4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_dir.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@intel.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <obd_class.h>
42 #include <obd_support.h>
43 #include <lustre_mds.h>
44 #include <lustre_fid.h>
46 #include "mdd_internal.h"
48 static const char dot[] = ".";
49 static const char dotdot[] = "..";
51 static struct lu_name lname_dotdot = {
52 .ln_name = (char *) dotdot,
53 .ln_namelen = sizeof(dotdot) - 1,
57 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
59 if (!lu_name_is_valid(ln))
61 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
67 /* Get FID from name and parent */
69 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
70 const struct lu_attr *pattr, const struct lu_name *lname,
71 struct lu_fid* fid, int mask)
73 const char *name = lname->ln_name;
74 const struct dt_key *key = (const struct dt_key *)name;
75 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
76 struct mdd_device *m = mdo2mdd(pobj);
77 struct dt_object *dir = mdd_object_child(mdd_obj);
81 if (unlikely(mdd_is_dead_obj(mdd_obj)))
84 if (!mdd_object_exists(mdd_obj))
87 if (mdd_object_remote(mdd_obj)) {
88 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
89 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
92 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
97 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
98 dt_try_as_dir(env, dir)))
99 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
106 int mdd_lookup(const struct lu_env *env,
107 struct md_object *pobj, const struct lu_name *lname,
108 struct lu_fid *fid, struct md_op_spec *spec)
110 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
114 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
118 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
119 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
123 /** Read the link EA into a temp buffer.
124 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
125 * A pointer to the buffer is stored in \a ldata::ld_buf.
129 static int __mdd_links_read(const struct lu_env *env,
130 struct mdd_object *mdd_obj,
131 struct linkea_data *ldata)
135 if (!mdd_object_exists(mdd_obj))
138 /* First try a small buf */
139 LASSERT(env != NULL);
140 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
142 if (ldata->ld_buf->lb_buf == NULL)
145 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
147 /* Buf was too small, figure out what we need. */
148 lu_buf_free(ldata->ld_buf);
149 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
153 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
154 if (ldata->ld_buf->lb_buf == NULL)
156 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
160 lu_buf_free(ldata->ld_buf);
161 ldata->ld_buf = NULL;
165 return linkea_init(ldata);
168 static int mdd_links_read(const struct lu_env *env,
169 struct mdd_object *mdd_obj,
170 struct linkea_data *ldata)
174 rc = __mdd_links_read(env, mdd_obj, ldata);
176 rc = linkea_init(ldata);
181 static int mdd_links_read_with_rec(const struct lu_env *env,
182 struct mdd_object *mdd_obj,
183 struct linkea_data *ldata)
187 rc = __mdd_links_read(env, mdd_obj, ldata);
189 rc = linkea_init_with_rec(ldata);
195 * Get parent FID of the directory
197 * Read parent FID from linkEA, if that fails, then do lookup
198 * dotdot to get the parent FID.
200 * \param[in] env execution environment
201 * \param[in] obj object from which to find the parent FID
202 * \param[in] attr attribute of the object
203 * \param[out] fid fid to get the parent FID
205 * \retval 0 if getting the parent FID succeeds.
206 * \retval negative errno if getting the parent FID fails.
208 static inline int mdd_parent_fid(const struct lu_env *env,
209 struct mdd_object *obj,
210 const struct lu_attr *attr,
213 struct mdd_thread_info *info = mdd_env_info(env);
214 struct linkea_data ldata = { NULL };
215 struct lu_buf *buf = &info->mti_link_buf;
216 struct lu_name lname;
221 LASSERT(S_ISDIR(mdd_object_type(obj)));
223 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
224 if (buf->lb_buf == NULL)
225 GOTO(lookup, rc = 0);
228 rc = mdd_links_read_with_rec(env, obj, &ldata);
232 LASSERT(ldata.ld_leh != NULL);
233 /* Directory should only have 1 parent */
234 if (ldata.ld_leh->leh_reccount > 1)
237 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
239 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
240 if (likely(fid_is_sane(fid)))
243 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
248 * For root fid use special function, which does not compare version component
249 * of fid. Version component is different for root fids on all MDTs.
251 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
253 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
254 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
258 * return 1: if lf is the fid of the ancestor of p1;
261 * return -EREMOTE: if remote object is found, in this
262 * case fid of remote object is saved to @pf;
264 * otherwise: values < 0, errors.
266 static int mdd_is_parent(const struct lu_env *env,
267 struct mdd_device *mdd,
268 struct mdd_object *p1,
269 const struct lu_attr *attr,
270 const struct lu_fid *lf,
273 struct mdd_object *parent = NULL;
278 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
279 pfid = &mdd_env_info(env)->mti_fid;
281 /* Check for root first. */
282 if (mdd_is_root(mdd, mdo2fid(p1)))
286 /* this is done recursively */
287 rc = mdd_parent_fid(env, p1, attr, pfid);
290 if (mdd_is_root(mdd, pfid))
292 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
294 if (lu_fid_eq(pfid, lf))
297 mdd_object_put(env, parent);
299 parent = mdd_object_find(env, mdd, pfid);
301 GOTO(out, rc = PTR_ERR(parent));
303 if (!mdd_object_exists(parent))
304 GOTO(out, rc = -EINVAL);
310 if (parent && !IS_ERR(parent))
311 mdd_object_put(env, parent);
316 * No permission check is needed.
318 * returns 1: if fid is ancestor of @mo;
319 * returns 0: if fid is not an ancestor of @mo;
321 * returns EREMOTE if remote object is found, fid of remote object is saved to
324 * returns < 0: if error
326 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
327 const struct lu_fid *fid, struct lu_fid *sfid)
329 struct mdd_device *mdd = mdo2mdd(mo);
330 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
334 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
337 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
341 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
345 } else if (rc == 1) {
346 /* found @fid is parent */
354 * Check that @dir contains no entries except (possibly) dot and dotdot.
359 * -ENOTDIR not a directory object
360 * -ENOTEMPTY not empty
364 static int mdd_dir_is_empty(const struct lu_env *env,
365 struct mdd_object *dir)
368 struct dt_object *obj;
369 const struct dt_it_ops *iops;
373 obj = mdd_object_child(dir);
374 if (!dt_try_as_dir(env, obj))
377 iops = &obj->do_index_ops->dio_it;
378 it = iops->init(env, obj, LUDA_64BITHASH);
380 result = iops->get(env, it, (const struct dt_key *)"");
383 for (result = 0, i = 0; result == 0 && i < 3; ++i)
384 result = iops->next(env, it);
387 else if (result == 1)
389 } else if (result == 0)
391 * Huh? Index contains no zero key?
398 result = PTR_ERR(it);
403 * Determine if the target object can be hard linked, and right now it only
404 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
405 * directory nlink count might exceed the maximum link count(see
406 * osd_object_ref_add), so it only check nlink for non-directories.
408 * \param[in] env thread environment
409 * \param[in] obj object being linked to
410 * \param[in] la attributes of \a obj
412 * \retval 0 if \a obj can be hard linked
413 * \retval negative error if \a obj is a directory or has too
416 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
417 const struct lu_attr *la)
419 struct mdd_device *m = mdd_obj2mdd_dev(obj);
424 /* Subdir count limitation can be broken through
425 * (see osd_object_ref_add), so only check non-directory here. */
426 if (!S_ISDIR(la->la_mode) &&
427 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
434 * Check whether it may create the cobj under the pobj.
436 * \param[in] env execution environment
437 * \param[in] pobj the parent directory
438 * \param[in] pattr the attribute of the parent directory
439 * \param[in] cobj the child to be created
440 * \param[in] check_perm if check WRITE|EXEC permission for parent
442 * \retval = 0 create the child under this dir is allowed
443 * \retval negative errno create the child under this dir is
446 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
447 const struct lu_attr *pattr, struct mdd_object *cobj,
453 if (cobj && mdd_object_exists(cobj))
456 if (mdd_is_dead_obj(pobj))
460 rc = mdd_permission_internal_locked(env, pobj, pattr,
461 MAY_WRITE | MAY_EXEC,
467 * Check whether can unlink from the pobj in the case of "cobj == NULL".
469 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
470 const struct lu_attr *pattr, const struct lu_attr *attr)
475 if (mdd_is_dead_obj(pobj))
478 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
481 rc = mdd_permission_internal_locked(env, pobj, pattr,
482 MAY_WRITE | MAY_EXEC,
487 if (pattr->la_flags & LUSTRE_APPEND_FL)
494 * pobj == NULL is remote ops case, under such case, pobj's
495 * VTX feature has been checked already, no need check again.
497 static inline int mdd_is_sticky(const struct lu_env *env,
498 struct mdd_object *pobj,
499 const struct lu_attr *pattr,
500 struct mdd_object *cobj,
501 const struct lu_attr *cattr)
503 struct lu_ucred *uc = lu_ucred_assert(env);
506 LASSERT(pattr != NULL);
507 if (!(pattr->la_mode & S_ISVTX) ||
508 (pattr->la_uid == uc->uc_fsuid))
512 LASSERT(cattr != NULL);
513 if (cattr->la_uid == uc->uc_fsuid)
516 return !md_capable(uc, CFS_CAP_FOWNER);
519 static int mdd_may_delete_entry(const struct lu_env *env,
520 struct mdd_object *pobj,
521 const struct lu_attr *pattr,
526 LASSERT(pobj != NULL);
527 if (!mdd_object_exists(pobj))
530 if (mdd_is_dead_obj(pobj))
535 rc = mdd_permission_internal_locked(env, pobj, pattr,
536 MAY_WRITE | MAY_EXEC,
542 if (pattr->la_flags & LUSTRE_APPEND_FL)
549 * Check whether it may delete the cobj from the pobj.
552 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
553 const struct lu_attr *tpattr, struct mdd_object *tobj,
554 const struct lu_attr *tattr, const struct lu_attr *cattr,
555 int check_perm, int check_empty)
561 LASSERT(tpattr != NULL);
562 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
570 if (!mdd_object_exists(tobj))
573 if (mdd_is_dead_obj(tobj))
576 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
579 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
582 /* additional check the rename case */
584 if (S_ISDIR(cattr->la_mode)) {
585 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
587 if (!S_ISDIR(tattr->la_mode))
590 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
592 } else if (S_ISDIR(tattr->la_mode))
596 if (S_ISDIR(tattr->la_mode) && check_empty)
597 rc = mdd_dir_is_empty(env, tobj);
603 * Check whether it can create the link file(linked to @src_obj) under
604 * the target directory(@tgt_obj), and src_obj has been locked by
607 * \param[in] env execution environment
608 * \param[in] tgt_obj the target directory
609 * \param[in] tattr attributes of target directory
610 * \param[in] lname the link name
611 * \param[in] src_obj source object for link
612 * \param[in] cattr attributes for source object
614 * \retval = 0 it is allowed to create the link file under tgt_obj
615 * \retval negative error not allowed to create the link file
617 static int mdd_link_sanity_check(const struct lu_env *env,
618 struct mdd_object *tgt_obj,
619 const struct lu_attr *tattr,
620 const struct lu_name *lname,
621 struct mdd_object *src_obj,
622 const struct lu_attr *cattr)
624 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
628 if (!mdd_object_exists(src_obj))
631 if (mdd_is_dead_obj(src_obj))
634 /* Local ops, no lookup before link, check filename length here. */
635 rc = mdd_name_check(m, lname);
639 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
642 if (S_ISDIR(mdd_object_type(src_obj)))
645 LASSERT(src_obj != tgt_obj);
646 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
650 rc = __mdd_may_link(env, src_obj, cattr);
655 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
656 const char *name, struct thandle *handle)
658 struct dt_object *next = mdd_object_child(pobj);
662 if (dt_try_as_dir(env, next))
663 rc = dt_delete(env, next, (struct dt_key *)name, handle);
670 static int __mdd_index_insert_only(const struct lu_env *env,
671 struct mdd_object *pobj,
672 const struct lu_fid *lf, __u32 type,
674 struct thandle *handle)
676 struct dt_object *next = mdd_object_child(pobj);
680 if (dt_try_as_dir(env, next)) {
681 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
682 struct lu_ucred *uc = lu_ucred_check(env);
686 rec->rec_type = type;
687 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
688 rc = dt_insert(env, next, (const struct dt_rec *)rec,
689 (const struct dt_key *)name, handle,
697 /* insert named index, add reference if isdir */
698 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
699 const struct lu_fid *lf, __u32 type,
700 const char *name, struct thandle *handle)
705 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
706 if (rc == 0 && S_ISDIR(type)) {
707 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
708 mdo_ref_add(env, pobj, handle);
709 mdd_write_unlock(env, pobj);
715 /* delete named index, drop reference if isdir */
716 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
717 const char *name, int is_dir,
718 struct thandle *handle)
723 rc = __mdd_index_delete_only(env, pobj, name, handle);
724 if (rc == 0 && is_dir) {
725 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
726 mdo_ref_del(env, pobj, handle);
727 mdd_write_unlock(env, pobj);
733 static int mdd_llog_record_calc_size(const struct lu_env *env,
734 const struct lu_name *tname,
735 const struct lu_name *sname)
737 const struct lu_ucred *uc = lu_ucred(env);
738 enum changelog_rec_flags crf = CLF_EXTRA_FLAGS;
739 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
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 enum changelog_rec_type type,
756 const struct lu_name *tname,
757 const struct lu_name *sname,
758 struct thandle *handle)
760 struct obd_device *obd = mdd2obd_dev(mdd);
761 struct llog_ctxt *ctxt;
762 struct llog_changelog_rec *rec;
764 struct thandle *llog_th;
768 if (!mdd_changelog_enabled(env, mdd, type))
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);
836 /* time to recover some space ?? */
837 if (likely(!mdd->mdd_changelog_gc ||
838 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
839 mdd->mdd_changelog_min_gc_interval >=
840 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
841 /* save a spin_lock trip */
843 spin_lock(&mdd->mdd_cl.mc_lock);
844 if (likely(mdd->mdd_changelog_gc &&
845 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
846 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
847 mdd->mdd_changelog_min_gc_interval)) {
848 if (unlikely(llog_cat_free_space(ctxt->loc_handle) <=
849 mdd->mdd_changelog_min_free_cat_entries ||
850 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD))) {
851 CWARN("%s:%s low on changelog_catalog free entries, "
852 "starting ChangeLog garbage collection thread\n",
854 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
857 /* indicate further kthread run will occur outside
858 * right after current journal transaction filling has
861 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
863 /* next check in mdd_changelog_min_gc_interval anyway
865 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
867 spin_unlock(&mdd->mdd_cl.mc_lock);
875 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
876 const struct lu_fid *sfid,
877 const struct lu_fid *spfid,
878 const struct lu_name *sname)
880 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
883 LASSERT(sfid != NULL);
884 LASSERT(spfid != NULL);
885 LASSERT(sname != NULL);
887 extsize = sname->ln_namelen + 1;
889 rnm->cr_sfid = *sfid;
890 rnm->cr_spfid = *spfid;
892 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
893 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
894 rec->cr_namelen += extsize;
897 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
899 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
901 if (jobid == NULL || jobid[0] == '\0')
904 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
907 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
909 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
911 ef->cr_extra_flags = eflags;
914 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
915 __u64 uid, __u64 gid)
917 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
919 uidgid->cr_uid = uid;
920 uidgid->cr_gid = gid;
923 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
926 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
931 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, int flags)
933 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
935 omd->cr_openflags = (__u32)flags;
938 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
939 const char *xattr_name)
941 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
943 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
946 /** Store a namespace change changelog record
947 * If this fails, we must fail the whole transaction; we don't
948 * want the change to commit without the log entry.
949 * \param target - mdd_object of change
950 * \param tpfid - target parent dir/object fid
951 * \param sfid - source object fid
952 * \param spfid - source parent fid
953 * \param tname - target name string
954 * \param sname - source name string
955 * \param handle - transaction handle
957 int mdd_changelog_ns_store(const struct lu_env *env,
958 struct mdd_device *mdd,
959 enum changelog_rec_type type,
960 enum changelog_rec_flags crf,
961 struct mdd_object *target,
962 const struct lu_fid *tpfid,
963 const struct lu_fid *sfid,
964 const struct lu_fid *spfid,
965 const struct lu_name *tname,
966 const struct lu_name *sname,
967 struct thandle *handle)
969 const struct lu_ucred *uc = lu_ucred(env);
970 struct llog_changelog_rec *rec;
973 __u64 xflags = CLFE_INVALID;
977 if (!mdd_changelog_enabled(env, mdd, type))
980 LASSERT(tpfid != NULL);
981 LASSERT(tname != NULL);
982 LASSERT(handle != NULL);
984 reclen = mdd_llog_record_calc_size(env, tname, sname);
985 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
986 if (buf->lb_buf == NULL)
991 crf |= CLF_EXTRA_FLAGS;
994 if (uc->uc_jobid[0] != '\0')
996 xflags |= CLFE_UIDGID;
1005 rec->cr.cr_flags = crf;
1007 if (crf & CLF_EXTRA_FLAGS) {
1008 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1009 if (xflags & CLFE_UIDGID)
1010 mdd_changelog_rec_extra_uidgid(&rec->cr,
1011 uc->uc_uid, uc->uc_gid);
1012 if (xflags & CLFE_NID)
1013 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1016 rec->cr.cr_type = (__u32)type;
1017 rec->cr.cr_pfid = *tpfid;
1018 rec->cr.cr_namelen = tname->ln_namelen;
1019 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1021 if (crf & CLF_RENAME)
1022 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1024 if (crf & CLF_JOBID)
1025 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1027 if (likely(target != NULL)) {
1028 rec->cr.cr_tfid = *mdo2fid(target);
1029 target->mod_cltime = ktime_get();
1031 fid_zero(&rec->cr.cr_tfid);
1034 rc = mdd_changelog_store(env, mdd, rec, handle);
1036 CERROR("%s: cannot store changelog record: type = %d, "
1037 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1038 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1039 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1046 static int __mdd_links_add(const struct lu_env *env,
1047 struct mdd_object *mdd_obj,
1048 struct linkea_data *ldata,
1049 const struct lu_name *lname,
1050 const struct lu_fid *pfid,
1051 int first, int check)
1055 if (ldata->ld_leh == NULL) {
1056 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1060 rc = linkea_data_new(ldata,
1061 &mdd_env_info(env)->mti_link_buf);
1068 rc = linkea_links_find(ldata, lname, pfid);
1069 if (rc && rc != -ENOENT)
1075 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1076 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1080 linkea_add_buf(ldata, lname, tfid);
1083 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1084 linkea_add_buf(ldata, lname, pfid);
1086 return linkea_add_buf(ldata, lname, pfid);
1089 static int __mdd_links_del(const struct lu_env *env,
1090 struct mdd_object *mdd_obj,
1091 struct linkea_data *ldata,
1092 const struct lu_name *lname,
1093 const struct lu_fid *pfid)
1097 if (ldata->ld_leh == NULL) {
1098 rc = mdd_links_read(env, mdd_obj, ldata);
1103 rc = linkea_links_find(ldata, lname, pfid);
1107 linkea_del_buf(ldata, lname);
1111 static int mdd_linkea_prepare(const struct lu_env *env,
1112 struct mdd_object *mdd_obj,
1113 const struct lu_fid *oldpfid,
1114 const struct lu_name *oldlname,
1115 const struct lu_fid *newpfid,
1116 const struct lu_name *newlname,
1117 int first, int check,
1118 struct linkea_data *ldata)
1123 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1126 LASSERT(oldpfid != NULL || newpfid != NULL);
1128 if (mdd_obj->mod_flags & DEAD_OBJ)
1129 /* Unnecessary to update linkEA for dead object. */
1132 if (oldpfid != NULL) {
1133 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1135 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1138 /* No changes done. */
1143 /* If renaming, add the new record */
1144 if (newpfid != NULL)
1145 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1151 int mdd_links_rename(const struct lu_env *env,
1152 struct mdd_object *mdd_obj,
1153 const struct lu_fid *oldpfid,
1154 const struct lu_name *oldlname,
1155 const struct lu_fid *newpfid,
1156 const struct lu_name *newlname,
1157 struct thandle *handle,
1158 struct linkea_data *ldata,
1159 int first, int check)
1164 if (ldata == NULL) {
1165 ldata = &mdd_env_info(env)->mti_link_data;
1166 memset(ldata, 0, sizeof(*ldata));
1167 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1168 newpfid, newlname, first, check, ldata);
1173 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1174 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1180 if (newlname == NULL)
1181 CERROR("link_ea add failed %d "DFID"\n",
1182 rc, PFID(mdd_object_fid(mdd_obj)));
1183 else if (oldpfid == NULL)
1184 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1185 newlname->ln_namelen, newlname->ln_name, rc,
1186 PFID(mdd_object_fid(mdd_obj)));
1187 else if (newpfid == NULL)
1188 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1189 oldlname->ln_namelen, oldlname->ln_name, rc,
1190 PFID(mdd_object_fid(mdd_obj)));
1192 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1193 "\n", oldlname->ln_namelen, oldlname->ln_name,
1194 newlname->ln_namelen, newlname->ln_name, rc,
1195 PFID(mdd_object_fid(mdd_obj)));
1198 if (is_vmalloc_addr(ldata->ld_buf))
1199 /* if we vmalloced a large buffer drop it */
1200 lu_buf_free(ldata->ld_buf);
1205 static inline int mdd_links_add(const struct lu_env *env,
1206 struct mdd_object *mdd_obj,
1207 const struct lu_fid *pfid,
1208 const struct lu_name *lname,
1209 struct thandle *handle,
1210 struct linkea_data *ldata, int first)
1212 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1213 pfid, lname, handle, ldata, first, 0);
1216 static inline int mdd_links_del(const struct lu_env *env,
1217 struct mdd_object *mdd_obj,
1218 const struct lu_fid *pfid,
1219 const struct lu_name *lname,
1220 struct thandle *handle)
1222 return mdd_links_rename(env, mdd_obj, pfid, lname,
1223 NULL, NULL, handle, NULL, 0, 0);
1226 /** Read the link EA into a temp buffer.
1227 * Uses the name_buf since it is generally large.
1228 * \retval IS_ERR err
1229 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1231 struct lu_buf *mdd_links_get(const struct lu_env *env,
1232 struct mdd_object *mdd_obj)
1234 struct linkea_data ldata = { NULL };
1237 rc = mdd_links_read(env, mdd_obj, &ldata);
1238 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1241 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1242 struct linkea_data *ldata, struct thandle *handle)
1244 const struct lu_buf *buf;
1247 if (ldata == NULL || ldata->ld_buf == NULL ||
1248 ldata->ld_leh == NULL)
1251 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1255 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1256 ldata->ld_leh->leh_len);
1257 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1258 if (unlikely(rc == -ENOSPC)) {
1259 rc = linkea_overflow_shrink(ldata);
1267 static int mdd_declare_links_add(const struct lu_env *env,
1268 struct mdd_object *mdd_obj,
1269 struct thandle *handle,
1270 struct linkea_data *ldata)
1276 if (ldata != NULL && ldata->ld_leh != NULL) {
1277 ea_len = ldata->ld_leh->leh_len;
1278 linkea = ldata->ld_buf->lb_buf;
1280 ea_len = MAX_LINKEA_SIZE;
1284 rc = mdo_declare_xattr_set(env, mdd_obj,
1285 mdd_buf_get_const(env, linkea, ea_len),
1286 XATTR_NAME_LINK, 0, handle);
1291 static inline int mdd_declare_links_del(const struct lu_env *env,
1292 struct mdd_object *c,
1293 struct thandle *handle)
1297 /* For directory, the linkEA will be removed together
1298 * with the object. */
1299 if (!S_ISDIR(mdd_object_type(c)))
1300 rc = mdd_declare_links_add(env, c, handle, NULL);
1305 static int mdd_declare_link(const struct lu_env *env,
1306 struct mdd_device *mdd,
1307 struct mdd_object *p,
1308 struct mdd_object *c,
1309 const struct lu_name *name,
1310 struct thandle *handle,
1312 struct linkea_data *data)
1314 struct lu_fid tfid = *mdo2fid(c);
1317 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1318 tfid.f_oid = cfs_fail_val;
1320 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1321 name->ln_name, handle);
1325 rc = mdo_declare_ref_add(env, c, handle);
1329 la->la_valid = LA_CTIME | LA_MTIME;
1330 rc = mdo_declare_attr_set(env, p, la, handle);
1334 la->la_valid = LA_CTIME;
1335 rc = mdo_declare_attr_set(env, c, la, handle);
1339 rc = mdd_declare_links_add(env, c, handle, data);
1343 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1349 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1350 struct md_object *src_obj, const struct lu_name *lname,
1353 const char *name = lname->ln_name;
1354 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1355 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1356 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1357 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1358 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1359 struct mdd_device *mdd = mdo2mdd(src_obj);
1360 struct thandle *handle;
1361 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1362 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1366 rc = mdd_la_get(env, mdd_sobj, cattr);
1370 rc = mdd_la_get(env, mdd_tobj, tattr);
1375 * If we are using project inheritance, we only allow hard link
1376 * creation in our tree when the project IDs are the same;
1377 * otherwise the tree quota mechanism could be circumvented.
1379 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1380 (tattr->la_projid != cattr->la_projid))
1383 handle = mdd_trans_create(env, mdd);
1385 GOTO(out_pending, rc = PTR_ERR(handle));
1387 memset(ldata, 0, sizeof(*ldata));
1389 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1390 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1392 /* Note: even this function will change ldata, but it comes from
1393 * thread_info, which is completely temporary and only seen in
1394 * this function, so we do not need reset ldata once it fails.*/
1395 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1396 lname, 0, 0, ldata);
1400 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1405 rc = mdd_trans_start(env, mdd, handle);
1409 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1410 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1413 GOTO(out_unlock, rc);
1415 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1416 rc = mdo_ref_add(env, mdd_sobj, handle);
1418 GOTO(out_unlock, rc);
1421 *tfid = *mdo2fid(mdd_sobj);
1422 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1423 tfid->f_oid = cfs_fail_val;
1425 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1426 mdd_object_type(mdd_sobj), name, handle);
1428 mdo_ref_del(env, mdd_sobj, handle);
1429 GOTO(out_unlock, rc);
1432 la->la_valid = LA_CTIME | LA_MTIME;
1433 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1435 GOTO(out_unlock, rc);
1437 la->la_valid = LA_CTIME;
1438 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1440 /* Note: The failure of links_add should not cause the
1441 * link failure, so do not check return value. */
1442 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1443 lname, handle, ldata, 0);
1447 mdd_write_unlock(env, mdd_sobj);
1449 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1450 mdo2fid(mdd_tobj), NULL, NULL,
1451 lname, NULL, handle);
1453 rc = mdd_trans_stop(env, mdd, rc, handle);
1454 if (is_vmalloc_addr(ldata->ld_buf))
1455 /* if we vmalloced a large buffer drop it */
1456 lu_buf_free(ldata->ld_buf);
1461 static int mdd_mark_orphan_object(const struct lu_env *env,
1462 struct mdd_object *obj, struct thandle *handle,
1465 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1468 if (!S_ISDIR(mdd_object_type(obj)))
1471 attr->la_valid = LA_FLAGS;
1472 attr->la_flags = LUSTRE_ORPHAN_FL;
1475 rc = mdo_declare_attr_set(env, obj, attr, handle);
1477 rc = mdo_attr_set(env, obj, attr, handle);
1482 static int mdd_declare_finish_unlink(const struct lu_env *env,
1483 struct mdd_object *obj,
1484 struct thandle *handle)
1488 /* Sigh, we do not know if the unlink object will become orphan in
1489 * declare phase, but fortunately the flags here does not matter
1490 * in current declare implementation */
1491 rc = mdd_mark_orphan_object(env, obj, handle, true);
1495 rc = mdo_declare_destroy(env, obj, handle);
1499 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1503 return mdd_declare_links_del(env, obj, handle);
1506 /* caller should take a lock before calling */
1507 int mdd_finish_unlink(const struct lu_env *env,
1508 struct mdd_object *obj, struct md_attr *ma,
1509 const struct mdd_object *pobj,
1510 const struct lu_name *lname,
1514 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1517 LASSERT(mdd_write_locked(env, obj) != 0);
1519 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1520 /* add new orphan and the object
1521 * will be deleted during mdd_close() */
1522 obj->mod_flags |= DEAD_OBJ;
1523 if (obj->mod_count) {
1524 rc = mdd_orphan_insert(env, obj, th);
1526 CDEBUG(D_HA, "Object "DFID" is inserted into "
1527 "orphan list, open count = %d\n",
1528 PFID(mdd_object_fid(obj)),
1531 CERROR("Object "DFID" fail to be an orphan, "
1532 "open count = %d, maybe cause failed "
1534 PFID(mdd_object_fid(obj)),
1537 /* mark object as an orphan here, not
1538 * before mdd_orphan_insert() as racing
1539 * mdd_la_get() may propagate ORPHAN_OBJ
1540 * causing the asserition */
1541 rc = mdd_mark_orphan_object(env, obj, th, false);
1543 rc = mdo_destroy(env, obj, th);
1545 } else if (!is_dir) {
1546 /* old files may not have link ea; ignore errors */
1547 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1555 * has mdd_write_lock on cobj already, but not on pobj yet
1557 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1558 const struct lu_attr *pattr,
1559 struct mdd_object *cobj,
1560 const struct lu_attr *cattr)
1565 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1570 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1571 struct mdd_object *p, struct mdd_object *c,
1572 const struct lu_name *name, struct md_attr *ma,
1573 struct thandle *handle, int no_name, int is_dir)
1575 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1578 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1579 if (likely(no_name == 0)) {
1580 rc = mdo_declare_index_delete(env, p, name->ln_name,
1587 rc = mdo_declare_ref_del(env, p, handle);
1593 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1594 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1595 la->la_valid = LA_CTIME | LA_MTIME;
1596 rc = mdo_declare_attr_set(env, p, la, handle);
1601 rc = mdo_declare_ref_del(env, c, handle);
1605 rc = mdo_declare_ref_del(env, c, handle);
1609 la->la_valid = LA_CTIME;
1610 rc = mdo_declare_attr_set(env, c, la, handle);
1614 rc = mdd_declare_finish_unlink(env, c, handle);
1618 /* FIXME: need changelog for remove entry */
1619 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1627 * test if a file has an HSM archive
1628 * if HSM attributes are not found in ma update them from
1631 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1632 struct mdd_object *obj,
1637 if (!(ma->ma_valid & MA_HSM)) {
1638 /* no HSM MD provided, read xattr */
1639 struct lu_buf *hsm_buf;
1640 const size_t buflen = sizeof(struct hsm_attrs);
1643 hsm_buf = mdd_buf_get(env, NULL, 0);
1644 lu_buf_alloc(hsm_buf, buflen);
1645 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1646 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1647 lu_buf_free(hsm_buf);
1651 ma->ma_valid |= MA_HSM;
1653 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1659 * Delete name entry and the object.
1660 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1661 * does not exist for this object, and it could only happen during resending
1662 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1663 * is also needed in this case(needed by changelog), so we have to add another
1664 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1665 * the ENOENT failure should be able to be fixed by redo mechanism.
1667 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1668 struct md_object *cobj, const struct lu_name *lname,
1669 struct md_attr *ma, int no_name)
1671 const char *name = lname->ln_name;
1672 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1673 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1674 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1675 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1676 struct mdd_object *mdd_cobj = NULL;
1677 struct mdd_device *mdd = mdo2mdd(pobj);
1678 struct thandle *handle;
1679 int rc, is_dir = 0, cl_flags = 0;
1682 /* let shutdown to start */
1683 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1685 /* cobj == NULL means only delete name entry */
1686 if (likely(cobj != NULL)) {
1687 mdd_cobj = md2mdd_obj(cobj);
1688 if (mdd_object_exists(mdd_cobj) == 0)
1692 rc = mdd_la_get(env, mdd_pobj, pattr);
1696 if (likely(mdd_cobj != NULL)) {
1698 rc = mdd_la_get(env, mdd_cobj, cattr);
1702 is_dir = S_ISDIR(cattr->la_mode);
1703 /* search for an existing archive.
1704 * we should check ahead as the object
1705 * can be destroyed in this transaction */
1706 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1707 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1710 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1714 handle = mdd_trans_create(env, mdd);
1716 RETURN(PTR_ERR(handle));
1718 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1719 lname, ma, handle, no_name, is_dir);
1723 rc = mdd_trans_start(env, mdd, handle);
1727 if (likely(mdd_cobj != NULL))
1728 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1730 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1731 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1736 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1737 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1738 GOTO(cleanup, rc = 0);
1740 if (likely(mdd_cobj != NULL)) {
1741 rc = mdo_ref_del(env, mdd_cobj, handle);
1743 __mdd_index_insert_only(env, mdd_pobj,
1745 mdd_object_type(mdd_cobj),
1752 mdo_ref_del(env, mdd_cobj, handle);
1754 /* fetch updated nlink */
1755 rc = mdd_la_get(env, mdd_cobj, cattr);
1760 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1761 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1763 la->la_valid = LA_CTIME | LA_MTIME;
1764 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1768 /* Enough for only unlink the entry */
1769 if (unlikely(mdd_cobj == NULL))
1772 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1773 /* update ctime of an unlinked file only if it is still
1774 * opened or a link still exists */
1775 la->la_valid = LA_CTIME;
1776 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1781 /* XXX: this transfer to ma will be removed with LOD/OSP */
1782 ma->ma_attr = *cattr;
1783 ma->ma_valid |= MA_INODE;
1784 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1788 /* fetch updated nlink */
1789 rc = mdd_la_get(env, mdd_cobj, cattr);
1790 /* if object is removed then we can't get its attrs,
1792 if (rc == -ENOENT) {
1793 cattr->la_nlink = 0;
1797 if (cattr->la_nlink == 0) {
1798 ma->ma_attr = *cattr;
1799 ma->ma_valid |= MA_INODE;
1804 if (likely(mdd_cobj != NULL))
1805 mdd_write_unlock(env, mdd_cobj);
1808 if (cattr->la_nlink == 0)
1809 cl_flags |= CLF_UNLINK_LAST;
1811 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1813 rc = mdd_changelog_ns_store(env, mdd,
1814 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1815 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1820 rc = mdd_trans_stop(env, mdd, rc, handle);
1826 * The permission has been checked when obj created, no need check again.
1828 static int mdd_cd_sanity_check(const struct lu_env *env,
1829 struct mdd_object *obj)
1834 if (!obj || mdd_is_dead_obj(obj))
1840 static int mdd_create_data(const struct lu_env *env,
1841 struct md_object *pobj,
1842 struct md_object *cobj,
1843 const struct md_op_spec *spec,
1846 struct mdd_device *mdd = mdo2mdd(cobj);
1847 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1848 struct mdd_object *son = md2mdd_obj(cobj);
1849 struct thandle *handle;
1850 const struct lu_buf *buf;
1851 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1852 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1856 rc = mdd_cd_sanity_check(env, son);
1860 if (!md_should_create(spec->sp_cr_flags))
1864 * there are following use cases for this function:
1865 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1866 * striping can be specified or not
1869 rc = mdd_la_get(env, son, attr);
1873 /* calling ->ah_make_hint() is used to transfer information from parent */
1874 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1876 handle = mdd_trans_create(env, mdd);
1878 GOTO(out_free, rc = PTR_ERR(handle));
1881 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1882 * Should this be fixed?
1884 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1885 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1886 spec->sp_cr_flags, spec->no_create);
1888 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1889 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1890 spec->u.sp_ea.eadatalen);
1895 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
1896 XATTR_NAME_LOV, 0, handle);
1900 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
1905 rc = mdd_trans_start(env, mdd, handle);
1909 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
1915 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
1918 rc = mdd_trans_stop(env, mdd, rc, handle);
1924 static int mdd_declare_object_initialize(const struct lu_env *env,
1925 struct mdd_object *parent,
1926 struct mdd_object *child,
1927 const struct lu_attr *attr,
1928 struct thandle *handle)
1933 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
1934 if (!S_ISDIR(attr->la_mode))
1937 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
1942 rc = mdo_declare_ref_add(env, child, handle);
1946 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
1952 static int mdd_object_initialize(const struct lu_env *env,
1953 const struct lu_fid *pfid,
1954 struct mdd_object *child,
1955 struct lu_attr *attr, struct thandle *handle,
1956 const struct md_op_spec *spec)
1961 if (S_ISDIR(attr->la_mode)) {
1962 /* Add "." and ".." for newly created dir */
1963 mdo_ref_add(env, child, handle);
1964 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1965 S_IFDIR, dot, handle);
1967 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
1970 mdo_ref_del(env, child, handle);
1977 * This function checks whether it can create a file/dir under the
1978 * directory(@pobj). The directory(@pobj) is not being locked by
1981 * \param[in] env execution environment
1982 * \param[in] pobj the directory to create files
1983 * \param[in] pattr the attributes of the directory
1984 * \param[in] lname the name of the created file/dir
1985 * \param[in] cattr the attributes of the file/dir
1986 * \param[in] spec create specification
1988 * \retval = 0 it is allowed to create file/dir under
1990 * \retval negative error not allowed to create file/dir
1991 * under the directory
1993 static int mdd_create_sanity_check(const struct lu_env *env,
1994 struct md_object *pobj,
1995 const struct lu_attr *pattr,
1996 const struct lu_name *lname,
1997 struct lu_attr *cattr,
1998 struct md_op_spec *spec)
2000 struct mdd_thread_info *info = mdd_env_info(env);
2001 struct lu_fid *fid = &info->mti_fid;
2002 struct mdd_object *obj = md2mdd_obj(pobj);
2003 struct mdd_device *m = mdo2mdd(pobj);
2004 bool check_perm = true;
2009 if (mdd_is_dead_obj(obj))
2013 * In some cases this lookup is not needed - we know before if name
2014 * exists or not because MDT performs lookup for it.
2015 * name length check is done in lookup.
2017 if (spec->sp_cr_lookup) {
2019 * Check if the name already exist, though it will be checked in
2020 * _index_insert also, for avoiding rolling back if exists
2023 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2024 MAY_WRITE | MAY_EXEC);
2026 RETURN(rc ? : -EEXIST);
2028 /* Permission is already being checked in mdd_lookup */
2032 if (S_ISDIR(cattr->la_mode) &&
2033 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2034 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2035 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2037 if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
2038 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
2039 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2041 CERROR("%s: invalid lmv_user_md: magic = %x, "
2042 "stripe_offset = %d, stripe_count = %u: "
2043 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2044 le32_to_cpu(lum->lum_magic),
2045 (int)le32_to_cpu(lum->lum_stripe_offset),
2046 le32_to_cpu(lum->lum_stripe_count), rc);
2051 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2056 if (pattr->la_mode & S_ISGID) {
2057 cattr->la_gid = pattr->la_gid;
2058 if (S_ISDIR(cattr->la_mode)) {
2059 cattr->la_mode |= S_ISGID;
2060 cattr->la_valid |= LA_MODE;
2064 /* Inherit project ID from parent directory */
2065 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2066 cattr->la_projid = pattr->la_projid;
2067 if (S_ISDIR(cattr->la_mode)) {
2068 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2069 cattr->la_valid |= LA_FLAGS;
2071 cattr->la_valid |= LA_PROJID;
2074 rc = mdd_name_check(m, lname);
2078 switch (cattr->la_mode & S_IFMT) {
2080 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2082 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2083 RETURN(-ENAMETOOLONG);
2102 static int mdd_declare_create_object(const struct lu_env *env,
2103 struct mdd_device *mdd,
2104 struct mdd_object *p, struct mdd_object *c,
2105 struct lu_attr *attr,
2106 struct thandle *handle,
2107 const struct md_op_spec *spec,
2108 struct lu_buf *def_acl_buf,
2109 struct lu_buf *acl_buf,
2110 struct dt_allocation_hint *hint)
2112 const struct lu_buf *buf;
2115 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2120 #ifdef CONFIG_FS_POSIX_ACL
2121 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2122 /* if dir, then can inherit default ACl */
2123 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2124 XATTR_NAME_ACL_DEFAULT,
2130 if (acl_buf->lb_len > 0) {
2131 rc = mdo_declare_attr_set(env, c, attr, handle);
2135 rc = mdo_declare_xattr_set(env, c, acl_buf,
2136 XATTR_NAME_ACL_ACCESS, 0, handle);
2141 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2145 /* replay case, create LOV EA from client data */
2146 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2147 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2148 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2149 spec->u.sp_ea.eadatalen);
2150 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2156 if (S_ISLNK(attr->la_mode)) {
2157 const char *target_name = spec->u.sp_symname;
2158 int sym_len = strlen(target_name);
2159 const struct lu_buf *buf;
2161 buf = mdd_buf_get_const(env, target_name, sym_len);
2162 rc = dt_declare_record_write(env, mdd_object_child(c),
2168 if (spec->sp_cr_file_secctx_name != NULL) {
2169 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2170 spec->sp_cr_file_secctx_size);
2171 rc = mdo_declare_xattr_set(env, c, buf,
2172 spec->sp_cr_file_secctx_name, 0,
2181 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2182 struct mdd_object *p, struct mdd_object *c,
2183 const struct lu_name *name,
2184 struct lu_attr *attr,
2185 struct thandle *handle,
2186 const struct md_op_spec *spec,
2187 struct linkea_data *ldata,
2188 struct lu_buf *def_acl_buf,
2189 struct lu_buf *acl_buf,
2190 struct dt_allocation_hint *hint)
2194 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2195 def_acl_buf, acl_buf, hint);
2199 if (S_ISDIR(attr->la_mode)) {
2200 rc = mdo_declare_ref_add(env, p, handle);
2205 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2206 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2210 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2211 enum changelog_rec_type type;
2213 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2214 name->ln_name, handle);
2218 rc = mdd_declare_links_add(env, c, handle, ldata);
2223 la->la_valid = LA_CTIME | LA_MTIME;
2224 rc = mdo_declare_attr_set(env, p, la, handle);
2228 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2229 S_ISREG(attr->la_mode) ? CL_CREATE :
2230 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2232 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2241 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2242 struct lu_attr *la, struct lu_buf *def_acl_buf,
2243 struct lu_buf *acl_buf)
2248 if (S_ISLNK(la->la_mode)) {
2249 acl_buf->lb_len = 0;
2250 def_acl_buf->lb_len = 0;
2254 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2255 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2256 XATTR_NAME_ACL_DEFAULT);
2257 mdd_read_unlock(env, pobj);
2259 /* If there are default ACL, fix mode/ACL by default ACL */
2260 def_acl_buf->lb_len = rc;
2261 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2262 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2263 acl_buf->lb_len = rc;
2264 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2267 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2268 /* If there are no default ACL, fix mode by mask */
2269 struct lu_ucred *uc = lu_ucred(env);
2271 /* The create triggered by MDT internal events, such as
2272 * LFSCK reset, will not contain valid "uc". */
2273 if (unlikely(uc != NULL))
2274 la->la_mode &= ~uc->uc_umask;
2276 acl_buf->lb_len = 0;
2277 def_acl_buf->lb_len = 0;
2284 * Create a metadata object and initialize it, set acl, xattr.
2286 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2287 struct mdd_object *son, struct lu_attr *attr,
2288 struct md_op_spec *spec, struct lu_buf *acl_buf,
2289 struct lu_buf *def_acl_buf,
2290 struct dt_allocation_hint *hint,
2291 struct thandle *handle)
2293 const struct lu_buf *buf;
2296 mdd_write_lock(env, son, MOR_TGT_CHILD);
2297 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2302 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2303 * created in declare phase, they also needs to be added to master
2304 * object as sub-directory entry. So it has to initialize the master
2305 * object, then set dir striped EA.(in mdo_xattr_set) */
2306 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2309 GOTO(err_destroy, rc);
2312 * in case of replay we just set LOVEA provided by the client
2313 * XXX: I think it would be interesting to try "old" way where
2314 * MDT calls this xattr_set(LOV) in a different transaction.
2315 * probably this way we code can be made better.
2318 /* During creation, there are only a few cases we need do xattr_set to
2320 * 1. regular file: see comments above.
2321 * 2. dir: inherit default striping or pool settings from parent.
2322 * 3. create striped directory with provided stripeEA.
2323 * 4. create striped directory because inherit default layout from the
2326 if (spec->no_create ||
2327 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2328 S_ISDIR(attr->la_mode)) {
2329 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2330 spec->u.sp_ea.eadatalen);
2331 rc = mdo_xattr_set(env, son, buf,
2332 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2336 GOTO(err_destroy, rc);
2339 #ifdef CONFIG_FS_POSIX_ACL
2340 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2341 S_ISDIR(attr->la_mode)) {
2342 /* set default acl */
2343 rc = mdo_xattr_set(env, son, def_acl_buf,
2344 XATTR_NAME_ACL_DEFAULT, 0,
2347 GOTO(err_destroy, rc);
2349 /* set its own acl */
2350 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2351 rc = mdo_xattr_set(env, son, acl_buf,
2352 XATTR_NAME_ACL_ACCESS,
2355 GOTO(err_destroy, rc);
2359 if (S_ISLNK(attr->la_mode)) {
2360 struct lu_ucred *uc = lu_ucred_assert(env);
2361 struct dt_object *dt = mdd_object_child(son);
2362 const char *target_name = spec->u.sp_symname;
2363 int sym_len = strlen(target_name);
2366 buf = mdd_buf_get_const(env, target_name, sym_len);
2367 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2369 CFS_CAP_SYS_RESOURCE_MASK);
2374 GOTO(err_initlized, rc = -EFAULT);
2377 if (spec->sp_cr_file_secctx_name != NULL) {
2378 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2379 spec->sp_cr_file_secctx_size);
2380 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2383 GOTO(err_initlized, rc);
2387 if (unlikely(rc != 0)) {
2389 if (S_ISDIR(attr->la_mode)) {
2390 /* Drop the reference, no need to delete "."/"..",
2391 * because the object to be destroied directly. */
2392 rc2 = mdo_ref_del(env, son, handle);
2396 rc2 = mdo_ref_del(env, son, handle);
2400 mdo_destroy(env, son, handle);
2403 mdd_write_unlock(env, son);
2407 static int mdd_index_delete(const struct lu_env *env,
2408 struct mdd_object *mdd_pobj,
2409 struct lu_attr *cattr,
2410 const struct lu_name *lname)
2412 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2413 struct thandle *handle;
2417 handle = mdd_trans_create(env, mdd);
2419 RETURN(PTR_ERR(handle));
2421 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2426 if (S_ISDIR(cattr->la_mode)) {
2427 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2432 /* Since this will only be used in the error handler path,
2433 * Let's set the thandle to be local and not mess the transno */
2434 handle->th_local = 1;
2435 rc = mdd_trans_start(env, mdd, handle);
2439 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2440 S_ISDIR(cattr->la_mode), handle);
2444 rc = mdd_trans_stop(env, mdd, rc, handle);
2450 * Create object and insert it into namespace.
2452 * Two operations have to be performed:
2454 * - an allocation of a new object (->do_create()), and
2455 * - an insertion into a parent index (->dio_insert()).
2457 * Due to locking, operation order is not important, when both are
2458 * successful, *but* error handling cases are quite different:
2460 * - if insertion is done first, and following object creation fails,
2461 * insertion has to be rolled back, but this operation might fail
2462 * also leaving us with dangling index entry.
2464 * - if creation is done first, is has to be undone if insertion fails,
2465 * leaving us with leaked space, which is not good but not fatal.
2467 * It seems that creation-first is simplest solution, but it is sub-optimal
2473 * case, because second mkdir is bound to create object, only to
2474 * destroy it immediately.
2476 * To avoid this follow local file systems that do double lookup:
2478 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2479 * 1. create (mdd_create_object_internal())
2480 * 2. insert (__mdd_index_insert(), lookup again)
2482 * \param[in] pobj parent object
2483 * \param[in] lname name of child being created
2484 * \param[in,out] child child object being created
2485 * \param[in] spec additional create parameters
2486 * \param[in] ma attributes for new child object
2488 * \retval 0 on success
2489 * \retval negative errno on failure
2491 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2492 const struct lu_name *lname, struct md_object *child,
2493 struct md_op_spec *spec, struct md_attr *ma)
2495 struct mdd_thread_info *info = mdd_env_info(env);
2496 struct lu_attr *la = &info->mti_la_for_fix;
2497 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2498 struct mdd_object *son = md2mdd_obj(child);
2499 struct mdd_device *mdd = mdo2mdd(pobj);
2500 struct lu_attr *attr = &ma->ma_attr;
2501 struct thandle *handle;
2502 struct lu_attr *pattr = &info->mti_pattr;
2503 struct lu_buf acl_buf;
2504 struct lu_buf def_acl_buf;
2505 struct linkea_data *ldata = &info->mti_link_data;
2506 const char *name = lname->ln_name;
2507 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2512 rc = mdd_la_get(env, mdd_pobj, pattr);
2516 /* Sanity checks before big job. */
2517 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2521 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2522 GOTO(out_free, rc = -EINPROGRESS);
2524 handle = mdd_trans_create(env, mdd);
2526 GOTO(out_free, rc = PTR_ERR(handle));
2528 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2529 mdd->mdd_dt_conf.ddp_max_ea_size);
2530 acl_buf = info->mti_xattr_buf;
2531 def_acl_buf.lb_buf = info->mti_key;
2532 def_acl_buf.lb_len = sizeof(info->mti_key);
2533 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2537 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2539 memset(ldata, 0, sizeof(*ldata));
2540 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2541 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2544 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2545 &tfid, lname, 1, 0, ldata);
2547 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2548 mdd_object_fid(mdd_pobj),
2549 lname, 1, 0, ldata);
2552 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2553 handle, spec, ldata, &def_acl_buf, &acl_buf,
2558 rc = mdd_trans_start(env, mdd, handle);
2562 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2563 &def_acl_buf, hint, handle);
2567 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2568 mdd_write_lock(env, son, MOR_TGT_CHILD);
2569 son->mod_flags |= VOLATILE_OBJ;
2570 rc = mdd_orphan_insert(env, son, handle);
2571 GOTO(out_volatile, rc);
2573 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2574 attr->la_mode, name, handle);
2576 GOTO(err_created, rc);
2578 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2581 /* update parent directory mtime/ctime */
2583 la->la_valid = LA_CTIME | LA_MTIME;
2584 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2586 GOTO(err_insert, rc);
2592 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2593 rc2 = mdd_orphan_delete(env, son, handle);
2595 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2596 S_ISDIR(attr->la_mode),
2602 mdd_write_lock(env, son, MOR_TGT_CHILD);
2603 if (S_ISDIR(attr->la_mode)) {
2604 /* Drop the reference, no need to delete "."/"..",
2605 * because the object is to be destroyed directly. */
2606 rc2 = mdo_ref_del(env, son, handle);
2608 mdd_write_unlock(env, son);
2613 /* For volatile files drop one link immediately, since there is
2614 * no filename in the namespace, and save any error returned. */
2615 rc2 = mdo_ref_del(env, son, handle);
2617 mdd_write_unlock(env, son);
2618 if (unlikely(rc == 0))
2623 /* Don't destroy the volatile object on success */
2624 if (likely(rc != 0))
2625 mdo_destroy(env, son, handle);
2626 mdd_write_unlock(env, son);
2629 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2630 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2631 rc = mdd_changelog_ns_store(env, mdd,
2632 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2633 S_ISREG(attr->la_mode) ? CL_CREATE :
2634 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2635 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2638 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2640 /* If creation fails, it is most likely due to the remote update
2641 * failure, because local transaction will mostly succeed at
2642 * this stage. There is no easy way to rollback all of previous
2643 * updates, so let's remove the object from namespace, and
2644 * LFSCK should handle the orphan object. */
2645 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2646 mdd_index_delete(env, mdd_pobj, attr, lname);
2650 if (is_vmalloc_addr(ldata->ld_buf))
2651 /* if we vmalloced a large buffer drop it */
2652 lu_buf_free(ldata->ld_buf);
2654 /* The child object shouldn't be cached anymore */
2656 set_bit(LU_OBJECT_HEARD_BANSHEE,
2657 &child->mo_lu.lo_header->loh_flags);
2662 * Get locks on parents in proper order
2663 * RETURN: < 0 - error, rename_order if successful
2671 static int mdd_rename_order(const struct lu_env *env,
2672 struct mdd_device *mdd,
2673 struct mdd_object *src_pobj,
2674 const struct lu_attr *pattr,
2675 struct mdd_object *tgt_pobj)
2677 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2681 if (src_pobj == tgt_pobj)
2682 RETURN(MDD_RN_SAME);
2684 /* compared the parent child relationship of src_p&tgt_p */
2685 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2687 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2690 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2704 /* has not mdd_write{read}_lock on any obj yet. */
2705 static int mdd_rename_sanity_check(const struct lu_env *env,
2706 struct mdd_object *src_pobj,
2707 const struct lu_attr *pattr,
2708 struct mdd_object *tgt_pobj,
2709 const struct lu_attr *tpattr,
2710 struct mdd_object *sobj,
2711 const struct lu_attr *cattr,
2712 struct mdd_object *tobj,
2713 const struct lu_attr *tattr)
2718 /* XXX: when get here, sobj must NOT be NULL,
2719 * the other case has been processed in cld_rename
2720 * before mdd_rename and enable MDS_PERM_BYPASS. */
2724 * If we are using project inheritance, we only allow renames
2725 * into our tree when the project IDs are the same; otherwise
2726 * tree quota mechanism would be circumvented.
2728 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2729 tpattr->la_projid != cattr->la_projid) ||
2730 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2731 (pattr->la_projid != tpattr->la_projid)))
2734 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2738 /* XXX: when get here, "tobj == NULL" means tobj must
2739 * NOT exist (neither on remote MDS, such case has been
2740 * processed in cld_rename before mdd_rename and enable
2742 * So check may_create, but not check may_unlink. */
2744 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2745 (src_pobj != tgt_pobj));
2747 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2748 (src_pobj != tgt_pobj), 1);
2750 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2751 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2756 static int mdd_declare_rename(const struct lu_env *env,
2757 struct mdd_device *mdd,
2758 struct mdd_object *mdd_spobj,
2759 struct mdd_object *mdd_tpobj,
2760 struct mdd_object *mdd_sobj,
2761 struct mdd_object *mdd_tobj,
2762 const struct lu_name *sname,
2763 const struct lu_name *tname,
2765 struct linkea_data *ldata,
2766 struct thandle *handle)
2768 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2771 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2772 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2778 /* name from source dir */
2779 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2783 /* .. from source child */
2784 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2785 /* source child can be directory,
2786 * counted by source dir's nlink */
2787 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2790 if (mdd_spobj != mdd_tpobj) {
2791 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2796 rc = mdo_declare_index_insert(env, mdd_sobj,
2798 S_IFDIR, dotdot, handle);
2803 /* new target child can be directory,
2804 * counted by target dir's nlink */
2805 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2810 la->la_valid = LA_CTIME | LA_MTIME;
2811 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2815 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2819 la->la_valid = LA_CTIME;
2820 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2824 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2829 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2830 mdd_object_type(mdd_sobj),
2831 tname->ln_name, handle);
2835 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2836 /* delete target child in target parent directory */
2837 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2842 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2846 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2847 /* target child can be directory,
2848 * delete "." reference in target child directory */
2849 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2853 /* delete ".." reference in target parent directory */
2854 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2859 la->la_valid = LA_CTIME;
2860 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2864 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2869 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
2877 /* src object can be remote that is why we use only fid and type of object */
2878 static int mdd_rename(const struct lu_env *env,
2879 struct md_object *src_pobj, struct md_object *tgt_pobj,
2880 const struct lu_fid *lf, const struct lu_name *lsname,
2881 struct md_object *tobj, const struct lu_name *ltname,
2884 const char *sname = lsname->ln_name;
2885 const char *tname = ltname->ln_name;
2886 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2887 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2888 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2889 struct mdd_device *mdd = mdo2mdd(src_pobj);
2890 struct mdd_object *mdd_sobj = NULL; /* source object */
2891 struct mdd_object *mdd_tobj = NULL;
2892 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2893 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2894 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2895 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2896 struct thandle *handle;
2897 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2898 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2899 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2902 bool tobj_locked = 0;
2903 unsigned cl_flags = 0;
2907 /* let unlink to complete and commit */
2908 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
2911 mdd_tobj = md2mdd_obj(tobj);
2913 mdd_sobj = mdd_object_find(env, mdd, lf);
2914 if (IS_ERR(mdd_sobj))
2915 RETURN(PTR_ERR(mdd_sobj));
2917 rc = mdd_la_get(env, mdd_sobj, cattr);
2919 GOTO(out_pending, rc);
2921 rc = mdd_la_get(env, mdd_spobj, pattr);
2923 GOTO(out_pending, rc);
2926 rc = mdd_la_get(env, mdd_tobj, tattr);
2928 GOTO(out_pending, rc);
2929 /* search for an existing archive.
2930 * we should check ahead as the object
2931 * can be destroyed in this transaction */
2932 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
2933 cl_flags |= CLF_RENAME_LAST_EXISTS;
2936 rc = mdd_la_get(env, mdd_tpobj, tpattr);
2938 GOTO(out_pending, rc);
2940 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
2941 mdd_sobj, cattr, mdd_tobj, tattr);
2943 GOTO(out_pending, rc);
2945 rc = mdd_name_check(mdd, ltname);
2947 GOTO(out_pending, rc);
2949 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2950 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
2952 GOTO(out_pending, rc);
2954 handle = mdd_trans_create(env, mdd);
2956 GOTO(out_pending, rc = PTR_ERR(handle));
2958 memset(ldata, 0, sizeof(*ldata));
2959 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
2960 lsname, mdd_object_fid(mdd_tpobj), ltname,
2965 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2966 mdd_tobj, lsname, ltname, ma, ldata, handle);
2970 rc = mdd_trans_start(env, mdd, handle);
2974 is_dir = S_ISDIR(cattr->la_mode);
2976 /* Remove source name from source directory */
2977 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
2981 /* "mv dir1 dir2" needs "dir1/.." link update */
2982 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2983 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
2985 GOTO(fixup_spobj2, rc);
2987 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
2990 GOTO(fixup_spobj, rc);
2993 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
2994 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
2996 /* tname might been renamed to something else */
2997 GOTO(fixup_spobj, rc);
3000 /* Insert new fid with target name into target dir */
3001 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3004 GOTO(fixup_tpobj, rc);
3006 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3007 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3009 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3010 la->la_valid = LA_CTIME;
3011 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3013 GOTO(fixup_tpobj, rc);
3015 /* Update the linkEA for the source object */
3016 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3017 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3018 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3021 /* Old files might not have EA entry */
3022 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3023 lsname, handle, NULL, 0);
3024 mdd_write_unlock(env, mdd_sobj);
3025 /* We don't fail the transaction if the link ea can't be
3026 updated -- fid2path will use alternate lookup method. */
3029 /* Remove old target object
3030 * For tobj is remote case cmm layer has processed
3031 * and set tobj to NULL then. So when tobj is NOT NULL,
3032 * it must be local one.
3034 if (tobj && mdd_object_exists(mdd_tobj)) {
3035 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
3037 if (mdd_is_dead_obj(mdd_tobj)) {
3038 /* shld not be dead, something is wrong */
3039 CERROR("tobj is dead, something is wrong\n");
3043 mdo_ref_del(env, mdd_tobj, handle);
3045 /* Remove dot reference. */
3046 if (S_ISDIR(tattr->la_mode))
3047 mdo_ref_del(env, mdd_tobj, handle);
3050 /* fetch updated nlink */
3051 rc = mdd_la_get(env, mdd_tobj, tattr);
3053 CERROR("%s: Failed to get nlink for tobj "
3055 mdd2obd_dev(mdd)->obd_name,
3056 PFID(tpobj_fid), rc);
3057 GOTO(fixup_tpobj, rc);
3060 la->la_valid = LA_CTIME;
3061 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3063 CERROR("%s: Failed to set ctime for tobj "
3065 mdd2obd_dev(mdd)->obd_name,
3066 PFID(tpobj_fid), rc);
3067 GOTO(fixup_tpobj, rc);
3070 /* XXX: this transfer to ma will be removed with LOD/OSP */
3071 ma->ma_attr = *tattr;
3072 ma->ma_valid |= MA_INODE;
3073 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3076 CERROR("%s: Failed to unlink tobj "
3078 mdd2obd_dev(mdd)->obd_name,
3079 PFID(tpobj_fid), rc);
3080 GOTO(fixup_tpobj, rc);
3083 /* fetch updated nlink */
3084 rc = mdd_la_get(env, mdd_tobj, tattr);
3085 if (rc == -ENOENT) {
3086 /* the object got removed, let's
3087 * return the latest known attributes */
3088 tattr->la_nlink = 0;
3090 } else if (rc != 0) {
3091 CERROR("%s: Failed to get nlink for tobj "
3093 mdd2obd_dev(mdd)->obd_name,
3094 PFID(tpobj_fid), rc);
3095 GOTO(fixup_tpobj, rc);
3097 /* XXX: this transfer to ma will be removed with LOD/OSP */
3098 ma->ma_attr = *tattr;
3099 ma->ma_valid |= MA_INODE;
3101 if (tattr->la_nlink == 0)
3102 cl_flags |= CLF_RENAME_LAST;
3104 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3107 la->la_valid = LA_CTIME | LA_MTIME;
3108 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3110 GOTO(fixup_tpobj, rc);
3112 if (mdd_spobj != mdd_tpobj) {
3113 la->la_valid = LA_CTIME | LA_MTIME;
3114 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3116 GOTO(fixup_tpobj, rc);
3123 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3125 CWARN("tp obj fix error %d\n",rc2);
3127 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3128 !mdd_is_dead_obj(mdd_tobj)) {
3130 mdo_ref_add(env, mdd_tobj, handle);
3132 mdo_ref_add(env, mdd_tobj, handle);
3135 rc2 = __mdd_index_insert(env, mdd_tpobj,
3137 mdd_object_type(mdd_tobj),
3140 CWARN("tp obj fix error: rc = %d\n", rc2);
3145 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3146 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3148 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3149 mdd2obd_dev(mdd)->obd_name, rc2);
3152 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3155 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3156 mdd2obd_dev(mdd)->obd_name, rc2);
3161 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3162 mdd_object_type(mdd_sobj), sname,
3165 CWARN("sp obj fix error: rc = %d\n", rc2);
3170 mdd_write_unlock(env, mdd_tobj);
3173 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3174 mdd_tobj, tpobj_fid, lf, spobj_fid,
3175 ltname, lsname, handle);
3178 rc = mdd_trans_stop(env, mdd, rc, handle);
3181 mdd_object_put(env, mdd_sobj);
3186 * During migration once the parent FID has been changed,
3187 * we need update the parent FID in linkea.
3189 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3190 struct mdd_object *parent,
3191 struct mdd_object *newparent,
3192 struct mdd_object *child,
3193 const char *name, int namelen,
3194 struct thandle *handle,
3197 struct mdd_thread_info *info = mdd_env_info(env);
3198 struct linkea_data ldata = { NULL };
3199 struct lu_buf *buf = &info->mti_link_buf;
3205 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3206 if (buf->lb_buf == NULL)
3210 rc = mdd_links_read(env, child, &ldata);
3212 if (rc == -ENOENT || rc == -ENODATA)
3217 LASSERT(ldata.ld_leh != NULL);
3218 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3219 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3220 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3221 struct lu_name lname;
3224 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3227 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3228 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3229 ldata.ld_lee = (struct link_ea_entry *)
3230 ((char *)ldata.ld_lee +
3235 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3236 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3237 lname.ln_namelen, lname.ln_name,
3238 PFID(mdd_object_fid(newparent)));
3239 /* update to the new parent fid */
3240 linkea_entry_pack(ldata.ld_lee, &lname,
3241 mdd_object_fid(newparent));
3243 rc = mdd_declare_links_add(env, child, handle, &ldata);
3245 rc = mdd_links_write(env, child, &ldata, handle);
3251 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3252 struct mdd_object *parent,
3253 struct mdd_object *newparent,
3254 struct mdd_object *child,
3255 const char *name, int namelen,
3256 struct thandle *handle)
3258 return mdd_linkea_update_child_internal(env, parent, newparent,
3260 namelen, handle, true);
3263 static int mdd_linkea_update_child(const struct lu_env *env,
3264 struct mdd_object *parent,
3265 struct mdd_object *newparent,
3266 struct mdd_object *child,
3267 const char *name, int namelen,
3268 struct thandle *handle)
3270 return mdd_linkea_update_child_internal(env, parent, newparent,
3272 namelen, handle, false);
3275 static int mdd_update_linkea_internal(const struct lu_env *env,
3276 struct mdd_object *mdd_pobj,
3277 struct mdd_object *mdd_sobj,
3278 struct mdd_object *mdd_tobj,
3279 const struct lu_name *child_name,
3280 struct linkea_data *ldata,
3281 struct thandle *handle,
3284 struct mdd_thread_info *info = mdd_env_info(env);
3289 LASSERT(ldata->ld_buf != NULL);
3290 LASSERT(ldata->ld_leh != NULL);
3292 /* If it is mulitple links file, we need update the name entry for
3294 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3295 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3296 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3297 struct mdd_object *pobj;
3298 struct lu_name lname;
3301 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3303 pobj = mdd_object_find(env, mdd, &fid);
3305 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3306 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3311 if (!mdd_object_exists(pobj)) {
3312 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3313 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3317 if (pobj == mdd_pobj &&
3318 lname.ln_namelen == child_name->ln_namelen &&
3319 strncmp(lname.ln_name, child_name->ln_name,
3320 lname.ln_namelen) == 0) {
3321 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3322 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3327 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3328 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3329 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3332 /* Remove source name from source directory */
3333 /* Insert new fid with target name into target dir */
3334 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3339 rc = mdo_declare_index_insert(env, pobj,
3340 mdd_object_fid(mdd_tobj),
3341 mdd_object_type(mdd_tobj),
3342 lname.ln_name, handle);
3346 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3350 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3354 char *tmp_name = info->mti_key;
3356 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3357 /* lnamelen is too big(> NAME_MAX + 16),
3358 * something wrong about this linkea, let's
3360 CWARN("%s: the name %.*s is too long under "
3361 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3362 lname.ln_namelen, lname.ln_name,
3367 /* Note: lname might be without \0 at the end, see
3368 * linkea_entry_unpack(), let's add extra \0 by
3370 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3371 lname.ln_namelen, lname.ln_name);
3372 lname.ln_name = tmp_name;
3374 /* Let's check if this linkEA still valid, before
3375 * it might be packed into the RPC buffer. */
3376 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3377 &info->mti_fid, NULL);
3378 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3379 mdd_object_fid(mdd_sobj)))
3380 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3382 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3386 rc = __mdd_index_insert(env, pobj,
3387 mdd_object_fid(mdd_tobj),
3388 mdd_object_type(mdd_tobj),
3393 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3394 rc = mdo_ref_add(env, mdd_tobj, handle);
3395 mdd_write_unlock(env, mdd_tobj);
3399 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3400 mdo_ref_del(env, mdd_sobj, handle);
3401 mdd_write_unlock(env, mdd_sobj);
3404 mdd_object_put(env, pobj);
3408 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3415 static int mdd_migrate_xattrs(const struct lu_env *env,
3416 struct mdd_object *mdd_sobj,
3417 struct mdd_object *mdd_tobj)
3419 struct mdd_thread_info *info = mdd_env_info(env);
3420 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3422 struct thandle *handle;
3428 struct lu_buf list_xbuf;
3431 /* retrieve xattr list from the old object */
3432 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3433 if (list_xsize == -ENODATA)
3439 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3440 if (info->mti_big_buf.lb_buf == NULL)
3443 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3444 list_xbuf.lb_len = list_xsize;
3445 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3450 xname = list_xbuf.lb_buf;
3452 xlen = strnlen(xname, rem - 1) + 1;
3453 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3454 strcmp(XATTR_NAME_LMV, xname) == 0)
3457 /* For directory, if there are default layout, migrate here */
3458 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3459 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3462 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3463 if (xsize == -ENODATA)
3468 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3469 if (info->mti_link_buf.lb_buf == NULL)
3470 GOTO(out, rc = -ENOMEM);
3472 xbuf.lb_len = xsize;
3473 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3474 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3480 handle = mdd_trans_create(env, mdd);
3482 GOTO(out, rc = PTR_ERR(handle));
3484 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3487 GOTO(stop_trans, rc);
3488 /* Note: this transaction is part of migration, and it is not
3489 * the last step of migration, so we set th_local = 1 to avoid
3490 * update last rcvd for this transaction */
3491 handle->th_local = 1;
3492 rc = mdd_trans_start(env, mdd, handle);
3494 GOTO(stop_trans, rc);
3497 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3499 GOTO(stop_trans, rc = 0);
3501 if (unlikely(rc == -ENOSPC &&
3502 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3503 rc = linkea_overflow_shrink(
3504 (struct linkea_data *)(xbuf.lb_buf));
3505 if (likely(rc > 0)) {
3512 GOTO(stop_trans, rc);
3514 rc = mdd_trans_stop(env, mdd, rc, handle);
3519 memmove(xname, xname + xlen, rem);
3525 static int mdd_declare_migrate_create(const struct lu_env *env,
3526 struct mdd_object *mdd_pobj,
3527 struct mdd_object *mdd_sobj,
3528 struct mdd_object *mdd_tobj,
3529 struct md_op_spec *spec,
3531 union lmv_mds_md *mgr_ea,
3532 struct linkea_data *ldata,
3533 struct thandle *handle)
3535 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3536 const struct lu_buf *buf;
3540 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3541 handle, spec, NULL);
3545 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3550 if (S_ISLNK(la->la_mode)) {
3551 const char *target_name = spec->u.sp_symname;
3552 int sym_len = strlen(target_name);
3553 const struct lu_buf *buf;
3555 buf = mdd_buf_get_const(env, target_name, sym_len);
3556 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3560 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3561 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3566 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3567 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3568 spec->u.sp_ea.eadatalen);
3569 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3575 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3576 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3577 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3582 la_flag->la_valid = LA_FLAGS;
3583 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3584 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3589 static int mdd_migrate_create(const struct lu_env *env,
3590 struct mdd_object *mdd_pobj,
3591 struct mdd_object *mdd_sobj,
3592 struct mdd_object *mdd_tobj,
3593 const struct lu_name *lname,
3596 struct mdd_thread_info *info = mdd_env_info(env);
3597 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3598 struct md_op_spec *spec = &info->mti_spec;
3599 struct lu_buf lmm_buf = { NULL };
3600 struct lu_buf link_buf = { NULL };
3601 struct lu_buf mgr_buf;
3602 struct thandle *handle;
3603 struct lmv_mds_md_v1 *mgr_ea;
3604 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3605 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3607 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3611 /* prepare spec for create */
3612 memset(spec, 0, sizeof(*spec));
3613 spec->sp_cr_lookup = 0;
3614 spec->sp_feat = &dt_directory_features;
3615 if (S_ISLNK(la->la_mode)) {
3616 const struct lu_buf *buf;
3618 buf = lu_buf_check_and_alloc(
3619 &mdd_env_info(env)->mti_big_buf,
3622 link_buf.lb_len = la->la_size + 1;
3623 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3624 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3626 rc = rc != 0 ? rc : -EFAULT;
3627 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3628 mdd2obd_dev(mdd)->obd_name,
3629 PFID(mdd_object_fid(mdd_sobj)), rc);
3632 spec->u.sp_symname = link_buf.lb_buf;
3633 } else if (S_ISREG(la->la_mode)) {
3634 /* retrieve lov of the old object */
3635 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3636 if (rc != 0 && rc != -ENODATA)
3638 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3639 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3640 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3641 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3643 } else if (S_ISDIR(la->la_mode)) {
3644 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3645 if (rc == -ENODATA) {
3646 /* ignore the non-linkEA error */
3654 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3655 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3656 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3657 mgr_buf.lb_len = mgr_easize;
3658 mgr_ea = mgr_buf.lb_buf;
3659 memset(mgr_ea, 0, sizeof(*mgr_ea));
3660 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3661 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3662 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3663 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3664 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3665 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3667 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3669 handle = mdd_trans_create(env, mdd);
3671 GOTO(out_free, rc = PTR_ERR(handle));
3673 /* Note: this transaction is part of migration, and it is not
3674 * the last step of migration, so we set th_local = 1 to avoid
3675 * update last rcvd for this transaction */
3676 handle->th_local = 1;
3677 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3678 la, mgr_buf.lb_buf, ldata, handle);
3680 GOTO(stop_trans, rc);
3682 rc = mdd_trans_start(env, mdd, handle);
3684 GOTO(stop_trans, rc);
3686 /* don't set nlink from the original object */
3687 la->la_valid &= ~LA_NLINK;
3689 /* create the target object */
3690 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3693 GOTO(stop_trans, rc);
3695 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3696 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3698 GOTO(stop_trans, rc);
3701 /* Set MIGRATE EA on the source inode, so once the migration needs
3702 * to be re-done during failover, the re-do process can locate the
3703 * target object which is already being created. */
3704 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3706 GOTO(stop_trans, rc);
3708 /* Set immutable flag, so any modification is disabled until
3709 * the migration is done. Once the migration is interrupted,
3710 * if the resume process find the migrating object has both
3711 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3712 * flag and approve the migration */
3713 la_flag->la_valid = LA_FLAGS;
3714 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3715 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3718 rc = mdd_trans_stop(env, mdd, rc, handle);
3720 if (lmm_buf.lb_buf != NULL)
3721 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3725 static int mdd_migrate_entries(const struct lu_env *env,
3726 struct mdd_object *mdd_sobj,
3727 struct mdd_object *mdd_tobj)
3729 struct dt_object *next = mdd_object_child(mdd_sobj);
3730 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3731 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3732 struct thandle *handle;
3734 const struct dt_it_ops *iops;
3736 struct lu_dirent *ent;
3740 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3744 if (!dt_try_as_dir(env, next))
3745 GOTO(out_ent, rc = -ENOTDIR);
3747 * iterate directories
3749 iops = &next->do_index_ops->dio_it;
3750 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3752 GOTO(out_ent, rc = PTR_ERR(it));
3754 rc = iops->load(env, it, 0);
3756 rc = iops->next(env, it);
3760 * At this point and across for-loop:
3762 * rc == 0 -> ok, proceed.
3763 * rc > 0 -> end of directory.
3767 struct mdd_object *child;
3768 char *name = mdd_env_info(env)->mti_key;
3771 bool target_exist = false;
3773 len = iops->key_size(env, it);
3777 result = iops->rec(env, it, (struct dt_rec *)ent,
3778 LUDA_FID | LUDA_TYPE);
3779 if (result == -ESTALE)
3786 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3788 /* Insert new fid with target name into target dir */
3789 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3790 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3791 ent->lde_name[1] == '.'))
3794 child = mdd_object_find(env, mdd, &ent->lde_fid);
3796 GOTO(out, rc = PTR_ERR(child));
3798 /* child may not exist, but lu_object_attr will assert this,
3799 * get type from loh_attr directly */
3800 is_dir = S_ISDIR(child->mod_obj.mo_lu.lo_header->loh_attr);
3802 mdd_write_lock(env, child, MOR_SRC_CHILD);
3804 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3806 /* Check whether the name has been inserted to the target */
3807 if (dt_try_as_dir(env, dt_tobj)) {
3808 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3810 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3811 (struct dt_key *)name);
3812 if (unlikely(rc == 0))
3813 target_exist = true;
3816 handle = mdd_trans_create(env, mdd);
3818 GOTO(out_put, rc = PTR_ERR(handle));
3820 /* Note: this transaction is part of migration, and it is not
3821 * the last step of migration, so we set th_local = 1 to avoid
3822 * updating last rcvd for this transaction */
3823 handle->th_local = 1;
3824 if (likely(!target_exist)) {
3825 rc = mdo_declare_index_insert(env, mdd_tobj,
3827 child->mod_obj.mo_lu.lo_header->loh_attr,
3833 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3839 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3844 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3848 /* Update .. for child */
3849 rc = mdo_declare_index_delete(env, child, dotdot,
3854 rc = mdo_declare_index_insert(env, child,
3855 mdd_object_fid(mdd_tobj),
3856 S_IFDIR, dotdot, handle);
3861 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3868 rc = mdd_trans_start(env, mdd, handle);
3870 CERROR("%s: transaction start failed: rc = %d\n",
3871 mdd2obd_dev(mdd)->obd_name, rc);
3875 if (likely(!target_exist)) {
3876 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3877 child->mod_obj.mo_lu.lo_header->loh_attr, name,
3883 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3888 rc = __mdd_index_delete_only(env, child, dotdot,
3893 rc = __mdd_index_insert_only(env, child,
3894 mdd_object_fid(mdd_tobj), S_IFDIR,
3900 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3902 strlen(name), handle);
3905 mdd_write_unlock(env, child);
3906 mdd_object_put(env, child);
3907 rc = mdd_trans_stop(env, mdd, rc, handle);
3911 result = iops->next(env, it);
3912 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
3913 GOTO(out, rc = -EINTR);
3915 if (result == -ESTALE)
3917 } while (result == 0);
3920 iops->fini(env, it);
3922 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
3926 static int mdd_declare_update_linkea(const struct lu_env *env,
3927 struct mdd_object *mdd_pobj,
3928 struct mdd_object *mdd_sobj,
3929 struct mdd_object *mdd_tobj,
3930 const struct lu_name *child_name,
3931 struct linkea_data *ldata,
3932 struct thandle *handle)
3934 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3935 child_name, ldata, handle, 1);
3938 static int mdd_update_linkea(const struct lu_env *env,
3939 struct mdd_object *mdd_pobj,
3940 struct mdd_object *mdd_sobj,
3941 struct mdd_object *mdd_tobj,
3942 const struct lu_name *child_name,
3943 struct linkea_data *ldata,
3944 struct thandle *handle)
3946 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
3947 child_name, ldata, handle, 0);
3950 static int mdd_declare_migrate_update_name(const struct lu_env *env,
3951 struct mdd_object *mdd_pobj,
3952 struct mdd_object *mdd_sobj,
3953 struct mdd_object *mdd_tobj,
3954 const struct lu_name *lname,
3956 struct lu_attr *parent_la,
3957 struct linkea_data *ldata,
3958 struct thandle *handle)
3960 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3961 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
3964 /* Revert IMMUTABLE flag */
3965 la_flag->la_valid = LA_FLAGS;
3966 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
3967 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3971 /* delete entry from source dir */
3972 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
3976 if (ldata->ld_buf != NULL) {
3977 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
3978 mdd_tobj, lname, ldata, handle);
3983 if (S_ISREG(mdd_object_type(mdd_sobj))) {
3984 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
3989 handle->th_complex = 1;
3990 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
3992 LU_XATTR_REPLACE, handle);
3997 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3998 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
4004 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
4005 mdd_object_type(mdd_tobj),
4006 lname->ln_name, handle);
4010 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
4014 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4015 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
4020 /* delete old object */
4021 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4025 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4026 /* delete old object */
4027 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4030 /* set nlink to 0 */
4031 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
4036 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
4040 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
4044 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4050 static int mdd_migrate_update_name(const struct lu_env *env,
4051 struct mdd_object *mdd_pobj,
4052 struct mdd_object *mdd_sobj,
4053 struct mdd_object *mdd_tobj,
4054 const struct lu_name *lname,
4057 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
4058 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4059 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4060 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4061 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
4062 struct thandle *handle;
4063 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
4064 const char *name = lname->ln_name;
4068 /* update time for parent */
4069 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
4070 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
4071 p_la->la_valid = LA_CTIME;
4073 rc = mdd_la_get(env, mdd_sobj, so_attr);
4077 ldata->ld_buf = NULL;
4078 rc = mdd_links_read(env, mdd_sobj, ldata);
4079 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4082 handle = mdd_trans_create(env, mdd);
4084 RETURN(PTR_ERR(handle));
4086 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4087 lname, so_attr, p_la, ldata,
4090 /* If the migration can not be fit in one transaction, just
4091 * leave it in the original MDT */
4093 GOTO(stop_trans, rc = 0);
4095 GOTO(stop_trans, rc);
4098 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4099 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4100 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4101 PFID(mdd_object_fid(mdd_tobj)));
4103 rc = mdd_trans_start(env, mdd, handle);
4105 GOTO(stop_trans, rc);
4107 /* Revert IMMUTABLE flag */
4108 la_flag->la_valid = LA_FLAGS;
4109 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4110 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4112 GOTO(stop_trans, rc);
4114 /* Remove source name from source directory */
4115 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4117 GOTO(stop_trans, rc);
4119 if (ldata->ld_buf != NULL) {
4120 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4121 lname, ldata, handle);
4123 GOTO(stop_trans, rc);
4125 /* linkea update might decrease the source object
4126 * nlink, let's get the attr again after ref_del */
4127 rc = mdd_la_get(env, mdd_sobj, so_attr);
4129 GOTO(stop_trans, rc);
4132 if (S_ISREG(so_attr->la_mode)) {
4133 if (so_attr->la_nlink == 1) {
4134 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4136 if (rc != 0 && rc != -ENODATA)
4137 GOTO(stop_trans, rc);
4139 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4141 LU_XATTR_REPLACE, handle);
4143 GOTO(stop_trans, rc);
4147 /* Insert new fid with target name into target dir */
4148 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4149 mdd_object_type(mdd_tobj), name, handle);
4151 GOTO(stop_trans, rc);
4153 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4155 mdd_sobj->mod_flags |= DEAD_OBJ;
4156 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4158 GOTO(out_unlock, rc);
4160 rc = mdd_orphan_insert(env, mdd_sobj, handle);
4162 GOTO(out_unlock, rc);
4164 mdo_ref_del(env, mdd_sobj, handle);
4166 mdo_ref_del(env, mdd_sobj, handle);
4168 /* Get the attr again after ref_del */
4169 rc = mdd_la_get(env, mdd_sobj, so_attr);
4171 GOTO(out_unlock, rc);
4173 ma->ma_attr = *so_attr;
4174 ma->ma_valid |= MA_INODE;
4176 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4178 GOTO(out_unlock, rc);
4180 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4181 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4182 mdo2fid(mdd_pobj), lname, lname, handle);
4184 CWARN("%s: changelog for migrate %s "DFID
4185 "under "DFID" failed: rc = %d\n",
4186 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4187 PFID(mdd_object_fid(mdd_sobj)),
4188 PFID(mdd_object_fid(mdd_pobj)), rc);
4189 /* Sigh, there are no easy way to migrate back the object, so
4190 * let's reset the result to 0 for now XXX */
4194 mdd_write_unlock(env, mdd_sobj);
4197 rc = mdd_trans_stop(env, mdd, rc, handle);
4202 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4203 const struct lu_fid *fid, __u32 *mdt_index)
4205 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4206 struct seq_server_site *ss;
4209 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4211 range->lsr_flags = LU_SEQ_RANGE_MDT;
4212 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4216 *mdt_index = range->lsr_index;
4221 * Check whether we should migrate the file/dir
4223 * < 0 permission check failed or other error.
4224 * = 0 the file can be migrated.
4225 * > 0 the file does not need to be migrated, mostly
4226 * for multiple link file
4228 static int mdd_migrate_sanity_check(const struct lu_env *env,
4229 struct mdd_object *pobj,
4230 const struct lu_attr *pattr,
4231 struct mdd_object *sobj,
4232 struct lu_attr *sattr)
4234 struct mdd_thread_info *info = mdd_env_info(env);
4235 struct linkea_data *ldata = &info->mti_link_data;
4236 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4238 struct lu_buf *mgr_buf;
4244 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4245 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4246 if (mgr_buf->lb_buf == NULL)
4249 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4251 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4253 /* If the object has migrateEA, it means IMMUTE flag
4254 * is being set by previous migration process, so it
4255 * needs to override the IMMUTE flag, otherwise the
4256 * following sanity check will fail */
4257 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4258 LMV_HASH_FLAG_MIGRATION) {
4259 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4261 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4262 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4263 mdd2obd_dev(mdd)->obd_name,
4264 PFID(mdd_object_fid(sobj)));
4268 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4269 sobj, sattr, NULL, NULL);
4273 /* Then it will check if the file should be migrated. If the file
4274 * has mulitple links, we only need migrate the file if all of its
4275 * entries has been migrated to the remote MDT */
4276 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4279 rc = mdd_links_read(env, sobj, ldata);
4281 /* For multiple links files, if there are no linkEA data at all,
4282 * means the file might be created before linkEA is enabled, and
4283 * all of its links should not be migrated yet, otherwise it
4284 * should have some linkEA there */
4285 if (rc == -ENOENT || rc == -ENODATA)
4290 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4291 /* If there are still links locally, then the file will not be
4293 LASSERT(ldata->ld_leh != NULL);
4295 /* If the linkEA is overflow, then means there are some unknown name
4296 * entries under unknown parents, that will prevent the migration. */
4297 if (unlikely(ldata->ld_leh->leh_overflow_time))
4300 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4301 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4302 struct lu_name lname;
4304 __u32 parent_mdt_index;
4306 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4308 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4311 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4315 /* Migrate the object only if none of its parents are on the
4317 if (parent_mdt_index != mdt_index)
4320 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4321 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4322 lname.ln_name, PFID(&fid));
4330 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4331 struct md_object *sobj, const struct lu_name *lname,
4332 struct md_object *tobj, struct md_attr *ma)
4334 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4335 struct mdd_device *mdd = mdo2mdd(pobj);
4336 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4337 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4338 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4339 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4340 bool created = false;
4344 /* If the file will being migrated, it will check whether
4345 * the file is being opened by someone else right now */
4346 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4347 if (mdd_sobj->mod_count > 0) {
4349 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4350 mdd2obd_dev(mdd)->obd_name,
4351 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4352 mdd_sobj->mod_count, -EBUSY);
4353 mdd_read_unlock(env, mdd_sobj);
4354 GOTO(put, rc = -EBUSY);
4356 mdd_read_unlock(env, mdd_sobj);
4358 rc = mdd_la_get(env, mdd_sobj, so_attr);
4362 rc = mdd_la_get(env, mdd_pobj, pattr);
4366 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4373 /* Sigh, it is impossible to finish all of migration in a single
4374 * transaction, for example migrating big directory entries to the
4375 * new MDT, it needs insert all of name entries of children in the
4378 * So migration will be done in multiple steps and transactions.
4380 * 1. create an orphan object on the remote MDT in one transaction.
4381 * 2. migrate extend attributes to the new target file/directory.
4382 * 3. For directory, migrate the entries to the new MDT and update
4383 * linkEA of each children. Because we can not migrate all entries
4384 * in a single transaction, so the migrating directory will become
4385 * a striped directory during migration, so once the process is
4386 * interrupted, the directory is still accessible. (During lookup,
4387 * client will locate the name by searching both original and target
4389 * 4. Finally, update the name/FID to point to the new file/directory
4390 * in a separate transaction.
4393 /* step 1: Check whether the orphan object has been created, and create
4394 * orphan object on the remote MDT if needed */
4395 if (!mdd_object_exists(mdd_tobj)) {
4396 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4403 LASSERT(mdd_object_exists(mdd_tobj));
4404 /* step 2: migrate xattr */
4405 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4409 /* step 3: migrate name entries to the orphan object */
4410 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4411 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4414 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4415 OBD_FAIL_MDS_REINT_NET_REP)))
4418 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4421 LASSERT(mdd_object_exists(mdd_tobj));
4422 /* step 4: update name entry to the new object */
4423 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4428 /* newly created target was not locked, don't cache its attributes */
4430 mdd_invalidate(env, tobj);
4435 const struct md_dir_operations mdd_dir_ops = {
4436 .mdo_is_subdir = mdd_is_subdir,
4437 .mdo_lookup = mdd_lookup,
4438 .mdo_create = mdd_create,
4439 .mdo_rename = mdd_rename,
4440 .mdo_link = mdd_link,
4441 .mdo_unlink = mdd_unlink,
4442 .mdo_create_data = mdd_create_data,
4443 .mdo_migrate = mdd_migrate,