname);
2859 GOTO(out_pending, rc);
2861 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2862 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2864 GOTO(out_pending, rc);
2866 handle = mdd_trans_create(env, mdd);
2868 GOTO(out_pending, rc = PTR_ERR(handle));
2870 memset(ldata, 0, sizeof(*ldata));
2871 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
2872 lsname, mdd_object_fid(mdd_tpobj), ltname,
2877 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2878 mdd_tobj, lsname, ltname, ma, ldata, handle);
2882 rc = mdd_trans_start(env, mdd, handle);
2886 is_dir = S_ISDIR(cattr->la_mode);
2888 /* Remove source name from source directory */
2889 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2893 /* "mv dir1 dir2" needs "dir1/.." link update */
2894 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2895 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2897 GOTO(fixup_spobj2, rc);
2899 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2902 GOTO(fixup_spobj, rc);
2905 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2906 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2908 /* tname might been renamed to something else */
2909 GOTO(fixup_spobj, rc);
2912 /* Insert new fid with target name into target dir */
2913 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
2916 GOTO(fixup_tpobj, rc);
2918 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2919 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2921 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2922 la->la_valid = LA_CTIME;
2923 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
2925 GOTO(fixup_tpobj, rc);
2927 /* Update the linkEA for the source object */
2928 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2929 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2930 mdo2fid(mdd_tpobj), ltname, handle, ldata,
2933 /* Old files might not have EA entry */
2934 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2935 lsname, handle, NULL, 0);
2936 mdd_write_unlock(env, mdd_sobj);
2937 /* We don't fail the transaction if the link ea can't be
2938 updated -- fid2path will use alternate lookup method. */
2941 /* Remove old target object
2942 * For tobj is remote case cmm layer has processed
2943 * and set tobj to NULL then. So when tobj is NOT NULL,
2944 * it must be local one.
2946 if (tobj && mdd_object_exists(mdd_tobj)) {
2947 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2949 if (mdd_is_dead_obj(mdd_tobj)) {
2950 /* shld not be dead, something is wrong */
2951 CERROR("tobj is dead, something is wrong\n");
2955 mdo_ref_del(env, mdd_tobj, handle);
2957 /* Remove dot reference. */
2958 if (S_ISDIR(tattr->la_mode))
2959 mdo_ref_del(env, mdd_tobj, handle);
2962 /* fetch updated nlink */
2963 rc = mdd_la_get(env, mdd_tobj, tattr);
2965 CERROR("%s: Failed to get nlink for tobj "
2967 mdd2obd_dev(mdd)->obd_name,
2968 PFID(tpobj_fid), rc);
2969 GOTO(fixup_tpobj, rc);
2972 la->la_valid = LA_CTIME;
2973 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
2975 CERROR("%s: Failed to set ctime for tobj "
2977 mdd2obd_dev(mdd)->obd_name,
2978 PFID(tpobj_fid), rc);
2979 GOTO(fixup_tpobj, rc);
2982 /* XXX: this transfer to ma will be removed with LOD/OSP */
2983 ma->ma_attr = *tattr;
2984 ma->ma_valid |= MA_INODE;
2985 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
2988 CERROR("%s: Failed to unlink tobj "
2990 mdd2obd_dev(mdd)->obd_name,
2991 PFID(tpobj_fid), rc);
2992 GOTO(fixup_tpobj, rc);
2995 /* fetch updated nlink */
2996 rc = mdd_la_get(env, mdd_tobj, tattr);
2997 if (rc == -ENOENT) {
2998 /* the object got removed, let's
2999 * return the latest known attributes */
3000 tattr->la_nlink = 0;
3002 } else if (rc != 0) {
3003 CERROR("%s: Failed to get nlink for tobj "
3005 mdd2obd_dev(mdd)->obd_name,
3006 PFID(tpobj_fid), rc);
3007 GOTO(fixup_tpobj, rc);
3009 /* XXX: this transfer to ma will be removed with LOD/OSP */
3010 ma->ma_attr = *tattr;
3011 ma->ma_valid |= MA_INODE;
3013 if (tattr->la_nlink == 0)
3014 cl_flags |= CLF_RENAME_LAST;
3016 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3019 la->la_valid = LA_CTIME | LA_MTIME;
3020 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3022 GOTO(fixup_tpobj, rc);
3024 if (mdd_spobj != mdd_tpobj) {
3025 la->la_valid = LA_CTIME | LA_MTIME;
3026 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3028 GOTO(fixup_tpobj, rc);
3035 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3037 CWARN("tp obj fix error %d\n",rc2);
3039 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3040 !mdd_is_dead_obj(mdd_tobj)) {
3042 mdo_ref_add(env, mdd_tobj, handle);
3044 mdo_ref_add(env, mdd_tobj, handle);
3047 rc2 = __mdd_index_insert(env, mdd_tpobj,
3049 mdd_object_type(mdd_tobj),
3052 CWARN("tp obj fix error: rc = %d\n", rc2);
3057 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3058 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3060 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3061 mdd2obd_dev(mdd)->obd_name, rc2);
3064 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3067 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3068 mdd2obd_dev(mdd)->obd_name, rc2);
3073 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3074 mdd_object_type(mdd_sobj), sname,
3077 CWARN("sp obj fix error: rc = %d\n", rc2);
3082 mdd_write_unlock(env, mdd_tobj);
3085 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3086 mdd_tobj, tpobj_fid, lf, spobj_fid,
3087 ltname, lsname, handle);
3090 rc = mdd_trans_stop(env, mdd, rc, handle);
3093 mdd_object_put(env, mdd_sobj);
3098 * During migration once the parent FID has been changed,
3099 * we need update the parent FID in linkea.
3101 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3102 struct mdd_object *parent,
3103 struct mdd_object *newparent,
3104 struct mdd_object *child,
3105 const char *name, int namelen,
3106 struct thandle *handle,
3109 struct mdd_thread_info *info = mdd_env_info(env);
3110 struct linkea_data ldata = { NULL };
3111 struct lu_buf *buf = &info->mti_link_buf;
3117 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3118 if (buf->lb_buf == NULL)
3122 rc = mdd_links_read(env, child, &ldata);
3124 if (rc == -ENOENT || rc == -ENODATA)
3129 LASSERT(ldata.ld_leh != NULL);
3130 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3131 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3132 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3133 struct lu_name lname;
3136 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3139 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3140 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3141 ldata.ld_lee = (struct link_ea_entry *)
3142 ((char *)ldata.ld_lee +
3147 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3148 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3149 lname.ln_namelen, lname.ln_name,
3150 PFID(mdd_object_fid(newparent)));
3151 /* update to the new parent fid */
3152 linkea_entry_pack(ldata.ld_lee, &lname,
3153 mdd_object_fid(newparent));
3155 rc = mdd_declare_links_add(env, child, handle, &ldata);
3157 rc = mdd_links_write(env, child, &ldata, handle);
3163 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3164 struct mdd_object *parent,
3165 struct mdd_object *newparent,
3166 struct mdd_object *child,
3167 const char *name, int namelen,
3168 struct thandle *handle)
3170 return mdd_linkea_update_child_internal(env, parent, newparent,
3172 namelen, handle, true);
3175 static int mdd_linkea_update_child(const struct lu_env *env,
3176 struct mdd_object *parent,
3177 struct mdd_object *newparent,
3178 struct mdd_object *child,
3179 const char *name, int namelen,
3180 struct thandle *handle)
3182 return mdd_linkea_update_child_internal(env, parent, newparent,
3184 namelen, handle, false);
3187 static int mdd_update_linkea_internal(const struct lu_env *env,
3188 struct mdd_object *mdd_pobj,
3189 struct mdd_object *mdd_sobj,
3190 struct mdd_object *mdd_tobj,
3191 const struct lu_name *child_name,
3192 struct linkea_data *ldata,
3193 struct thandle *handle,
3196 struct mdd_thread_info *info = mdd_env_info(env);
3201 LASSERT(ldata->ld_buf != NULL);
3202 LASSERT(ldata->ld_leh != NULL);
3204 /* If it is mulitple links file, we need update the name entry for
3206 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3207 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3208 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3209 struct mdd_object *pobj;
3210 struct lu_name lname;
3213 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3215 pobj = mdd_object_find(env, mdd, &fid);
3217 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3218 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3223 if (!mdd_object_exists(pobj)) {
3224 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3225 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3229 if (pobj == mdd_pobj &&
3230 lname.ln_namelen == child_name->ln_namelen &&
3231 strncmp(lname.ln_name, child_name->ln_name,
3232 lname.ln_namelen) == 0) {
3233 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3234 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3239 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3240 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3241 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3244 /* Remove source name from source directory */
3245 /* Insert new fid with target name into target dir */
3246 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3251 rc = mdo_declare_index_insert(env, pobj,
3252 mdd_object_fid(mdd_tobj),
3253 mdd_object_type(mdd_tobj),
3254 lname.ln_name, handle);
3258 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3262 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3266 char *tmp_name = info->mti_key;
3268 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3269 /* lnamelen is too big(> NAME_MAX + 16),
3270 * something wrong about this linkea, let's
3272 CWARN("%s: the name %.*s is too long under "
3273 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3274 lname.ln_namelen, lname.ln_name,
3279 /* Note: lname might be without \0 at the end, see
3280 * linkea_entry_unpack(), let's add extra \0 by
3282 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3283 lname.ln_namelen, lname.ln_name);
3284 lname.ln_name = tmp_name;
3286 /* Let's check if this linkEA still valid, before
3287 * it might be packed into the RPC buffer. */
3288 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3289 &info->mti_fid, NULL);
3290 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3291 mdd_object_fid(mdd_sobj)))
3292 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3294 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3298 rc = __mdd_index_insert(env, pobj,
3299 mdd_object_fid(mdd_tobj),
3300 mdd_object_type(mdd_tobj),
3305 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3306 rc = mdo_ref_add(env, mdd_tobj, handle);
3307 mdd_write_unlock(env, mdd_tobj);
3311 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3312 mdo_ref_del(env, mdd_sobj, handle);
3313 mdd_write_unlock(env, mdd_sobj);
3316 mdd_object_put(env, pobj);
3320 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3327 static int mdd_migrate_xattrs(const struct lu_env *env,
3328 struct mdd_object *mdd_sobj,
3329 struct mdd_object *mdd_tobj)
3331 struct mdd_thread_info *info = mdd_env_info(env);
3332 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3334 struct thandle *handle;
3340 struct lu_buf list_xbuf;
3343 /* retrieve xattr list from the old object */
3344 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3345 if (list_xsize == -ENODATA)
3351 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3352 if (info->mti_big_buf.lb_buf == NULL)
3355 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3356 list_xbuf.lb_len = list_xsize;
3357 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3362 xname = list_xbuf.lb_buf;
3364 xlen = strnlen(xname, rem - 1) + 1;
3365 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3366 strcmp(XATTR_NAME_LMV, xname) == 0)
3369 /* For directory, if there are default layout, migrate here */
3370 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3371 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3374 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3375 if (xsize == -ENODATA)
3380 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3381 if (info->mti_link_buf.lb_buf == NULL)
3382 GOTO(out, rc = -ENOMEM);
3384 xbuf.lb_len = xsize;
3385 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3386 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3392 handle = mdd_trans_create(env, mdd);
3394 GOTO(out, rc = PTR_ERR(handle));
3396 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3399 GOTO(stop_trans, rc);
3400 /* Note: this transaction is part of migration, and it is not
3401 * the last step of migration, so we set th_local = 1 to avoid
3402 * update last rcvd for this transaction */
3403 handle->th_local = 1;
3404 rc = mdd_trans_start(env, mdd, handle);
3406 GOTO(stop_trans, rc);
3409 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3411 GOTO(stop_trans, rc = 0);
3413 if (unlikely(rc == -ENOSPC &&
3414 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3415 rc = linkea_overflow_shrink(
3416 (struct linkea_data *)(xbuf.lb_buf));
3417 if (likely(rc > 0)) {
3424 GOTO(stop_trans, rc);
3426 rc = mdd_trans_stop(env, mdd, rc, handle);
3431 memmove(xname, xname + xlen, rem);
3437 static int mdd_declare_migrate_create(const struct lu_env *env,
3438 struct mdd_object *mdd_pobj,
3439 struct mdd_object *mdd_sobj,
3440 struct mdd_object *mdd_tobj,
3441 struct md_op_spec *spec,
3443 union lmv_mds_md *mgr_ea,
3444 struct linkea_data *ldata,
3445 struct thandle *handle)
3447 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3448 const struct lu_buf *buf;
3452 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3453 handle, spec, NULL);
3457 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3462 if (S_ISLNK(la->la_mode)) {
3463 const char *target_name = spec->u.sp_symname;
3464 int sym_len = strlen(target_name);
3465 const struct lu_buf *buf;
3467 buf = mdd_buf_get_const(env, target_name, sym_len);
3468 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3472 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3473 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3478 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3479 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3480 spec->u.sp_ea.eadatalen);
3481 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3487 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3488 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3489 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3494 la_flag->la_valid = LA_FLAGS;
3495 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3496 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3501 static int mdd_migrate_create(const struct lu_env *env,
3502 struct mdd_object *mdd_pobj,
3503 struct mdd_object *mdd_sobj,
3504 struct mdd_object *mdd_tobj,
3505 const struct lu_name *lname,
3508 struct mdd_thread_info *info = mdd_env_info(env);
3509 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3510 struct md_op_spec *spec = &info->mti_spec;
3511 struct lu_buf lmm_buf = { NULL };
3512 struct lu_buf link_buf = { NULL };
3513 struct lu_buf mgr_buf;
3514 struct thandle *handle;
3515 struct lmv_mds_md_v1 *mgr_ea;
3516 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3517 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3519 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3523 /* prepare spec for create */
3524 memset(spec, 0, sizeof(*spec));
3525 spec->sp_cr_lookup = 0;
3526 spec->sp_feat = &dt_directory_features;
3527 if (S_ISLNK(la->la_mode)) {
3528 const struct lu_buf *buf;
3530 buf = lu_buf_check_and_alloc(
3531 &mdd_env_info(env)->mti_big_buf,
3534 link_buf.lb_len = la->la_size + 1;
3535 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3536 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3538 rc = rc != 0 ? rc : -EFAULT;
3539 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3540 mdd2obd_dev(mdd)->obd_name,
3541 PFID(mdd_object_fid(mdd_sobj)), rc);
3544 spec->u.sp_symname = link_buf.lb_buf;
3545 } else if (S_ISREG(la->la_mode)) {
3546 /* retrieve lov of the old object */
3547 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3548 if (rc != 0 && rc != -ENODATA)
3550 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3551 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3552 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3553 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3555 } else if (S_ISDIR(la->la_mode)) {
3556 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3557 if (rc == -ENODATA) {
3558 /* ignore the non-linkEA error */
3566 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3567 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3568 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3569 mgr_buf.lb_len = mgr_easize;
3570 mgr_ea = mgr_buf.lb_buf;
3571 memset(mgr_ea, 0, sizeof(*mgr_ea));
3572 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3573 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3574 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3575 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3576 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3577 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3579 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3581 handle = mdd_trans_create(env, mdd);
3583 GOTO(out_free, rc = PTR_ERR(handle));
3585 /* Note: this transaction is part of migration, and it is not
3586 * the last step of migration, so we set th_local = 1 to avoid
3587 * update last rcvd for this transaction */
3588 handle->th_local = 1;
3589 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3590 la, mgr_buf.lb_buf, ldata, handle);
3592 GOTO(stop_trans, rc);
3594 rc = mdd_trans_start(env, mdd, handle);
3596 GOTO(stop_trans, rc);
3598 /* don't set nlink from the original object */
3599 la->la_valid &= ~LA_NLINK;
3601 /* create the target object */
3602 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3605 GOTO(stop_trans, rc);
3607 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3608 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3610 GOTO(stop_trans, rc);
3613 /* Set MIGRATE EA on the source inode, so once the migration needs
3614 * to be re-done during failover, the re-do process can locate the
3615 * target object which is already being created. */
3616 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3618 GOTO(stop_trans, rc);
3620 /* Set immutable flag, so any modification is disabled until
3621 * the migration is done. Once the migration is interrupted,
3622 * if the resume process find the migrating object has both
3623 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3624 * flag and approve the migration */
3625 la_flag->la_valid = LA_FLAGS;
3626 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3627 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3630 rc = mdd_trans_stop(env, mdd, rc, handle);
3632 if (lmm_buf.lb_buf != NULL)
3633 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3637 static int mdd_migrate_entries(const struct lu_env *env,
3638 struct mdd_object *mdd_sobj,
3639 struct mdd_object *mdd_tobj)
3641 struct dt_object *next = mdd_object_child(mdd_sobj);
3642 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3643 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3644 struct thandle *handle;
3646 const struct dt_it_ops *iops;
3648 struct lu_dirent *ent;
3652 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3656 if (!dt_try_as_dir(env, next))
3657 GOTO(out_ent, rc = -ENOTDIR);
3659 * iterate directories
3661 iops = &next->do_index_ops->dio_it;
3662 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3664 GOTO(out_ent, rc = PTR_ERR(it));
3666 rc = iops->load(env, it, 0);
3668 rc = iops->next(env, it);
3672 * At this point and across for-loop:
3674 * rc == 0 -> ok, proceed.
3675 * rc > 0 -> end of directory.
3679 struct mdd_object *child;
3680 char *name = mdd_env_info(env)->mti_key;
3683 bool target_exist = false;
3685 len = iops->key_size(env, it);
3689 result = iops->rec(env, it, (struct dt_rec *)ent,
3690 LUDA_FID | LUDA_TYPE);
3691 if (result == -ESTALE)
3698 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3700 /* Insert new fid with target name into target dir */
3701 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3702 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3703 ent->lde_name[1] == '.'))
3706 child = mdd_object_find(env, mdd, &ent->lde_fid);
3708 GOTO(out, rc = PTR_ERR(child));
3710 mdd_write_lock(env, child, MOR_SRC_CHILD);
3711 is_dir = S_ISDIR(mdd_object_type(child));
3713 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3715 /* Check whether the name has been inserted to the target */
3716 if (dt_try_as_dir(env, dt_tobj)) {
3717 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3719 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3720 (struct dt_key *)name);
3721 if (unlikely(rc == 0))
3722 target_exist = true;
3725 handle = mdd_trans_create(env, mdd);
3727 GOTO(out_put, rc = PTR_ERR(handle));
3729 /* Note: this transaction is part of migration, and it is not
3730 * the last step of migration, so we set th_local = 1 to avoid
3731 * updating last rcvd for this transaction */
3732 handle->th_local = 1;
3733 if (likely(!target_exist)) {
3734 rc = mdo_declare_index_insert(env, mdd_tobj,
3736 mdd_object_type(child),
3742 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3748 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3753 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3757 /* Update .. for child */
3758 rc = mdo_declare_index_delete(env, child, dotdot,
3763 rc = mdo_declare_index_insert(env, child,
3764 mdd_object_fid(mdd_tobj),
3765 S_IFDIR, dotdot, handle);
3770 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3777 rc = mdd_trans_start(env, mdd, handle);
3779 CERROR("%s: transaction start failed: rc = %d\n",
3780 mdd2obd_dev(mdd)->obd_name, rc);
3784 if (likely(!target_exist)) {
3785 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3786 mdd_object_type(child),
3792 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3797 rc = __mdd_index_delete_only(env, child, dotdot,
3802 rc = __mdd_index_insert_only(env, child,
3803 mdd_object_fid(mdd_tobj), S_IFDIR,
3809 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3811 strlen(name), handle);
3814 mdd_write_unlock(env, child);
3815 mdd_object_put(env, child);
3816 rc = mdd_trans_stop(env, mdd, rc, handle);
3820 result = iops->next(env, it);
3821 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3822 GOTO(out, rc = -EINTR);
3824 if (result == -ESTALE)
3826 } while (result == 0);
3829 iops->fini(env, it);
3831 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3835 static int mdd_declare_update_linkea(const struct lu_env *env,
3836 struct mdd_object *mdd_pobj,
3837 struct mdd_object *mdd_sobj,
3838 struct mdd_object *mdd_tobj,
3839 const struct lu_name *child_name,
3840 struct linkea_data *ldata,
3841 struct thandle *handle)
3843 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3844 child_name, ldata, handle, 1);
3847 static int mdd_update_linkea(const struct lu_env *env,
3848 struct mdd_object *mdd_pobj,
3849 struct mdd_object *mdd_sobj,
3850 struct mdd_object *mdd_tobj,
3851 const struct lu_name *child_name,
3852 struct linkea_data *ldata,
3853 struct thandle *handle)
3855 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3856 child_name, ldata, handle, 0);
3859 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3860 struct mdd_object *mdd_pobj,
3861 struct mdd_object *mdd_sobj,
3862 struct mdd_object *mdd_tobj,
3863 const struct lu_name *lname,
3865 struct lu_attr *parent_la,
3866 struct linkea_data *ldata,
3867 struct thandle *handle)
3869 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3870 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3873 /* Revert IMMUTABLE flag */
3874 la_flag->la_valid = LA_FLAGS;
3875 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3876 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3880 /* delete entry from source dir */
3881 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3885 if (ldata->ld_buf != NULL) {
3886 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3887 mdd_tobj, lname, ldata, handle);
3892 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3893 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3898 handle->th_complex = 1;
3899 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3901 LU_XATTR_REPLACE, handle);
3906 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3907 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
3913 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
3914 mdd_object_type(mdd_tobj),
3915 lname->ln_name, handle);
3919 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
3923 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3924 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
3929 /* delete old object */
3930 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3934 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3935 /* delete old object */
3936 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3939 /* set nlink to 0 */
3940 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3945 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
3949 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
3953 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
3958 static int mdd_migrate_update_name(const struct lu_env *env,
3959 struct mdd_object *mdd_pobj,
3960 struct mdd_object *mdd_sobj,
3961 struct mdd_object *mdd_tobj,
3962 const struct lu_name *lname,
3965 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
3966 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
3967 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3968 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3969 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3970 struct thandle *handle;
3971 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
3972 const char *name = lname->ln_name;
3976 /* update time for parent */
3977 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3978 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
3979 p_la->la_valid = LA_CTIME;
3981 rc = mdd_la_get(env, mdd_sobj, so_attr);
3985 ldata->ld_buf = NULL;
3986 rc = mdd_links_read(env, mdd_sobj, ldata);
3987 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
3990 handle = mdd_trans_create(env, mdd);
3992 RETURN(PTR_ERR(handle));
3994 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
3995 lname, so_attr, p_la, ldata,
3998 /* If the migration can not be fit in one transaction, just
3999 * leave it in the original MDT */
4001 GOTO(stop_trans, rc = 0);
4003 GOTO(stop_trans, rc);
4006 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4007 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4008 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4009 PFID(mdd_object_fid(mdd_tobj)));
4011 rc = mdd_trans_start(env, mdd, handle);
4013 GOTO(stop_trans, rc);
4015 /* Revert IMMUTABLE flag */
4016 la_flag->la_valid = LA_FLAGS;
4017 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4018 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4020 GOTO(stop_trans, rc);
4022 /* Remove source name from source directory */
4023 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4025 GOTO(stop_trans, rc);
4027 if (ldata->ld_buf != NULL) {
4028 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4029 lname, ldata, handle);
4031 GOTO(stop_trans, rc);
4033 /* linkea update might decrease the source object
4034 * nlink, let's get the attr again after ref_del */
4035 rc = mdd_la_get(env, mdd_sobj, so_attr);
4037 GOTO(stop_trans, rc);
4040 if (S_ISREG(so_attr->la_mode)) {
4041 if (so_attr->la_nlink == 1) {
4042 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4044 if (rc != 0 && rc != -ENODATA)
4045 GOTO(stop_trans, rc);
4047 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4049 LU_XATTR_REPLACE, handle);
4051 GOTO(stop_trans, rc);
4055 /* Insert new fid with target name into target dir */
4056 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4057 mdd_object_type(mdd_tobj), name, handle);
4059 GOTO(stop_trans, rc);
4061 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4063 mdd_sobj->mod_flags |= DEAD_OBJ;
4064 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4066 GOTO(out_unlock, rc);
4068 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4070 GOTO(out_unlock, rc);
4072 mdo_ref_del(env, mdd_sobj, handle);
4074 mdo_ref_del(env, mdd_sobj, handle);
4076 /* Get the attr again after ref_del */
4077 rc = mdd_la_get(env, mdd_sobj, so_attr);
4079 GOTO(out_unlock, rc);
4081 ma->ma_attr = *so_attr;
4082 ma->ma_valid |= MA_INODE;
4084 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4086 GOTO(out_unlock, rc);
4088 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4089 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4090 mdo2fid(mdd_pobj), lname, lname, handle);
4092 CWARN("%s: changelog for migrate %s "DFID
4093 "under "DFID" failed: rc = %d\n",
4094 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4095 PFID(mdd_object_fid(mdd_sobj)),
4096 PFID(mdd_object_fid(mdd_pobj)), rc);
4097 /* Sigh, there are no easy way to migrate back the object, so
4098 * let's reset the result to 0 for now XXX */
4102 mdd_write_unlock(env, mdd_sobj);
4105 rc = mdd_trans_stop(env, mdd, rc, handle);
4110 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4111 const struct lu_fid *fid, __u32 *mdt_index)
4113 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4114 struct seq_server_site *ss;
4117 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4119 range->lsr_flags = LU_SEQ_RANGE_MDT;
4120 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4124 *mdt_index = range->lsr_index;
4129 * Check whether we should migrate the file/dir
4131 * < 0 permission check failed or other error.
4132 * = 0 the file can be migrated.
4133 * > 0 the file does not need to be migrated, mostly
4134 * for multiple link file
4136 static int mdd_migrate_sanity_check(const struct lu_env *env,
4137 struct mdd_object *pobj,
4138 const struct lu_attr *pattr,
4139 struct mdd_object *sobj,
4140 struct lu_attr *sattr)
4142 struct mdd_thread_info *info = mdd_env_info(env);
4143 struct linkea_data *ldata = &info->mti_link_data;
4144 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4146 struct lu_buf *mgr_buf;
4152 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4153 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4154 if (mgr_buf->lb_buf == NULL)
4157 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4159 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4161 /* If the object has migrateEA, it means IMMUTE flag
4162 * is being set by previous migration process, so it
4163 * needs to override the IMMUTE flag, otherwise the
4164 * following sanity check will fail */
4165 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4166 LMV_HASH_FLAG_MIGRATION) {
4167 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4169 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4170 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4171 mdd2obd_dev(mdd)->obd_name,
4172 PFID(mdd_object_fid(sobj)));
4176 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4177 sobj, sattr, NULL, NULL);
4181 /* Then it will check if the file should be migrated. If the file
4182 * has mulitple links, we only need migrate the file if all of its
4183 * entries has been migrated to the remote MDT */
4184 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4187 rc = mdd_links_read(env, sobj, ldata);
4189 /* For multiple links files, if there are no linkEA data at all,
4190 * means the file might be created before linkEA is enabled, and
4191 * all of its links should not be migrated yet, otherwise it
4192 * should have some linkEA there */
4193 if (rc == -ENOENT || rc == -ENODATA)
4198 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4199 /* If there are still links locally, then the file will not be
4201 LASSERT(ldata->ld_leh != NULL);
4203 /* If the linkEA is overflow, then means there are some unknown name
4204 * entries under unknown parents, that will prevent the migration. */
4205 if (unlikely(ldata->ld_leh->leh_overflow_time))
4208 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4209 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4210 struct lu_name lname;
4212 __u32 parent_mdt_index;
4214 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4216 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4219 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4223 /* Migrate the object only if none of its parents are on the
4225 if (parent_mdt_index != mdt_index)
4228 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4229 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4230 lname.ln_name, PFID(&fid));
4238 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4239 struct md_object *sobj, const struct lu_name *lname,
4240 struct md_object *tobj, struct md_attr *ma)
4242 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4243 struct mdd_device *mdd = mdo2mdd(pobj);
4244 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4245 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4246 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4247 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4248 bool created = false;
4252 /* If the file will being migrated, it will check whether
4253 * the file is being opened by someone else right now */
4254 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4255 if (mdd_sobj->mod_count > 0) {
4257 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4258 mdd2obd_dev(mdd)->obd_name,
4259 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4260 mdd_sobj->mod_count, -EBUSY);
4261 mdd_read_unlock(env, mdd_sobj);
4262 GOTO(put, rc = -EBUSY);
4264 mdd_read_unlock(env, mdd_sobj);
4266 rc = mdd_la_get(env, mdd_sobj, so_attr);
4270 rc = mdd_la_get(env, mdd_pobj, pattr);
4274 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4281 /* Sigh, it is impossible to finish all of migration in a single
4282 * transaction, for example migrating big directory entries to the
4283 * new MDT, it needs insert all of name entries of children in the
4286 * So migration will be done in multiple steps and transactions.
4288 * 1. create an orphan object on the remote MDT in one transaction.
4289 * 2. migrate extend attributes to the new target file/directory.
4290 * 3. For directory, migrate the entries to the new MDT and update
4291 * linkEA of each children. Because we can not migrate all entries
4292 * in a single transaction, so the migrating directory will become
4293 * a striped directory during migration, so once the process is
4294 * interrupted, the directory is still accessible. (During lookup,
4295 * client will locate the name by searching both original and target
4297 * 4. Finally, update the name/FID to point to the new file/directory
4298 * in a separate transaction.
4301 /* step 1: Check whether the orphan object has been created, and create
4302 * orphan object on the remote MDT if needed */
4303 if (!mdd_object_exists(mdd_tobj)) {
4304 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4311 LASSERT(mdd_object_exists(mdd_tobj));
4312 /* step 2: migrate xattr */
4313 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4317 /* step 3: migrate name entries to the orphan object */
4318 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4319 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4322 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4323 OBD_FAIL_MDS_REINT_NET_REP)))
4326 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4329 LASSERT(mdd_object_exists(mdd_tobj));
4330 /* step 4: update name entry to the new object */
4331 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4336 /* newly created target was not locked, don't cache its attributes */
4338 mdd_invalidate(env, tobj);
4343 const struct md_dir_operations mdd_dir_ops = {
4344 .mdo_is_subdir = mdd_is_subdir,
4345 .mdo_lookup = mdd_lookup,
4346 .mdo_create = mdd_create,
4347 .mdo_rename = mdd_rename,
4348 .mdo_link = mdd_link,
4349 .mdo_unlink = mdd_unlink,
4350 .mdo_create_data = mdd_create_data,
4351 .mdo_migrate = mdd_migrate,