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 <linux/kthread.h>
43 #include <obd_class.h>
44 #include <obd_support.h>
45 #include <lustre_mds.h>
46 #include <lustre_fid.h>
48 #include "mdd_internal.h"
50 static const char dot[] = ".";
51 static const char dotdot[] = "..";
53 static struct lu_name lname_dotdot = {
54 .ln_name = (char *) dotdot,
55 .ln_namelen = sizeof(dotdot) - 1,
59 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
61 if (!lu_name_is_valid(ln))
63 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
69 /* Get FID from name and parent */
71 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
72 const struct lu_attr *pattr, const struct lu_name *lname,
73 struct lu_fid* fid, int mask)
75 const char *name = lname->ln_name;
76 const struct dt_key *key = (const struct dt_key *)name;
77 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
78 struct mdd_device *m = mdo2mdd(pobj);
79 struct dt_object *dir = mdd_object_child(mdd_obj);
83 if (unlikely(mdd_is_dead_obj(mdd_obj)))
86 if (!mdd_object_exists(mdd_obj))
89 if (mdd_object_remote(mdd_obj)) {
90 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
91 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
94 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
99 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
100 dt_try_as_dir(env, dir)))
101 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
108 int mdd_lookup(const struct lu_env *env,
109 struct md_object *pobj, const struct lu_name *lname,
110 struct lu_fid *fid, struct md_op_spec *spec)
112 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
116 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
120 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
121 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
125 /** Read the link EA into a temp buffer.
126 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
127 * A pointer to the buffer is stored in \a ldata::ld_buf.
131 static int __mdd_links_read(const struct lu_env *env,
132 struct mdd_object *mdd_obj,
133 struct linkea_data *ldata)
137 if (!mdd_object_exists(mdd_obj))
140 /* First try a small buf */
141 LASSERT(env != NULL);
142 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
144 if (ldata->ld_buf->lb_buf == NULL)
147 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
149 /* Buf was too small, figure out what we need. */
150 lu_buf_free(ldata->ld_buf);
151 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
155 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
156 if (ldata->ld_buf->lb_buf == NULL)
158 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
162 lu_buf_free(ldata->ld_buf);
163 ldata->ld_buf = NULL;
167 return linkea_init(ldata);
170 static int mdd_links_read(const struct lu_env *env,
171 struct mdd_object *mdd_obj,
172 struct linkea_data *ldata)
176 rc = __mdd_links_read(env, mdd_obj, ldata);
178 rc = linkea_init(ldata);
183 static int mdd_links_read_with_rec(const struct lu_env *env,
184 struct mdd_object *mdd_obj,
185 struct linkea_data *ldata)
189 rc = __mdd_links_read(env, mdd_obj, ldata);
191 rc = linkea_init_with_rec(ldata);
197 * Get parent FID of the directory
199 * Read parent FID from linkEA, if that fails, then do lookup
200 * dotdot to get the parent FID.
202 * \param[in] env execution environment
203 * \param[in] obj object from which to find the parent FID
204 * \param[in] attr attribute of the object
205 * \param[out] fid fid to get the parent FID
207 * \retval 0 if getting the parent FID succeeds.
208 * \retval negative errno if getting the parent FID fails.
210 static inline int mdd_parent_fid(const struct lu_env *env,
211 struct mdd_object *obj,
212 const struct lu_attr *attr,
215 struct mdd_thread_info *info = mdd_env_info(env);
216 struct linkea_data ldata = { NULL };
217 struct lu_buf *buf = &info->mti_link_buf;
218 struct lu_name lname;
223 LASSERT(S_ISDIR(mdd_object_type(obj)));
225 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
226 if (buf->lb_buf == NULL)
227 GOTO(lookup, rc = 0);
230 rc = mdd_links_read_with_rec(env, obj, &ldata);
234 LASSERT(ldata.ld_leh != NULL);
235 /* Directory should only have 1 parent */
236 if (ldata.ld_leh->leh_reccount > 1)
239 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
241 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
242 if (likely(fid_is_sane(fid)))
245 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
250 * For root fid use special function, which does not compare version component
251 * of fid. Version component is different for root fids on all MDTs.
253 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
255 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
256 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
260 * return 1: if lf is the fid of the ancestor of p1;
263 * return -EREMOTE: if remote object is found, in this
264 * case fid of remote object is saved to @pf;
266 * otherwise: values < 0, errors.
268 static int mdd_is_parent(const struct lu_env *env,
269 struct mdd_device *mdd,
270 struct mdd_object *p1,
271 const struct lu_attr *attr,
272 const struct lu_fid *lf,
275 struct mdd_object *parent = NULL;
280 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
281 pfid = &mdd_env_info(env)->mti_fid;
283 /* Check for root first. */
284 if (mdd_is_root(mdd, mdo2fid(p1)))
288 /* this is done recursively */
289 rc = mdd_parent_fid(env, p1, attr, pfid);
292 if (mdd_is_root(mdd, pfid))
294 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
296 if (lu_fid_eq(pfid, lf))
299 mdd_object_put(env, parent);
301 parent = mdd_object_find(env, mdd, pfid);
303 GOTO(out, rc = PTR_ERR(parent));
305 if (!mdd_object_exists(parent))
306 GOTO(out, rc = -EINVAL);
312 if (parent && !IS_ERR(parent))
313 mdd_object_put(env, parent);
318 * No permission check is needed.
320 * returns 1: if fid is ancestor of @mo;
321 * returns 0: if fid is not an ancestor of @mo;
323 * returns EREMOTE if remote object is found, fid of remote object is saved to
326 * returns < 0: if error
328 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
329 const struct lu_fid *fid, struct lu_fid *sfid)
331 struct mdd_device *mdd = mdo2mdd(mo);
332 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
336 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
339 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
343 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
347 } else if (rc == 1) {
348 /* found @fid is parent */
356 * Check that @dir contains no entries except (possibly) dot and dotdot.
361 * -ENOTDIR not a directory object
362 * -ENOTEMPTY not empty
366 static int mdd_dir_is_empty(const struct lu_env *env,
367 struct mdd_object *dir)
370 struct dt_object *obj;
371 const struct dt_it_ops *iops;
375 obj = mdd_object_child(dir);
376 if (!dt_try_as_dir(env, obj))
379 iops = &obj->do_index_ops->dio_it;
380 it = iops->init(env, obj, LUDA_64BITHASH);
382 result = iops->get(env, it, (const struct dt_key *)"");
385 for (result = 0, i = 0; result == 0 && i < 3; ++i)
386 result = iops->next(env, it);
389 else if (result == 1)
391 } else if (result == 0)
393 * Huh? Index contains no zero key?
400 result = PTR_ERR(it);
405 * Determine if the target object can be hard linked, and right now it only
406 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
407 * directory nlink count might exceed the maximum link count(see
408 * osd_object_ref_add), so it only check nlink for non-directories.
410 * \param[in] env thread environment
411 * \param[in] obj object being linked to
412 * \param[in] la attributes of \a obj
414 * \retval 0 if \a obj can be hard linked
415 * \retval negative error if \a obj is a directory or has too
418 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
419 const struct lu_attr *la)
421 struct mdd_device *m = mdd_obj2mdd_dev(obj);
426 /* Subdir count limitation can be broken through
427 * (see osd_object_ref_add), so only check non-directory here. */
428 if (!S_ISDIR(la->la_mode) &&
429 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
436 * Check whether it may create the cobj under the pobj.
438 * \param[in] env execution environment
439 * \param[in] pobj the parent directory
440 * \param[in] pattr the attribute of the parent directory
441 * \param[in] cobj the child to be created
442 * \param[in] check_perm if check WRITE|EXEC permission for parent
444 * \retval = 0 create the child under this dir is allowed
445 * \retval negative errno create the child under this dir is
448 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
449 const struct lu_attr *pattr, struct mdd_object *cobj,
455 if (cobj && mdd_object_exists(cobj))
458 if (mdd_is_dead_obj(pobj))
462 rc = mdd_permission_internal_locked(env, pobj, pattr,
463 MAY_WRITE | MAY_EXEC,
469 * Check whether can unlink from the pobj in the case of "cobj == NULL".
471 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
472 const struct lu_attr *pattr, const struct lu_attr *attr)
477 if (mdd_is_dead_obj(pobj))
480 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
483 rc = mdd_permission_internal_locked(env, pobj, pattr,
484 MAY_WRITE | MAY_EXEC,
489 if (pattr->la_flags & LUSTRE_APPEND_FL)
496 * pobj == NULL is remote ops case, under such case, pobj's
497 * VTX feature has been checked already, no need check again.
499 static inline int mdd_is_sticky(const struct lu_env *env,
500 struct mdd_object *pobj,
501 const struct lu_attr *pattr,
502 struct mdd_object *cobj,
503 const struct lu_attr *cattr)
505 struct lu_ucred *uc = lu_ucred_assert(env);
508 LASSERT(pattr != NULL);
509 if (!(pattr->la_mode & S_ISVTX) ||
510 (pattr->la_uid == uc->uc_fsuid))
514 LASSERT(cattr != NULL);
515 if (cattr->la_uid == uc->uc_fsuid)
518 return !md_capable(uc, CFS_CAP_FOWNER);
521 static int mdd_may_delete_entry(const struct lu_env *env,
522 struct mdd_object *pobj,
523 const struct lu_attr *pattr,
528 LASSERT(pobj != NULL);
529 if (!mdd_object_exists(pobj))
532 if (mdd_is_dead_obj(pobj))
537 rc = mdd_permission_internal_locked(env, pobj, pattr,
538 MAY_WRITE | MAY_EXEC,
544 if (pattr->la_flags & LUSTRE_APPEND_FL)
551 * Check whether it may delete the cobj from the pobj.
554 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
555 const struct lu_attr *tpattr, struct mdd_object *tobj,
556 const struct lu_attr *tattr, const struct lu_attr *cattr,
557 int check_perm, int check_empty)
563 LASSERT(tpattr != NULL);
564 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
572 if (!mdd_object_exists(tobj))
575 if (mdd_is_dead_obj(tobj))
578 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
581 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
584 /* additional check the rename case */
586 if (S_ISDIR(cattr->la_mode)) {
587 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
589 if (!S_ISDIR(tattr->la_mode))
592 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
594 } else if (S_ISDIR(tattr->la_mode))
598 if (S_ISDIR(tattr->la_mode) && check_empty)
599 rc = mdd_dir_is_empty(env, tobj);
605 * Check whether it can create the link file(linked to @src_obj) under
606 * the target directory(@tgt_obj), and src_obj has been locked by
609 * \param[in] env execution environment
610 * \param[in] tgt_obj the target directory
611 * \param[in] tattr attributes of target directory
612 * \param[in] lname the link name
613 * \param[in] src_obj source object for link
614 * \param[in] cattr attributes for source object
616 * \retval = 0 it is allowed to create the link file under tgt_obj
617 * \retval negative error not allowed to create the link file
619 static int mdd_link_sanity_check(const struct lu_env *env,
620 struct mdd_object *tgt_obj,
621 const struct lu_attr *tattr,
622 const struct lu_name *lname,
623 struct mdd_object *src_obj,
624 const struct lu_attr *cattr)
626 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
630 if (!mdd_object_exists(src_obj))
633 if (mdd_is_dead_obj(src_obj))
636 /* Local ops, no lookup before link, check filename length here. */
637 rc = mdd_name_check(m, lname);
641 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
644 if (S_ISDIR(mdd_object_type(src_obj)))
647 LASSERT(src_obj != tgt_obj);
648 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
652 rc = __mdd_may_link(env, src_obj, cattr);
657 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
658 const char *name, struct thandle *handle)
660 struct dt_object *next = mdd_object_child(pobj);
664 if (dt_try_as_dir(env, next))
665 rc = dt_delete(env, next, (struct dt_key *)name, handle);
672 static int __mdd_index_insert_only(const struct lu_env *env,
673 struct mdd_object *pobj,
674 const struct lu_fid *lf, __u32 type,
676 struct thandle *handle)
678 struct dt_object *next = mdd_object_child(pobj);
682 if (dt_try_as_dir(env, next)) {
683 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
684 struct lu_ucred *uc = lu_ucred_check(env);
688 rec->rec_type = type;
689 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
690 rc = dt_insert(env, next, (const struct dt_rec *)rec,
691 (const struct dt_key *)name, handle,
699 /* insert named index, add reference if isdir */
700 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
701 const struct lu_fid *lf, __u32 type,
702 const char *name, struct thandle *handle)
707 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
708 if (rc == 0 && S_ISDIR(type)) {
709 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
710 mdo_ref_add(env, pobj, handle);
711 mdd_write_unlock(env, pobj);
717 /* delete named index, drop reference if isdir */
718 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
719 const char *name, int is_dir,
720 struct thandle *handle)
725 rc = __mdd_index_delete_only(env, pobj, name, handle);
726 if (rc == 0 && is_dir) {
727 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
728 mdo_ref_del(env, pobj, handle);
729 mdd_write_unlock(env, pobj);
735 static int mdd_llog_record_calc_size(const struct lu_env *env,
736 const struct lu_name *tname,
737 const struct lu_name *sname)
739 const struct lu_ucred *uc = lu_ucred(env);
740 enum changelog_rec_flags crf = CLF_EXTRA_FLAGS;
741 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
746 if (uc != NULL && uc->uc_jobid[0] != '\0')
749 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
750 changelog_rec_offset(crf, crfe) +
751 (tname != NULL ? tname->ln_namelen : 0) +
752 (sname != NULL ? 1 + sname->ln_namelen : 0));
755 int mdd_declare_changelog_store(const struct lu_env *env,
756 struct mdd_device *mdd,
757 enum changelog_rec_type type,
758 const struct lu_name *tname,
759 const struct lu_name *sname,
760 struct thandle *handle)
762 struct obd_device *obd = mdd2obd_dev(mdd);
763 struct llog_ctxt *ctxt;
764 struct llog_changelog_rec *rec;
766 struct thandle *llog_th;
770 if (!mdd_changelog_enabled(env, mdd, type))
773 reclen = mdd_llog_record_calc_size(env, tname, sname);
774 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
775 if (buf->lb_buf == NULL)
779 rec->cr_hdr.lrh_len = reclen;
780 rec->cr_hdr.lrh_type = CHANGELOG_REC;
782 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
786 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
788 GOTO(out_put, rc = PTR_ERR(llog_th));
790 rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th);
798 struct mdd_changelog_gc {
799 struct mdd_device *mcgc_mdd;
802 __u64 mcgc_maxindexes;
806 /* return first registered ChangeLog user idle since too long
807 * use ChangeLog's user plain LLOG mtime for this */
808 static int mdd_changelog_gc_cb(const struct lu_env *env,
809 struct llog_handle *llh,
810 struct llog_rec_hdr *hdr, void *data)
812 struct llog_changelog_user_rec *rec;
813 struct mdd_changelog_gc *mcgc = (struct mdd_changelog_gc *)data;
814 struct mdd_device *mdd = mcgc->mcgc_mdd;
817 if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
820 rec = container_of(hdr, struct llog_changelog_user_rec,
823 /* find oldest idle user, based on last record update/cancel time (new
824 * behavior), or for old user records, last record index vs current
825 * ChangeLog index. Late users with old record format will be treated
826 * first as we assume they could be idle since longer
828 if (rec->cur_time != 0) {
829 __u32 time_now = (__u32)get_seconds();
830 __u32 time_out = rec->cur_time +
831 mdd->mdd_changelog_max_idle_time;
832 __u32 idle_time = time_now - rec->cur_time;
834 /* treat oldest idle user first, and if no old format user
835 * has been already selected
837 if (time_after32(time_now, time_out) &&
838 idle_time > mcgc->mcgc_maxtime &&
839 mcgc->mcgc_maxindexes == 0) {
840 mcgc->mcgc_maxtime = idle_time;
841 mcgc->mcgc_id = rec->cur_id;
842 mcgc->mcgc_found = true;
845 /* old user record with no idle time stamp, so use empirical
846 * method based on its current index/position
850 idle_indexes = mdd->mdd_cl.mc_index - rec->cur_endrec;
852 /* treat user with the oldest/smallest current index first */
853 if (idle_indexes >= mdd->mdd_changelog_max_idle_indexes &&
854 idle_indexes > mcgc->mcgc_maxindexes) {
855 mcgc->mcgc_maxindexes = idle_indexes;
856 mcgc->mcgc_id = rec->cur_id;
857 mcgc->mcgc_found = true;
864 /* recover space from long-term inactive ChangeLog users */
865 static int mdd_chlg_garbage_collect(void *data)
867 struct mdd_device *mdd = (struct mdd_device *)data;
868 struct lu_env *env = NULL;
870 struct llog_ctxt *ctxt;
871 struct mdd_changelog_gc mcgc = {
875 .mcgc_maxindexes = 0,
879 CDEBUG(D_HA, "%s: ChangeLog garbage collect thread start\n",
880 mdd2obd_dev(mdd)->obd_name);
884 GOTO(out, rc = -ENOMEM);
886 rc = lu_env_init(env, LCT_MD_THREAD);
891 ctxt = llog_get_context(mdd2obd_dev(mdd),
892 LLOG_CHANGELOG_USER_ORIG_CTXT);
894 (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
895 GOTO(out_env, rc = -ENXIO);
897 rc = llog_cat_process(env, ctxt->loc_handle,
898 mdd_changelog_gc_cb, &mcgc, 0, 0);
899 if (rc != 0 || mcgc.mcgc_found == false)
903 CWARN("%s: Force deregister of ChangeLog user cl%d idle more "
904 "than %us\n", mdd2obd_dev(mdd)->obd_name, mcgc.mcgc_id,
907 mdd_changelog_user_purge(env, mdd, mcgc.mcgc_id);
909 /* try again to search for another candidate */
910 mcgc.mcgc_found = false;
911 mcgc.mcgc_maxtime = 0;
912 mcgc.mcgc_maxindexes = 0;
924 mdd->mdd_cl.mc_gc_task = NULL;
928 /** Add a changelog entry \a rec to the changelog llog
931 * \param handle - currently ignored since llogs start their own transaction;
932 * this will hopefully be fixed in llog rewrite
935 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
936 struct llog_changelog_rec *rec, struct thandle *th)
938 struct obd_device *obd = mdd2obd_dev(mdd);
939 struct llog_ctxt *ctxt;
940 struct thandle *llog_th;
942 bool run_gc_task = false;
944 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
945 changelog_rec_varsize(&rec->cr));
947 /* llog_lvfs_write_rec sets the llog tail len */
948 rec->cr_hdr.lrh_type = CHANGELOG_REC;
949 rec->cr.cr_time = cl_time();
951 spin_lock(&mdd->mdd_cl.mc_lock);
952 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
953 * but as long as the MDD transactions are ordered correctly for e.g.
954 * rename conflicts, I don't think this should matter. */
955 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
956 spin_unlock(&mdd->mdd_cl.mc_lock);
958 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
962 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
964 GOTO(out_put, rc = PTR_ERR(llog_th));
966 /* nested journal transaction */
967 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
969 /* time to recover some space ?? */
970 spin_lock(&mdd->mdd_cl.mc_lock);
971 if (unlikely(mdd->mdd_changelog_gc && (ktime_get_real_seconds() -
972 mdd->mdd_cl.mc_gc_time > mdd->mdd_changelog_min_gc_interval) &&
973 mdd->mdd_cl.mc_gc_task == NULL &&
974 llog_cat_free_space(ctxt->loc_handle) <=
975 mdd->mdd_changelog_min_free_cat_entries)) {
976 CWARN("%s: low on changelog_catalog free entries, starting "
977 "ChangeLog garbage collection thread\n", obd->obd_name);
979 /* indicate further kthread run will occur outside right after
982 mdd->mdd_cl.mc_gc_task = (struct task_struct *)(-1);
985 spin_unlock(&mdd->mdd_cl.mc_lock);
987 struct task_struct *gc_task;
989 gc_task = kthread_run(mdd_chlg_garbage_collect, mdd,
991 if (IS_ERR(gc_task)) {
992 CERROR("%s: cannot start ChangeLog garbage collection "
993 "thread: rc = %ld\n", obd->obd_name,
995 mdd->mdd_cl.mc_gc_task = NULL;
997 CDEBUG(D_HA, "%s: ChangeLog garbage collection thread "
998 "has started with Pid %d\n", obd->obd_name,
1000 mdd->mdd_cl.mc_gc_task = gc_task;
1001 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
1005 llog_ctxt_put(ctxt);
1011 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
1012 const struct lu_fid *sfid,
1013 const struct lu_fid *spfid,
1014 const struct lu_name *sname)
1016 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
1017 size_t extsize = sname->ln_namelen + 1;
1019 LASSERT(sfid != NULL);
1020 LASSERT(spfid != NULL);
1021 LASSERT(sname != NULL);
1023 rnm->cr_sfid = *sfid;
1024 rnm->cr_spfid = *spfid;
1026 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
1027 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
1028 rec->cr_namelen += extsize;
1031 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
1033 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
1035 if (jobid == NULL || jobid[0] == '\0')
1038 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
1041 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
1043 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
1045 ef->cr_extra_flags = eflags;
1048 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
1049 __u64 uid, __u64 gid)
1051 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
1053 uidgid->cr_uid = uid;
1054 uidgid->cr_gid = gid;
1057 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
1060 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
1062 clnid->cr_nid = nid;
1065 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, int flags)
1067 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
1069 omd->cr_openflags = (__u32)flags;
1072 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
1073 const char *xattr_name)
1075 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
1077 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
1080 /** Store a namespace change changelog record
1081 * If this fails, we must fail the whole transaction; we don't
1082 * want the change to commit without the log entry.
1083 * \param target - mdd_object of change
1084 * \param tpfid - target parent dir/object fid
1085 * \param sfid - source object fid
1086 * \param spfid - source parent fid
1087 * \param tname - target name string
1088 * \param sname - source name string
1089 * \param handle - transaction handle
1091 int mdd_changelog_ns_store(const struct lu_env *env,
1092 struct mdd_device *mdd,
1093 enum changelog_rec_type type,
1094 enum changelog_rec_flags crf,
1095 struct mdd_object *target,
1096 const struct lu_fid *tpfid,
1097 const struct lu_fid *sfid,
1098 const struct lu_fid *spfid,
1099 const struct lu_name *tname,
1100 const struct lu_name *sname,
1101 struct thandle *handle)
1103 const struct lu_ucred *uc = lu_ucred(env);
1104 struct llog_changelog_rec *rec;
1107 __u64 xflags = CLFE_INVALID;
1111 if (!mdd_changelog_enabled(env, mdd, type))
1114 LASSERT(tpfid != NULL);
1115 LASSERT(tname != NULL);
1116 LASSERT(handle != NULL);
1118 reclen = mdd_llog_record_calc_size(env, tname, sname);
1119 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1120 if (buf->lb_buf == NULL)
1124 crf &= CLF_FLAGMASK;
1125 crf |= CLF_EXTRA_FLAGS;
1128 if (uc->uc_jobid[0] != '\0')
1130 xflags |= CLFE_UIDGID;
1139 rec->cr.cr_flags = crf;
1141 if (crf & CLF_EXTRA_FLAGS) {
1142 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1143 if (xflags & CLFE_UIDGID)
1144 mdd_changelog_rec_extra_uidgid(&rec->cr,
1145 uc->uc_uid, uc->uc_gid);
1146 if (xflags & CLFE_NID)
1147 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1150 rec->cr.cr_type = (__u32)type;
1151 rec->cr.cr_pfid = *tpfid;
1152 rec->cr.cr_namelen = tname->ln_namelen;
1153 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1155 if (crf & CLF_RENAME)
1156 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1158 if (crf & CLF_JOBID)
1159 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1161 if (likely(target != NULL)) {
1162 rec->cr.cr_tfid = *mdo2fid(target);
1163 target->mod_cltime = ktime_get();
1165 fid_zero(&rec->cr.cr_tfid);
1168 rc = mdd_changelog_store(env, mdd, rec, handle);
1170 CERROR("%s: cannot store changelog record: type = %d, "
1171 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1172 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1173 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1180 static int __mdd_links_add(const struct lu_env *env,
1181 struct mdd_object *mdd_obj,
1182 struct linkea_data *ldata,
1183 const struct lu_name *lname,
1184 const struct lu_fid *pfid,
1185 int first, int check)
1189 if (ldata->ld_leh == NULL) {
1190 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1194 rc = linkea_data_new(ldata,
1195 &mdd_env_info(env)->mti_link_buf);
1202 rc = linkea_links_find(ldata, lname, pfid);
1203 if (rc && rc != -ENOENT)
1209 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1210 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1214 linkea_add_buf(ldata, lname, tfid);
1217 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1218 linkea_add_buf(ldata, lname, pfid);
1220 return linkea_add_buf(ldata, lname, pfid);
1223 static int __mdd_links_del(const struct lu_env *env,
1224 struct mdd_object *mdd_obj,
1225 struct linkea_data *ldata,
1226 const struct lu_name *lname,
1227 const struct lu_fid *pfid)
1231 if (ldata->ld_leh == NULL) {
1232 rc = mdd_links_read(env, mdd_obj, ldata);
1237 rc = linkea_links_find(ldata, lname, pfid);
1241 linkea_del_buf(ldata, lname);
1245 static int mdd_linkea_prepare(const struct lu_env *env,
1246 struct mdd_object *mdd_obj,
1247 const struct lu_fid *oldpfid,
1248 const struct lu_name *oldlname,
1249 const struct lu_fid *newpfid,
1250 const struct lu_name *newlname,
1251 int first, int check,
1252 struct linkea_data *ldata)
1257 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1260 LASSERT(oldpfid != NULL || newpfid != NULL);
1262 if (mdd_obj->mod_flags & DEAD_OBJ)
1263 /* Unnecessary to update linkEA for dead object. */
1266 if (oldpfid != NULL) {
1267 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1269 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1272 /* No changes done. */
1277 /* If renaming, add the new record */
1278 if (newpfid != NULL)
1279 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1285 int mdd_links_rename(const struct lu_env *env,
1286 struct mdd_object *mdd_obj,
1287 const struct lu_fid *oldpfid,
1288 const struct lu_name *oldlname,
1289 const struct lu_fid *newpfid,
1290 const struct lu_name *newlname,
1291 struct thandle *handle,
1292 struct linkea_data *ldata,
1293 int first, int check)
1298 if (ldata == NULL) {
1299 ldata = &mdd_env_info(env)->mti_link_data;
1300 memset(ldata, 0, sizeof(*ldata));
1301 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1302 newpfid, newlname, first, check, ldata);
1307 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1308 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1314 if (newlname == NULL)
1315 CERROR("link_ea add failed %d "DFID"\n",
1316 rc, PFID(mdd_object_fid(mdd_obj)));
1317 else if (oldpfid == NULL)
1318 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1319 newlname->ln_namelen, newlname->ln_name, rc,
1320 PFID(mdd_object_fid(mdd_obj)));
1321 else if (newpfid == NULL)
1322 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1323 oldlname->ln_namelen, oldlname->ln_name, rc,
1324 PFID(mdd_object_fid(mdd_obj)));
1326 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1327 "\n", oldlname->ln_namelen, oldlname->ln_name,
1328 newlname->ln_namelen, newlname->ln_name, rc,
1329 PFID(mdd_object_fid(mdd_obj)));
1332 if (is_vmalloc_addr(ldata->ld_buf))
1333 /* if we vmalloced a large buffer drop it */
1334 lu_buf_free(ldata->ld_buf);
1339 static inline int mdd_links_add(const struct lu_env *env,
1340 struct mdd_object *mdd_obj,
1341 const struct lu_fid *pfid,
1342 const struct lu_name *lname,
1343 struct thandle *handle,
1344 struct linkea_data *ldata, int first)
1346 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1347 pfid, lname, handle, ldata, first, 0);
1350 static inline int mdd_links_del(const struct lu_env *env,
1351 struct mdd_object *mdd_obj,
1352 const struct lu_fid *pfid,
1353 const struct lu_name *lname,
1354 struct thandle *handle)
1356 return mdd_links_rename(env, mdd_obj, pfid, lname,
1357 NULL, NULL, handle, NULL, 0, 0);
1360 /** Read the link EA into a temp buffer.
1361 * Uses the name_buf since it is generally large.
1362 * \retval IS_ERR err
1363 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1365 struct lu_buf *mdd_links_get(const struct lu_env *env,
1366 struct mdd_object *mdd_obj)
1368 struct linkea_data ldata = { NULL };
1371 rc = mdd_links_read(env, mdd_obj, &ldata);
1372 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1375 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1376 struct linkea_data *ldata, struct thandle *handle)
1378 const struct lu_buf *buf;
1381 if (ldata == NULL || ldata->ld_buf == NULL ||
1382 ldata->ld_leh == NULL)
1385 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1389 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1390 ldata->ld_leh->leh_len);
1391 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1392 if (unlikely(rc == -ENOSPC)) {
1393 rc = linkea_overflow_shrink(ldata);
1401 static int mdd_declare_links_add(const struct lu_env *env,
1402 struct mdd_object *mdd_obj,
1403 struct thandle *handle,
1404 struct linkea_data *ldata)
1410 if (ldata != NULL && ldata->ld_leh != NULL) {
1411 ea_len = ldata->ld_leh->leh_len;
1412 linkea = ldata->ld_buf->lb_buf;
1414 ea_len = MAX_LINKEA_SIZE;
1418 rc = mdo_declare_xattr_set(env, mdd_obj,
1419 mdd_buf_get_const(env, linkea, ea_len),
1420 XATTR_NAME_LINK, 0, handle);
1425 static inline int mdd_declare_links_del(const struct lu_env *env,
1426 struct mdd_object *c,
1427 struct thandle *handle)
1431 /* For directory, the linkEA will be removed together
1432 * with the object. */
1433 if (!S_ISDIR(mdd_object_type(c)))
1434 rc = mdd_declare_links_add(env, c, handle, NULL);
1439 static int mdd_declare_link(const struct lu_env *env,
1440 struct mdd_device *mdd,
1441 struct mdd_object *p,
1442 struct mdd_object *c,
1443 const struct lu_name *name,
1444 struct thandle *handle,
1446 struct linkea_data *data)
1448 struct lu_fid tfid = *mdo2fid(c);
1451 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1452 tfid.f_oid = cfs_fail_val;
1454 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1455 name->ln_name, handle);
1459 rc = mdo_declare_ref_add(env, c, handle);
1463 la->la_valid = LA_CTIME | LA_MTIME;
1464 rc = mdo_declare_attr_set(env, p, la, handle);
1468 la->la_valid = LA_CTIME;
1469 rc = mdo_declare_attr_set(env, c, la, handle);
1473 rc = mdd_declare_links_add(env, c, handle, data);
1477 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1483 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1484 struct md_object *src_obj, const struct lu_name *lname,
1487 const char *name = lname->ln_name;
1488 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1489 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1490 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1491 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1492 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1493 struct mdd_device *mdd = mdo2mdd(src_obj);
1494 struct thandle *handle;
1495 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1496 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1500 rc = mdd_la_get(env, mdd_sobj, cattr);
1504 rc = mdd_la_get(env, mdd_tobj, tattr);
1509 * If we are using project inheritance, we only allow hard link
1510 * creation in our tree when the project IDs are the same;
1511 * otherwise the tree quota mechanism could be circumvented.
1513 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1514 (tattr->la_projid != cattr->la_projid))
1517 handle = mdd_trans_create(env, mdd);
1519 GOTO(out_pending, rc = PTR_ERR(handle));
1521 memset(ldata, 0, sizeof(*ldata));
1523 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1524 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1526 /* Note: even this function will change ldata, but it comes from
1527 * thread_info, which is completely temporary and only seen in
1528 * this function, so we do not need reset ldata once it fails.*/
1529 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1530 lname, 0, 0, ldata);
1534 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1539 rc = mdd_trans_start(env, mdd, handle);
1543 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1544 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1547 GOTO(out_unlock, rc);
1549 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1550 rc = mdo_ref_add(env, mdd_sobj, handle);
1552 GOTO(out_unlock, rc);
1555 *tfid = *mdo2fid(mdd_sobj);
1556 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1557 tfid->f_oid = cfs_fail_val;
1559 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1560 mdd_object_type(mdd_sobj), name, handle);
1562 mdo_ref_del(env, mdd_sobj, handle);
1563 GOTO(out_unlock, rc);
1566 la->la_valid = LA_CTIME | LA_MTIME;
1567 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1569 GOTO(out_unlock, rc);
1571 la->la_valid = LA_CTIME;
1572 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1574 /* Note: The failure of links_add should not cause the
1575 * link failure, so do not check return value. */
1576 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1577 lname, handle, ldata, 0);
1581 mdd_write_unlock(env, mdd_sobj);
1583 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1584 mdo2fid(mdd_tobj), NULL, NULL,
1585 lname, NULL, handle);
1587 rc = mdd_trans_stop(env, mdd, rc, handle);
1588 if (is_vmalloc_addr(ldata->ld_buf))
1589 /* if we vmalloced a large buffer drop it */
1590 lu_buf_free(ldata->ld_buf);
1595 static int mdd_mark_orphan_object(const struct lu_env *env,
1596 struct mdd_object *obj, struct thandle *handle,
1599 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1602 if (!S_ISDIR(mdd_object_type(obj)))
1605 attr->la_valid = LA_FLAGS;
1606 attr->la_flags = LUSTRE_ORPHAN_FL;
1609 rc = mdo_declare_attr_set(env, obj, attr, handle);
1611 rc = mdo_attr_set(env, obj, attr, handle);
1616 static int mdd_declare_finish_unlink(const struct lu_env *env,
1617 struct mdd_object *obj,
1618 struct thandle *handle)
1622 /* Sigh, we do not know if the unlink object will become orphan in
1623 * declare phase, but fortunately the flags here does not matter
1624 * in current declare implementation */
1625 rc = mdd_mark_orphan_object(env, obj, handle, true);
1629 rc = mdo_declare_destroy(env, obj, handle);
1633 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1637 return mdd_declare_links_del(env, obj, handle);
1640 /* caller should take a lock before calling */
1641 int mdd_finish_unlink(const struct lu_env *env,
1642 struct mdd_object *obj, struct md_attr *ma,
1643 const struct mdd_object *pobj,
1644 const struct lu_name *lname,
1648 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1651 LASSERT(mdd_write_locked(env, obj) != 0);
1653 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1654 /* add new orphan and the object
1655 * will be deleted during mdd_close() */
1656 obj->mod_flags |= DEAD_OBJ;
1657 if (obj->mod_count) {
1658 rc = mdd_orphan_insert(env, obj, th);
1660 CDEBUG(D_HA, "Object "DFID" is inserted into "
1661 "orphan list, open count = %d\n",
1662 PFID(mdd_object_fid(obj)),
1665 CERROR("Object "DFID" fail to be an orphan, "
1666 "open count = %d, maybe cause failed "
1668 PFID(mdd_object_fid(obj)),
1671 /* mark object as an orphan here, not
1672 * before mdd_orphan_insert() as racing
1673 * mdd_la_get() may propagate ORPHAN_OBJ
1674 * causing the asserition */
1675 rc = mdd_mark_orphan_object(env, obj, th, false);
1677 rc = mdo_destroy(env, obj, th);
1679 } else if (!is_dir) {
1680 /* old files may not have link ea; ignore errors */
1681 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1689 * has mdd_write_lock on cobj already, but not on pobj yet
1691 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1692 const struct lu_attr *pattr,
1693 struct mdd_object *cobj,
1694 const struct lu_attr *cattr)
1699 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1704 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1705 struct mdd_object *p, struct mdd_object *c,
1706 const struct lu_name *name, struct md_attr *ma,
1707 struct thandle *handle, int no_name, int is_dir)
1709 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1712 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1713 if (likely(no_name == 0)) {
1714 rc = mdo_declare_index_delete(env, p, name->ln_name,
1721 rc = mdo_declare_ref_del(env, p, handle);
1727 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1728 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1729 la->la_valid = LA_CTIME | LA_MTIME;
1730 rc = mdo_declare_attr_set(env, p, la, handle);
1735 rc = mdo_declare_ref_del(env, c, handle);
1739 rc = mdo_declare_ref_del(env, c, handle);
1743 la->la_valid = LA_CTIME;
1744 rc = mdo_declare_attr_set(env, c, la, handle);
1748 rc = mdd_declare_finish_unlink(env, c, handle);
1752 /* FIXME: need changelog for remove entry */
1753 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1761 * test if a file has an HSM archive
1762 * if HSM attributes are not found in ma update them from
1765 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1766 struct mdd_object *obj,
1771 if (!(ma->ma_valid & MA_HSM)) {
1772 /* no HSM MD provided, read xattr */
1773 struct lu_buf *hsm_buf;
1774 const size_t buflen = sizeof(struct hsm_attrs);
1777 hsm_buf = mdd_buf_get(env, NULL, 0);
1778 lu_buf_alloc(hsm_buf, buflen);
1779 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1780 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1781 lu_buf_free(hsm_buf);
1785 ma->ma_valid |= MA_HSM;
1787 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1793 * Delete name entry and the object.
1794 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1795 * does not exist for this object, and it could only happen during resending
1796 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1797 * is also needed in this case(needed by changelog), so we have to add another
1798 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1799 * the ENOENT failure should be able to be fixed by redo mechanism.
1801 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1802 struct md_object *cobj, const struct lu_name *lname,
1803 struct md_attr *ma, int no_name)
1805 const char *name = lname->ln_name;
1806 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1807 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1808 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1809 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1810 struct mdd_object *mdd_cobj = NULL;
1811 struct mdd_device *mdd = mdo2mdd(pobj);
1812 struct thandle *handle;
1813 int rc, is_dir = 0, cl_flags = 0;
1816 /* cobj == NULL means only delete name entry */
1817 if (likely(cobj != NULL)) {
1818 mdd_cobj = md2mdd_obj(cobj);
1819 if (mdd_object_exists(mdd_cobj) == 0)
1823 rc = mdd_la_get(env, mdd_pobj, pattr);
1827 if (likely(mdd_cobj != NULL)) {
1829 rc = mdd_la_get(env, mdd_cobj, cattr);
1833 is_dir = S_ISDIR(cattr->la_mode);
1834 /* search for an existing archive.
1835 * we should check ahead as the object
1836 * can be destroyed in this transaction */
1837 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1838 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1841 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1845 handle = mdd_trans_create(env, mdd);
1847 RETURN(PTR_ERR(handle));
1849 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1850 lname, ma, handle, no_name, is_dir);
1854 rc = mdd_trans_start(env, mdd, handle);
1858 if (likely(mdd_cobj != NULL))
1859 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1861 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1862 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1867 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1868 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1869 GOTO(cleanup, rc = 0);
1871 if (likely(mdd_cobj != NULL)) {
1872 rc = mdo_ref_del(env, mdd_cobj, handle);
1874 __mdd_index_insert_only(env, mdd_pobj,
1876 mdd_object_type(mdd_cobj),
1883 mdo_ref_del(env, mdd_cobj, handle);
1885 /* fetch updated nlink */
1886 rc = mdd_la_get(env, mdd_cobj, cattr);
1891 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1892 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1894 la->la_valid = LA_CTIME | LA_MTIME;
1895 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1899 /* Enough for only unlink the entry */
1900 if (unlikely(mdd_cobj == NULL))
1903 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1904 /* update ctime of an unlinked file only if it is still
1905 * opened or a link still exists */
1906 la->la_valid = LA_CTIME;
1907 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1912 /* XXX: this transfer to ma will be removed with LOD/OSP */
1913 ma->ma_attr = *cattr;
1914 ma->ma_valid |= MA_INODE;
1915 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1919 /* fetch updated nlink */
1920 rc = mdd_la_get(env, mdd_cobj, cattr);
1921 /* if object is removed then we can't get its attrs,
1923 if (rc == -ENOENT) {
1924 cattr->la_nlink = 0;
1928 if (cattr->la_nlink == 0) {
1929 ma->ma_attr = *cattr;
1930 ma->ma_valid |= MA_INODE;
1935 if (likely(mdd_cobj != NULL))
1936 mdd_write_unlock(env, mdd_cobj);
1939 if (cattr->la_nlink == 0)
1940 cl_flags |= CLF_UNLINK_LAST;
1942 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1944 rc = mdd_changelog_ns_store(env, mdd,
1945 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1946 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1951 rc = mdd_trans_stop(env, mdd, rc, handle);
1957 * The permission has been checked when obj created, no need check again.
1959 static int mdd_cd_sanity_check(const struct lu_env *env,
1960 struct mdd_object *obj)
1965 if (!obj || mdd_is_dead_obj(obj))
1971 static int mdd_create_data(const struct lu_env *env,
1972 struct md_object *pobj,
1973 struct md_object *cobj,
1974 const struct md_op_spec *spec,
1977 struct mdd_device *mdd = mdo2mdd(cobj);
1978 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1979 struct mdd_object *son = md2mdd_obj(cobj);
1980 struct thandle *handle;
1981 const struct lu_buf *buf;
1982 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1983 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1987 rc = mdd_cd_sanity_check(env, son);
1991 if (!md_should_create(spec->sp_cr_flags))
1995 * there are following use cases for this function:
1996 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1997 * striping can be specified or not
2000 rc = mdd_la_get(env, son, attr);
2004 /* calling ->ah_make_hint() is used to transfer information from parent */
2005 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2007 handle = mdd_trans_create(env, mdd);
2009 GOTO(out_free, rc = PTR_ERR(handle));
2012 * XXX: Setting the lov ea is not locked but setting the attr is locked?
2013 * Should this be fixed?
2015 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
2016 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
2017 spec->sp_cr_flags, spec->no_create);
2019 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
2020 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2021 spec->u.sp_ea.eadatalen);
2026 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
2027 XATTR_NAME_LOV, 0, handle);
2031 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
2036 rc = mdd_trans_start(env, mdd, handle);
2040 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
2046 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
2049 rc = mdd_trans_stop(env, mdd, rc, handle);
2055 static int mdd_declare_object_initialize(const struct lu_env *env,
2056 struct mdd_object *parent,
2057 struct mdd_object *child,
2058 const struct lu_attr *attr,
2059 struct thandle *handle)
2064 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
2065 if (!S_ISDIR(attr->la_mode))
2068 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
2073 rc = mdo_declare_ref_add(env, child, handle);
2077 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
2083 static int mdd_object_initialize(const struct lu_env *env,
2084 const struct lu_fid *pfid,
2085 struct mdd_object *child,
2086 struct lu_attr *attr, struct thandle *handle,
2087 const struct md_op_spec *spec)
2092 if (S_ISDIR(attr->la_mode)) {
2093 /* Add "." and ".." for newly created dir */
2094 mdo_ref_add(env, child, handle);
2095 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
2096 S_IFDIR, dot, handle);
2098 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2101 mdo_ref_del(env, child, handle);
2108 * This function checks whether it can create a file/dir under the
2109 * directory(@pobj). The directory(@pobj) is not being locked by
2112 * \param[in] env execution environment
2113 * \param[in] pobj the directory to create files
2114 * \param[in] pattr the attributes of the directory
2115 * \param[in] lname the name of the created file/dir
2116 * \param[in] cattr the attributes of the file/dir
2117 * \param[in] spec create specification
2119 * \retval = 0 it is allowed to create file/dir under
2121 * \retval negative error not allowed to create file/dir
2122 * under the directory
2124 static int mdd_create_sanity_check(const struct lu_env *env,
2125 struct md_object *pobj,
2126 const struct lu_attr *pattr,
2127 const struct lu_name *lname,
2128 struct lu_attr *cattr,
2129 struct md_op_spec *spec)
2131 struct mdd_thread_info *info = mdd_env_info(env);
2132 struct lu_fid *fid = &info->mti_fid;
2133 struct mdd_object *obj = md2mdd_obj(pobj);
2134 struct mdd_device *m = mdo2mdd(pobj);
2135 bool check_perm = true;
2140 if (mdd_is_dead_obj(obj))
2144 * In some cases this lookup is not needed - we know before if name
2145 * exists or not because MDT performs lookup for it.
2146 * name length check is done in lookup.
2148 if (spec->sp_cr_lookup) {
2150 * Check if the name already exist, though it will be checked in
2151 * _index_insert also, for avoiding rolling back if exists
2154 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2155 MAY_WRITE | MAY_EXEC);
2157 RETURN(rc ? : -EEXIST);
2159 /* Permission is already being checked in mdd_lookup */
2163 if (S_ISDIR(cattr->la_mode) &&
2164 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2165 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2166 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2168 if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
2169 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
2170 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2172 CERROR("%s: invalid lmv_user_md: magic = %x, "
2173 "stripe_offset = %d, stripe_count = %u: "
2174 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2175 le32_to_cpu(lum->lum_magic),
2176 (int)le32_to_cpu(lum->lum_stripe_offset),
2177 le32_to_cpu(lum->lum_stripe_count), rc);
2182 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2187 if (pattr->la_mode & S_ISGID) {
2188 cattr->la_gid = pattr->la_gid;
2189 if (S_ISDIR(cattr->la_mode)) {
2190 cattr->la_mode |= S_ISGID;
2191 cattr->la_valid |= LA_MODE;
2195 /* Inherit project ID from parent directory */
2196 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2197 cattr->la_projid = pattr->la_projid;
2198 if (S_ISDIR(cattr->la_mode)) {
2199 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2200 cattr->la_valid |= LA_FLAGS;
2202 cattr->la_valid |= LA_PROJID;
2205 rc = mdd_name_check(m, lname);
2209 switch (cattr->la_mode & S_IFMT) {
2211 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2213 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2214 RETURN(-ENAMETOOLONG);
2233 static int mdd_declare_create_object(const struct lu_env *env,
2234 struct mdd_device *mdd,
2235 struct mdd_object *p, struct mdd_object *c,
2236 struct lu_attr *attr,
2237 struct thandle *handle,
2238 const struct md_op_spec *spec,
2239 struct lu_buf *def_acl_buf,
2240 struct lu_buf *acl_buf,
2241 struct dt_allocation_hint *hint)
2243 const struct lu_buf *buf;
2246 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2251 #ifdef CONFIG_FS_POSIX_ACL
2252 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2253 /* if dir, then can inherit default ACl */
2254 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2255 XATTR_NAME_ACL_DEFAULT,
2261 if (acl_buf->lb_len > 0) {
2262 rc = mdo_declare_attr_set(env, c, attr, handle);
2266 rc = mdo_declare_xattr_set(env, c, acl_buf,
2267 XATTR_NAME_ACL_ACCESS, 0, handle);
2272 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2276 /* replay case, create LOV EA from client data */
2277 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2278 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2279 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2280 spec->u.sp_ea.eadatalen);
2281 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2287 if (S_ISLNK(attr->la_mode)) {
2288 const char *target_name = spec->u.sp_symname;
2289 int sym_len = strlen(target_name);
2290 const struct lu_buf *buf;
2292 buf = mdd_buf_get_const(env, target_name, sym_len);
2293 rc = dt_declare_record_write(env, mdd_object_child(c),
2299 if (spec->sp_cr_file_secctx_name != NULL) {
2300 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2301 spec->sp_cr_file_secctx_size);
2302 rc = mdo_declare_xattr_set(env, c, buf,
2303 spec->sp_cr_file_secctx_name, 0,
2312 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2313 struct mdd_object *p, struct mdd_object *c,
2314 const struct lu_name *name,
2315 struct lu_attr *attr,
2316 struct thandle *handle,
2317 const struct md_op_spec *spec,
2318 struct linkea_data *ldata,
2319 struct lu_buf *def_acl_buf,
2320 struct lu_buf *acl_buf,
2321 struct dt_allocation_hint *hint)
2325 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2326 def_acl_buf, acl_buf, hint);
2330 if (S_ISDIR(attr->la_mode)) {
2331 rc = mdo_declare_ref_add(env, p, handle);
2336 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2337 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2341 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2342 enum changelog_rec_type type;
2344 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2345 name->ln_name, handle);
2349 rc = mdd_declare_links_add(env, c, handle, ldata);
2354 la->la_valid = LA_CTIME | LA_MTIME;
2355 rc = mdo_declare_attr_set(env, p, la, handle);
2359 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2360 S_ISREG(attr->la_mode) ? CL_CREATE :
2361 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2363 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2372 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2373 struct lu_attr *la, struct lu_buf *def_acl_buf,
2374 struct lu_buf *acl_buf)
2379 if (S_ISLNK(la->la_mode)) {
2380 acl_buf->lb_len = 0;
2381 def_acl_buf->lb_len = 0;
2385 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2386 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2387 XATTR_NAME_ACL_DEFAULT);
2388 mdd_read_unlock(env, pobj);
2390 /* If there are default ACL, fix mode/ACL by default ACL */
2391 def_acl_buf->lb_len = rc;
2392 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2393 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2394 acl_buf->lb_len = rc;
2395 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2398 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2399 /* If there are no default ACL, fix mode by mask */
2400 struct lu_ucred *uc = lu_ucred(env);
2402 /* The create triggered by MDT internal events, such as
2403 * LFSCK reset, will not contain valid "uc". */
2404 if (unlikely(uc != NULL))
2405 la->la_mode &= ~uc->uc_umask;
2407 acl_buf->lb_len = 0;
2408 def_acl_buf->lb_len = 0;
2415 * Create a metadata object and initialize it, set acl, xattr.
2417 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2418 struct mdd_object *son, struct lu_attr *attr,
2419 struct md_op_spec *spec, struct lu_buf *acl_buf,
2420 struct lu_buf *def_acl_buf,
2421 struct dt_allocation_hint *hint,
2422 struct thandle *handle)
2424 const struct lu_buf *buf;
2427 mdd_write_lock(env, son, MOR_TGT_CHILD);
2428 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2433 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2434 * created in declare phase, they also needs to be added to master
2435 * object as sub-directory entry. So it has to initialize the master
2436 * object, then set dir striped EA.(in mdo_xattr_set) */
2437 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2440 GOTO(err_destroy, rc);
2443 * in case of replay we just set LOVEA provided by the client
2444 * XXX: I think it would be interesting to try "old" way where
2445 * MDT calls this xattr_set(LOV) in a different transaction.
2446 * probably this way we code can be made better.
2449 /* During creation, there are only a few cases we need do xattr_set to
2451 * 1. regular file: see comments above.
2452 * 2. dir: inherit default striping or pool settings from parent.
2453 * 3. create striped directory with provided stripeEA.
2454 * 4. create striped directory because inherit default layout from the
2457 if (spec->no_create ||
2458 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2459 S_ISDIR(attr->la_mode)) {
2460 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2461 spec->u.sp_ea.eadatalen);
2462 rc = mdo_xattr_set(env, son, buf,
2463 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2467 GOTO(err_destroy, rc);
2470 #ifdef CONFIG_FS_POSIX_ACL
2471 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2472 S_ISDIR(attr->la_mode)) {
2473 /* set default acl */
2474 rc = mdo_xattr_set(env, son, def_acl_buf,
2475 XATTR_NAME_ACL_DEFAULT, 0,
2478 GOTO(err_destroy, rc);
2480 /* set its own acl */
2481 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2482 rc = mdo_xattr_set(env, son, acl_buf,
2483 XATTR_NAME_ACL_ACCESS,
2486 GOTO(err_destroy, rc);
2490 if (S_ISLNK(attr->la_mode)) {
2491 struct lu_ucred *uc = lu_ucred_assert(env);
2492 struct dt_object *dt = mdd_object_child(son);
2493 const char *target_name = spec->u.sp_symname;
2494 int sym_len = strlen(target_name);
2497 buf = mdd_buf_get_const(env, target_name, sym_len);
2498 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2500 CFS_CAP_SYS_RESOURCE_MASK);
2505 GOTO(err_initlized, rc = -EFAULT);
2508 if (spec->sp_cr_file_secctx_name != NULL) {
2509 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2510 spec->sp_cr_file_secctx_size);
2511 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2514 GOTO(err_initlized, rc);
2518 if (unlikely(rc != 0)) {
2520 if (S_ISDIR(attr->la_mode)) {
2521 /* Drop the reference, no need to delete "."/"..",
2522 * because the object to be destroied directly. */
2523 rc2 = mdo_ref_del(env, son, handle);
2527 rc2 = mdo_ref_del(env, son, handle);
2531 mdo_destroy(env, son, handle);
2534 mdd_write_unlock(env, son);
2538 static int mdd_index_delete(const struct lu_env *env,
2539 struct mdd_object *mdd_pobj,
2540 struct lu_attr *cattr,
2541 const struct lu_name *lname)
2543 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2544 struct thandle *handle;
2548 handle = mdd_trans_create(env, mdd);
2550 RETURN(PTR_ERR(handle));
2552 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2557 if (S_ISDIR(cattr->la_mode)) {
2558 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2563 /* Since this will only be used in the error handler path,
2564 * Let's set the thandle to be local and not mess the transno */
2565 handle->th_local = 1;
2566 rc = mdd_trans_start(env, mdd, handle);
2570 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2571 S_ISDIR(cattr->la_mode), handle);
2575 rc = mdd_trans_stop(env, mdd, rc, handle);
2581 * Create object and insert it into namespace.
2583 * Two operations have to be performed:
2585 * - an allocation of a new object (->do_create()), and
2586 * - an insertion into a parent index (->dio_insert()).
2588 * Due to locking, operation order is not important, when both are
2589 * successful, *but* error handling cases are quite different:
2591 * - if insertion is done first, and following object creation fails,
2592 * insertion has to be rolled back, but this operation might fail
2593 * also leaving us with dangling index entry.
2595 * - if creation is done first, is has to be undone if insertion fails,
2596 * leaving us with leaked space, which is not good but not fatal.
2598 * It seems that creation-first is simplest solution, but it is sub-optimal
2604 * case, because second mkdir is bound to create object, only to
2605 * destroy it immediately.
2607 * To avoid this follow local file systems that do double lookup:
2609 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2610 * 1. create (mdd_create_object_internal())
2611 * 2. insert (__mdd_index_insert(), lookup again)
2613 * \param[in] pobj parent object
2614 * \param[in] lname name of child being created
2615 * \param[in,out] child child object being created
2616 * \param[in] spec additional create parameters
2617 * \param[in] ma attributes for new child object
2619 * \retval 0 on success
2620 * \retval negative errno on failure
2622 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2623 const struct lu_name *lname, struct md_object *child,
2624 struct md_op_spec *spec, struct md_attr *ma)
2626 struct mdd_thread_info *info = mdd_env_info(env);
2627 struct lu_attr *la = &info->mti_la_for_fix;
2628 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2629 struct mdd_object *son = md2mdd_obj(child);
2630 struct mdd_device *mdd = mdo2mdd(pobj);
2631 struct lu_attr *attr = &ma->ma_attr;
2632 struct thandle *handle;
2633 struct lu_attr *pattr = &info->mti_pattr;
2634 struct lu_buf acl_buf;
2635 struct lu_buf def_acl_buf;
2636 struct linkea_data *ldata = &info->mti_link_data;
2637 const char *name = lname->ln_name;
2638 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2643 rc = mdd_la_get(env, mdd_pobj, pattr);
2647 /* Sanity checks before big job. */
2648 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2652 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2653 GOTO(out_free, rc = -EINPROGRESS);
2655 handle = mdd_trans_create(env, mdd);
2657 GOTO(out_free, rc = PTR_ERR(handle));
2659 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2660 mdd->mdd_dt_conf.ddp_max_ea_size);
2661 acl_buf = info->mti_xattr_buf;
2662 def_acl_buf.lb_buf = info->mti_key;
2663 def_acl_buf.lb_len = sizeof(info->mti_key);
2664 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2668 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2670 memset(ldata, 0, sizeof(*ldata));
2671 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2672 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2675 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2676 &tfid, lname, 1, 0, ldata);
2678 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2679 mdd_object_fid(mdd_pobj),
2680 lname, 1, 0, ldata);
2683 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2684 handle, spec, ldata, &def_acl_buf, &acl_buf,
2689 rc = mdd_trans_start(env, mdd, handle);
2693 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2694 &def_acl_buf, hint, handle);
2698 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2699 mdd_write_lock(env, son, MOR_TGT_CHILD);
2700 son->mod_flags |= VOLATILE_OBJ;
2701 rc = mdd_orphan_insert(env, son, handle);
2702 GOTO(out_volatile, rc);
2704 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2705 attr->la_mode, name, handle);
2707 GOTO(err_created, rc);
2709 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2712 /* update parent directory mtime/ctime */
2714 la->la_valid = LA_CTIME | LA_MTIME;
2715 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2717 GOTO(err_insert, rc);
2723 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2724 rc2 = mdd_orphan_delete(env, son, handle);
2726 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2727 S_ISDIR(attr->la_mode),
2733 mdd_write_lock(env, son, MOR_TGT_CHILD);
2734 if (S_ISDIR(attr->la_mode)) {
2735 /* Drop the reference, no need to delete "."/"..",
2736 * because the object is to be destroyed directly. */
2737 rc2 = mdo_ref_del(env, son, handle);
2739 mdd_write_unlock(env, son);
2744 /* For volatile files drop one link immediately, since there is
2745 * no filename in the namespace, and save any error returned. */
2746 rc2 = mdo_ref_del(env, son, handle);
2748 mdd_write_unlock(env, son);
2749 if (unlikely(rc == 0))
2754 /* Don't destroy the volatile object on success */
2755 if (likely(rc != 0))
2756 mdo_destroy(env, son, handle);
2757 mdd_write_unlock(env, son);
2760 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2761 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2762 rc = mdd_changelog_ns_store(env, mdd,
2763 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2764 S_ISREG(attr->la_mode) ? CL_CREATE :
2765 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2766 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2769 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2771 /* If creation fails, it is most likely due to the remote update
2772 * failure, because local transaction will mostly succeed at
2773 * this stage. There is no easy way to rollback all of previous
2774 * updates, so let's remove the object from namespace, and
2775 * LFSCK should handle the orphan object. */
2776 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2777 mdd_index_delete(env, mdd_pobj, attr, lname);
2781 if (is_vmalloc_addr(ldata->ld_buf))
2782 /* if we vmalloced a large buffer drop it */
2783 lu_buf_free(ldata->ld_buf);
2785 /* The child object shouldn't be cached anymore */
2787 set_bit(LU_OBJECT_HEARD_BANSHEE,
2788 &child->mo_lu.lo_header->loh_flags);
2793 * Get locks on parents in proper order
2794 * RETURN: < 0 - error, rename_order if successful
2802 static int mdd_rename_order(const struct lu_env *env,
2803 struct mdd_device *mdd,
2804 struct mdd_object *src_pobj,
2805 const struct lu_attr *pattr,
2806 struct mdd_object *tgt_pobj)
2808 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2812 if (src_pobj == tgt_pobj)
2813 RETURN(MDD_RN_SAME);
2815 /* compared the parent child relationship of src_p&tgt_p */
2816 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2818 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2821 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2835 /* has not mdd_write{read}_lock on any obj yet. */
2836 static int mdd_rename_sanity_check(const struct lu_env *env,
2837 struct mdd_object *src_pobj,
2838 const struct lu_attr *pattr,
2839 struct mdd_object *tgt_pobj,
2840 const struct lu_attr *tpattr,
2841 struct mdd_object *sobj,
2842 const struct lu_attr *cattr,
2843 struct mdd_object *tobj,
2844 const struct lu_attr *tattr)
2849 /* XXX: when get here, sobj must NOT be NULL,
2850 * the other case has been processed in cld_rename
2851 * before mdd_rename and enable MDS_PERM_BYPASS. */
2855 * If we are using project inheritance, we only allow renames
2856 * into our tree when the project IDs are the same; otherwise
2857 * tree quota mechanism would be circumvented.
2859 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2860 tpattr->la_projid != cattr->la_projid) ||
2861 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2862 (pattr->la_projid != tpattr->la_projid)))
2865 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2869 /* XXX: when get here, "tobj == NULL" means tobj must
2870 * NOT exist (neither on remote MDS, such case has been
2871 * processed in cld_rename before mdd_rename and enable
2873 * So check may_create, but not check may_unlink. */
2875 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2876 (src_pobj != tgt_pobj));
2878 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2879 (src_pobj != tgt_pobj), 1);
2881 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2882 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2887 static int mdd_declare_rename(const struct lu_env *env,
2888 struct mdd_device *mdd,
2889 struct mdd_object *mdd_spobj,
2890 struct mdd_object *mdd_tpobj,
2891 struct mdd_object *mdd_sobj,
2892 struct mdd_object *mdd_tobj,
2893 const struct lu_name *sname,
2894 const struct lu_name *tname,
2896 struct linkea_data *ldata,
2897 struct thandle *handle)
2899 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2902 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2903 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2909 /* name from source dir */
2910 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2914 /* .. from source child */
2915 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2916 /* source child can be directory,
2917 * counted by source dir's nlink */
2918 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2921 if (mdd_spobj != mdd_tpobj) {
2922 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2927 rc = mdo_declare_index_insert(env, mdd_sobj,
2929 S_IFDIR, dotdot, handle);
2934 /* new target child can be directory,
2935 * counted by target dir's nlink */
2936 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2941 la->la_valid = LA_CTIME | LA_MTIME;
2942 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2946 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2950 la->la_valid = LA_CTIME;
2951 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2955 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2960 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2961 mdd_object_type(mdd_sobj),
2962 tname->ln_name, handle);
2966 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2967 /* delete target child in target parent directory */
2968 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2973 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2977 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2978 /* target child can be directory,
2979 * delete "." reference in target child directory */
2980 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2984 /* delete ".." reference in target parent directory */
2985 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2990 la->la_valid = LA_CTIME;
2991 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2995 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
3000 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
3008 /* src object can be remote that is why we use only fid and type of object */
3009 static int mdd_rename(const struct lu_env *env,
3010 struct md_object *src_pobj, struct md_object *tgt_pobj,
3011 const struct lu_fid *lf, const struct lu_name *lsname,
3012 struct md_object *tobj, const struct lu_name *ltname,
3015 const char *sname = lsname->ln_name;
3016 const char *tname = ltname->ln_name;
3017 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
3018 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
3019 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
3020 struct mdd_device *mdd = mdo2mdd(src_pobj);
3021 struct mdd_object *mdd_sobj = NULL; /* source object */
3022 struct mdd_object *mdd_tobj = NULL;
3023 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
3024 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3025 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
3026 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3027 struct thandle *handle;
3028 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3029 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
3030 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
3033 bool tobj_locked = 0;
3034 unsigned cl_flags = 0;
3039 mdd_tobj = md2mdd_obj(tobj);
3041 mdd_sobj = mdd_object_find(env, mdd, lf);
3042 if (IS_ERR(mdd_sobj))
3043 RETURN(PTR_ERR(mdd_sobj));
3045 rc = mdd_la_get(env, mdd_sobj, cattr);
3047 GOTO(out_pending, rc);
3049 rc = mdd_la_get(env, mdd_spobj, pattr);
3051 GOTO(out_pending, rc);
3054 rc = mdd_la_get(env, mdd_tobj, tattr);
3056 GOTO(out_pending, rc);
3057 /* search for an existing archive.
3058 * we should check ahead as the object
3059 * can be destroyed in this transaction */
3060 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3061 cl_flags |= CLF_RENAME_LAST_EXISTS;
3064 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3066 GOTO(out_pending, rc);
3068 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3069 mdd_sobj, cattr, mdd_tobj, tattr);
3071 GOTO(out_pending, rc);
3073 rc = mdd_name_check(mdd, ltname);
3075 GOTO(out_pending, rc);
3077 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3078 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3080 GOTO(out_pending, rc);
3082 handle = mdd_trans_create(env, mdd);
3084 GOTO(out_pending, rc = PTR_ERR(handle));
3086 memset(ldata, 0, sizeof(*ldata));
3087 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3088 lsname, mdd_object_fid(mdd_tpobj), ltname,
3093 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3094 mdd_tobj, lsname, ltname, ma, ldata, handle);
3098 rc = mdd_trans_start(env, mdd, handle);
3102 is_dir = S_ISDIR(cattr->la_mode);
3104 /* Remove source name from source directory */
3105 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3109 /* "mv dir1 dir2" needs "dir1/.." link update */
3110 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3111 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3113 GOTO(fixup_spobj2, rc);
3115 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3118 GOTO(fixup_spobj, rc);
3121 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3122 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3124 /* tname might been renamed to something else */
3125 GOTO(fixup_spobj, rc);
3128 /* Insert new fid with target name into target dir */
3129 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3132 GOTO(fixup_tpobj, rc);
3134 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3135 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3137 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3138 la->la_valid = LA_CTIME;
3139 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3141 GOTO(fixup_tpobj, rc);
3143 /* Update the linkEA for the source object */
3144 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3145 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3146 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3149 /* Old files might not have EA entry */
3150 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3151 lsname, handle, NULL, 0);
3152 mdd_write_unlock(env, mdd_sobj);
3153 /* We don't fail the transaction if the link ea can't be
3154 updated -- fid2path will use alternate lookup method. */
3157 /* Remove old target object
3158 * For tobj is remote case cmm layer has processed
3159 * and set tobj to NULL then. So when tobj is NOT NULL,
3160 * it must be local one.
3162 if (tobj && mdd_object_exists(mdd_tobj)) {
3163 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
3165 if (mdd_is_dead_obj(mdd_tobj)) {
3166 /* shld not be dead, something is wrong */
3167 CERROR("tobj is dead, something is wrong\n");
3171 mdo_ref_del(env, mdd_tobj, handle);
3173 /* Remove dot reference. */
3174 if (S_ISDIR(tattr->la_mode))
3175 mdo_ref_del(env, mdd_tobj, handle);
3178 /* fetch updated nlink */
3179 rc = mdd_la_get(env, mdd_tobj, tattr);
3181 CERROR("%s: Failed to get nlink for tobj "
3183 mdd2obd_dev(mdd)->obd_name,
3184 PFID(tpobj_fid), rc);
3185 GOTO(fixup_tpobj, rc);
3188 la->la_valid = LA_CTIME;
3189 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3191 CERROR("%s: Failed to set ctime for tobj "
3193 mdd2obd_dev(mdd)->obd_name,
3194 PFID(tpobj_fid), rc);
3195 GOTO(fixup_tpobj, rc);
3198 /* XXX: this transfer to ma will be removed with LOD/OSP */
3199 ma->ma_attr = *tattr;
3200 ma->ma_valid |= MA_INODE;
3201 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3204 CERROR("%s: Failed to unlink tobj "
3206 mdd2obd_dev(mdd)->obd_name,
3207 PFID(tpobj_fid), rc);
3208 GOTO(fixup_tpobj, rc);
3211 /* fetch updated nlink */
3212 rc = mdd_la_get(env, mdd_tobj, tattr);
3213 if (rc == -ENOENT) {
3214 /* the object got removed, let's
3215 * return the latest known attributes */
3216 tattr->la_nlink = 0;
3218 } else if (rc != 0) {
3219 CERROR("%s: Failed to get nlink for tobj "
3221 mdd2obd_dev(mdd)->obd_name,
3222 PFID(tpobj_fid), rc);
3223 GOTO(fixup_tpobj, rc);
3225 /* XXX: this transfer to ma will be removed with LOD/OSP */
3226 ma->ma_attr = *tattr;
3227 ma->ma_valid |= MA_INODE;
3229 if (tattr->la_nlink == 0)
3230 cl_flags |= CLF_RENAME_LAST;
3232 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3235 la->la_valid = LA_CTIME | LA_MTIME;
3236 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3238 GOTO(fixup_tpobj, rc);
3240 if (mdd_spobj != mdd_tpobj) {
3241 la->la_valid = LA_CTIME | LA_MTIME;
3242 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3244 GOTO(fixup_tpobj, rc);
3251 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3253 CWARN("tp obj fix error %d\n",rc2);
3255 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3256 !mdd_is_dead_obj(mdd_tobj)) {
3258 mdo_ref_add(env, mdd_tobj, handle);
3260 mdo_ref_add(env, mdd_tobj, handle);
3263 rc2 = __mdd_index_insert(env, mdd_tpobj,
3265 mdd_object_type(mdd_tobj),
3268 CWARN("tp obj fix error: rc = %d\n", rc2);
3273 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3274 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3276 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3277 mdd2obd_dev(mdd)->obd_name, rc2);
3280 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3283 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3284 mdd2obd_dev(mdd)->obd_name, rc2);
3289 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3290 mdd_object_type(mdd_sobj), sname,
3293 CWARN("sp obj fix error: rc = %d\n", rc2);
3298 mdd_write_unlock(env, mdd_tobj);
3301 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3302 mdd_tobj, tpobj_fid, lf, spobj_fid,
3303 ltname, lsname, handle);
3306 rc = mdd_trans_stop(env, mdd, rc, handle);
3309 mdd_object_put(env, mdd_sobj);
3314 * During migration once the parent FID has been changed,
3315 * we need update the parent FID in linkea.
3317 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3318 struct mdd_object *parent,
3319 struct mdd_object *newparent,
3320 struct mdd_object *child,
3321 const char *name, int namelen,
3322 struct thandle *handle,
3325 struct mdd_thread_info *info = mdd_env_info(env);
3326 struct linkea_data ldata = { NULL };
3327 struct lu_buf *buf = &info->mti_link_buf;
3333 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3334 if (buf->lb_buf == NULL)
3338 rc = mdd_links_read(env, child, &ldata);
3340 if (rc == -ENOENT || rc == -ENODATA)
3345 LASSERT(ldata.ld_leh != NULL);
3346 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3347 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3348 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3349 struct lu_name lname;
3352 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3355 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3356 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3357 ldata.ld_lee = (struct link_ea_entry *)
3358 ((char *)ldata.ld_lee +
3363 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3364 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3365 lname.ln_namelen, lname.ln_name,
3366 PFID(mdd_object_fid(newparent)));
3367 /* update to the new parent fid */
3368 linkea_entry_pack(ldata.ld_lee, &lname,
3369 mdd_object_fid(newparent));
3371 rc = mdd_declare_links_add(env, child, handle, &ldata);
3373 rc = mdd_links_write(env, child, &ldata, handle);
3379 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3380 struct mdd_object *parent,
3381 struct mdd_object *newparent,
3382 struct mdd_object *child,
3383 const char *name, int namelen,
3384 struct thandle *handle)
3386 return mdd_linkea_update_child_internal(env, parent, newparent,
3388 namelen, handle, true);
3391 static int mdd_linkea_update_child(const struct lu_env *env,
3392 struct mdd_object *parent,
3393 struct mdd_object *newparent,
3394 struct mdd_object *child,
3395 const char *name, int namelen,
3396 struct thandle *handle)
3398 return mdd_linkea_update_child_internal(env, parent, newparent,
3400 namelen, handle, false);
3403 static int mdd_update_linkea_internal(const struct lu_env *env,
3404 struct mdd_object *mdd_pobj,
3405 struct mdd_object *mdd_sobj,
3406 struct mdd_object *mdd_tobj,
3407 const struct lu_name *child_name,
3408 struct linkea_data *ldata,
3409 struct thandle *handle,
3412 struct mdd_thread_info *info = mdd_env_info(env);
3417 LASSERT(ldata->ld_buf != NULL);
3418 LASSERT(ldata->ld_leh != NULL);
3420 /* If it is mulitple links file, we need update the name entry for
3422 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3423 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3424 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3425 struct mdd_object *pobj;
3426 struct lu_name lname;
3429 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3431 pobj = mdd_object_find(env, mdd, &fid);
3433 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3434 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3439 if (!mdd_object_exists(pobj)) {
3440 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3441 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3445 if (pobj == mdd_pobj &&
3446 lname.ln_namelen == child_name->ln_namelen &&
3447 strncmp(lname.ln_name, child_name->ln_name,
3448 lname.ln_namelen) == 0) {
3449 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3450 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3455 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3456 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3457 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3460 /* Remove source name from source directory */
3461 /* Insert new fid with target name into target dir */
3462 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3467 rc = mdo_declare_index_insert(env, pobj,
3468 mdd_object_fid(mdd_tobj),
3469 mdd_object_type(mdd_tobj),
3470 lname.ln_name, handle);
3474 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3478 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3482 char *tmp_name = info->mti_key;
3484 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3485 /* lnamelen is too big(> NAME_MAX + 16),
3486 * something wrong about this linkea, let's
3488 CWARN("%s: the name %.*s is too long under "
3489 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3490 lname.ln_namelen, lname.ln_name,
3495 /* Note: lname might be without \0 at the end, see
3496 * linkea_entry_unpack(), let's add extra \0 by
3498 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3499 lname.ln_namelen, lname.ln_name);
3500 lname.ln_name = tmp_name;
3502 /* Let's check if this linkEA still valid, before
3503 * it might be packed into the RPC buffer. */
3504 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3505 &info->mti_fid, NULL);
3506 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3507 mdd_object_fid(mdd_sobj)))
3508 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3510 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3514 rc = __mdd_index_insert(env, pobj,
3515 mdd_object_fid(mdd_tobj),
3516 mdd_object_type(mdd_tobj),
3521 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3522 rc = mdo_ref_add(env, mdd_tobj, handle);
3523 mdd_write_unlock(env, mdd_tobj);
3527 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3528 mdo_ref_del(env, mdd_sobj, handle);
3529 mdd_write_unlock(env, mdd_sobj);
3532 mdd_object_put(env, pobj);
3536 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3543 static int mdd_migrate_xattrs(const struct lu_env *env,
3544 struct mdd_object *mdd_sobj,
3545 struct mdd_object *mdd_tobj)
3547 struct mdd_thread_info *info = mdd_env_info(env);
3548 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3550 struct thandle *handle;
3556 struct lu_buf list_xbuf;
3559 /* retrieve xattr list from the old object */
3560 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3561 if (list_xsize == -ENODATA)
3567 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3568 if (info->mti_big_buf.lb_buf == NULL)
3571 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3572 list_xbuf.lb_len = list_xsize;
3573 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3578 xname = list_xbuf.lb_buf;
3580 xlen = strnlen(xname, rem - 1) + 1;
3581 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3582 strcmp(XATTR_NAME_LMV, xname) == 0)
3585 /* For directory, if there are default layout, migrate here */
3586 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3587 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3590 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3591 if (xsize == -ENODATA)
3596 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3597 if (info->mti_link_buf.lb_buf == NULL)
3598 GOTO(out, rc = -ENOMEM);
3600 xbuf.lb_len = xsize;
3601 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3602 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3608 handle = mdd_trans_create(env, mdd);
3610 GOTO(out, rc = PTR_ERR(handle));
3612 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3615 GOTO(stop_trans, rc);
3616 /* Note: this transaction is part of migration, and it is not
3617 * the last step of migration, so we set th_local = 1 to avoid
3618 * update last rcvd for this transaction */
3619 handle->th_local = 1;
3620 rc = mdd_trans_start(env, mdd, handle);
3622 GOTO(stop_trans, rc);
3625 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3627 GOTO(stop_trans, rc = 0);
3629 if (unlikely(rc == -ENOSPC &&
3630 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3631 rc = linkea_overflow_shrink(
3632 (struct linkea_data *)(xbuf.lb_buf));
3633 if (likely(rc > 0)) {
3640 GOTO(stop_trans, rc);
3642 rc = mdd_trans_stop(env, mdd, rc, handle);
3647 memmove(xname, xname + xlen, rem);
3653 static int mdd_declare_migrate_create(const struct lu_env *env,
3654 struct mdd_object *mdd_pobj,
3655 struct mdd_object *mdd_sobj,
3656 struct mdd_object *mdd_tobj,
3657 struct md_op_spec *spec,
3659 union lmv_mds_md *mgr_ea,
3660 struct linkea_data *ldata,
3661 struct thandle *handle)
3663 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3664 const struct lu_buf *buf;
3668 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3669 handle, spec, NULL);
3673 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3678 if (S_ISLNK(la->la_mode)) {
3679 const char *target_name = spec->u.sp_symname;
3680 int sym_len = strlen(target_name);
3681 const struct lu_buf *buf;
3683 buf = mdd_buf_get_const(env, target_name, sym_len);
3684 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3688 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3689 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3694 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3695 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3696 spec->u.sp_ea.eadatalen);
3697 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3703 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3704 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3705 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3710 la_flag->la_valid = LA_FLAGS;
3711 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3712 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3717 static int mdd_migrate_create(const struct lu_env *env,
3718 struct mdd_object *mdd_pobj,
3719 struct mdd_object *mdd_sobj,
3720 struct mdd_object *mdd_tobj,
3721 const struct lu_name *lname,
3724 struct mdd_thread_info *info = mdd_env_info(env);
3725 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3726 struct md_op_spec *spec = &info->mti_spec;
3727 struct lu_buf lmm_buf = { NULL };
3728 struct lu_buf link_buf = { NULL };
3729 struct lu_buf mgr_buf;
3730 struct thandle *handle;
3731 struct lmv_mds_md_v1 *mgr_ea;
3732 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3733 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3735 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3739 /* prepare spec for create */
3740 memset(spec, 0, sizeof(*spec));
3741 spec->sp_cr_lookup = 0;
3742 spec->sp_feat = &dt_directory_features;
3743 if (S_ISLNK(la->la_mode)) {
3744 const struct lu_buf *buf;
3746 buf = lu_buf_check_and_alloc(
3747 &mdd_env_info(env)->mti_big_buf,
3750 link_buf.lb_len = la->la_size + 1;
3751 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3752 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3754 rc = rc != 0 ? rc : -EFAULT;
3755 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3756 mdd2obd_dev(mdd)->obd_name,
3757 PFID(mdd_object_fid(mdd_sobj)), rc);
3760 spec->u.sp_symname = link_buf.lb_buf;
3761 } else if (S_ISREG(la->la_mode)) {
3762 /* retrieve lov of the old object */
3763 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3764 if (rc != 0 && rc != -ENODATA)
3766 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3767 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3768 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3769 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3771 } else if (S_ISDIR(la->la_mode)) {
3772 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3773 if (rc == -ENODATA) {
3774 /* ignore the non-linkEA error */
3782 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3783 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3784 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3785 mgr_buf.lb_len = mgr_easize;
3786 mgr_ea = mgr_buf.lb_buf;
3787 memset(mgr_ea, 0, sizeof(*mgr_ea));
3788 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3789 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3790 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3791 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3792 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3793 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3795 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3797 handle = mdd_trans_create(env, mdd);
3799 GOTO(out_free, rc = PTR_ERR(handle));
3801 /* Note: this transaction is part of migration, and it is not
3802 * the last step of migration, so we set th_local = 1 to avoid
3803 * update last rcvd for this transaction */
3804 handle->th_local = 1;
3805 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3806 la, mgr_buf.lb_buf, ldata, handle);
3808 GOTO(stop_trans, rc);
3810 rc = mdd_trans_start(env, mdd, handle);
3812 GOTO(stop_trans, rc);
3814 /* don't set nlink from the original object */
3815 la->la_valid &= ~LA_NLINK;
3817 /* create the target object */
3818 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3821 GOTO(stop_trans, rc);
3823 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3824 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3826 GOTO(stop_trans, rc);
3829 /* Set MIGRATE EA on the source inode, so once the migration needs
3830 * to be re-done during failover, the re-do process can locate the
3831 * target object which is already being created. */
3832 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3834 GOTO(stop_trans, rc);
3836 /* Set immutable flag, so any modification is disabled until
3837 * the migration is done. Once the migration is interrupted,
3838 * if the resume process find the migrating object has both
3839 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3840 * flag and approve the migration */
3841 la_flag->la_valid = LA_FLAGS;
3842 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3843 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3846 rc = mdd_trans_stop(env, mdd, rc, handle);
3848 if (lmm_buf.lb_buf != NULL)
3849 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3853 static int mdd_migrate_entries(const struct lu_env *env,
3854 struct mdd_object *mdd_sobj,
3855 struct mdd_object *mdd_tobj)
3857 struct dt_object *next = mdd_object_child(mdd_sobj);
3858 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3859 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3860 struct thandle *handle;
3862 const struct dt_it_ops *iops;
3864 struct lu_dirent *ent;
3868 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3872 if (!dt_try_as_dir(env, next))
3873 GOTO(out_ent, rc = -ENOTDIR);
3875 * iterate directories
3877 iops = &next->do_index_ops->dio_it;
3878 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3880 GOTO(out_ent, rc = PTR_ERR(it));
3882 rc = iops->load(env, it, 0);
3884 rc = iops->next(env, it);
3888 * At this point and across for-loop:
3890 * rc == 0 -> ok, proceed.
3891 * rc > 0 -> end of directory.
3895 struct mdd_object *child;
3896 char *name = mdd_env_info(env)->mti_key;
3899 bool target_exist = false;
3901 len = iops->key_size(env, it);
3905 result = iops->rec(env, it, (struct dt_rec *)ent,
3906 LUDA_FID | LUDA_TYPE);
3907 if (result == -ESTALE)
3914 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3916 /* Insert new fid with target name into target dir */
3917 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3918 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3919 ent->lde_name[1] == '.'))
3922 child = mdd_object_find(env, mdd, &ent->lde_fid);
3924 GOTO(out, rc = PTR_ERR(child));
3926 /* child may not exist, but lu_object_attr will assert this,
3927 * get type from loh_attr directly */
3928 is_dir = S_ISDIR(child->mod_obj.mo_lu.lo_header->loh_attr);
3930 mdd_write_lock(env, child, MOR_SRC_CHILD);
3932 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3934 /* Check whether the name has been inserted to the target */
3935 if (dt_try_as_dir(env, dt_tobj)) {
3936 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3938 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3939 (struct dt_key *)name);
3940 if (unlikely(rc == 0))
3941 target_exist = true;
3944 handle = mdd_trans_create(env, mdd);
3946 GOTO(out_put, rc = PTR_ERR(handle));
3948 /* Note: this transaction is part of migration, and it is not
3949 * the last step of migration, so we set th_local = 1 to avoid
3950 * updating last rcvd for this transaction */
3951 handle->th_local = 1;
3952 if (likely(!target_exist)) {
3953 rc = mdo_declare_index_insert(env, mdd_tobj,
3955 child->mod_obj.mo_lu.lo_header->loh_attr,
3961 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3967 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3972 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3976 /* Update .. for child */
3977 rc = mdo_declare_index_delete(env, child, dotdot,
3982 rc = mdo_declare_index_insert(env, child,
3983 mdd_object_fid(mdd_tobj),
3984 S_IFDIR, dotdot, handle);
3989 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3996 rc = mdd_trans_start(env, mdd, handle);
3998 CERROR("%s: transaction start failed: rc = %d\n",
3999 mdd2obd_dev(mdd)->obd_name, rc);
4003 if (likely(!target_exist)) {
4004 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
4005 child->mod_obj.mo_lu.lo_header->loh_attr, name,
4011 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
4016 rc = __mdd_index_delete_only(env, child, dotdot,
4021 rc = __mdd_index_insert_only(env, child,
4022 mdd_object_fid(mdd_tobj), S_IFDIR,
4028 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
4030 strlen(name), handle);
4033 mdd_write_unlock(env, child);
4034 mdd_object_put(env, child);
4035 rc = mdd_trans_stop(env, mdd, rc, handle);
4039 result = iops->next(env, it);
4040 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
4041 GOTO(out, rc = -EINTR);
4043 if (result == -ESTALE)
4045 } while (result == 0);
4048 iops->fini(env, it);
4050 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
4054 static int mdd_declare_update_linkea(const struct lu_env *env,
4055 struct mdd_object *mdd_pobj,
4056 struct mdd_object *mdd_sobj,
4057 struct mdd_object *mdd_tobj,
4058 const struct lu_name *child_name,
4059 struct linkea_data *ldata,
4060 struct thandle *handle)
4062 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4063 child_name, ldata, handle, 1);
4066 static int mdd_update_linkea(const struct lu_env *env,
4067 struct mdd_object *mdd_pobj,
4068 struct mdd_object *mdd_sobj,
4069 struct mdd_object *mdd_tobj,
4070 const struct lu_name *child_name,
4071 struct linkea_data *ldata,
4072 struct thandle *handle)
4074 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4075 child_name, ldata, handle, 0);
4078 static int mdd_declare_migrate_update_name(const struct lu_env *env,
4079 struct mdd_object *mdd_pobj,
4080 struct mdd_object *mdd_sobj,
4081 struct mdd_object *mdd_tobj,
4082 const struct lu_name *lname,
4084 struct lu_attr *parent_la,
4085 struct linkea_data *ldata,
4086 struct thandle *handle)
4088 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4089 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4092 /* Revert IMMUTABLE flag */
4093 la_flag->la_valid = LA_FLAGS;
4094 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
4095 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
4099 /* delete entry from source dir */
4100 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
4104 if (ldata->ld_buf != NULL) {
4105 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
4106 mdd_tobj, lname, ldata, handle);
4111 if (S_ISREG(mdd_object_type(mdd_sobj))) {
4112 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4117 handle->th_complex = 1;
4118 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
4120 LU_XATTR_REPLACE, handle);
4125 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4126 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
4132 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
4133 mdd_object_type(mdd_tobj),
4134 lname->ln_name, handle);
4138 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
4142 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4143 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
4148 /* delete old object */
4149 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4153 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4154 /* delete old object */
4155 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4158 /* set nlink to 0 */
4159 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
4164 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
4168 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
4172 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4178 static int mdd_migrate_update_name(const struct lu_env *env,
4179 struct mdd_object *mdd_pobj,
4180 struct mdd_object *mdd_sobj,
4181 struct mdd_object *mdd_tobj,
4182 const struct lu_name *lname,
4185 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
4186 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4187 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4188 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4189 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
4190 struct thandle *handle;
4191 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
4192 const char *name = lname->ln_name;
4196 /* update time for parent */
4197 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
4198 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
4199 p_la->la_valid = LA_CTIME;
4201 rc = mdd_la_get(env, mdd_sobj, so_attr);
4205 ldata->ld_buf = NULL;
4206 rc = mdd_links_read(env, mdd_sobj, ldata);
4207 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4210 handle = mdd_trans_create(env, mdd);
4212 RETURN(PTR_ERR(handle));
4214 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4215 lname, so_attr, p_la, ldata,
4218 /* If the migration can not be fit in one transaction, just
4219 * leave it in the original MDT */
4221 GOTO(stop_trans, rc = 0);
4223 GOTO(stop_trans, rc);
4226 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4227 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4228 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4229 PFID(mdd_object_fid(mdd_tobj)));
4231 rc = mdd_trans_start(env, mdd, handle);
4233 GOTO(stop_trans, rc);
4235 /* Revert IMMUTABLE flag */
4236 la_flag->la_valid = LA_FLAGS;
4237 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4238 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4240 GOTO(stop_trans, rc);
4242 /* Remove source name from source directory */
4243 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4245 GOTO(stop_trans, rc);
4247 if (ldata->ld_buf != NULL) {
4248 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4249 lname, ldata, handle);
4251 GOTO(stop_trans, rc);
4253 /* linkea update might decrease the source object
4254 * nlink, let's get the attr again after ref_del */
4255 rc = mdd_la_get(env, mdd_sobj, so_attr);
4257 GOTO(stop_trans, rc);
4260 if (S_ISREG(so_attr->la_mode)) {
4261 if (so_attr->la_nlink == 1) {
4262 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4264 if (rc != 0 && rc != -ENODATA)
4265 GOTO(stop_trans, rc);
4267 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4269 LU_XATTR_REPLACE, handle);
4271 GOTO(stop_trans, rc);
4275 /* Insert new fid with target name into target dir */
4276 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4277 mdd_object_type(mdd_tobj), name, handle);
4279 GOTO(stop_trans, rc);
4281 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4283 mdd_sobj->mod_flags |= DEAD_OBJ;
4284 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4286 GOTO(out_unlock, rc);
4288 rc = mdd_orphan_insert(env, mdd_sobj, handle);
4290 GOTO(out_unlock, rc);
4292 mdo_ref_del(env, mdd_sobj, handle);
4294 mdo_ref_del(env, mdd_sobj, handle);
4296 /* Get the attr again after ref_del */
4297 rc = mdd_la_get(env, mdd_sobj, so_attr);
4299 GOTO(out_unlock, rc);
4301 ma->ma_attr = *so_attr;
4302 ma->ma_valid |= MA_INODE;
4304 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4306 GOTO(out_unlock, rc);
4308 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4309 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4310 mdo2fid(mdd_pobj), lname, lname, handle);
4312 CWARN("%s: changelog for migrate %s "DFID
4313 "under "DFID" failed: rc = %d\n",
4314 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4315 PFID(mdd_object_fid(mdd_sobj)),
4316 PFID(mdd_object_fid(mdd_pobj)), rc);
4317 /* Sigh, there are no easy way to migrate back the object, so
4318 * let's reset the result to 0 for now XXX */
4322 mdd_write_unlock(env, mdd_sobj);
4325 rc = mdd_trans_stop(env, mdd, rc, handle);
4330 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4331 const struct lu_fid *fid, __u32 *mdt_index)
4333 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4334 struct seq_server_site *ss;
4337 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4339 range->lsr_flags = LU_SEQ_RANGE_MDT;
4340 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4344 *mdt_index = range->lsr_index;
4349 * Check whether we should migrate the file/dir
4351 * < 0 permission check failed or other error.
4352 * = 0 the file can be migrated.
4353 * > 0 the file does not need to be migrated, mostly
4354 * for multiple link file
4356 static int mdd_migrate_sanity_check(const struct lu_env *env,
4357 struct mdd_object *pobj,
4358 const struct lu_attr *pattr,
4359 struct mdd_object *sobj,
4360 struct lu_attr *sattr)
4362 struct mdd_thread_info *info = mdd_env_info(env);
4363 struct linkea_data *ldata = &info->mti_link_data;
4364 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4366 struct lu_buf *mgr_buf;
4372 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4373 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4374 if (mgr_buf->lb_buf == NULL)
4377 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4379 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4381 /* If the object has migrateEA, it means IMMUTE flag
4382 * is being set by previous migration process, so it
4383 * needs to override the IMMUTE flag, otherwise the
4384 * following sanity check will fail */
4385 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4386 LMV_HASH_FLAG_MIGRATION) {
4387 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4389 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4390 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4391 mdd2obd_dev(mdd)->obd_name,
4392 PFID(mdd_object_fid(sobj)));
4396 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4397 sobj, sattr, NULL, NULL);
4401 /* Then it will check if the file should be migrated. If the file
4402 * has mulitple links, we only need migrate the file if all of its
4403 * entries has been migrated to the remote MDT */
4404 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4407 rc = mdd_links_read(env, sobj, ldata);
4409 /* For multiple links files, if there are no linkEA data at all,
4410 * means the file might be created before linkEA is enabled, and
4411 * all of its links should not be migrated yet, otherwise it
4412 * should have some linkEA there */
4413 if (rc == -ENOENT || rc == -ENODATA)
4418 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4419 /* If there are still links locally, then the file will not be
4421 LASSERT(ldata->ld_leh != NULL);
4423 /* If the linkEA is overflow, then means there are some unknown name
4424 * entries under unknown parents, that will prevent the migration. */
4425 if (unlikely(ldata->ld_leh->leh_overflow_time))
4428 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4429 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4430 struct lu_name lname;
4432 __u32 parent_mdt_index;
4434 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4436 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4439 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4443 /* Migrate the object only if none of its parents are on the
4445 if (parent_mdt_index != mdt_index)
4448 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4449 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4450 lname.ln_name, PFID(&fid));
4458 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4459 struct md_object *sobj, const struct lu_name *lname,
4460 struct md_object *tobj, struct md_attr *ma)
4462 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4463 struct mdd_device *mdd = mdo2mdd(pobj);
4464 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4465 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4466 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4467 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4468 bool created = false;
4472 /* If the file will being migrated, it will check whether
4473 * the file is being opened by someone else right now */
4474 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4475 if (mdd_sobj->mod_count > 0) {
4477 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4478 mdd2obd_dev(mdd)->obd_name,
4479 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4480 mdd_sobj->mod_count, -EBUSY);
4481 mdd_read_unlock(env, mdd_sobj);
4482 GOTO(put, rc = -EBUSY);
4484 mdd_read_unlock(env, mdd_sobj);
4486 rc = mdd_la_get(env, mdd_sobj, so_attr);
4490 rc = mdd_la_get(env, mdd_pobj, pattr);
4494 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4501 /* Sigh, it is impossible to finish all of migration in a single
4502 * transaction, for example migrating big directory entries to the
4503 * new MDT, it needs insert all of name entries of children in the
4506 * So migration will be done in multiple steps and transactions.
4508 * 1. create an orphan object on the remote MDT in one transaction.
4509 * 2. migrate extend attributes to the new target file/directory.
4510 * 3. For directory, migrate the entries to the new MDT and update
4511 * linkEA of each children. Because we can not migrate all entries
4512 * in a single transaction, so the migrating directory will become
4513 * a striped directory during migration, so once the process is
4514 * interrupted, the directory is still accessible. (During lookup,
4515 * client will locate the name by searching both original and target
4517 * 4. Finally, update the name/FID to point to the new file/directory
4518 * in a separate transaction.
4521 /* step 1: Check whether the orphan object has been created, and create
4522 * orphan object on the remote MDT if needed */
4523 if (!mdd_object_exists(mdd_tobj)) {
4524 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4531 LASSERT(mdd_object_exists(mdd_tobj));
4532 /* step 2: migrate xattr */
4533 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4537 /* step 3: migrate name entries to the orphan object */
4538 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4539 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4542 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4543 OBD_FAIL_MDS_REINT_NET_REP)))
4546 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4549 LASSERT(mdd_object_exists(mdd_tobj));
4550 /* step 4: update name entry to the new object */
4551 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4556 /* newly created target was not locked, don't cache its attributes */
4558 mdd_invalidate(env, tobj);
4563 const struct md_dir_operations mdd_dir_ops = {
4564 .mdo_is_subdir = mdd_is_subdir,
4565 .mdo_lookup = mdd_lookup,
4566 .mdo_create = mdd_create,
4567 .mdo_rename = mdd_rename,
4568 .mdo_link = mdd_link,
4569 .mdo_unlink = mdd_unlink,
4570 .mdo_create_data = mdd_create_data,
4571 .mdo_migrate = mdd_migrate,