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);
1019 LASSERT(sfid != NULL);
1020 LASSERT(spfid != NULL);
1021 LASSERT(sname != NULL);
1023 extsize = sname->ln_namelen + 1;
1025 rnm->cr_sfid = *sfid;
1026 rnm->cr_spfid = *spfid;
1028 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
1029 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
1030 rec->cr_namelen += extsize;
1033 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
1035 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
1037 if (jobid == NULL || jobid[0] == '\0')
1040 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
1043 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
1045 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
1047 ef->cr_extra_flags = eflags;
1050 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
1051 __u64 uid, __u64 gid)
1053 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
1055 uidgid->cr_uid = uid;
1056 uidgid->cr_gid = gid;
1059 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
1062 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
1064 clnid->cr_nid = nid;
1067 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, int flags)
1069 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
1071 omd->cr_openflags = (__u32)flags;
1074 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
1075 const char *xattr_name)
1077 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
1079 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
1082 /** Store a namespace change changelog record
1083 * If this fails, we must fail the whole transaction; we don't
1084 * want the change to commit without the log entry.
1085 * \param target - mdd_object of change
1086 * \param tpfid - target parent dir/object fid
1087 * \param sfid - source object fid
1088 * \param spfid - source parent fid
1089 * \param tname - target name string
1090 * \param sname - source name string
1091 * \param handle - transaction handle
1093 int mdd_changelog_ns_store(const struct lu_env *env,
1094 struct mdd_device *mdd,
1095 enum changelog_rec_type type,
1096 enum changelog_rec_flags crf,
1097 struct mdd_object *target,
1098 const struct lu_fid *tpfid,
1099 const struct lu_fid *sfid,
1100 const struct lu_fid *spfid,
1101 const struct lu_name *tname,
1102 const struct lu_name *sname,
1103 struct thandle *handle)
1105 const struct lu_ucred *uc = lu_ucred(env);
1106 struct llog_changelog_rec *rec;
1109 __u64 xflags = CLFE_INVALID;
1113 if (!mdd_changelog_enabled(env, mdd, type))
1116 LASSERT(tpfid != NULL);
1117 LASSERT(tname != NULL);
1118 LASSERT(handle != NULL);
1120 reclen = mdd_llog_record_calc_size(env, tname, sname);
1121 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1122 if (buf->lb_buf == NULL)
1126 crf &= CLF_FLAGMASK;
1127 crf |= CLF_EXTRA_FLAGS;
1130 if (uc->uc_jobid[0] != '\0')
1132 xflags |= CLFE_UIDGID;
1141 rec->cr.cr_flags = crf;
1143 if (crf & CLF_EXTRA_FLAGS) {
1144 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1145 if (xflags & CLFE_UIDGID)
1146 mdd_changelog_rec_extra_uidgid(&rec->cr,
1147 uc->uc_uid, uc->uc_gid);
1148 if (xflags & CLFE_NID)
1149 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1152 rec->cr.cr_type = (__u32)type;
1153 rec->cr.cr_pfid = *tpfid;
1154 rec->cr.cr_namelen = tname->ln_namelen;
1155 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1157 if (crf & CLF_RENAME)
1158 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1160 if (crf & CLF_JOBID)
1161 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1163 if (likely(target != NULL)) {
1164 rec->cr.cr_tfid = *mdo2fid(target);
1165 target->mod_cltime = ktime_get();
1167 fid_zero(&rec->cr.cr_tfid);
1170 rc = mdd_changelog_store(env, mdd, rec, handle);
1172 CERROR("%s: cannot store changelog record: type = %d, "
1173 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1174 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1175 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1182 static int __mdd_links_add(const struct lu_env *env,
1183 struct mdd_object *mdd_obj,
1184 struct linkea_data *ldata,
1185 const struct lu_name *lname,
1186 const struct lu_fid *pfid,
1187 int first, int check)
1191 if (ldata->ld_leh == NULL) {
1192 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1196 rc = linkea_data_new(ldata,
1197 &mdd_env_info(env)->mti_link_buf);
1204 rc = linkea_links_find(ldata, lname, pfid);
1205 if (rc && rc != -ENOENT)
1211 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1212 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1216 linkea_add_buf(ldata, lname, tfid);
1219 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1220 linkea_add_buf(ldata, lname, pfid);
1222 return linkea_add_buf(ldata, lname, pfid);
1225 static int __mdd_links_del(const struct lu_env *env,
1226 struct mdd_object *mdd_obj,
1227 struct linkea_data *ldata,
1228 const struct lu_name *lname,
1229 const struct lu_fid *pfid)
1233 if (ldata->ld_leh == NULL) {
1234 rc = mdd_links_read(env, mdd_obj, ldata);
1239 rc = linkea_links_find(ldata, lname, pfid);
1243 linkea_del_buf(ldata, lname);
1247 static int mdd_linkea_prepare(const struct lu_env *env,
1248 struct mdd_object *mdd_obj,
1249 const struct lu_fid *oldpfid,
1250 const struct lu_name *oldlname,
1251 const struct lu_fid *newpfid,
1252 const struct lu_name *newlname,
1253 int first, int check,
1254 struct linkea_data *ldata)
1259 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1262 LASSERT(oldpfid != NULL || newpfid != NULL);
1264 if (mdd_obj->mod_flags & DEAD_OBJ)
1265 /* Unnecessary to update linkEA for dead object. */
1268 if (oldpfid != NULL) {
1269 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1271 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1274 /* No changes done. */
1279 /* If renaming, add the new record */
1280 if (newpfid != NULL)
1281 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1287 int mdd_links_rename(const struct lu_env *env,
1288 struct mdd_object *mdd_obj,
1289 const struct lu_fid *oldpfid,
1290 const struct lu_name *oldlname,
1291 const struct lu_fid *newpfid,
1292 const struct lu_name *newlname,
1293 struct thandle *handle,
1294 struct linkea_data *ldata,
1295 int first, int check)
1300 if (ldata == NULL) {
1301 ldata = &mdd_env_info(env)->mti_link_data;
1302 memset(ldata, 0, sizeof(*ldata));
1303 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1304 newpfid, newlname, first, check, ldata);
1309 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1310 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1316 if (newlname == NULL)
1317 CERROR("link_ea add failed %d "DFID"\n",
1318 rc, PFID(mdd_object_fid(mdd_obj)));
1319 else if (oldpfid == NULL)
1320 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1321 newlname->ln_namelen, newlname->ln_name, rc,
1322 PFID(mdd_object_fid(mdd_obj)));
1323 else if (newpfid == NULL)
1324 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1325 oldlname->ln_namelen, oldlname->ln_name, rc,
1326 PFID(mdd_object_fid(mdd_obj)));
1328 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1329 "\n", oldlname->ln_namelen, oldlname->ln_name,
1330 newlname->ln_namelen, newlname->ln_name, rc,
1331 PFID(mdd_object_fid(mdd_obj)));
1334 if (is_vmalloc_addr(ldata->ld_buf))
1335 /* if we vmalloced a large buffer drop it */
1336 lu_buf_free(ldata->ld_buf);
1341 static inline int mdd_links_add(const struct lu_env *env,
1342 struct mdd_object *mdd_obj,
1343 const struct lu_fid *pfid,
1344 const struct lu_name *lname,
1345 struct thandle *handle,
1346 struct linkea_data *ldata, int first)
1348 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1349 pfid, lname, handle, ldata, first, 0);
1352 static inline int mdd_links_del(const struct lu_env *env,
1353 struct mdd_object *mdd_obj,
1354 const struct lu_fid *pfid,
1355 const struct lu_name *lname,
1356 struct thandle *handle)
1358 return mdd_links_rename(env, mdd_obj, pfid, lname,
1359 NULL, NULL, handle, NULL, 0, 0);
1362 /** Read the link EA into a temp buffer.
1363 * Uses the name_buf since it is generally large.
1364 * \retval IS_ERR err
1365 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1367 struct lu_buf *mdd_links_get(const struct lu_env *env,
1368 struct mdd_object *mdd_obj)
1370 struct linkea_data ldata = { NULL };
1373 rc = mdd_links_read(env, mdd_obj, &ldata);
1374 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1377 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1378 struct linkea_data *ldata, struct thandle *handle)
1380 const struct lu_buf *buf;
1383 if (ldata == NULL || ldata->ld_buf == NULL ||
1384 ldata->ld_leh == NULL)
1387 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1391 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1392 ldata->ld_leh->leh_len);
1393 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1394 if (unlikely(rc == -ENOSPC)) {
1395 rc = linkea_overflow_shrink(ldata);
1403 static int mdd_declare_links_add(const struct lu_env *env,
1404 struct mdd_object *mdd_obj,
1405 struct thandle *handle,
1406 struct linkea_data *ldata)
1412 if (ldata != NULL && ldata->ld_leh != NULL) {
1413 ea_len = ldata->ld_leh->leh_len;
1414 linkea = ldata->ld_buf->lb_buf;
1416 ea_len = MAX_LINKEA_SIZE;
1420 rc = mdo_declare_xattr_set(env, mdd_obj,
1421 mdd_buf_get_const(env, linkea, ea_len),
1422 XATTR_NAME_LINK, 0, handle);
1427 static inline int mdd_declare_links_del(const struct lu_env *env,
1428 struct mdd_object *c,
1429 struct thandle *handle)
1433 /* For directory, the linkEA will be removed together
1434 * with the object. */
1435 if (!S_ISDIR(mdd_object_type(c)))
1436 rc = mdd_declare_links_add(env, c, handle, NULL);
1441 static int mdd_declare_link(const struct lu_env *env,
1442 struct mdd_device *mdd,
1443 struct mdd_object *p,
1444 struct mdd_object *c,
1445 const struct lu_name *name,
1446 struct thandle *handle,
1448 struct linkea_data *data)
1450 struct lu_fid tfid = *mdo2fid(c);
1453 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1454 tfid.f_oid = cfs_fail_val;
1456 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1457 name->ln_name, handle);
1461 rc = mdo_declare_ref_add(env, c, handle);
1465 la->la_valid = LA_CTIME | LA_MTIME;
1466 rc = mdo_declare_attr_set(env, p, la, handle);
1470 la->la_valid = LA_CTIME;
1471 rc = mdo_declare_attr_set(env, c, la, handle);
1475 rc = mdd_declare_links_add(env, c, handle, data);
1479 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1485 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1486 struct md_object *src_obj, const struct lu_name *lname,
1489 const char *name = lname->ln_name;
1490 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1491 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1492 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1493 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1494 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1495 struct mdd_device *mdd = mdo2mdd(src_obj);
1496 struct thandle *handle;
1497 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1498 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1502 rc = mdd_la_get(env, mdd_sobj, cattr);
1506 rc = mdd_la_get(env, mdd_tobj, tattr);
1511 * If we are using project inheritance, we only allow hard link
1512 * creation in our tree when the project IDs are the same;
1513 * otherwise the tree quota mechanism could be circumvented.
1515 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1516 (tattr->la_projid != cattr->la_projid))
1519 handle = mdd_trans_create(env, mdd);
1521 GOTO(out_pending, rc = PTR_ERR(handle));
1523 memset(ldata, 0, sizeof(*ldata));
1525 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1526 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1528 /* Note: even this function will change ldata, but it comes from
1529 * thread_info, which is completely temporary and only seen in
1530 * this function, so we do not need reset ldata once it fails.*/
1531 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1532 lname, 0, 0, ldata);
1536 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1541 rc = mdd_trans_start(env, mdd, handle);
1545 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1546 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1549 GOTO(out_unlock, rc);
1551 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1552 rc = mdo_ref_add(env, mdd_sobj, handle);
1554 GOTO(out_unlock, rc);
1557 *tfid = *mdo2fid(mdd_sobj);
1558 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1559 tfid->f_oid = cfs_fail_val;
1561 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1562 mdd_object_type(mdd_sobj), name, handle);
1564 mdo_ref_del(env, mdd_sobj, handle);
1565 GOTO(out_unlock, rc);
1568 la->la_valid = LA_CTIME | LA_MTIME;
1569 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1571 GOTO(out_unlock, rc);
1573 la->la_valid = LA_CTIME;
1574 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1576 /* Note: The failure of links_add should not cause the
1577 * link failure, so do not check return value. */
1578 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1579 lname, handle, ldata, 0);
1583 mdd_write_unlock(env, mdd_sobj);
1585 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1586 mdo2fid(mdd_tobj), NULL, NULL,
1587 lname, NULL, handle);
1589 rc = mdd_trans_stop(env, mdd, rc, handle);
1590 if (is_vmalloc_addr(ldata->ld_buf))
1591 /* if we vmalloced a large buffer drop it */
1592 lu_buf_free(ldata->ld_buf);
1597 static int mdd_mark_orphan_object(const struct lu_env *env,
1598 struct mdd_object *obj, struct thandle *handle,
1601 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1604 if (!S_ISDIR(mdd_object_type(obj)))
1607 attr->la_valid = LA_FLAGS;
1608 attr->la_flags = LUSTRE_ORPHAN_FL;
1611 rc = mdo_declare_attr_set(env, obj, attr, handle);
1613 rc = mdo_attr_set(env, obj, attr, handle);
1618 static int mdd_declare_finish_unlink(const struct lu_env *env,
1619 struct mdd_object *obj,
1620 struct thandle *handle)
1624 /* Sigh, we do not know if the unlink object will become orphan in
1625 * declare phase, but fortunately the flags here does not matter
1626 * in current declare implementation */
1627 rc = mdd_mark_orphan_object(env, obj, handle, true);
1631 rc = mdo_declare_destroy(env, obj, handle);
1635 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1639 return mdd_declare_links_del(env, obj, handle);
1642 /* caller should take a lock before calling */
1643 int mdd_finish_unlink(const struct lu_env *env,
1644 struct mdd_object *obj, struct md_attr *ma,
1645 const struct mdd_object *pobj,
1646 const struct lu_name *lname,
1650 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1653 LASSERT(mdd_write_locked(env, obj) != 0);
1655 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1656 /* add new orphan and the object
1657 * will be deleted during mdd_close() */
1658 obj->mod_flags |= DEAD_OBJ;
1659 if (obj->mod_count) {
1660 rc = mdd_orphan_insert(env, obj, th);
1662 CDEBUG(D_HA, "Object "DFID" is inserted into "
1663 "orphan list, open count = %d\n",
1664 PFID(mdd_object_fid(obj)),
1667 CERROR("Object "DFID" fail to be an orphan, "
1668 "open count = %d, maybe cause failed "
1670 PFID(mdd_object_fid(obj)),
1673 /* mark object as an orphan here, not
1674 * before mdd_orphan_insert() as racing
1675 * mdd_la_get() may propagate ORPHAN_OBJ
1676 * causing the asserition */
1677 rc = mdd_mark_orphan_object(env, obj, th, false);
1679 rc = mdo_destroy(env, obj, th);
1681 } else if (!is_dir) {
1682 /* old files may not have link ea; ignore errors */
1683 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1691 * has mdd_write_lock on cobj already, but not on pobj yet
1693 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1694 const struct lu_attr *pattr,
1695 struct mdd_object *cobj,
1696 const struct lu_attr *cattr)
1701 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1706 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1707 struct mdd_object *p, struct mdd_object *c,
1708 const struct lu_name *name, struct md_attr *ma,
1709 struct thandle *handle, int no_name, int is_dir)
1711 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1714 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1715 if (likely(no_name == 0)) {
1716 rc = mdo_declare_index_delete(env, p, name->ln_name,
1723 rc = mdo_declare_ref_del(env, p, handle);
1729 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1730 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1731 la->la_valid = LA_CTIME | LA_MTIME;
1732 rc = mdo_declare_attr_set(env, p, la, handle);
1737 rc = mdo_declare_ref_del(env, c, handle);
1741 rc = mdo_declare_ref_del(env, c, handle);
1745 la->la_valid = LA_CTIME;
1746 rc = mdo_declare_attr_set(env, c, la, handle);
1750 rc = mdd_declare_finish_unlink(env, c, handle);
1754 /* FIXME: need changelog for remove entry */
1755 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1763 * test if a file has an HSM archive
1764 * if HSM attributes are not found in ma update them from
1767 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1768 struct mdd_object *obj,
1773 if (!(ma->ma_valid & MA_HSM)) {
1774 /* no HSM MD provided, read xattr */
1775 struct lu_buf *hsm_buf;
1776 const size_t buflen = sizeof(struct hsm_attrs);
1779 hsm_buf = mdd_buf_get(env, NULL, 0);
1780 lu_buf_alloc(hsm_buf, buflen);
1781 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1782 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1783 lu_buf_free(hsm_buf);
1787 ma->ma_valid |= MA_HSM;
1789 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1795 * Delete name entry and the object.
1796 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1797 * does not exist for this object, and it could only happen during resending
1798 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1799 * is also needed in this case(needed by changelog), so we have to add another
1800 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1801 * the ENOENT failure should be able to be fixed by redo mechanism.
1803 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1804 struct md_object *cobj, const struct lu_name *lname,
1805 struct md_attr *ma, int no_name)
1807 const char *name = lname->ln_name;
1808 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1809 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1810 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1811 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1812 struct mdd_object *mdd_cobj = NULL;
1813 struct mdd_device *mdd = mdo2mdd(pobj);
1814 struct thandle *handle;
1815 int rc, is_dir = 0, cl_flags = 0;
1818 /* cobj == NULL means only delete name entry */
1819 if (likely(cobj != NULL)) {
1820 mdd_cobj = md2mdd_obj(cobj);
1821 if (mdd_object_exists(mdd_cobj) == 0)
1825 rc = mdd_la_get(env, mdd_pobj, pattr);
1829 if (likely(mdd_cobj != NULL)) {
1831 rc = mdd_la_get(env, mdd_cobj, cattr);
1835 is_dir = S_ISDIR(cattr->la_mode);
1836 /* search for an existing archive.
1837 * we should check ahead as the object
1838 * can be destroyed in this transaction */
1839 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1840 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1843 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1847 handle = mdd_trans_create(env, mdd);
1849 RETURN(PTR_ERR(handle));
1851 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1852 lname, ma, handle, no_name, is_dir);
1856 rc = mdd_trans_start(env, mdd, handle);
1860 if (likely(mdd_cobj != NULL))
1861 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1863 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1864 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1869 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1870 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1871 GOTO(cleanup, rc = 0);
1873 if (likely(mdd_cobj != NULL)) {
1874 rc = mdo_ref_del(env, mdd_cobj, handle);
1876 __mdd_index_insert_only(env, mdd_pobj,
1878 mdd_object_type(mdd_cobj),
1885 mdo_ref_del(env, mdd_cobj, handle);
1887 /* fetch updated nlink */
1888 rc = mdd_la_get(env, mdd_cobj, cattr);
1893 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1894 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1896 la->la_valid = LA_CTIME | LA_MTIME;
1897 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1901 /* Enough for only unlink the entry */
1902 if (unlikely(mdd_cobj == NULL))
1905 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1906 /* update ctime of an unlinked file only if it is still
1907 * opened or a link still exists */
1908 la->la_valid = LA_CTIME;
1909 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1914 /* XXX: this transfer to ma will be removed with LOD/OSP */
1915 ma->ma_attr = *cattr;
1916 ma->ma_valid |= MA_INODE;
1917 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1921 /* fetch updated nlink */
1922 rc = mdd_la_get(env, mdd_cobj, cattr);
1923 /* if object is removed then we can't get its attrs,
1925 if (rc == -ENOENT) {
1926 cattr->la_nlink = 0;
1930 if (cattr->la_nlink == 0) {
1931 ma->ma_attr = *cattr;
1932 ma->ma_valid |= MA_INODE;
1937 if (likely(mdd_cobj != NULL))
1938 mdd_write_unlock(env, mdd_cobj);
1941 if (cattr->la_nlink == 0)
1942 cl_flags |= CLF_UNLINK_LAST;
1944 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1946 rc = mdd_changelog_ns_store(env, mdd,
1947 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1948 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1953 rc = mdd_trans_stop(env, mdd, rc, handle);
1959 * The permission has been checked when obj created, no need check again.
1961 static int mdd_cd_sanity_check(const struct lu_env *env,
1962 struct mdd_object *obj)
1967 if (!obj || mdd_is_dead_obj(obj))
1973 static int mdd_create_data(const struct lu_env *env,
1974 struct md_object *pobj,
1975 struct md_object *cobj,
1976 const struct md_op_spec *spec,
1979 struct mdd_device *mdd = mdo2mdd(cobj);
1980 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1981 struct mdd_object *son = md2mdd_obj(cobj);
1982 struct thandle *handle;
1983 const struct lu_buf *buf;
1984 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1985 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1989 rc = mdd_cd_sanity_check(env, son);
1993 if (!md_should_create(spec->sp_cr_flags))
1997 * there are following use cases for this function:
1998 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1999 * striping can be specified or not
2002 rc = mdd_la_get(env, son, attr);
2006 /* calling ->ah_make_hint() is used to transfer information from parent */
2007 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2009 handle = mdd_trans_create(env, mdd);
2011 GOTO(out_free, rc = PTR_ERR(handle));
2014 * XXX: Setting the lov ea is not locked but setting the attr is locked?
2015 * Should this be fixed?
2017 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
2018 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
2019 spec->sp_cr_flags, spec->no_create);
2021 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
2022 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2023 spec->u.sp_ea.eadatalen);
2028 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
2029 XATTR_NAME_LOV, 0, handle);
2033 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
2038 rc = mdd_trans_start(env, mdd, handle);
2042 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
2048 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
2051 rc = mdd_trans_stop(env, mdd, rc, handle);
2057 static int mdd_declare_object_initialize(const struct lu_env *env,
2058 struct mdd_object *parent,
2059 struct mdd_object *child,
2060 const struct lu_attr *attr,
2061 struct thandle *handle)
2066 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
2067 if (!S_ISDIR(attr->la_mode))
2070 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
2075 rc = mdo_declare_ref_add(env, child, handle);
2079 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
2085 static int mdd_object_initialize(const struct lu_env *env,
2086 const struct lu_fid *pfid,
2087 struct mdd_object *child,
2088 struct lu_attr *attr, struct thandle *handle,
2089 const struct md_op_spec *spec)
2094 if (S_ISDIR(attr->la_mode)) {
2095 /* Add "." and ".." for newly created dir */
2096 mdo_ref_add(env, child, handle);
2097 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
2098 S_IFDIR, dot, handle);
2100 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2103 mdo_ref_del(env, child, handle);
2110 * This function checks whether it can create a file/dir under the
2111 * directory(@pobj). The directory(@pobj) is not being locked by
2114 * \param[in] env execution environment
2115 * \param[in] pobj the directory to create files
2116 * \param[in] pattr the attributes of the directory
2117 * \param[in] lname the name of the created file/dir
2118 * \param[in] cattr the attributes of the file/dir
2119 * \param[in] spec create specification
2121 * \retval = 0 it is allowed to create file/dir under
2123 * \retval negative error not allowed to create file/dir
2124 * under the directory
2126 static int mdd_create_sanity_check(const struct lu_env *env,
2127 struct md_object *pobj,
2128 const struct lu_attr *pattr,
2129 const struct lu_name *lname,
2130 struct lu_attr *cattr,
2131 struct md_op_spec *spec)
2133 struct mdd_thread_info *info = mdd_env_info(env);
2134 struct lu_fid *fid = &info->mti_fid;
2135 struct mdd_object *obj = md2mdd_obj(pobj);
2136 struct mdd_device *m = mdo2mdd(pobj);
2137 bool check_perm = true;
2142 if (mdd_is_dead_obj(obj))
2146 * In some cases this lookup is not needed - we know before if name
2147 * exists or not because MDT performs lookup for it.
2148 * name length check is done in lookup.
2150 if (spec->sp_cr_lookup) {
2152 * Check if the name already exist, though it will be checked in
2153 * _index_insert also, for avoiding rolling back if exists
2156 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2157 MAY_WRITE | MAY_EXEC);
2159 RETURN(rc ? : -EEXIST);
2161 /* Permission is already being checked in mdd_lookup */
2165 if (S_ISDIR(cattr->la_mode) &&
2166 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2167 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2168 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2170 if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
2171 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
2172 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2174 CERROR("%s: invalid lmv_user_md: magic = %x, "
2175 "stripe_offset = %d, stripe_count = %u: "
2176 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2177 le32_to_cpu(lum->lum_magic),
2178 (int)le32_to_cpu(lum->lum_stripe_offset),
2179 le32_to_cpu(lum->lum_stripe_count), rc);
2184 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2189 if (pattr->la_mode & S_ISGID) {
2190 cattr->la_gid = pattr->la_gid;
2191 if (S_ISDIR(cattr->la_mode)) {
2192 cattr->la_mode |= S_ISGID;
2193 cattr->la_valid |= LA_MODE;
2197 /* Inherit project ID from parent directory */
2198 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2199 cattr->la_projid = pattr->la_projid;
2200 if (S_ISDIR(cattr->la_mode)) {
2201 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2202 cattr->la_valid |= LA_FLAGS;
2204 cattr->la_valid |= LA_PROJID;
2207 rc = mdd_name_check(m, lname);
2211 switch (cattr->la_mode & S_IFMT) {
2213 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2215 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2216 RETURN(-ENAMETOOLONG);
2235 static int mdd_declare_create_object(const struct lu_env *env,
2236 struct mdd_device *mdd,
2237 struct mdd_object *p, struct mdd_object *c,
2238 struct lu_attr *attr,
2239 struct thandle *handle,
2240 const struct md_op_spec *spec,
2241 struct lu_buf *def_acl_buf,
2242 struct lu_buf *acl_buf,
2243 struct dt_allocation_hint *hint)
2245 const struct lu_buf *buf;
2248 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2253 #ifdef CONFIG_FS_POSIX_ACL
2254 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2255 /* if dir, then can inherit default ACl */
2256 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2257 XATTR_NAME_ACL_DEFAULT,
2263 if (acl_buf->lb_len > 0) {
2264 rc = mdo_declare_attr_set(env, c, attr, handle);
2268 rc = mdo_declare_xattr_set(env, c, acl_buf,
2269 XATTR_NAME_ACL_ACCESS, 0, handle);
2274 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2278 /* replay case, create LOV EA from client data */
2279 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2280 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2281 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2282 spec->u.sp_ea.eadatalen);
2283 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2289 if (S_ISLNK(attr->la_mode)) {
2290 const char *target_name = spec->u.sp_symname;
2291 int sym_len = strlen(target_name);
2292 const struct lu_buf *buf;
2294 buf = mdd_buf_get_const(env, target_name, sym_len);
2295 rc = dt_declare_record_write(env, mdd_object_child(c),
2301 if (spec->sp_cr_file_secctx_name != NULL) {
2302 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2303 spec->sp_cr_file_secctx_size);
2304 rc = mdo_declare_xattr_set(env, c, buf,
2305 spec->sp_cr_file_secctx_name, 0,
2314 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2315 struct mdd_object *p, struct mdd_object *c,
2316 const struct lu_name *name,
2317 struct lu_attr *attr,
2318 struct thandle *handle,
2319 const struct md_op_spec *spec,
2320 struct linkea_data *ldata,
2321 struct lu_buf *def_acl_buf,
2322 struct lu_buf *acl_buf,
2323 struct dt_allocation_hint *hint)
2327 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2328 def_acl_buf, acl_buf, hint);
2332 if (S_ISDIR(attr->la_mode)) {
2333 rc = mdo_declare_ref_add(env, p, handle);
2338 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2339 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2343 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2344 enum changelog_rec_type type;
2346 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2347 name->ln_name, handle);
2351 rc = mdd_declare_links_add(env, c, handle, ldata);
2356 la->la_valid = LA_CTIME | LA_MTIME;
2357 rc = mdo_declare_attr_set(env, p, la, handle);
2361 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2362 S_ISREG(attr->la_mode) ? CL_CREATE :
2363 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2365 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2374 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2375 struct lu_attr *la, struct lu_buf *def_acl_buf,
2376 struct lu_buf *acl_buf)
2381 if (S_ISLNK(la->la_mode)) {
2382 acl_buf->lb_len = 0;
2383 def_acl_buf->lb_len = 0;
2387 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2388 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2389 XATTR_NAME_ACL_DEFAULT);
2390 mdd_read_unlock(env, pobj);
2392 /* If there are default ACL, fix mode/ACL by default ACL */
2393 def_acl_buf->lb_len = rc;
2394 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2395 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2396 acl_buf->lb_len = rc;
2397 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2400 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2401 /* If there are no default ACL, fix mode by mask */
2402 struct lu_ucred *uc = lu_ucred(env);
2404 /* The create triggered by MDT internal events, such as
2405 * LFSCK reset, will not contain valid "uc". */
2406 if (unlikely(uc != NULL))
2407 la->la_mode &= ~uc->uc_umask;
2409 acl_buf->lb_len = 0;
2410 def_acl_buf->lb_len = 0;
2417 * Create a metadata object and initialize it, set acl, xattr.
2419 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2420 struct mdd_object *son, struct lu_attr *attr,
2421 struct md_op_spec *spec, struct lu_buf *acl_buf,
2422 struct lu_buf *def_acl_buf,
2423 struct dt_allocation_hint *hint,
2424 struct thandle *handle)
2426 const struct lu_buf *buf;
2429 mdd_write_lock(env, son, MOR_TGT_CHILD);
2430 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2435 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2436 * created in declare phase, they also needs to be added to master
2437 * object as sub-directory entry. So it has to initialize the master
2438 * object, then set dir striped EA.(in mdo_xattr_set) */
2439 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2442 GOTO(err_destroy, rc);
2445 * in case of replay we just set LOVEA provided by the client
2446 * XXX: I think it would be interesting to try "old" way where
2447 * MDT calls this xattr_set(LOV) in a different transaction.
2448 * probably this way we code can be made better.
2451 /* During creation, there are only a few cases we need do xattr_set to
2453 * 1. regular file: see comments above.
2454 * 2. dir: inherit default striping or pool settings from parent.
2455 * 3. create striped directory with provided stripeEA.
2456 * 4. create striped directory because inherit default layout from the
2459 if (spec->no_create ||
2460 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2461 S_ISDIR(attr->la_mode)) {
2462 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2463 spec->u.sp_ea.eadatalen);
2464 rc = mdo_xattr_set(env, son, buf,
2465 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2469 GOTO(err_destroy, rc);
2472 #ifdef CONFIG_FS_POSIX_ACL
2473 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2474 S_ISDIR(attr->la_mode)) {
2475 /* set default acl */
2476 rc = mdo_xattr_set(env, son, def_acl_buf,
2477 XATTR_NAME_ACL_DEFAULT, 0,
2480 GOTO(err_destroy, rc);
2482 /* set its own acl */
2483 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2484 rc = mdo_xattr_set(env, son, acl_buf,
2485 XATTR_NAME_ACL_ACCESS,
2488 GOTO(err_destroy, rc);
2492 if (S_ISLNK(attr->la_mode)) {
2493 struct lu_ucred *uc = lu_ucred_assert(env);
2494 struct dt_object *dt = mdd_object_child(son);
2495 const char *target_name = spec->u.sp_symname;
2496 int sym_len = strlen(target_name);
2499 buf = mdd_buf_get_const(env, target_name, sym_len);
2500 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2502 CFS_CAP_SYS_RESOURCE_MASK);
2507 GOTO(err_initlized, rc = -EFAULT);
2510 if (spec->sp_cr_file_secctx_name != NULL) {
2511 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2512 spec->sp_cr_file_secctx_size);
2513 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2516 GOTO(err_initlized, rc);
2520 if (unlikely(rc != 0)) {
2522 if (S_ISDIR(attr->la_mode)) {
2523 /* Drop the reference, no need to delete "."/"..",
2524 * because the object to be destroied directly. */
2525 rc2 = mdo_ref_del(env, son, handle);
2529 rc2 = mdo_ref_del(env, son, handle);
2533 mdo_destroy(env, son, handle);
2536 mdd_write_unlock(env, son);
2540 static int mdd_index_delete(const struct lu_env *env,
2541 struct mdd_object *mdd_pobj,
2542 struct lu_attr *cattr,
2543 const struct lu_name *lname)
2545 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2546 struct thandle *handle;
2550 handle = mdd_trans_create(env, mdd);
2552 RETURN(PTR_ERR(handle));
2554 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2559 if (S_ISDIR(cattr->la_mode)) {
2560 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2565 /* Since this will only be used in the error handler path,
2566 * Let's set the thandle to be local and not mess the transno */
2567 handle->th_local = 1;
2568 rc = mdd_trans_start(env, mdd, handle);
2572 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2573 S_ISDIR(cattr->la_mode), handle);
2577 rc = mdd_trans_stop(env, mdd, rc, handle);
2583 * Create object and insert it into namespace.
2585 * Two operations have to be performed:
2587 * - an allocation of a new object (->do_create()), and
2588 * - an insertion into a parent index (->dio_insert()).
2590 * Due to locking, operation order is not important, when both are
2591 * successful, *but* error handling cases are quite different:
2593 * - if insertion is done first, and following object creation fails,
2594 * insertion has to be rolled back, but this operation might fail
2595 * also leaving us with dangling index entry.
2597 * - if creation is done first, is has to be undone if insertion fails,
2598 * leaving us with leaked space, which is not good but not fatal.
2600 * It seems that creation-first is simplest solution, but it is sub-optimal
2606 * case, because second mkdir is bound to create object, only to
2607 * destroy it immediately.
2609 * To avoid this follow local file systems that do double lookup:
2611 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2612 * 1. create (mdd_create_object_internal())
2613 * 2. insert (__mdd_index_insert(), lookup again)
2615 * \param[in] pobj parent object
2616 * \param[in] lname name of child being created
2617 * \param[in,out] child child object being created
2618 * \param[in] spec additional create parameters
2619 * \param[in] ma attributes for new child object
2621 * \retval 0 on success
2622 * \retval negative errno on failure
2624 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2625 const struct lu_name *lname, struct md_object *child,
2626 struct md_op_spec *spec, struct md_attr *ma)
2628 struct mdd_thread_info *info = mdd_env_info(env);
2629 struct lu_attr *la = &info->mti_la_for_fix;
2630 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2631 struct mdd_object *son = md2mdd_obj(child);
2632 struct mdd_device *mdd = mdo2mdd(pobj);
2633 struct lu_attr *attr = &ma->ma_attr;
2634 struct thandle *handle;
2635 struct lu_attr *pattr = &info->mti_pattr;
2636 struct lu_buf acl_buf;
2637 struct lu_buf def_acl_buf;
2638 struct linkea_data *ldata = &info->mti_link_data;
2639 const char *name = lname->ln_name;
2640 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2645 rc = mdd_la_get(env, mdd_pobj, pattr);
2649 /* Sanity checks before big job. */
2650 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2654 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2655 GOTO(out_free, rc = -EINPROGRESS);
2657 handle = mdd_trans_create(env, mdd);
2659 GOTO(out_free, rc = PTR_ERR(handle));
2661 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2662 mdd->mdd_dt_conf.ddp_max_ea_size);
2663 acl_buf = info->mti_xattr_buf;
2664 def_acl_buf.lb_buf = info->mti_key;
2665 def_acl_buf.lb_len = sizeof(info->mti_key);
2666 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2670 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2672 memset(ldata, 0, sizeof(*ldata));
2673 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2674 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2677 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2678 &tfid, lname, 1, 0, ldata);
2680 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2681 mdd_object_fid(mdd_pobj),
2682 lname, 1, 0, ldata);
2685 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2686 handle, spec, ldata, &def_acl_buf, &acl_buf,
2691 rc = mdd_trans_start(env, mdd, handle);
2695 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2696 &def_acl_buf, hint, handle);
2700 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2701 mdd_write_lock(env, son, MOR_TGT_CHILD);
2702 son->mod_flags |= VOLATILE_OBJ;
2703 rc = mdd_orphan_insert(env, son, handle);
2704 GOTO(out_volatile, rc);
2706 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2707 attr->la_mode, name, handle);
2709 GOTO(err_created, rc);
2711 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2714 /* update parent directory mtime/ctime */
2716 la->la_valid = LA_CTIME | LA_MTIME;
2717 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2719 GOTO(err_insert, rc);
2725 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2726 rc2 = mdd_orphan_delete(env, son, handle);
2728 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2729 S_ISDIR(attr->la_mode),
2735 mdd_write_lock(env, son, MOR_TGT_CHILD);
2736 if (S_ISDIR(attr->la_mode)) {
2737 /* Drop the reference, no need to delete "."/"..",
2738 * because the object is to be destroyed directly. */
2739 rc2 = mdo_ref_del(env, son, handle);
2741 mdd_write_unlock(env, son);
2746 /* For volatile files drop one link immediately, since there is
2747 * no filename in the namespace, and save any error returned. */
2748 rc2 = mdo_ref_del(env, son, handle);
2750 mdd_write_unlock(env, son);
2751 if (unlikely(rc == 0))
2756 /* Don't destroy the volatile object on success */
2757 if (likely(rc != 0))
2758 mdo_destroy(env, son, handle);
2759 mdd_write_unlock(env, son);
2762 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2763 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2764 rc = mdd_changelog_ns_store(env, mdd,
2765 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2766 S_ISREG(attr->la_mode) ? CL_CREATE :
2767 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2768 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2771 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2773 /* If creation fails, it is most likely due to the remote update
2774 * failure, because local transaction will mostly succeed at
2775 * this stage. There is no easy way to rollback all of previous
2776 * updates, so let's remove the object from namespace, and
2777 * LFSCK should handle the orphan object. */
2778 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2779 mdd_index_delete(env, mdd_pobj, attr, lname);
2783 if (is_vmalloc_addr(ldata->ld_buf))
2784 /* if we vmalloced a large buffer drop it */
2785 lu_buf_free(ldata->ld_buf);
2787 /* The child object shouldn't be cached anymore */
2789 set_bit(LU_OBJECT_HEARD_BANSHEE,
2790 &child->mo_lu.lo_header->loh_flags);
2795 * Get locks on parents in proper order
2796 * RETURN: < 0 - error, rename_order if successful
2804 static int mdd_rename_order(const struct lu_env *env,
2805 struct mdd_device *mdd,
2806 struct mdd_object *src_pobj,
2807 const struct lu_attr *pattr,
2808 struct mdd_object *tgt_pobj)
2810 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2814 if (src_pobj == tgt_pobj)
2815 RETURN(MDD_RN_SAME);
2817 /* compared the parent child relationship of src_p&tgt_p */
2818 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2820 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2823 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2837 /* has not mdd_write{read}_lock on any obj yet. */
2838 static int mdd_rename_sanity_check(const struct lu_env *env,
2839 struct mdd_object *src_pobj,
2840 const struct lu_attr *pattr,
2841 struct mdd_object *tgt_pobj,
2842 const struct lu_attr *tpattr,
2843 struct mdd_object *sobj,
2844 const struct lu_attr *cattr,
2845 struct mdd_object *tobj,
2846 const struct lu_attr *tattr)
2851 /* XXX: when get here, sobj must NOT be NULL,
2852 * the other case has been processed in cld_rename
2853 * before mdd_rename and enable MDS_PERM_BYPASS. */
2857 * If we are using project inheritance, we only allow renames
2858 * into our tree when the project IDs are the same; otherwise
2859 * tree quota mechanism would be circumvented.
2861 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2862 tpattr->la_projid != cattr->la_projid) ||
2863 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2864 (pattr->la_projid != tpattr->la_projid)))
2867 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2871 /* XXX: when get here, "tobj == NULL" means tobj must
2872 * NOT exist (neither on remote MDS, such case has been
2873 * processed in cld_rename before mdd_rename and enable
2875 * So check may_create, but not check may_unlink. */
2877 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2878 (src_pobj != tgt_pobj));
2880 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2881 (src_pobj != tgt_pobj), 1);
2883 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2884 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2889 static int mdd_declare_rename(const struct lu_env *env,
2890 struct mdd_device *mdd,
2891 struct mdd_object *mdd_spobj,
2892 struct mdd_object *mdd_tpobj,
2893 struct mdd_object *mdd_sobj,
2894 struct mdd_object *mdd_tobj,
2895 const struct lu_name *sname,
2896 const struct lu_name *tname,
2898 struct linkea_data *ldata,
2899 struct thandle *handle)
2901 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2904 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2905 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2911 /* name from source dir */
2912 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2916 /* .. from source child */
2917 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2918 /* source child can be directory,
2919 * counted by source dir's nlink */
2920 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2923 if (mdd_spobj != mdd_tpobj) {
2924 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2929 rc = mdo_declare_index_insert(env, mdd_sobj,
2931 S_IFDIR, dotdot, handle);
2936 /* new target child can be directory,
2937 * counted by target dir's nlink */
2938 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2943 la->la_valid = LA_CTIME | LA_MTIME;
2944 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2948 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2952 la->la_valid = LA_CTIME;
2953 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2957 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2962 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2963 mdd_object_type(mdd_sobj),
2964 tname->ln_name, handle);
2968 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2969 /* delete target child in target parent directory */
2970 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2975 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2979 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2980 /* target child can be directory,
2981 * delete "." reference in target child directory */
2982 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2986 /* delete ".." reference in target parent directory */
2987 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2992 la->la_valid = LA_CTIME;
2993 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2997 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
3002 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
3010 /* src object can be remote that is why we use only fid and type of object */
3011 static int mdd_rename(const struct lu_env *env,
3012 struct md_object *src_pobj, struct md_object *tgt_pobj,
3013 const struct lu_fid *lf, const struct lu_name *lsname,
3014 struct md_object *tobj, const struct lu_name *ltname,
3017 const char *sname = lsname->ln_name;
3018 const char *tname = ltname->ln_name;
3019 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
3020 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
3021 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
3022 struct mdd_device *mdd = mdo2mdd(src_pobj);
3023 struct mdd_object *mdd_sobj = NULL; /* source object */
3024 struct mdd_object *mdd_tobj = NULL;
3025 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
3026 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3027 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
3028 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3029 struct thandle *handle;
3030 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3031 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
3032 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
3035 bool tobj_locked = 0;
3036 unsigned cl_flags = 0;
3041 mdd_tobj = md2mdd_obj(tobj);
3043 mdd_sobj = mdd_object_find(env, mdd, lf);
3044 if (IS_ERR(mdd_sobj))
3045 RETURN(PTR_ERR(mdd_sobj));
3047 rc = mdd_la_get(env, mdd_sobj, cattr);
3049 GOTO(out_pending, rc);
3051 rc = mdd_la_get(env, mdd_spobj, pattr);
3053 GOTO(out_pending, rc);
3056 rc = mdd_la_get(env, mdd_tobj, tattr);
3058 GOTO(out_pending, rc);
3059 /* search for an existing archive.
3060 * we should check ahead as the object
3061 * can be destroyed in this transaction */
3062 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3063 cl_flags |= CLF_RENAME_LAST_EXISTS;
3066 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3068 GOTO(out_pending, rc);
3070 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3071 mdd_sobj, cattr, mdd_tobj, tattr);
3073 GOTO(out_pending, rc);
3075 rc = mdd_name_check(mdd, ltname);
3077 GOTO(out_pending, rc);
3079 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3080 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3082 GOTO(out_pending, rc);
3084 handle = mdd_trans_create(env, mdd);
3086 GOTO(out_pending, rc = PTR_ERR(handle));
3088 memset(ldata, 0, sizeof(*ldata));
3089 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3090 lsname, mdd_object_fid(mdd_tpobj), ltname,
3095 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3096 mdd_tobj, lsname, ltname, ma, ldata, handle);
3100 rc = mdd_trans_start(env, mdd, handle);
3104 is_dir = S_ISDIR(cattr->la_mode);
3106 /* Remove source name from source directory */
3107 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3111 /* "mv dir1 dir2" needs "dir1/.." link update */
3112 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3113 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3115 GOTO(fixup_spobj2, rc);
3117 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3120 GOTO(fixup_spobj, rc);
3123 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3124 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3126 /* tname might been renamed to something else */
3127 GOTO(fixup_spobj, rc);
3130 /* Insert new fid with target name into target dir */
3131 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3134 GOTO(fixup_tpobj, rc);
3136 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3137 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3139 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3140 la->la_valid = LA_CTIME;
3141 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3143 GOTO(fixup_tpobj, rc);
3145 /* Update the linkEA for the source object */
3146 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3147 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3148 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3151 /* Old files might not have EA entry */
3152 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3153 lsname, handle, NULL, 0);
3154 mdd_write_unlock(env, mdd_sobj);
3155 /* We don't fail the transaction if the link ea can't be
3156 updated -- fid2path will use alternate lookup method. */
3159 /* Remove old target object
3160 * For tobj is remote case cmm layer has processed
3161 * and set tobj to NULL then. So when tobj is NOT NULL,
3162 * it must be local one.
3164 if (tobj && mdd_object_exists(mdd_tobj)) {
3165 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
3167 if (mdd_is_dead_obj(mdd_tobj)) {
3168 /* shld not be dead, something is wrong */
3169 CERROR("tobj is dead, something is wrong\n");
3173 mdo_ref_del(env, mdd_tobj, handle);
3175 /* Remove dot reference. */
3176 if (S_ISDIR(tattr->la_mode))
3177 mdo_ref_del(env, mdd_tobj, handle);
3180 /* fetch updated nlink */
3181 rc = mdd_la_get(env, mdd_tobj, tattr);
3183 CERROR("%s: Failed to get nlink for tobj "
3185 mdd2obd_dev(mdd)->obd_name,
3186 PFID(tpobj_fid), rc);
3187 GOTO(fixup_tpobj, rc);
3190 la->la_valid = LA_CTIME;
3191 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3193 CERROR("%s: Failed to set ctime for tobj "
3195 mdd2obd_dev(mdd)->obd_name,
3196 PFID(tpobj_fid), rc);
3197 GOTO(fixup_tpobj, rc);
3200 /* XXX: this transfer to ma will be removed with LOD/OSP */
3201 ma->ma_attr = *tattr;
3202 ma->ma_valid |= MA_INODE;
3203 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3206 CERROR("%s: Failed to unlink tobj "
3208 mdd2obd_dev(mdd)->obd_name,
3209 PFID(tpobj_fid), rc);
3210 GOTO(fixup_tpobj, rc);
3213 /* fetch updated nlink */
3214 rc = mdd_la_get(env, mdd_tobj, tattr);
3215 if (rc == -ENOENT) {
3216 /* the object got removed, let's
3217 * return the latest known attributes */
3218 tattr->la_nlink = 0;
3220 } else if (rc != 0) {
3221 CERROR("%s: Failed to get nlink for tobj "
3223 mdd2obd_dev(mdd)->obd_name,
3224 PFID(tpobj_fid), rc);
3225 GOTO(fixup_tpobj, rc);
3227 /* XXX: this transfer to ma will be removed with LOD/OSP */
3228 ma->ma_attr = *tattr;
3229 ma->ma_valid |= MA_INODE;
3231 if (tattr->la_nlink == 0)
3232 cl_flags |= CLF_RENAME_LAST;
3234 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3237 la->la_valid = LA_CTIME | LA_MTIME;
3238 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3240 GOTO(fixup_tpobj, rc);
3242 if (mdd_spobj != mdd_tpobj) {
3243 la->la_valid = LA_CTIME | LA_MTIME;
3244 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3246 GOTO(fixup_tpobj, rc);
3253 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3255 CWARN("tp obj fix error %d\n",rc2);
3257 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3258 !mdd_is_dead_obj(mdd_tobj)) {
3260 mdo_ref_add(env, mdd_tobj, handle);
3262 mdo_ref_add(env, mdd_tobj, handle);
3265 rc2 = __mdd_index_insert(env, mdd_tpobj,
3267 mdd_object_type(mdd_tobj),
3270 CWARN("tp obj fix error: rc = %d\n", rc2);
3275 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3276 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3278 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3279 mdd2obd_dev(mdd)->obd_name, rc2);
3282 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3285 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3286 mdd2obd_dev(mdd)->obd_name, rc2);
3291 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3292 mdd_object_type(mdd_sobj), sname,
3295 CWARN("sp obj fix error: rc = %d\n", rc2);
3300 mdd_write_unlock(env, mdd_tobj);
3303 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3304 mdd_tobj, tpobj_fid, lf, spobj_fid,
3305 ltname, lsname, handle);
3308 rc = mdd_trans_stop(env, mdd, rc, handle);
3311 mdd_object_put(env, mdd_sobj);
3316 * During migration once the parent FID has been changed,
3317 * we need update the parent FID in linkea.
3319 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3320 struct mdd_object *parent,
3321 struct mdd_object *newparent,
3322 struct mdd_object *child,
3323 const char *name, int namelen,
3324 struct thandle *handle,
3327 struct mdd_thread_info *info = mdd_env_info(env);
3328 struct linkea_data ldata = { NULL };
3329 struct lu_buf *buf = &info->mti_link_buf;
3335 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3336 if (buf->lb_buf == NULL)
3340 rc = mdd_links_read(env, child, &ldata);
3342 if (rc == -ENOENT || rc == -ENODATA)
3347 LASSERT(ldata.ld_leh != NULL);
3348 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3349 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3350 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3351 struct lu_name lname;
3354 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3357 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3358 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3359 ldata.ld_lee = (struct link_ea_entry *)
3360 ((char *)ldata.ld_lee +
3365 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3366 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3367 lname.ln_namelen, lname.ln_name,
3368 PFID(mdd_object_fid(newparent)));
3369 /* update to the new parent fid */
3370 linkea_entry_pack(ldata.ld_lee, &lname,
3371 mdd_object_fid(newparent));
3373 rc = mdd_declare_links_add(env, child, handle, &ldata);
3375 rc = mdd_links_write(env, child, &ldata, handle);
3381 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3382 struct mdd_object *parent,
3383 struct mdd_object *newparent,
3384 struct mdd_object *child,
3385 const char *name, int namelen,
3386 struct thandle *handle)
3388 return mdd_linkea_update_child_internal(env, parent, newparent,
3390 namelen, handle, true);
3393 static int mdd_linkea_update_child(const struct lu_env *env,
3394 struct mdd_object *parent,
3395 struct mdd_object *newparent,
3396 struct mdd_object *child,
3397 const char *name, int namelen,
3398 struct thandle *handle)
3400 return mdd_linkea_update_child_internal(env, parent, newparent,
3402 namelen, handle, false);
3405 static int mdd_update_linkea_internal(const struct lu_env *env,
3406 struct mdd_object *mdd_pobj,
3407 struct mdd_object *mdd_sobj,
3408 struct mdd_object *mdd_tobj,
3409 const struct lu_name *child_name,
3410 struct linkea_data *ldata,
3411 struct thandle *handle,
3414 struct mdd_thread_info *info = mdd_env_info(env);
3419 LASSERT(ldata->ld_buf != NULL);
3420 LASSERT(ldata->ld_leh != NULL);
3422 /* If it is mulitple links file, we need update the name entry for
3424 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3425 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3426 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3427 struct mdd_object *pobj;
3428 struct lu_name lname;
3431 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3433 pobj = mdd_object_find(env, mdd, &fid);
3435 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3436 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3441 if (!mdd_object_exists(pobj)) {
3442 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3443 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3447 if (pobj == mdd_pobj &&
3448 lname.ln_namelen == child_name->ln_namelen &&
3449 strncmp(lname.ln_name, child_name->ln_name,
3450 lname.ln_namelen) == 0) {
3451 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3452 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3457 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3458 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3459 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3462 /* Remove source name from source directory */
3463 /* Insert new fid with target name into target dir */
3464 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3469 rc = mdo_declare_index_insert(env, pobj,
3470 mdd_object_fid(mdd_tobj),
3471 mdd_object_type(mdd_tobj),
3472 lname.ln_name, handle);
3476 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3480 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3484 char *tmp_name = info->mti_key;
3486 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3487 /* lnamelen is too big(> NAME_MAX + 16),
3488 * something wrong about this linkea, let's
3490 CWARN("%s: the name %.*s is too long under "
3491 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3492 lname.ln_namelen, lname.ln_name,
3497 /* Note: lname might be without \0 at the end, see
3498 * linkea_entry_unpack(), let's add extra \0 by
3500 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3501 lname.ln_namelen, lname.ln_name);
3502 lname.ln_name = tmp_name;
3504 /* Let's check if this linkEA still valid, before
3505 * it might be packed into the RPC buffer. */
3506 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3507 &info->mti_fid, NULL);
3508 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3509 mdd_object_fid(mdd_sobj)))
3510 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3512 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3516 rc = __mdd_index_insert(env, pobj,
3517 mdd_object_fid(mdd_tobj),
3518 mdd_object_type(mdd_tobj),
3523 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3524 rc = mdo_ref_add(env, mdd_tobj, handle);
3525 mdd_write_unlock(env, mdd_tobj);
3529 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3530 mdo_ref_del(env, mdd_sobj, handle);
3531 mdd_write_unlock(env, mdd_sobj);
3534 mdd_object_put(env, pobj);
3538 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3545 static int mdd_migrate_xattrs(const struct lu_env *env,
3546 struct mdd_object *mdd_sobj,
3547 struct mdd_object *mdd_tobj)
3549 struct mdd_thread_info *info = mdd_env_info(env);
3550 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3552 struct thandle *handle;
3558 struct lu_buf list_xbuf;
3561 /* retrieve xattr list from the old object */
3562 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3563 if (list_xsize == -ENODATA)
3569 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3570 if (info->mti_big_buf.lb_buf == NULL)
3573 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3574 list_xbuf.lb_len = list_xsize;
3575 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3580 xname = list_xbuf.lb_buf;
3582 xlen = strnlen(xname, rem - 1) + 1;
3583 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3584 strcmp(XATTR_NAME_LMV, xname) == 0)
3587 /* For directory, if there are default layout, migrate here */
3588 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3589 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3592 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3593 if (xsize == -ENODATA)
3598 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3599 if (info->mti_link_buf.lb_buf == NULL)
3600 GOTO(out, rc = -ENOMEM);
3602 xbuf.lb_len = xsize;
3603 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3604 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3610 handle = mdd_trans_create(env, mdd);
3612 GOTO(out, rc = PTR_ERR(handle));
3614 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3617 GOTO(stop_trans, rc);
3618 /* Note: this transaction is part of migration, and it is not
3619 * the last step of migration, so we set th_local = 1 to avoid
3620 * update last rcvd for this transaction */
3621 handle->th_local = 1;
3622 rc = mdd_trans_start(env, mdd, handle);
3624 GOTO(stop_trans, rc);
3627 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3629 GOTO(stop_trans, rc = 0);
3631 if (unlikely(rc == -ENOSPC &&
3632 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3633 rc = linkea_overflow_shrink(
3634 (struct linkea_data *)(xbuf.lb_buf));
3635 if (likely(rc > 0)) {
3642 GOTO(stop_trans, rc);
3644 rc = mdd_trans_stop(env, mdd, rc, handle);
3649 memmove(xname, xname + xlen, rem);
3655 static int mdd_declare_migrate_create(const struct lu_env *env,
3656 struct mdd_object *mdd_pobj,
3657 struct mdd_object *mdd_sobj,
3658 struct mdd_object *mdd_tobj,
3659 struct md_op_spec *spec,
3661 union lmv_mds_md *mgr_ea,
3662 struct linkea_data *ldata,
3663 struct thandle *handle)
3665 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3666 const struct lu_buf *buf;
3670 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3671 handle, spec, NULL);
3675 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3680 if (S_ISLNK(la->la_mode)) {
3681 const char *target_name = spec->u.sp_symname;
3682 int sym_len = strlen(target_name);
3683 const struct lu_buf *buf;
3685 buf = mdd_buf_get_const(env, target_name, sym_len);
3686 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3690 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3691 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3696 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3697 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3698 spec->u.sp_ea.eadatalen);
3699 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3705 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3706 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3707 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3712 la_flag->la_valid = LA_FLAGS;
3713 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3714 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3719 static int mdd_migrate_create(const struct lu_env *env,
3720 struct mdd_object *mdd_pobj,
3721 struct mdd_object *mdd_sobj,
3722 struct mdd_object *mdd_tobj,
3723 const struct lu_name *lname,
3726 struct mdd_thread_info *info = mdd_env_info(env);
3727 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3728 struct md_op_spec *spec = &info->mti_spec;
3729 struct lu_buf lmm_buf = { NULL };
3730 struct lu_buf link_buf = { NULL };
3731 struct lu_buf mgr_buf;
3732 struct thandle *handle;
3733 struct lmv_mds_md_v1 *mgr_ea;
3734 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3735 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3737 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3741 /* prepare spec for create */
3742 memset(spec, 0, sizeof(*spec));
3743 spec->sp_cr_lookup = 0;
3744 spec->sp_feat = &dt_directory_features;
3745 if (S_ISLNK(la->la_mode)) {
3746 const struct lu_buf *buf;
3748 buf = lu_buf_check_and_alloc(
3749 &mdd_env_info(env)->mti_big_buf,
3752 link_buf.lb_len = la->la_size + 1;
3753 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3754 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3756 rc = rc != 0 ? rc : -EFAULT;
3757 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3758 mdd2obd_dev(mdd)->obd_name,
3759 PFID(mdd_object_fid(mdd_sobj)), rc);
3762 spec->u.sp_symname = link_buf.lb_buf;
3763 } else if (S_ISREG(la->la_mode)) {
3764 /* retrieve lov of the old object */
3765 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3766 if (rc != 0 && rc != -ENODATA)
3768 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3769 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3770 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3771 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3773 } else if (S_ISDIR(la->la_mode)) {
3774 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3775 if (rc == -ENODATA) {
3776 /* ignore the non-linkEA error */
3784 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3785 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3786 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3787 mgr_buf.lb_len = mgr_easize;
3788 mgr_ea = mgr_buf.lb_buf;
3789 memset(mgr_ea, 0, sizeof(*mgr_ea));
3790 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3791 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3792 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3793 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3794 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3795 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3797 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3799 handle = mdd_trans_create(env, mdd);
3801 GOTO(out_free, rc = PTR_ERR(handle));
3803 /* Note: this transaction is part of migration, and it is not
3804 * the last step of migration, so we set th_local = 1 to avoid
3805 * update last rcvd for this transaction */
3806 handle->th_local = 1;
3807 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3808 la, mgr_buf.lb_buf, ldata, handle);
3810 GOTO(stop_trans, rc);
3812 rc = mdd_trans_start(env, mdd, handle);
3814 GOTO(stop_trans, rc);
3816 /* don't set nlink from the original object */
3817 la->la_valid &= ~LA_NLINK;
3819 /* create the target object */
3820 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3823 GOTO(stop_trans, rc);
3825 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3826 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3828 GOTO(stop_trans, rc);
3831 /* Set MIGRATE EA on the source inode, so once the migration needs
3832 * to be re-done during failover, the re-do process can locate the
3833 * target object which is already being created. */
3834 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3836 GOTO(stop_trans, rc);
3838 /* Set immutable flag, so any modification is disabled until
3839 * the migration is done. Once the migration is interrupted,
3840 * if the resume process find the migrating object has both
3841 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3842 * flag and approve the migration */
3843 la_flag->la_valid = LA_FLAGS;
3844 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3845 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3848 rc = mdd_trans_stop(env, mdd, rc, handle);
3850 if (lmm_buf.lb_buf != NULL)
3851 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3855 static int mdd_migrate_entries(const struct lu_env *env,
3856 struct mdd_object *mdd_sobj,
3857 struct mdd_object *mdd_tobj)
3859 struct dt_object *next = mdd_object_child(mdd_sobj);
3860 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3861 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3862 struct thandle *handle;
3864 const struct dt_it_ops *iops;
3866 struct lu_dirent *ent;
3870 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3874 if (!dt_try_as_dir(env, next))
3875 GOTO(out_ent, rc = -ENOTDIR);
3877 * iterate directories
3879 iops = &next->do_index_ops->dio_it;
3880 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3882 GOTO(out_ent, rc = PTR_ERR(it));
3884 rc = iops->load(env, it, 0);
3886 rc = iops->next(env, it);
3890 * At this point and across for-loop:
3892 * rc == 0 -> ok, proceed.
3893 * rc > 0 -> end of directory.
3897 struct mdd_object *child;
3898 char *name = mdd_env_info(env)->mti_key;
3901 bool target_exist = false;
3903 len = iops->key_size(env, it);
3907 result = iops->rec(env, it, (struct dt_rec *)ent,
3908 LUDA_FID | LUDA_TYPE);
3909 if (result == -ESTALE)
3916 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3918 /* Insert new fid with target name into target dir */
3919 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3920 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3921 ent->lde_name[1] == '.'))
3924 child = mdd_object_find(env, mdd, &ent->lde_fid);
3926 GOTO(out, rc = PTR_ERR(child));
3928 /* child may not exist, but lu_object_attr will assert this,
3929 * get type from loh_attr directly */
3930 is_dir = S_ISDIR(child->mod_obj.mo_lu.lo_header->loh_attr);
3932 mdd_write_lock(env, child, MOR_SRC_CHILD);
3934 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3936 /* Check whether the name has been inserted to the target */
3937 if (dt_try_as_dir(env, dt_tobj)) {
3938 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3940 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3941 (struct dt_key *)name);
3942 if (unlikely(rc == 0))
3943 target_exist = true;
3946 handle = mdd_trans_create(env, mdd);
3948 GOTO(out_put, rc = PTR_ERR(handle));
3950 /* Note: this transaction is part of migration, and it is not
3951 * the last step of migration, so we set th_local = 1 to avoid
3952 * updating last rcvd for this transaction */
3953 handle->th_local = 1;
3954 if (likely(!target_exist)) {
3955 rc = mdo_declare_index_insert(env, mdd_tobj,
3957 child->mod_obj.mo_lu.lo_header->loh_attr,
3963 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3969 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3974 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3978 /* Update .. for child */
3979 rc = mdo_declare_index_delete(env, child, dotdot,
3984 rc = mdo_declare_index_insert(env, child,
3985 mdd_object_fid(mdd_tobj),
3986 S_IFDIR, dotdot, handle);
3991 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3998 rc = mdd_trans_start(env, mdd, handle);
4000 CERROR("%s: transaction start failed: rc = %d\n",
4001 mdd2obd_dev(mdd)->obd_name, rc);
4005 if (likely(!target_exist)) {
4006 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
4007 child->mod_obj.mo_lu.lo_header->loh_attr, name,
4013 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
4018 rc = __mdd_index_delete_only(env, child, dotdot,
4023 rc = __mdd_index_insert_only(env, child,
4024 mdd_object_fid(mdd_tobj), S_IFDIR,
4030 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
4032 strlen(name), handle);
4035 mdd_write_unlock(env, child);
4036 mdd_object_put(env, child);
4037 rc = mdd_trans_stop(env, mdd, rc, handle);
4041 result = iops->next(env, it);
4042 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
4043 GOTO(out, rc = -EINTR);
4045 if (result == -ESTALE)
4047 } while (result == 0);
4050 iops->fini(env, it);
4052 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
4056 static int mdd_declare_update_linkea(const struct lu_env *env,
4057 struct mdd_object *mdd_pobj,
4058 struct mdd_object *mdd_sobj,
4059 struct mdd_object *mdd_tobj,
4060 const struct lu_name *child_name,
4061 struct linkea_data *ldata,
4062 struct thandle *handle)
4064 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4065 child_name, ldata, handle, 1);
4068 static int mdd_update_linkea(const struct lu_env *env,
4069 struct mdd_object *mdd_pobj,
4070 struct mdd_object *mdd_sobj,
4071 struct mdd_object *mdd_tobj,
4072 const struct lu_name *child_name,
4073 struct linkea_data *ldata,
4074 struct thandle *handle)
4076 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4077 child_name, ldata, handle, 0);
4080 static int mdd_declare_migrate_update_name(const struct lu_env *env,
4081 struct mdd_object *mdd_pobj,
4082 struct mdd_object *mdd_sobj,
4083 struct mdd_object *mdd_tobj,
4084 const struct lu_name *lname,
4086 struct lu_attr *parent_la,
4087 struct linkea_data *ldata,
4088 struct thandle *handle)
4090 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4091 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4094 /* Revert IMMUTABLE flag */
4095 la_flag->la_valid = LA_FLAGS;
4096 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
4097 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
4101 /* delete entry from source dir */
4102 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
4106 if (ldata->ld_buf != NULL) {
4107 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
4108 mdd_tobj, lname, ldata, handle);
4113 if (S_ISREG(mdd_object_type(mdd_sobj))) {
4114 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4119 handle->th_complex = 1;
4120 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
4122 LU_XATTR_REPLACE, handle);
4127 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4128 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
4134 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
4135 mdd_object_type(mdd_tobj),
4136 lname->ln_name, handle);
4140 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
4144 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4145 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
4150 /* delete old object */
4151 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4155 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4156 /* delete old object */
4157 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4160 /* set nlink to 0 */
4161 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
4166 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
4170 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
4174 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
4180 static int mdd_migrate_update_name(const struct lu_env *env,
4181 struct mdd_object *mdd_pobj,
4182 struct mdd_object *mdd_sobj,
4183 struct mdd_object *mdd_tobj,
4184 const struct lu_name *lname,
4187 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
4188 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4189 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4190 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4191 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
4192 struct thandle *handle;
4193 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
4194 const char *name = lname->ln_name;
4198 /* update time for parent */
4199 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
4200 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
4201 p_la->la_valid = LA_CTIME;
4203 rc = mdd_la_get(env, mdd_sobj, so_attr);
4207 ldata->ld_buf = NULL;
4208 rc = mdd_links_read(env, mdd_sobj, ldata);
4209 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4212 handle = mdd_trans_create(env, mdd);
4214 RETURN(PTR_ERR(handle));
4216 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4217 lname, so_attr, p_la, ldata,
4220 /* If the migration can not be fit in one transaction, just
4221 * leave it in the original MDT */
4223 GOTO(stop_trans, rc = 0);
4225 GOTO(stop_trans, rc);
4228 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4229 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4230 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4231 PFID(mdd_object_fid(mdd_tobj)));
4233 rc = mdd_trans_start(env, mdd, handle);
4235 GOTO(stop_trans, rc);
4237 /* Revert IMMUTABLE flag */
4238 la_flag->la_valid = LA_FLAGS;
4239 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4240 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4242 GOTO(stop_trans, rc);
4244 /* Remove source name from source directory */
4245 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4247 GOTO(stop_trans, rc);
4249 if (ldata->ld_buf != NULL) {
4250 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4251 lname, ldata, handle);
4253 GOTO(stop_trans, rc);
4255 /* linkea update might decrease the source object
4256 * nlink, let's get the attr again after ref_del */
4257 rc = mdd_la_get(env, mdd_sobj, so_attr);
4259 GOTO(stop_trans, rc);
4262 if (S_ISREG(so_attr->la_mode)) {
4263 if (so_attr->la_nlink == 1) {
4264 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4266 if (rc != 0 && rc != -ENODATA)
4267 GOTO(stop_trans, rc);
4269 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4271 LU_XATTR_REPLACE, handle);
4273 GOTO(stop_trans, rc);
4277 /* Insert new fid with target name into target dir */
4278 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4279 mdd_object_type(mdd_tobj), name, handle);
4281 GOTO(stop_trans, rc);
4283 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4285 mdd_sobj->mod_flags |= DEAD_OBJ;
4286 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4288 GOTO(out_unlock, rc);
4290 rc = mdd_orphan_insert(env, mdd_sobj, handle);
4292 GOTO(out_unlock, rc);
4294 mdo_ref_del(env, mdd_sobj, handle);
4296 mdo_ref_del(env, mdd_sobj, handle);
4298 /* Get the attr again after ref_del */
4299 rc = mdd_la_get(env, mdd_sobj, so_attr);
4301 GOTO(out_unlock, rc);
4303 ma->ma_attr = *so_attr;
4304 ma->ma_valid |= MA_INODE;
4306 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4308 GOTO(out_unlock, rc);
4310 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4311 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4312 mdo2fid(mdd_pobj), lname, lname, handle);
4314 CWARN("%s: changelog for migrate %s "DFID
4315 "under "DFID" failed: rc = %d\n",
4316 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4317 PFID(mdd_object_fid(mdd_sobj)),
4318 PFID(mdd_object_fid(mdd_pobj)), rc);
4319 /* Sigh, there are no easy way to migrate back the object, so
4320 * let's reset the result to 0 for now XXX */
4324 mdd_write_unlock(env, mdd_sobj);
4327 rc = mdd_trans_stop(env, mdd, rc, handle);
4332 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4333 const struct lu_fid *fid, __u32 *mdt_index)
4335 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4336 struct seq_server_site *ss;
4339 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4341 range->lsr_flags = LU_SEQ_RANGE_MDT;
4342 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4346 *mdt_index = range->lsr_index;
4351 * Check whether we should migrate the file/dir
4353 * < 0 permission check failed or other error.
4354 * = 0 the file can be migrated.
4355 * > 0 the file does not need to be migrated, mostly
4356 * for multiple link file
4358 static int mdd_migrate_sanity_check(const struct lu_env *env,
4359 struct mdd_object *pobj,
4360 const struct lu_attr *pattr,
4361 struct mdd_object *sobj,
4362 struct lu_attr *sattr)
4364 struct mdd_thread_info *info = mdd_env_info(env);
4365 struct linkea_data *ldata = &info->mti_link_data;
4366 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4368 struct lu_buf *mgr_buf;
4374 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4375 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4376 if (mgr_buf->lb_buf == NULL)
4379 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4381 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4383 /* If the object has migrateEA, it means IMMUTE flag
4384 * is being set by previous migration process, so it
4385 * needs to override the IMMUTE flag, otherwise the
4386 * following sanity check will fail */
4387 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4388 LMV_HASH_FLAG_MIGRATION) {
4389 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4391 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4392 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4393 mdd2obd_dev(mdd)->obd_name,
4394 PFID(mdd_object_fid(sobj)));
4398 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4399 sobj, sattr, NULL, NULL);
4403 /* Then it will check if the file should be migrated. If the file
4404 * has mulitple links, we only need migrate the file if all of its
4405 * entries has been migrated to the remote MDT */
4406 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4409 rc = mdd_links_read(env, sobj, ldata);
4411 /* For multiple links files, if there are no linkEA data at all,
4412 * means the file might be created before linkEA is enabled, and
4413 * all of its links should not be migrated yet, otherwise it
4414 * should have some linkEA there */
4415 if (rc == -ENOENT || rc == -ENODATA)
4420 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4421 /* If there are still links locally, then the file will not be
4423 LASSERT(ldata->ld_leh != NULL);
4425 /* If the linkEA is overflow, then means there are some unknown name
4426 * entries under unknown parents, that will prevent the migration. */
4427 if (unlikely(ldata->ld_leh->leh_overflow_time))
4430 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4431 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4432 struct lu_name lname;
4434 __u32 parent_mdt_index;
4436 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4438 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4441 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4445 /* Migrate the object only if none of its parents are on the
4447 if (parent_mdt_index != mdt_index)
4450 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4451 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4452 lname.ln_name, PFID(&fid));
4460 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4461 struct md_object *sobj, const struct lu_name *lname,
4462 struct md_object *tobj, struct md_attr *ma)
4464 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4465 struct mdd_device *mdd = mdo2mdd(pobj);
4466 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4467 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4468 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4469 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4470 bool created = false;
4474 /* If the file will being migrated, it will check whether
4475 * the file is being opened by someone else right now */
4476 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4477 if (mdd_sobj->mod_count > 0) {
4479 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4480 mdd2obd_dev(mdd)->obd_name,
4481 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4482 mdd_sobj->mod_count, -EBUSY);
4483 mdd_read_unlock(env, mdd_sobj);
4484 GOTO(put, rc = -EBUSY);
4486 mdd_read_unlock(env, mdd_sobj);
4488 rc = mdd_la_get(env, mdd_sobj, so_attr);
4492 rc = mdd_la_get(env, mdd_pobj, pattr);
4496 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4503 /* Sigh, it is impossible to finish all of migration in a single
4504 * transaction, for example migrating big directory entries to the
4505 * new MDT, it needs insert all of name entries of children in the
4508 * So migration will be done in multiple steps and transactions.
4510 * 1. create an orphan object on the remote MDT in one transaction.
4511 * 2. migrate extend attributes to the new target file/directory.
4512 * 3. For directory, migrate the entries to the new MDT and update
4513 * linkEA of each children. Because we can not migrate all entries
4514 * in a single transaction, so the migrating directory will become
4515 * a striped directory during migration, so once the process is
4516 * interrupted, the directory is still accessible. (During lookup,
4517 * client will locate the name by searching both original and target
4519 * 4. Finally, update the name/FID to point to the new file/directory
4520 * in a separate transaction.
4523 /* step 1: Check whether the orphan object has been created, and create
4524 * orphan object on the remote MDT if needed */
4525 if (!mdd_object_exists(mdd_tobj)) {
4526 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4533 LASSERT(mdd_object_exists(mdd_tobj));
4534 /* step 2: migrate xattr */
4535 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4539 /* step 3: migrate name entries to the orphan object */
4540 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4541 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4544 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4545 OBD_FAIL_MDS_REINT_NET_REP)))
4548 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4551 LASSERT(mdd_object_exists(mdd_tobj));
4552 /* step 4: update name entry to the new object */
4553 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4558 /* newly created target was not locked, don't cache its attributes */
4560 mdd_invalidate(env, tobj);
4565 const struct md_dir_operations mdd_dir_ops = {
4566 .mdo_is_subdir = mdd_is_subdir,
4567 .mdo_lookup = mdd_lookup,
4568 .mdo_create = mdd_create,
4569 .mdo_rename = mdd_rename,
4570 .mdo_link = mdd_link,
4571 .mdo_unlink = mdd_unlink,
4572 .mdo_create_data = mdd_create_data,
4573 .mdo_migrate = mdd_migrate,